SYNTHETIC APERTURE RADAR DATA EMPLOYED FOR SOIL MOISTURE ESTIMATION IN THE PIKETBERG REGION, SOUTH AFRICA

Jeanine Engelbrecht

Council for Geoscience, South Africa, jengelbrecht@geoscience.org.za

ABSTRACT

Information on the distribution of surface soil moisture is important for a number of applications. Due to the high temporal and spatial variability, and consequently the cost of monitoring by field observations, a means of remote monitoring of soil moisture content using remote sensing data is needed. The aim of this study was to test soil moisture retrieval algorithms based on synthetic aperture radar data (SAR). This includes the use of Envisat ASAR and ALOS PALSAR data, which was provided by the European Space Agency. Both linear regression and multiple-polarization models were applied for soil moisture quantification. The results could not be validated due to a lack of distributed field-based measurements but were compared to rainfall figures over the same period. Though inconclusive, the results suggest that the techniques show promise in their ability to quantify surface soil moisture.

Index Terms— Soil moisture, Envisat ASAR, ALOS PALSAR

1. INTRODUCTION TO THE STUDY

Information about the distribution of surface soil moisture is important for a number of applications including the precision management of agriculture [9]. However, it is difficult and impractical to map soil moisture in the field over large regions [10][13]. Therefore, a means of remote monitoring of soil moisture content is needed for long-term monitoring of surface soil moisture. Radar remote sensing data from orbiting satellites have been used extensively in an aim to quantify surface soil moisture [3][14][15][5][13].

The overall objective of this study was to test a variety of soil moisture quantification algorithms and to determine the usefulness and applicability of using radar remote sensing data as a tool for extracting soil moisture measurements over time. This includes linear regression models and multiple-polarization models. Both Envisat ASAR, captured in dual polarization mode, and ALOS PALSAR, captured in polarimetric mode, data were used. The data was provided by the European Space Agency. Envisat ASAR data were captured on 2006-06-25 (wet period), 2006-10-15 (beginning of dry period) and 2006-10-

29 (dry period). The ALOS data was captured on 2007-04-01 (end of dry period).

2. THE STUDY AREA AND FIELD MEASUREMENTS

The study area, situated in the Piketberg region in the Western Cape Province of South Africa (Figure 1), encompasses a quaternary catchment in which commercial agriculture plays an important role. The area experiences winter rainfall. Since South Africa is a water scarce country, it is important for water resource managers to have accurate information on all aspects of water resource occurrence and use. This includes the measuring and monitoring of surface soil moisture over time. Remote sensing based on aerial photography and satellite imagery can be used to quantify parameters related to surface and ground water storage and use.

The agricultural fields on one of the farms in the study area are the subject of precision agricultural practices (Figure 1). These practices include the measurement of soil moisture content at different depths of the soil profile using neutron probes. These measurements were used as input into linear regression models with the aim of quantifying soil moisture.

Figure 1: Boundary of the study area with location of field measurements indicated

II - 214978-1-4244-3395-7/09/$25.00 ©2009 IEEE IGARSS 2009

3. SOIL MOISTURE QUANTIFICATION WITH SAR DATA

The principle of radar-based soil moisture estimation relies on the existing relationship between the backscattering coefficient 0σ (in dB) and the dielectric properties of the

observed soils [1]. The dielectric properties of the observed soils are directly related to the moisture content of soils. Various authors ascribe the relationship between the radar backscattering coefficient ( 0σ ) and dielectric constant ( ) to

parameters including soil moisture content [2][11][1][7]. The measurement of soil moisture using radar remote sensing is based on the large contrast between the dielectric properties of dry soil ( 2.5 to 3.5) and water ( 80) [8][6][4][7]. As the soil water content increases, the dielectric constant increases from approximately 2.5 to 3.5 when dry to about 25-30 under saturated conditions. This translates to an increase in reflected power by almost 8 dB for wet soil compared to dry soil.

