Season- and depth-dependent time stability for characterising representative monitoring locations of soil water storage in a hummocky landscape

Catena 116 (2014) 38–50

Contents lists available at ScienceDirect

Catena journal homepage: www.elsevier.com/locate/catena

Season- and depth-dependent time stability for characterising representative monitoring locations of soil water storage in a hummocky landscape Asim Biswas ⁎ Department of Natural Resource Sciences, McGill University, 21111 Lakeshore Road, Ste-Anne-de-Bellevue, QC, H9X 3V9, Canada

a r t i c l e

i n f o

Article history: Received 5 July 2013 Received in revised form 11 December 2013 Accepted 21 December 2013 Keywords: Root zone Prairie pothole region Spatial pattern Inter-season Intra-season Inter-annual

a b s t r a c t Characterizing the inherent spatial variability of soil water is intensive in terms of sampling effort and therefore costly. The objectives of this study were to examine time stability of the spatial pattern of soil water storage (SWS) at different seasons and depths, to identify representative monitoring locations (RMLs) that consistently reflect field average SWS, and to recommend efficient ways to identify these locations. Soil water was measured down to 1.4 m at 0.20-m vertical depth intervals using time domain reflectometry and neutron backscattering along a hummocky transect in semiarid central Canada. Strong Spearman rank correlation coefficients between SWS spatial series indicated similarity of spatial patterns, which showed strong seasonal dependence: intraseason time stability was stronger than inter-annual time stability, which was stronger than inter-season time stability. The observed time stability was used to identify an RML. The RML for the surface layer (0–0.20 m) was different from the root zone (0–0.60 m) and the total active soil profile (0–1.20 m), the latter two having adjacent RMLs. The bias (~3%–10% maximum) associated with using an RML to predict field-averaged SWS was examined using regression. However, the prediction of field-averaged SWS using an RML was better compared to the prediction using field extreme locations in terms of error associated with the prediction. Predictions using more than one RML yielded better results than when a single RML was used. Successful and rapid identification of RMLs greatly reduced the number of observations needed to characterize the average soil water behaviour of a field. The measurement of SWS through representative monitoring location(s) may be used for environmental monitoring and modelling, irrigation scheduling, nutrient recommendations, and predicting greenhouse gas emissions. © 2013 Elsevier B.V. All rights reserved.

1. Introduction Soil water is the principal limiting factor in semiarid agricultural production. It also influences environmental health through its role in transporting materials (e.g., agricultural chemicals and nutrients) to environmentally sensitive assets such as rivers and groundwater (Sun, 1986). Soil water content is a key variable for describing many hydrological and climatic processes such as the partitioning of rainfall and snowmelt water into infiltration and runoff and the evapotranspiration process (Famiglietti and Wood, 1994). Improved knowledge of soil water patterns is a prerequisite for better hydrologic and climatic modelling. However, understanding these processes is a major challenge in hydrology because the distribution of soil water is controlled by several factors and processes including soil properties, topography, vegetation, ground water, and climatic and water routing processes. The individual and combined effect of these factors and processes yield large spatiotemporal heterogeneity in soil water distribution

⁎ Tel.: +1 514 398 7620; fax: +1 514 398 7990. E-mail address: [email protected]. 0341-8162/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.catena.2013.12.008

within a field (Brocca et al., 2010; Seyfried, 1998). Therefore, a large number of samples are usually required to characterize the soil water content of a field. A conventional way of the determining soil water content of a field is by measuring a number of randomly selected locations and by averaging them over the measurement area. Such conventional sampling assumes that soil water content is randomly distributed in a field (Bell et al., 1980). Random water content distribution necessitates a large number of samples to account for spatial variability and to obtain a desired level of accuracy and precision in soil water measurement. Fortunately, the factors controlling soil water exhibit non-random patterns in a field and give rise to patterns in soil water and its storage (Grayson and Western, 1998; Kachanoski and Dejong, 1988). The similarity of this pattern over time was first observed by Vachaud et al. (1985). They introduced this concept as ‘time stability’ (TS), which is defined as a time invariant association between spatial location and classical statistical measures of soil water, most often the mean (Grayson and Western, 1998). A large body of literature (Brocca et al., 2009; Grayson and Western, 1998; Hu et al., 2010a; Kachanoski and Dejong, 1988; Tallon and Si, 2004) has used this concept to explain the temporal dynamics of soil water over a variety of areas (from a few m2 to 109 m2), time periods,

