Forest Ecology and Management 256 (2008) 1984–1994

Incorporating hydrologic dynamics into buffer strip design on the sub-humidBoreal Plain of Alberta

I.F. Creed a,b,*, G.Z. Sass a, M.B. Wolniewicz b, K.J. Devito c

a Department of Biology, The University of Western Ontario, London, Ontario N6A 5B7, Canadab Department of Geography, The University of Western Ontario, London, Ontario N6A 5C2, Canadac Department of Biological Sciences, University of Alberta, Edmonton, Alberta T6G 2G8, Canada

A R T I C L E I N F O

Article history:

Received 31 January 2008

Received in revised form 24 May 2008

Accepted 19 July 2008

Keywords:

Wet areas

Boreal forest

Hydrology

Buffer strip

Remote sensing

Probability

RADARSAT

Forest management

Boreal Plain

Wetland

A B S T R A C T

The status quo in forestry practice is to leave standard width buffers around water bodies in order to

prevent the excess transport of sediments and nutrients from terrestrial to aquatic systems. This practice

does not seem to be effective in the sub-humid boreal forest where climatic and physiographic variability

produces complex hydrologic pathways not well protected by standard width buffers. We developed a

remote sensing technique that forms a hydrologic basis for buffer strip design. Synthetic aperture radar

(SAR) imagery was used to map, both inundated and saturated areas (herein called wet areas) amenable

for surface transport of nutrients and sediments on a low relief landscape in northern Alberta, Canada.

Wet areas coverage of the Moose Lake drainage basin showed a semi-logarithmic relationship with daily

discharge (r2 = 0.72, p < 0.001, n = 18). This relationship was used to define a flow–duration curve for wet

areas that could be used as an aspatial assessment of what proportion of a drainage basin should be

protected. A probability map of wet areas formation was calculated from the database of 18 images. We

demonstrated how the probability map may be used to design adaptive buffer strips for the mitigation of

increased nutrient loading to boreal lakes following timber harvesting.

� 2008 Elsevier B.V. All rights reserved.

Contents lists available at ScienceDirect

Forest Ecology and Management

journal homepage: www.elsev ier .com/ locate / foreco

1. Introduction

The Canadian boreal forest is one of the world’s largest forestregions. Increasing demands for natural resources have resulted inabout 50% of the Canadian boreal forest being allocated for timberharvesting and for oil and gas exploration (Global Forest WatchCanada, 2002). Such activities may have a range of impacts onaquatic systems including alteration of water, sediment, andnutrient fluxes from disturbed sites to surface waters and blockageof amphibian and/or fish migration corridors. While considerableeffort has focused on assessing the effects of these disturbances onaquatic systems (Carignan et al., 2000; Prepas et al., 2001),coherent models are lacking that provide an understanding of thehydrological processes that control water, sediment, and/ornutrient transport from forested areas and the implications ofdisturbances on these processes (Buttle et al., 2000, 2005).

* Corresponding author at: Department of Biology, The University of Western

Ontario, London, Ontario N6A 5B7, Canada. Tel.: +1 519 661 4265;

fax: +1 519 661 3935.

E-mail address: icreed@uwo.ca (I.F. Creed).

0378-1127/$ – see front matter � 2008 Elsevier B.V. All rights reserved.

doi:10.1016/j.foreco.2008.07.021

Devito et al. (2000) presented a hierarchy of landscape featuresthat control the range in natural variation in the nutrient statusamong boreal lakes and the potential susceptibility of the nutrientstatus of these lakes to disturbance. Sass et al. (2008) tested thishierarchy in a relatively undisturbed region of the Boreal Plainecozone and found that indicators of surface and subsurfacehydrologic connections between terrestrial and aquatic systemsexplained 72% of the natural variability in trophic status of 40shallow lakes. One of the key surface hydrological indicators of laketrophic status was the percent of the drainage basin occupied bywet areas. Sass et al. (2008) defined wet areas as parts of thelandscape that experience either saturation or inundation. Heargued that wet areas provide effective surface pathways of waterdirectly to lakes and/or streams, and thus influence the export ofnutrients and ultimately the trophic status of lakes.

The effectiveness of wet areas in transporting matter differsbased on their position within the landscape. Wet areas will act assources of water and associated dissolved or particulate materialsonly if they are connected to receiving water bodies. On the BorealShield, which is characterized by a humid climate and/or ruggedterrain, previous studies have observed strong positive relation-ships between the proportion of wet areas covering a drainagebasin and the amount of surface and/or near-surface transport of

I.F. Creed et al. / Forest Ecology and Management 256 (2008) 1984–1994 1985

water and water-soluble nutrients to lakes (e.g., Dillon and Molot,1997, 2005). On the Boreal Plain, which is characterized by a sub-humid climate and flat terrain, a significant relationship wasobserved not between the proportion of wet areas covering adrainage basin and the nutrient status of lakes, but rather betweenthe proportion of wet areas connected to the lake and nutrientstatus (Devito et al., 2000). On the Boreal Plain, wet areas locatedwithin topographic depressions and flats may not regularlyconnect through surface or near-surface hydrologic pathways tothe lakes. The relatively well-organized hydrologic pathways thatoccur on the Boreal Shield, where wet areas connect directly to thelake or indirectly through streams to the lake (Creed et al., 2003),contrast with the relatively disorganized hydrologic pathways thatoccur on the Boreal Plain, where wet areas shift from a state ofbeing ‘disconnected’ during dry periods to ‘connected’ during wetperiods (Devito et al., 2005a; Sass and Creed, 2008). A large portionof Canada’s boreal forest is characterized by flat landscapes withpoorly defined networks of wet areas, and yet, we know little of theimportance and the implications of disruptions and/or enlarge-ment in these hydrologically active areas on the structure andfunction of boreal ecosystems.

