ground penetrating radar for the parameterisation of

272
Ground Penetrating Radar for the parameterisation of subsurface hydrological properties By Andrew Howe A thesis submitted to the University of London for the degree of Doctor of Philosophy Department of Geography King’s College London September 2000

Upload: others

Post on 16-Apr-2022

8 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Ground Penetrating Radar for the parameterisation of

Ground Penetrating Radar for the parameterisation of

subsurface hydrological properties

By

Andrew Howe

A thesis submitted to the University of London for the

degree of Doctor of Philosophy

Department of Geography

King’s College London

September 2000

Page 2: Ground Penetrating Radar for the parameterisation of

Andrew Howe

2

Plynlimon, Wales.

Page 3: Ground Penetrating Radar for the parameterisation of

Andrew Howe

3

Acknowledgements

I would like to thank all those who have contributed towards this thesis. In

particular my supervisor, Dr. Mark Mulligan, who has helped me through all

stages of this project.

Without dedicated field workers this project would not have been completed.

Key members of the team were Rebecca Chadwick, Matthew Charlton and

David Hewitt. Staff at the Institute of Hydrology, Plynlimon, in particular Jim

Hudson, provided invaluable data and expertise. Thanks also to my colleagues

at King’s College London, Elias Symeonakis, Benito Meza Diaz, Jim Griffiths,

Sophia Burke, Sotirios Koukoulas, Mauricio Rinc n-Romero and Matthew

Charlton for numerous ideas and discussions.

Funding for this research was provided by the Natural Environment Research

Council and Sensors & Software Inc., Ontario (NERC CASE studentship

GT4/96/175/EO).

Page 4: Ground Penetrating Radar for the parameterisation of

Andrew Howe

4

Abstract

The subsurface plays a fundamental role in catchment hydrological response to

rainfall events, but is often a highly simplified component of hydrological

models due to the lack of high resolution spatial subsurface data required for

model parameterisation and validation. This study uses ground penetrating radar

(GPR) in order to quantify soil thickness and soil moisture in a fully distributed

manner for one of the Plynlimon catchments in Wales, UK. Soil thickness is a

hydrologically important parameter since it exerts a strong control on total

profile moisture storage capacity and soil response to precipitation events, yet

distributed models often apply a single mean depth value to a catchment (Boer

et al., 1993). Similarly the soil moisture of the upper soil layers is crucial in

determining rates of infiltration and plays a critical role in the partitioning of

precipitation into runoff or subsurface flow (Orlandini et al., 1996). Both soil

thickness and soil moisture are known to be highly spatially variable so it is

inappropriate to handle them in a lumped manner for a distributed model.

Soil thickness has been sampled at 32 sites using a PulseEKKO 1000 system

(Sensors & Software Inc.). Results show GPR depth estimates to bedrock

accurate to 0.12m (RMS) compared with auger readings, once the appropriate

interface has been identified using auguring techniques. Analysis of velocity

data from multiple frequency common midpoint surveys (CMP’s) of these same

sites can provide a non-destructive estimate of near surface soil moisture

through use of an empirical equation relating propagation velocity and moisture

content (Topp et al., 1980). GPR is able to measure soil moisture with a RMS

error of 0.04m3/m3 compared with Theta probe and gravimetric techniques for

the same soil space. Using a Geographic Information System (GIS) based

dynamic hydrological model, a 25m grid size digital elevation model and spatial

maps of GPR derived soil thickness, catchment response to precipitation events

is modelled. Validation is carried out at the plot scale using an 18 month time

series of hourly rainfall, overland flow, throughflow and soil moisture and at the

catchment scale using hourly flow data provided by the Institute of Hydrology.

Page 5: Ground Penetrating Radar for the parameterisation of

Andrew Howe

5

Evaluation of model output at the catchment scale shows that scenarios which

use distributed soil thickness, derived from GPR, provide an improved

prediction of the catchment hydrograph (RMS error 0.34m3/s), compared to

scenarios in which a lumped parameter for soil thickness is assumed (RMS error

0.29-0.40m3/s).

Page 6: Ground Penetrating Radar for the parameterisation of

Andrew Howe

6

Table of Contents

Chapter 1: Overview1.1 Introduction 16

1.2 Current issues in subsurface hydrology 17

1.3 Research objectives 24

Chapter 2: Ground Penetrating Radar2.1 Introduction 26

2.2 GPR theory 28

2.3 GPR survey types 37

2.4 Relation between dielectric constant and hydrologic parameters 40

2.5 Contemporary soil and hydrological research using GPR 43

2.6 Conclusion 48

Chapter 3: Hydrological Modelling and Model Development3.1 Introduction 49

3.2 Hydrological models 51

3.2.1 Empirical models 51

3.2.2 Physically based models 52

3.2.3 Lumped vs. distributed models 52

3.2.4 Hydrological modelling and GIS 54

3.2.5 Why another hydrological model? 55

3.3 PCRaster 57

3.4 Building an integrated hydrological – GIS model 58

3.4.1 Precipitation 60

3.4.2 Evaporation 61

3.4.3 Infiltration 63

3.4.4 Soil moisture 69

3.4.5 Subsurface flow 72

3.4.6 Surface runoff 79

3.5 Conclusion 82

Page 7: Ground Penetrating Radar for the parameterisation of

Andrew Howe

7

Chapter 4: Fieldwork4.1 Introduction 83

4.2 The Plynlimon catchments 84

4.2.1 Climate 86

4.2.2 Hydrology 87

4.2.3 Geomorphology and soils 87

4.2.4 Vegetation 89

4.3 Data collection of hydrological parameters 90

4.3.1 Plot-scale collection of hydrological parameters 90

4.3.2 Catchment-scale collection of hydrological parameters 93

4.4 Topography 94

4.4.1 The catchment DEM 95

4.4.2 The hillslope DEM 98

4.5 Site selection using terrain attributes 101

4.5.1 Locating sample sites using topographic indices 105

4.5.2 Site location in the field 106

4.6 GPR survey methodology 107

4.6.1 Reflection survey design 108

4.6.2 Time window 111

4.6.3 Station spacing 111

4.6.4 Sampling interval 113

4.6.5 CMP survey design 114

4.6.6 Stacking 114

4.7 Validation of GPR subsurface data 115

4.7.1 Soil thickness 115

4.7.2 Soil moisture 116

4.7.3 Soil physical properties 117

4.8 Errors 119

4.9 Conclusion 120

Page 8: Ground Penetrating Radar for the parameterisation of

Andrew Howe

8

Chapter 5: Soil Thickness Data Analysis5.1 Introduction 121

5.2 Analysis methodology 121

5.2.1 Initial GPR profile display 123

5.2.2 Derivation of subsurface velocities 126

5.2.3 From time domain to depth domain 128

5.2.4 Direct wave masking of shallow reflectors 134

5.2.5 GPR depth measurement errors 135

5.2.6 Theoretical resolution of reflectors 137

5.3 Analysis of individual GPR traces for depth extraction 138

5.3.1 Reflector identification 138

5.3.2 Picking horizon position within a GPR trace 140

5.3.3 Peat soils (Class 1) 140

5.3.4 Simple hillslope soils (Class 2) 147

5.3.5 Complex hillslope soils (Class 3) 151

5.4 Soil thickness measurement using GPR 155

5.5 The spatial variation of soil thickness 161

5.5.1 Plan Curvature 164

5.5.2 Profile Curvature 165

5.5.3 Slope Angle 166

5.5.4 Elevation 168

5.5.5 Upslope Area 169

5.5.6 Wetness Index 170

5.6 Conclusion 176

Chapter 6: Soil Moisture Data Analysis6.1 Introduction 178

6.2 Analysis methodology 178

6.3 Comparing GPR VMC with conventional VMC measurement

techniques 182

6.4 Soil moisture from GPR surveys 186

6.5 Conclusion 191

Page 9: Ground Penetrating Radar for the parameterisation of

Andrew Howe

9

Chapter 7: Hydrological Modelling Results7.1 Introduction 192

7.2 The catchment model 192

7.3 Reminder: The effect of soil thickness on modelled discharge 193

7.4 Catchment model results 195

7.4.1 High flows 198

7.4.2 Intermediate and low flows 200

7.4.3 Catchment discharge frequency distributions 202

7.5 The spatial distribution of runoff 207

7.6 Field plot data and model results 213

7.6.1 Overland flow 217

7.6.2 Subsurface flow 224

7.6.3 Plot soil moisture dynamics 229

7.7 Catchment soil moisture dynamics and GPR 233

7.8 Sensitivity Analysis 236

7.9 Conclusion 237

Chapter 8: Conclusions 239

Bibliography 242

Appendix I 251

Appendix II 259

Appendix III 270

Page 10: Ground Penetrating Radar for the parameterisation of

Andrew Howe

10

List of symbols

c velocity of light (m/s)f frequency (MHz)ft transition frequency (MHz)K dielectric constantσ material electrical conductivityR reflection coefficientλ wavelength (m)z depth (m)v velocity (m/s)vrms root mean square velocity (m/s)t time (s)θ Volumetric soil moisture content (m3/m3)θsat Saturated volumetric soil moisture content (m3/m3)φ Porosity (m/m)Sw Saturation (m/m)M Cementation indexKsat Soil saturated hydraulic conductivity (m/s)Epot Potential evaporation (mm/hr)E Actual evaporation (mm/hr)RN Net radiation (W/m2)π piγ latent heat of vaporisation (J)ρb bulk density (g/cm3)ρd particle density (g/cm3)b pore interaction termσg geometric standard deviation of particle diameterd g geometric mean of particle diameterQ discharge (m3/s)A Area (m2)H Hydraulic head (m)x length (m)y width (m)β slope angle (degrees)a upslope area (m2)V Voltage (V)a0 theta probe coefficient 1a1 theta probe coefficient 2D total GPR pulse duration (ns)W Wetness index valued Soil thickness (m)

Page 11: Ground Penetrating Radar for the parameterisation of

Andrew Howe

11

List of figures

Figure 2.1. Theoretical EM pathways between transmitter and receiver

antennae.

27

Figure 2.2. a) Random orientation of electric dipoles due to thermal motion, b)

Polarization in an electric field, thermal motion active.

30

Figure 2.3. GPR footprint dimensions. 33Figure 2.4. Maximum theoretical resolution for the frequency range 100 – 1200

MHz for 2 dielectric values (plot derived using equation 2.12).

36

Figure 2.5. Reflection Survey - constant offset. 37Figure 2.6. Common midpoint survey (CMP) - variable offset. 38Figure 2.7. Transillumination survey. 39Figure 3.1. Precipitation module.

Figure 3.2. Evaporation module.

6062

Figure 3.3. Theoretical depth – porosity relationship. 65Figure 3.4. Saturated hydraulic conductivity with varying depth of wetting front.

Figure 3.5. Infiltration module.

Figure 3.6. Soil moisture module.

666871

Figure 3.7. Possible drainage paths from a grid cell using the D8 flow algorithm. 73Figure 3.8. Cross-section of two adjacent grid cells with different elevation, soil

thickness and water table depth.

75

Figure 3.9. Subsurface flow module. 78

Figure 3.10. Surface runoff module 81Figure 4.1. Cyff catchment aerial photograph with watershed and stream

network superimposed.

85

Figure 4.2. Total monthly rainfall recorded by the field plot TBR. 86Figure 4.3. A Cyff soil profile at ~500m above sea level. 88Figure 4.4. Runoff - subsurface flow field plot. 92Figure 4.5. 25m grid resolution DEM of the Cyff catchment, overlain with a map

of the logarithm of upslope drainage area for each cell.

97

Figure 4.6. Elevation map with survey points marked in red and location of the

field plot marked in blue.

99

Figure 4.7. 5m grid DEM of the hillslope overlain with LN(upslope area) map and

local drainage direction network (white lines indicate cell drainage paths).

100

Figure 4.8. Frequency distribution of pixel values for the three topographic

indices used to locate potential GPR survey sites.

104

Figure 4.9. Soil texture data from thirty samples. 108Figure 4.10. The impact of reducing profile length on interpreting subsurface

features.

109

Page 12: Ground Penetrating Radar for the parameterisation of

Andrew Howe

12

Figure 4.11. The change in coefficient of variation with an increasing number of

auger samples.

110

Figure 4.12. Long-axis footprint size for variable dielectric constant. 112Figure 4.13. Depth – mean porosity relationship derived for the Cyff catchment 118Figure 5.1. Deriving soil thickness from GPR surveys 122Figure 5.2. 900MHz GPR profile (site 20) showing the difference in detail

displayed using a constant gain compared to a SEC gain.

124

Figure 5.3. CMP surveys for site 56 and site 50 using 900MHz antennae. 127Figure 5.4. Initial 10 traces from site 50 plotted using Picker software. Airwave

arrival time marked by red, ground wave arrival by green and first picked reflector

by blue lines.

127

Figure 5.5. Theoretical wave travel path for offset antennae. 130Figure 5.6. Comparison between linear and non-linear depth calculation.

(900MHz, v=0.06m/ns, x=0.17m)

132

Figure 5.7. Application of a constant gain of 2 (a) and a power gain of 2 (b) for

site 40, 0m.

139

Figure 5.8. Dry bulk density measured at 18 sites within the Cyff catchment,

depth measured using a steel ruler.

141

Figure 5.9. GPR traces: - Peat soils. a) 2m GPR profile site 33, b) Raw

amplitude extracted from trace at position 1.00m, c) Amplitude squared at

trace position 1.00m

142

Figure 5.10. Site 37 profile using 450MHz and 900MHz antennae. 145Figure 5.11. Trace amplitude comparison for site 37 bedrock boundary (O/C

horizon) for 900 MHz and 450 MHz GPR trace at the same location.

146

Figure 5.12. GPR profile: - Class 2 hillslope soils. (Site 8, 900MHz antennae,

constant gain 10).

149

Figure 5.13. GPR trace amplitude characteristics for site 8, 900MHz GPR

survey.

150

Figure 5.14. GPR profile: - Class 3 hillslope soils. (Site 54). 152Figure 5.15. Stone fragments located at 0.5 - 0.6m depth. 153Figure 5.16. Radar traces corresponding to the location of four auger points 154Figure 5.17. Soil auger measurements plotted against GPR derived soil

thickness to the B-horizon.

157

Figure 5.18. Soil auger measurements plotted against GPR derived soil

thickness to the C-horizon.

157

Figure 5.19. Average soil auger depth for each site plotted against GPR average

site soil thickness to the B-horizon and C-horizon.

158

Figure 5.20. The variation between physically measured and GPR derived soil

thickness, where physical measurements are considered to be true depth.

159

Page 13: Ground Penetrating Radar for the parameterisation of

Andrew Howe

13

Figure 5.21. Plan curvature. 164Figure 5.22. Profile curvature 165Figure 5.23. Slope angle 166

Figure 5.24. The distribution of slope angle less than 10o in the catchment 167

Figure 5.25. Elevation. 168Figure 5.26. Natural logarithm (LN) of Upslope Area. 169Figure 5.27. Wetness Index 170Figure 5.28. Cell soil thickness (m) predicted from the distribution of wetness

index.

172

Figure 5.29. Grid cells ≥ 445,000m2 upslope area 174

Figure 5.30. Cell soil thickness (m) predicted from the distribution of wetness

index with stream network cells assigned a value of 0.01m soil thickness.

175

Figure 6.1. Soil moisture data analysis using CMP data. 179Figure 6.2. VMC – depth relationship with trend lines for site 54. 184Figure 6.3. Soil thickness - VMC relationship for site 48 with strong soil horizon

control on VMC.

185

Figure 6.4. GPR measured moisture and gravimetric moisture. 187Figure 6.5. GPR measured moisture and theta probe moisture. 187Figure 6.6. Frequency distribution of the error between GPR water depth and a)

theta probe measurements, b) gravimetric soil samples.

188

Figure 7.1. Modelled and recorded minimum and maximum catchment

discharge, from table 7.1.

196

Figure 7.2. Outflow hydrograph of measured discharge and GPR variable soil

thickness modelled discharge.

197

Figure 7.3. A period of high flow from 27th February 1999 to 3rd March 1999. 198Figure 7.4. Low flow hydrograph. 201

Figure 7.5. Total catchment discharge (m3/s) histograms (measured and

modelled).

203

Figure 7.6. The spatial distribution of total overland flow for model scenarios of

0.2m and 1.6m soil thickness.

207

Figure 7.7. The spatial distribution of total overland flow and subsurface flow

from the GPR variable soil thickness scenario.

208

Figure 7.8. GPR variable soil thickness model: - cell water depth vs. cell soil

thickness.

211

Figure 7.9. Rainfall, overland flow and throughflow measured at the field plot. 214Figure 7.10. Output variable change (totals) in response to changing plot cell soil

thickness.

216

Figure 7.11. Modelled plot cell overland flow (25m DEM). 217

Page 14: Ground Penetrating Radar for the parameterisation of

Andrew Howe

14

Figure 7.12. The changing pattern of the local drainage direction (LDD) network

after decreasing cell size from 25m to 5m.

219

Figure 7.13. Model output values for 5m-grid size. 220Figure 7.14. Modelled plot overland flow data for 5m grid resolution. 221

Figure 7.15. Frequency distributions of plot overland flow (m3/hr). 223

Figure 7.16. Modelled plot cell subsurface flow for three soil depth scenarios. 224Figure 7.17. Mean and maximum subsurface flow for all soil depth scenarios and

measured data.

225

Figure 7.18. Frequency distributions of plot throughflow (m3/hr). 226Figure 7.19. Field plot measured and modelled VMC for distributed soil

thickness scenario.

230

Figure 7.20. GPR measured VMC and variable depth scenario VMC for 29 sites

across the Cyff. The trend line is fitted to a 1:1 relationship, not through data

points.

234

Page 15: Ground Penetrating Radar for the parameterisation of

Andrew Howe

15

List of tables

Table 2.1. Typical values of dielectric constant (K), velocity, attenuation and

conductivity for earth materials in the GPR frequency range.

32

Table 2. 2. Ideal resolution (from equation 2.12) and footprint dimensions (from

equation 2.10 & 2.11) at 1m depth for selected operating frequencies.

35

Table 2.3. Potential methods for the collection of soil thickness and soil

moisture data using GPR.

47

Table 4.1. Field plot data record. 91Table 4.2. Data collected by Institute of Hydrology hydrometric network. 93Table 4.3. Terrain attributes of the plot cell for variable DEM grid size. 98Table 4.4. GPS and DEM feature co-ordinates. 106Table 4.5. Soil thickness variation at the field plot. 109Table 4.6. Theta probe calibration values. 116Table 5.1. Reflector depth using linear/non-linear methods of calculation. 131Table 5.2. Calculation of minimum resolution depth using an average velocity

of 0.06m/ns, derived from 29 CMP ground wave velocities.

135

Table 5.3. Positioning error effects on calculated velocities. 136Table 5.4. Soil thickness measurements for site 8 using an auger and 900MHz

GPR survey.

148

Table 5.5. Summary soil thickness data for all sample sites 156Table 5.6. Correlation coefficients of soil thickness variation with respect to

DEM derived terrain attributes.

163

Table 6.1. Site Vrms and VMC values calculated for the first subsurface reflector

after groundwave arrival. Only those sites with validation data are included.

181

Table 6.2. The difference between GPR measured water and water depth

measured using gravimetric and theta probe methods.

186

Table 6.3. Signed-rank test results. 189Table 6.4. CMP measured VMC for sites without validation data. 190Table 7.1. Catchment outflow summary data for all modelled scenarios (1501-

15469 hours)

195

Table 7.2. High flow statistics for event 27th February 1999 to 3rd March 1999. 199Table 7.3. Low and intermediate discharge statistics from 17th May 1999 to

3rdJune 1999.

200

Table 7.4 Kruskal-Wallis results for catchment discharge. 206Table 7.5. Kruskal-Wallis results for field plot runoff. 222Table 7.6. Kruskal-Wallis results for field plot subsurface flow. 228Table 7.7. Modelled and measured VMC summary statistics for the field plot. 232

Page 16: Ground Penetrating Radar for the parameterisation of

Andrew Howe

16

Chapter 1: Overview

1.1 Introduction

Soils form an interface between the atmosphere and lithosphere. Soil is notable

for high spatial variability over scales which range from millimetres to

kilometres, a reflection of the variable processes involved in soil formation. Soil

properties have been recognised as one factor in controlling the degree of

surface and subsurface flow which occurs at a point (Mulligan & Thornes, In

Press), although catchment morphometry and the duration and intensity of

precipitation are also important influences.

This thesis is concerned with the measurement of those subsurface soil

properties that have often been a highly simplified component of the analysis of

hydrological response to precipitation. Historically this has been due both to the

lack of detailed subsurface data required for model parameterisation and

validation of results and the complexity of simulating water movement across

three dimensions. Complex models of hydrological response have been

developed, but the lack of high-resolution subsurface data remains a very

current problem. The lack of data concerned with the subsurface extends into

both the temporal and spatial domains. The problem is accentuated by the

complexity of modelling non-linear processes at a range of scales in an

environment characterised by heterogeneity and anisotropy. Quantifying the

response of a hillslope to precipitation events and post-event readjustment is

essential if the physical processes acting at the hillslope scale are to be

modelled. The importance of identifying the contributions of different

hydrological pathways to water status within a catchment has long been

recognised, but an achievable means to sample and validate subsurface

hydrological parameters over a wide area at a grid scale has been lacking.

Ground penetrating radar (GPR) potentially provides a solution as a surface

based non-intrusive tool for sampling subsurface properties.

Page 17: Ground Penetrating Radar for the parameterisation of

Andrew Howe

17

1.2 Current issues in subsurface hydrology

The project aims to improve the understanding of subsurface water dynamics,

using GPR to rapidly quantify soil thickness and soil moisture variability across

a hillslope and ultimately an entire catchment. A better understanding of

hydrological response is a necessary requirement for improved flood

forecasting, pollutant movement, soil erosion estimation and water management

purposes. Globally, water is an intensively managed and sensitive resource,

essential to industry, agriculture, domestic users, and to the natural

environment. In the UK the 1990’s have seen an increased interest in water

quality issues, water supply, flood defence and the environmental impact of

over-abstraction from rivers and groundwater. Many of these issues require an

improved understanding of the patterns of water movement and storage in

catchment systems.

Water flow through a river basin is a non-linear process following a complex

three-dimensional geometry (Beven, 1994). In modelling scenarios some

simplification of this geometry is necessary so that in the case of the Systeme

Hydrologique Europeen (SHE) model (Binley & Beven, 1993), subsurface flow

paths are assumed to follow surface elevations and tend to be represented by

one-dimensional vertical unsaturated and two-dimensional saturated flow

components. Alternatively the Institute of Hydrology’s distributed model

(IHDM) uses a coupled two-dimensional saturated-unsaturated component for

simplicity (Beven, 1985). TOPMODEL (Beven et al., 1984) uses a topographic

index to reflect the tendency of water to accumulate at points within the

catchment. Subsurface flow is represented by different water table positions and

the downslope flow rate is exponentially related to local water table position

(Chappell et al., 1993). Without detailed subsurface data hydrological modellers

necessarily make assumptions regarding subsurface storage and hydrological

fluxes. Two common assumptions are that subsurface flow pathways are

Page 18: Ground Penetrating Radar for the parameterisation of

Andrew Howe

18

controlled by surface topography and flow rates are proportional to the

hydraulic gradient, derived from local slope (Quinn et al., 1993a).

Hydrological models can be described as either lumped or distributed depending

on the area over which parameters are determined. Lumped models use spatially

averaged parameters usually at the catchment scale in which vegetation type,

soils, topography and precipitation are considered constant over the entire area.

Distributed type models account for catchment heterogeneity by subdividing a

catchment into regular (grid based) or irregular zones (triangulated irregular

networks), each of which possesses a set of parameters reflecting the values

within each cell. Simulation and validation of distributed models requires that

the parameter set for each cell is known. Often these data are unavailable,

particularly for subsurface attributes. Soil thickness exerts a strong control over

moisture storage and soil response to precipitation yet distributed models often

apply a single ‘mean’ value for an entire catchment (Boer et al., 1996). The lack

of data for other fundamental hydrological parameters e.g. hydraulic

conductivity, porosity, depth, texture and bulk density, is largely due to the

prohibitive task of collecting enough samples to adequately characterise

catchment soil characteristics. Sample collection and analysis is a time and

labour intensive process.

The requirement for more data from the vadose zone stems from hydrological

model parameterisation and validation requirements, as well as the need for an

improved conceptual understanding of subsurface water fluxes. The large

number of parameters contained within distributed models can produce the

same output from a number of different model scenarios (Beven, 1993b). Model

evaluation becomes complex in these situations since validation must take place

at the subcatchment scale. If data sets exist for comparison between modelled

and actual results they are often of poor spatial resolution and typically point

measures which require interpolation to produce estimates of properties for an

area or volume of space (Blöschl & Sivapalan, 1995). Blöschl & Sivapalan

identify spatial data as a key requirement that is presently lacking from

Page 19: Ground Penetrating Radar for the parameterisation of

Andrew Howe

19

distributed model validation. They suggest that monitoring the spatial patterns

of state variables could be used to evaluate model performance, the most

hydrologically useful pattern being that of soil moisture distribution to depths of

several decimetres. This requires an accurate, non-invasive technique capable of

quick data collection before appreciable changes in variable state occur. GPR is

a possible tool; a surface based remote sensing technique that offers a rapid and

non-destructive method for subsurface data collection, dependent upon a low

level of soil electrical conductivity. Soil and rock conductivity of less than

1mS/m allows penetration in ideal environments to maximum depths of up to

50m with low frequency systems (Davis & Annan, 1989). In pedological and

hydrological applications the depth of interest is often less than a tenth of this

amount, potentially allowing use of higher frequency systems with the benefits

of improved spatial resolution. Shih & Doolitle (1984), Truman et al., (1988),

and Collins et al., (1989) have successfully measured soil thickness using GPR.

A number of papers have recently appeared in the literature concerning the use

of GPR as a tool for soil water content estimation including those by Chanzy et

al., (1996), Greaves et al., (1996), van Overmeeren et al., (1997) and Gilson et

al., (1996). None however are working within a hydrological modelling

framework where detailed moisture data is required for model validation and

calibration.

A detailed overview of GPR techniques relevant to hydrology is given in

Chapter 2. The following section identifies a selection of subsurface

hydrological parameters of interest and reviews some current data collection

methods.

Soil structure is of primary importance to hydrological modelling since structure

exerts strong control over the water retention and transmission characteristics of

a soil. The transmission of water through a profile is controlled by a soil’s

hydraulic conductivity, largely determined by total porosity, pore size

distribution and pore connectivity. It is a non-linear function of volumetric

water content (Rawis et al., 1992) and determines whether infiltration-excess

Page 20: Ground Penetrating Radar for the parameterisation of

Andrew Howe

20

flow is generated (Anderson & Burt, 1990). Soil water retention is controlled by

soil texture. The maximum amount of water held in a soil depends on the total

porosity of the soil profile. Alternatively for two identical soils of equally

decreasing porosity with depth and equal initial soil moisture it is the thickness

of soil that controls the maximum amount of water storage. A thinner soil will

saturate first during storm events and becomes a source area for saturation

excess overland flow. Soil thickness and the spatial variation of soil thickness

across the drainage basin can therefore be seen as an important parameter in

hydrological modelling.

The water balance of the upper soil layer is the most dynamic since it forms the

boundary between atmosphere and subsurface. The soil moisture of the upper

layers is critical in determining rates of infiltration and plays a crucial role in the

partitioning of precipitation into either runoff or subsurface flow (Orlandini et

al., 1996). An understanding of subsurface water dynamics is required for

catchment response to be modelled (O’Loughlin, 1990). Anderson and Burt

(1990) note that,

‘soil moisture is a key variable in the hydrological cycle and flow overand within the soil has a strong and direct influence on the timing andmagnitude of the basin hydrograph’.

As yet no hydrological model uses fully distributed soil thickness and moisture

values within its structure. This has been recognised, but the problem remains of

finding effective techniques for measuring these parameters in the field. Soil

moisture is particularly variable owing to the range of processes acting to

redistribute over scales ranging from centimetres to kilometres. Western et al.,

(1999) identify the spatial variability in precipitation and evapotranspiration

along with topography, soils, geology and vegetation as important controls on

moisture patterns.

Measurement of subsurface properties is problematic because accurate

measurement of soil characteristics such as depth, texture and porosity are

Page 21: Ground Penetrating Radar for the parameterisation of

Andrew Howe

21

currently only possible through the excavation of trial pits or auguring. Soil

moisture can be measured through direct or indirect methods (Rawis et al.,

1992). Direct measurement of water content is achieved using gravimetric

methods. Indirect measurement requires that a measurable soil property can be

related to moisture content. This approach includes electrical conductance and

resistivity methods, time-domain reflectometry (TDR), radiological methods

(neutron thermalization or gamma attenuation), and remote sensing. The

indirect methods of moisture measurement explored by Rawis et al., (1992) are

termed indirect because soil water is a function of a surrogate variable such as

soil resistivity. Many of these methods of soil moisture derivation require

excavation for access to different profile depths, resulting in the alteration of

subsurface properties. The exception are remote sensing methods which

includes satellite, aerial and surface-based techniques (e.g. GPR).

Soil moisture measurement using remote sensing offers the potential to

repeatedly sample large areas. In the last five years the measurement of soil

moisture using satellite and airborne systems has concentrated on active and

passive microwave methods, although the reflected visible and thermal infrared

wavelengths are areas of on-going research (Pietroniro & Leconte, 2000).

Active microwave methods provide spatial resolutions of tens of metres,

compared with tens of kilometres for passive methods, but the extraction of soil

moisture from radar images is complicated by the equally important effects of

surface roughness and vegetation properties on radar backscatter (Verhoest et

al., 1998).

Due to limitations in sensor spatial resolution, the extraction of soil moisture

data for hillslope scale hydrology would seem to preclude the use of passive

sensors, while active instruments require improved parameterisation of

vegetation and surface roughness effects (Engman & Chauhan, 1995). The issue

of scale is a key factor in the ability of remote sensing to provide soil moisture

data of use to the hydrologist. Van Oevelen (1998) concludes that averaged

estimates of moisture content from remote sensing can be very different from

Page 22: Ground Penetrating Radar for the parameterisation of

Andrew Howe

22

field measurements made at the same location due to the presence of

geomorphological features undetected by the sensor. Although active

microwave methods have been applied to extract near-surface soil moisture

data, the sampled depth is typically only 0-5cm (Pietroniro & Leconte, 2000).

Point sampling of soil properties is destructive and samples only a small volume

of any catchment. An alternative methodology uses landscape position and

digital elevation models (DEM’s) to derive terrain attributes (Brubaker et al.,

1994; Moore et al., 1993b respectively). Boer et al (1996) use this method to

map soil thickness classes over three distinct geological types in south-east

Spain, generating accurate results over 40% - 78% of the area. This is only a

viable methodology if a strong identifiable correlation exists between slope

position and soil attributes, which is not always the case. While this method

may be useful for large-scale generalisations of soil classes it is of less use to

the detailed study of hydrological processes. This approach still requires

widespread validation, a problem that cannot easily be overcome without either

an expensive and manually intensive program of data collection or a faster non-

intrusive survey technique capable of collecting the parameters required.

The requirements for, and problems associated with, the current methods of

subsurface data collection have been outlined. Geophysical techniques are one

method of obtaining information about the subterranean environment. These

techniques have the advantage of being non-invasive, but are surface-based

rather than airborne/space platform-based. As a result the power available for

signal transmission into the subsurface (and detection of a reflected signal) is

substantially increased, allowing greater depths of investigation than alternative

remote sensing techniques. The volume sampled by surface based geophysical

instruments falls in intermediate scale between the point measurements typical

of a hydrological investigation and the large areas sampled by aerial platforms.

Advances in computer capabilities and technology have led to the development

of ground penetrating radar systems capable of high-resolution data collection

obtained by interpreting the velocity and attenuation changes in an

Page 23: Ground Penetrating Radar for the parameterisation of

Andrew Howe

23

electromagnetic pulse passing through a material. Inversion of velocity and

attenuation changes can be used to produce maps of electrical conductivity and

dielectric permittivity. These values can then be transformed given that valid

relationships exist between hydrogeology and dielectric properties (Knoll &

Knight, 1994).

Page 24: Ground Penetrating Radar for the parameterisation of

Andrew Howe

24

1.3 Research Objectives

The aims of this study are to use distributed soil thickness measurements from

GPR as a key parameter to the subsurface component of a distributed

hydrological model. This model will then be validated using GPR derived soil

moisture measurements collected from a sample of field sites. The initial stage

is then to evaluate the capability of GPR to measure these two parameters using

conventional intrusive measurement techniques for validation.

A simple hydrological model has subsequently been designed which uses

spatially distributed soil thickness as one input and grid cell soil moisture as an

output, along with total catchment discharge. Model validation will therefore be

carried out at two scales; the grid scale and the catchment scale. Validation will

be achieved at the catchment scale using a lumped approach, comparing

predicted model discharge with actual recorded discharge over the same time

period. Unlike the current group of physically-based distributed hydrological

models the aim is also to perform an internal state validation at the grid scale by

comparing model predicted soil moisture distribution with field moisture

patterns measured using GPR, conventional moisture measurements and a field

runoff – throughflow plot. In this way it may be possible to confirm or refute

the hypothesis that a model producing a realistic hydrograph response at the

catchment scale may not adequately simulate processes at the scale of the

individual grid cell.

Page 25: Ground Penetrating Radar for the parameterisation of

Andrew Howe

25

Specific objectives of this study are:

I. To quantify the variation in soil thickness across the study catchment at

the grid scale using data from GPR field measurements.

II. To use GPR, gravimetric and theta probe (Delta T Instruments,

Cambridge) field measurements to quantify spatial changes in soil

moisture at the grid scale.

III. To develop a simple distributed GIS based hydrological model which

uses soil thickness data obtained from GPR field surveys as an input

parameter to the model.

IV. To validate the predicted spatial pattern of soil moisture generated as

model output using GPR, theta probe (Delta T Instruments, Cambridge)

and field plot measurements of moisture content for sample sites located

across the catchment.

V. To compare measured catchment discharge and field plot measurements

with predicted discharge (catchment scale) and field plot runoff and

subsurface flow (grid scale) for different scenarios of cell soil thickness

over the catchment. Specifically to assess whether variable soil thickness

derived using GPR provides improved model results over model

scenarios in which soil thickness is a lumped parameter.

Page 26: Ground Penetrating Radar for the parameterisation of

Andrew Howe

26

Chapter 2: Ground Penetrating Radar

2.1 Introduction

Ground Penetrating Radar (GPR) is a technique of subsurface sampling using

electromagnetic (EM) radiation to construct an image of soil and geological

spatial variation. GPR transmits high frequency pulses (10-1000MHz) into the

ground and records the subsequent reflections. Reflected energy is caused by an

abrupt change in the dielectric properties of a material, and within soils this may

be associated with a change in texture, moisture or conductivity. Reviews of

GPR operation theory and methodology are presented in Annan & Davis

(1977), Daniels et al., (1988), and Davis & Annan (1989).

Subsurface mapping through the reflection of EM waves from dielectric

boundaries was first used to locate ore deposits and water table depth by

Leimbach and Lowy in 1910 (Daniels et al., 1988). Advances in electrical

engineering, GPR system design and an increased understanding of material

properties have improved the range, resolution and performance of radar

systems to the present.

GPR uses the principle of detecting reflections from layers or objects below the

ground to create a subsurface image. The first return is a direct airwave

followed by a ground wave and thereafter returns from subsurface objects

(figure 2.1). The time taken for an emitted signal to travel to and return from the

target provides an estimation of target depth. The magnitude of the reflected

signal gives an indication of target characteristics and the velocity of wave

propagation is dependent on the characteristics of the medium through which

the electromagnetic wave has passed.

Page 27: Ground Penetrating Radar for the parameterisation of

Andrew Howe

27

Figure 2.1. Theoretical EM pathways between transmitter and receiver antennae.

The potential advantage of GPR over conventional mapping of the subsurface is

its non-invasive nature; there is no requirement for borehole drilling or soil

augering. These invasive techniques are costly, time consuming and

fundamentally alter the structure of the sampled and adjacent area. GPR surveys

are non-destructive and in theory faster, more economical and provide a better

spatial resolution than conventional methods (Doolittle & Collins, 1995).

Page 28: Ground Penetrating Radar for the parameterisation of

Andrew Howe

28

2.2 GPR Theory

The basis of GPR operation is the measurement of reflected electromagnetic

waves from material boundaries. The propagation of electromagnetic waves

through a medium depends on the material properties at a specific frequency.

Wave propagation is dependent on the conductivity, dielectric constant and

magnetic permeability of a material as a response to an applied electric field.

When an electric field is applied, for example an electromagnetic pulse radiated

into a medium as a sine wave function, the electron cloud around each atom

within becomes similarly orientated by the field, forming a dipole. This is

polarization. Non conducting materials which are polarized by applying an

electric field are called dielectrics (Alonso & Finn, 1992). Polarization is a

vector quantity (P) proportional to the strength of the applied electric field

where

(2.1)

P = ε0 χe δ

χe is a measure of atomic response to an electric field, termed material

susceptibility and is dependent on the molecular properties of the material. εO is

the vacuum permittivity (8.85 × 10-12 farads/metre). In fact it is more common

to talk about material properties in terms of their relative permittivity (εR) which

is dimensionless, where

(2.2)

εR = ε / ε0 = 1 + χe

Relative permittivity is also known as dielectric constant (K). This is the

terminology that will be used throughout the remainder of this thesis. Dielectric

constants for selected earth materials at radar frequencies of 50 - 1000 MHz are

Page 29: Ground Penetrating Radar for the parameterisation of

Andrew Howe

29

shown in table 2.1.

When an electromagnetic wave propagates through a medium the electric field

generated causes electric charges to move. Two categories of movement exist,

these being conduction and displacement currents. Conduction currents arise

when an electric field causes free charges i.e. electrons, to move within the

material. Moving charges collide with stationary molecules, dissipating energy

as heat into the medium. The electrical conductivity of most minerals is very

low although there are exceptions to this rule e.g. magnetite, carbon, and pyrite.

In the majority of soils however conductivity is controlled by porosity, moisture

content, the concentration of electrolytes, water temperature and the amount and

composition of colloids (McNeill, 1980), rather than mineral type. Materials in

which conduction currents dominate are poor environments for GPR surveys

since electromagnetic fields will disperse (Annan, 1996). Soils with high

conductivity include those with high clay contents and soils with high pore

water conductivity due to high levels of total dissolved solids (TDS). The actual

degree of dissipation depends on the conductivity of the sample.

Displacement currents (polarization) occur when a charge may only move a

constrained distance. Four polarization mechanisms exist (Powers, 1997). These

are electronic polarization, which occurs when circular electron clouds become

elliptical, molecular polarization where charged molecules distort, orientational

polarization, where dipole molecules rotate in the presence of a electronic field,

and interfacial polarization which involves the accumulation of ions at material

interfaces. Applying an electric field causes charges to move to a new stable

equilibrium storing energy in the process. Once the electric field is removed the

charges return to their original equilibrium configuration releasing energy

(Annan, 1997). Water molecules are dipolar and respond by aligning to an

electric field, producing a high dielectric constant at radar frequencies.

Page 30: Ground Penetrating Radar for the parameterisation of

Andrew Howe

30

Figure 2.2. a) Random orientation of electric dipoles due to thermal motion, b)Polarization in an electric field, thermal motion active. (Alonso & Finn, 1992).

In a material of constant electrical conductivity (σ) the dominant current

mechanism is determined by signal frequency. As frequency increases

displacement frequencies become the dominant processes over conduction

(Annan, 1997). The point at which the two are equal is termed the transition

frequency (ft) and is calculated by

(2.3)

ft = σ / 2π × K × 8.85×10-12

where K is the material dielectric constant.

The ratio of conduction (Jc) to displacement (Jd) currents at a frequency (f) is

given by the loss tangent

(2.4)

tan δ = Jc/Jd = σ / (2πf × K × 8.85×10-12)

The dielectric constant (K) of any material is a complex number of the form

K = KR + j Ki (2.5)

a) b)

Page 31: Ground Penetrating Radar for the parameterisation of

Andrew Howe

31

The real part (KR) describes the propagation of the wave in a medium, the

imaginary component (KI) describes the attenuation losses due to conduction

processes. For non-zero conductivity, which always occurs in field situations,

both components are frequency dependent (Schmugge 1980, Power 1997).