Various models have been proposed for the quantification of soil moisture content including the use of linear regression models [10][1][6] and multiple polarization models [4][13]. The following sections describe two models and present the result of the application of these models for quantification of soil moisture content.

3.1. Linear regression models

When using linear regression models for soil moisture quantification, one assumes a linear relationship between radar backscattering coefficient ( 0σ ) and volumetric soil

moisture content. The linear relationship can be expressed as bVSMa +×=0σ

where VSM is the volumetric soil moisture content, the slope coefficient a is related to the signal’s sensitivity to change in soil moisture concentration and the intercept coefficient b represents the backscatter of dry soil.

The field data was used as input into the model with the aim of estimating the a and b coefficients. The values for a and b were retrieved by linear regression analysis on the fields with known moisture content values. The volumetric soil moisture content could be derived by solving for VSM using the radar backscatter, and a and b coefficients as input. If one assumes that the signal’s sensitivity to change in soil moisture concentration (coefficient a) and the backscatter of dry soil (coefficient b) does not change between datasets, the same coefficient values can be applied to all the data sets. However, such an assumption would only be valid if the sensor configuration does not change between image acquisitions. The results of the soil moisture quantification using linear regression models are presented and discussed in Section 4.

A well documented problem in the use of linear regression models for soil moisture quantification is that one assumes a linear relationship between soil moisture and radar backscatter while in reality, radar backscatter is also influenced by surface roughness as well as by vegetation cover in the area of interest, both of which remains unaccounted for in this technique [13][1][6][7][4][2]. The same researchers also state that, in cases, the sensitivity of radar backscatter may be greater than the sensitivity to soil moisture, and should therefore be accounted for. The attenuation by vegetation is also expected to be significant due to the agricultural nature of the area of interest.

It is suggested that the strong interaction of the backscattered signal with soil as well as vegetation cannot be expressed by a simple linear function. In this regard, various authors describe the combination of microwave signatures at different frequencies and/or polarizations, which may provide additional information on soil and vegetation conditions which may prove to be more useful. The following section addresses the limitations of linear regression models for soil moisture quantifications by describing multiple polarization models for surface soil moisture retrieval.

3.2. Multiple polarization models

Because SAR responds to several surface features which are usually unknown, it is necessary to use either several polarizations at one point in time, or alternatively, several images over time to develop reliable interpretation algorithms. By using models developed for multi-polarized radar data both soil moisture contents and surface roughness can be determined [4]. Models have been developed for multi-frequency, multi-polarization SAR data which remove the need for prior information on surface roughness. One of the multiple-polarization models, known as the Dubois model, relates two co-polarised backscatter responses

0vvσ and 0

hhσ to the RMS surface height, the dielectric constant, the SAR wavelength, and local incidence angle. These equations are expressed as follows:

7.04.1tan028.05

5.175.20 )sin(10

sincos10 λθ

θθσ θε khhh

−=

and

7.01.1tan046.03

335.20 )sin(10

sincos10 λθ

θθσ θε khvv

−=

Where λπ2=k , h is the RMS surface height, ε is the

dielectric constant, λ is the wavelength and θ is the local incidence angle. These equations were used to obtain estimates of dielectric constant and surface height by

II - 215

simultaneous solving of 0vvσ and 0

hhσ (also known as inverse modelling) [13].

Multiple-polarization models provide the user with the opportunity to quantify soil moisture content within the scene in the absence of ancillary data on surface conditions, including surface roughness. Furthermore, no initial field-based soil moisture measurements are needed for the estimation of soil moisture content in the area, as is the case with linear regression models.