As a consequence of inadequate knowledge on how environ-mental changes in terrestrial and aquatic ecosystems effect wetareas and the surface hydrologic pathways that connect thesehydrologic features to receiving water bodies, the governmentalpolicies that try to address these potential impacts on the BorealPlain are general and largely imported from other jurisdictions.For example, a common practice in forest management is to leavebuffer strips of riparian forest between harvested areas andadjacent surface waters to trap dissolved and particulatenutrients in hydrologic flows, and thereby, reduce nutrientloading from harvested areas to surface waters (Castelle et al.,1994; Lee et al., 2004). Most jurisdictions apply a rule-basedapproach when it comes to assigning minimum buffer widths.Current guidelines on the Boreal Plain (Alberta) require that astandardized width of buffer strip be left around wet areas (20 m),streams (30 m), rivers (60 m) and lakes (100 m) (AlbertaEnvironmental Protection, 1994). The guidelines exclude streamsthat are less than 0.5 m in width or lakes less than 4 ha in area andthese guidelines assume that a uniform and diffuse hydrologicflow from contributing hillslopes dominates runoff. These guide-lines fail to incorporate the spatial and temporal dynamics of thehydrologic system.

The failure to incorporate hydrologic dynamics in the design ofbuffer strips may explain conflicting results among studies thatevaluated the effectiveness of standard widths of buffer strips onsurface water quality in the boreal forest. The effectiveness (e.g.,Carignan et al., 2000) versus ineffectiveness (e.g., Steedman, 2000;Prepas et al., 2001; Pinel-Alloul et al., 2002) of buffer strips inminimizing nutrient loading to surface waters may in part beattributed to the dynamics of hydrologic pathways within theseboreal watersheds. For example, in the TROLS experiment on theBoreal Plain, an increase in total phosphorous (TP) observed in theexperimental lakes following timber harvesting was not related tothe width of riparian buffer strips (Prepas et al., 2001). The failureto find a significant relation between buffer strips and nutrientloading in the TROLS study may be attributed to the (1)heterogeneity in hydrologic pathways among the experimentaldrainage basins, and (2) variability in the climatic conditions forthe duration of the study. Specifically, subsurface and surfaceconnectivity (through wet areas) were implicated as important inidentifying the susceptibility of receiving aquatic systems todisturbances and in determining the effectiveness of buffer stripsin mitigating the impacts of disturbances (Devito et al., 2000). Priorto offering better guidelines for buffer strips, we need a better

understanding of the spatial and temporal dynamics of wet areasand how these vary in different landscapes.

Although government policy and therefore most operationalguidelines incorporate standard width buffers to protect waterresources, there are increasing calls from the scientific communityto incorporate hydrologic dynamics in designing managementplans not only for buffer but also road and bridge placement (Smithet al., 2003; Lee and Barker, 2005). At its simplest, hydrologicallyinformed buffer design has been based on the idea that differentstream segments have differential loading potential based on theirspecific contributing areas (i.e., convergent areas have higherloading than divergent areas) and therefore, buffer widths havebeen recommended to be normalized to hydrologic loadingpotential (Bren, 1998, 2000). So far, to our knowledge, hydro-logically informed buffer design has exclusively used topographicinformation to define wet areas of the landscape, which are proneto generate saturation overland flow. However, even in humidcatchments where surface topography is a good replica of thewater-table, the topographic index alone can not identify all runoffgenerating areas (Western et al., 2004). On the sub-humid BorealPlain where there is a water deficit during most years, surfacetopography is a poor replica of the water table and thereforetopographical information alone will not be very useful inidentifying wet areas (Devito et al., 2005b). On the other hand,remote sensing techniques offer ways to detect areas of saturationand inundation over large geographic scales and independent oflandscape position (Frazier et al., 2003; Moran et al., 2004).

Recent work on the Boreal Plain has indicated that radarimagery is useful at mapping wet areas at regional scales (Sass andCreed, 2008; Clark et al., 2008). Clark et al. (2008) used EuropeanResources Satellite (ERS) imagery collected over an eleven yearperiod to delineate wet areas in a regional drainage basin andsubsequently to derive a map of probability of inundation andsaturation. Similarly, Sass and Creed (2008) mapped wet areasusing ERS imagery, however, their focus was on wet area dynamicsin the context of climatic variability. Whereas both Clark et al.(2008) and Sass and Creed (2008) focused on wet area mapping inundisturbed drainage basins, we applied their mapping techniqueto another part of the Boreal Plain, one which has experiencedsignificant amount of forest harvesting.

The focus of this paper was on using satellite-derived saturatedand inundated areas as a basis for designing hydrologicallymeaningful buffers. The objectives were to: (1) map saturatedand inundated areas (collectively termed wet areas) within adrainage basin; (2) establish links between wet area extent andhydro-climatic metrics; (3) generate probability of wet areaformation aspatially and spatially; and (4) demonstrate howprobability of wet area formation can be used to aid in the design ofbuffer strips that incorporates the natural range of variation in wetarea cover. Future studies will examine the difference betweenstandard width and hydrologically based buffer design incontrolling the nutrient loading to lakes.

2. Drainage basin description

The study area is centered on the Moose Lake drainage basin(5580705800N, 11184601700W), which is a sub-basin of the largerLogan River drainage basin. The drainage basin lies within theMixed-Wood Boreal Forest ecoregion (Rowe, 1972), which is partof the Boreal Plain ecozone (Ecological Stratification WorkingGroup, 1995) in northern Alberta, Canada (Fig. 1). The drainagebasin is 22.5 km2 in area and it is characterized by gentletopography ranging from approximately 570–650 m in elevation.The underlying glacial drift ranges from 20 to 200 m in thickness,with a stratigraphy that contains complex inter-beds of clays,

Fig. 1. Map of the Moose Lake drainage basin showing location of harvest blocks and significant hydrological features. Inset map showing the location of the drainage basin

within the mixed-wood Boreal Forest ecoregion (Rowe, 1972) and the Boreal Plain ecozone (Ecological Stratification Working Group, 1995).