Between 10 - 1000 MHz however a window exists for optimal GPR

performance where the velocity and attenuation of an electromagnetic wave are

relatively independent of frequency (Davis & Annan, 1989, Annan, 1996). At

frequencies below ~10 MHz losses are dispersive. Above approximately 1000

MHz the value of KI increases rapidly as the rotational relaxation of water is

approached at 10 GHz. This relaxation effect is of great importance since

almost all earth materials contain some water. Scattering losses also increase as

wavelengths approach similar magnitudes to particle sizes.

For low loss materials tan δ << 1 and attenuation (α) approximates to

α = 1.69×103 σ / ( KR)1/2 (2.6)

Similarly when tan δ << 1 the velocity of electromagnetic waves can be

approximated by equation 2.7, where c is the plane wave propagation velocity

of electromagnetic radiation in free space (3×108 m/s).

(2.7)

This equation allows conversion from wave velocity to dielectric constants.

In terms of depth (z) to a reflector

(2.8)zv t

=.2

21

R )K(

c=v

Page 32: Ground Penetrating Radar for the parameterisation of

Andrew Howe

32

where t is two way travel time.

Page 33: Ground Penetrating Radar for the parameterisation of

Andrew Howe

33

Table 2.1. Typical values of dielectric constant (K), velocity, attenuation andconductivity for earth materials in the GPR frequency range. (From: Davis &Annan, 1989).

Material K Velocity (m/ns) Attenuation α (dB/m) Conductivity σ

(mS/m)

Air 1 0.30 0 0

Distilled Water 80 0.033 0.002 0.01

Fresh Water 80 0.033 0.1 0.5

Sea Water 80 0.01 1000 300000

Granite 4-6 0.13 0.01-1 0.01-1

Ice 3-4 0.16 0.01 0.01

Limestone 4-8 0.12 0.4-1 0.5-2

Shales 5-15 0.09 1-100 1-100

Dry Salt 5-6 0.13 0.01-1 0.01-1

Silts 5-30 0.07 1-100 1-100

Clays 5-40 0.06 1-300 2-1000

Dry Sand 3-5 0.15 0.01 0.01

Saturated Sand 20-30 0.06 0.03-0.3 0.1-1

Reflection of an electromagnetic wave occurs at any interface between two

materials with different dielectric values. The reflection coefficient (R) is a

function of the difference in electrical properties between the two materials and

is calculated as

(2.9)

Reflection also depends on the thickness of the layer and on the frequency (and

hence wavelength) of the incident signal. The sign of R depends upon the values

of K1 and K2 with R being negative if the wave travels from a low K to a high K

and positive in the reverse case. This raises the possibility of material

RK KK K

=−

+1 2

1 2

Page 34: Ground Penetrating Radar for the parameterisation of

Andrew Howe

34

characterisation from examining the returned signal for coefficient sign. The

power reflection coefficient is R2.

Due to spherical spreading of the electromagnetic wave, signals are returned

from an area beneath the radar rather than a discrete point. The size of this area

or footprint depends on the operating wavelength (λ), depth to reflector (z) and

the dielectric constant of the material (K) through which the signal passes.

Increased survey depth leads to increased footprint size (figure 2.3) since the

wave spreads as an elliptic based cone from the source.

Figure 2.3. GPR footprint dimensions.

Transmitter

xy

z

Page 35: Ground Penetrating Radar for the parameterisation of

Andrew Howe

35

Increasing the dielectric constant reduces spread; a water-saturated soil has a

smaller footprint than does the same soil when dry (at equivalent depths).

Higher frequencies also reduce footprint dimensions. The equation for

determining footprint dimensions is given by Annan, (1997) as:

(2.10)

(2.11)

The volume of soil (V) sampled can subsequently be calculated using the

standard mathematical equation for the volume of an elliptical cone,

(2.12)

Antenna frequency and the dielectric properties of the ground determine the

achievable spatial resolution. Resolution in this case is the ability of a system to

distinguish between two signals separated by a small time interval (Davis &

Annan, 1989). For the frequencies used in standard GPR systems (10 – 1000

MHz) this is of the order of centimetres to metres. In all ground radar

applications a trade off exists between penetration depth and resolution. High

frequency, short wavelength antennae have improved resolution but suffer from

increased signal attenuation resulting in shallower investigation depths.

Scattering losses also increase as frequency approaches a similar magnitude to

particle size. For GPR systems the maximum theoretical resolution (z) is one

quarter pulse wavelength (Reynolds, 1997)

xz

K

yx

= +−

=

λ4 1

2

xyzV π31

=

Page 36: Ground Penetrating Radar for the parameterisation of

Andrew Howe

36

(2.12)

Where v equals medium velocity and f is the frequency of operation (Hz). Table

2.2 summarises the theoretical resolution and change in footprint dimensions at

three frequencies for a dry (K=4) and saturated sand (K=25).

Table 2. 2. Ideal resolution (from equation 2.12) and footprint dimensions (fromequation 2.10 & 2.11) at 1m depth for selected operating frequencies.

Frequency

(MHz)

Maximum

Resolution

(m) K=4

Maximum

Resolution

(m) K=25

Footprint (m)

Dry Sand (K=4)

x y

Footprint (m) Wet

Sand (K=25)

x y

225 0.17 0.07 0.91 0.45 0.53 0.27450 0.08 0.03 0.75 0.37 0.37 0.19900 0.04 0.02 0.66 0.33 0.28 0.14

For all frequencies in this range that the effect of increasing the water content of

the medium is one of reducing footprint dimensions. The volume returning the

transmitted signal is a dynamic property of the dielectric value of the material.

Increasing soil moisture results in a reduction in the volume of soil contributing

to the returning signal, in effect a ‘tightening’ of the radar beam. Increasing the

dielectric constant of a material also results in a theoretical increase in

resolution.

zvf

=4

Page 37: Ground Penetrating Radar for the parameterisation of

Andrew Howe

37

Figure 2.4. Maximum theoretical resolution for the frequency range 100 – 1200MHz for 2 dielectric values (plot derived using equation 2.12).

0.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35

0.40

100

200

300

400

500

600

700

800

900

1000

1100

1200

Frequency (MHz)

Max

imum

Res

olut

ion

(m) Res (m) K=4

Res (m) K=25

Page 38: Ground Penetrating Radar for the parameterisation of

Andrew Howe

38

2.3 GPR Survey Types

The three standard survey configurations are described. Most commonly used is

the ‘reflection survey’ in which the transmitter and receiver antennae are

located on the ground surface and maintained at a constant offset (figure 2.5).

Variation in dielectric value causes partial reflection of the transmitted pulse

which is subsequently detected and recorded. Moving the system across the

ground to the next survey point delineates the spatial variation in subsurface

features. Reflection surveys enable rapid collection of data and displayed as a

two-dimensional profile from which it is possible to pick out linear and point

reflectors. In reality as discussed previously the zone of response is a volume of

soil, not a single point and therefore targets off centre of the system will

contribute to the returned signal. It is this characteristic of GPR operation that

produces the hyperbolic pattern and allows point reflectors i.e. stones and pipe

cross-sections to be identified. The concept of what constitutes a point reflector

is somewhat arbitrary since it is dependent on the frequency at which the survey

is undertaken.

Figure 2.5. Reflection Survey - constant offset.

Tx Rx

1 2 3 4

GPR surveys describe the variation in soil characteristics with respect to the

time of travel between transmitter and receiver antennae. The depth at which

Page 39: Ground Penetrating Radar for the parameterisation of

Andrew Howe

39

reflectors are located is determined by measuring the velocity of the EM pulse

through the material. Two methods of velocity measurement are the ‘common

midpoint survey’ (CMP) and ‘transillumination survey’.

For a CMP survey (figure 2.6) the antennae are initially placed on the ground

surface separated by one stepsize. Each antenna is then moved outwards in

increments of one half the stepsize defined for the survey. Increasing separation

(x) between antennae results in longer travel times (t) and by comparing the

change in antenna separation with the time taken for the pulse to return from

reflectors an estimate of the velocity for each layer can be calculated. The

gradient of reflection events plotted as x2 against t2 provides a measure of wave

velocity. A major assumption of CMP’s is that linear reflecting horizons must

underlie the survey area. If reflectors are inclined significantly from the radar

horizontal plane the time for reflections is not only a function of antenna

spacing but also of the varying depth to the reflector.

Figure 2.6. Common midpoint survey (CMP) - variable offset.

Tx Rx

Transillumination survey schematics are shown in figure 2.7. In this case the

transmitter and receiver antennae are separated by a constant horizontal distance

and moved vertically parallel to one another. Two data collection methods are

possible. A zero offset gather (ZOG) ensures no vertical difference between

Page 40: Ground Penetrating Radar for the parameterisation of

Andrew Howe

40

each antenna. Alternatively a multiple offset gather (MOG) requires that one

antenna be fixed while the second moves, producing a more detailed picture

using multiple raypaths. Transillumination has been used for internal imaging of

solid structures e.g. concrete blocks, and near-surface deposits using boreholes

(cross-hole tomography) (Gilson et al., 1996. Eppstein & Dougherty, 1998).

Variations in the propagation velocity between antennae indicates changing

dielectric values which in may indicate related changes in hydrogeological

parameters.

Figure 2.7. Transillumination survey.

Tx

TxRx

Rx

ZOG MOG

Both CMP and tomographic surveys use travel time over a known distance to

provide velocity estimates of the electromagnetic wave. Using equation 2.7

these velocities can be converted to values of dielectric constant. Variations in

velocity and hence dielectric value imply variations in material properties.

Page 41: Ground Penetrating Radar for the parameterisation of

Andrew Howe

41

2.4 Relation between dielectric constant and hydrologicparameters

Using GPR to collect soil information requires calculation of dielectric values

which can be related to soil properties, a process known as inversion.

Laboratory investigations have suggested that the most important parameters

controlling the dielectric response of geologic materials are water content,

porosity, clay content and measurement frequency (Knoll & Knight, 1994).

Topp et al (1980) demonstrated that soil water content is the primary control

over soil dielectric value. This is due to the unique atomic structure of water

which results in its high dielectric constant of 80 compared to dielectrics of 3-20

for most minerals. Other factors that have been found to affect dielectric values

include soil salinity and temperature. Salinity has an effect since increasing

water content leads to an increase in the dissolved ions within the water. The

effects of salinity are low at frequencies around 1000 MHz and above, but

increase with decreasing frequency, rising sharply below 25 MHz (Wensink,

1993). Over the range of 1 - 1000 MHz Topp et al (1980) found that the

temperature of liquid water did not exert a significant control on recorded

results, but upon freezing the dielectric constant of water drops from

approximately 80 to 3.

Soil water content is itself a function of available pore space and hence bulk

density. Porosity controls the maximum amount of water held in a soil and is a

factor in determining soil drainage. Schmugge (1980) has shown that textural

composition is also an important consideration, since the velocity and therefore

frequency of the electromagnetic wave is largely controlled by free moisture

held within pores. Below a transition value, water molecules are adsorbed to

soil particles. Response to the electromagnetic wave is limited until molecules

are less strongly held. The amount of water held as adsorbed water is directly

related to particle surface area meaning that clay soils are able to hold a larger

amount of water in this state than sands. The dielectric response of soil is

Page 42: Ground Penetrating Radar for the parameterisation of

Andrew Howe

42

therefore not a straightforward measure of total water content, but could provide

a figure of available water content, of use for hydrological modelling and

agricultural purposes.

Investigation into the relation between soil dielectric value and soil properties

has predominantly been carried out using time domain reflectometry (TDR).

This technique involves the insertion of metal transmission lines into the target

environment. The propagation time of an electromagnetic pulse through a

transmission line of known length can be used to calculate velocity and related

to dielectric value using equation 2.7. Two decades of research has identified a

positive relationship between volumetric soil moisture and soil dielectric

constant (Topp et al., 1980; Topp & Davis, 1985; Roth et al., 1992; Whalley,

1993; Weiler et al., 1998).

Determination of soil moisture from dielectric constant has been attempted

using both empirical and theoretical approaches. Topp et al., (1980) derived a

third order polynomial (2.13) using regression analysis based on 18 experiments

using four mineral soils. 93% of the data was within ± 0.025 of measured

volumetric moisture content. Therefore by calculating the velocity of a wave

through a soil layer a corresponding value of K can be derived. This value can

then be substituted into equation 2.13 to provide an estimate of volumetric

moisture content (θ) where

(2.13)

θ = -0.053 + 0.0292K – 0.00055K2 + 0.0000043K3

A more physically based theoretical methodology has been developed using a

mixture modelling approach. Given that soil is a three phase mixture consisting

of mineral grains, air and water, each with very different dielectric values, the

volume fractions of these constituents are important in determining the effective

dielectric constant of a soil unit. Knoll (1996) reviews a selection including the

semi-empirical complex refractive index method (CRIM) and the effective

medium approach of the Bruggeman-Hanai-Sen (BHS) model. Both the CRIM

Page 43: Ground Penetrating Radar for the parameterisation of

Andrew Howe

43

and BHS are two or three phase mixing formulae and can therefore be used to

represent the volumes of air, water and soil components with their associated

dielectric values. The BHS model also includes a micro geometry power term to

account for grain shape (Greaves et al., 1996).

CRIM (three phase)

(2.14)

At high frequencies where tan δ << 1 the BHS (two phase) model results in the

implicit expression (equation 2.15), which can only be solved explicitly by

assuming that the sample is totally saturated (Greaves et al., 1996).

(2.15)

Where Ks, Kw, Kg, Ka are the dielectric constants of the sample, water, mineral

grains and air respectively, Sw is sample saturation (0-1) and φ is sample

porosity. m (equation 2.15) is a cementation index ranging from 1.5 for

spherically grained unconsolidated sands to 2.0 for consolidated sandstones

with oblate grains (Greaves et al., 1996).

Three phase versions of CRIM and BHS require that the two unknowns,

saturation and porosity of a sample are solved from a value of K. There is

however no unique mathematical solution to this problem. The two-phase

variants assume total sample saturation i.e. Sw equals unity, allowing an

estimation of porosity below the water table. Extrapolating this value of

porosity to unconsolidated heterogeneous sediments which form the soil mantle

is likely to produce erroneous estimates of water content. Therefore despite the

K = S K K S Ks w w g w a φ φ φ+ − + −( ) ( )1 1

K K

KKKK

g

w

g

s

s wm

m

=−

φ1

1

Page 44: Ground Penetrating Radar for the parameterisation of

Andrew Howe

44

disadvantages, the Topp equation remains the simplest method for estimating

soil water content from velocity changes derived from GPR data.

2.5 Contemporary soil and hydrological research using GPR

Shallow sub-surface investigations suffer from a number of complicating

factors. The near surface environment is the most complex in electromagnetic

terms and exhibits great heterogeneity over short distances. The shallow depths

common to soil investigation result in short travel times between reflections

causing problems in the resolution of different layers. At short time windows

the direct wave dominates the received image. The small separation transmitter-

receiver offset geometry commonly used for reflection surveys can cause air

and ground wave interference. Soils are an electrically lossy environment and

the presence of conductive clays limits the use of radar as a survey tool.

Despite these difficulties the potential application of GPR as an aid to soil

survey has been recognised. Its main advantages over conventional survey

techniques are an increase in sampling rate and the ability to produce

continuous real time profiles rather than point observations. High quality data

can be collected over a large area without the need for excavation. Excavation is

time consuming, labour intensive and infeasible over large areas. It destroys the

subsurface at the site and will result in changes in hydrological flow pathways

as a result of changing soil bulk density. To date field investigations have been

concerned with investigating depth to bedrock (Collins et al., 1989) and the

thickness of soil horizons (Shih & Doolittle, 1984). Truman et al., (1988)

successfully used GPR to image depth and lateral extent of argillic horizons and

depth to water table in a course textured coastal soil. Collins et al., (1989)

working in an upland area of New England found that 120 MHz radar profiles

gave a better estimation of soil overburden than auguring. Test pits dug for

validation found that 87% of radar derived depths were within 10cm of actual

depth compared to which only 7% of auger data was within 15cm. Collins

Page 45: Ground Penetrating Radar for the parameterisation of

Andrew Howe

45

proposed that the auger failed to penetrate fully due to the coarse texture and

varying size distribution of the underlying till deposits.

Characterisation of subsurface heterogeneity using GPR has been attempted by

Rea & Knight (1998) using radar profiles and geostatistical analysis of a

digitized photograph of a cliff face. The cliff consisted of two alternating

lithologies, coarse sands interlaced with silts and fine clay. Visual comparison

between the GPR profile and the photograph showed good agreement in

bedding structure. Geostatistical analysis of wave amplitude after correction for

spreading losses to create a semivariogram for radar data compared well with

the semivariogram produced for the photograph. They concluded that GPR was

able to image the spatial distribution of these lithologies.

From a hydrological viewpoint GPR has primarily been used as an experimental

tool for water table detection and subsurface contaminant mapping and

monitoring. The latter has concentrated on detection of dense non-aqueous

phase liquids (DNAPL’s) and light non-aqueous phase liquids (LNAPL’s).

DNAPL’s are solvents and degreasers, which form a major source of subsurface

contaminants in the industrial world. DNAPL’s are highly mobile being low

viscosity, high density liquids which are not affected by groundwater movement

(Annan et al., 1991b, Brewster et al., 1992). LNAPL’s include all hydrocarbons

and are a second widespread contaminant type. Being less dense than water

LNAPL’s accumulate at the water table surface and impermeable horizons,

releasing toxins into the water supply (Redman et al., 1994).

Many hydrological studies have been undertaken from a theoretical viewpoint.

One approach is to create a model of a potential target or interface response.

Often this is of primary importance prior to field application to see if radar is an

applicable detection method. Assessment of the viability of radar use depends

on the frequency used, target dimensions, depth to target and the dielectric

properties of both target and host material. The general model for assessing

radar suitability is described by the radar range equation (Annan & Davis,

Page 46: Ground Penetrating Radar for the parameterisation of

Andrew Howe

46

1977). Theoretical assessment of radar for water table detection is outlined in

Annan et al., (1991a) while theoretical detection of preferential flow pathways

within sands is presented by Kung & Lu (1993). Beres and Haeni (1991) use

GPR to investigate depth to the water table and delineate shallow drift deposits.

More recently, studies using GPR to quantify soil moisture in the vadose zone

have been undertaken.

Soil water content measurement using geophysical techniques depends on

finding a measurable parameter that accurately represents the amount of

moisture within a soil profile. Du and Rummel (1994) and Greaves et al.,

(1996) use the Topp equation to transform dielectric constants calculated from

velocity measurements into estimates of subsurface water content. Neither of

these studies validated their results with gravimetric or TDR measurements and

therefore only succeeded in producing profiles of relative soil water contents.

As discussed a common non-invasive method of deriving wave velocity is via a

common midpoint survey (CMP). Greaves et al., (1996) used multi-offset

CMP’s to construct an image of subsurface velocity variations that was then

converted to a profile of moisture distribution. Du and Rummel (1994) use wide

angle reflection and refraction to initially identify the ground wave and set the

optimum antenna separation. The ground wave is the signal that travels from

transmitter to receiver through the ground without reflecting. At small antennae

offsets the air wave and ground wave arrive within a time interval which can be

below the spatial resolution of the system which results in interference patterns

between wave fronts. This precludes the use of the direct wave amplitude as a

measure of moisture content since it is dominated by antenna-ground coupling

effects. Increasing the antenna separation allows separate discrimination of both

return signals since the speed of an electromagnetic wave in air is 0.3m/ns

compared to a typical soil velocity of 0.06-0.15m/ns depending on material

dielectric value.

Page 47: Ground Penetrating Radar for the parameterisation of

Andrew Howe

47

Van Overmeeren et al., (1997) use conventional CMP’s and an alternative

method of deriving wave velocities. In this case the two way travel time of a

wave to reflector is correlated with physical measurements of depth taken by

soil cores and piezometric detection of the water table surface. Van Overmeeren

looked at two sites in the Netherlands with sandy soils over a period of 13

months and compared soil water content in the vadose zone with capacitance

probe measurements taken at the same times in access tubes at each site. Lateral

and seasonal variations in moisture content included in the paper as graphs gave

similar results for both methods although no statistical comparison between data

sets is included.

A comparison between TDR and GPR measurements of soil moisture was

carried out by Weiler et al., (1998). 52 soil samples with corresponding TDR

dielectric values were used to derive a site-specific calibration for soil moisture

and TDR readings. Subsequently the relationship was found to be in good

agreement with equation 2.13, with a maximum difference of 0.03m3/m3 VMC

for a moisture content of 0.4m3/m3. Divergence between VMC is most likely

because Weiler et al. use a linear regression based equation whereas equation

2.13 is a third-order polynomial. GPR measured water content was found to

consistently under predict by on average 0.05m3/m3 compared with that

measured using TDR. Possible reasons for this include the difference in

frequency range between the two methods, zero time shift of GPR and,

mentioned briefly, the increased sample volume of GPR.

A mixture modelling approach to define intrinsic permeability and saturation

characteristics has been used by Hubbard et al., (1997) for estimation of these

parameters for two facies. Cross well tomography was used to build up a picture

of subsurface dielectric constant values. For each stratigraphy each dielectric

constant value produced two possible values of permeability and saturation.

Maximum likelihood technique was used to select the most probable value for

each location..

Page 48: Ground Penetrating Radar for the parameterisation of

Andrew Howe

48

An alternative method for deriving dielectric constant from GPR data is

described by Chanzy et al., (1996). They use ground wave amplitude and

reflection coefficient rather than velocity changes to calculate soil moisture in

ground and airborne modes. However while the results look promising the exact

effect of antenna - ground coupling, wave pattern interference, antenna

orientation and site characteristics on ground wave amplitude is poorly

understood. Data gathered in the air are also a function of look angle, surface

roughness and vegetation interaction. This theme requires considerably more

research and is beyond the scope of this project.

Table 2.3. Potential methods for the collection of soil thickness and soil moisturedata using GPR.

Parameter Method Authors ofprevious

work

Advantages Dis-advantages

Assumptions

Depth tobedrock

1.ReflectionSurvey

Shih &Doolittle,(1984),Truman et al.,(1988), Collins et al.,(1989)

Non intrusive,fast method ofdata collection.

How canreflections bepositivelyidentified asbedrock?

Low loss media,reliant on theaccuratederivation ofsubsurfacevelocities.

Soilmoisture

1. Velocityfrom Trans-illumination

2. Velocityfrom CMPsurvey

3.Amplitudeanalysisfromreflectionsurvey/CMP survey

Hubbard etal., (1998), Eppstein etal., (1998)

Du & Rummel(1994)Greaves et al.,(1996), VanOvermeerenet al., (1997), Weiler et al.,(1998).

Chanzy et al.,(1996)

Detailed,multiple raypathProven K – vrelationship.

Non intrusive,fast.CMP requiredfor reflectionsurvey anyway.Proven K – vrelationship.

Easy to extractamplitude datafrom reflectionsurveys.

Intrusive,requires 2adjacentboreholes.

Low loss media.

Near horizontalreflectors.

Uniquerelationship existsbetween K andamplitude.

Page 49: Ground Penetrating Radar for the parameterisation of

Andrew Howe

49

2.6 Conclusion

This chapter defines fundamental concepts of GPR, introduces methods of

deriving subsurface hydrological parameters from dielectric values, and

provides a review of current research relevant to this thesis, in particular

contemporary GPR methods for measurement of soil thickness and soil

moisture.

Analysis of the current literature shows that reflection profiling is the standard

technique used to derive the depth of soil overlying bedrock. Soil moisture

estimation is most feasible using CMP data, while transillumination is an

intrusive technique in the field scenario, requiring borehole construction. The

use of signal amplitude data for moisture estimation is at an early stage of

investigation and optimally researched in a controlled environment. Therefore

reflection surveys and CMP surveys will be used in this thesis to define soil

thickness and moisture respectively.

Page 50: Ground Penetrating Radar for the parameterisation of

Andrew Howe

50

Chapter 3: Hydrological Modelling and Model Development

3.1 Introduction

Hydrological models are required both as decision making tools for applied

water resource problems and as tools to improve our understanding of the

interaction between different components of the hydrological cycle.

Hydrological models can be broadly grouped into predictive or investigative

types (Blöschl & Sivapalan, 1995). The latter tend to have greater data

requirements and a more complex structure but provide some insight into

interactions between subsystems. Predictive models are generally concerned

with the answers to specific questions rather than specific system

characteristics. In reality some degree of overlap between these types exists in

the majority of hydrological models. Ultimately investigative models should be

able to make some predictions within appropriate error margins concerning

specific system response to input for a range of system states. Increasingly

hydrological models are required to predict both total outputs and the internal

spatial distribution of variables such as soil moisture, water quality, pollutant

mapping and erosion (Quinn et al., 1993; Chappell & Ternan, 1993.

The trend in hydrological modelling has been one of increasing complexity both

in terms of the equations used to describe hydrological processes and the

number of processes considered to be acting across a catchment. At present the

primary aims of catchment modelling can be summarised as; 1) Prediction of

streamflow response to storm events. 2) Determination of the nature of the

processes creating surface and subsurface flow in the catchment and the spatial

distribution of these processes (Quinn et al., 1995, Orlandini et al., 1996).

Advances in computer power and availability in recent decades has resulted in a

marked increase in both the number of hydrological models and the complexity

Page 51: Ground Penetrating Radar for the parameterisation of

Andrew Howe

51

of the numerical methods used to simulate hydrological processes. Simulation

of coupled saturated and unsaturated three-dimensional flow is now possible

using approximate numerical solutions by computer, which can solve for

millions of nodes over millions of timesteps. Numerous models are now

available which simulate water movement over and through the landscape at a

variety of spatio-temporal scales and at various levels of complexity.

Unfortunately, creating increasingly sophisticated mathematical models is no

guarantee of improved model performance. Many authors have suggested that

while the development of complex mathematical models has been beneficial

from a theoretical viewpoint, the collection of field data to validate these

models has been lacking (Anderson & Burt, 1990; Blöschl & Sivapalan, 1995).

Without validation data the usefulness of these models to applied hydrology is

limited.

Page 52: Ground Penetrating Radar for the parameterisation of

Andrew Howe

52

3.2 Hydrological Models

Current hydrological models can be classified on the basis of how the model

simulates hydrological processes. The relationship between precipitation and

the catchment outflow hydrograph can be described either as a simple empirical

function or as a series of physically based equations that aim to simulate the

actual processes taking place within the catchment using physically meaningful

parameters. Between these two extremes are a wide range of models which

incorporate interaction between different hydrological processes, but use both

empirical and physically based relationships to model these processes. A

conceptual model is commonly displayed as a flow diagram showing the basic

theories of hydrological interaction, and frequently form the underlying basis of

mathematical models (Blöschl & Sivapalan, 1995).

3.2.1 Empirical Models

Empirical models take the form of input, function, output, and are typically

based upon identification of statistical relationships between recorded inputs

and outputs. Regression analysis and extreme frequency analysis are commonly

used to derive the form of the transfer function. Empirical models do not

explain why events occur and therefore do not increase our understanding of

system behaviour. Extrapolating beyond the range of observed data is error

prone, particularly for extreme events (Anderson & Burt, 1990). Since the

influence of different hydrological processes is highly spatially dependent,

using this model type for ungauged catchments can produce highly inaccurate

results. Nevertheless for catchments with an observed hydrograph this is a

popular method being mathematically simple and requiring few parameters.

Page 53: Ground Penetrating Radar for the parameterisation of

Andrew Howe

53

3.2.2 Physically Based Models

Physically based models represent the process of water movement from

precipitation to channel outflow via a series of stores and fluxes described by

elemental physical equations. Typically these would include mathematical

equations to simulate rainfall interception by vegetation, evaporation, surface

runoff, soil moisture storage and groundwater storage. Examples include the

Institute of Hydrology lumped model (Blackie & Eeles, 1985), TOPMODEL

(Beven et al., 1984) and SHE (Abbott et al., 1986). Each component of the

system is described by equations using meaningful physical parameters derived

from field data.

3.2.3 Lumped vs. Distributed Models

Model classification is also possible by considering how hydrological

parameters are represented across the study area. Catchments are characterised

by their extreme heterogeneity (e.g. soil properties) and great variability in

fluxes (e.g. runoff) and state variables (e.g. soil moisture). ‘Lumped’ models

consider a catchment to be spatially homogenous with respect to the inputs and

parameters used (Wood, 1995). Heterogeneity and variability within the

catchment are described using a single ‘effective’ parameter set. The parameters

are ‘effective’ rather than ‘real’ since they are not directly measurable in the

field, instead being derived from model calibration with actual data. Lumped

models provide insight into overall catchment response to events but no insight

into processes internal to the system.

Distributed models attempt to better represent the variability within a system by

dividing a catchment into sub-units. A catchment can therefore consist of sub-

catchments, hillslopes, hydrological response units or regular grid elements.

Parameter sets are, in theory, derived for each sub-unit and therefore units are

considered to be homogenous and sub unit variability is neglected. Aggregation

Page 54: Ground Penetrating Radar for the parameterisation of

Andrew Howe

54

of each unit response over time using flow routing provides a spatio-temporal

model of catchment behaviour. Distributed models therefore require a large

number of parameters to describe the hydrological characteristics of each sub-

unit. In reality many supposedly distributed models lump subsurface parameters

at the catchment scale resulting in uncertain subsurface flow predictions

(Chappell & Ternan, 1993). An example is the treatment of saturated hydraulic

conductivity (Ksat) within a catchment. Ksat is almost exclusively assigned a

single parameter value per catchment or per soil type, a somewhat unrealistic

scenario since specific soil horizons can show variations in Ksat of several

orders of magnitude. Ksat between soil types in a heterogeneous catchment can

vary to an even greater extent (Chappell & Ternan, 1993).

The requirement for distributed models stems from a need for improved

predictions of erosion, chemical and nutrient transport, sedimentation and

understanding of changing land-use strategies within catchments (Abbott et al.,

1986; Quinn et al., 1993). A further advantage of distributed models is their

potential for validation at the grid cell scale as opposed to the validation

methods for lumped models, which consist of comparing the modelled

hydrograph with the observed hydrograph. Validation at the grid cell scale, also

known as internal state validation, is perceived as the next major goal of

hydrological modelling (Quinn et al., 1995).

Beven (1989) has questioned the extent to which the current set of physically

based distributed models accurately represent reality. Distributed models have

been criticised because of their large data requirements, high computational cost

and the fact that they are often as highly parameterised as lumped models

(Kalma et al., 1995). These problems are a result of trying to adequately

represent inter-cell variation and sub-grid variability in heterogeneous

catchments (Blöschl & Sivapalan, 1995). A comparison between physically

based and conceptual type models has shown that the results achieved are

comparable (Refsgaard & Knudsen, 1996). Distributed type models typically

consist of a large number of parameters, often too many to collect truly

Page 55: Ground Penetrating Radar for the parameterisation of

Andrew Howe

55

distributed measurements for. As a result some parameters are necessarily

treated as lumped values. This is often the case for subsurface values such as

hydraulic conductivity (Binley & Beven, 1991). The large number of

parameters involved makes subsequent model evaluation difficult. Beven &

Binley (1992) and Beven (1993b) question the assumption that an optimum

parameter set exists for each modelled catchment. Instead the concept of

equifinality has been suggested, the idea that different model structures or

parameter sets can produce the same model predictions. If this is the case, it is

important that validation occurs at a grid rather than catchment scale so the

internal dynamics of throughflow, runoff and soil moisture distribution can be

identified and validated against field data. It is increasingly important that as

noted by Klemes, (1986), models must produce not only the right results, but

also the right results for the right reasons.

3.2.4 Hydrological modelling and GIS

The use of GIS for hydrological modelling has become more commonplace

over the past decade. GIS is defined as a computing system for management and

analysis of spatial data (Drayton et al., 1993) and is therefore well placed to aid

in distributed modelling due to the capacity for storage and analysis of large

quantities of spatially distributed information. GIS also provides a graphical

interface with model results generally displayed as a series of maps showing the

spatial variation of selected model outputs. Interaction between GIS and

hydrological models can be on a number of levels ranging from simple

hydrologic assessment to fully integrated distributed models (Maidment, 1993).

Increasingly GIS and hydrological models are approaching this latter phase and

becoming embedded such that dynamic modelling of a river basin is possible

within the GIS. This level of integration removes the need for separate GIS and

modelling programmes with the associated problems of data transfer between

packages. In dynamic packages the results from the previous time step can be

used as inputs at the next time step. This thesis uses an integrated GIS -

Page 56: Ground Penetrating Radar for the parameterisation of

Andrew Howe

56

hydrological software package called PCRaster for development of a distributed

hydrological model.

3.2.5 Why another hydrological model?

The requirements for another hydrological model are summarised below:

1. The primary aims of this thesis are concerned with using GPR to provide

spatially distributed soil thickness data as a model input and GPR data as a

validation tool for the spatial pattern of soil moisture for selected grid cells.

To the authors knowledge no hydrological models exist which use

distributed soil thickness measurements as a key input, so a new, simple

model structure is needed to incorporate this extra data. All model

components will use standard physically based equations to simulate the

hydrological processes occurring at the catchment scale.

2. Truly integrated GIS – hydrological models are still in the very early stages

of development and therefore using an existing hydrological model would

require rewriting of model code within the GIS software. This process may

be as time consuming as developing a new model code using existing

physically based relationships between parameters.

3. The advantage of running an integrated GIS model is the spatially

distributed nature of the input and output data. Existing models that were

not developed in this system do not take full advantage of this benefit.

A widely held view is that the current set of spatially distributed models has

delivered disappointing results when compared with far simpler lumped

conceptual models (De Roo, 1998a). Given that the uncertainty in estimating the

value of the numerous input variables is one possible reason for this failure, this

Page 57: Ground Penetrating Radar for the parameterisation of

Andrew Howe

57

model will concentrate on modelling the dominant hydrological processes using

the minimum number of parameters possible.

Page 58: Ground Penetrating Radar for the parameterisation of

Andrew Howe

58

3.3 PCRaster

PCRaster is a GIS with the capability for dynamic modelling of spatial

processes. This is possible through use of a high-level spatial modelling

language designed specifically for environmental applications (Van Deursen,

1995; Wesseling et al., 1996; De Roo et al., 1998b). The PCRaster modelling

language consists of some 50 mathematical functions, both general algebraic

and logical operators, and functions specific to hydrological and environmental

modelling. Standard GIS operators are also included for the spatial analysis of

data, although the main emphasis of the software is on the dynamic modelling

of environmental processes. As yet PCRaster is unable to solve directly ordinary

or partial differential equations, nor implement vector field operations required

to simulate diffusion or advection (Wesseling et al., 1996).

PCRaster is a grid based GIS. Both point operations and global operations are

included in the package using mathematical operators. Flow routing within

PCRaster is controlled using an intrinsic operator which creates a map of the

catchment drainage network based on a DEM. The flow routing algorithm is a

deterministic 8-node (D8) type, which routes all cell outflow to one of eight

possible adjacent cells, depending on which cell has the lowest elevation. Each

cell is assigned one of eight possible flow directions or flagged as a sink area

when all surrounding cells have higher elevations. The resulting map is known

as a local drainage direction (ldd) map. The D8 algorithm tends to produce

parallel flow lines and is unable to account for flow dispersion (Moore, 1996).

Quinn et al. (1993a) found the inability to model multiple flow directions from a

single grid cell increases errors at coarse grid resolutions, but for grid cells

smaller than 50m this effect is lessened. A grid size of less than 50m was also

found to produce improved results in an analysis of the sensitivity of

TOPMODEL to time and space resolution (Bruneau et al., 1994). A reduction

in errors is to be expected with increasing grid resolution since the averaged

Page 59: Ground Penetrating Radar for the parameterisation of

Andrew Howe

59

area within grid cells is less resulting in effective parameter values derived from

smaller areas in which parameter variation may also be smaller.

3.4 Building an integrated GIS – Hydrological Model

The core modules used here are precipitation, evapo-transpiration, infiltration,

surface runoff, lateral subsurface flow and soil moisture storage. The

development of each module will be covered in detail in subsequent sections of

this chapter. This model is not designed to include all possible processes acting

in the catchment, nor does it use the most complicated and mathematically

intensive equations possible. The number of parameters used is kept to a

minimum and, where possible, these parameters are physically meaningful and

derivable from field data. The emphasis is on designing a simple hydrological

model with a distributed subsurface input parameter (soil thickness) which can

be used to determine the spatial variation in soil moisture for each time-step.

The model is designed to simulate the hydrological fluxes of a catchment using

regular grid cells. Cell length can be changed to any value bearing in mind the

findings of Bruneau et al., (1994). For this thesis a cell length of 25m was

chosen as the principle spacing, a similar size to many studies of hydrological

model performance. Varying the cell length is likely to have a small impact on

model results but is beyond the scope of this study.

Solutions to each equation are calculated on a cell by cell basis for each

timestep. The order of equation solution is important owing to the

interdependence of solution variables between sub-models. Solutions for all

sub-models are obtained using an iteration technique whereby the final solution

set for each cell at the current time are used as the initial state values at the next

timestep.

Page 60: Ground Penetrating Radar for the parameterisation of

Andrew Howe

60

GPR surveys of selected zones across the catchment have been used to measure

soil thickness. Survey sites were identified which covered the range of

hydrological areas present in the catchment on the basis of topographic indices.

A detailed explanation of the methodology used for GPR site selection can be

found in Chapter Five. Within this model soil thickness is an important

hydrological parameter, since depth multiplied by total profile porosity

determines the total volume of water that can be stored within each cell.

Shallow soils have a lower storage capacity and are therefore possible zones of

runoff generation (Burt & Butcher, 1985). Soil thickness is therefore an

important factor, along with precipitation input and soil conductivity, in the

control of soil moisture status. The available moisture content of the upper soil

controls the actual evaporation rate. Runoff occurs when direct precipitation and

flow from upstream exceed the local infiltration rate (infiltration excess type) or

is generated by return flow from the subsurface. Infiltration is partially a

function of cell moisture content. Lateral saturated subsurface flow between

cells is controlled by the differences in hydraulic gradient between them.

The following sections describe the component processes modelled. Where

appropriate examples of PCRaster modelling code are included as examples of

the method used to obtain solutions. PCRaster commands are printed in bold

uppercase. The full model code is listed in Appendix I. The standard units of

length are metres and the unit of time is hours.

Page 61: Ground Penetrating Radar for the parameterisation of

Andrew Howe

61

3.4.1 Precipitation

Rainfall data is available on an hourly time step as input data to the model.

Rainfall may be spatially distributed across the catchment or treated as a lumped

parameter. For this thesis rainfall was treated as lumped with an hourly

resolution collected by a tipping bucket raingauge located in the catchment.

While distributed information would potentially improve model accuracy, the

small catchment area of 3km2 should ensure that significant variation in rainfall

amount is limited.

Figure 3.1. Precipitation module

Page 62: Ground Penetrating Radar for the parameterisation of

Andrew Howe

62

3.4.2 Evaporation

The conversion of water to vapour is termed evaporation. The rate of

evaporation is determined by the amount of energy available and the availability

of water to evaporate. Following an approach outlined by Shuttleworth (1992),

the maximum evaporation rate (Epot) can be derived from net radiation input

(RN) to a grid cell per time step and the latent heat of vaporisation (γ). Hourly

net radiation data is available from an automatic weather station (AWS) located

in the headwaters of the catchment.

(3.1)

Epot is the rate of evaporation given an unlimited supply of water from a soil

column (Wood, 1995). Actual evaporation (E) is controlled both by the

potential rate and by the availability of local soil water (θ) for that time step. As

a result actual evaporation approaches zero when soils near wilting point and

attains the maximum (Epot) for saturated soils (θsat). This approach has been

successfully used by Beven et al., (1984) and Quinn & Beven, (1993) within

TOPMODEL. Local evaporation can be calculated as:

(3.2)

In common with the rainfall module, net radiation and hence potential

evaporation is a lumped parameter with the hourly-recorded value from the

AWS applied to all cells across the entire catchment. However maximum soil

moisture and actual cell soil moisture are spatially variable resulting in different

=

SatpotEE

θθ

γN

potR

E =

Page 63: Ground Penetrating Radar for the parameterisation of

Andrew Howe

63

actual evaporation rates between cells. Derivation of these parameters is

discussed in the soil moisture module.

Page 64: Ground Penetrating Radar for the parameterisation of

Andrew Howe

64

3.4.3 Infiltration

Infiltration is the process by which water enters the soil. The rate of infiltration

is determined by antecedent soil moisture, soil porosity, the moisture dependent

hydraulic conductivity and the rate of rainfall.