A. Biswas / Catena 116 (2014) 38–50

land uses, and locations and using different sampling schemes (random, grid, and transect) and sampling depths (0.05 m to ~2 m). The TS research has found that certain locations maintain their water content ranking and consistently show greater, average, or lesser water content than other locations irrespective of dry or wet climate conditions. Therefore, there will be a location that closely represents the field-averaged soil water. The relative difference between the average soil water of that location and the field will be close to zero and the location can be used to represent the field-averaged soil water at any time. The location is described as the representative monitoring location (RML), which is time invariant. Grayson and Western (1998) identified this location as a catchment average soil moisture monitoring (CASMM) site. Identifying an RML that consistently exhibits soil water content close to the field average would greatly reduce the number of samples or observations needed to characterize a field. In identifying the time-stable RML, the mean relative difference (MRD) and the standard deviation of the relative difference (SDRD) have been introduced by Vachaud et al. (1985). Various other indices have also been used to identify the locations that consistently exhibit behaviour similar to the mean through time. For example, Jacobs et al. (2004) introduced root mean squared error (RMSE) of the relative difference and associated variance. Guber et al. (2008) computed a new TS index, Tik, on the width of 90% empirical tolerance interval of empirical probability distribution functions of relative soil water content and used the root mean squared difference, Dik, to select the best locations. Hu et al. (2010b) have reported mean absolute biased error (MABE) to identify time-stable locations. However, the most widely accepted and used index is the SDRD and will be used in this study. Grayson and Western (1998) identified the location of catchment average soil moisture monitoring site s in areas that were neither topographically convergent nor divergent. These locations tended to be located near the mid-slopes or in areas that had topographic aspects close to the average and were identified as the ‘aspect neutral’ part of the transect. Jacobs et al. (2004) and Vivoni et al. (2008) also documented locations that consistently maintained TS as being located at mid slopes and mid-elevation, respectively. Most of the efforts in indentifying the time-stable locations have focussed on the surface soil water (Brocca et al., 2009; Grayson and Western, 1998; Vachaud et al., 1985; Zhao et al., 2010), only a few considered water deeper in the soil profile (Hu et al., 2009; Pachepsky et al., 2005). Only a small number of studies have explored the changes in TS of the spatial pattern with depth (Cassel et al., 2000; MartinezFernandez and Ceballos, 2003). Here I investigate the TS of the spatial pattern of soil water in the surface layer, root zone, and total active soil profile. The surface layer is the soil zone that is subject to climate forcing, the root zone is where most of the extraction of water by plants occurs, resulting in strong variability in soil water over time, and the total active soil profile is the zone below which the seasonal changes in soil water are suppressed. The hydrology of these soil layers is dictated by the dominant processes operating at the various depths. Knowledge of the TS of the spatial pattern of soil water storage (SWS) at different layers of soil with variable thickness would provide a better understanding of soil water dynamics from the surface throughout the whole soil profile. Researchers have made a distinction between wet and dry periods, or preferred states, in soil water patterns. (Grayson et al., 1997) separated the factors in ‘local’ (e.g., soil properties and micro-topography) and ‘nonlocal’ factors (e.g., drainage lines due to catchment topography). The local controls dominate in dry conditions, while the nonlocal controls dominate in wet conditions. While various authors (Famiglietti et al., 1998; Western et al., 1999) reported high degrees of time-stable spatial patterns during wet periods contributed from the higher topographic organization, others (Gomez-Plaza et al., 2000; MartinezFernandez and Ceballos, 2003) found strong time-stable spatial patterns during dry periods. There is no comprehensive information about how TS persists within a season (intra-season) with similar moisture condition, between seasons (inter-season) with different moisture condition,