4. RESULTS AND DISCUSSION

The results of the soil moisture quantification using linear regression analysis and the Dubois model are displayed in Figure 2. Soil moisture values for 2005-06-26, 2006-10-15 and 2006-10-29 were quantified using linear regression models while the values for 2007-04-01 were quantified using the Dubois model. The results depict estimated soil moisture values in volumetric percent. Visual inspection of the results indicates a flaw in radar remote sensing data for soil moisture quantification, caused by the radar image geometry. Due to the side-looking geometry of the radar sensors, mountainous regions show layover and foreshortening effects. Mountainous areas have a severe effect on radar image interpretability. In particular, anomalously high backscatter values are recorded on the ridges of mountains, which is not necessarily a true reflection of the backscatter values for a particular land surface.

It is difficult to assess the accuracy with which the soil moisture quantification algorithms were applied since distributed ground truth is unavailable. This causes large uncertainties with regards to the reliability of the results. To assess if soil moisture patterns compare with the rainfall patterns for the period during which the data was captured, 1km rainfall grids were obtained from the Agricultural Research Council’s Institute for Soil Climate and Water (ARC-ISCW), South Africa. The rainfall grids represent the cumulative rainfall over a ten day period, including the date of image capture. The soil moisture content and cumulative rainfall figures were analyzed as a function of landcover type using zonal statistics for different landcover types in the area. These results are displayed graphically in Figure 3 using average soil moisture content and average rainfall figures per landcover class. The results suggest that an increase in rainfall for a specific period is associated with an increase in soil moisture content for the same period. However, some variations in soil moisture values between landcover classes could not be explained through direct comparison with rainfall. This is especially true for the two images captured during the summer (2006-10-15 and 2006-10-29). Here derived soil moisture values increase with landcover class in the following order: 1) Commercial dryland (corresponding to wheat), 2) Wetlands, 3)

Figure 2: Quantified soil moisture in volumetric percent

Average soil moisture per landcover class

0.000

2.000

4.000

6.000

8.000

10.000

12.000

14.000

16.000

20050626 20061015 20061029 20070401

Image date

Soi

l moi

stur

e %

Commercial_dryland

Thicket, High fynbos

Shrubland, low fynbos

Forest, Plantations

cultivated, temp,Commercial, irrigatedCultivated, perm, comm,irrigatedWetlands

Average rainfall per landcover class

0.000

2.000

4.000

6.000

8.000

10.000

12.000

14.000

16.000

18.000

20050626 20061015 20061029 20070401

Image date

Rain

fall

(mm

)

Commercial_dryland

Thicket, High fynbos

Shrubland, low fynbos

Forest, Plantations

cultivated, temp,Commercial, irrigatedCultivated, perm, comm,irrigatedWetlands

Figure 3: Average soil moisture and average rainfall for each landcover class

II - 216

Cultivated temporary commercial irrigated (potatoes), 4) Shrubland and low fynbos, 5) Thicket and high fynbos, 6) Cultivated permanent commercial irrigated (corresponding to orchards), and 7) Forest and plantations. This suggests that the radar signal may be influenced by vegetation, with denser vegetation increasing the calculated soil moisture content which is not necessarily a true reflection of soil moisture content in the field.

For the soil moisture values derived using the Dubois model, it is important to note that the Dubois model was developed for data with wavelengths ranging between 20.5 cm and 2.8 cm and local incidence angle between 30 and 65o[13]. On the other hand, the ALOS PALSAR data used in this process operates at a wavelength of 23.6 cm and the local incidence angles were calculated to be less that the 30o

threshold for large portions of the scene. Detailed ground truth data will be needed to test the validity of the Dubois model under these conditions.

5. CONCLUDING REMARKS

Soil moisture variability is an important factor for many applications including precision agriculture and hydrological models. Radar imagery for the quantification of soil moisture content is the focus of many research programmes, and the results indicate that the data can be used with various levels of success. In general, SAR systems show a relatively high sensitivity to soil moisture due to the large contrast in the dielectric constant of dry and wet soils at microwave frequencies. Consequently, much hope has been put in the capability of SAR sensors to quantify soil moisture content. However, despite extensive efforts by the science communities, soil moisture retrieval from SAR imagery can still be considered to be in an experimental stage [12].