The time variant infiltration rate has been simulated by empirical equations i.e.

Horton (1933), approximate theory based models i.e. Green-Ampt (1911),

Philip (1957), and by rigorous, physically based methods i.e. the Richards

equation (Rawis et al., 1992). The latter two classes of model require a large

number of parameters to calculate infiltration and the Richards equation also

needs appropriate boundary conditions to be specified. A more suitable

approach for a conceptual type model is via an equation limited to a few

physically measurable parameters, ideally derived from field measurements.

Mulligan and Thornes (In press) propose that infiltration rates are primarily

controlled by changes in porosity and saturated hydraulic conductivity (Ksat)

with depth and developed the following model of infiltration under such

conditions.

Neglecting macropore flow, water entering the soil from the surface moves

down the profile with a defined wetting front. The sharp boundary between wet

and dry areas is produced as a result of the low hydraulic conductivity of the

drier soil beneath (Campbell, 1985). The transmission zone, a zone of constant

water content, extends from the wetting front to the surface while precipitation

equals or exceeds the movement of water through the profile. The rate of water

movement through the profile is controlled by the rate of advance of the wetting

front, which therefore controls the maximum rate of surface infiltration.

Campbell (1985) demonstrated that saturated hydraulic conductivity could be

calculated as a function of bulk density and soil texture. Texture refers to the

Page 65: Ground Penetrating Radar for the parameterisation of

Andrew Howe

65

sand, silt and clay fractions of a soil and is the simplest method for deriving soil

moisture retention properties (Rawis et al., 1992).

(3.3)

where mfc and mfs are the mass fraction of clay and sand respectively, ρb is bulk

density (g/cm3) and b is a pore interaction term calculated using σg, the

geometric standard deviation of particle diameter and dg, the geometric mean

particle diameter. Typical b values range from 2 to 24 (Campbell, 1985).

(3.4)

For homogenous soils with uniform texture and particle size, saturated

hydraulic conductivity is a function of bulk density. For a homogenous soil,

bulk density increases linearly with depth due to compaction by overburden,

resulting in a non-linear decrease in porespace. Decreasing porespace means

that a unit volume of water will saturate an increasing volume of soil as the

water body moves downwards.

The relationship between porosity (φ) and bulk density (ρb) is given by equation

3.5, where particle density (ρd) is taken from the literature to have a value of

2.65g/cm3 (Ellis et al., 1995).

(3.5)

[ ]3.7mfs-6.9mfc-3.1

3 e3.1104b

bsatK

×= −

ρ

51 g

gdb

σ+=

d

b

ρρ

1−=φ

Page 66: Ground Penetrating Radar for the parameterisation of

Andrew Howe

66

The measured relationship between average sample porosity and soil thickness

from field measurements is presented in Chapter 4. A theoretical example of the

expected reduction in porosity with increasing overburden thickness is shown in

figure 3.3.

Figure 3.3. Theoretical depth-porosity relationship

Increasing bulk density also leads to a reduction in average pore volume,

resulting in reduced hydraulic conductivity values. As the wetting front

progresses downward and bulk density increases, saturated hydraulic

conductivity decreases. Correspondingly the surface infiltration rate is reduced.

Figure 3.4 shows the expected reduction in saturated hydraulic conductivity

with increasing soil thickness, derived using equation 3.3.

Page 67: Ground Penetrating Radar for the parameterisation of

Andrew Howe

67

Figure 3.4. Saturated hydraulic conductivity with varying depth of wetting front.

Infiltration rate is considered to be equal to the rate of advance of the wetting

front (Ksat). The depth of the wetting front (WFdepth) within the soil column

therefore controls infiltration rate. Given an initial wetting front depth and

known water input the proportion of infiltration and surface runoff (localXS)

each timestep (t) can be calculated using equations 3.3 and 3.5, and assuming

that water is preferentially infiltrated. Subsequent to infiltration the new wetting

front depth is calculated on the basis that the depth of soil needed to hold the

corresponding amount of water can be derived from the depth (x) – porosity (φ)

relationship (equation 3.6)

The total depth of soil needed to store a known amount of water between two

depths x1 and x2 is equivalent to the integral of equation of the line in figure 3.3

between the limits x1 and x2.

(3.6)

dxex

x

x∫ −= 1

2

φ

Page 68: Ground Penetrating Radar for the parameterisation of

Andrew Howe

68

In the case of infiltration, φ = infiltrated water this timestep, and x1 = wetting

front depth from the previous timestep. Solving equation 3.6 and rearranging to

make x2 the subject gives the new wetting front depth for the current timestep.

To maintain model simplicity, multiple wetting fronts within a soil profile were

not considered. Rainfall is added to the same wetting front for as long as the

timestep rainfall amount exceeds zero. Once rainfall ceases the wetting front is

considered to break up and disperse, resulting in a new wetting front depth of

zero until the next storm event.

Page 69: Ground Penetrating Radar for the parameterisation of

Andrew Howe

69

Page 70: Ground Penetrating Radar for the parameterisation of

Andrew Howe

70

3.4.4 Soil Moisture

Soil moisture status has long been recognised as an important factor influencing

the hydrological response of catchments to storm events, for example in the

evaporation module actual evaporation is partially determined by available

moisture. In this module the store of soil moisture by the model and the impact

of soil moisture status on subsurface flow and surface runoff generation will be

explained.

The maximum water content of a soil is controlled by the amount of porespace

available. Highly porous soils hold a greater amount of water than do those with

low porosity. Equally, for soils of constant porosity the thicker the soil the

greater the amount of water required for saturation to occur (assuming rainfall

does not exceed the infiltration rate). In reality porosity is found to decrease

with increasing soil thickness. The exact form of the relationship obtained from

55 field measurements is presented in Chapter 4.

The maximum amount of water which can be stored in a cell can be calculated

using knowledge of each cell’s soil thickness from GPR measurements, and the

relationship between porosity and depth obtained from field measurements (fig

3.3). Since all cells have equal areas and as all model inputs are in metres,

(rainfall and evaporation rates) rather than areas or volumes, the model

calculations are kept as units of length. The total porosity of any depth of soil

can be calculated using the relationship between soil thickness and the

cumulative porosity to that point. Integration of the depth – porosity curve

(equation 3.6) between the limits of zero and total soil thickness is equal to the

total porespace per depth of soil.

The ratio of water filled pores (soil moisture) to maximum pore space

(maximum moisture) is the cell saturation percentage. Assuming water

preferentially reaches the bedrock boundary and subsequent water inputs build

Page 71: Ground Penetrating Radar for the parameterisation of

Andrew Howe

71

up towards the surface, the depth of water table for each cell can be calculated.

Saturation excess overland flow is generated once cell saturation reaches one

hundred percent, i.e. soil moisture equals maximum cell moisture. The

calculation of water table position is similar to calculation of the position of the

downward moving wetting front discussed in the previous infiltration module.

In this model profile saturation can arise from three scenarios. Saturation could

occur via the wetting front reaching the bottom of the profile (top-down type).

Alternatively total saturation can occur if lateral water input from neighbouring

cells exceeds outflow causing the water table to rise to the soil surface (bottom-

up type). The third case occurs when a downward moving wetting front reaches

the water table resulting in profile saturation.

For every grid cell soil thickness, the maximum moisture and actual moisture

content are known. Using the theoretical depth-porosity relationship of equation

3.6, the position of the water table can be calculated by integration between

maximum soil thickness (xmax) and new unknown water table depth (x) for a

known amount of water. The new water table depth (x) is referenced to the soil

surface, e.g. when actual moisture is equal to the maximum moisture value,

water table depth equals zero. Likewise when soil moisture equals zero the

water table depth is equivalent to the maximum cell soil thickness. The

cumulative porosity between the maximum depth (xmax) and an unknown depth

(x) can be calculated as the integral of equation 3.8 between xmax and x.

(3.8)

solving equation 3.8 to make x the subject allows calculation of the new water

table depth each timestep.

dxex

x

x∫ −=max

φ

Page 72: Ground Penetrating Radar for the parameterisation of

Andrew Howe

72

Page 73: Ground Penetrating Radar for the parameterisation of

Andrew Howe

73

3.4.5 Subsurface Flow

Flow under unsaturated conditions is driven by gravitational and matric

potentials (Campbell, 1985). In these situations a soils hydraulic conductivity is

dependent on the volumetric soil moisture content (Rawis et al., 1992). The

model simulates flow between cells which may be partially or fully saturated.

The approach used is similar to that of Xiao et al., (1996) who used a GIS to

develop a raster based surface-subsurface hydrological model for a 2511km2

catchment in Alaska. Although the Cyff catchment has an area of only 3km2, the

subsurface flow module used by Xiao et al., (1996) is applicable at this scale

since it is physically based and only requires three parameters for each cell;

surface elevation, soil thickness and the saturated hydraulic conductivity (Ksat).

While Xiao et al., only considered lateral subsurface flow to occur when a cell

was 100% saturated i.e. the water table was at the surface, in this model flow

will be modelled for soils which are less than fully saturated. In this case the

amount of water contributing to subsurface flow is proportional to water table

height above bedrock rather than total cell depth. Calculation of water table

height each time-step as a function of total soil thickness and the depth –

porosity relationship has been explained within the soil moisture section.

The rate of saturated flow is directly proportional to the difference in hydraulic

head (H) between two points and inversely proportional to the flow length (x)

between them. This constitutes Darcy’s Law, where the constant of

proportionality is described as the saturated hydraulic conductivity (Ksat).

(3.9)

xHAKQ sat

∆∆

−=

Page 74: Ground Penetrating Radar for the parameterisation of

Andrew Howe

74

The hydraulic head gradient depends on the change in elevation of the water

table over a horizontal distance. In terms of lateral saturated flow from cell i to

an adjacent grid cell i +1 Darcy’s law can be written as

(3.10)

For the grid structure used by PCRaster, x equals the distance to the next

downstream cell. For flow directions cardinal to the source cell this distance is

equal to cell length. The downstream distance for flow diagonal to a cell is

calculated using Pythagorean theorem (figure 3.7). PCRaster uses the

DOWNSTREAMDIST command to automate this process.

Figure 3.7. Possible drainage paths from a grid cell using the D8 flow algorithm.

−−

−=+

+

1

1

ii

iisat xx

hhAKQ

( )2length cell 2 RED

length cell BLACK

×=

=

x

x

Page 75: Ground Penetrating Radar for the parameterisation of

Andrew Howe

75

The change in head (H) between cells can be calculated from water table depth

and cell elevation. This is shown in figure 3.8 and expressed as

(3.11)

Hydraulic gradient is therefore

(3.12)

Typically subsurface flow is routed using surface topography. However while

hydraulic gradient is largely controlled by cell elevation, considered to be

constant for model runs (i.e. no significant erosion or deposition takes place),

water table depth is also an important factor. Water table position reflects the

soil moisture status of each cell and is a dynamic feature. Changes in water

table depth between cells may lead to a different pattern of subsurface flow

direction than that predicted solely on the basis of topography. The effect is

likely to be greatest in valley bottom areas with low slope angle and deeper soils

where differences in elevation between cells are minimal. In order to model this

effect a local drainage direction network based on water table height and cell

elevation is calculated for each timestep and this network used for the routing of

subsurface flow.

xhhhh

xH )()( 1324 −−−

=∆∆

le) water tab thedepth to -elevation (Celldownstream and

le) water tab thedepth to-elevation Cell( where

)()(

13

24

1324

=−

=−

−−−=

hh

hh

hhhhH

Page 76: Ground Penetrating Radar for the parameterisation of

Andrew Howe

76

Figure 3.8. Cross-section of two adjacent grid cells with different elevation, soilthickness and water table depth.

For those cells which are not fully saturated the lateral hydraulic conductivity

was calculated using a method outlined by Campbell (1974, 1985). In this case

hydraulic conductivity (K) is determined by the ratio of cell soil moisture to

maximum soil moisture, a pore interaction term (equation 3.4), and saturated

hydraulic conductivity, calculated by:

(3.13)

The potential amount of water available to move from a cell to a downstream

cell is a function of the hydraulic gradient (Hgrad) and the hydraulic

conductivity of the soil (K). The limiting factor is the amount of water available

h3

h4

Zero level

h2

h1Water table depth

Soil depth

Cell elevation

Bedrock

d1

d2

Bedrock

)32(

satθθ

+

=

b

satKK

Page 77: Ground Penetrating Radar for the parameterisation of

Andrew Howe

77

to move from a cell in the time period (soilmoist). Subsurface output from a cell

(as metres of water) can therefore be expressed as:

Cell output = IF (soilmoist > Hgrad × K THEN Hgrad × K ELSE soilmoist)

Input from upstream cells is calculated using the PCRaster UPSTREAM

command which sums the output from all cells immediately upstream of the

input cell using the local drainage direction (LDD) network.

Cell input = Σ (upstream cell output)

Considering only adjacent upstream cells to be contributors to subsurface flow

can be justified by the slow rates of lateral flow, typically of the order of one to

ten centimetres per hour. This is small compared to cell widths of 10 – 25m

which the model is designed for.

A balance equation is then used to check that the maximum moisture capacity

(maxmoist) of a cell has not been exceeded. If the maximum storage capacity of

the cell is exceeded in any timestep the excess water is considered to be return

flow (ROF) and added to the localXS total, where it is routed downslope using

the surface runoff module.

The local elevation of the water table is equal to cell elevation minus water table

depth. A equals the average flow area between adjacent cells and is calculated as

the product of cell width (y) and the difference between mean water table depth

(h1, h2) and mean soil thickness (d1, d2).

(3.14)

yhhddA

+

−+

=22

1212

Page 78: Ground Penetrating Radar for the parameterisation of

Andrew Howe

78

Therefore cell subsurface discharge (Q) for each timestep is calculated as:

(3.15)

yhhddx

hhhhKQ

+

−+

−−−

=22

)()( 12121324

Page 79: Ground Penetrating Radar for the parameterisation of

Andrew Howe

79

Page 80: Ground Penetrating Radar for the parameterisation of

Andrew Howe

80

3.4.6 Surface Runoff

Ponded surface water that cannot be infiltrated during a timestep becomes

runoff. Runoff is routed to the next downstream cell via the surface local

drainage direction network calculated from the catchment DEM. Runoff takes

the form of infiltration excess, saturation excess or return overland flow type.

Runoff in PCRaster can be modelled using a number of approaches. PCRaster

includes ‘ACCU…’, a group of functions designed to route inputs (water,

sediment etc) through and out of the catchment each timestep. These are useful

operators when the material transported is not considered to contribute to any

other model processes. It is of less use when the infiltration of surface runoff

downslope of the source cell is considered or for any processes where the

material travel time through the catchment is more than the model timestep. In

this case a more sophisticated routing algorithm is required. One example is to

use the manning equation to route runoff, another to use the kinematic wave

approximation.

The kinematic wave approximation for flow routing has been successfully

implemented in PCRaster by De Roo et al., (1998b). The algorithm is

implemented using the KINEMATIC command and requires the following

parameters

• Timestep (seconds)

• Cell length (m)

• Beta, equals 0.6 for sheet flow

• Q, side flow input, equals zero

• mannings n value

• Surface water height (m)

• Local slope gradient map (m/m)

• Local drainage direction map

Page 81: Ground Penetrating Radar for the parameterisation of

Andrew Howe

81

The dominant vegetation types found in the Cyff are grass species, often

cropped as a result of sheep grazing. These vegetation types correspond to a

mannings n value of 0.1-0.2 (Stephenson & Meadows, 1986). For the purposes

of this thesis a value of 0.15 for mannings n was used in all model simulations.

A height of surface water (localXS) for any cell within the catchment can be

generated by three mechanisms:

1. Direct precipitation to a cell for the timestep failing to infiltrate during that

timestep. This situation may occur either because rainfall intensity exceeds

the infiltration rate, or as a result of soil saturation at a cell.

2. Surface water (localXS) routed downslope as runoff from the previous

timestep, and that is not able to infiltrate.

3. Return overland flow (ROF) from cells that exceed the maximum moisture

capacity due to subsurface flow from upslope cells.

Page 82: Ground Penetrating Radar for the parameterisation of

Andrew Howe

82

Page 83: Ground Penetrating Radar for the parameterisation of

Andrew Howe

83

3.5 Conclusion

PCRaster provides a modelling framework embedded within a GIS for the

development of dynamic, spatially distributed models. The benefit of using an

integrated GIS-model is the capability for storage and display of dynamic,

spatially heterogeneous data, using the raster architecture of the software to

overlay different data maps. The overall structure and requirements of the

model have been explained. Where possible all modules are based on standard

routines used by other hydrological models in order that the emphasis is on

using GPR data for hydrological modelling, not on developing an entirely new

model. The potential of using GPR to derive distributed soil thickness has

however resulted in this model using depth as a key parameter, for example as a

control on the maximum moisture content of each cell.

Page 84: Ground Penetrating Radar for the parameterisation of

Andrew Howe

84

Chapter 4: Fieldwork

4.1 Introduction

This chapter describes the methodology used to collect the hydrological data

required for model input, parameterisation and validation. An initial outline of

previous catchment studies at Plynlimon and a justification for using this area is

presented, followed by a brief description of the climatic and geomorphological

background of the region. Subsequent sections explain which hydrological

parameters were collected and the methods used. The final part of this chapter

deals with creation of the catchment and hillslope digital elevation models

(DEM’s).

Page 85: Ground Penetrating Radar for the parameterisation of

Andrew Howe

85

4.2 The Plynlimon Catchments

The Plynlimon catchments are located in an upland region of Central Wales.

Two adjacent ‘paired’ catchments, the Wye and the Severn, have formed the

basis for an experimental programme set up under the encouragement of the

International Hydrological Decade, 1965-1974. The primary reason for

establishing Plynlimon was to investigate the differences in annual water budget

between grassland and forested catchments, in particular to test the hypothesis

that for UK conditions, evaporative losses were greater from coniferous forests

than pasture (Hudson, 1988).

In order to answer this question a network of weather stations, rain gauges and

flumes were established throughout the two catchments and maintained by the

Institute of Hydrology. The network now provides comprehensive spatial

coverage of precipitation, channel discharge and other climatic parameters

recorded by weather stations at daily and better temporal resolutions and

stretching back 30 years. The high spatial and temporal resolution of data has

resulted in Plynlimon becoming the focus for numerous hydrological studies.

These have included channel studies (Newson et al., 1978), soil piping studies

(Gilman & Newson, 1980), the impact of soil moisture storage to upland

catchment water balances (Hudson, 1988) and catchment scale model

simulations (Beven et al., 1984), (Quinn et al., 1993).

Plynlimon was chosen as the site for this study because of the availability of

hydrometric data for model input and validation, and because of the knowledge

available from previous studies of the soils and hydrological processes active in

the catchments. The study area for this thesis is a sub-catchment of the Wye

watershed, the Cyff. The Cyff was selected because of the location of a channel

flume measuring hourly discharge at the catchment outflow and two automatic

weather stations (AWS), one in the headwaters of the catchment and one

located 500m downstream of the outflow. The area of the Cyff is approximately

3.1km2 (derived from the catchment DEM) and a 1:5000 topographic map

Page 86: Ground Penetrating Radar for the parameterisation of

Andrew Howe

86

(Hunting Survey Ltd) of the area was available to use as source for the

catchment DEM, as well as aerial photographs from 9th July 1995. In terms of

the suitability for GPR use the Cyff is accessible by farm tracks and there are no

forestry plantations therefore radar profiles can be taken anywhere within the

catchment. Textural analysis of 29 soil samples from the catchment showed clay

content ranged from 0–39 % (by mass), generally increasing with depth, but

preliminary site investigation with GPR indicated that signal penetration depths

of 1-2m were possible using GPR frequencies of 225-900MHz.

Figure 4.1. Cyff catchment aerial photograph with watershed and stream networksuperimposed (NERC 1995).

Page 87: Ground Penetrating Radar for the parameterisation of

Andrew Howe

87

4.2.1 Climate

Plynlimon is one of the wetter regions of the UK due to the high elevation (320

to 740m) and the proximity of the Atlantic Ocean. The 1961-1990 average

annual rainfall for the Cyff catchment is 2416mm. (NERC, 1999). The amount

of received precipitation in the Plynlimon catchments has been shown to

increase with altitude, but no statistical relationship was found to exist between

precipitation and either slope angle or aspect (Clarke et al., 1973).

For the purposes of this study a calibrated tipping bucket raingauge (TBR) was

installed in the Cyff catchment at the field plot, located at an altitude of 454m A

total precipitation of 2559.6mm was recorded for the 1998 calendar year.

Monthly totals from October 1997 to June 1999, the TBR operational period,

are shown in figure 4.2.

Figure 4.2. Total monthly rainfall recorded by the field plot TBR.

Page 88: Ground Penetrating Radar for the parameterisation of

Andrew Howe

88

4.2.2 Hydrology

A network of ephemeral and perennial open channels, ephemeral soil pipes and

perennial flushes provides drainage within the Wye and Severn catchments.

Afforestation of the Severn required excavation of drainage ditches to a depth of

1.5m; in the Cyff no afforestation was undertaken but a network of open drains

of approximately 1m cross-section were excavated in the valley bottom bog.

These were excavated to improve pasture, but were not maintained (Newson &

Harrison, 1978). In 1999 there is little evidence of these artificial drainage

channels remaining.

4.2.3 Geomorphology and soils

The adjacent catchments of the upper Wye and upper Severn exhibit similar

morphologies with total areas of 10.55 km2 and 8.70 km2 respectively, similar

aspects, mean altitudes of 450m and similar drainage densities of 2.04 km/km2

and 2.40 km/km2. 91% of the Cyff lies between 340 - 539m above sea level,

median slope angle is 10.6o while the dominant drainage direction along the

central valley is to the south east (Newson, 1976). Significant differences in

landuse do exist between the catchments. The Wye is primarily used as rough

pasture for stock grazing while the Severn was 68.3% forested (Hudson, 1988).

At the present time many forest stands have reached maturity and as a result the

Severn is undergoing extensive deforestation. The stability of landuse in the

Wye is an added advantage for this study.

The soils of Plynlimon have been extensively documented by Rudeforth (1970),

Newson (1976) and Chappell & Ternan (1993) amongst others. The soils

present in the Cyff have evolved through the interaction between bedrock,

glacial and periglacial deposits of the tertiary and climate. The parent material

consists of mudstones and shales and is dominated by illite, quartz and iron rich

chlorite. Three major soil types have been recognised in this area. Ferric

Page 89: Ground Penetrating Radar for the parameterisation of

Andrew Howe

89

podzols are found on the mid-slopes and characterised by depositions of

aluminium/iron (Al/Fe) from the Ea horizon over a 5-10cm depth of the Bs

horizon. Placic podzols are found both upslope and downslope of the ferric

podzols on slopes with lower gradients. In these soils the Al/Fe is precipitated

and forms an ironpan, the Es horizon becomes gleyed and the organic layer

above the Es horizon forms a peat. The Bs horizon below the pan remains

relatively permeable compared with the Bs horizon of the ferric podzol (Adams,

1974). Moving towards the crest-slope and toe-slopes peat depth increases and

the Bs horizon becomes gleyed resulting hydraulic conductivities several orders

of magnitude lower than those of the peat layer (Chappell et al., 1993). The

soils described belong to the Hiraethog series, and in this area are underlain by a

C horizon of solid rock (shale), unconsolidated scree or compact drift

(Rudeforth, 1970).

Figure 4.3. A Cyff soil profile at ~500m above sea level. Note the dark organichorizon ~20cm deep, overlying a yellow/brown horizon and a C-horizonconsisting of coarse, angular shale.

Page 90: Ground Penetrating Radar for the parameterisation of

Andrew Howe

90

4.2.4 Vegetation

Acid grassland communities characterised by the grasses Nardus spp.

(matgrass) and Festuca spp. (fescue) dominate on placic podzol slope systems.

Mesotrophic mires are widespread in the valley bottoms and extend upslope via

‘rush flushes’. These mire areas are dominated by Juncus spp. (rush),

Eriophorum spp. (cotton grass) and Mollinia spp. (purple moor grass), all of

which thrive in saturated soils and are tolerant of submergence common during

the winter/spring months. While plant species can be taken as an indicator of

moisture status, vegetation distribution is also affected by nutrient status and

human interference. In the case of the Cyff this has taken the form of improved

pastures using surface treatments (Newson, 1976).

Page 91: Ground Penetrating Radar for the parameterisation of

Andrew Howe

91

4.3 Data collection of hydrological parameters

Field measurements of hydrological parameters are required as inputs to the

model and to enable model validation. As discussed in the modelling chapter,

the hydrological inputs comprise rainfall and potential evaporation data. Model

validation will be carried out in two ways. Firstly a lumped approach involving

a comparison between predicted surface and subsurface discharge and recorded

data obtained from a field plot within the catchment. This will provide an

internal state validation for one grid cell within the catchment as proposed by

Quinn et al., (1995). At the catchment scale a lumped validation will be

achieved by comparing the model hydrograph to the actual hydrograph over the

same time period. Secondly it is planned to carry out internal state validation of

the model, validating generated soil moisture levels per cell with field data

collected using GPR. Field data is also required for the parameterisation of the

hydrological model (e.g. the infiltration equation) and as a validation of radar

derived measurements of soil thickness and volumetric water content for the

catchment.

4.3.1 Plot-scale collection of hydrological parameters

A field site was established in April 1997 in the Cyff. The plot is south facing

and approximately 454m above sea level. The underlying soil is a ferric podzol

with a well defined organic and Bs boundary. A site within the catchment was

needed to measure the hydrological variables required as model input (rainfall)

and for the internal validation of the catchment model (runoff, subsurface flow

and soil moisture). Construction of a throughflow/runoff plot required

excavation of a 2.5m long trench approximately 40cm deep and perpendicular

to the line of steepest slope. Two plastic gutters 2m long were placed at depths

of 2cm and 30cm for the interception of runoff and throughflow respectively.

The guttering was sealed at one end and slightly inclined to allow water flow

Page 92: Ground Penetrating Radar for the parameterisation of

Andrew Howe

92

via funnels into the respective tipping buckets located in a covered pit to one

side of the plot. Sheet aluminium was used to provide two lips over which water

from overland flow and throughflow could drain into each gutter from the soil

face. Two aluminium covers were placed over the open top of each gutter to

prevent direct input by precipitation and sediment. The trench was back-filled in

an attempt to limit the impact of the excavation on hydraulic flow pathways.

The rain gauge was installed 2m downslope and to the right of the trench in a pit

to reduce the effect of wind turbulence on rain catch. The plot area after

installation of the runoff/throughflow measurement structure is shown in figure

4.4.

Between 29th September 1997 and 7th July 1999 the field site recorded surface

runoff, subsurface flow and rainfall via three tipping bucket gauges connected

to a datalogger. Soil moisture at three depths was also monitored between 28th

September 1998 and 7th July 1999. Prior to 29th September 1997 field data were

poor due to technical problems and these data are not considered in this thesis.

The data coverage provided by field plot instruments is summarised in table 4.1

Table 4.1. Field plot data record.

MeasuredVariable

From To TemporalResolution

Required for

Rainfall* 29/09/97 07/07/99 Hourly Model input

Runoff* 29/09/97 07/07/99 Hourly Internal model validation

Subsurface

flow*

29/09/97 07/07/99 Hourly Internal model validation

Soil moisture*

at depths of:

0.08m

0.23m

0.38m

28/09/98 07/07/99 Hourly Internal model validation

*No data for periods between 27th September 1998 - 29th September 1998 and 12th April 1999 -

20th April 1999.

Page 93: Ground Penetrating Radar for the parameterisation of

Andrew Howe

93

Figure 4.4. Runoff - subsurface flow field plot.

Page 94: Ground Penetrating Radar for the parameterisation of

Andrew Howe

94

4.3.2 Catchment-scale collection of hydrological parameters

Two parameters were needed at the catchment scale. Hourly potential

evaporation figures for input into the evaporation module were calculated using

hourly weather station data. Hourly river discharge from the basin outflow

provides the means for comparison between the measured and modelled

hydrograph. Both the channel flow gauge and AWS form part of the monitoring

network maintained by the Institute of Hydrology. The period for which

discharge and AWS data are available is summarised in table 4.2

Table 4.2. Data collected by Institute of Hydrology hydrometric network.

MeasuredVariable

From To TemporalResolution

Required for

Discharge 29/09/97 31/03/99 Hourly Model validation

Net Radiation 01/01/97 14/03/99 Hourly Model input

Page 95: Ground Penetrating Radar for the parameterisation of

Andrew Howe

95

4.4 Topography

Topography is a critical factor in the hydrological response of a catchment. Not

only is topography a major control on surface and subsurface flow pathways

(Quinn et al., 1994), it is also an important factor in the downslope

redistribution of soil moisture during interstorm periods (Wood, 1995). For

many distributed models topography is the only truly spatially distributed

parameter available and as a result it is often used as a covariate to model the

spatial distribution of other variables (Blöschl et al., 1995).

Topography has been recognised as one factor important to the short-term

catchment hydrological response and the long-term evolution of hillslope form.

Over longer time scales topography can be considered to influence soil

development. The local influence of topography includes slope gradient which

controls rates of water movement over and through soils, and also the rates and

process of soil movement downslope (Selby, 1993). Gravity driven transport

processes typically result in thin soils on topographic ridges and those areas

adjacent to the watershed, while accumulation zones are located in valley

bottoms (Heimsath et al., 1999). As a result of the importance of topography to

hydrology, and the widespread availability of detailed topographic maps, digital

elevation models (DEM’s) are extensively used in distributed hydrological

modelling. This thesis uses a 1:5000 map of the Cyff catchment to produce a

catchment DEM. This forms the basis for modelling the internal dynamics of

hydrological response to precipitation events through routing and moisture

distribution. The area bounding and upslope of the hydrological monitoring

station was surveyed to produce a finer resolution DEM of local topography.

This was carried out to quantify the area contributing to surface runoff and also

potentially the source area of contributing to subsurface flow.

Page 96: Ground Penetrating Radar for the parameterisation of

Andrew Howe

96

4.4.1 The catchment DEM

Conversion from the 1:5000 contour map of the Cyff catchment to a raster

DEM was achieved by scanning the map and digitising the contour lines using

ARC/INFO. The vector file containing contours, stream network location and

spot heights was converted to a raster map using ARC/INFO-TOPOGRID and

exported to PCRaster. The catchment drainage network was derived in

PCRaster using the local drainage direction algorithm discussed previously. A

25m grid size catchment DEM is shown in figure 4.5.

An intrinsic part of DEM creation is the choice of grid size. Grid size is

important because the description of processes logically requires grid

resolutions of similar or finer scale. Coarse grids are unable to represent

detailed catchment features, for example gully features that exert control on

surface flow routing. Secondly internal validation of distributed models using

point data cannot be achieved with coarse resolutions. There seems to be no

optimum grid size for hydrological studies, rather the choice of grid resolution

depends on the output required by the modeller. The production of hydrographs

from coarse scale DEM’s is possible, but if internal state validation is required

grid resolutions finer than 50m are needed (Quinn et al., 1995).

Grid size not only determines the resolution of topographic features but also

affects the values of terrain attributes derived from DEM analysis. The impact

of changing grid size on terrain attributes and hydrology was explored by

Bruneau at al., (1994) who found that catchment area decreases as grid

resolution is increased and consequently simulated discharge is also reduced.

The topographic index was found to be sensitive to changing grid size, with grid

cells greater than 50m primarily controlled by local slope and finer grids

showing increased sensitivity to cumulative upslope area. Quinn et al., (1994)

downscaled from a 12.5m to a 50m grid to analyse the impact on topographic

Page 97: Ground Penetrating Radar for the parameterisation of

Andrew Howe

97

index values. They found an increased percentage of higher index values, due to

the increase in minimum cell area.

Page 98: Ground Penetrating Radar for the parameterisation of

Andrew Howe

98

Figure 4.5. 25m grid resolution DEM of the Cyff catchment, overlain with a map ofthe logarithm of upslope drainage area for each cell.

Page 99: Ground Penetrating Radar for the parameterisation of

Andrew Howe

99

4.4.2 The hillslope DEM

In order to quantify the upstream area contributing to the plot, and thereby

enable comparison between plot measured surface/subsurface flow and model

predictions, a higher resolution DEM of the immediate area was required. 277

points were surveyed over the hillslope adjacent and upslope of the plot, and

used to derive surface topography using kriging (figure 4.6 and figure 4.7).

Visually the plot is located on the toe-slope of a topographic ridge with a

pronounced gully system to one side. The drainage area upslope of the plot is

likely to be relatively small and this is confirmed by measurement of the cell

drainage area from the hillslope DEM, in this case 200m2. The effect of changes

in DEM grid resolution on topographic attribute values is summarised in table

4.3.

Table 4.3. Terrain attributes of the plot cell for variable DEM grid size.

Source DEM GridSize (m)

Slope Angle(degrees)

Area

m2

Area asnumber of

cells

WetnessIndex

LN(a/tanβ)

Catchment DEM 25 15.1 625 1 7.751

Catchment DEM 10 15.8 400 4 7.252

Slope DEM 5 15.9 200 8 6.556

Page 100: Ground Penetrating Radar for the parameterisation of

Andrew Howe

100

Figure 4.6. Elevation map with survey points marked in red and location of thefield plot marked in blue.

Page 101: Ground Penetrating Radar for the parameterisation of

Andrew Howe

101

Figure 4.7. 5m grid DEM of the hillslope overlain with LN(upslope area) map andlocal drainage direction network (white lines indicate cell drainage paths).

Page 102: Ground Penetrating Radar for the parameterisation of

Andrew Howe

102

4.5. Site selection using terrain attributes

DEMs provide a method to derive maps of terrain attributes potentially of use to

hydrological studies. This is because of the relationship that exists between

topography and hydrological fluxes. The first order derivatives of topographic

surfaces are slope and aspect; second order derivatives include plan and profile

curvature (Burrough & McDonnell, 1998). Terrain attributes or topographic

indices have been used as surrogate indicators to describe the spatial

distribution of variables, for example soil moisture. They are therefore useful

for designing sampling schemes when little prior knowledge of the spatial

variation in parameters is known. In this study topographic indices were used

initially to identify potential sites for subsurface investigation using GPR.

Kirkby & Chorley (1967) investigated the significance of topography in

defining zones of water accumulation and hence the likelihood of contributing

to subsurface runoff and saturation excess flow. Hillslope hollows and low

gradient slopes were found to contribute to this mechanism of flow initiation.

Both of these zones are identifiable through analysis of a DEM which can be

used to derive maps of the spatial distribution of plan curvature, upslope

drainage area and local slope. Burt et al., (1985) explored the use of topographic

indices for predicting soil moisture status for a UK catchment, and found that an

index based on upslope drainage area provided the greatest correlation with

observed soil moisture, particularly during wetter conditions.

The importance of topography has led to the development of a number of

topographic indices that predict the local state of moisture flux. One such

topographic index, also referred to as a wetness index, is calculated on the basis

of local slope angle (β) and cumulative upslope area (a) and reflects the

tendency for water to accumulate at any point. The value of accumulated

upslope area is considered proportional to the volume of water moving through

a cell, and local slope angle is a measure of the capability of a cell to transport

water (Quinn et al., 1993). On shallow gradient slopes subsurface flow is

Page 103: Ground Penetrating Radar for the parameterisation of

Andrew Howe

103

limited by reduced hydraulic gradients and saturation of soils is more likely

(Burt et al., 1985). The wetness index is therefore a possible indicator of soil

moisture status, which is seen as an important factor in the generation of

saturation excess flow. The exact form of the equation is given by

(4.1)

Other topographic indices include a/β and plan curvature. Plan curvature

provides an indication of flow convergence or divergence, identifying

topographic hollows that concentrate flow and are therefore potential zones of

saturation. These and combinations of topographic attributes have been used to

investigate the changing distribution of soil moisture, runoff, erosion and

deposition, and catenary development through time and space.

The concept of using topography as a covariate for soil moisture status has been

extended to include prediction of other soil attributes, including soil thickness.

Jenny (1941) summarised the key factors in soil development as climate, parent

material, topography and biotic activity. For a small catchment area, such as the

one used for this study, climate, parent material and biotic activity can be

considered to act uniformly across the basin. Topographic indices therefore

have the potential to partially explain the spatial variability of soil attributes,

including soil thickness, for many environments. One hypothesis considered is

that for landscapes experiencing dominantly fluvial processes the soil catena

develops in response to water movement (Moore et al., 1993). Heimsath et al.,

(1999) consider the dominant transport processes in upland environments to be

mass wasting and overland flow, resulting in shallow soils adjacent to

catchment divides and on steep slopes, and the accumulation of soils in valley

bottoms. Upland Wales is one such landscape.

Boer et al., (1996) and Moore et al., (1993) used terrain attributes derived from

DEM’s to estimate soil thickness variability in Southeast Spain and Colorado

respectively. Both studies used local slope angle, aspect, profile curvature and

=

βtan ln aIndexWetness

Page 104: Ground Penetrating Radar for the parameterisation of

Andrew Howe

104

wetness index as covariates for the prediction of soil thickness. Moore et al.,

(1993) found that up to 50% of soil thickness variability could be explained

with slope and wetness indices. Boer et al., (1996) found that these indices

provided an accuracy of 40-81% in terms of predicted and measured soil

thickness, highly dependent on lithology.

In this study topographic indices were initially used to develop a sampling

strategy for GPR site selection. The size of the catchment precluded a survey of

the entire area. Instead the catchment DEM was used to derive maps of the

spatial variation in slope gradient, upslope drainage area and wetness index.

Given that these topographic indices have been shown to explain a certain

amount of soil moisture and soil thickness variability, one hypothesis would be

that sampling areas which cover the range of index values would result in

adequate spatial coverage of the variables of interest. The indices chosen as

guides for initial site selection were slope, upslope drainage area and wetness

index.

For each index a histogram distribution function of pixel values was produced.

Figure 4.8 shows the distribution functions for slope, log(upslope area) and

log(wetness) for the 25m-grid cell DEM. The histogram for each index was

used to identify the range of pixel values present in the catchment. The number

of cells sampled for each index range was determined by the frequency of cells

in each index range. A large frequency of cells with a particular value resulted

in the allocation of a larger number of sample sites than for lower frequency

values.

Page 105: Ground Penetrating Radar for the parameterisation of

Andrew Howe

105

Figure 4.8. Frequency distribution of pixel values for the three topographicindices used to locate potential GPR survey sites.

Page 106: Ground Penetrating Radar for the parameterisation of

Andrew Howe

106

4.5.1 Locating sample sites using topographic indices

Given the large number of potential sites associated with each index value a

method of displaying only those cells within a specific index range was

required. Using boolean operators within PCRaster a series of maps were

created which identified the location of cells which met the search criteria. For

each of the three indices, cells were found with index values within their

respective histogram bin using greater than and less than operators. The

resulting map allocated a true value to all cells that met the criteria and a false

value to those that did not. Using this method a series of simple maps was

created showing the location of all possible sample sites for a particular index

range. Maps of slope, wetness and upslope area were overlain and combinations

of these index values used to display pixels that met the search criteria. In total

60 sites were identified as potential survey areas and 32 were actually sampled

during subsequent fieldwork. Due to the physical weight of the GPR system and

the lack of off-road transport it was necessary to choose sample sites which

were within approximately one kilometre of vehicle tracks. As a result the

majority of field sites are located on the south facing slopes of the Cyff, where

access is provided by a track running south-east to north-west (appendix II,

figure II.1.). The co-ordinates of the centre of each cell chosen for sampling

were recorded for subsequent use with a GPS system for cell location in the

field. The co-ordinates and key terrain indices for each sample site are listed in

appendix II, table II.1.

Page 107: Ground Penetrating Radar for the parameterisation of

Andrew Howe

107

4.5.2 Site location in the field

A GARMIN differential GPS system was used to locate sites in the field. The

GARMIN unit is a handheld system and in this case used two satellite receivers

for differential positioning. In differential mode the accuracy of the unit is 1-

5m. Sites were located on September 29th and 30th 1998 and marked with

wooden stakes and identification numbers. Given the 25m DEM grid size even

a worst case scenario with a GPS error of 5m would still position the site

marker within the correct grid square. Whilst in the field the GPS positions of

the catchment outflow and a road cross were recorded to allow comparison

between GPS and map co-ordinate values. The results are shown in table 4.4.

Table 4.4. GPS and DEM feature co-ordinates.

Feature GPS x(m)

GPS y(m)

DEM x(m)

DEM y(m)

x Error(m)

y Error(m)

Catchment

–outlet.