Fig. 2. Inter-annual variability of precipitation (P), discharge (Q), and mean

temperature (T) for the Moose Lake drainage basin. For P and T, data were compiled

using an inverse distance-weighted average of measurements taken at four

meteorological recording stations that had a continuous record and that were

closest to Moose Lake. For Q, the Logan River discharge recording station data were

used which was the closest station to Moose Lake.

I.F. Creed et al. / Forest Ecology and Management 256 (2008) 1984–19941986

sands, and gravels (Tokarsky and Epp, 1987). The soils are highlyvariable ranging from well to poorly drained, with Gray Luvisolsand Eutric Brunisols in uplands and Gleysols in lowlands (Tokarskyand Epp, 1987). Tree species such as jack pine (Pinus banksiana

Lamb.) and balsam fir (Abies balsamea (L.) Mill.) are found in dryhabitats (Rowe, 1972). Trembling aspen (Populus tremuloides

Michx.) and white spruce (Picea glauca (Moench) Voss) arecommon in the mesic habitats, while balsam poplar (Populus

balsamifera L.), paper birch (Betula papyrifera Marsh.), tamarack(Larix laricina (Du Roi) K. Koch), and black spruce (Picea mariana

(Mill.) B.S.P.) are common in wet habitats (Rowe, 1972). Tremblingaspen is the dominant species within the drainage basin (Rowe,1972). The Boreal Plain ecozone contains numerous wetlands,including bog and fen peatlands, with marshes and swampscommon along lake margins, adding to the complexity of thehydrology in the region.

3. Hydro-climatic context

The average annual precipitation (P) and mean-temperature (T)were 457 mm year�1 and 1.6 8C, respectively, from 1970 to 2005. P

and T were spatially interpolated using inverse distance weightingfrom the closest four meteorological recording stations. Averageannual discharge (Q) was 98 mm year�1 based on the availablerecord from 1984 to 2005 measured at Logan River dischargerecording station (5581004400N, 11184304200W) located 20 km fromMoose Lake. The average runoff coefficient was 22% with amaximum of 45% reached in 1997. Inter-annual hydrologicvariability did not show any significant trends based on average

annual P, T or Q (Fig. 2). However, there were periodic peaks in bothprecipitation and discharge occurring at a frequency of 3–7 years(Fig. 2).

Intra-annual hydrologic activity was driven by spring rain-on-snow and rain events and summer convective storms (Fig. 3). Onaverage, 60% of the annual precipitation fell between May andSeptember (Fig. 3A). The period with above-freezing temperatureslasted on average from April 1 to October 31 (Fig. 3B). The springdischarge period (April 1–May 31) accounted for approximately

Fig. 3. Long term daily median and inter quartile range for calendar year of: (A)

precipitation; (B) mean temperature; and (C) discharge based on data from 1970 to

2005 for the Moose Lake drainage basin (Note: same dataset as used in Fig. 2).

Table 1Results of Kolmogorov–Smirnov two-sample tests comparing precipitation (P),

effective precipitation (P-PET), antecedent precipitation index (API), and discharge

(Q) frequency distributions for the day of and different aggregation periods

compiled from long-term record versus the same frequency distributions for the

image acquisition dates

Days (current + preceding) Z p

P

1 1.84 0.002

7 0.60 0.868

14 1.25 0.211

30 0.23 0.820

P-PET

1 1.18 0.237

7 0.20 0.838

14 0.59 0.558

30 0.76 0.451

API

1 1.94 0.053

7 2.18 0.029

14 1.37 0.047

30 0.99 0.286

Q

1 0.58 0.885

7 0.76 0.616

14 0.77 0.589

30 0.64 0.640

Note: p < 0.05 indicates statistically different distributions.

I.F. Creed et al. / Forest Ecology and Management 256 (2008) 1984–1994 1987

23% of annual discharge, whereas the summer discharge period(June 1–August 31) accounted for approximately 55% of annualdischarge (Fig. 3C).

4. Methods

Moose Lake was one of 12 experimental lakes used in a multi-year study analyzing the efficacy of standard width buffer designon reducing nutrient loading following forest harvesting (Prepaset al., 2001). We chose this lake as the topography and vegetationcover in the drainage basin was representative of this region. Weused a combination of ground and satellite-based data tocharacterize the dynamics of wet areas within the Moose Lakedrainage basin. We calculated the probability of wet areaformation, which was then used to provide a hydrologic basisfor an alternative buffer design strategy.

4.1. Satellite-based detection of wet areas

We used Radarsat Standard Beam Mode 1 (S1) images acquiredduring the growing season to capture surface hydrological

dynamics (i.e., post-snowpack; early to late growing season; andpre-snowpack). This radar sensor operates in C-band (5.66 cmwavelength), which means that the microwave can penetrate thetop 5 cm of soils. The acquired signal, taken at incidence anglesbetween 208 and 278 from nadir, is HH polarized (the signal ishorizontally transmitted and received), which investigators havefound to better penetrate vegetation and therefore detect wetnessthan VV polarized ERS images (van der Sanden et al., 2001). Allimages were acquired in ascending orbit, at 18:00 h local time.

A total of 18 images were available for analysis (1 each from1996 and 2003, 5 from 1999 and 11 from 2000). The goal was tocapture both within-year hydrologic variability and inter-annualhydrologic variability of the past >30 years. Three of the imagescaptured the spring period, 11 images captured the summer periodand 4 images captured the fall period. Overall, we found nostatistical difference between the hydro-climatic conditionscaptured by the 18 images and the 36-year record (Table 1),suggesting that our images were representative of spring-to-fallhydrologic conditions of this region.

We followed the methodology of Sass and Creed (2008) todelineate wet areas (areas of both saturation and inundation). First,images were geo-registered and rectified to the geographiccoordinates on the digital elevation model (DEM) of the drainagebasin. Following these geometric corrections, the images wereprocessed for the reduction of noise (i.e., speckle filtering) using agamma filter (Lopes et al., 1993).