6. REFERENCES

[1] J. Álvarez-Mozos, J. Casali, M. González-Audícana, and N.E.C. Verhoest, “Correlation between ground measured soil moisture and RADARSAT-1 derived backscattering coefficient over an agricultural catchment of Navarre (North of Spain)”, Biosystems Engineering, 92, pp 119-133, 2005.

[2] K. Dabrowska-Zielinska, Y. Inoue, W. Kowalik, and M. Gruszczynska “Inferring the effect of plant and soil variables on C-band and L-band SAR backscatter over agricultural fields, based on model analysis” Advances in Space Research 39: pp 139 – 148, 2007.

[3] G. D’Urso, and M. Minacapilli, “A semi-empirical approach for surface soil water content estimation from radar data without a-priori information on surface roughness,” Journal of Hydrology321, pp. 297-310 2006.

[4] T.E. Engmen, and N. Chauhan, “Status of microwave soil moisture measurements with remote sensing,” Remote Sensing of Environment, 51, pp.189-198, 1995.

[5] N. Holah, N. Baghdadi, M. Zribi, A, Bruand, and C. King, “Potential of ASAR/ENVISAT for the characterization of soil surface parameters over bare agricultural fields,” Remote Sensing of Environment, 96, pp. 78-86, 2005.

[6] M.S. Moran, D.C. Hymer, J. Qi, and E.E. Sano, “Soil moisture evaluation using multi-temporal synthetic aperture radar (SAR) in semiarid rangeland,” Agricultural and Forest Meteorology, 105, pp. 69-80, 2000.

[7] M.S. Moran, S. McElroy, J.M. Watts, and C.D. Peters-Lidard, “Radar remote sensing for estimation of surface soil moisture at the watershed scale,” [online]. Available at http://www.tucson.ars.ag.gov/unit/Publications/PDFfiles/1566.pdf. [Accessed November 2006].

[8] R.M. Narayanan, and P.P. Hirsave, “Soil moisture estimation models using SIR-C SAR data: a case study in New Hampshire, USA,” Remote Sensing of Environment, 75, pp. 385 – 396, 2001.

[9] M.M. Rahman, M.S. Moran, D.P. Thoma, R. Bryant, C.D Holifield Collins, T. Jackson, B.J. Orr, and M. Tischler, “Mapping surface roughness and soil moisture using multi-angle radar imagery without ancillary data,” Remote Sensing of Environment,112, pp. 391-402, 2008.

[10] T. Svoray, and M. Shoshany, “Multi-scale analysis of intrinsic soil factors from SAR-based mapping of drying rates” Remote Sensing of Environment, 92, pp. 233-246, 2004.

[11] A. Turesson, “Water content and porosity estimated from ground-penetrating radar and resistivity,” Journal of Applied Geophysics, 58, pp. 99-111, 2006.

[12] W. Wagner, and C. Pathe, “Has SAR failed in soil moisture retrieval?,” Proceedings of the 2004 Envisat & ERS Symposium, Salzburg, Austria 6 – 10 September 2004.

[13] A.W. Western, R.B. Grayson, T. Sadek, and H. Turral, “On the ability of AirSAR to measure patterns of dielectric constant at hillslope scale,” Journal of Hydrology, 289, pp. 9-22, 2004.

[14] M. Zribi, N. Baghdadi, N. Holah, and O. Fafin, “New methodology for soil surface moisture estimation and its application to ENVISAT-ASAR multi-incidence data inversion” Remote Sensing of Environment, 96, pp. 485-496, 2005.

[15] M. Zribi, N. Baghdadi, N. Holah, O. Fafin, and C. Guérin, “Evaluation of a rough soil surface description with ASAR-ENVISAT radar data,” Remote Sensing of Environment, 95, pp. 67-76, 2005.

II - 217