282374 284634 282374 284627 0 7

Road cross. 281514 285706 281524 285702 10 4

Page 108: Ground Penetrating Radar for the parameterisation of

Andrew Howe

108

4.6 GPR survey methodology

Initial site investigation took place during the Spring and Summer of 1997. A

key requirement was field testing the radar, as GPR is limited as a survey tool in

electrically conductive environments. Initial investigations to assess the

suitability of using GPR at Plynlimon concentrated on a 20m2 hillslope area,

subsequently used as the site for the hydrological monitoring station. All

available antennae frequencies (225, 450 and 900 MHz) were used to collect

subsurface data from four 20m survey lines in both reflection and CMP modes

of operation. GPR data was complemented by augur holes every metre and two

trial pits. 30 soil samples were collected for textural analysis in order to quantify

clay content at differing depths. The results shown in figure 4.9 detail a general

increase in silt with depth, and a peak in soil clay content between 0.1m - 0.2m

depth of sample. The percentage of gravel particles (>2mm) increases with

increasing sample depth. Soil textural classification used the British Standards

system (Ellis et al., 1995).

The conclusions drawn from the preliminary investigations were:

1. GPR was a viable survey tool in this area due to the relatively low clay

content of the soils (0-32%).

2. Higher frequency antennae (900MHz) with associated higher resolution

provided the optimum method for locating the soil-bedrock interface across

much of the catchment. Excavation and augur measurements showed that

the local soils were relatively thin, with an organic horizon typically less

than 0.2m thick and bedrock located within 1m of the surface.

3. Areas of peat bog in the valley floor and to an extent, drainage channels,

proved the exception to the above. In these areas of the catchment the

deeper soils required 450MHz antennae be used since the lower frequency

provided improved penetration to the bedrock boundary.

Page 109: Ground Penetrating Radar for the parameterisation of

Andrew Howe

109

Figure 4.9. Soil texture data from thirty samples.

4.6.1 Reflection survey design

GPR reflection surveys were carried out in order to image the subsurface, and

with the specific aim of locating the soil-bedrock boundary and measuring soil

thickness. To achieve this radar profiles were taken along a 2m-survey line at

each site. The choice of line length was based on the following reasoning. The

standard use of GPR is one of subsurface feature identification, for example

pipe location or bedrock structure. In these and for the case of bedrock

boundary identification, interpretation of subsurface structure is achieved by

examining the spatial pattern of subsurface response. Closely spaced radar

returns allow an image of the soil to be built up and the nature of features to be

resolved. It is only through the accumulation of adjacent radar traces that a

reflector can be identified as a linear, dipping or point feature, or as noise. The

actual number of adjacent traces required to identify a feature is dependent on

the survey aim and spatial variability of the target. Ultimately human input is

required to judge the precise nature of a reflector. Using the GPR profiles

Page 110: Ground Penetrating Radar for the parameterisation of

Andrew Howe

110

collected during the preliminary survey stage a survey line length of 2m was

found visually to be useful in interpreting linear reflectors which corresponded

to soil horizons. Figure 4.10 shows the effect of reducing survey line length on

feature identification.

Figure 4.10. The impact of reducing profile length on interpreting subsurfacefeatures. The profile, from a 225MHz survey, is displayed with DEWOW and aconstant gain of 20.

A statistical approach for the determination of an optimum GPR survey length

considered the variation in soil thickness along each profile. Using auger data

collected at one-metre intervals along the four 20m survey lines, upslope of the

field plot, the variation in depth was analysed using simple statistics and

presented in table 4.5.

Table 4.5. Soil thickness variation at the field plot.

Survey line Line1 Line2 Line3 Line4

Minimum depth 0.65 0.52 0.4 0.37

Maximum depth 0.28 0.21 0.25 0.12

Range 0.37 0.31 0.15 0.25

Mean depth 0.39 0.34 0.33 0.27

Standard deviation 0.095 0.078 0.046 0.069

Page 111: Ground Penetrating Radar for the parameterisation of

Andrew Howe

111

Individual depth measurements within each profile are subject to high

variability. Averaging two or more adjacent depth measurements reduces this

variability and leads to a decline in the value of standard deviation. A

dimensionless measure of variability is provided by the coefficient of variation,

which is equal to standard deviation divided by the sample mean. Figure 4.11

shows the effect of increasing the number of adjacent points used to calculate

average depth.

Figure 4.11. The change in coefficient of variation with an increasing number ofauger samples.

Initially the coefficient of variation is high, but decreases as the number of

auger measurements used to calculate the depth at any point are increased. The

rate of decrease is initially high, but the rate of change decreases after the

number of observations reaches 3-4. After this point the coefficient of variation

still declines, but at a lesser rate. For these data sets a useful auger profile length

for soil thickness measurement is 3-4 observations when the spacing between

measurement points is one metre. This is equivalent to a GPR profile length of

2-3m. Using this information it was decided to sample each site using a two-

metre survey line, orientated parallel to the maximum slope.

0 .000

0 .050

0 .100

0 .150

0 .200

0 .250

0 .300

0 1 2 3 4 5 6 7 8 9 10 11

N u m ber o f o bservation s used to averag e

coef

ficie

nt o

f var

iatio

n

L ine1L ine2L ine3L ine4

Page 112: Ground Penetrating Radar for the parameterisation of

Andrew Howe

112

4.6.2 Time Window

The time window refers to the length of time the GPR system records a

returning signal. Signals that take longer in time to return to the receiver have,

in general, travelled a greater distance through the subsurface. Selection of an

appropriate time window is an important component of GPR survey design

since it determines the limits of imaging depth. Augering carried out at 1m

intervals along the initial four 20m profiles showed that soil thickness ranged

between 0.12 – 0.65m over the area investigated. Analysis of four CMP’s

carried out in the adjacent area produced velocity estimates ranging from 0.06 –

0.08m/ns. These low velocity values reflect the wet soil conditions prevalent in

upland Wales and provided an initial basis for the choice of time window.

Combining the greatest depth with the slowest estimate of wave velocity

produced a minimum two-way time window of 22ns. During actual GPR

surveys the time window was set at between 60 - 100ns, given the likelihood of

soils exceeding 0.65m depth and the possibility of slower electromagnetic wave

velocities. Assuming a velocity of 0.06m/ns throughout the subsurface these

time window values are equivalent to an imaging depth of 1.8 – 3.0m.

4.6.3 Station Spacing

Station spacing refers to the horizontal interval between adjacent radar traces

along a survey line and is a key factor in the definition of subsurface features.

As the distance between sample points is increased the likelihood of

inadequately sampling the spatial variation of steeply dipping reflectors also

increases and results in the spatial aliasing of subsurface images. In this context

spatial aliasing refers to radar profiles in which the spatial variation in reflectors

is poorly defined due to station spacing of a greater interval than the variability

of the reflector. Flat or shallow dipping reflectors exhibit little variation in depth

over short distances, whereas steeply dipping reflectors are characterised by

sharp variation in depth over similar survey lengths. While the bedrock

boundary in this scenario does not dip greatly over a distance of two metres,

optimum station spacing is still required in order that profiles are visually

Page 113: Ground Penetrating Radar for the parameterisation of

Andrew Howe

113

interpretable.

Optimum station spacing is a function of antennae frequency and the dielectric

value of the host environment (Annan, 1997). The actual volume of subsurface

sampled at each point is a function of these parameters, and the depth of the

reflector, since the radar footprint can be thought of as a cone with an elliptical

base expanding from the radar system. The dimensions of the conical base are

given by equation 2.12. and tend to increase with depth and decrease with

increased dielectric values. (Annan and Cosway, 1992). In the standard radar

transmitter – receiver configuration the long axis of the cone is orientated

parallel to the survey line, and this is the configuration used for all data

collection in this thesis. A degree of overlap in sample volume between

successive sample points is required to reduce spatial aliasing. Figure 4.12

shows the theoretical dimensions of the long axis radar footprint for 900 MHz

and 450 MHz antennae for the range of dielectric values (K) likely to be

encountered in the field. Two reflectors are considered, located at depths of

0.2m and 1.0m respectively.

Figure 4.12. Long-axis footprint size for variable dielectric constant

Page 114: Ground Penetrating Radar for the parameterisation of

Andrew Howe

114

For a station spacing of 0.05m, overlap between adjacent traces occurs at 0.02m

depth for K equals 4 and 0.14m depth for K equals 20. These are very shallow

depths for the GPR system and subsequent analysis of profile data has shown

that the airwave and groundwave mask reflectors at these depths. Therefore a

station spacing of 0.05m was used for all 900MHz surveys. 450MHz surveys

were used in scenarios where it was considered likely that soil thickness

exceeded one metre and 900MHz surveys would be unable to penetrate. These

sites were located in the valley bottoms. At these sites 450MHz reflection

surveys were carried out with a station spacing of 0.1m.

4.6.4 Sampling Interval

A returning GPR trace consists of a continuous waveform. Sampling of this

waveform must be carried out at the appropriate temporal resolution in order

that the waveform can be reconstructed. Under-sampling results in a loss of data

and over-sampling leads to the collection of unnecessary data, slowing survey

time and increasing file size. The maximum sampling interval is described by

the Nyquist frequency and is one-half the period of the highest frequency. In

GPR systems the highest frequency is approximately 1.5 times the centre

frequency, therefore following the Nyquist theorem results in a sample rate

twice this, or three times the centre frequency. Annan (1997) suggests doubling

this sample rate to ensure that no GPR signal is under-sampled. Following this

guideline results in a maximum sampling interval of 0.185ns for a 900MHz

survey and 0.370ns for a 450MHz survey. For all GPR surveys carried out as

part of this thesis the sampling interval was 0.05ns, well above the minimum

Nyquist frequency.

Page 115: Ground Penetrating Radar for the parameterisation of

Andrew Howe

115

4.6.5 Common Midpoint survey design

A common midpoint survey was carried out at 29 of the 32 sampled sites.

CMP’s were required to measure the subsurface velocity of the electromagnetic

waves and thereby enable the calculation of reflector depth from travel time

data collected by GPR reflection surveys. In conjunction with gravimetric and

theta probe measurements of subsurface moisture, CMP’s were also used to

investigate the relationship between velocity and moisture content in a field

environment.

At each site, one CMP survey was collected along the same 2m line used for the

reflection survey. The antennae were arranged around the one metre centre

point and moved stepwise away from the centre. A Step size of 0.05m and

0.10m was chosen for surveys using the 900MHz and 450MHz antenna

respectively in order that the moveout be comparable with that used for the

reflection profiles. Each CMP was continued until either the transmitted signal

was not visible to the receiver for several traces, or antenna offset reached 2m.

4.6.6 Stacking

Stacking is the term used to define the number of times a single point

measurement is averaged to produce a final GPR trace. In electrically noisy

environments high stacking values reduce the effects of spurious background

interference since the averaging process smoothes the final output trace.

Increasing stacking to 128 measurements (the maximum allowed using the

PulseEKKO software) increases the time needed to measure at each point and

therefore slows the rate of data collection. For this study, stacking was set to 32

measurements as a compromise to ensure good data quality without appreciably

lengthening the time required to measure at each sample site.

Page 116: Ground Penetrating Radar for the parameterisation of

Andrew Howe

116

4.7 Validation of GPR subsurface data

Subsurface data is required for the validation of GPR derived soil thickness and

moisture, and needed for parameterisation of the subsurface component of the

hydrological model. There have been a number of studies which have used GPR

to measure soil thickness e.g. Shih et al., (1984), Collins et al., (1987), Truman

et al., (1988), Collins et al., (1989), and soil moisture e.g. Greaves et al.,

(1996), Hubbard et al., (1997) and van Overmeeren et al., (1997). The

techniques are however a current research issue. Physical data collection is

required to validate the results obtained by GPR and to assess the reliability and

accuracy of GPR for measurement of these parameters.

4.7.1 Soil thickness

Soil thickness was measured in the field using a combination of augering and

trial pits. The aim of physically sampling the subsurface at each site was to

identify any correlation between specific soil horizons and reflectors evident in

radar profiles. In order that soil horizons could be associated with specific

reflectors an auger was used to extract soil cores and provide a measurement of

total depth to bedrock. Auger samples were taken at each end of a two-metre

survey line and at the centre point if no trial pit was excavated. Trial pits were

excavated to allow more detailed examination of soil horizons over a greater

area, and were used to sample soil moisture, bulk density and soil texture.

Page 117: Ground Penetrating Radar for the parameterisation of

Andrew Howe

117

4.7.2 Soil Moisture

A Delta T theta probe was used to sample soil moisture at the surface and

subsurface for GPR sites. Measurements of subsurface moisture were carried

out for those sites with an excavated trial pit. Readings were taken at intervals

of 0.05m or 0.1m until insertion of the theta probe was impossible due to high

stone content or the bedrock surface was reached. In conjunction with theta

probe measurements, soil samples of a known volume were collected from the

same points for laboratory analysis of volumetric moisture content and

calibration of theta probe measurements.

Theta probes measure the soil dielectric constant and relate this to volumetric

soil moisture using a polynomial relationship (Theta probe User Manual, 1998).

Eight soil samples from four different profile depths were used to perform a soil

specific calibration for the theta probe. Table 4.6 summarises the measured

values of a0 and a1, required for derivation of volumetric soil moisture (θ) from

measured voltage (V) (equation 4.2).

Table 4.6. Theta probe calibration values.

Depth Number of soil samples a0 a1

0-10cm 2 1.32 8.5910-20cm 2 1.37 9.1320-30cm 2 1.41 8.6530-40cm 2 1.47 6.68Mean 1.39 8.26

(4.2)

( )1

032 7.44.64.607.1

aaVVV −+−+

Page 118: Ground Penetrating Radar for the parameterisation of

Andrew Howe

118

4.7.3 Soil physical properties

The two soil properties required as hydrological model parameters are bulk

density and texture. The variation in bulk density with depth is used by the

model to calculate soil porosity at any soil thickness, and subsequently to

calculate wetting front depth (equation 3.7) and water table depth (equation 3.9)

for each timestep. Soil texture (mass fraction sand and clay) is required as input

to the infiltration module (equation 3.3). The soil samples collected for analysis

of moisture content were also used for bulk density measurement and textural

analysis. A total of 55 bulk density measurements and 30 texture samples were

collected from trial pits excavated at sites within the catchment.

Analysis of field data shows an exponential decline in porosity with increasing

soil thickness, derived from 55 bulk density samples for 17 sites and at varying

depths across the catchment. Figure 4.13 shows the relationship between

average sample porosity and soil overburden thickness derived from field

measurements. The relationship between depth (x) and porosity (φ) from 55

field soil samples is

(4.2)

8286.0 848.0 2757.0 == − re xφ

Page 119: Ground Penetrating Radar for the parameterisation of

Andrew Howe

119

Figure 4.13. Depth – mean porosity relationship derived for the Cyff catchment

The total depth of soil needed to store a known amount of water between two

depths x1 and x2 is equivalent to the integral of equation 4.3 between the limits

x1 and x2.

(4.3)

In the case of infiltration from rainfall or upslope runoff, φ = infiltrated water

this timestep, and x1 = wetting front depth from the previous timestep. Solving

equation 4.3 and rearranging to make x2 the subject gives the new wetting front

depth for the current timestep.

dxex

x

x∫ −= 1

2

848.0 757.0φ

Page 120: Ground Penetrating Radar for the parameterisation of

Andrew Howe

120

4.8 Errors

Soil depth measurement used a steel tape to record the thickness of each soil

horizon core contained within the auger. A steel tape is unaffected by any

stretch which could occur if a plastic measuring tape were used. The tape used

was marked at millimetre intervals, allowing measurement of depth to ± 2mm.

Theta probe measurement error was minimised using the soil specific

calibration technique described in the Theta Probe User Manual (1998). Using

this method, soil moisture measurements are accurate to ± 0.02 m3/m3 over a

temperature range of 0-40oC (Theta Probe User Manual, 1998).

Measurement of wet and dry soil weight for the gravimetric method of soil

moisture content were carried out in the laboratory using a balance sensitive to

one hundredth of a gram. Using the mean of all eighty dry and wet sample

weights (Appendix III, table III.1), the effect of a measurement error of ± 0.01g

applied to wet and dry soil samples represents a soil moisture error of ± 0.001

m3/m3. More general problems associated with the gravimetric method concern

the oxidation of organic material during the drying process and the potential for

underestimating water contents, especially in clay soils where adsorbed water is

not fully evaporated during oven heating at 105oC (Ward & Robinson, 2000).

Page 121: Ground Penetrating Radar for the parameterisation of

Andrew Howe

121

4.9 Conclusion

A background to Plynlimon has been presented along with a justification for

choosing the Cyff as a study catchment. The hydrological parameters required

for model input and validation and the methodology used to collect the data at

scales appropriate to modelling requirements has been discussed. Topography is

an important factor in any hydrological study given that it is a control on

hydrological flow rates and pathways, and determines some of the spatial

variation of variables including soil moisture. Recreating an elevation surface as

a grid structure is problematic since the choice of cell resolution effects the

resulting terrain attribute values derived from the DEM.

The influence of topography on catchment hydrology and soils required the use

of slope, upslope area and wetness topographic indices for selection of sample

sites for subsurface investigation using GPR and physical sampling. Preliminary

investigation using frequencies of 225, 450 and 900MHz showed that GPR was

a viable survey tool for the subsurface in this area. Subsequent detailed

investigation of soil thickness and soil moisture spatial variation was undertaken

using 900MHz and 450MHz antennae for high resolution of subsurface

reflectors. Physical sampling of soil thickness, moisture, texture and bulk

density was carried out at sites to validate GPR measurements of site specific

depth and moisture, and to provide parameter values for the hydrological model.

Page 122: Ground Penetrating Radar for the parameterisation of

Andrew Howe

122

Chapter 5: Soil Thickness Data Analysis

5.1 Introduction

This chapter examines the methods used to derive soil thickness using radar.

The results obtained are compared with those measured by physical excavation

at each site using a soil auger and trial pits. Using the catchment DEM the

spatial variation in soil thickness is compared with a number of spatial

variables, notably local slope gradient, upslope drainage area and the

topographic wetness index. These variables are subsequently used to extrapolate

from individual site soil thickness to a catchment wide distributed map of soil

thickness variation, required as an input to the subsurface module of the

hydrological model.

5.2 Analysis Methodology

Figure 5.1 shows the processing methodology followed to analyse collected

GPR data for soil thickness measurement at each survey site. An initial radar

profile is visually displayed and simple display parameters are manipulated to

aid in identification of subsurface features, in this case the identification of soil

horizons and total soil thickness to bedrock. Visual examination of a profile is

helpful in subsequent analysis of individual radar trace characteristics. CMP

analysis is a key component of the processing stream given that accurate

derivation of the depth to a reflector relies on a detailed knowledge of the

pattern of subsurface velocities. Extraction of individual traces corresponding to

the location of physical samples and the application of the local velocity

distribution in a spreadsheet package enables a quantitative approach to

reflector characterisation. An observed correlation between reflections in a

series of individual traces and physical sampling of the soil enables a

measurement of the average depth to a reflector in a profile.

Page 123: Ground Penetrating Radar for the parameterisation of

Andrew Howe

123

Figure 5.1. Deriving soil thickness from GPR surveys.

GPR reflection profile

Visual display of original profile &hardcopy printout

ID potential reflectors of interestusing visual interpretation

Apply DEWOW/DCSHIFT &simple filters/gains

Data conversion(PE to ASCII)

CMP survey

Extract individual trace amplitudedata corresponding to sample

points

Import trace data tospreadsheet and apply

velocity field

Identify key reflectors in atrace which correspond to

soil features

Apply PE velocityanalysis module

Measure gradient ofgroundwave/reflectors to

derive depth-velocitystructure of the subsurface

Visual display of original CMP &hardcopy printout

Reconstruct travel time-velocitystructure of the subsurface

Apply to entire GPRprofile

Compare GPR resultswith auger measurements

of soil depth

Derive mean site soilthickness

Page 124: Ground Penetrating Radar for the parameterisation of

Andrew Howe

124

5.2.1 Initial GPR profile display

Visual display of each site profile was undertaken using the PulseEKKO

software. The software enables limited processing of a profile, without altering

the original amplitude data file. Initial data processing was confined to the

permanent dewowing of all GPR profiles. Wow is the term applied to a low

frequency component of the initial transmit pulse which decays slowly over

time. It is present to an extent in all GPR data and depends on ground conditions

and antennae separation. This low frequency component obscures the return

signal from reflectors but can be removed effectively by a high pass filter.

Historically this is known as dewowing (Sensors & Software Inc., 1993). The

high pass filter does, however, create a system artefact of one half-period cycle,

located before time-zero.

The second stage in display is gain recovery. Gains may take the form of a

constant multiplier to all raw amplitude data or a time-variant function. Time-

variant gains compensate for the increased spreading and attenuation of an

emitted pulse with increased travel time. As a result of these effects those

reflectors at longer travel times are of lower amplitude compared with those of

shorter travel times (Fisher et al., 1996). Two time variant functions available in

the PulseEKKO software are spreading and exponential compensation (SEC)

and automatic gain control (AGC). As an initial step all profiles were displayed

with a low constant gain of 5-10 in order to pick the highest amplitude reflectors

present in a profile. In many instances application of a constant gain clearly

showed linear reflectors within 0-30ns of the surface. In other cases a constant

gain points to zones of potential interest within a profile, usually at greater

travel times. Subsequent use of SEC or AGC gain to offset spreading and

attenuation effects yields an improved image of deeper subsurface structure.

Figure 5.2 shows the differing effect caused by the application of a constant

gain and a SEC gain to the same radar profile, site 20. The upper boundary

position of the B-horizon and C-horizon are marked in red and blue

respectively, physically identified by auger measurements.

Page 125: Ground Penetrating Radar for the parameterisation of

Andrew Howe

125

Figure 5.2. 900MHz GPR profile (site 20) showing the difference in detaildisplayed using a constant gain compared to a SEC gain. The depth axis valueassumes a single velocity value (0.06m/ns) applies to the entire profile.

Further improvements in image display are provided through trace averaging.

Two averaging tools were used to smooth profile images, acting between

adjacent traces (spatial) and down traces (temporal). All averaging acted on the

displayed image rather than the raw data and in the majority of cases averaging

was avoided. Spatial (trace to trace) averaging acts to smooth the spatial

variation in linear reflectors by averaging adjacent traces within a user defined

window. Temporal (down trace) averaging results in a smoothing of amplitude

data where the number of points included in the averaging operation is user

specified. Temporal averaging can lead to a reduction in high frequency noise,

which is a common feature in data from longer travel times. Image quality is

degraded by high frequency amplitude variation caused by signal degradation.

Application of a gain to data, for example application of an AGC gain to

equalise arrivals from all depths tend to show increased noise. In all instances

Page 126: Ground Penetrating Radar for the parameterisation of

Andrew Howe

126

where averaging was used the maximum number of adjacent traces/points used

were set to two.

Examination of each site profile is an essential first step in GPR data

processing, particularly for the identification of linear features which in this

study were potential soil horizon boundaries. Display of radar images is a

highly subjective method of interpretation as is the application of gains and

averaging operators, which can result in the creation of image artefacts. To

reduce this potential all gains have been kept as simple and as limited as was

possible to reduce the probability of artefact creation. Investigation of the

response of individual trace amplitude to varying soil horizons is the next stage

of a quantitative method to soil thickness measurement using GPR.

Page 127: Ground Penetrating Radar for the parameterisation of

Andrew Howe

127

5.2.2 Derivation of subsurface velocities

Given knowledge of the nature of subsurface velocity variation, travel time can

be recalculated as reflector depth. The following outlines the method used to

derive soil thickness from travel time data. Derivation of depth from any GPR

profile relies on the accurate estimation of the propagation velocity of the layer

which the EM wave passes through. The Common Midpoint (CMP) survey is

the standard method for deriving velocity profiles and therefore forms a key

requirement for depth estimation. The velocity analysis program is based on the

assumption that signal travel time from reflectors varies hyperbolically as the

separation between transmitter and receiver increases. A CMP survey was

carried out at each of 29 sites and analysed using Picker (Sensors & Software

Inc.). CMP surveys were imported into Picker, a software package that enabled

the increase in travel time (t) with increasing antennae separation (x) to be

measured and used to calculate velocity. The process was based on the normal

moveout procedure, in which a plot of increasing t against x shows a linear

change in airwave and groundwave arrival, while reflectors from subsurface

layers exhibit a hyperbolic shift in arrival time (figure 5.3). The inverse

gradients of reflectors plotted as t2 against x2 are equal to an average velocity

value from zero time (ground surface) to the reflector, termed a root mean

square velocity (vrms) (Reynolds, 1997) and calculated by

(5.1)

Figures 5.3 and 5.4 show the increase in airwave, groundwave and reflector

travel time with increasing travel time, both as a PulseEKKO CMP profile and

displayed within the Picker software enabling measurement of the change in

travel time with antennae moveout. All CMP derived velocities for sample sites

are listed in table II.4, Appendix II.

21

21

22

21

22

−−

=ttxxvrms

Page 128: Ground Penetrating Radar for the parameterisation of

Andrew Howe

128

Figure 5.3. CMP surveys for site 56 and site 50 using 900MHz antennae.

Figure 5.4. Initial 10 traces from site 50 plotted using Picker software. Airwavearrival time marked by red, ground wave arrival by green and first pickedreflector by blue lines.

Page 129: Ground Penetrating Radar for the parameterisation of

Andrew Howe

129

Actual layer velocity (Vint) can be estimated using the Dix equation (5.2), (Dix,

1955) which should provide a more accurate value for soil thickness and

potentially water content (Chapter 6). The Dix equation was applied to the CMP

velocities to produce a corresponding interval velocity. The interval velocity

was then used in all subsequent time to depth conversions of profiles.

(5.2)

Where Tn and Tn-1 are two way travel times to the nth and n-1 reflectors

respectively.

5.2.3 From time domain to depth domain

The derivation of accurate estimates of reflector depth from GPR requires that

two way travel time (T) is transformed to a value of depth using a velocity for

the layer(s) above the reflector in question. In most cases this conversion is

simply T halved and multiplied by the velocity v to give reflector depth d

(equation 5.3), which gives an adequate depth value for most GPR applications.

(5.3)

However due to the finite distance between the transmitter and receiver the path

taken by an emitted EM pulse is not vertical but forms an isosceles triangle with

the lowest vertex located on the reflector midway between the transmitter (Tx)

vTd ×

=

2

( ) ( )( )

21

1

12

1.2

.int

−−

=−

−−

nn

nnRMSnnRMS

TTTVTVV

Page 130: Ground Penetrating Radar for the parameterisation of

Andrew Howe

130

and receiver (Rx) (figure 5.5). Actual depth to a reflector (d) is therefore

somewhat less than half two-way travel time. As depth increases T tends to 2d

and the approximation of reflector depth equalling one way travel time is a valid

one. At shallow depths however this is not the case since T >> 2d. In this case

equation 5.3 produces apparent depths in excess of actual depth. In order to

compensate for this effect the PE1000 software module displays an initial non-

linear depth axis compared to the time axis.

This study of the use of GPR as a tool for evaluating subsurface soil structure

and hydrology requires manipulation and evaluation of data in a manner not

covered by the PulseEKKO software module. The majority of interpretation is

carried out in a spreadsheet package and one component requires the validation

of GPR derived soil thickness with soil auger depths. The auger data collected

from the Cyff fieldsites suggests that soil thickness falls in the range of 0.1 -

2.0m and therefore inclusion of a non linear time - depth conversion within the

spreadsheet is essential for the accurate positioning of GPR reflector depths and

for validation of these shallow reflectors with auger data.

Without an initial non-linear time-depth conversion shallow reflectors will

appear to be located at greater depths below the ground surface than they are in

reality. A good example is the ground wave itself, which travels directly

between Tx-Rx and therefore arrives at a time of antenna separation divided by

ground velocity. Plotted on a depth axis which increases linearly from time zero

places this reflector at a non-zero depth.

Page 131: Ground Penetrating Radar for the parameterisation of

Andrew Howe

131

Figure 5.5. Theoretical wave travel path for offset antennae.

The relationship between actual reflector depth (d), antennae separation (x), two

way travel time (T) and the layer velocity (v) is shown in figure 5.5. One way

travel time forms the hypotenuse of a right-angled triangle allowing the true

depth to reflector to be calculated as

(5.4)

Implementation of equation 5.4 in a spreadsheet is straightforward when

applying a single average velocity to all layers, as is the case for the

PulseEKKO software. For each increase in one way travel time the equation is

applied to calculate the depth. However if incremental depth values are

calculated from the start, the time increments are always small compared to

antenna offset distance, resulting in very small values for d or error values

caused by negative square root calculations.

21

22

22

×=

xvTd

Page 132: Ground Penetrating Radar for the parameterisation of

Andrew Howe

132

To solve this problem equation 5.4 was implemented for the first velocity layer

only while for deeper reflectors depth was calculated in increments using the

standard assumption that depth is equivalent to one way travel time (equation

5.3). This method can be justified by the decrease in importance of antenna

offset as depth increases. Neglecting the antennae offset effect results in the

largest errors in actual depth for shallow reflectors. Given that in the case of the

Cyff the initial velocity layer has a mean two way travel time of 12ns (T) for

900 MHz this method seems valid. Values of T in excess of 12ns produce little

difference in calculated reflector depth value between the two methods of

calculation (equation 5.3 & 5.4). The difference in reflector depth between

equations 5.3 and 5.4 is 0.08m. Table 5.1 shows comparisons between depth

estimates for the 900MHz (offset 0.17m) and assuming an initial layer velocity

of 0.06m/ns (the average for the data set). Figure 5.6 shows the predicted depths

calculated using standard depth estimates (equation 5.3) and incorporating

antennae offset (equation 5.4).

Table 5.1. Reflector depth using linear/non-linear methods of calculation.

FrequencyMHz

Two WayTime (ns)

Depth = Equation 5.4 (m) Depth = Equation 5.3 (m)

900 1.0 0.030 0.000

900 2.5 0.075 0.000

900 5.0 0.150 0.124

900 10.0 0.300 0.290

900 15.0 0.450 0.442

900 20.0 0.600 0.594

Page 133: Ground Penetrating Radar for the parameterisation of

Andrew Howe

133

Figure 5.6. Comparison between linear and non-linear depth calculation.(900MHz, v=0.06m/ns, x=0.17m)

A second concern is the calculation of total travel time. Within the PE software,

time zero is taken as the time at which the airwave arrives at the receiver and all

subsequent arrival times are zeroed from this point. Ground zero is taken as the

time taken for a direct wave to travel from Tx - Rx minus airwave travel time.

However due to the finite offset of Tx Rx the airwave does not arrive

instantaneously at Rx but takes a short period of time to travel between the two

antennae. This is calculated as

(5.5)

Where x is antenna separation and vAIR is airwave velocity (0.3m/ns).

This is the time at which the initial pulse was transmitted and should therefore

be added to all time values to give a travel time from pulse transmission rather

than from when the airwave arrives. Again when considering reflectors at

depths greater than 0.2 - 0.3m this extra time is insignificant considering the

unknown accuracy of velocity estimates for the subsurface, but at very early

times (and shallow reflectors) this can be a significant time period.

AIRvxt =

Page 134: Ground Penetrating Radar for the parameterisation of

Andrew Howe

134

Accurate depth measurements from GPR traces require that ground zero is

located in the correct position. Theoretically a PulseEKKO wavelet is 1.5

cycles. The first signal arrival is the airwave followed by the ground wave and it

is possible to pick these individual arrivals when antennae offsets are large as is

the case during CMP surveys. In reality at small antennae offsets these two

arrivals overlap causing interference patterns which causes problems when

attempting to visually pick ground zero with any accuracy. In this case

knowledge of ground surface velocity (vg) and antenna separation (x) could be

used to calculate theoretical ground zero. Ground-zero can be calculated as

(5.6)

This is the time from initial signal transmission, but it is more common to fix

zero time as the time at which the airwave arrives rather than actual signal

transmission. In these terms the time of ground wave arrival (t) after time-zero

(airwave arrival) is given by

(5.7)

Since time-zero can be identified from a graph of signal response as the first

zero crossing, ground-zero can now be located assuming that the ground

velocity value is correct. Within a spreadsheet model this allows the position of

the ground surface to be located and subsequent calculations for successive

reflector depths calculated using equations 5.3 and 5.4.

ggzero v

xt =

0ttt gzero −=

Page 135: Ground Penetrating Radar for the parameterisation of

Andrew Howe

135

5.2.4 Direct wave masking of shallow reflectors

The direct wave, a combination of the air and ground wave, acts to mask very

shallow reflections from the subsurface. This effect is termed transmitter

blanking (Annan, 1997) and results from the requirement for two events to be a

minimum of one half of the envelope width of the electromagnetic pulse in

order for these events to be resolved as separate reflectors. In terms of the

minimum path length that can be resolved this can be calculated using

knowledge of antennae frequency and direct wave travel time. The minimum

depth at which a reflector can be distinguished from the direct wave is

important in this instance because of the shallow depths of organic horizons

across the catchment.

For a 900MHz survey with antennae separation (x) of 0.17m and a ground

velocity (v) of 0.06m/ns the theoretical direct wave travel time is 2.833ns from

transmitter to receiver. GPR systems are designed such that the centre frequency

is inversely proportional to the pulse period (Reynolds, 1997), therefore a

900MHz antennae has a pulse period of 1.11ns. The output pulse envelope from

a PE1000 system is 1.5 cycle waveform and therefore in this scenario total pulse

duration (D) is a determined by pulse period and pulse envelope length. Annan

(1997) states that two events can only be distinguished as separate reflectors if

they are separated in time by a difference of half the envelope width, in which

case the minimum travel time (tmin) at which an event can be resolved is given

by equation 5.8.

(5.8)

tmin = Direct wave travel time + Pulse envelope + One half pulse envelope

DDvxt

43

23 min ++=

Page 136: Ground Penetrating Radar for the parameterisation of

Andrew Howe

136

Minimum travel time is more useful when converted to a minimum depth

estimate. Substituting tmin as T into equation 5.8 calculates the minimum vertical

depth to a reflector. Reflectors occurring at depths less than this value will be

indistinguishable from the direct wave return. Table 5.2 summarises the

minimum reflector depths calculated for 900 and 450MHz surveys.

Table 5.2. Calculation of minimum resolution depth using an average velocity of0.06m/ns, derived from 29 CMP ground wave velocities.

900MHz 450MHz

Antennae separation (m) 0.17 0.25

Pulse period (n/s) 1.11 2.22

Direct wave travel time (n/s) 2.83 4.17

tmin 5.33 9.17

Minimum depth (m) 0.14 0.24

5.2.5 GPR depth measurement errors

Error in GPR depth measurement is a result of a) error in the measurement of

subsurface velocities, and b) error in the measurement of signal travel time. The

magnitude of these errors is calculated using standard error methods which

enable the combination of errors in velocity (m/ns) and travel time (ns)

measurement to produce a single reflector depth error.

In order to quantify errors from measurements with different units, the precision

(Py) of each measurement (y) is calculated, where ∆y is the error associated with

the measurement (equation 5.9).

(5.9)y

ypy∆

=100

Page 137: Ground Penetrating Radar for the parameterisation of

Andrew Howe

137

Combination of the precision (Px) of different measurements is found by

(5.10)

Using this final precision and reflection depth (d) in metres, the total error (A)

can be calculated as

(5.11)

Error in velocity measurement can be caused by a) incorrect measurement of

travel time to the reflector with increasing antennae separation, and b) errors in

the positioning of GPR antennae along the CMP line. Positioning error is a

result of the inaccurate placement of GPR antennae along the CMP line. For this

study each CMP was carried out along a two-metre line on uneven ground using

a plastic tape with half-centimetre interval markings. The error between antenna

location for each measurement is estimated to be ±0.02m

Sources of error associated with measurement of the change in travel time to a

reflector include error in picking reflections, with Picker software able to

resolve to 0.01ns. However a greater limiting factor is the GPR sampling

interval which was set at 0.05ns for all surveys in this study. The exact point of

time zero for each trace, identified using Picker, is conservatively estimated as

accurate to 0.5ns. No filters and only constant gains were applied within the

Picker software to ensure reflector travel times were not shifted.

Using the above values of error in time and distance in equations 5.9 and 5.10

the error in velocity measurement was calculated, using a survey line length of

1m and an average velocity of 0.06m/ns, calculated from all non-air wave

velocities (table II.4, appendix II ). The calculated precision is 4.4% which

equates to a velocity error of 0.003m/ns.

22zyx ppP +=

100xPdA ×

=

Page 138: Ground Penetrating Radar for the parameterisation of

Andrew Howe

138

The same method was used to calculate the error in GPR depth measurement,

using the previously calculated velocity precision and combining with the

calculated error in travel time measurement with a travel time of 60ns used in

equation 5.10. Using these error values for velocity and travel time results in a

precision of 4.5%. Table 5.3 shows the absolute error calculated using equation

5.11 for reflectors at varying depths.

Table 5.3. Error associated with increasing depth to a reflection, assuming aconstant velocity of 0.06m/ns and two-way travel time of 60ns.

Reflector depth (m) Absolute error (m)

0.20 0.009

0.50 0.022

0.75 0.034

1.00

1.50

0.045

0.067

5.2.6 Theoretical resolution of reflectors

For GPR systems the maximum theoretical resolution r is one quarter pulse

wavelength (Reynolds, 1997).

(5.12)

The average of velocities calculated from analysis of site CMP’s is 0.06m/ns.

Using this value for v in equation 5.12 predicts a maximum resolution of 0.02m

for 900MHz antennae and 0.03m for 450MHz antennae.

fvr

4=

Page 139: Ground Penetrating Radar for the parameterisation of

Andrew Howe

139

5.3 Analysis of individual GPR traces for depth extraction

Application of the velocity data derived from a site specific CMP and

incorporation of the initial non-linear time to depth conversion enables the

response of signal amplitude with changing depth to be derived for any trace. In

the Cyff catchment the observed, visually well defined, changes in soil horizon

and associated changes in soil properties (e.g. bulk density, porosity, texture and

stone content) were thought to result in the formation of a soil profile with

defined dielectric boundaries. The presence of dielectric boundaries is evident

from standard GPR profile data by the large number of reflectors typically

shown on these profiles. However the hypothesised correlation between specific

soil horizon boundaries and zones of high/low amplitude signal requires a more

detailed analysis of the profile data. To achieve this, GPR traces along each

radar profile corresponding to soil auger and trial pit points were used to

examine the correlation of minimum and maximum amplitude values with the

physical measurement of soil horizon thickness.

5.3.1 Reflector identification

The PulseEKKO transmit pulse takes the form of 1.5 cycle waveform.

Subsequent reflections retain this waveform and therefore the location of

subsurface reflectors are identifiable by the presence of this shape within a

trace. The situation is complicated by factors including: 1) closely spaced

reflectors compared with signal wavelengths which create multiple returns

which cannot be individually recognised as specific reflectors. 2) amplitude

decay due to spreading and attenuation losses which leads to problems

identifying deeper reflectors against background noise of a similar amplitude

level. Applying a time variant SEC or AGC gain can compensate for this second

problem to an extent, but both require a knowledge of ground attenuation

characteristics and an arbitrary choice of maximum and minimum gain levels.

Applying a constant gain to trace amplitude data multiplies all data values by a

Page 140: Ground Penetrating Radar for the parameterisation of

Andrew Howe

140

defined amount and therefore acts indiscriminately on actual reflectors and

background noise. An alternative approach used by this study applies a power

gain to amplitude data, whereby each raw data value is raised to a user-defined

power, resulting in the suppression of low amplitude values and enhancement of

higher amplitudes. This approach assumes that low amplitudes are generally

noise and that the suppression of these parts of the trace enables easier visual

identification of potential reflectors. The difference between application of a

constant and power gain is shown in figure 5.7 using a GPR trace from site 40.

Figure 5.7. Application of a constant gain of 2 (a) and a power gain of 2 (b) forsite 40, 0m.

Page 141: Ground Penetrating Radar for the parameterisation of

Andrew Howe

141

5.3.2 Picking horizon position within a GPR trace

For individual trace data the position of a horizon boundary within a profile was

recorded as the first zero crossing preceding a positive peak of magnitude

greater than the preceding maximum. In general at each site, two auger samples

and one excavated pit at a spacing of one metre were compared with three GPR

traces at equivalent positions. For a number of sites located on peat deposits the

horizontal sample frequency was increased to an auger sample every 0.5m

primarily because the low bulk density of the subsurface at these sites enabled

the auger to be inserted to bedrock by hand.