The radar backscatter images (units of decibels) were convertedto volumetric soil moisture (%) based on a regression model, whichwas generated from ground-based volumetric soil moisture andcoincident backscatter signal. For the regression model, ground-based soil moisture measurements and coincident satellite-basedradar imagery were collected during three sampling dates in 2000(May 8, July 19, and October 16). Ground-based soil moisture wasmeasured using a portable soil moisture meter (TH2O) (DynamaxInc., Houston, TX), which measures the dielectric permittivity(units of millivolts) of the soil. Dielectric permittivity wasconverted to volumetric soil moisture using conversions given

I.F. Creed et al. / Forest Ecology and Management 256 (2008) 1984–19941988

by the manufacturer for mineral and organic soils. Soil moisturemeasurements were made at 7 plots representing the full range incanopy cover, including wetlands and uplands. Each plot waslocated within a larger homogenous area (minimum area of 1 ha)with respect to vegetation cover to ensure that coincidentbackscatter signal would be minimally influenced by mixed-pixeleffects (Griffiths and Wooding, 1996). At each plot, soil moisturemeasurements were taken at the center (0 m) and at 3, 6, 9, 12, and15 m along transects radiating out from the center in the N, E, S, W,NE, SE, NW, and SW compass directions, resulting in a total of 41samples. At each sampling point, the surface cover was removed,and the TH2O meter was inserted horizontally into a litter-fibric-hemic (LFH) layer in the uplands or into the 0–10 cm layer in theorganic rich lowlands. Satellite-based radar images that wereacquired coincident to the ground-based sampling were used toderive an average backscatter coefficient for each plot sampled onthe ground. Backscatter coefficients were extracted from homo-genous polygons with respect to vegetation cover, which ranged insize from 1 to 3 ha.

Regression models explained between 74 and 94% of the variancein backscatter coefficient. Analysis of covariance showed the threeregression models to be statistically similar (no difference in slopes,p = 0.84, or intercepts, p = 0.58) and the data were therefore pooled.The regression model based on pooled data (r2 = 0.81, S.E. = 0.89;p < 0.01; n = 21) was used to convert backscatter coefficients tovolumetric soil moisture for all 18 images.

We classified the images into three classes: inundated,saturated, and unsaturated. We defined unsaturated as 0–60%VSM and saturated as 60% > VSM. Inundated parts of the landscapecorresponded to anything less than �15.1 dB. In order to accountfor uncertainty in the classification we allowed a fuzzy transitionbetween classes (Sass and Creed, 2008). We determined the degreeof overlap based on frequency distributions of backscattercoefficients separated based on the ground-based volumetric soilmoisture measurements (Fig. 4). The final classified maps werereduced to a binary non-wet/wet classification by combining thesaturated and inundated classes. We masked out open water areasusing a provincial hydrography layer in order to reduce noise bywind (Sass and Creed, 2008). For each image, we computed thepercent wet area cover (WET) of the Moose Lake drainage basin

Fig. 4. Distributions of backscatter coefficients extracted under masks defined by

field-based unsaturated and saturated soil moisture data and open-water lake

inundation data. Kruskal–Wallis H test indicated statistically significant differences

between the three classes (x2 = 1095; p < 0.0001).

(open-lake areas were masked out), as well as the connected wetarea cover (WETcon). Connectivity was based on connections to alllakes greater than 4 ha as well as to connections to permanentstreams.

4.2. Wet areas and the probability of their formation

The correlations between climatic and hydrologic variables andWET were computed, in order to understand the hydro-climaticcontrols on the formation of wet areas (inundated and saturatedareas). Besides P and Q, we also computed effective precipitation(P-PET) (Hamon, 1964) and an antecedent precipitation index(after Linsley et al., 1958). The influence of the hydro-climaticvariables was tested by using the daily and the previous 7-, 14-,and 30-day totals. For P-PET, we also summed values from currentday to beginning of growing season (Sass and Creed, 2008). Weselected the variable-pair with the highest correlation coefficientfor regression modeling.

To predict the probability of wet area formation aspatially, wederived a flow-duration curve based on all daily dischargemeasurements (1984–2005). Based on the statistically significantrelationship between discharge and wet area cover, we assignedthe exceedence probabilities of daily discharge observed on theimage acquisition dates to the wet area extent derived from theimages. We also mapped the probability of wet area formationacross the Moose Lake drainage basin. We calculated the pixel-based probability surface by summing the 18 wet area maps andscaling to 0–100%. The binary wet area maps were each weightedbased on their exceedence probabilities. While the intent ofgenerating a spatial probability layer was to create somethinganalogous to the aspatial wet area–duration curve, the two werenot the same (i.e., the total wet area coverage at a given probabilityfrom the spatial layer was different from the total wet areacoverage at the same exceedence probability from Fig. 7). In ourdiscussion we refer to the spatial probability as ‘‘probability of wetarea formation’’. Finally, we used the wet area probability map tosuggest an alternative buffer design.

5. Results and discussion

5.1. Wet area dynamics: establishing the range of natural variation

The incorporation of a long hydrologic time-series that capturesthe range of natural variation is important when consideringhydrological dynamics since shorter periods (i.e., <5 years) maymiss hydrologic extremes that are important in influencing thehydro-ecological behavior of drainage basins. We considered amulti-decadal time-frame for our study (1970–2005 for P and T;1984–2005 for Q) and used it as the reference condition to whichwe compared our wet area characterization. Overall, the satellite-image database captured the long-term hydro-climatic character-istics of the Moose Lake region (Table 1).

There was substantial temporal variation in WET, ranging from alow of 5–26%. This range of wet area coverage was similar to whatother studies have found in a regional drainage basin on the BorealPlain (Sass and Creed, 2008) and a subcatchment of the Moose Lakedrainage basin (Devito et al., 2005a). The spatial organization of wetareas also showed large variability (Fig. 5). Early in the growingseason (post-snowpack), about 22% of the drainage basin wascovered by wet areas, many of them connected to lakes and streams(9%). During the middle part of the growing season before thesummer rains, WET decreased to 12%. This decrease most likelyreflected the substantial increase in evapotranspiration due to leafgrowth by end of May (Hogg et al., 2000). Summer rainssubstantially spiked the discharge, which was reflected in increased

Fig. 5. A time-series of hydrologically classified images from 2000 showing within-year hydrological dynamics of wet areas from early to late growing season. The

corresponding hyetograph and hydrograph are shown below the maps. (Note: same dataset as used in Fig. 2).

Table 2Summary of correlation between hydro-climatic variables (precipitation [P],

effective precipitation [P-PET], antecedent precipitation index [API], and discharge

[Q] and percent wet area cover (unconnected and connected to lakes and streams)

Spearman’s rho Days (current + preceding) Unconnected Connected

P 1 0.66** 0.56**

7 0.38 0.30

14 0.28 0.20

30 0.30 0.15

P-PET 1 0.27 0.35

7 0.13 0.20

14 0.22 0.21

30 0.65** 0.49*

From beginning of growing season 0.52* 0.59**

API 1 0.34 0.23

7 0.31 0.20

14 0.30 0.22

30 0.20 0.11

Q 1 0.81** 0.73**

7 0.78** 0.68**

14 0.74** 0.64**

30 0.43 0.35

Note: statistical significance * p < 0.05, ** p < 0.01.

I.F. Creed et al. / Forest Ecology and Management 256 (2008) 1984–1994 1989

WET (25%). Summer convective storms are the primary hydrologicinput into this system (Devito et al., 2005a), which translate into thehighest discharge period (Fig. 3C). It was during this wet periodwhen connectivity increased to its highest level (13%). Late in thegrowing season (pre-snowpack), WET contracted again (16%) butthey were more connected than during the pre-rain drier period.Hydrologic inputs were minimal during this fall period.

We investigated the relationship between WET and hydro-climatic variables, in order to build a model that could be used toforecast wet area behavior based on commonly measured variablessuch as discharge and precipitation. WET was significantly andpositively correlated to short term P, seasonal P-PET, and shortterm Q (Table 2). One-day Q showed the highest correlation withWET (rho = 0.81, p < 0.001). Q, like WET, is an integrative variable,reflecting cumulative system behavior. The high correlationbetween WET and P-PET (cumulated at least 30 days) showedthat WET reflects long-term memory of hydrologic inputs.Correlation analysis between 1-day Q and P-PET cumulated overdifferent time-periods also revealed that 1-day Q reflects long-term behavior in the system.

The scatter-plot of Q versus WET revealed a semi-logarithmicrelationship (Fig. 6) with Q explaining 72% of the variance in WET.This relationship signified that at high daily discharges, WETreached a maximum (near 25%) and no further increase in wet areasize could be expected with increased discharge. It is likely thatduring these very wet periods, areas of saturation were beginningto get inundated and grow more in volume rather than area.

5.2. Predicting the probability of wet area formation

The flow–duration curve for all daily measurements of daily Q

from 1984 and 2005 showed the typical rotated S pattern with high

flows having the lowest exceedence probabilities and low flowshaving the highest exceedence probabilities (Fig. 7). The wet area–duration curve showed a similar pattern, however, lack of datameant that the distribution extremes were not well defined. Theexceedence probability of 5–10% WET of the drainage basin,reflective of dry conditions, was approximately 60%. For very wetconditions (�25% WET), the exceedence probability was approxi-mately 5%. From this we concluded that a majority of the time WET

Fig. 6. Relation between daily discharge (day of image acquisition) and percentage

wet area cover for the Moose Lake drainage basin.

Fig. 8. Map of probability of wet area formation, where 100% depicts highest

likelihood of finding wet (saturated or inundated) areas.

I.F. Creed et al. / Forest Ecology and Management 256 (2008) 1984–19941990

can be expected to fluctuate between 5 and 10%, only duringinfrequent wet periods can WET be expected to increasesubstantially.

The map of probability of wet area formation showedqualitatively a well-defined spatial structure (Fig. 8). Parts of thelandscape with the highest probability of wet area formation wereobserved near lake and stream riparian zones, although there wereother smaller pockets of high probability wet areas away fromthese bodies of receiving water (Fig. 8). Most of the existingcutblocks fell within parts of the landscape of low probability ofwet area formation, however, there were some areas where theyintersected wet areas and streams (Fig. 9). These are areas ofconcern, where the highest impacts could be expected as a result ofincreased nutrient or sediment loading.

5.3. Implications for design of buffer strips

Timber harvesting has the potential to alter wet area dynamics,either within or adjacent to these areas. Harvesting within andadjacent wet areas may lead to more frequent saturation andinundation because of decreased evapotranspiration after plantremoval and enhanced channelization due to heavy equipmentimpacts and subsequently more efficient transfer of sediments andnutrients. To mitigate these impacts, the standard practice forforest companies is that they leave standard widths of riparianforest buffer strips around surface water bodies to serve as sinks forenhanced particulate and dissolved nutrients in hydrologic flowsresulting from harvest activities (Castelle et al., 1994; Lee et al.,2004). Current forest management guidelines in Alberta require a

Fig. 7. Flow–duration curve for daily discharge (1984–2005) and wet area–duration

curve for satellite-images based on exceedence probabilities of daily discharge

observed on image acquisition dates.

standard width of buffer strip of 20 m around wetlands, 30 maround streams, 60 m around rivers, and 100 m around lakes(Alberta Environmental Protection, 1994) (Fig. 10A). In Alberta, thedelineation of these hydrologic features is based on photo-interpretation of aerial photographs at cartographic scales of1:20,000 to 1:50,000 (Alberta Vegetation Inventory, 2007). Loggingplanners use these hydrographic maps to allocate buffers andsubsequently, in combination with maps of forest productivity,delineate areas that can be harvested. However, these standardbuffer widths do not fully reflect the hydrologic characteristics ofthe landscape, especially in landscapes where the substrate has animportant control on hydrologic fluxes (Buttle, 2002). Conse-quently, the standardized widths of buffer strips may offer toomuch or not enough protection for surface water quality (Belt andO’Laughlin, 1994).