The process of picking reflectors for each trace emphasised the differences

between the hillslope soils, and the peat soils of the valley bottom and ‘rush

flushes’. Rudeforth (1970) also makes a distinction between soils of the

Hiraethog series and undifferentiated peat complexes on the basis of soil

horizon properties. From the results of this study both categories exhibit

physical differences in vegetation types, soil physical characteristics and GPR

properties. Hillslope soils were further subdivided into those that were ‘simple’

and those that were ‘complex’ environments in terms of GPR profile

interpretation. These three subsurface classes are explained below.

5.3.3 Peat soils (Class 1)

The toe slopes of the Cyff are dominated by raw peats (histosols) which grade

into humic gleysols on the crest-slope shoulders and areas immediately above

the toe-slopes (Chappell & Ternan, 1993). These peat deposits were also found

to extend upslope in side valleys, features known as ‘rush flushes’ that act as

drainage routes for the adjacent hillslopes. Seven out of a total of thirty GPR

sites have been classified as belonging to class 1 subsurface environments on

the basis of GPR profile and trace characteristics discussed below. Physical

sampling of the soils at 18 of these sites show the differences in bulk density

Page 142: Ground Penetrating Radar for the parameterisation of

Andrew Howe

142

(figure 5.8) between peat sites and those sites classified as hillslope soils (Class

2 & Class 3). Assuming a constant soil particle density, bulk density was used

to derive soil porosity. Given the importance of water on the dielectric

properties of materials and the propagation velocity of an electromagnetic wave

through a medium, soil porosity is of interest when conducting GPR site

investigation.

Figure 5.8. Dry bulk density measured at 18 sites within the Cyff catchment,depth measured using a steel ruler.

In terms of GPR profile interpretation, peat soils provided an excellent medium

for the use of radar as a method for deriving soil thickness to bedrock. This is

due to the absence of multiple GPR reflectors within the soil profiles, rather, the

highest amplitude after the direct wave arrival correlates well with the measured

bedrock depth using a soil auger. An example radar profile from site 33 and a

single trace extracted from the 1m along line position is shown in figure 5.9.

The profile exhibits characteristics common to all GPR profiles collected from

peat sites in the catchment.

Page 143: Ground Penetrating Radar for the parameterisation of

Andrew Howe

143

Figure 5.9. GPR traces: - Peat soils. a) 2m GPR profile site 33, b) Raw amplitudeextracted from trace at position 1.00m, c) Amplitude squared at trace position1.00m.

Page 144: Ground Penetrating Radar for the parameterisation of

Andrew Howe

144

GPR profile and trace features characteristic of the peat soils found in the Cyff

are listed:

1. GPR profiles of peat soils show shallow depth horizontal banding,

effectively blanking any returns from reflectors up to 15-20ns deep (two

way travel time). This phenomenon is a result of the arrival of the high

amplitude direct air and ground waves during this time interval.

2. Subsequent high amplitude signals are predominantly caused by the

reflection of the electromagnetic wave from the boundary between the

organic peat deposits and the C-horizon. Auger samples confirmed that the

underlying material at this boundary was either solid rock or rock fragments,

given the underlying geology noted by previous soil surveys (Rudeforth,

1970) and field observation of bedrock outcrops, this material is slate and/or

shales.

3. Analysis of extracted GPR traces found that after omission of the direct

wave from each trace, the zero crossing that precedes the greatest or second

greatest peak amplitude correlates well with the measured auger depth to the

C-horizon. Often the greatest and second greatest amplitude after the direct

wave form the same PulseEKKO waveform, but no set pattern is identifiable

from these data sets as to which amplitude arrives first. This rule is

applicable when traces have had no gain or a constant gain applied to the

profile.

4. GPR profiles of peat soils were collected using both 450MHz and 900MHz

antennae. In terms of radar profiles, the lower frequency generally provided

a visually clearer indication of the soil - bedrock interface (figure 5.10.).

Examination of individual traces taken using both frequencies at the same

location (figure 5.11.) confirms that, for each frequency, the highest or

second highest positive amplitude value after the direct wave corresponds to

the bedrock boundary position. In addition the absolute values associated

Page 145: Ground Penetrating Radar for the parameterisation of

Andrew Howe

145

with the maximum amplitude in each trace are dependent on the frequency

used. In this example the 450MHz trace has a peak positive amplitude of

4087 compared with a positive peak value of 668 for 900MHz. This trend is

continued throughout the profile and for other sites located on peat soils. In

general peak positive amplitudes recorded using 450MHz antennae are 5-12

times larger than peak amplitudes recorded for the same position using

900MHz antennae (raw amplitude data). The resulting effect is one of

increased magnitude reflectors that are visually evident within a

profile/trace. This effect is likely to be a direct result of the increased

attenuation of electromagnetic waves at higher frequencies thereby resulting

in a reduction in the recorded amplitude at dielectric boundaries.

Page 146: Ground Penetrating Radar for the parameterisation of

Andrew Howe

146

Figure 5.10. Site 37 profile using 450MHz and 900MHz antennae.

Page 147: Ground Penetrating Radar for the parameterisation of

Andrew Howe

147

Figure 5.11. Trace amplitude comparison for site 37 bedrock boundary (O/Chorizon) for 900 MHz and 450 MHz GPR trace at the same location.

Page 148: Ground Penetrating Radar for the parameterisation of

Andrew Howe

148

5.3.4 Simple hillslope soils (Class 2)

57% of sites investigated (17 of 30) were classified as simple hillslope soils.

Excavation of the subsurface at these sites showed the existence of well-defined

soil horizons, generally consisting of a 0.1-0.2m thick dark brown/black organic

(O) horizon, a grey/white Ea horizon overlaying a yellow-red-brown B horizon

of high stone content. The middle and upper-slope positions which these sites

occupy are typically underlain by several metres of soliflucted regolith (Watson,

1967), and bedrock consisting of shales and mudstones (Rudeforth, 1970).

GPR profiles of these sites exhibit a more intricate subsurface image with a

greater number of reflectors than are present in the class 1 peat soils. The

presence of defined horizons within the soil profile should result in

corresponding dielectric boundaries being evident in GPR data. The assumption

is that the observed variation in soil physical properties, most importantly

moisture content, but also texture, results in a response at the frequencies used

by GPR for this thesis. Given that three distinct horizons are evident from

excavated soil profiles these three horizon boundaries should be distinguishable

within GPR data. Actual GPR profiles taken at these class 2 sites have been

found to contain a greater number of reflectors, reflecting the complexity of the

subsurface and the influence of stones within the soil profile on radar traces.

Analysis of soil thickness data for class 2, simple hillslope sites, showed that the

B and C-horizon boundary depths derived from GPR traces correspond well

with auger and trial pit measurements. However the majority of sites exhibit

shallow depth organic horizons (Organic <0.20m for 88% of class 2 sites)

resulting in no possible measurement of O horizon depth using GPR due to the

interference of the direct wave arrival at these time windows.

Figure 5.12 shows a profile from site 8, a class 2 site with a mean depth to C-

horizon of 0.43m from GPR and 0.47m from auger measurements. Figure 5.13

details traces at zero, one and two metres along the transect line which

Page 149: Ground Penetrating Radar for the parameterisation of

Andrew Howe

149

correspond to auger sample points. At site 8, the C-horizon boundary defined by

auger data corresponded with the zero-crossing point prior to the greatest

positive amplitude after a depth of 0.30m. Likewise the A-B soil horizon

boundary corresponded with the zero-crossing point prior to the peak positive

amplitude value and occurring after the direct wave, in a depth window between

0.25 and 0.35m. The actual points picked as GPR depth estimates are marked by

blue (B-C boundary) and red (A-B boundary) vertical lines. Table 5.4

summarises the difference in measured depth between GPR and auger data. This

analysis was carried out for each of the 30 sites investigated.

Table 5.4. Soil thickness measurements for site 8 using an auger and 900MHzGPR survey.

Position A-B auger(m)

A-B GPR(m)

Difference(m)

B-C auger(m)

B-C GPR(m)

Difference(m)

0m 0.30 0.30 0.00 0.45 0.39 -0.06

1m 0.30 0.28 -0.02 0.50 0.43 -0.07

2m 0.26 0.33 0.07 0.46 0.47 0.01

Site Mean 0.29 0.30 0.47 0.43

GPR depth error was calculated using equations 5.9 –5.11 and the site mean depths to each horizon. The

resultant errors are 0.013m for the A-B horizon and 0.021m for the B-C horizon, assuming a constant velocity

of 0.06m/ns.

Page 150: Ground Penetrating Radar for the parameterisation of

Andrew Howe

150

Figure 5.12. GPR profile: - Class 2 hillslope soils. (Site 8, 900MHz antennae,constant gain 10).

Page 151: Ground Penetrating Radar for the parameterisation of

Andrew Howe

151

Figure 5.13. GPR trace amplitude characteristics for site 8, 900MHz GPR survey.

Page 152: Ground Penetrating Radar for the parameterisation of

Andrew Howe

152

5.3.5 Complex hillslope soils (Class 3)

Class 3 sites composed six out of the total of thirty sites investigated (20%).

Complex hillslope soil sites were found to commonly lack coherent linear

reflectors within the GPR profile, or contained individual traces that correlated

poorly with auger observations of soil horizon position or total soil thickness.

Site 54 is an example of the latter scenario, in which four auger measurements

along the 2m-survey line varied in soil thickness to bedrock from 0.35m to

0.84m. A GPR profile obtained along the same line showed a linear boundary at

a depth of approximately 0.5-0.6m throughout the profile (figure 5.14). The site

was subsequently excavated to a depth of 0.6m along the 2m-survey line. At

this depth the B-horizon graded into a deposit of angular rock fragments

forming the upper C-horizon. The non-cohesive and brittle nature of the

underlying rock fragments had allowed auger penetration into the C-horizon

materials to a depth of 0.3m at points along the survey line, until encountering

larger rock fragments. Figure 5.15 shows the range of fragment sizes removed

from between 0.5-0.6m depth.

Page 153: Ground Penetrating Radar for the parameterisation of

Andrew Howe

153

Figure 5.14. GPR profile: - Class 3 hillslope soils. (Site 54).

Page 154: Ground Penetrating Radar for the parameterisation of

Andrew Howe

154

Figure 5.15. Stone fragments located at 0.5 - 0.6m depth.

Post processing of radar traces (DEWOW) and extraction of raw data for

amplitude – depth investigation provided an improved picture of subsurface

structure. The four trace amplitude signatures corresponding to auger samples

taken at 0, 0.5, 0.75 and 2 metres along the survey line are shown in figure 5.16.

Page 155: Ground Penetrating Radar for the parameterisation of

Andrew Howe

155

Figure 5.16. Radar traces corresponding to the location of four auger points

Page 156: Ground Penetrating Radar for the parameterisation of

Andrew Howe

156

In summary, GPR trace information from class 3 sites can be used to derive

depth to horizon measurements, but these sites generally require more extensive

physical excavation in order to positively identify the nature of specific

reflectors with a degree of certainty.

5.4 Soil thickness measurement using GPR

From the 30 sites investigated using GPR and augering, 111 physical

measurements of the depth to C-horizon and 46 measurements of B-horizon

depth were compared with GPR traces at the corresponding point. Table 5.5

summarises the mean GPR and auger measurements of B and C-horizon depth

on a site by site basis. The full table of individual auger and GPR measurements

is detailed in Appendix II, table II.5. Figures 5.17 – 5.19 show the relationship

between the depth to a specific horizon for individual measurements and as a

function of the mean horizon depth at each site.

Page 157: Ground Penetrating Radar for the parameterisation of

Andrew Howe

157

Table 5.5. Summary soil thickness data for all sample sites

Mean site depth (m) to B-horizon Mean site depth (m) to C-horizonSite ID Class Auger GPR Difference Auger GPR Difference

5 2 0.25 0.26 -0.01 0.5 0.43 0.076 2 0.17 0.18 -0.01 0.26 0.27 -0.018 2 0.29 0.3 -0.01 0.47 0.43 0.049 1 0.94 0.97 -0.0317 3 0.16 0.17 -0.01 0.93 0.85 0.0818 2 0.47 0.45 0.02 0.71 0.67 0.0419 2 0.26 0.23 0.03 0.42 0.42 0.0020 2 0.19 0.2 -0.01 0.47 0.44 0.0321 3 0.46 0.45 0.01 0.78 0.75 0.0322 2 0.23 0.23 0.00 0.66 0.62 0.0427 2 0.36 0.43 -0.0732 2 0.18 0.21 -0.03 0.3 0.34 -0.0437 1 0.85 0.85 0.0039 1 0.82 0.94 -0.1241 1 0.83 0.89 -0.0649 2 0.27 0.25 0.0250 2 0.29 0.34 -0.0552 2 0.22 0.26 -0.0454 3 0.24 0.25 -0.01 0.56 0.44 0.1260 2 no data 0.2913 3 0.25 0.25 0.00 0.71 0.62 0.0914 3 0.25 0.24 0.01 0.68 0.44 0.2415 2 0.14 0.2 -0.06 0.71 0.73 -0.0216 2 0.43 0.44 -0.01 0.48 0.44 0.0423 2 0.30 0.29 0.0124 2 0.37 0.29 0.0833 1 0.64 0.61 0.0340 1 no data 1.4448 3 0.59 0.52 0.07 0.85 0.7 0.1553 2 0.31 0.36 -0.05 0.62 0.5 0.1256 1 1.00 1.04 -0.04

The individual measurements of soil thickness by auger and GPR methods

(appendix II, table II.5) were used to calculate the RMS of the difference

between the two methods of depth measurement to the B and C soil horizon.

The results show a RMS of 0.04m to the B horizon and 0.12m to the C horizon.

Page 158: Ground Penetrating Radar for the parameterisation of

Andrew Howe

158

Figure 5.17. Soil auger measurements plotted against GPR derived soil thicknessto the B-horizon.

Figure 5.18. Soil auger measurements plotted against GPR derived soil thicknessto the C-horizon.

Page 159: Ground Penetrating Radar for the parameterisation of

Andrew Howe

159

Figure 5.19. Average soil auger depth for each site plotted against GPR averagesite soil thickness to the B-horizon and C-horizon.

Inspection of figures 5.17 – 5.19 shows no evidence of a trend away from the

1:1 line, indicating that soil auger and GPR measured depths offer equivalent

measurement results. The GPR method of depth measurement does however

require at least one physical measurement of depth at each site to ensure that the

reflector picked corresponds to the correct soil horizon. Once identified, GPR

can then be used to rapidly quantify the spatial variation of soil horizon depth.

Physical measurements of depth to soil horizons were taken in the field to

enable comparison between radar derived depth and actual depth. Derivation of

reflector depth required the application of subsurface velocities to radar data,

with velocities calculated from a site CMP survey. Because CMP velocities

were used to calculate radar reflector depth, physical measurements can be used

as an independent check on soil horizon depth. The assumption of independence

is valid since although auger measurements were used to identify specific

horizons in a GPR trace, the radar depth axis were not altered, being defined by

the velocity values measured and evaluated from the site CMP.

Page 160: Ground Penetrating Radar for the parameterisation of

Andrew Howe

160

Using a single auger core at each survey line, a GPR reflector within a single

trace was identified and flagged as corresponding to a soil horizon interface.

Following this reflector through the GPR profile of the survey line allowed the

variation in reflector depth to be quantified and compared with other physical

measurements of depth along the survey section. The results of the measured

variation in GPR and auger depths at the same point are shown in figure 5.20. A

regression line has been added to show that no systematic trend was found in

the variation of GPR measurements compared to physical measurements of

horizon depth. In this study GPR measured depth was as likely to under record

as over record soil depth at any location.

Figure 5.20. The variation between physically measured and GPR derived soilthickness, where physical measurements are considered to be true depth.

Using the data from all sites displayed in figure 5.20 the root mean square

between auger and GPR measurement of the C-horizon depth was 0.12m and

Page 161: Ground Penetrating Radar for the parameterisation of

Andrew Howe

161

0.04m for depth to B-horizon. The increase in GPR measurement variability

with increasing depth of measurement is potentially a result of two factors (a)

the use of higher frequency antennae for shallow depth measurement and

therefore an improved resolution of reflector position and (b) the application of

an incorrect subsurface velocity having a cumulative effect on reflector depth,

increasing the error in position as travel time (depth) increases.

Three outlying residuals are evident in figure 5.20. with values of 0.43, 0.35 and

−0.48m The two positive values are associated with two separate type 1 sites

(peat soils). In each case the remaining nine error values at each site are less

than 49% of the maximum residual value, indicating that the waveform

identified as the bedrock boundary is incorrect for that position only. The

benefit of a survey technique which samples a volume of subsurface rather than

a single point measurement is evident. The error of -0.48m is associated with a

type 3 site (complex hillslope soil). The remaining two auger-GPR comparison

measurements for this site give values of 0.01m and -0.25m. Overall the mean

absolute error between auger and GPR depth is 0.24m, the largest error value of

the 30 sites investigated.

Page 162: Ground Penetrating Radar for the parameterisation of

Andrew Howe

162

5.5 The spatial variation of soil thickness

An objective of this study is to investigate the use of GPR to parameterise and

validate a fully distributed hydrological model of which the subsurface is a key

component. Parameterisation has concentrated on the use of GPR to gather soil

thickness measurements from a sample of locations within the catchment with

different topographic and therefore hydrological and pedological characteristics.

Hillslope sediments have a degree of order associated with their spatial

distribution, for example the progressive increase in soil thickness with position

downslope (Daniels & Hammer, 1992). The relationship between topography,

hydrology and soil characteristics is useful since it enables a relationship

between sampled locations and topographic attributes to predict the spatial

variability of a soil characteristic for an entire catchment, over which it is

infeasible to undertake a GPR survey. The method discussed in chapter 5

proposed that topography is an important controlling factor on soil development

and as a result topographic attributes derived from the catchment DEM could be

used as surrogate indicators for soil thickness. The key topographic attributes

considered in this study are slope, upslope area and wetness index (figure 5.21 –

5.27). A number of other topographic attributes are obtainable from the

catchment DEM and have also been compared with the variation in soil

thickness over the catchment, namely altitude, aspect, plan curvature and profile

curvature. Plan curvature provides an indication of converging/diverging flow,

profile curvature a measure of the rate of change of slope and therefore

indication of flow acceleration and the likelihood of deposition/ erosion

(Burrough & McDonnell, 1998). Both plan and profile curvatures are therefore

expected to influence soil development over long time periods and water

movement over much shorter time-scales. A relationship between one of these

variables and GPR derived soil thickness would allow the prediction of soil

thickness for each cell within the catchment.

In order to compare cell topographic attributes with soil thickness it is necessary

to first obtain an average depth from each profile. Two methods of trace

Page 163: Ground Penetrating Radar for the parameterisation of

Andrew Howe

163

averaging were considered. The first, averaging within the radar software and

producing an averaged trace for each profile is computationally simple to

implement and produces a single trace which can then be exported to a

spreadsheet package and subsequently reflector depths picked. However while

this method may work well in sites with well-defined linear reflectors it does

less well in noisy complex environments as are characteristic of this area. If

reflectors are coherent i.e. they appear at the same depth then this method will

work well, but a shift in reflector depth or the addition of other reflectors leads

to destructive interference patterns and degradation in trace quality. In particular

the pulse and a half wave pattern that characterises a reflector is often lost

during this process, making interpretation of specific horizon depths very

difficult. This method creates an effective depth value that may not relate to any

actual observed depths. The second approach and the one used for this study

uses individual trace analysis, in this case traces which corresponded to augur

sample points, to identify reflector depth. Depths derived using this method

were then averaged to give a mean profile value. The advantage of this method

is the clarity of single rather than averaged reflection wavelets and use of augur

data to verify horizon position. Soil thickness derived using GPR and calculated

using the latter method were compared against values of slope, upslope area,

wetness, plan curvature and profile curvature for each sample site. The

relationship between these attributes and soil thickness is summarised in table

5.6.

Page 164: Ground Penetrating Radar for the parameterisation of

Andrew Howe

164

Table 5.6. Correlation coefficients of soil thickness variation with respect to DEMderived terrain attributes.

25m DEMderived

Terrain Attribute

Correlation coefficient (r) with GPR measured depth

A/B Horizon B/C Horizon

Number of samples 15 31

Plan Curvature 0.303 0.454

Profile Curvature -0.411 -0.190

Elevation -0.554 -0.324

Aspect -0.166 -0.229

Slope 0.156 -0.306

LN (Upslope Area) 0.392 0.831

Wetness Index 0.330 0.855

As a result of the poor correlation between depth to B-horizon and terrain

indices the subsequent use of these indices as a predictor for catchment wide B-

horizon thickness was not attempted. The lack of a good correlation may be a

result of the scarcity of collected data, and reflects the complex nature and

inherent heterogeneity of B-horizon depth in the Cyff. What can be concluded

from these data sets is that for this catchment the B-horizon is an intermittent

feature, absent in class 1 peat sites. The subsequent discussion of results deals

only with total soil thickness, defined as the depth measured from the ground

surface to the boundary of the C-horizon with bedrock.

Page 165: Ground Penetrating Radar for the parameterisation of

Andrew Howe

165

5.5.1 Plan Curvature

Total soil thickness to C-horizon shows a positive correlation with plan

curvature. 11 sites have a negative plan curvature value (indicative of possible

flow divergence) with a soil thickness range between 0.27m and 0.73m. Sites

with positive plan curvature (convergent zones) exhibit a soil thickness range of

0.26m to 1.44m. The relationship between erosion and deposition acting on

those sites located in convergent grid cells may explain the high degree of

variability in soil thickness with calculated plan curvature. However the impact

of local slope angle and upslope contributing area is also likely to have a strong

influence on soil thickness distribution.

Figure 5.21. Plan curvature.

Page 166: Ground Penetrating Radar for the parameterisation of

Andrew Howe

166

5.5.2 Profile Curvature

Concave sites (negative curvature, 21 sites) exhibit soil thickness ranging from

0.26m to 1.44m. Convex sites (positive curvature; 10 sites) exhibit a reduced

soil thickness, ranging from 0.25m to 0.97m. The overall trend is a slightly

negative, but the degree of scatter evident in the plot and corresponding low

correlation coefficient of –0.190 suggests that profile curvature is a poor

indicator of soil thickness distribution for the data sets.

Figure 5.22. Profile curvature.

Page 167: Ground Penetrating Radar for the parameterisation of

Andrew Howe

167

5.5.3 Slope Angle

Figure 5.23 shows that soil thickness to C-horizon generally decreases with

increased slope angle for the catchment, however shallow soil thickness occur

across the entire range of slope angles. Soil thickness could reasonably be

expected to decrease with steeper slope angles given the likelihood of increased

instability and potential for erosion of soils on steeper hillslopes.

Figure 5.23. Slope angle.

The greatest range of measured soil thickness was found on slopes of 0 – 10o

with depth ranging between 0.26 – 1.44m. Examination of the 10 sites with

slope angles less than 10o show that these are often associated with areas of the

catchment adjacent to the catchment divide or in areas adjacent to the river

course, figure 5.24 shows the spatial distribution of these slope angles over the

catchment. Those sample sites located adjacent to the catchment divide also

exhibit soil thickness typically less than 0.5-0.6m, compared with depths

Page 168: Ground Penetrating Radar for the parameterisation of

Andrew Howe

168

ranging from 0.85 to 1.44m for sites located adjacent to the river. The upslope

drainage area provides a method of distinguishing between sites of similar slope

angle but with a wide range of soil thickness. Due to the occurrence of similar

slope angles at very different locations in the catchment it is unlikely that slope

alone is a good predictor of soil thickness. However when linked with another

terrain attribute such as upslope drainage area, slope angle is an important

descriptor of soil thickness distribution within this catchment.

Figure 5.24. The distribution of slope angle less than 10o in the catchment.

Page 169: Ground Penetrating Radar for the parameterisation of

Andrew Howe

169

5.5.4 Elevation

Soil thickness shows a high degree of variability with elevation. Using this

information no strong relationship can be identified between soil thickness and

cell elevation.

Figure 5.25. Elevation.

Page 170: Ground Penetrating Radar for the parameterisation of

Andrew Howe

170

5.5.5 Upslope Area

The plot shows a general trend of increasing soil thickness with increasing area

upslope of a grid cell. Of the terrain attributes examined so far this relationship

is statistically the best with a correlation coefficient of 0.831. Given the range of

upslope area values (625m2 – 40,000m2) generated for all 31 sites a logarithmic

scale provided a more satisfactory plot of the variation of soil thickness with

area (figure 5.26).

Figure 5.26. Natural logarithm (LN) of Upslope Area.

Page 171: Ground Penetrating Radar for the parameterisation of

Andrew Howe

171

5.5.6 Wetness Index

Wetness index combines upslope area with local slope (equation 4.1). Figure

5.27 shows the observed relationship between wetness index and soil thickness

at each site. Given the relationship observed between depth and upslope area,

and the potential for upslope area to explain the observed distribution of soil

thickness at varying slope angles, wetness index could be expected to provide a

good relationship to explain depth distribution between sample sites. This is

found to be the case, with wetness index providing the highest correlation (r =

0.855) with measured soil thickness at sample sites.

Figure 5.27. Wetness Index.

Page 172: Ground Penetrating Radar for the parameterisation of

Andrew Howe

172

The upslope area and wetness indices provide the clearest statistical relationship

between soil thickness and terrain attributes. Prior to applying the relationship

to predict soil thickness for each grid cell across the catchment the physical

basis to using wetness index as a predictor of depth is examined. In particular

are the processes causing soil accumulation and removal similar over the

drainage basin? In this case the catchment concerned is relatively small ~ 3.1

km2 and although located in an upland region of the UK the climate is temperate

and altitude ranges from 355m to 698m (Cyff DEM). In terms of the

environmental factors governing soil creation the limited size of this catchment

and the small altitude range suggests that the climate and therefore also

weathering processes and organisms responsible for soil formation would be

expected to be constant over the entire catchment. Likewise parent material is

uniform throughout this area. Changes in soil thickness must be a result of

transport processes redistributing particulate matter according to local slope and

the capacity of the transport mechanism to move material, in this case by water.

In the long term soil thickness at a point depends on the upstream inputs and

accumulation of biomass in situ. The effect of variable rates of biomass

accumulation across the catchment from peat bogs to grazed pasture is

accounted for within the data by the fact that a cross-section of sites with a

range of upslope areas and vegetation types has been sampled. The difference in

vegetation type and hence biomass input into the soil between bog and pasture

zones may explain why the increase in soil thickness with increasing upslope

area is more complicated than a simple linear increase.

Figure 5.28 shows the spatial distribution of soil thickness after applying the

calculated relationship between wetness index (W) and soil thickness (d) below,

(5.13)

7308.0r 1376.1173.0 2 =−= Wd

Page 173: Ground Penetrating Radar for the parameterisation of

Andrew Howe

173

Figure 5.28. Cell soil thickness (m) predicted from the distribution of wetnessindex.

Page 174: Ground Penetrating Radar for the parameterisation of

Andrew Howe

174

Two features of the predicted spatial pattern of depths are evident. The impact

of the local drainage direction network (LDDN) upon predicted soil thickness is

large and the thickest soils, not unexpectedly, occur in drainage channels. The

LDDN is calculated using elevation data in which the drainage direction out of

a cell is determined by the lowest surrounding cell. The shortcomings of

different routing methods are well-documented (Moore et al. 1996, Burrough &

McDonnell, 1998) and the D8 algorithm implemented in PCRaster cannot

model flow dispersion, resulting in predicted flow networks defined as parallel

lines. For this catchment adjacent cells may have very different upslope

drainage areas, therefore resulting in abrupt changes in predicted soil thickness

between cells.

The result of using an equation to predict soil thickness that is partly a function

of drainage area is that the deepest soils occur in grid cells occupied by the

permanent stream channel. Field observation of the Cyff stream network

however shows that the lower reaches of the channel have a bedrock or boulder

bed structure and therefore soil thickness assumes a minimum value at these

locations. Within the model a permanent stream channel requires a soil

thickness of zero in order that all water entering a channel cell is treated as

overland flow. This is necessary for subsequent modelling of catchment water

flux in response to precipitation events since the catchment outflow hydrograph

is required as one method of model validation. In order to allocate zero depth

values to stream channel cells the permanent stream network was overlain by

enforcing zero depth dependent on the upslope drainage area into a cell. A

threshold of 445,000m2 was chosen, such that cells equal to or exceeding this

value were classified as stream cells (figure 5.29). This threshold value was

selected as providing the best visual fit with work carried out by Newson &

Harrison (1978) into channel characteristics in the Plynlimon catchments. The

authors present a channel classification and map for Plynlimon that classifies

the drainage network into bedrock, boulder or alluvial reaches and also specifies

the location of flushes, gullies and drains. Figure 5.30 details the spatial

variability of soil thickness with stream cells incorporated.

Page 175: Ground Penetrating Radar for the parameterisation of

Andrew Howe

175

Figure 5.29. Grid cells ≥ 445,000m2 upslope area

Page 176: Ground Penetrating Radar for the parameterisation of

Andrew Howe

176

Figure 5.30. Cell soil thickness (m) predicted from the distribution of wetnessindex with stream network cells assigned a value of 0.01m soil thickness.

Page 177: Ground Penetrating Radar for the parameterisation of

Andrew Howe

177

5.6 Conclusion

• Three site classifications have been developed using GPR profile and trace

amplitude and reflector characteristics. Class 1 sites are located in areas of

high moisture content and are characterised by peat deposits > 0.5m

overlying bedrock. These sites are dominated by high amplitude reflectors

which are associated with the peat/bedrock boundary. Class 2 and class 3

sites are both located on hillslope soils, but are differentiated by the extent

of lateral horizon definition evident within GPR profiles, and the degree of

correlation between GPR trace data and physical sampling of soil horizon

depth. Class 2 soils compare well with observed horizon position and allow

the extrapolation of point measurements of horizon type to an entire GPR

profile. Class 3 sites are more problematic and are treated as radar complex

environments. Intensive subsurface investigation is essential to quantify

total soil thickness. Both class 2 and 3 sites exhibit multiple reflectors over

the survey depth and require at least one physical soil sample in order to

relate GPR reflectors to specific soil horizon boundaries.

• GPR for soil thickness measurement has been used successfully in

conjunction with excavation and augering to identify soil horizon position

within a soil profile. In this catchment soil thickness has been measured

using GPR at 900 and 450 MHz. Using root mean square to calculate the

difference between radar and physically recorded depths shows that GPR is

accurate to 0.12 m of auger readings to the B/C-horizon boundary. The

accuracy of GPR improves when measuring the A/B horizon interface with

an absolute error of 0.04m compared with auger measurements. However at

least one physical measurement using a soil auger was required to check

reflector – horizon type and confirm that calculated depth for GPR profiles

corresponded with actual physical depth to horizons. The depth of organic

horizon was not generally measurable with GPR due to the shallow nature

of this reflector, with organic horizons in this catchment ranging between

0.1 and 0.2m below the surface.

Page 178: Ground Penetrating Radar for the parameterisation of

Andrew Howe

178

• From the 30 sites investigated for soil thickness, the spatial distribution of

soil thickness to the upper boundary of the C-horizon is strongly correlated

(r = 0.855) with wetness index derived from the 25m grid cell DEM. No

statistical relationship was identified between a terrain index and depth to

the upper boundary of the B-horizon. The observed relationship between

depth to the C-horizon and wetness index was subsequently used to

calculate distributed soil thickness to bedrock for the entire 3.1km2

catchment.

Page 179: Ground Penetrating Radar for the parameterisation of

Andrew Howe

179

Chapter 6: Soil Moisture Data Analysis

6.1 Introduction

The following chapter presents the analysis methods and results obtained using

GPR for measurement of shallow subsurface soil volumetric water content

(VMC) at 18 sites across the study catchment. VMC derived for different depths

using GPR are compared against VMC measurements obtained at the same

location and depths using gravimetric and theta probe methods for sampling soil

moisture. The measured value of soil moisture at each site is a potential method

for internal hydrological model validation at the scale of the individual grid cell.

6.2 Analysis methodology

Figure 6.1 outlines the data analysis process followed to derive soil moisture

from CMP surveys. Initial processing required that site CMP surveys were

displayed and subsurface velocity derived by measuring the gradient of

reflectors as the horizontal distance between the transmitter and receiver

separation was progressively increased about a common mid-point.

Page 180: Ground Penetrating Radar for the parameterisation of

Andrew Howe

180

Figure 6.1. Soil moisture analysis using CMP data.

Values of shallow subsurface velocities and subsequently VMC can be obtained

through measurement of groundwave velocity from a CMP survey. Using this

method with a 100MHz GPR system, Lesmes et al. (1999) found that GPR

measured VMC followed the general trend in the variation in soil moisture

derived from soil samples taken from the top 30cm of soil. A mean discrepancy

of 0.1m3/m3 was found to exist between GPR and gravimetric VMC

measurements which the authors attributed to the unknown depth of soil

sampled by the groundwave. In order to avoid the problem of an unknown soil

Page 181: Ground Penetrating Radar for the parameterisation of

Andrew Howe

181

volume sampled by the groundwave, this study instead used the velocity

calculated for the first reflector following the groundwave return as input to

calculate soil dielectric values and subsequently VMC. The depth to this

reflector was calculated as the one-way travel time at zero antennae offset,

therefore quantifying the depth over which subsurface velocity is measured. In

this study the depth to the first reflector at survey sites was of a similar value to

maximum depths sampled by theta probe and gravimetric soil cores for

validation of VMC (table 6.1) thereby allowing validation of GPR soil water

measurement. Using equation 5.1, vrms values calculated from each site CMP survey were used

to quantify volumetric soil moisture changes with depth using the established

relationship between velocity and dielectric value (K) (equation 2.13). Values of

K were substituted into the Topp equation (equation 2.13) to calculate the VMC

of a volume of soil. Table 6.1 summarises the calculated VMC from the soil

surface to the specified depth to reflector using GPR at each site. These

estimates of VMC were then compared with two standard and intrusive methods

of VMC sampling: theta probe measurements and soil cores.

Page 182: Ground Penetrating Radar for the parameterisation of

Andrew Howe

182

Table 6.1. Site Vrms and VMC values calculated for the first subsurface reflectorafter groundwave arrival. Only those sites with validation data are included.

Site Maximum depth ofvalidation data (m)

GPR CMP Reflector depth

(m)

GPRVrms

(m/ns)

Dielectricconstant

K

VMC fromGPR (m3/m3)

13 0.34 0.35 0.056 29.0 0.44

14 0.43 0.33 0.046 43.1 0.53

16 0.55 0.53 0.051 35.0 0.48

17 0.15 0.41 0.048 38.9 0.50

18 0.30 0.42 0.055 30.3 0.45

19 0.25 0.46 0.061 24.2 0.39

20 0.05 0.31 0.052 32.9 0.47

22 0.25 0.39 0.053 29.6 0.44

23 0.20 0.34 0.052 33.5 0.47

24 0.20 0.33 0.054 30.8 0.45

27 0.20 0.57 0.058 27.2 0.42

32 0.20 0.32 0.062 23.2 0.38

33 0.55 0.59 0.041 52.6 0.59

48 0.65 0.30 0.052 32.7 0.46

49 0.10 0.25 0.063 22.9 0.38

53 0.45 0.43 0.062 23.5 0.39

54 0.40 0.26 0.047 40.1 0.51

56 0.70 0.25 0.043 48.7 0.56

Page 183: Ground Penetrating Radar for the parameterisation of

Andrew Howe

183

6.3 Comparing GPR VMC with conventional VMC measurementtechniques

Validation of GPR derived measurements of VMC with conventional methods

is required in order to check the accuracy of this technique. Direct comparison

between VMC measured using GPR and VMC measured using a theta probe

and soil core is not immediately possible due to the different volume of soil

sampled by each technique. The theta probe samples a volume of 30cm3 (Theta

probe User Manual, 1998) and soil cores a volume of 44cm3. The measurements

derived from GPR are of an average VMC sampled from the ground surface to

the depth of the reflector providing the return signal. The sample volume is

therefore a function of reflector depth and the GPR footprint area returning the

signal (equations 2.10, 2.11 and 2.12). The geometry of an electromagnetic

wave propagating through the soil is treated in a theoretical manner and is a

current topic for research. In this study the GPR signal is assumed to propagate

as an expanding cone from a point source (the transmitter). Therefore the total

soil volume sampled by GPR can be calculated as the volume of a cone with an

elliptical base equal to the radar footprint area (figure 2.3) and a vertical height

equal to the reflector depth. For a reflector located at a depth of 0.2m and an

average velocity of 0.06m/ns this equates to a volume of 252cm3, compared

with a volume of 4420cm3 for a reflector occurring at 0.6m depth. Compared to

GPR measurements, those obtained using a theta probe or soil core are point

measures of VMC, and have been found in this study to provide similar VMC

results when comparing VMC measurements taken at the same depth positions

at a site. Closely spaced measurements were taken using these two methods,

with a depth interval of between 0.05 – 0.2m, dependant on visual examination

of the variation in subsurface soil structure at a site. Samples were taken until

subsurface stone content made insertion of a theta probe or soil core rings

impossible.

The variation in site VMC with increasing depth from the ground surface was

calculated from theta probe and soil core measurements by fitting trend lines to

the recorded data points for each of these methods. Depending on the trend of

Page 184: Ground Penetrating Radar for the parameterisation of

Andrew Howe

184

data points and goodness of fit, a linear or polynomial trend line was fitted to

theta and/or soil sample VMC data points over the range of depths sampled.

This assumes that the trend line adequately described the behaviour of VMC

between closely spaced sample points. The VMC at any depth can then be

calculated by substituting depth into the trend line equation. Extrapolation

beyond the sampled depth assumes that soil moisture does not change from that

predicted by the trend line, however given the observed effect of soil horizon

changes on VMC (figure 6.3) this activity was minimised.

Page 185: Ground Penetrating Radar for the parameterisation of

Andrew Howe

185

Figure 6.2. VMC – depth relationship with trend lines for site 54.

Trend lines do not directly provide an average value for VMC, but do enable the

calculation of point VMC for a depth. Average moisture between two depths

can be calculated as the area bounded by the trend line and the depth axis using

a method of integration. Given that VMC is a ratio, the area beneath each trend

line is equivalent to a depth of water per depth of soil over which integration is

calculated.

At all but one site (site 48, figure 6.3), integration of the trend line between the

soil surface and depth n was used to calculate the depth of water present in a

given depth n of soil at each site. Depth n was always chosen as the depth at

which a CMP reflection was present with a corresponding value of VMC for the

overlying soil to the surface. Given that GPR VMC is calculated using vrms and

vrms is an average of the velocity between the ground surface to the reflector,

Page 186: Ground Penetrating Radar for the parameterisation of

Andrew Howe

186

integration was between the limits of soil surface to the depth of reflector.

Wherever possible a CMP reflector depth was used which was located at a

similar depth to which theta probe and/or soil cores had been collected in order

to minimise extrapolation of VMC values beyond the range of measured data.

Figure 6.3. Soil thickness - VMC relationship for site 48 with strong soil horizoncontrol on VMC. The sharp increase in VMC between 0.25m and 0.30m depthcorresponds to the position of the O/A boundary. Manual calculation of the areabeneath the observations was used in this one case where a trend line wasunable to simulate the variation in VMC with depth.

Page 187: Ground Penetrating Radar for the parameterisation of

Andrew Howe

187

6.4 Soil moisture from GPR surveys

Figure 6.4 and figure 6.5 present the results of the calculation of total moisture

depth measured using GPR and theta probe, and GPR and gravimetric soil

moisture data. Raw data for soil moisture depth for each site using the three

techniques are presented in tabular format in appendix II, table II.2.

The calculated root-mean-square (RMS) error between GPR and theta probe

measured moisture is 0.03m3/m3 and 0.05m3/m3 between GPR and gravimetric

measured moisture, an average error between techniques of 0.04m3/m3.

Solving the fitted best-fit line for each graph show that for a given depth of soil

water, measured using either gravimetric or theta probe techniques, the GPR

over measures the amount of water at low water contents (depths) and under

measures the water content at higher water contents (depths). These data are

summarised in table 6.2.

Table 6.2. The difference between GPR measured water and water depthmeasured using gravimetric and theta probe methods.

Gravimetric GPR % difference Theta probe GPR % difference

0.05 0.058 +0.16 0.05 0.061 +0.22

0.15 0.141 -0.06 0.15 0.152 +0.01

0.35 0.308 -0.12 0.35 0.334 -0.05

Page 188: Ground Penetrating Radar for the parameterisation of

Andrew Howe

188

Figure 6.4. GPR measured moisture and gravimetric moisture.