An alternative to the standard width design of buffer strips thatincorporates the range in natural variation of hydrologic pathwayswithin the Moose Lake drainage basin is demonstrated in(Fig. 10B). Under this proposed system, the level of protection isbased on the risk tolerance of the policy maker or resourcemanager. The riskiest approach would be to only protect wet areasthat have a high probability of formation (e.g., 75% or higher)(Fig. 10B). This level of protection would only target a small areaand would most likely lead to higher impacts as many wet areaswould not be protected. In contrast, the safest approach would beto target wet areas that have a much lower probability of formation(e.g., 25% or higher) (Fig. 10B). At this probability level, a muchlarger area would be protected, similar in size to the area protectedunder the standard buffer width (Fig. 10A).

For the Moose Lake drainage basin, a buffer based on aprobability of wet area formation of 25% would protect an area ofsimilar size as one with the standard approach (approximately 15%of the drainage basin). However, the spatial distribution of the wetareas and associated buffers would be somewhat different underthe two approaches (63% of the probability based wet area wascontained within the standard approach). The differences couldmost likely be attributed to the inability of photo-interpretivemethods to capture hydrological dynamics, in ephemeral channels

Fig. 9. Map of probability of wet area formation superimposed with cutblock locations. The two highlighted areas depict areas where the cutblocks intersect hydrologic

features (i.e., wetlands, streams) that are connected to Moose Lake.

I.F. Creed et al. / Forest Ecology and Management 256 (2008) 1984–1994 1991

for example, that were not reflected by hydric vegetation or bymapping wet areas that do not wet-up at a probability level of 25%.Some of the differences could also be attributed to misclassifica-tion of the radar images from which the probability layer of wetareas was calculated. Sass and Creed (2008) reported a classifica-tion accuracy of 88% in differentiating saturated from non-saturated parts of the landscape.

The selection of specific probabilities to form a basis of ahydrologically based design for buffer strips should consider anumber of factors including the specific goals of the buffer such asnutrient or sediment reduction and economics. For example, ifsediment movement is the concern than wet areas with a lowprobability of formation (e.g., >1%) would be selected. If nutrienttransfer was the concern, the probability could perhaps be setbetween 20 and 30%. For example, Walter et al. (2000) advocatedusing a probability of 30% in identifying wet areas for protection ina nutrient loading context. Eventually this probability level couldbe quantified based on field-experiments although the difficulty incontrolling for climate and landscape differences, as shown by(Prepas et al., 2001), would be significant. Research is also neededin determining how the timing of harvesting operations (e.g.,winter or dry season harvesting) influences the probability levelchosen for minimizing impacts. Harvesting during hydrologically

inactive parts of the year will minimize impacts (e.g. rutting) andfacilitate quicker recovery rates.

From an economic perspective, forest managers would also beinterested in how much of the productive land-base (for timber) dothe different levels of buffer designs prevent from being harvested.In the Moose Lake drainage basin, the total productive land area(loggable resource) was 13.9 km2 or 62% of the total drainage basinarea (Alberta Vegetation Inventory). Of this productive land-base,92.5% would have been available for logging under the standardbuffer width buffer design. Under the probability approach, theavailable productive land area would have decreased from near97% at a probability of 99% to about 37% at a probability of 1%(Fig. 11). At a probability of 25%, the productive land base wouldhave decreased to 87.1% (Fig. 11). By combining the hydrologicaland economic perspectives, the optimum probability of wet areaformation would minimize the ‘loss’ of loggable resource andmaximize the wet areas protection. In our example, this optimumprobability would be approximately 20% (Fig. 11).

The general application of a remote sensing derived probabilitybased way of selecting buffers will face two important considera-tions: (1) the hydrologic relevance of assessing hydrologicaldynamics by capturing a time-series of radar images; and (2)technical feasibility of deriving hydrodynamics from a time-series

Fig. 10. Identification of riparian buffers based on (A) standard width approach and (B) probability of wet area formation. For B, a narrow definition of buffers might only

protect areas that have at least 75% probability of being wet. A broader definition of buffers would target all areas of the drainage basin that have a probability of being wet 25%

of the time or more.

I.F. Creed et al. / Forest Ecology and Management 256 (2008) 1984–19941992

of radar images. With respect to relevance, in humid landscapeswhere there are well formed drainage networks and topography isthe main control on water flow, knowledge of the stream networksand identification of wet areas using terrain analysis will suffice

Fig. 11. Trade-off between wet areas protection and available loggable resource at

different levels of probability of wet area formation.

(e.g., Chappell et al., 2006). Therefore, the problem of omittinghydrologically active areas when using standard width bufferzones would be less acute. However, even in topographicallycontrolled landscapes, the temporal dynamics of the systemshould be incorporated into selecting the appropriate protectionfor wet areas. For example, a wetness index could be calibratedusing soil moisture information to reflect something akin to aprobability of wet area formation. More complex hydrologicalmodels might also form a useful basis to predict the temporalvariability in wet areas (e.g., Agnew et al., 2006). In landscapeswhere geology and soils supersede topography in controlling themovement of water, radar imagery offers an excellent way toderive a probability layer of wet area formation. However, thisinvolves technical and financial considerations. On the technicalside, dense forest canopies make it difficult for the radar signal topenetrate to the ground. In deciduous forests, this could becircumvented by imaging during leaf-off conditions. A newgeneration of satellites, including RADARSAT-2, will offer muchimproved vegetation penetration as a result of multi-polarized andmulti-frequency imaging. In terms of finances, the total commer-cial cost of 18 archived RADARSAT-1 images would be US$ 65,000

I.F. Creed et al. / Forest Ecology and Management 256 (2008) 1984–1994 1993

(MacDonald, Dettwiler and Associates Ltd, http://gs.mdacorpor-ation.com/). This does not include programming fees for theacquisition of new images and image processing in general. Forexample, if we assume US$ 100,000 for total costs, the cost per1 km2 of productive land area would be US$ 16, probably areasonable cost for most companies.