Figure 6.5. GPR measured moisture and theta probe moisture.

Page 189: Ground Penetrating Radar for the parameterisation of

Andrew Howe

189

The frequency histograms of residuals (figure 6.6) for the both techniques show

very different distributions. A normal distribution is associated with the theta

probe – GPR moisture measurement technique, whilst the gravimetric – GPR

technique shows an increased frequency of higher magnitude residuals. There is

also a degree of positive skewness associated with the gravimetric technique

with GPR tending to measure less water than found in soil samples.

Figure 6.6. Frequency distribution of the error between GPR water depth and a)theta probe measurements, b) gravimetric soil samples.

The signed-rank test was applied to test the significance of the relationship

between GPR measured soil moisture and theta probe and gravimetric estimates

of moisture over the same depth of subsurface. The signed-rank test is a non-

parametric statistical test allowing comparison between pairs of observations

and examination of the difference between data pairs (Hirsch et al., 1993). In

this case the value of GPR water depth calculated for each site is paired with the

associated theta probe and gravimetric integrated value. A non-parametric test

was chosen given the non-normal distribution assumed by the residuals of GPR-

gravimetric samples. The null hypothesis (H0) for the signed-rank test states

that both data series come from the same population. The alternative hypothesis

(H1) states that a statistically significant difference between series exists.

Page 190: Ground Penetrating Radar for the parameterisation of

Andrew Howe

190

Assigning confidence limits of 0.95, the critical test value (ZCRIT) is 1.96.

Table 6.3 shows the calculated Z value for GPR-theta probe and GPR-

gravimetric paired water depth measurements.

Table 6.3. Signed-rank test results.

GPR-theta probe GPR-gravimetric

Number ofobservations

18 12

Z -0.305 -1.608

Action H0 accepted, no difference

exists between samples.

H0 accepted, no difference exists

between samples.

The overall correlation values between GPR and theta probe and soil sampled

VMC are equivalent (r2 equals 0.80 for each), despite the distribution of

residuals differing. The signed-rank test calculates a figure closer to zero for

GPR-theta probe observations although both data samples exhibit a Z value less

than the critical value and therefore the null hypothesis, that no statistically

significant difference exists between populations, can be retained. In this case

GPR measurements of shallow subsurface moisture content is statistically

comparable to the integrated value of moisture over the same depth derived

from a trend line fitted through a series of point measurements.

Given the relationship found between GPR and standard measurements of

VMC, CMP first reflector velocities were calculated for the remaining 11 sites

at which CMP surveys had been carried out but no validation data were

collected. The results are presented in table 6.4.

0Reject HZZ CRIT ⇒>

Page 191: Ground Penetrating Radar for the parameterisation of

Andrew Howe

191

Table 6.4. CMP measured VMC for sites without validation data.

Site CMP Reflector depth(m)

Vrms (m/ns) Dielectricconstant

GPR VMC(m3/m3)

5 0.56 0.051 35.2 0.48

6 0.38 0.047 40.3 0.51

8 0.35 0.054 31.3 0.45

9 0.46 0.059 25.9 0.41

21 0.50 0.044 47.5 0.55

37 0.55 0.040 57.5 0.61

39 0.93 0.041 53.2 0.59

40 0.90 0.041 52.2 0.58

41 0.75 0.040 55.9 0.61

50 0.38 0.050 36.5 0.48

60 0.59 0.080 14.2 0.26

The VMC data measured using GPR forms a set of observations of site soil

moisture for 29 sites across the catchment. These data sets provide a method of

comparing the spatial pattern of soil moisture dynamics predicted in the

catchment using the hydrological model with distributed soil thickness.

Page 192: Ground Penetrating Radar for the parameterisation of

Andrew Howe

192

6.5 Conclusion

Soil moisture derived using GPR CMP surveys is statistically comparable to the

depth of water predicted to be present in a depth of soil sampled by theta probe

and soil sample measurements. In order to compare GPR derived moisture with

physical soil moisture measurements there is a requirement to scale up from

point samples to the volumes sampled by GPR. The method used in this study

compares the two techniques by integration of measured soil sample and theta

probe VMC to estimate the depth of moisture per depth of soil. The two key

assumptions to this process are that soil moisture variation with depth at each

site can be described by the fitting of a trend line to observed data, and that an

expanding cone centred on the transmitter adequately describes the GPR

subsurface sample volume. The similarity in results between methods suggests

that these assumptions are valid in this case.

Page 193: Ground Penetrating Radar for the parameterisation of

Andrew Howe

193

Chapter 7: Hydrological Modelling Results

7.1 Introduction

This chapter presents the results obtained using the hydrological model run on a

25m-grid resolution to simulate the effect of varying soil thickness distribution

on the catchment and cell hydrological response to precipitation. Differing

scenarios of catchment soil thickness were investigated, with soil thickness

assumed to be either constant or variable over the entire catchment. The overall

response of changing soil thickness distribution on hydrological behaviour was

examined using a) total catchment discharge characteristics for each scenario,

and b) the predicted status of overland flow, subsurface flow and soil moisture

simulated for the field plot. Comparison between modelled and recorded data at

both these scales is used to examine the impact of varying soil thickness on

these hydrological variables.

7.2 The catchment model

Nine soil thickness scenarios were used to examine the hydrological response to

varying soil thickness. These consisted of eight scenarios in which the

catchment soil was a constant depth value, ranging from a minimum depth of

0.2m and increasing to a maximum depth of 1.6m in increments of 0.2m each

modelled scenario. This range of depth was chosen as being similar to the range

of depth sampled through field investigation (0.2m - 1.5m). A single model

scenario in which soil thickness was considered variable was also investigated.

The distributed pattern of catchment soil thickness used the GPR soil thickness

and soil wetness index relationship detailed in chapters six and seven, using a

25m-grid cell size. Each of the nine simulations was run at an hourly time step

for the period 1st October 1997 to 7th July 1999. This spanned a period of 15,469

hours, for which hourly rainfall and potential evaporation were available for

model input and catchment outflow discharge values for model validation at the

Page 194: Ground Penetrating Radar for the parameterisation of

Andrew Howe

194

catchment scale. Field plot overland flow and subsurface flow were available

over the same time period, and volumetric soil moisture monitored by hourly

theta probe readings from 28th September 1998 to 7th July 1999.

All model parameter values, input maps and initial values were kept constant

for each model run except for cell soil thickness. Model parameters and initial

values are listed in Appendix I. In all analysis of model catchment hydrograph

compared with the recorded hydrograph the first 1500 hours of data are rejected

to allow the model to reach a state of equilibrium with catchment conditions.

7.3 Reminder: The effect of soil thickness on modelleddischarge

Soil thickness is an important input parameter to this model and can be

considered either as a uniform value for the entire catchment or spatially

distributed for each grid cell contained within the catchment. The depth of soil

present in any cell controls the maximum moisture storage capacity of that cell.

The relationship between soil thickness and maximum moisture (equation 3.6)

was derived from 55 field measurements of soil bulk density at 19 sites, for a

range of depths between 0.03m and 0.50m.

Actual water input to a cell is via a combination of rainfall, and for those cells

with upslope drainage sources, subsurface throughflow. Cell soil thickness, by

controlling the maximum moisture capacity, also controls the partition of

overland flow/throughflow generation for each cell. When actual cell moisture

equals the maximum cell moisture, subsequent rainfall and subsurface

throughflow inputs are considered to be surface runoff for that cell, although re-

infiltration can occur in downstream cells. This model routine simulates

saturated overland flow.

The model also simulates the process of infiltration excess overland flow when

rainfall intensities exceed the maximum rate of infiltration. Infiltration (equation

3.3, Campbell, 1985) is a function of soil texture and bulk density. Water

Page 195: Ground Penetrating Radar for the parameterisation of

Andrew Howe

195

infiltrating into the soil is considered to form a defined boundary between non-

saturated soil (below the wetting front and above the water table) and saturated

soil (above the wetting front, extending to the surface). Using equation 3.6, the

rate of advance of the wetting front and therefore infiltration rates are controlled

by the depth of the wetting front within the soil profile, since soil bulk density is

a function of depth.

The volume of subsurface throughflow is controlled by the position of the water

table in each cell and the lateral hydraulic conductivity. The vertical difference

in water table elevation between a cell and its downstream neighbour is used to

calculate the hydraulic gradient for throughflow. Based on a model developed

by Xiao et al. (1996), the average soil thickness contributing to flow between

cells is the average of the sum of cell soil thickness. In this model throughflow

occurs between cells which are not fully saturated, instead using water table

position within both cells to calculate the average depth of saturated flow.

Page 196: Ground Penetrating Radar for the parameterisation of

Andrew Howe

196

7.4 Catchment model results

The effects of varying soil thickness model scenarios on catchment flow

response to precipitation and evaporation are summarised in table 7.1. The

outflow hydrograph generated from each model run is used to examine the

impact of soil thickness variability on catchment outflow for the same time

period. It should be noted that while a high r2 value indicates that a scenario

hydrograph has a similar pattern to the measured hydrograph, the root mean

square (RMS) error indicates the difference in magnitude between modelled and

observed hourly discharge.

Table 7.1. Catchment outflow summary data for all modelled scenarios (1501-15469 hours).

Modelled soilthicknessscenario

MeanQ

(m3/s)

Max Q (m3/s)

Min Q (m3/s)

Totaldischarge

(×106 m3)

Linear trendline R2

value: measured vs.modelled hourly

discharge

RMSError

Constant, 0.2m 0.185 5.476 0.000 9.31 0.741 0.287

Constant, 0.4m 0.186 4.000 0.002 9.33 0.578 0.334

Constant, 0.6m 0.186 3.227 0.005 9.37 0.447 0.357

Constant, 0.8m 0.187 2.812 0.007 9.39 0.334 0.371

Constant, 1.0m 0.186 2.584 0.009 9.37 0.242 0.380

Constant, 1.2m 0.185 2.439 0.010 9.31 0.170 0.387

Constant, 1.4m 0.183 2.329 0.011 9.22 0.116 0.392

Constant, 1.6m 0.181 2.249 0.012 9.13 0.076 0.396

Variable, GPR

measured

depth.

0.186 3.960 0.006 9.36 0.530 0.343

Discharge

values recorded

at gauging

station.

0.265 6.110 0.019 13.30 N/A N/A

Page 197: Ground Penetrating Radar for the parameterisation of

Andrew Howe

197

Figure 7.1. Modelled and recorded minimum and maximum catchment discharge,from table 7.1.

Figure 7.1 shows two key trends; decreasing maximum river discharge as soil

thickness increases and a decrease in the minimum modelled discharge as

catchment soil thickness is reduced. The variable soil thickness model shows

characteristics of both, with lower peak discharge than the shallow depth

scenarios but a larger minimum flow during drier periods. For those model

scenarios considering a catchment constant soil thickness, the rate of change in

minimum and maximum discharge is reduced as depth increases. Soils deeper

than 0.8-1.0m show increased hydrological stability, increasingly damped

against the minimum and maximum extreme flow values which are

characteristic of shallower depth soil scenarios.

The pattern of modelled maximum and minimum discharge is detailed by the

scenario hydrograph. Figure 7.2 shows the recorded hydrograph and the

simulated hydrograph for the GPR measured soil thickness scenario.

Page 198: Ground Penetrating Radar for the parameterisation of

Andrew Howe

198

Figure 7.2. Outflow hydrograph of measured discharge and GPR variable soilthickness modelled discharge.

Page 199: Ground Penetrating Radar for the parameterisation of

Andrew Howe

199

7.4.1 High Flows

One high flow period is used as an example of model behaviour (figure 7.3).

Three soil thickness scenarios are considered: - 0.2m catchment soil thickness

and 1.6m soil thickness which are the lowest and highest values of soil

thickness used, and the variable soil thickness scenario. These three scenarios

enable a comparison to be made between a lumped and distributed catchment

soil thickness simulation and actual flow measurements. The period examined

extends from 27th February 1999 to 3rd March 1999 (100 hours) during which

time a total of 127.2mm rainfall was recorded.

Figure 7.3. A period of high flow from 27th February 1999 to 3rd March 1999.

Page 200: Ground Penetrating Radar for the parameterisation of

Andrew Howe

200

Table 7.2. High flow statistics for event 27th February 1999 to 3rd March 1999.

Model

scenario

Mean

absolute

error (m3/s)

Maximum

flow (m3/s)

Minimum

flow (m3/s)

Linear trendline R2

value of measured

vs. modelled

hourly discharge

RMS

Error

Measured

flow

N/A 5.822

(100%)

0.151

(100%)

N/A N/A

0.2m depth

flow

0.664 3.990

(69%)

0.141

(93%)

0.813 1.147

1.6m depth

flow

0.995 1.861

(32%)

0.189

(125%)

0.589 1.682

GPR

variable

depth flow

0.771 3.132

(54%)

0.176

(117%)

0.677 1.367

While the shape of the storm hydrograph for all scenarios is similar to the

hydrograph recorded by the Institute of Hydrology flume, no model scenario is

able to recreate the magnitude of recorded peak flows. Those model scenarios

with constant shallow soil thickness provide the highest discharges and lowest

RMS error (1.147m3/s), constant deep soils the lowest discharge and highest

RMS error (1.682m3/s). The distributed soil thickness model performs more

closely to the shallow constant model particularly for the third hydrograph peak

shown in the above example. The inability of any model scenario to generate

equivalent flows to those recorded may be due to the lack of a model

component enabling macropore flow through the subsurface.

Page 201: Ground Penetrating Radar for the parameterisation of

Andrew Howe

201

7.4.2 Intermediate and low flows

350 hours of low flow occurred between 17th May 1999 and 1st June 1999.

During this time period 21.4mm of rainfall was recorded and the hydrograph is

dominated by base flow. Model output is presented in figure 7.4 and table 7.3

for soil thickness scenarios of 0.2m, 1.6m and GPR derived soil thickness.

Table 7.3. Low and intermediate discharge statistics from 17th May 1999 to 3rd

June 1999.

Model

scenario

Mean

absolute

error

(m3/s)

Mean

flow

(m3/s)

Maximum

flow

(m3/s)

Minimum

flow

(m3/s)

Linear trendline

R2 value of

measured vs.

modelled hourly

discharge

RMS

Error

Measured

flow

N/A 0.042

(100%)

0.073

(100%)

0.030

(100%)

N/A N/A

0.2m

depth flow

0.020 0.022

(52%)

0.046

(63%)

0.004

(13%)

0.615 0.020

1.6m

depth flow

0.009 0.049

(117%)

0.088

(121%)

0.022

(73%)

0.224 0.012

GPR

variable

depth flow

0.007 0.037

(88%)

0.068

(93%)

0.014

(47%)

0.529 0.009

Page 202: Ground Penetrating Radar for the parameterisation of

Andrew Howe

202

Figure 7.4. Low flow hydrograph

Throughout this period the 0.2m soil thickness scenario predicts mean discharge

values 48% less than the mean observed discharge. This characteristic is due to

the reduced amount of total soil pore volume compared with deeper soils and

therefore a reduction in the amount of subsurface water stored in each grid cell.

During periods of reduced or no precipitation shallow soil models are unable to

simulate catchment discharge due to the lack of soil water available for release

into the river channel.

The 1.6m soil thickness scenario maintains discharge values approximately

double those of the 0.2m depth model. This increase is the result of the greater

water storage capacity of deeper soils with greater pore volume per cell. The

distributed soil thickness scenario provides the closest simulation of actual flow

with the smallest RMS error, 0.009m3/s, between modelled and observed

discharge during this period.

Page 203: Ground Penetrating Radar for the parameterisation of

Andrew Howe

203

7.4.3. Catchment discharge frequency distributions

Analysis of the differences in discharge between varying scenarios of soil

thickness can also be achieved by examining the frequency distribution of

discharge for each scenario and recorded data. Flow results for each model

scenario and measured river flow are shown as frequency distributions in figure

7.5, where the x-axis shows the discharge interval and the y-axis the percentage

of all hourly flows falling within each interval.

The Kruskal-Wallis statistical test was used to examine for differences between

modelled scenarios and the observed catchment outflow frequency distribution.

Kruskal-Wallis is a one-way analysis of variance (ANOVA) by ranks test,

which tests for the difference between the medians of two or more samples. The

Kruskal-Wallis test is a non-parametric version of the F-test, and therefore does

not assume a normal distribution or homoscedasticity (Rogerson, 2001).

In this case the test was applied to frequency distributions of catchment

discharge by pairing measured discharge against each model scenario discharge

in turn. The null and alternative hypothesis for this test were specified

respectively as:

H0 = no statistical difference exists between observed and the modelled flow

frequency distribution.

H1 = the modelled flow frequency distribution is statistically different from

observed flow.

The results of comparison between model scenarios of soil depth and measured

flow are presented in table 7.4.

Page 204: Ground Penetrating Radar for the parameterisation of

Andrew Howe

204

Figure 7.5. Total catchment discharge (m3/s) histograms (measured andmodelled).

Page 205: Ground Penetrating Radar for the parameterisation of

Andrew Howe

205

The frequency distribution of measured catchment discharge is positively

skewed, with 95% of hourly flow values occurring between 0.03m3/s and

1.00m3/s. The distribution is typical of many catchment flow regimes, in which

high magnitude flows are low frequency events. The frequency distributions of

modelled soil thickness show that none are able to fully simulate the distribution

of observed discharges.

Increasing catchment soil thickness leads to a reduction in the range of

discharges observed, achieved through a reduction in the frequency of high flow

and low flow values. The effect of a reduction in the range of discharges

modelled is to increase the contribution of ‘medium’ flows to the hydrograph.

This is seen in the frequency distribution as an increase in the maximum

percentages of discharge, for example discharges in the range 0.12m3/s to

0.25m3/s contribute 52% of discharge for the 1.6m soil thickness scenario

compared with 31% for the 0.2m scenario. As soil thickness increases,

catchment discharge is shown to decrease in variability, with a reduction in both

the highest and lowest values, a damping of discharge in response to

hydrological input, also highlighted by figure 7.1.

Shallow, constant depth soils perform well in comparison to deeper, constant

depth soils when simulating high discharges. The 0.2m depth frequency

distribution shows that 2.1% of flows occurring are greater than 1.0m3/s,

compared with 0.6% for the 1.6m depth and 4.4% measured in the field. The

reverse is true for low flows. Shallow soil model scenarios over-predict the

frequency of low flow occurrence. The 0.2m depth scenario has 5.1% of

discharge in the range 0-0.01m3/s. Neither the 1.6m depth scenario nor the

measured hydrograph recorded discharges below 0.01m3/s.

The GPR, variable depth frequency distribution exhibits characteristics of both

shallow and deep soils, unsurprising given that cell soil thickness in this

scenario range from 0.22m to 1.70m, with an average soil thickness of 0.51m.

The variable depth histogram is not, however, an average of the frequency

Page 206: Ground Penetrating Radar for the parameterisation of

Andrew Howe

206

distributions produced for 0.4m and 0.6m constant soil thickness scenarios.

Maximum discharge frequencies (1.4% between 3.7-4.0m3/s) are equivalent to

those simulated by the 0.4m scenario, while the frequency of discharges

between 0-0.01m3/s is 0.3%, compared with 0.4% and 2.4% for scenarios of

0.6m and 0.4m constant soil thickness.

For this model, soil thickness is one control on both the range of catchment

discharge and the distribution of discharges. The effect of increasing soil

thickness is to de-sensitize the model to periods of extreme hydrological

behaviour. Peak flows are lower than those simulated by shallow soils, while

discharge during periods of low flow are greater. Using the frequency

distributions of scenario discharge, it can be seen that scenarios with deeper,

uniform soils produce a smaller range of flows, with the greatest reduction

occurring for the high values of discharge. The reduction of hydrological

response to periods of intense and zero precipitation is due to the increased

moisture storage capacity of deeper soils. These soils are able to store greater

volumes of water prior to saturation, thereby reducing the amount of overland

flow generated by a rainfall event. During inter-storm periods this water is

released into the river network, maintaining discharges above zero.

Shallow soils exhibit a range of discharges similar to those recorded at the

gauging station. However the frequency distributions shows that for shallow

soil scenarios during interstorm periods, discharge volumes are underestimated.

The volume of water contributing to catchment outflow is limited during these

periods because of the limited soil moisture capacity of each cell.

The variable soil thickness scenario improved simulation of the catchment

hydrograph, primarily for periods of low and intermediate flow. The frequency

distribution for this scenario showed that the frequency of discharge less than

0.01m/s was less than for scenarios of similar uniform depth, but that the

distribution of high discharge values was similar to that achieved by the

uniform, shallow depth model runs.

Page 207: Ground Penetrating Radar for the parameterisation of

Andrew Howe

207

Table 7.4 Kruskal-Wallis results for catchment discharge frequency distributions.

Modelled soildepth scenario

Kruskal-Wallis test value 1

p - value H0

Variable depth 3.08 0.079 Accept

0.2m depth 1.27 0.259 Accept

0.4m depth 2.61 0.106 Accept

0.6m depth 3.63 0.057 Accept

0.8m depth 4.53 0.033 Reject

1.0m depth 5.37 0.021 Reject

1.2m depth 5.90 0.015 Reject

1.4m depth 6.17 0.013 Reject

1.6m depth 6.68 0.010 Reject

1 Critical test statistic value 3.841 (p = 0.05) using chi square distribution for 1 degree

of freedom (Rogerson, 2001).

Table 7.4 shows that the null hypothesis (H0) is rejected for all soil depth

scenarios between 0.8m and 1.6m deep. The frequency distributions for these

scenarios are statistically different to the observed distribution of catchment

discharge. The results are significant, since the p-values for each of these 5

distributions are below the alpha value of 0.05, such that the probability of

making a type I error (rejecting the null hypothesis when it is in fact true) is less

than 5% in all cases.

Variable soil depth and scenarios of constant depths of 0.2m, 0.4m and 0.6m

calculated test values less than the critical value, and therefore the null

hypothesis was not rejected. In each of these scenarios the modelled distribution

of flows was not statistically different from that observed. Again the p-values

for these 4 scenarios confirm this finding.

Page 208: Ground Penetrating Radar for the parameterisation of

Andrew Howe

208

7.5 The spatial distribution of catchment runoff

Figure 7.6 shows the variation in source area for total overland flow exceeding

an arbitrary threshold, in this case 0.1m depth, generated over the model

simulation period. The percentage catchment area contributing to overland flow

using this threshold is 64% for the 0.2m depth scenario compared with 36% for

the 1.6m soil scenario. The spatial pattern of overland flow occurrence on

deeper soils shows a well-defined network, with overland flow concentrated

along lines of flow equivalent to stream channels. In contrast the shallow soil

simulation shows less defined linearity, with an increasing number of cells

adjacent to the drainage divide contributing to overland flow.

Figure 7.6. The spatial distribution of total overland flow for model scenarios of0.2m and 1.6m soil thickness.

GPR soil thickness scenario results (figure 7.7) show the spatial distribution of

total overland flow and throughflow after removal of cells that contained the

permanent stream channel within them. Stream channel cells were assigned a

constant soil thickness of 0.01m to simulate the bedrock form of these channels

at Plynlimon.

Page 209: Ground Penetrating Radar for the parameterisation of

Andrew Howe

209

Figure 7.7. The spatial distribution of total overland flow and subsurface flowfrom the GPR variable soil thickness scenario.

Page 210: Ground Penetrating Radar for the parameterisation of

Andrew Howe

210

Total overland flow shows greater network linearity than patterns of subsurface

flow. This is due to the influence of upslope flow area and the local drainage

network on the distribution of surface flow. Despite the model clearly defining

stream networks, every catchment cell records an amount of overland flow for

at least one timestep of the model simulation extending to cells adjacent to the

catchment divide. Figure 7.8. a) shows that soil thickness has a large control on

total overland flow, because of the influence of upslope contributing area on

both routed overland flow and, in this model, soil thickness through the use of

the wetness index as a covariate for cell soil thickness.

Cell subsurface flow totals show a less defined channel network and a more

smoothed spatial distribution of total flow. This is a result of subsurface flow

being partially controlled by cell soil thickness, along with subsurface drainage

direction. Adjacent cells are likely to have similar magnitude soil thickness

because of the likelihood that adjacent cells have similar wetness index values

and hence soil depths.

The range of total subsurface flow is of a similar magnitude to overland flow

totals for the modelled period. The variation in total flow depth is responsive to

soil thickness variability, given that soil thickness controls the soil water storage

capacity of each cell within the model. Graphs of modelled overland and

throughflow cell totals plotted against cell soil thickness show a trend of

increasing depths of water for deeper soils (figure 7.8). All cells within the

catchment produce both types of flow at some point within the simulated

period, but total runoff depths follow a pattern similar to throughflow. This is

due to the high values of throughflow compared with overland flow depths for

the same time period.

Page 211: Ground Penetrating Radar for the parameterisation of

Andrew Howe

211

The increase in throughflow for deeper soils is a result of two factors:

(a) In this model, deeper soils have a larger moisture capacity than shallow

soils, and therefore a larger depth of water to potentially transmit to

downstream cells.

(b) Deeper soils exhibit increased maximum cross-sectional area available to

transmit water to a downstream cell.

Page 212: Ground Penetrating Radar for the parameterisation of

Andrew Howe

212

Figure 7.8. GPR variable soil thickness model: - cell water depth vs. cell soilthickness.

(a) Overland flow.

(b) Subsurface throughflow.

Page 213: Ground Penetrating Radar for the parameterisation of

Andrew Howe

213

(c) Total cell runoff.

Page 214: Ground Penetrating Radar for the parameterisation of

Andrew Howe

214

7.6 Field plot data and model results

The instrumented field plot provided an event record of subsurface flow,

overland flow and soil moisture. In order to compare logged data with the

hourly values produced by the model, event records were processed to provide

hourly totals for each parameter (figure 7.9). These records provided validation

for the modelled hydrological state at the plot location within the model,

allowing internal state validation for part of a single cell in addition to the

aggregated catchment validation outlined above.

Page 215: Ground Penetrating Radar for the parameterisation of

Andrew Howe

215

Figure 7.9. Rainfall, overland flow and throughflow measured at the field plot.

Plot m e as ure d ove r land flow

0

0.005

0.01

0.015

0.02

0.025

0.03

1-Oct-97

1-Nov-97

1-Dec-97

1-Jan-98

1-Feb-98

1-Mar-98

1-Apr-98

1-May-98

1-Jun-98

1-Jul-98

1-Aug-98

1-Sep-98

1-Oct-98

1-Nov-98

1-Dec-98

1-Jan-99

1-Feb-99

1-Mar-99

1-Apr-99

1-May-99

1-Jun-99

1-Jul-99

Tim e

Flo

w (

m3/

hr)

Plot m e as ure d throughflow

0

0.005

0.01

1-Oct-97

1-Nov-97

1-Dec-97

1-Jan-98

1-Feb-98

1-Mar-98

1-Apr-98

1-May-98

1-Jun-98

1-Jul-98

1-Aug-98

1-Sep-98

1-Oct-98

1-Nov-98

1-Dec-98

1-Jan-99

1-Feb-99

1-Mar-99

1-Apr-99

1-May-99

1-Jun-99

1-Jul-99

Tim e

Flo

w (

m3/

hr)

Hour ly rainfall (m m )

0

2

4

6

8

10

12

14

16

1-Oct-97

1-Nov-97

1-Dec-97

1-Jan-98

1-Feb-98

1-Mar-98

1-Apr-98

1-May-98

1-Jun-98

1-Jul-98

1-Aug-98

1-Sep-98

1-Oct-98

1-Nov-98

1-Dec-98

1-Jan-99

1-Feb-99

1-Mar-99

1-Apr-99

1-May-99

1-Jun-99

1-Jul-99

Tim e

Ho

url

y ra

infa

ll (m

m)

Page 216: Ground Penetrating Radar for the parameterisation of

Andrew Howe

216

Because of the scale difference between the 25m-model grid size and the 2m-

field plot cross-section three assumptions were made regarding volumes

contributing flow compared to the volume of the grid cell.

1. In order to directly compare between recorded data for the grid cell and

model results, modelled subsurface flow and overland flow measurements

were reduced to simulate a 50m2 contributing area. A 25m model grid cell

resolution has an area of 625m2, therefore model values of overland flow

and subsurface flow were reduced to 8% of the cell value each hour to allow

direct comparison with hourly recorded plot measurements.

2. Overland flow and subsurface flow were measured as the amount of water

entering two lengths of two-metre length of pipe located perpendicular to

the line of steepest slope gradient. Surveying of the immediate area

(approximately 100m2, from which the hillslope 5m grid cell DEM was

derived) calculated that an upslope area of 50m2 contributed overland flow

to the plot. The same area was also assumed to contribute to subsurface

flow.

3. Soil moisture measured by three theta probes at depths of 0.08m, 0.23m and

0.37m was assumed to be representative of soil water over the entire 25m

square grid cell.

To examine the differences between the variable depth scenario and constant

depth scenarios for key model output variables all results are expressed as a

fraction of the variable soil thickness output values. Therefore the output from

the variable depth scenario is assigned a value of unity, and all other scenario

results are expressed as a fraction of this value. This enables the magnitude and

direction of all output variables for all scenarios to be compared in a single

graph (figure 7.10). The soil thickness (x-axis) for each scenario is expressed as

the fraction of the variable soil thickness for the field plot cell. The output

Page 217: Ground Penetrating Radar for the parameterisation of

Andrew Howe

217

variable (y-axis) is calculated as the fraction of the value of the variable soil

thickness model.

Figure 7.10. Output variable change (totals) in response to changing plot cell soilthickness.

Increasing plot soil thickness results in a decline in simulated overland flow to

zero, but an increase in the amounts of subsurface flow and, not unexpectedly,

increased total soil moisture within the cell. Actual evapo-transpiration values

are relatively insensitive to an increase in cell soil thickness, indicating that, as

expected for a temperate climate with relatively high annual rainfall, evapo-

transpiration is limited by net radiation rather than the available soil moisture.

Page 218: Ground Penetrating Radar for the parameterisation of

Andrew Howe

218

7.6.1 Overland flow

The model predicts that overland flow at the field plot cell is only initiated for

model scenarios of 0.2m-catchment soil thickness and distributed soil thickness.

Scenarios with deeper soils fail to generate any overland flow at the field plot

cell for the period simulated. Modelled data for those two model runs that do

generate overland flow compares poorly with the frequency and magnitude of

measured flow. Figure 7.11 shows the model simulation of overland flow events

between 1st October 1997 and 7th July 1999 for the plot cell.

Figure 7.11. Modelled plot cell overland flow (25m DEM)

One explanation for the poor reproduction of overland flow events by the model

is grid scale. Overland flow that was observed on hillslopes in the catchment

was highly spatially variable. Flow on the slopes adjacent to the plot was

concentrated in narrow channels generally less than one metre wide and moved

downslope in response to micro topography. While the overall modelled flow

direction may be correct, within any cell actual flow is concentrated into areas

far smaller than the 625m2 cell area can resolve.

Page 219: Ground Penetrating Radar for the parameterisation of

Andrew Howe

219

To examine the impact of changing the grid resolution on modelled overland

flow occurring at the plot cell, the model was run for the hillslope immediately

upslope and containing the plot cell for a 5m-cell size. The hillslope DEM was

derived from surveying measurements of the area, not interpolation of the 25m

grid DEM. Figure 7.12 details the change in the local drainage direction (LDD)

when using a 25m and a 5m DEM of the same hillslope. In the case of the 25m

DEM, the plot is contained within a cell with no drainage from upslope cells.

Due to the improved resolution of the 5m DEM the plot cell is shown located

with a maximum of 6 cells upslope possibly contributing to flow.

Page 220: Ground Penetrating Radar for the parameterisation of

Andrew Howe

220

Figure 7.12. The changing pattern of the local drainage direction (LDD) networkafter decreasing cell size from 25m to 5m.

Page 221: Ground Penetrating Radar for the parameterisation of

Andrew Howe

221

Soil thickness for the upslope area and plot was measured using a 30m GPR

profile extending from 2m downslope of the plot, upslope to the sub-catchment

drainage divide. All other model parameters retained the values assigned for the

catchment simulation. As was the case for the catchment model, soil thickness

of 0.2m to 1.6m in increments of 0.2m were used to examine model output

using lumped soil thickness scenarios. The mean values of model output derived

from this second simulation run on a 5m grid are shown in figure 7.13.

Figure 7.13. Model output values for 5m-grid size.

The effect of the reduction in grid size on modelled overland flow generation is

striking. Compared with the 25m grid in which only the GPR and 0.2m depth

scenario produced overland flow, all scenarios generated overland flow. The

trend is similar to the 25m model, with shallow soils generating a larger volume

of overland flow than deeper soils. This is due to the saturation of shallow soils

for longer time periods than deeper soils with their greater soil moisture storage

capacity.

Page 222: Ground Penetrating Radar for the parameterisation of

Andrew Howe

222

In common with model results at the 25m scale, volumes of throughflow are

linearly related to an increase in soil thickness because the model calculates cell

subsurface outflow as a function of Ksat, hydraulic gradient and the amount of

water stored in each cell. Evaporation is least affected by changing soil

thickness, only slightly increasing with increased depth (and therefore stored

moisture).

Soil moisture storage is a non-linear function of depth because the controlling

equation on soil moisture uses field sampled porosity data to reconstruct the

measured decrease in pore volume with increasing sample depth. In this case

the relationship between depth and porosity (bulk density) was found to be

exponential (equation 3.6).

The frequency and magnitude of plot equivalent modelled overland flow after

compensation for the 5m-cell size used is shown in figure 7.14. The reduction in

grid size and the creation of cells upslope which contribute to plot overland

flow leads to an increased frequency of events and greater predicted discharge.

Figure 7.14. Modelled plot overland flow data for 5m grid resolution.

Page 223: Ground Penetrating Radar for the parameterisation of

Andrew Howe

223

Frequency distributions of simulated overland flow for all nine soil thickness

scenarios are shown in figure 7.15. Compared with the distribution of observed

overland flow, the model over predicts the frequency of high volumes of surface

flow and does not recreate overland flow of similar magnitude to that measured

by the field plot. However the effect of increasing model catchment soil

thickness is to reduce the total volume of flow occurring and reduce the overall

frequency of these events for the modelled period.

The Kruskal-Wallis 1-way ANOVA was again applied to compare each

modelled distribution of flow with that observed at the field plot. The results are

presented in table 7.5.

Table 7.5. Kruskal-Wallis results applied to frequency distributions of field plotrunoff.

Modelled soildepth scenario

Kruskal-Wallis test value 1

p - value H0

Variable depth 0.40 0.525 Accept

0.2m depth 1.56 0.212 Accept

0.4m depth 0.32 0.571 Accept

0.6m depth 0.01 0.958 Accept

0.8m depth 0.37 0.543 Accept

1.0m depth 0.94 0.332 Accept

1.2m depth 1.58 0.209 Accept

1.4m depth 2.25 0.134 Accept

1.6m depth 4.24 0.039 Reject

1 Critical test statistic value 3.841 (p = 0.05) using chi square distribution for 1 degree

of freedom (Rogerson, 2001).

Page 224: Ground Penetrating Radar for the parameterisation of

Andrew Howe

224

Figure 7.15. Frequency distributions of plot overland flow (m3/hr).

M e as ure d plot runoff

0

100

200

300

400

5002.

5E-0

5

1.0E

-04

5.0E

-04

1.0E

-03

8.0E

-03

5.0E

-02

5.0E

-01

GPR var iable s oil de pth s ce nar io

0

100

200

300

400

500

2.5E

-05

1.0E

-04

5.0E

-04

1.0E

-03

8.0E

-03

5.0E

-02

5.0E

-01

0.2m s oil de pth s ce nar io

0

100

200

300

400

500

2.5E

-05

1.0E

-04

5.0E

-04

1.0E

-03

8.0E

-03

5.0E

-02

5.0E

-01

0.4m s oil de pth s ce nar io

0

100

200

300

400

500

2.5E

-05

1.0E

-04

5.0E

-04

1.0E

-03

8.0E

-03

5.0E

-02

5.0E

-01

0.6m s oil de pth s ce nario

0

100

200

300

400

500

2.5E

-05

1.0E

-04

5.0E

-04

1.0E

-03

8.0E

-03

5.0E

-02

5.0E

-01

0.8m s oil de pth s ce nar io

0

100

200

300

400

500

2.5E

-05

1.0E

-04

5.0E

-04

1.0E

-03

8.0E

-03

5.0E

-02

5.0E

-01

1.0m s oil de pth s ce nar io

0

100

200

300

400

500

2.5E

-05

1.0E

-04

5.0E

-04

1.0E

-03

8.0E

-03

5.0E

-02

5.0E

-01

1.2m s oil de pth s ce nar io

0

100

200

300

400

500

2.5E

-05

1.0E

-04

5.0E

-04

1.0E

-03

8.0E

-03

5.0E

-02

5.0E

-01

1.4m s oil de pth s ce nar io

0

100

200

300

400

500

2.5E

-05

1.0E

-04

5.0E

-04

1.0E

-03

8.0E

-03

5.0E

-02

5.0E

-01

1.6m s oil de pth s ce nar io

0

100

200

300

400

500

2.5E

-05

1.0E

-04

5.0E

-04

1.0E

-03

8.0E

-03

5.0E

-02

5.0E

-01

Page 225: Ground Penetrating Radar for the parameterisation of

Andrew Howe

225

7.6.2 Subsurface flow

Modelled subsurface flow (figure 7.16) follows a similar pattern to the

catchment hydrograph and shows continuous flow occurring. Measured

subsurface flow (figure 7.9) is instead event based, with discrete time periods

during which no flow is recorded for up to several weeks. Within the measured

record, a period of zero flow is recorded between 26th March 1998 and 21st July

1998, following a large rainfall event. A subsequent site visit found that the

equipment funnel was blocked, resulting in any subsurface flow bypassing the

tipping bucket. This period of data is therefore suspect and given the frequency

of subsurface flow occurring throughout the remaining record, is not used in

subsequent analysis.

Figure 7.16. Modelled plot cell subsurface flow for three soil depth scenarios.

Page 226: Ground Penetrating Radar for the parameterisation of

Andrew Howe

226

Figure 7.17 shows the trend in subsurface flow for all model scenarios and

recorded data. Compared with the modelled overland flow results, modelled

subsurface flow values are of a similar magnitude to those measured. Reducing

cell soil thickness results in a linear reduction in average flow from the cell

while the maximum simulated flow within the record is approximately

0.005m3/hr for depths ranging from 1.6m to 0.4m.

Figure 7.17. Mean and maximum subsurface flow for all soil depth scenarios andmeasured data.

The probability distribution functions for recorded and modelled plot cell

throughflow are shown in figure 7.18. The results show that the effect of

increasing model soil thickness is to increase the total volume of cell

throughflow but to reduce the minimum flow value simulated over the time

period. The frequency distribution of measured flow is skewed to the lower

range of hourly flow totals, with the greatest frequency of observed throughflow

occurring between 0 and 0.00005m3/hr which corresponds to the range in which

a single bucket tip (0.000047m3/hr) is located.

Page 227: Ground Penetrating Radar for the parameterisation of

Andrew Howe

227

Figure 7.18. Frequency distributions of plot throughflow (m3/hr).

M e as ure d plot throughflow

0

100

200

300

400

5.0E

-05

1.0E

-04

2.5E

-04

5.0E

-04

7.5E

-04

1.0E

-03

2.5E

-03

5.0E

-03

7.5E

-03

1.0E

-02

5.0E

-02

GPR m e as ure d s oil de pth s ce nar io

0100020003000400050006000

5.0E

-05

2.5E

-04

7.5E

-04

2.5E

-03

7.5E

-03

5.0E

-02

0.2m s oil de pth s ce nar io

0100020003000

400050006000

5.0E

-05

2.5E

-04

7.5E

-04

2.5E

-03

7.5E

-03

5.0E

-02

0.4m s oil de pth s ce nar io

0100020003000400050006000

5.0E

-05

2.5E

-04

7.5E

-04

2.5E

-03

7.5E

-03

5.0E

-02

0.6m s oil de pth s ce nario

0100020003000400050006000

5.0E

-05

2.5E

-04

7.5E

-04

2.5E

-03

7.5E

-03

5.0E

-02

0.8m s oil de pth s ce nar io

0100020003000400050006000

5.0E

-05

2.5E

-04

7.5E

-04

2.5E

-03

7.5E

-03

5.0E

-02

1.0m s oil de pth s ce nario

0100020003000400050006000

5.0E

-05

2.5E

-04

7.5E

-04

2.5E

-03

7.5E

-03

5.0E

-02

1.2m s oil de pth s ce nar io

0100020003000400050006000

5.0E

-05

2.5E

-04

7.5E

-04

2.5E

-03

7.5E

-03

5.0E

-02

1.4m s oil de pth s ce nar io

0100020003000400050006000

5.0E

-05

2.5E

-04

7.5E

-04

2.5E

-03

7.5E

-03

5.0E

-02

1.6m s oil de pth s ce nar io

0100020003000400050006000

5.0E

-05

2.5E

-04

7.5E

-04

2.5E

-03

7.5E

-03

5.0E

-02

Page 228: Ground Penetrating Radar for the parameterisation of

Andrew Howe

228

The mean of the distributed GPR measured soil thickness of the plot cell is

0.20m. This is equivalent to the value of 0.2m depth for the lumped scenario

and it would therefore seem logical that the two scenarios would produce

equivalent results, given that no upslope cells drain into the plot cell. This is not

observed in model subsurface flow results however, where the mean flow

generated by the variable depth model is double that of the lumped 0.2m depth

scenario. Since the plot cell soil thickness are equal in both scenarios and no

cells are located upslope of the plot cell this response must be due to

downstream grid cells.