We recognize that buffer design has multiple hydro-ecologicaland socio-economic objectives. These values need to be prioritizedgiven the locale and buffer design needs to reflect these priorities.For instance, in an area where nutrient loading and ecologicalcorridors are both high priorities, buffer design could be based on ahybrid approach where probability based buffers define areas ofprotection for nutrient and sediment transfer considerations and inaddition standard width buffers are placed around some water-courses for ecological connectivity to facilitate movement oforganisms for example. An industry-wide implementation ofapproaches other than the current standard width approach willhinge on the practicality and economics of these alternativeapproaches.

6. Conclusions

Surface saturated or inundated areas facilitate the movement ofnutrients and sediment from terrestrial to aquatic systems, andprovide habitat and migratory pathways for flora and fauna. Weshowed how hydrological dynamics of such wet areas could be builtinto forest management practices to better protect the integrity ofboreal forest ecosystems on the Boreal Plain. We developed aremote-sensing based approach that predicted the probability ofwet area formation based on the range in the natural variation inclimate. This approach offers an alternative to the standard widthdesign of buffer strips for mitigating nutrient loads to surface watersfollowing harvesting activities by (1) recognizing that the naturalvariation in wet areas, which regulate nutrient loading to lakes mustbe maintained, and (2) identifying the probability of formation ofwet areas at any given point in the watershed. The alternative to thestandard width design of buffer strips would prevent harvestactivities within areas of high probability of wet area formation andthus locate buffer strips along the boundaries of wet areas, wheredrainage of waters from harvested areas occur. Future research willfocus on testing the effectiveness of hydrologically based buffers inreducing nutrient loading under a range of climatic, landscape anddisturbance conditions.

Acknowledgements

This research was funded by the Canadian Network of Centresof Excellence on Sustainable Forest Management (NCE-SFM) grantand Premier’s Research Excellence Award awarded to I.F. Creed.Research data were provided by the Government of Alberta –Alberta Environment (K. Dutchak, R. Sleep), Alberta Pacific ForestIndustries (M. Spafford, C. Kemble), and the Government of Canada– Environment Canada (P. Chambers, N. Hedstrom).

References

Agnew, L.J., Lyon, S., Gerard-Marchant, P., Collins, V.B., Lembo, A.J., Steenhuis, T.S.,Walter, M.T., 2006. Identifying hydrologically sensitive areas: bridging the gapbetween science and application. J. Environ. Manage. 78, 63–76.

Alberta Environmental Protection, 1994. Alberta timber harvest planning andoperating ground rules, Natural Resources Service, Recreate and ProtectedAreas Division. Edmonton, Alberta.

Alberta Vegetation Inventory, 2007. Alberta Vegetation Inventory Standard Version2.1.1. http://srd.alberta.ca/forests/researcheducation/vegetationinventory.aspx(Accessed April 2008).

Belt, G.H., O’Laughlin, J., 1994. Buffer strip design for protecting water quality andfish habitat. West. J. Appl. For. 9, 41–45.

Bren, L.J., 1998. The geometry of a constant buffer-loading design method for humidwatersheds. Forest Ecol. Manage. 110, 113–125.

Bren, L.J., 2000. A case study in the use of threshold measures of hydrologic loadingin the design of stream buffer strips. Forest Ecol. Manage. 132, 243–257.

Buttle, J.M., 2002. Rethinking the donut: the case for hydrologically relevant bufferzones. Hydrol. Process 18, 3093–3096.

Buttle, J.M., Creed, I.F., Pomeroy, J.W., 2000. Advances in Canadian forest hydrology,1995–1998. Hydrol. Process 14, 1551–1578.

Buttle, J.M., Creed, I.F., Moore, R.D., 2005. Progress in Canadian forest hydrology:1999–2003. Hydrol. Process 19, 169–200, doi:10.1002/hyp.5773.

Carignan, R., D’Arcy, P., Lamontagne, S., 2000. Comparative impacts of fire and forestharvesting on water quality in Boreal Shield lakes. Can. J. Fish. Aquat. Sci. 57,105–117.

Castelle, A.J., Johnson, A.W., Conolly, C., 1994. Wetland and stream buffer sizerequirements: A review. J. Environ. Qual. 23, 878–882.

Chappell, N.A., Vongtanaboon, S., Jiang, Y., Tangtham, N., 2006. Return-flow pre-diction and buffer designation in two rainforest headwaters. Forest Ecol.Manage. 224, 131–146.

Clark, R.B., Creed, I.F., Sass, G.Z., 2008. Mapping hydrologically sensitive areas on theBoreal Plain: A multi-temporal analysis of ERS synthetic aperture radar data.Int. J. Remote Sens., in press.

Creed, S.E., Sanford, S.E., Beall, F.D., Molot, L.A., Dillon, P.J., 2003. Cryptic wetlands:integrating hidden wetlands in regression models of the export of dissolvedorganic carbon from forested landscapes. Hydrol. Process 17, 3629–3648.

Devito, K.J., Creed, I.F., Rothwell, R.L., Prepas, E.E., 2000. Landscape controls onphosphorous loading to boreal lakes: implications for the potential impacts offorest harvesting. Can. J. Fish. Aquat. Sci. 57, 1977–1984.

Devito, K.J., Creed, I.F., Fraser, C., 2005a. Controls on runoff from a partiallyharvested aspen forested headwater catchment, Boreal Plain, Canada. Hydrol.Process 19, 3–25.