The influence of downstream cells on subsurface flow is controlled by the

model routine calculating hydraulic gradient. As explained in chapter 3, the

movement of subsurface water between cells is calculated using D’Arcy’s law

(equation 3.9) such that flow is a function of soil hydraulic conductivity and the

hydraulic gradient between cells. In this model hydraulic conductivity is based

on Campbell’s 1974 formula (equation 3.3) which requires soil texture and bulk

density as input parameters. Texture and bulk density values were based on the

measurement of 68 soil samples taken from 19 sites across the catchment and is

assigned as a lumped model parameter. Model hydraulic gradient is calculated

as the difference in water table elevation above sea level between a cell and its

downstream neighbour in the subsurface drainage network. It is a distributed

parameter based on cell soil moisture and soil thickness. Cell water table depth

is calculated by the model for each time step based on the amount of soil water

in the profile and the depth – porosity relationship derived from field samples.

When a cell and the downstream cell are fully saturated, i.e. water table depth

equals zero, hydraulic gradient is the difference in cell elevation divided by the

horizontal distance between cells. The depth of soil in a cell controls the

maximum soil water table depth and is one important control on hydraulic

gradient, along with surface elevation differences between cells. Soil thickness

therefore affects the potential rate of subsurface flow generation within cells,

Page 229: Ground Penetrating Radar for the parameterisation of

Andrew Howe

229

although actual flow to the downstream cell is also controlled by the hydraulic

conductivity of the soil.

The Kruskal-Wallis 1-way ANOVA was again applied to compare each

modelled distribution of throughflow with that observed at the field plot. The

results are presented in table 7.6.

Table 7.6. Kruskal-Wallis results applied to frequency distributions of field plotsubsurface flow.

Modelled soildepth scenario

Kruskal-Wallis test value 1

p - value H0

Variable depth 2.89 0.089 Accept

0.2m depth 0.00 0.977 Accept

0.4m depth 2.85 0.091 Accept

0.6m depth 8.90 0.003 Reject

0.8m depth 10.78 0.001 Reject

1.0m depth 7.07 0.008 Reject

1.2m depth 5.52 0.019 Reject

1.4m depth 4.35 0.037 Reject

1.6m depth 4.35 0.037 Reject

1 Critical test statistic value 3.841 (p = 0.05) using chi square distribution for 1 degree

of freedom (Rogerson, 2001).

Page 230: Ground Penetrating Radar for the parameterisation of

Andrew Howe

230

7.6.3 Plot soil moisture dynamics

Theta probe voltage measurements were used to derive VMC at the field plot by

solving equation 4.2. The hydrological model, in its current format, simulates

the depth of water, not the VMC of each cell. The model is designed in this

manner because the actual depth of water is, for this model, a more useful value

for the calculation of model water table depth, subsurface flow and infiltration

rate each timestep. VMC can however be easily calculated from model

generated cell soil moisture using the following relationship,

(7.1)

Applying this transformation to modelled moisture data allows direct

comparison between theta probe and model simulated VMC. In this section the

temporal variation in model plot cell VMC is contrasted with that obtained by

the three theta probes installed at the same location. The model scenario used is

one of variable soil thickness. The results obtained are shown in figure 7.19.

CELL

CELLCELL soildepth

erdepthofwatVMC =

Page 231: Ground Penetrating Radar for the parameterisation of

Andrew Howe

231

Figure 7.19. Field plot measured and modelled VMC for distributed soil thicknessscenario.

Page 232: Ground Penetrating Radar for the parameterisation of

Andrew Howe

232

The theta probe located at 0.08m below the ground surface maintains a

relatively constant VMC of 0.55 – 0.56 m3/m3 from record start on 28th

September 1998 to 14th March 1999. Subsequently near-surface VMC exhibits a

greater range of VMC values as precipitation is reduced and evapo-transpiration

increases in the summer months, until the record end on 7th July 1999.

Theta probes located at 0.23m and 0.37m below ground level follow a pattern of

year round fluctuation and do not exhibit the plateau characteristic of the

shallow probe during the winter months. Both however show an increased

frequency of wetting events during the first 3000 hours of records compared

with the more pronounced periods of soil drying evident in the latter 4000 hours

of records as precipitation decreases and evapo-transpiration increases during

the spring and summer period.

Modelled field plot VMC is an integrated value of moisture across the entire

soil profile from bedrock to soil surface. Table 7.7 summarises the key changes

in VMC with different soil thickness scenarios. As depth is increased the

average value of VMC for the cell is reduced and the range of values within the

record also lessens. The model predicts that deeper soils result in a less

responsive VMC trace and an overall reduction in VMC due to the dispersal of

water over a deeper soil profile. It is however the variable soil thickness and

0.2m depth model scenarios which predict mean values closest to the mean

theta probe VMC for the duration of the record. The range of VMC measured

over the period is less well simulated, with the model predicting VMC greater

than that recorded during wetting cycles whilst the reverse occurs during

periods of drying. In this case the modelled response of VMC for the variable

depth scenario follows a pattern similar in frequency to that of the two deeper

probes than the shallower probe. The soil drying curve is less well simulated

because the model initially reduces VMC at a rate less than that recorded, but

continues to reduce VMC below the level at which observed VMC stabilised,

approximately 0.45 m3/m3. The effect is of a modelled soil hydrograph with a

Page 233: Ground Penetrating Radar for the parameterisation of

Andrew Howe

233

greater range of VMC values, but the correct pattern of response to

precipitation.

Thresholds could be implemented within the model to prevent modelled VMC

from exceeding the limits of VMC observed in the field. The problem may be

linked to the simplistic depth – porosity relationship derived from soil samples

taken from sites across the catchment, not only from the field plot.

Implementation of an upper VMC value for this model would result in reduced

maximum soil water storage as pore volume in the soil profile becomes less.

The frequency and magnitude of overland flow would increase as cells fully

saturate for lower amounts of rainfall. Application of a lower VMC threshold

would limit cell subsurface drainage during periods of low rainfall, resulting in

an event rather than continuous subsurface flow record. However the

implementation of thresholds would be based on observational data rather than

the physical basis of soil water behaviour and has not been attempted in this

thesis.

Table 7.7. Modelled and measured VMC summary statistics for the field plot.

Modelled Values Mean VMC (m3/m

3) VMC range (m

3/m

3)

0.2m soil thickness 0.507 0.440

0.4m soil thickness 0.479 0.323

0.6m soil thickness 0.449 0.256

0.8m soil thickness 0.421 0.214

1.0m soil thickness 0.395 0.184

1.2m soil thickness 0.372 0.162

1.4m soil thickness 0.349 0.145

1.6m soil thickness 0.329 0.131

Variable soil thickness 0.490 0.438

Measured Values Mean VMC (m3/m

3) VMC range (m

3/m

3)

Probe at 0.37m depth 0.494 0.210

Probe at 0.23m depth 0.478 0.123

Probe at 0.08m depth 0.541 0.096

Average theta probe VMC 0.504 0.143

Page 234: Ground Penetrating Radar for the parameterisation of

Andrew Howe

234

7.7 Catchment soil moisture dynamics and GPR

The spatio-temporal dynamics of soil moisture are an intrinsic component of

this hydrological model. Soil moisture partially determines the magnitude of

evaporation (equation 3.2), overland flow generation and subsurface flow,

through the relationship between soil moisture and hydraulic conductivity

(equation 3.13). GPR offers a relatively fast and non-invasive method for the

internal validation of modelled soil moisture using GPR measured values.

From analysis of CMP results (chapter 5), 29 GPR derived VMC measurements

were available for 29 individual sites within the catchment. The data was

collected during two periods of fieldwork during October 1998 and June/July

1999. The VMC calculated by the variable soil thickness model was used to

compare the VMC measured by GPR at the corresponding site in the catchment.

In order to compare between GPR and model VMC for each sampled grid cell

the model soil moisture result closest to the time at which the GPR

measurement were taken was used as the source of VMC data. Figure 7.20.

shows the resulting plot of GPR VMC against model VMC, site details and

results are listed in table II.3, appendix II.

Page 235: Ground Penetrating Radar for the parameterisation of

Andrew Howe

235

Figure 7.20. GPR measured VMC and variable depth scenario VMC for 29 sitesacross the Cyff. The trend line is fitted to a 1:1 relationship, not through datapoints.

Figure 7.20 shows that an increase in GPR measured VMC are linearly related

to the VMC for the associated model grid cell. The largest deviation from the

1:1 line (site 60) recorded an RMS velocity of 0.08m/ns, the highest value of the

data set, and is probably an erroneous figure resulting from the complexity of

CMP – velocity processing.

Those points with the greatest deviation from the ordinate are located above the

line. Data points that plot above the line indicate that modelled VMC for the

cell is greater than the VMC measured in a subsection of the grid cell using

GPR. Assuming that GPR VMC measured over a 2m-survey line is

representative of VMC over the 25m-grid cell this indicates that specific cells

with the model are storing greater volumes of soil moisture than is physically

measured. Neglecting site 60, the six data points with residuals greater than

Page 236: Ground Penetrating Radar for the parameterisation of

Andrew Howe

236

0.1m3/m3 from the ordinate were measured in the field between 6th – 16th

October 1998. This period coincides with a rapid rise in VMC recorded by field

plot theta probes (figure 7.19) and VMC simulated at the plot cell. The

overestimation of VMC by the model for these six sites indicates that the depth

– porosity relationship, and/or subsurface flow, is poorly modelled at these

locations. The remaining 12 sites at which VMC data was measured over the

same time period have soil VMC values varying between -0.02 to +0.09m3/m3

from the ordinate. The fact that GPR and model VMC correspond for some sites

and not others for the same time period is an indication that the model simulates

some areas of the catchment less well than others.

Although VMC data using GPR was only available for two periods throughout

the model simulation, the results show the potential of GPR to quantify soil

VMC for selected cells in a distributed hydrological model. Divergence of

modelled cell VMC from measured values could indicate areas within the

catchment that required further investigation. Although this analysis has

concentrated on the spatial variability of soil moisture, the quantification of soil

moisture variation for the same grid cells over time is also possible. Divergence

between observed and modelled data would provide information on model

performance at the grid scale for different types of events, such as model

response to storm events or to prolonged periods of drought.

Page 237: Ground Penetrating Radar for the parameterisation of

Andrew Howe

237

7.8 Sensitivity Analysis

Sensitivity analysis is used in modelling to assess the effect of uncertainty in

model parameters and variables on model outputs. The complex nature of many

environmental models requires a mathematical approach since the interactions

between model parameters and their relative effects on model response cannot

be easily understood. The relative importance of different parameters on model

response can be defined using sensitivity analysis (Campolongo & Braddock,

1999; Cryer & Havens, 1999).

In this thesis, model sensitivity analysis has not been formally carried out.

However the model was designed to examine the impact of varying soil depth

on catchment and hillslope scale discharge, and the effect of changes to this

parameter on model output have been clearly shown.

A full sensitivity analysis would (a) throw light upon the manner in which soil

thickness affects at-a-point and spatially integrated hydrological responses

through the various hydrological processes and (b) help understand the

magnitude of soil thickness sensitivity compared with the sensitivity to other

parameters, thereby indicating the significance of soil thickness sensitivity.

This would be an appropriate activity for further work in this research area.

Page 238: Ground Penetrating Radar for the parameterisation of

Andrew Howe

238

7.9 Conclusion

a) No model scenario could simulate peak discharge during storm events and

this is likely to be due to the lack of a macropore subsurface flow

mechanism within the model. Analysis of the recorded hydrograph with all

model scenario discharge shows that shallow soil models are best able to

simulate maximum flows. RMS errors of 0.29m3/s, 0.34m3/s and 0.40m3/s

are associated with soil thickness scenarios of 0.2m, variable and 1.6m.

Shallow soil scenarios perform poorly during periods of low flow, when

predicted discharge approaches zero (RMS error = 0.020m3/s). Deeper soil

scenarios maintain baseflow throughout periods of low rainfall (RMS error

= 0.012m3/s), but are unable to simulate the peak flows generated during

precipitation events. The variable soil thickness model has a range of soil

thickness distributed over the catchment and behaves similarly to shallow

soil models during high flows whilst simulating well intermediate and low

flows, with an RMS error of 0.009m3/s for the event studied.

b) The model is unable to simulate plot overland flow when implemented at a

25m-grid size probably because hillslope overland flow occurs at sub-grid

scales. Large scale grids poorly define the area contributing to flow, but

smaller grids have an improved capacity to model the flow paths and

network of contributing areas. A reduction from a 25m to a 5m-grid size

resulted in a greater frequency and magnitude of modelled overland flow

events. A finer grid also resulted in the creation of upslope cells and is

approaching the scales at which observed overland flow occurred.

Increasing soil thickness at this scale resulted in a decline in cell overland

flow.

c) In this model subsurface flow is affected not only by the soil thickness of

the cell examined but also by downstream cells. This is in response to the

calculation of hydraulic gradient, which drives flow and depends both on

cell elevation and cell water table depth from the soil surface. With all other

Page 239: Ground Penetrating Radar for the parameterisation of

Andrew Howe

239

parameters constant, similar depths of soil at the plot cell produce different

subsurface flow because of the difference in soil thickness of the

downstream cell.

d) Modelled subsurface flow occurs continuously in all depth scenarios at the

25m-grid size. Actual subsurface flow is event based, but of a similar

magnitude to modelled results. Soil moisture modelled using the variable

depth model has a similar wetting frequency and mean VMC, but the range

of observed values is smaller than those modelled. Both the range of VMC

and the frequency of subsurface flow could be curtailed using upper and

lower thresholds for flow and VMC, these thresholds would be without

physical basis and may reflect the need for an improved, distributed soil

thickness – porosity relationship. This is currently a lumped parameter

within the model.

Page 240: Ground Penetrating Radar for the parameterisation of

Andrew Howe

240

Chapter 8: Conclusions

This thesis has used GPR as a tool for the measurement of two hydrological

parameters across a catchment; soil thickness and soil moisture. However it is

recognised that because of the complex nature of GPR output, some form of

physical measurement of depth and water content is required to calibrate and

validate GPR data.

GPR was successfully applied to the Cyff catchment to measure the variation of

soil thickness at 30 sites, and soil moisture at 29 sites. Using a simple, GIS,

physically-based model with an emphasis on subsurface hydrology, GPR was

used along with physical measurements to parameterise the model for

distributed soil depth and subsequently to perform an internal state validation of

individual cell soil moisture for two periods over the model simulation period.

The results show an improved method for soil thickness and moisture

measurement, with GPR used in conjunction with other techniques to

interpolate between extractive measurements. Correlation of the variability

between GPR and physical measurements of soil thickness showed that the

GPR method was not biased to either over or under recording soil thickness,

although the magnitude of variation between the two methods increased as

measurement depth increased.

The RMS of the difference between auger and GPR measured soil thickness

was calculated as 0.04m to the A-B horizon boundary and 0.12m to the soil-

bedrock B-C boundary. The increase in thickness variation with greater depths

of sampling probably results from the cumulative effects of errors in the

calculation of GPR velocity values, and because the soil-bedrock interface is a

gradual, rather than abrupt boundary.

The observed relationship between GPR soil thickness and TOPMODEL

wetness index, derived from a 25m-grid cell DEM, enabled the prediction of

Page 241: Ground Penetrating Radar for the parameterisation of

Andrew Howe

241

soil thickness over the entire catchment. Spatially distributed soil thickness in

conjunction with 55 soil bulk density measurements allowed quantification of

the maximum soil moisture storage capacity for each cell within the catchment.

Soil moisture was derived from CMP radar surveys. Subsurface velocities were

converted to dielectric values and subsequently volumetric moisture content

using standard methods. Validation of GPR measured soil moisture was

provided by theta probe and gravimetric measurements. Because the volume

sampled by GPR in this study is of the order of 0.1-5.0m3, compared to <0.1m3

for point samples, comparison between data sets required scaling up. This was

achieved using a method of integration to sum total soil moisture over a known

depth given the relationship between sample depth and theta probe and

gravimetric moisture. Results show a high degree of correlation between theta

probe and GPR measured VMC (r2 = 0.80, RMS error = 0.03m3/m3) and

gravimetric and GPR measured VMC (r2 = 0.80, RMS error = 0.05m3/m3).

The results of multiple model simulations run using 18-months of hydrological

data show that soil thickness is an important control both on total catchment

discharge and cell overland/subsurface flow frequency and magnitude. Initially,

when considering soil thickness to be lumped at the catchment scale, shallow

soils reproduce hydrograph peaks more accurately than deeper soil scenarios.

The reverse is true during periods of reduced precipitation, when deeper soil

scenarios simulate the decline in river flow more accurately. This behaviour

results from the impact of soil thickness on the different amounts of soil

moisture that can be stored during each simulation.

The distributed GPR derived model of catchment soil thickness contains,

unsurprisingly, an element of both shallow and deep soil models. This enables

the model to simulate low flows and to an extent the peak flows recorded. One

reason for the limited success in simulating peak flows may be the lack of a

model component to simulate subsurface macropore flow, which would rapidly

contribute soil water to the stream network.

Page 242: Ground Penetrating Radar for the parameterisation of

Andrew Howe

242

A comparison between GPR measured and the distributed soil thickness model

scenario VMC shows the ability of GPR to perform an internal state validation

of soil moisture. Site GPR VMC were plotted against model VMC for the same

location and time to compare results. Divergence from the expected pattern

indicates potential areas and time periods within the catchment that are poorly

simulated, requiring more detailed investigation and improved parameterisation.

Future work could also include verification of modelled soil thickness at other

locations within the catchment not measured as part of this study.

Page 243: Ground Penetrating Radar for the parameterisation of

Andrew Howe

243

Bibliography

Abbott, M.B. Bathurst, J.C. Cunge, J.A. O’Connell, P.E. Rasmussen, J. 1986.An introduction to the European Hydrological System – SystèmeHydrologique Europèen, SHE, 2: structure of a physically-based, distributedmodelling system. Journal of Hydrology 87: 61-77.

Adams, W.A. 1974. Pedogenesis of soils derived from Lower Palaeozoicsediments in mid-Wales. Welsh Soils Discussion Group Annual Report 15:61-69.

Alonso, M. Finn, E.J. 1992. Physics. Addison-Wesley Publishers Ltd.

Anderson, M.G. Burt, T.P. 1990. Subsurface Runoff. In: Process Studies inHillslope Hydrology. Eds. Anderson, M.G. and Burt, T.P. Wiley; 365-400.

Annan, A.P. Davis, J.L. 1977. Radar range analysis for geological materials.PEMD #10. Sensors & Software Inc., Mississauga, Ontario.

Annan, A.P. Cosway, S.W. Redman, J.D. 1991a. Water table detection withground penetrating radar. Expanded Abstracts. Society of ExplorationGeophysicists sixty-first annual meeting, Houston, Texas. PEMD #60.Sensors & Software Inc., Mississauga , Ontario.

Annan, A.P. Bauman, P. Greenhouse, J.P. Redman, J.D. 1991b. Geophysics andDNAPLs. PEMD #69. Sensors & Software Inc., Mississauga, Ontario.

Annan, A.P. Cosway, S.W. 1992. Ground-penetrating radar survey design:Proc. Of the Symp. On the Application of Geophysics to Eng. And Environ.Problems (SAGEEP).

Annan, A.P. 1996. Transmission dispersion and GPR. Journal of ExplorationGeophysics 1: 125-136.

Annan, A.P. 1997. Ground Penetrating Radar Workshop Notes. Sensors &Software Inc., Mississauga, Ontario.

Theta probe User Manual. 1998. Theta probe soil moisture sensor. User ManualML2-UM-1, Delta-T Devices Ltd. Cambridge, UK.

Beres, M. Haeni, F.P. 1991. Application of ground-penetrating radar methods inhydrogeologic studies. Ground Water 29: 375-386.

Beven, K.J. Kirkby, M.J. Schofield, N. Tagg, A.F. 1984. Testing a physically-based flood forecasting model TOPMODEL for three U.K. catchments.Journal of Hydrology 69: 119-143.

Page 244: Ground Penetrating Radar for the parameterisation of

Andrew Howe

244

Beven, K.J. 1985. Distributed Models. In: Hydrological Forecasting. Anderson,M.G. and Burt, T.P. (Ed.). Wiley, 405-436.

Beven, K.J. 1989. Changing ideas in hydrology. The case of physically basedmodels. Journal of Hydrology 105: 157-172.

Beven, K. J. Binley, A. 1993a. The future of distributed models: Modelcalibration and uncertainty prediction. In: Terrain analysis and distributedmodelling in hydrology. Beven, K.J. and Moore, I.D. (Ed.). Wiley; 227-246.

Beven, K.J. 1993b. Prophecy, reality and uncertainty in distributed hydrologicalmodelling. Advances in Water Resources 16: 41-51.

Beven, K. J. 1994. The limits of splitting: Hydrology. NATO ARW onEcosystem modelling: Delineating the possible from the impossible.Unpublished report.

Binley, A.M. Beven, K.J. 1991. Physically based modelling of catchmenthydrology: a likelihood approach to reducing predictive uncertainty. In:Computer modelling in the environmental sciences. Farmer, D.G. andRycroft, M.J. (Ed.). Clarendon.

Binley, A. Beven, K. 1993. Three-dimensional modelling of hillslopehydrology. In: Terrain analysis and distributed modelling in hydrology.Beven, K.J. and Moore, I.D. (Ed.). Wiley, 107-119.

Blackie, J.R. Eeles, C.W.O. 1985. Lumped Catchment Models. In: HydrologicalForecasting. Anderson, M.G. and Burt, T.P. (Ed.). Wiley; 311-346.

Blöschl, G. Sivapalan, M. 1995. Scale issues in hydrological modelling: Areview. In: Scale Issues in Hydrological Modelling. Wiley; 9-48.

Boer, M. Del Barrio, G. Puigdefábregas, J. 1996. Mapping soil depth classes indry Mediterranean areas using terrain attributes derived from a digitalelevation model. Geoderma 72: 99-118.

Brewster, M.L. Redman, J.D. Annan, A.P. 1992. Monitoring a controlledinjection of perchloroethylene in a sandy aquifer with ground penetratingradar and time domain reflectometry. Proceedings of the Symposium on theApplication of Geophysics to Engineering and Environmental Problems.Oakbrook, Illinios; 611-618.

Brubaker, S.C. Jones, A.J. Frank, K. Lewis, D.T. 1994. Regression models forestimating soil properties by landscape position. Soil Sci. Soc. Am. J. 58:1763-1767.

Page 245: Ground Penetrating Radar for the parameterisation of

Andrew Howe

245

Bruneau, P. Gascuel-Odoux, C. Robin, P. Merot, P. Beven, K.J. 1995.Sensitivity to space and time resolution of a hydrological model using digitalelevation data. Hydrological Processes 9: 69-81.

Burrough, P.A. McDonnell, R.A. 1998. Principles of Geographical InformationSystems. Oxford University Press.

Burt, T.P. Butcher, D.P. 1985. Topographic controls of soil moisturedistributions. Journal of Soil Science 36: 469-486.

Campbell, G.S. 1974. A simple method for determining unsaturatedconductivity from moisture retention data. Soil Science 117: 311-314.

Campbell, G.S. 1985. Soil Physics with BASIC. Transport Models for Soil-Plant Systems. Developments in Soil Science 14. Department of Agronomyand Soils, Washington State University. Elsevier: New York.

Campolongo, F. Braddock, R. 1999. Sensitivity Analysis of the IMAGEGreenhouse model. Environmental Modelling & Software 14: 275-282.

Chanzy, A. Tarussov, A. Judge, A. Bonn, F. 1996. Soil water contentdetermination using a digital ground-penetrating radar. Soil Sci. Soc. Am. J.60: 1318-1326.

Chappell, N. Ternan, L. 1993. Flow path dimensionality and hydrologicalmodelling. In: Terrain analysis and distributed modelling in hydrology.Beven, K.J. & Moore, I.D. (Ed.). Wiley, 121-140.

Clarke, R.T. Leese, M.N. Newson, A.J. 1973. Analysis of data from Plynlimonraingauge networks: April 1971 – March 1973. Report 27. NaturalEnvironment Research Council - Institute of Hydrology, Wallingford, UK.

Collins, M.E. Doolittle, J.A. 1987. Using ground-penetrating radar to study soilmicrovariability. Soil Sci. Soc. Am. J. 51: 491-493.

Collins, M.E. Doolittle, J.A. Rourke, R.V. 1989. Mapping depth to bedrock on aglaciated landscape with ground-penetrating radar. Soil Sci. Soc. Am. J. 53:1806-1812.

Cryer, S.A. & Havens, P.L. 1999. Regional sensitivity analysis using afractional method for the USDA model GLEAMS. Environmental Modelling& Software 14: 613-624.

Daniels, D.J. Gunton, D.J. Scott, H.F. 1988. Introduction to subsurface radar.IEE Proceedings 135: Part F, 278-319.

Daniels, R.B. Hammer, R.D. 1992. Soil Geomorphology. John Wiley & Sons,Inc. New York.

Page 246: Ground Penetrating Radar for the parameterisation of

Andrew Howe

246

Davis, J.L. Annan, A.P. 1985. High resolution shallow soundings using radarand reflection seismic methods. PEMD #29. Sensors & Software Inc.,Mississauga, Ontario.

Davis, J.L. Annan, A.P. 1989. Ground-penetrating radar for high-resolutionmapping of soil and rock stratigraphy. Geophysical Prospecting 37: 531-551.

De Roo, A.P.J. 1998a. Modelling runoff and sediment transport in catchmentsusing GIS. Hydrological Processes 12: 905-922.

De Roo, A.P.J. Wesseling, C.G. Van Deursen, W.P.A. 1998b. Physically-basedRiver Basin Modelling Within a GIS: The LISFLOOD Model.http://www.geog.port.ac.uk/geocomp/geo98/06gc_06.htm

Dix, C.D. 1955. Seismic velocities from surface measurements. GeophysicsXX: 68-86.

Doolittle, J.A. Collins, M.E. 1995. Use of soil information to determineapplication of ground penetrating radar. Journal of Applied Geophysics 33:101-108.

Drayton, R.S. Wilde, B.M. Harris, J.H.K. 1993. Geographical informationsystem approach to distributed modelling. In: Terrain analysis and distributedmodelling in hydrology. Beven, K.J. & Moore, I.D. (Ed.). Wiley; 361-368.

Du, S. Berktold, A. Rummel, P. 1994. The wave nature and propagationcharacteristics of the ground wave in GPR. Minutes for the Symposium forElectromagnetic Depth Exploration, Odenwald 1994. ISSN 0946-7467.

Ellis, S. Mellor, A. 1995. Soils and Environment. Routledge PhysicalEnvironment Series, Routledge, London.

Engman E.T, Chauhan N. 1995. Status of microwave soil-moisturemeasurements with remote- sensing. Remote Sensing of Environment 51: 189-198

Eppstein, M.J. Dougherty, D.E. 1998. Efficient three-dimensional inversion:Soil characterization and moisture monitoring from cross-well ground-penetrating radar at a Vermont test site. Water Resources Research 34: 1889-1900.

Fisher, S.C. Stewart, R.R. & Jol, H.M. 1996. Ground Penetrating Radar (GPR)Data Enhancement Using Seismic Techniques. JEEG 1: 89-96.

Gilson, E.W. Redman, J.D. Pilon, J. Annan, A.P. 1996. Near surfaceapplications of borehole radar. PEMD #126. Sensors & Software Inc.,Mississauga, Ontario.

Page 247: Ground Penetrating Radar for the parameterisation of

Andrew Howe

247

Gilman, K. Newson, M.D. 1980. Soil pipes and pipeflow - a hydrological studyin upland Wales. BGRG Research Monograph 1. Geo Abstracts, University ofEast Anglia, Norwich.

Greaves, R.J. Lesmes, D.P. Lee, J.M. Toksöz 1996. Velocity variations andwater content estimated from multi-offset, ground-penetrating radar.Geophysics 61: 683-695.

Green, W.H. Ampt, G.A. 1911. Studies in soil physics I. The flow of air andwater through soils. J. Agric. Sci 4: 1-24.

Heimsath, A.M. Dietrich, W.E. Nishiizumi, K. Finkel, R.C. 1999. Cosmogenicnuclides, topography, and the spatial variation of soil depth. Geomorphology27: 151-172.

Hirsch, R.M. Helsel, D.R. Cohn, T.A. & Gilroy, E.J. 1993. Statistical analysisof hydrologic data. In: Handbook of Hydrology. Maidment, D.R. (Ed.).McGraw-Hill Inc.

Horton, R.E. 1933. The role of infiltration in the hydrological cycle. Trans. Am.Geophys. Union 14: 446-460.

Hubbard, S.S. Rubin, Y. Majer, E. 1997. Ground-penetrating-radar-assistedsaturation and permeability estimation in bimodal systems. Water ResourcesResearch 33: 971-990.

Hudson, J.A. 1988. The contribution of soil moisture storage to the waterbalances of upland forested and grassland catchments. Hydrological SciencesJournal 33: 289-309.

Jenny, H. 1941. Factors in Soil Formation. McGraw-Hill, New York.

Kalma, J. Bates, B. Woods, R. 1995. Predicting catchment-scale soil moisturestatus with limited field measurements. Hydrological Processes 9: 445-467.

Kirkby, M.J. & Chorley, R.J. 1967. Throughflow, overland flow and erosion.Bulletin of the International Association of Scientific Hydrology 12: 5-21.

Klemes, V. 1986. Operational testing of hydrological simulation models.Hydrological Sciences Journal 31: 13-24.

Knoll, M.D. Knight, R. 1994. Relationships between dielectric andhydrogeologic properties of sand-clay mixtures. Proceedings of the 5th

International Conference on Ground Penetrating Radar. Kitchener, Ontario.45-61.

Page 248: Ground Penetrating Radar for the parameterisation of

Andrew Howe

248

Knoll, M.D. 1996. A petrophysical basis for ground penetrating radar and veryearly time electromagnetics: Electrical properties of sand-clay mixtures.Unpublished PhD Thesis, University of British Columbia.

Kung, K. Lu, Z.B. 1993. Using ground-penetrating radar to detect layers ofdiscontinuous dielectric constant. Soil Sci. Soc. Am. J. 57: 335-340.

Maidment, D.R. 1993. GIS and Hydrological Modeling. In: EnvironmentalModeling With GIS. Goodchild, Steyaert & Parks. (Ed.). Oxford Press, NewYork, 147-167.

McDonnell, R.A. 1996. Including the spatial dimension: Using geographicalinformation systems in hydrology. Progress in Physical Geography, 20: 159-177.

McNeill, J.D. 1980. Electrical conductivity of soils and rocks. Technical NoteTN-5, Geonics Ltd, Ontario.

Moore, I.D. Gessler, P.E. Nielsen, G.A. Peterson, G.A. 1993a. Soil attributeprediction using terrain analysis. Soil Sci. Soc. Am. J. 57: 443-452.

Moore, I.D. Turner, A.K. Wilson, J.P. Jenson, S.K. Band, L.W. 1993b. GIS andland-surface-subsurface process modeling. In: Environmental Modeling WithGIS. (Ed.). Goodchild, Steyaert and Parks. Oxford Press, New York; 196-230.

Moore, I.D. 1996. Hydrologic Modeling and GIS. In: GIS and EnvironmentalModeling: Progress and Research Issues. Goodchild, Steyaert & Parks. (Ed.).World Books; 143-148.

Mulligan, M. Thornes, J.B. In press. Catena verses cellular approaches tohydrological modelling through the plant community in seasonally dry anddry environments. Journal of Hydrology

NERC, AA & Crown Copyright. 1999. Flood Estimation Handbook CD-ROM.Version 1.0. CEH Institute of Hydrology, Wallingford, Oxfordshire, UK.

Newson, M.D. 1976. The physiography, deposits and vegetation of thePlynlimon catchments. Report 30. Natural Environment Research Council -Institute of Hydrology, Wallingford, UK.

Newson, M.D. Harrison, J.G. 1978. Channel studies in the Plynlimonexperimental catchments. Report 47. Natural Environment Research Council -Institute of Hydrology, Wallingford, UK.

O’Loughlin, E.M. 1990. Perspectives on hillslope research. In: Process Studiesin Hillslope Hydrology. Anderson, M.G. and Burt, T.P. (Ed.). Wiley; 501-516.

Page 249: Ground Penetrating Radar for the parameterisation of

Andrew Howe

249

Orlandini, S. Mancini, M. 1996. Local contributions to infiltration excess runofffor a conceptual catchment scale model. Water Resources Research 32: 2003-2012.

Philip, J.R. 1957. The theory of infiltration. Soil Science 84: 163-177.

Pietroniro, A. Leconte, R. 2000. A review of Canadian remote sensingapplications in hydrology, 1995-1999. Hydrological Processes 14: 1641-1666.

Powers, M.H. 1997. Modeling frequency-dependent GPR. The Leading Edge,SEG, 1657-1662.

Quinn, P. Beven, K. Chevallier, P. Planchon, O. 1993a. The prediction ofhillslope flow paths for distributed hydrological modelling using digitalterrain models. In: Terrain analysis and distributed modelling in hydrology.Beven, K.J. and Moore, I.D. (Ed.). Wiley; 63-83.

Quinn, P.F. Beven, K.J. 1993b. Spatial and temporal predictions of soil moisturedynamics, runoff, variable source areas and evapotranspiration for Plynlimon,mid-Wales. Hydrological Processes 7: 425-448.

Quinn, P.F. Beven, K.J. Lamb, R. 1995. The lnα/tanβ index: How to calculate itand how to use it within the TOPMODEL framework. HydrologicalProcesses 9: 161-182.

Rawis, W.J. Lajpat, R.A. Brakensiek, D.L. Shirmohammadi, A. 1992.Infiltration and soil water movement. In: Handbook of Hydrology. Maidment,D.R. (Ed.). McGraw-Hill Inc.

Rea, J. Knight, R. 1998. Geostatistical analysis of ground-penetrating radardata: A means of describing spatial variation in the subsurface. WaterResources Research 34: 329-339.

Redman, J.D. DeRyck, S.M. Annan, A.P. 1994. Detection of LNAPL pools withGPR: Theoretical modelling and surveys of a controlled spill. PEMD #155.Sensors & Software Inc., Mississauga, Ontario.

Refsgaard, J.C. Knudsen, J. 1996. Operational validation and intercomparisonof different types of hydrological models. Water Resources Research 32:2189-2202.

Reynolds, J.M. 1997. An Introduction to Applied and EnvironmentalGeophysics. Wiley. Bath.

Rogerson, P.A. 2001. Statistical methods for Geography. Sage Publications Ltd.London.

Page 250: Ground Penetrating Radar for the parameterisation of

Andrew Howe

250

Roth, C.H. Malicki, M.A. Plagge, R. 1992. Empirical-evaluation of therelationship between soil dielectric constant and volumetric water-content asthe basis for calibrating soil-moisture measurements by TDR. Journal of SoilScience 43: 1-13.

Rudeforth, C.C. 1970. Memoirs of the Soil Survey of Great Britain. NorthCardiganshire. Harpenden.

Saulnier, G. Beven, K. Obled, C. 1997. Including spatially variable effectivesoil depths in TOPMODEL. Journal of Hydrology 202: 158-172.

Schmugge, T.J. 1980. Effect of texture on microwave emission from soils. IEEETransactions on Geoscience and remote sensing GE-18: 353-361.

Schmugge, T.J. 1985. Remote sensing of soil moisture. In: HydrologicalForecasting. Anderson, M.G. and Burt, T.P. (Ed.). Wiley; 101-124.

Selby, M.J. 1993. Hillslope Materials and Processes. 2nd Edition. OxfordUniversity Press.

Sensors & Software Inc. 1993. PulseEKKO 1000 User’s Guide Version 1.1.Technical Manual 24. Sensors & Software Inc., Mississauga, Ontario, Canada.

Seyfried, M.S. Wilcox, B.P. 1995. Scale and the nature of spatial variability:Field examples having implications for hydrologic modeling. WaterResources Research 31: 173-184.

Shih, S.F. Doolittle, J.A. 1984. Using radar to investigate organic soil thicknessin the Florida everglades. Soil Sci. Soc. Am. J. 48: 651-656.

Shuttleworth, W.J. 1992. Evaporation. In: Handbook of Hydrology. Maidment,D.R. (Ed.). McGraw-Hill Inc.

Stephenson, D. Meadows, M.E. 1986. Kinematic hydrology and modelling.Developments in Water Science 26. Elsevier, Amsterdam.

Topp, G.C. Davis, J.L. Annan, A.P. 1980. Electromagnetic determination of soilwater content: measurements in coaxial transmission lines. Water ResourcesResearch 16: 574-582.

Topp, G.C. Davis, J.L. 1985. Measurement of soil water content using time-domain reflectometry (TDR): a field evaluation. Soil Sci. Soc. Am. J. 49: 19-24.

Truman, C.C. Perkins, H.F. Asmussen, L.E. Allison, H.D. 1988. Using groundpenetrating-radar to investigate variability in selected soil properties. Journalof Soil and Water Conservation 341-345.

Page 251: Ground Penetrating Radar for the parameterisation of

Andrew Howe

251

Van Oevelen P.J. 1998. Soil moisture variability: A comparison betweendetailed field measurements and remote sensing measurement techniques.Hydrological Sciences Journal-Journal Des Sciences Hydrologiques. 43: 511-520

Van Overmeeren, R.A. Sariowan, S.V. Gehrels, J.C. 1997. Ground penetratingradar for determining volumetric soil water content; results of comparativemeasurements at two test sites. Journal of Hydrology 197: 316-338.

Verhoest, N.E.C. Troch, P.A. Paniconi, C. Troch, F.P. 1998. Mapping basinscale variable source areas from multitemporal remotely sensed observationsof soil moisture behavior. Water Resources Research 43: 3235-3244.

Ward, R.C. Robinson, M. 2000. Principles of Hydrology. 4th Edition. McGraw-Hill.

Watson, E. 1967. The geomorphological factor in upland soil formation. In:Upland Soils, Welsh Soils Discussion Group Report 8. Jenkins, D. (Ed.)

Weiler, K. Steenhuis, T. Boll, J. Kung, K. 1998. Comparison of groundpenetrating radar and time-domain reflectometry as soil water sensors. SoilSci. Soc. Am. J. 62: 1237-1239.

Wensink, W.A. 1993. Dielectric properties of wet soils in the frequency range1-3000 MHz. Geophysical Prospecting 41: 671-696.

Wesseling, C.G. van Deursen, P.A. Burrough, P.A. 1996. A spatial modellinglanguage that unifies dynamic environmental models and GIS.http://www.ncgia.ucsb.edu/conf/SANTA_FE_CDROM/sf_papers/wesseling_cees/santa_fe.html

Western, A.W. Grayson, R.B. Bloschl, G. Willgoose, G.R. McMahon, T.A.1999. Observed spatial organization of soil moisture and its relation to terrainindices. Water Resources Research 35: 797-810.

Whalley, W.R. 1993. Considerations on the use of time domain reflectometry(TDR) for measuring soil water content. Journal of Soil Science 44: 1-9.

Wolock, D.M. Price, C.V. 1994. Effects of digital elevation model map scaleand data resolution on a topography-based watershed model. Water ResourcesResearch 30: 3041-3052.