Devito, K.J., Creed, I.F., Gan, T., Mendoza, C.A., Petrone, R., Silins, U., Smerdon, B.D.,2005b. A framework for broad-scale classification of hydrologic response unitson the Boreal Plain: is topography the last thing to consider? Hydrol. Process 19,1705–1714, doi:10.1002/hyp.5881.

Dillon, P.J., Molot, L.A., 1997. Effect of landscape form on export of dissolved organiccarbon, iron and phosphorous from forested stream catchments. Water Resour.Res. 33, 2591–2600.

Dillon, P.J., Molot, L.A., 2005. Long-term trends in catchment export and lakeretention of dissolved organic carbon, dissolved organic nitrogen, total iron,and total phosphorus: The Dorset, Ontario Study, 1978–1988. J. Geophys. Res.110, G01002, doi:10.1029/2004JG000003.

Ecological Stratification Working Group, 1995. A national ecological framework forCanada, Report and national map at 1:7 500 000 scale, Agriculture and Agri-FoodCanada, Research Branch, Centre for Land and Biological Resources Researchand Environment Canada, State of Environment Directorate, Ecozone AnalysisBranch, Ottawa/Hull.

Frazier, P., Page, K., Louis, J., Briggs, S., Robertson, A.I., 2003. Relating wetland inunda-tion to river flow using Landsat TM data. Int. J. Remote Sens. 24, 3755–3770.

Global Forest Watch Canada, 2002. Canada’s Forests at a Crossroads: An assessmentin the Year 2000, Edmonton, Alberta, 107 pp.

Griffiths, G.H., Wooding, M.G., 1996. Temporal monitoring of soil moisture usingERS-1 SAR data. Hydrol. Process 10, 1127–1138.

Hamon, R.W., 1964. Computation of direct runoff amounts from storm rainfall.International Association of Scientific Hydrology, Publication 63, pp. 52–62.

Hogg, E.H., Price, D.T., Black, T.A., 2000. Postulated feedbacks of deciduous forestphenology on seasonal climate patterns in the western Canadian interior. J.Climate 13, 4229–4243.

Lee, P., Barker, T., 2005. Impact of riparian buffer guidelines on old growth inwestern boreal forests of Canada. Forestry 78, 263–278.

Lee, P., Smyth, C., Boutin, S., 2004. Quantitative review of riparian buffer widthguidelines from Canada and the United States. J. Environ. Manage. 70, 165–180.

Linsley, R.K., Kohler, M.A., Paulhus, J.L.H., 1958. Hydrology for Engineers. McGraw-Hill Book Company Inc., New York.

Lopes, A., Nezry, E., Touzi, R., Laur, H., 1993. Structure detection and statisticaladaptive speckle filtering in SAR images. Int. J. Remote Sens. 14, 1735–1758.

Moran, M.S., Peters-Lidard, C.D., Watts, J.M., McElroy, S., 2004. Estimating soilmoisture at the watershed scale with satellite-based radar and land surfacemodels. Can. J. Remote Sens. 30, 805–826.

Pinel-Alloul, B., Prepas, E.E., Planas, D., Steedman, R., Charette, T., 2002. Watershedimpacts of logging and wildfire: case studies in Canada. Lake Reserv. Manage.18, 307–318.

Prepas, E.E., Pinel-Alloul, B., Planas, D., Methot, G., Paquet, S., Reedyk, S., 2001. Forestharvest impacts on water quality and aquatic biota on the Boreal Plain:Introduction to the TROLS lake program. Can. J. Fish. Aquat. Sci. 58, 421–436.

Rowe, J.S., 1972. Forest Regions of Canada. Department of the Environment,Canadian Forestry Service, Ottawa, Ontario, Canada (Publication Number 1300).

Sass, G.Z., Creed, I.F., Devito, K.J., 2008. Spatial heterogeneity in trophic status ofshallow lakes on the Boreal Plain: influence of hydrologic setting. Water Resour.Res. 44, W08444, doi:10.1029/2007WR006311.

Sass, G.Z., Creed, I.F., 2008. Characterizing hydrodynamics on boreal landscapesusing archived synthetic aperture radar imagery. Hydrol. Process 22, 1687–1699, doi:10.1002/hyp.6736.

Smith, D.W., Russell, J.S., Burke, J.M., Prepas, E.E., 2003. Expanding the forestmanagement framework in the province of Alberta to include landscape-basedresearch. J. Environ. Eng. Sci. 2, S15–S22.

I.F. Creed et al. / Forest Ecology and Management 256 (2008) 1984–19941994

Steedman, R.J., 2000. Effects of experimental clearcut logging on water qualityin three small boreal forest lake trout lakes. Can. J. Fish. Aquat. Sci. 57,92–96.

Tokarsky, O., Epp, A.E., 1987. Winnifred Lake 73M, Alberta Environment: EarthSciences Division for Water Resources Management Services.

van der Sanden, J.J., Budkewitsch, P., Flett, D., Landry, R., Lukowski, T.I., McNairn, H.,Pultz, T.J., Singhroy, V., Sokol, J., Touzi, R., Vachon, P.W., 2001. Applicationspotential of planned C-band SAR satellites: leading to RADARSAT-2. In:

Proceedings of the International Geoscience and Remote Sensing Symposium,Sydney, Australia, July 9–13, pp. 488–492, doi:10.1109/IGARSS.2001.976199.

Walter, M.T., Walter, M.F., Brooks, E.S., Steenhuis, T.S., Boll, J., Weiler, K., 2000.Hydrologically sensitive areas: variable source area hydrology implication forwater quality risk assessment. J. Soil Water Conserv. 55, 277–284.

Western, A.W., Zhou, S., Grayson, R.B., McMahon, T.A., Bloschl, G., Wilson, D.J., 2004.Spatial correlation of soil moisture in small catchments and its relationship todominant spatial hydrological processes. J. Hydrol. 286, 113–134.