Xiao, Q.F. Ustin, S.L. Wallender, W.W. 1996. A spatial and temporalcontinuous surface-subsurface hydrologic model. Journal of GeophysicalResearch 101: D23, 29,565-29,584.

Page 252: Ground Penetrating Radar for the parameterisation of

Andrew Howe

252

Appendix I. PCRaster model code and parameter values.

Variable GPR soil thickness. 25m grid# Version id = msd02. # Soil moisture model with evaporation, infiltration, runoff & throughflow modules# Non-linear porosity term included in subsurface calculation# Ksat is a function of bulk density (Campbell, 1985)# Depth = variable (sd02.map)# Drainage LDD calculated each timestep on the basis of hydraulic gradient# Surface routing uses kinematic flow# Model assumes one soil layer of variable depth derived from GPR measurements# Water calculated/input must be as depth NOT area or volume

#-----------------------------------------------------------------------------------------------------------------binding RainTimeSeries=Rain4.txt; # rainfall input timeseries PotEvapTime=PE4.txt; # Potential evapotranspiration time series

InitMoist=imc.map; # initial moisture content, m water /m soil thickness,set as 0.9 maxmoist

soildepth=sd02.map; # soil thickness from GPR - Upslope area results.Forced depth of 0.01m in river network

Ldd=Ldd25.map; # catchment drainage network dem=dtm25.map; # digital elevation model Slope=slope25.map; # slope angle (degrees), must be scalar

timestep=scalar(3600); # timestep in seconds cellwidth=scalar(25); # cell width (m)

beta=scalar(0.6); # kinematic wave parameter, 0.6 = sheet flow q=scalar(0.0); # kinematic wave parameter, side flow, not used MN=scalar(0.15); # mannings n value

sites=allsites.map; # Location of GPR sample sites Outflow=pit25.map; # catchment outflow cell location plot=plot.map; # location of field plot

b=scalar(2.4); # b value from Campbell (1985) mfc=scalar(0.150); # mass fraction of clay mfs=scalar(0.282); # mass fraction of sand

areamap bool25.map; # catchment mask map

timer 1 9999 1; # 9999 step simulation. Hourly. Start 0100 1 Oct 1997 reportdefault = 1+50..endtime; # selective storage of maps on harddisk (every fiftieth

stored) #-------------------------------------------------------------------------------------------------------------initial soilmoist=InitMoist; # Initial soil moisture infilwater=scalar(0); # cell surface water for infiltration equals zero

Page 253: Ground Penetrating Radar for the parameterisation of

Andrew Howe

253

localXS=scalar(0); # cell surface water height is zero (m)

WFdepth=scalar(0); # wetting front depth (m) grad=scalar(tan(slope25.map)); # DEM slope gradient (m/m) ROF=scalar(0); # return overland flow initially zero (m) WTdepth=scalar(0); # water table depth initially zero (m)

# soildepth=scalar(1.6); # Use for soil constant depth scenarios (m) # soil depth = 1.6m maxmoist=(((-0.848/0.757)*exp(-0.757*soildepth))-((-0.848/0.757)*exp(-0.757*0)));# new porosity profile from all bulk density samples (except bog samples)

# maximum total moisture in profile, f(changing porosity with depth)# integral maxmoist gives area under depth-porosity curve per cell (m porosity / m soil)# Non-linear porosity decrease with depth derived from Cyff soil samples.

sat_pc = (soilmoist/maxmoist*100); # Percentage saturation of cell soil

#----------------------------------------------------------------------------------------------------------------dynamic

Rain=timeinputscalar(RainTimeSeries,1); # Rainfall (mm) Raincell=Rain/1000 # Rainfall (m)

Epot=timeinputscalar(PotEvapTime,1); # Potential evaporation (m)

E=Epot/1000*(soilmoist/maxmoist); # Actual evaporation (m)

# Surface water infiltration module

Check0=Raincell+localXS; # Sum inputs this timestep

WFdepth = if(Raincell eq 0, 0, WFdepth); # If no rainfall in previous time step wetting front depth assumed to be zero

# WFdepth = if(WFdepth + (soildepth-WTdepth) ge soildepth,soildepth,WFdepth); # If wetting front plus water table height above bedrock >= soildepth, entire soil profileis saturated

# Calculate porosity at wetting front depth from previous timestep P = 0.848*exp(-0.757*WFdepth);

# Convert porosity to bulk density for input into Campbell (1985) equation BD = (1-P)*2.65;

Ksat=0.004*((1.3/BD)**(1.3*b))*exp(-6.9*mfc-3.7*mfs); # Campbell (1985) equation for Ksat

infilwater=min((Raincell+localXS),Ksat); # infiltration amount this timestep localXS=max((Raincell+localXS)-Ksat,0); # infiltration rate determines localXS Check1=Check0 - (localXS+infilwater); # Check previous inputs equal new

partition of water

Check2=(infilwater+soilmoist) gt maxmoist; # Boolean map of cells where maxmoist is exceeded. Should not occur.

Page 254: Ground Penetrating Radar for the parameterisation of

Andrew Howe

254

localXS=localXS+(max(infilwater+soilmoist-maxmoist,0));

soilmoist=min(infilwater+soilmoist,maxmoist);# if infilwater and current moisture > maxmoist the excess is added to localXS # Calculate new wetting front depth for the next timestep # i.e. real depth of soil taken by infiltrated water WFdepth = infilwater+(-0.848/0.757*exp(-0.757*WFdepth)); WFdepth = WFdepth/(-0.848/0.757); WFdepth = (ln(WFdepth))/-0.757; #-------------------------------------------------------------------------------------------------------- # Create ldd for subsurface flow # dynamic, responds to changes in water table position each timestep

# Initially calculates depth of water table (m) given: # 1. Water input (soilmoist) # 2. Total profile depth # 3. Knowledge of non-linear porosity decrease with depth

WTdepth = soilmoist - ((-0.848/0.757)*exp(-0.757*soildepth)); WTdepth = WTdepth/(0.848/0.757); WTdepth = (ln(WTdepth))/(-0.757);

# this value is with reference to the DEM surface # ie when soilmoist = maxmoist, WTdepth = 0 # and when soilmoist = 0, WTdepth = soildepth

Check3 = WTdepth gt soildepth; # check water table < soil depth

WTldd = lddcreate((dem-WTdepth),1e31,1e31,1e31,1e31);# ldd for soil moisture uses height of water table to calculate drainage net eachtimestep

distance = if(downstreamdist(WTldd) gt 0,downstreamdist(WTldd),cellwidth);# horizontal distance between cells use downstreamdist command except for pit cellswhich must be cellwidth

Hgrad = ((dem-WTdepth)-(downstream(WTldd,dem)-downstream(WTldd,WTdepth)))/distance; # Hydraulic gradient between cells# This gradient includes amount of soil depth contributing to flow so soils do not needto be fully saturated for flow to occur

#------------------------------------------------------------------------------------------------------------ # Subsurface throughflow module # Actual flow out of a cell depends on: # a) amount of potential water available to move, the soil moisture # b) hydraulic gradient between a cell and its downstream neighbour # c) cell K

K=Ksat*((soilmoist/maxmoist)**(2*b+3)); # Percolation rate, Campbell (1985)

out = min((K*Hgrad),soilmoist); # water output from a cell cannot exceed the water available to transport (m/timestep)

Page 255: Ground Penetrating Radar for the parameterisation of

Andrew Howe

255

in = upstream(WTldd,out); # water input from upstream cells

balance = soilmoist + in - out;

soilmoist = if(balance gt maxmoist,maxmoist,balance); ROF = max(balance-maxmoist,0); # if the new balance exceeds cell capacity the excess becomes return overland flow soilmoist=if(soilmoist gt E,soilmoist-E,0); # final soil moisture includes evaporation losses

sat_pc=(soilmoist/maxmoist*100); # percentage saturation of soil in a cell

soilm = soilmoist; # renamed to allow storage of results

#----------------------------------------------------------------------------------------------------------# Runoff module. Uses kinematic flow routing. From De Roo et al.,. (1998)

localXS=localXS+ROF; # New localXS includes any return flow from saturated cells

alpha=((MN*(cellwidth+2*localXS)**(2/3))/(sqrt(grad)))**beta;# alpha is a roughness parameter Q1=if(alpha gt 0.0, ((cellwidth*localXS)/alpha)**(1/beta), 0.0); Q2=kinematic(Ldd,Q1,q,alpha,beta,timestep,cellwidth);# overland flow discharge (m

3/s)

localXS=alpha*(Q2**beta)/cellwidth; # first approximation using old alpha alpha=((MN*(cellwidth+2*localXS)**(2/3))/(sqrt(grad)))**beta; # new alpha localXS=alpha*(Q2**beta)/cellwidth; # second approximation for localXS using new alpha Runoff=localXS gt 0; # Boolean map, true value shows runoff is occurring in a cell

#--------------------------------------------------------------------------------------------------------------- # Subsurface discharge calculation # Based on Xiao 1996 for saturated flow between cells

x = ((dem-WTdepth)+downstream(WTldd,(dem-WTdepth))/2)-((dem-soildepth)+(downstream(WTldd,dem)-downstream(WTldd,soildepth))/2); # mean depth of saturated soil contributing to subsurface flow between cells (m)

A = x*cellwidth; # Cross sectional area contributing to subsurface flow (m2)

ssQ = A*out; # subsurface discharge (cell/timestep)

#-------------------------------------------------------------- --------------------------------------------------# Storage of output as maps

report stack1_ = soilm; # soil moisture report stack2_ = WTldd; # subsurface flow LDD report stack3_ = sat_pc; # percentage cell saturation report stack4_ = ROF; # return overland flow

Page 256: Ground Penetrating Radar for the parameterisation of

Andrew Howe

256

report stack5_ = Hgrad; # hydraulic gradient report stack6_ = Q2; # overland flow report stack7_ = ssQ; # subsurface throughflow report stack8_ = WTdepth; # water table depth report stack9_ = WFdepth; # wetting front depth # TimeSeries output

report Q2ts=timeoutput(Outflow,Q2); # Q from kinematic routing report ssQts=timeoutput(Outflow,ssQ); # subsurface Q report XSts=timeoutput(Outflow,localXS); # water height of outflow cell report plotrun=timeoutput(plot,Q2); # plot runoff report plotsub=timeoutput(plot,ssQ); # plot subsurface flow report Evap=timeoutput(plot,E); # Evaporation report Kval=timeoutput(plot,K); # Hydraulic conductivity report smoist=timeoutput(plot,soilm); # Plot soil moisture status report infil=timeoutput(plot,infilwater); # Infiltration value report sat=timeoutput(plot,sat_pc); # Percentage cell soil saturation

Page 257: Ground Penetrating Radar for the parameterisation of

Andrew Howe

257

Model inputs and parameters

Table I.1 Model parameters and initial data requirements.

Name Description Value Required for Spatialdistribution

Temporalstatus

Unit

RainTimeSeries

Hourlyrainfall data

Variable Model input Lumped Dynamic mm

InitMoist Initial soilmoisturestatus

0.9 ×maxmoist

Initialisation Distributed N/A m

PotEvapTime

Hourlypotentialevaporation

Variable Model input Lumped Dynamic m

NetRadiation

Hourly netradiation

Variable Calculation ofpotentialevaporation

Lumped Dynamic W/m2

Lamda Latent heatofvaporisation

2.5 Calculation ofpotentialevaporation

Lumped Constant J/kg

soildepth Map of soildepth (fromGPR depthandtopographicrelationship)

Variable Maximum soilmoisturestoragecapacity

Distributed Constant m

DEM Catchmentelevation

Variable Ldd networkand slope

Distributed Constant m

Ldd Localdrainagedirectionnetworkfrom DEM

Variable Routing ofsurface runoff

Distributed Constant N/A

Slope Local slopeangle fromDEM

Variable Input forsurface runoff

Distributed Constant degs

Timestep Modeltimestep

3600 Kinematicequation

N/A N/A s

Cellwidth

Cell width 25 Kinematicequation

N/A N/A m

Outflow Catchmentmap withoutflow cellselected

TRUE Catchmenthydrograph

1 cell Constant N/A

Plot Catchmentmap withfield plotcell flagged

TRUE Recording thetemporal statusof plotoverland flow,throughflowand soilmoisture

1 cell Constant N/A

Bool25 Booleanmap of theentirecatchment

TRUE Definecatchmentextent duringmodel

N/A Constant N/A

Page 258: Ground Penetrating Radar for the parameterisation of

Andrew Howe

258

area simulation

Name Description Value Required for Spatialdistribution

Temporalstatus

Units

sites Map of allsitessampledusing GPR

TRUE Analysis ofcell soilmoisturedynamics

Distributed Constant N/A

beta Kinematicwaveparameter,sheet flow(De Roo etal, 1998b)

0.6 Kinematicequation

Lumped Constant N/A

q Kinematicwaveparameter,side flow(De Roo etal, 1998b)

0.0 Kinematicequation

Lumped Constant N/A

b Poreinteractionterm fromCampbell(1985)

2.4 Calculation ofKsat

Lumped Constant N/A

MN Manning’s nvalue forcroppedgrassland(Meadows&Stephenson,1985)

0.1 Kinematicequation

Lumped Constant N/A

Page 259: Ground Penetrating Radar for the parameterisation of

Andrew Howe

259

Table I.2 Model internal variables.

Name Description Value Required for Spatialdistribution

Temporalstatus

Unit

soilmoist Cell soilmoisture

Variable Infiltration,evaporation,runoff andsubsurfaceflow

Distributed Dynamic m

localXS Ponded surfacewater in a cell

Variable Surface runoff Distributed Dynamic m

infilwater Amount of waterinfiltrating eachtimestep

Variable Infiltration Distributed Dynamic m

WFdepth Wetting frontdepth

Variable Infiltration Distributed Dynamic m

WTdepth Water tabledepth

Variable Subsurfaceflow andWTlddcalculation

Distributed Dynamic m

WTldd Ldd calculatedfrom water tableheight and cellelevation

Variable Routing ofsubsurfaceflow

Distributed Dynamic

Hgrad Hydraulicgradient betweena cell and thedownstream cell

Variable Subsurfaceflow

Distributed Dynamic m/m

ROF Return overlandflow. From cellswhen subsurfaceflow exceedsmaxmoist

Variable Surface runoff Distributed Dynamic m

Sat_pc Cell saturationpercentage

Variable Cell soilmoistureanalysis

Distributed Dynamic %

A Average depthcontributing tosubsurface flowbetween cells

Variable Subsurfaceflow

Distributed Dynamic m

Runoff Cell surfacewater discharge

Variable Catchment andplothydrograph

Distributed Dynamic m3/s

Page 260: Ground Penetrating Radar for the parameterisation of

Andrew Howe

260

Appendix II. GPR sample site information.

Figure II.1. Location of GPR sample sites within the Cyff. See table II.1 for site co-ordinates, terrain attributes and average GPR derived soil thickness.

Page 261: Ground Penetrating Radar for the parameterisation of

Andrew Howe

261

Table II.1 Site attributes and GPR measured soil thickness to horizons.

Page 262: Ground Penetrating Radar for the parameterisation of

Andrew Howe

262

Table II.2. Summary data of trend line r2 value, the number of samples used to fitthe trend line, total soil thickness to the CMP reflector, and the depth of watercalculated within the soil column using GPR, theta probe and gravimetricanalysis methods.

Theta probe

measurements

Gravimetric

samples

Site r2 No. of

readingsr2 No. of

samples

Soil thickness

to reflector (m)

GPR water

depth (m)

Theta probe

water depth

(m)

Gravimetric

water depth

(m)

13 0.774 8 0.481 3 0.349 0.152 0.166 0.179

14 0.863 8 0.971 3 0.330 0.175 0.166 0.141

16 0.622 11 0.879 5 0.525 0.251 0.298 0.290

17 0.904 4 n/a 0 0.410 0.207 0.167 -

18 0.969 4 n/a 0 0.423 0.189 0.147 -

19 0.989 4 0.988 4 0.461 0.181 0.197 0.172

20 1.000 2 n/a 0 0.314 0.146 0.162 -

22 0.548 6 0.459 5 0.511 0.244 0.224 0.308

23 0.576 5 n/a 0 0.336 0.158 0.143 -

24 0.786 5 n/a 0 0.330 0.149 0.122 -

27 0.095 3 0.081 3 0.565 0.238 0.249 0.201

32 0.961 4 1.000 2 0.322 0.123 0.151 0.164

33 0.156 12 0.942 6 0.591 0.347 0.375 0.416

48 Manual 14 Manual 5 0.297 0.138 0.130 0.111

49 0.744 3 n/a 1 0.372 0.145 0.085 -

53 0.879 10 0.972 3 0.425 0.164 0.226 0.210

54 0.992 5 0.824 4 0.260 0.133 0.124 0.107

56 0.486 14 4E-06 3 0.253 0.142 0.164 0.212

RMS error 0.031 0.045

Page 263: Ground Penetrating Radar for the parameterisation of

Andrew Howe

263

Table II.3. Variable soil thickness model scenario and GPR VMC for 29 sites inthe Cyff catchment.

Site Date Time Cell soilmoisture (m)

Cell soilthickness (m)

VMC cell GPR VMC

13 29/06/99 13:00 0.25 0.53 0.46 0.4414 29/06/99 11:00 0.30 0.58 0.51 0.5316 29/06/99 16:00 0.26 0.53 0.49 0.4823 01/07/99 12:00 0.15 0.33 0.45 0.4724 01/07/99 14:00 0.11 0.25 0.42 0.4533 08/07/99 12:00 0.35 0.66 0.53 0.5940 08/07/99 13:00 0.62 1.09 0.57 0.6048 30/06/99 13:00 0.34 0.70 0.49 0.4653 30/06/99 17:00 0.23 0.52 0.45 0.3956 30/06/99 11:00 0.38 0.75 0.50 0.565 16/10/98 17:00 0.26 0.41 0.64 0.486 07/10/98 10:00 0.19 0.40 0.49 0.518 06/10/98 11:00 0.19 0.38 0.51 0.459 14/10/98 15:00 0.58 0.97 0.60 0.4117 15/10/98 13:00 0.33 0.54 0.63 0.5018 09/10/98 15:00 0.31 0.57 0.54 0.4519 09/10/98 11:00 0.30 0.55 0.55 0.3920 14/10/98 16:00 0.31 0.62 0.50 0.4721 07/10/98 15:00 0.49 0.77 0.63 0.5522 15/10/98 11:00 0.46 0.72 0.64 0.4827 08/10/98 13:00 0.08 0.19 0.40 0.4232 08/10/98 12:00 0.14 0.32 0.44 0.3837 15/10/98 15:00 0.59 0.99 0.60 0.6339 15/10/00 16:00 0.60 1.00 0.59 0.5941 08/10/98 16:00 0.61 1.041 0.59 0.6149 06/10/98 14:00 0.23 0.45 0.52 0.3950 06/10/98 16:00 0.15 0.31 0.47 0.4954 07/10/98 14:00 0.29 0.52 0.56 0.5160 16/10/98 11:00 0.21 0.32 0.64 0.26

Page 264: Ground Penetrating Radar for the parameterisation of

Andrew Howe

264

Table II.4. Variable soil thickness model scenario and GPR VMC for 29 sites inthe Cyff catchment.

Site GPR survey Layer 2-way Interval Calculated Layer Total K VMCfrequency ID travel

time(time) Velocity depth depth rms

MHz ns ns m/ns m m m3/m3

5 900 1 0.0 0.281 0.00 0.00 1.12 22.0 22.0 0.051 0.56 0.56 34.4 0.483 26.0 4.0 0.055 0.11 0.67 29.6 0.444 31.0 5.0 0.068 0.17 0.84 19.6 0.34

6 900 1 0.0 0.291 0.00 0.00 1.12 16.0 16.0 0.047 0.38 0.38 40.5 0.514 22.0 6.0 0.068 0.20 0.58 19.6 0.34

20 900 1 0.0 0.277 0.00 0.00 1.22 5.5 5.5 0.050 0.14 0.14 35.8 0.483 12.5 7.0 0.052 0.18 0.32 33.1 0.474 20.5 8.0 0.065 0.26 0.58 21.2 0.36

21 450 1 0.0 0.325 0.00 0.00 0.82 12.5 12.5 0.048 0.30 0.30 39.6 0.513 22.5 22.5 0.044 0.50 0.50 46.2 0.554 26.5 14.0 0.055 0.39 0.68 29.6 0.445 35.0 8.5 0.068 0.29 0.97 19.6 0.34

22 900 1 0.0 0.319 0.00 0.00 0.92 9.0 9.0 0.050 0.23 0.23 35.8 0.483 15.0 6.0 0.055 0.17 0.39 29.6 0.444 22.5 7.5 0.063 0.23 0.62 22.9 0.38

27 900 1 0.0 0.313 0.00 0.00 0.92 10.0 10.0 0.050 0.25 0.25 35.8 0.483 21.2 11.2 0.058 0.32 0.57 27.0 0.42

9 450 1 0.0 0.294 0.00 0.00 1.02 4.5 4.5 0.055 0.12 0.12 29.6 0.443 19.5 15.0 0.070 0.53 0.65 18.2 0.324 44.0 24.5 0.055 0.67 1.32 29.6 0.44

9 900 1 0.0 0.313 0.00 0.00 0.92 4.0 4.0 0.045 0.09 0.09 44.1 0.533 16.6 12.6 0.059 0.37 0.46 25.7 0.414 45.0 28.4 0.055 0.78 1.24 29.6 0.44

49 450 1 0.0 0.284 0.00 0.00 1.12 8.0 8.0 0.065 0.26 0.26 21.2 0.363 16.5 8.5 0.085 0.36 0.62 12.4 0.234 45.0 28.5 0.100 1.43 2.05 8.9 0.17

Page 265: Ground Penetrating Radar for the parameterisation of

Andrew Howe

265

Site GPR survey Layer 2-way Interval Calculated Layer Total K VMCfrequency ID travel

time(time) Velocity depth depth rms

MHz ns ns m/ns m m m3/m3

49 900 1 0.0 0.279 0.00 0.00 1.12 8.0 8.0 0.063 0.25 0.25 22.9 0.383 18.0 10.0 0.080 0.40 0.65 14.0 0.26

8 450 1 0.0 0.287 0.00 0.00 1.12 15.0 15.0 0.050 0.38 0.38 35.8 0.483 23.0 8.0 0.070 0.28 0.66 18.2 0.324 40.0 17.0 0.068 0.57 1.23 19.6 0.34

8 900 1 0.0 0.310 0.00 0.00 0.92 13.0 13.0 0.054 0.35 0.35 30.7 0.453 25.0 12.0 0.065 0.39 0.74 21.2 0.36

17 450 1 0.0 0.297 0.00 0.00 1.02 7.5 7.5 0.058 0.22 0.22 27.0 0.423 15.5 8.0 0.048 0.19 0.41 38.8 0.504 29.5 14.0 0.068 0.47 0.88 19.6 0.34

17 900 1 0.0 0.293 0.00 0.00 1.02 10.0 10.0 0.060 0.30 0.30 24.8 0.403 17.5 7.5 0.058 0.22 0.52 27.0 0.424 32.5 15.0 0.068 0.51 1.02 19.6 0.34

18 450 1 0.0 0.300 0.00 0.00 1.02 10.0 10.0 0.060 0.30 0.30 24.8 0.403 14.5 4.5 0.055 0.12 0.42 29.6 0.444 40.0 25.5 0.075 0.96 1.38 15.9 0.29

18 900 1 0.0 0.287 0.00 0.00 1.12 7.5 7.5 0.055 0.21 0.21 29.6 0.443 20.0 12.5 0.063 0.39 0.60 22.9 0.384 35.0 15.0 0.075 0.56 1.16 15.9 0.29

19 900 1 0.0 0.285 0.00 0.00 1.12 5.5 5.5 0.040 0.11 0.11 55.9 0.613 17.0 11.5 0.061 0.35 0.46 24.0 0.394 22.5 5.5 0.068 0.19 0.65 19.6 0.345 30.0 7.5 0.075 0.28 0.93 15.9 0.29

41 450 1 0.0 0.289 0.00 0.00 1.12 5.5 5.5 0.040 0.11 0.11 55.9 0.613 37.5 32.0 0.040 0.64 0.75 55.9 0.614 50.0 12.5 0.053 0.33 1.08 32.4 0.46

41 900 1 0.0 0.333 0.00 0.00 0.82 5.5 5.5 0.058 0.16 0.16 27.0 0.423 42.0 36.5 0.045 0.82 0.98 44.1 0.534 47.5 5.5 0.050 0.14 1.12 35.8 0.48

Page 266: Ground Penetrating Radar for the parameterisation of

Andrew Howe

266

Site GPR survey Layer 2-way Interval Calculated Layer Total K VMCfrequency ID travel

time(time) Velocity depth depth rms

MHz ns ns m/ns m m m3/m3

37 450 1 0.0 0.268 0.00 0.00 1.22 5.5 5.5 0.045 0.12 0.12 44.1 0.533 21.5 21.5 0.040 0.43 0.55 55.9 0.614 40.0 34.5 0.055 0.95 1.07 29.6 0.445 60.0 20.0 0.060 0.60 1.67 24.8 0.40

39 450 1 0.0 0.278 0.00 0.00 1.22 4.5 4.5 0.043 0.10 0.10 49.5 0.573 45.0 40.5 0.041 0.83 0.93 53.2 0.594 55.0 10.0 0.060 0.30 1.23 24.8 0.40

50 900 1 0.0 0.328 0.00 0.00 0.82 10.0 10.0 0.055 0.28 0.28 29.6 0.443 14.0 4.0 0.050 0.10 0.38 35.8 0.484 22.0 8.0 0.075 0.30 0.68 15.9 0.29

54 900 1 0.0 0.258 0.00 0.00 1.32 5.0 5.0 0.050 0.13 0.13 35.8 0.483 10.8 5.8 0.047 0.14 0.26 40.5 0.51

60 900 1 0.0 0.294 0.00 0.00 1.02 7.5 7.5 0.043 0.16 0.16 49.5 0.573 18.2 10.7 0.080 0.43 0.59 14.0 0.26

13 900 1 0.0 0.325 0.00 0.00 0.82 3.4 3.4 0.050 0.09 0.09 35.8 0.483 12.5 12.5 0.056 0.35 0.44 28.5 0.43

15 900 1 0.0 0.306 0.00 0.00 1.02 3.2 3.2 0.053 0.09 0.09 31.8 0.463 17.6 17.6 0.063 0.56 0.64 22.5 0.37

16 900 1 0.0 0.308 0 0.00 0.92 4.0 4.0 0.042 0.09 0.09 50.7 0.573 20.6 20.6 0.051 0.53 0.61 34.4 0.48

23 900 1 0.0 0.274 0 0.00 1.22 3.4 3.4 0.050 0.09 0.09 35.8 0.483 12.9 12.9 0.052 0.34 0.42 33.1 0.474 20.2 20.2 0.065 0.65 1.08 21.2 0.36

24 900 1 0.0 0.292 0 0.00 1.02 3.3 3.3 0.052 0.08 0.08 33.1 0.473 12.2 12.2 0.054 0.33 0.41 30.7 0.45

33 900 1 0.0 0.260 0 0.00 1.32 3.1 3.1 0.055 0.09 0.09 29.6 0.443 6.0 6.0 0.050 0.15 0.24 35.8 0.484 21.5 21.5 0.041 0.44 0.68 53.2 0.59

Page 267: Ground Penetrating Radar for the parameterisation of

Andrew Howe

267

Site GPR survey Layer 2-way Interval Calculated Layer Total K VMCfrequency ID travel

time(time) Velocity depth depth rms

MHz ns ns m/ns m m m3/m3

40 450 1 0.0 0.272 0 0.00 1.22 1.4 1.4 0.118 0.09 0.09 6.4 0.113 45 43.6 0.041 0.90 0.99 52.2 0.585 60.0 15.0 0.111 0.83 1.82 7.3 0.13

48 900 1 0.0 0.298 0 0.00 1.02 2.4 2.4 0.071 0.08 0.08 17.7 0.323 11.3 11.3 0.052 0.29 0.38 33.1 0.474 20.7 9.3 0.088 0.41 0.79 11.5 0.22

53 900 1 0.0 0.270 0 0.00 1.22 3.0 3.0 0.057 0.08 0.08 27.5 0.423 13.7 13.7 0.062 0.43 0.51 23.3 0.38

56 900 1 0.0 0.313 0 0.00 0.92 4.1 4.1 0.041 0.08 0.08 53.2 0.593 11.8 11.8 0.043 0.25 0.34 48.4 0.564 37.1 25.3 0.069 0.87 1.21 18.8 0.33

Page 268: Ground Penetrating Radar for the parameterisation of

Andrew Howe

268

Table II.5. Soil thickness measurements (m) to B and C horizons using auger andGPR methods.

site ID f (MHz) A-B auger A-B gpr Difference B-C auger B-C gpr Difference5 900 0.25 0.26 0.01 0.5 0.43 -0.076 900 0.25 0.21 -0.04

0.3 0.3 0.29 -0.010.17 0.18 0.01 0.25 0.26 0.01

no data 0.22 0.21 -0.01no data 0.29 0.33 0.04no data 0.25 0.29 0.04

8 900 0.3 0.3 0.00 0.45 0.39 -0.060.3 0.28 -0.02 0.5 0.43 -0.07

0.26 0.33 0.07 0.46 0.47 0.019 450 no data 0.94 0.93 -0.01

no data 1.0 1.01 0.01no data 0.87 no pick

9 900 no data 0.94 no pickno data 1.0 no pickno data 0.87 no pick

17 450 0.16 no pick 0.84 0.78 -0.060.175 no pick 0.95 0.84 -0.11

0.15 no pick 1.0 0.92 -0.0817 900 0.16 0.16 0.00 0.84 no pick

0.175 0.18 0.005 0.95 no pick0.15 0.18 0.03 1.0 no pick

18 900 0.42 0.45 0.03 0.77 0.61 -0.160.485 0.45 -0.035 0.66 0.63 -0.03

0.49 0.46 -0.03 0.70 0.77 0.0719 900 0.21 0.2 -0.01 0.40 0.34 -0.06

0.265 0.25 -0.015 0.35 0.55 0.200.35 0.24 -0.11 0.63 0.46 -0.17

Pit 0.21 0.22 0.01 0.30 0.33 0.0320 900 0.18 0.23 0.05 0.35 0.40 0.05

0.22 0.21 -0.01 0.34 0.46 0.120.23 0.18 -0.05 0.65 0.55 -0.10

0.155 0.18 0.025 0.56 0.33 -0.230.17 0.18 0.01 0.46 0.48 0.02

21 450 no data 0.47 0.73 0.67 -0.06no data 0.36 0.72 0.71 -0.01

0.375 0.36 -0.015 0.89 0.79 -0.100.54 0.47 -0.07 0.77 0.97 0.20

900 no data 0.46 0.73 0.7 -0.03no data 0.50 0.72 0.76 0.04

0.375 0.36 -0.015 0.89 0.68 -0.210.54 0.52 -0.02 0.77 0.68 -0.09

22 450 0.16 0.66 0.65 -0.010.25 0.78 0.85 0.070.31 0.89 0.88 -0.01

900 0.16 0.21 0.05 0.66 0.67 0.010.25 0.23 -0.02 0.78 0.65 -0.13

Page 269: Ground Penetrating Radar for the parameterisation of

Andrew Howe

269

0.31 0.27 -0.04 0.89 0.63 -0.26Pit 0.2 0.21 0.01 0.31 0.53 0.22

27 900 No B horizon present 0.15 0.39 0.24No B horizon present 0.48 0.55 0.07

Pit No B horizon present 0.3 0.29 -0.01No B horizon present 0.39 0.36 -0.03No B horizon present 0.5 0.54 0.04

32 900 0.21 0.24 0.03 0.29 0.35 0.060.15 0.18 0.03 0.29 0.35 0.060.18 0.22 0.04 0.31 0.33 0.02

37 450 0.90 0.88 -0.020.90 0.92 0.020.73 0.70 -0.030.60 0.95 0.351.00 0.94 -0.061.00 0.97 -0.030.55 0.5 -0.050.85 0.85 0.000.92 0.85 -0.071.00 0.91 -0.09

39 450 0.95 1.12 0.170.78 0.85 0.070.84 0.88 0.041.00 1.07 0.070.85 0.81 -0.040.9 1.12 0.22

0.72 1.15 0.430.76 0.93 0.170.7 0.82 0.12

0.65 0.68 0.0341 450 No data 0.88 0.98 0.10

No data 0.87 0.84 -0.03No data 0.74 0.85 0.11

49 900 No A horizon present 0.26 0.23 -0.03No A horizon present 0.3 0.26 -0.04No A horizon present 0.31 0.26 -0.05

Pit No A horizon present 0.2 0.26 0.0650 900 0.29 0.29 0.34 0.0552 900 No data 0.2 0.3 0.1

0.19 0.245 0.25 0.0050.1 0.25 0.27 0.02

No data 0.21 0.21 0.00No data 0.26 0.26 0.00Organic to C horizon 0.17 0.24 0.07

54 900 No data 0.35 0.31 -0.040.3 0.25 0.05 0.36 0.35 -0.01

No data 0.67 0.47 -0.200.15 0.84 0.64 -0.20

Pit 0.26 0.25 0.01 0.3+ See 2nd position forthis site

13 900 0.25 No match 0.7 No matchPit 0.28 0.27 -0.01 0.34+ 0.65

0.23 0.23 0.00 0.72 0.59 -0.13

Page 270: Ground Penetrating Radar for the parameterisation of

Andrew Howe

270

14 900 0.26 0.26 0.00 0.86 0.61 -0.250.23 0.28 0.05 0.37 0.38 0.010.25 0.19 -0.06 0.82 0.34 -0.48

15 900 0.1 Too shallow 0.8 0.69 -0.11Pit 0.12 0.15 0.03 0.17+ 0.68

0.20 0.25 0.05 0.62 0.81 0.1916 900 0.41 0.39 -0.02 0.41 0.39 -0.02

Pit 0.45 0.48 0.03 0.55 0.48 -0.0748 900 0.52 0.4 -0.12 0.79 0.73 -0.06

Pit 0.66 0.64 -0.02 0.66+ 0.640.64+ No match 0.9 0.73 -0.17

53 Pit 0.38 0.42 0.04 0.78 No match900 0.3 0.3 0.00 0.5 0.5 0.00

0.24 No match 0.57 0.49 -0.0856 900 No A or B horizon present 1.00 1.06 0.06

Pit No A or B horizon present 0.7+ 1.02No A or B horizon present 1.00 1.03 0.03

23 900 No A horizon present 0.32 0.32 0.00Pit No A horizon present 0.3 0.31 0.01

No A horizon present 0.29 0.23 -0.0624 900 no data 0.49 0.32 -0.17

Pit No A horizon present 0.32 0.29 -0.03No A horizon present 0.3 0.27 -0.03

33 450 No A horizon present 0.69 0.64 -0.05Pit No A horizon present 0.56 0.55 -0.01

No A horizon present 0.66 0.65 -0.0140 No A horizon present 1.04+ 1.27

No A horizon present 1.05+ 1.51No A horizon present 0.97+ 1.55

B Horizon C HorizonRMS ofdifference (m)

0.04 RMS ofdifference (m)

0.12

Page 271: Ground Penetrating Radar for the parameterisation of

Andrew Howe

271

Appendix III. Soil sample data.

Table III.1 Soil sample wet and dry weights.

Site ID Depth (m)from

groundsurface

Wetweight

(g)

Dryweight

(g)

VMC %(gravimetric)

BulkDensity(g/cm

3)

%Porosity

6 0.1 43.29 16.90 60.57 0.39 0.856 0.2 56.72 39.77 38.91 0.91 0.6632 0.1 44.06 18.95 57.64 0.43 0.8432 0.2 51.86 31.45 46.85 0.72 0.7319 0.1 44.23 18.01 60.18 0.41 0.8419 0.2 59.67 40.09 44.94 0.92 0.6519 0.15 50.63 27.44 53.23 0.63 0.7619 0.29 39.32 28.49 24.86 0.65 0.7527 0.04 36.34 15.39 48.09 0.35 0.8727 0.1 38.84 24.58 32.73 0.56 0.7927 0.18 53.16 34.66 42.46 0.80 0.7049 0.1 49.24 31.06 41.73 0.71 0.738 0.1 47.27 20.98 60.35 0.48 0.828 0.2 49.77 30.83 43.47 0.71 0.738 0.3 73.48 54.57 43.41 1.25 0.5350 0.1 49.38 26.55 52.40 0.61 0.7750 0.2 53.06 34.07 43.59 0.78 0.7054 0.04 47.60 23.64 55.00 0.54 0.8054 0.1 53.56 35.59 41.25 0.82 0.6954 0.2 55.27 43.01 28.14 0.99 0.6354 0.3 51.43 39.06 28.39 0.90 0.6622 0.05 32.97 9.73 53.34 0.22 0.9222 0.1 41.95 15.45 60.83 0.35 0.8722 0.15 66.16 46.76 44.53 1.07 0.5922 0.2 59.30 36.63 52.04 0.84 0.6822 0.25 44.16 26.25 41.11 0.60 0.7733 0.1 71.36 7.75 73.00 0.09 0.9733 0.15 73.40 8.28 74.74 0.10 0.9633 0.25 75.93 9.44 76.31 0.11 0.9633 0.35 77.82 8.17 79.94 0.09 0.9633 0.45 95.84 37.32 67.16 0.43 0.8433 0.55 60.70 36.51 55.52 0.42 0.8448 0.05 82.36 38.17 50.72 0.44 0.8348 0.2 103.52 89.57 16.01 1.03 0.6148 0.3 45.99 22.67 53.53 0.52 0.8048 0.4 52.91 30.07 52.43 0.69 0.7448 0.5 66.68 47.61 43.77 1.09 0.5916 0.05 87.10 28.23 67.56 0.32 0.8816 0.2 60.96 37.75 53.28 0.87 0.6716 0.3 55.84 32.04 54.63 0.74 0.7216 0.4 63.66 41.32 51.28 0.95 0.6416 0.5 67.78 49.17 42.72 1.13 0.57

Page 272: Ground Penetrating Radar for the parameterisation of

Andrew Howe

272

15 0.05 77.20 22.86 62.37 0.26 0.9015 0.1 51.14 26.87 55.71 0.62 0.7715 0.15 53.37 43.80 21.97 1.01 0.6214 0.05 85.45 40.60 51.47 0.47 0.8214 0.15 60.62 40.73 45.65 0.93 0.6514 0.25 58.05 42.85 34.89 0.98 0.6313 0.05 84.25 37.11 54.10 0.43 0.8413 0.15 108.69 58.38 57.74 0.67 0.7513 0.25 60.81 41.21 44.99 0.95 0.6453 0.15 41.84 17.79 55.20 0.41 0.8553 0.25 54.58 32.74 50.13 0.75 0.7253 0.35 64.54 46.75 40.83 1.07 0.6056 0.1 88.52 13.08 86.58 0.15 0.9456 0.3 80.58 11.40 79.40 0.13 0.9556 0.7 86.87 12.67 85.16 0.15 0.95

Field plot 0.03 43.61 17.62 59.66 0.40 0.85Field plot 0.1 49.94 22.19 63.70 0.51 0.81Field plot 0.19 64.72 46.75 41.25 1.07 0.60Field plot 0.2 67.97 47.90 46.07 1.10 0.59Field plot 0.3 48.42 34.19 32.66 0.78 0.70Field plot 0.05 42.69 18.45 55.65 0.42 0.84Field plot 0.1 61.13 40.84 46.57 0.94 0.65Field plot 0.15 50.40 30.08 46.64 0.69 0.74Field plot 0.2 49.75 27.49 51.09 0.63 0.76Field plot 0.25 57.14 39.94 39.48 0.92 0.65Field plot 0.33 64.56 35.65 66.36 0.82 0.69

37 0.2 40.59 8.88 72.79 0.20 0.9237 0.3 45.10 8.72 83.51 0.20 0.9237 0.4 42.40 7.06 81.12 0.16 0.9437 0.1 37.02 4.49 74.67 0.10 0.9637 0.5 41.85 5.79 82.77 0.13 0.9537 0.7 65.28 34.95 69.62 0.80 0.7037 0.7 53.73 20.85 75.47 0.48 0.8237 0.2 40.61 6.96 77.24 0.16 0.9437 0.3 38.70 6.76 73.31 0.16 0.9437 0.6 44.78 6.88 86.99 0.16 0.9437 0.4 41.45 5.54 82.43 0.13 0.9537 0.5 43.26 5.49 86.70 0.13 0.95

Mean 57.83 28.33 67.71