39

and between the same seasons of different years (inter-annual) with similar moisture condition. The similarity of the spatial patterns from submonthly to intra-annual time frame would allow understanding the similarity in hydrological dynamics. TS has been studied in different landscape conditions with varying slopes (0%–~40%). A comprehensive review of different study conditions was presented in a table by Brocca et al. (2009). Few studies have been carried out in rolling landscape (Jacobs et al., 2004; Zhao et al., 2010), and even fewer were completed in a hummocky landscape (Kachanoski and Dejong, 1988; Tallon and Si, 2004). The hummocky landscape contains a complex sequence of knolls and depressions. The depressions generally have no permanent inlet or outlet and work as micro-watersheds (Parsons et al., 2004). These depressions are known as potholes, and the area is known as the prairie pothole region. The hummocky landscape of the prairie pothole region covers more than 775,000 km2 area in the North American prairie region and provides various hydrological and ecological functions (National Wetlands Working Group, 1997). The area is known as the ‘duck factory’ of North America as the depressions provide 50%–80% of nesting habitat for North American waterfowl population. It acts as a sink for agriculture-derived nutrients (Whigham and Jordan, 2003) and a store for surface water, which can attenuate flood flows. In understanding the hydrology of this landscape, better information is necessary on the spatial pattern of SWS and its stability over time at different seasons and depths. The objectives of this study were (a) to examine the TS of spatial patterns in SWS at different time scales (submonthly to interannual) and depths and (b) to identify the time-stable location with fieldaveraged SWS (RML) in a hummocky landscape in semiarid climate. The time-stable location or RML is potential to improve the sampling efficiency by reducing the time, cost, and labour of monitoring SWS in field. Therefore, in this manuscript, I wanted to examine the seasondependent (intra-season, inter-season, and inter-annual) TS at the classified soil layers of different depths. With the increment of depths of soil layers (e.g., 0–20 cm for the surface layers, 0–60 cm for the root zone, and 0–120 cm for the total active soil profile), the hydrological processes changes and thus the spatial patterns. The behaviour of depth-dependent spatial patterns over time was examined in this manuscript. Based on the behaviour of the spatial patterns, an attempt was made to identify the most time-stable location to represent the field-averaged SWS. 2. Materials and methods 2.1. Site description A sampling transect extending north–south direction was established at St. Denis National Wildlife Area (SDNWA) (52°12′N latitude, 106°50′W longitude), which is located approximately 40 km east of Saskatoon, Saskatchewan, Canada (Fig. 1). A detailed description of the study site can be found in Biswas and Si (2011a, 2011b) and Biswas et al. (2012). Briefly, the landscape of the study area contains a complex sequence of slopes originating from different-sized rounded depressions to irregular complex knolls and knobs, making the terrain hummocky with 10% to 15% slope (Fig. 1). The study area is representative of the prairie pothole region. The major soil type of the study area is the Dark Brown Chernozem (Mollisol in US taxonomy) developed from moderately fine to fine textured, moderately calcareous, glaciolacustrine deposits and modified glacial till (Saskatchewan Centre for Soil Research, 1989). The climate of this area is mainly semiarid with the mean annual air temperature (at the Saskatoon airport) of 2 °C with the monthly mean of −19 °C in January and 18 °C in July. Snow generally covers the area for 5 to 6 months (November to April) and contributes approximately 30% of the total precipitation (Pomeroy et al., 2007). The 90-year mean annual precipitation in Saskatoon is 360 mm, of which 84 mm occurs in winter, mostly as snow. The study area received a total precipitation 489 mm in 2006, 366 mm in 2007, 331 mm in 2008, and 402 mm in 2009 (Fig. 2). The years 2010 and

40

A. Biswas / Catena 116 (2014) 38–50

Fig. 1. Geographic location of study site (St. Denis National Wildlife Area (SDNWA), Saskatchewan, Canada) and the transect position on the hummocky landscape of the Prairie Pothole Region of North America.

2011 received higher than normal spring and summer precipitation. April to September precipitation in 2010 was 645 mm (Environment Canada, 2011b)—almost double the average annual precipitation. 2.2. Data collection A 576-m long transect with 128 equally spaced points was established over several rounded knolls and depressions, representing different landform cycles (Fig. 1). The regular sampling interval of 4.5 m along the transect was chosen to capture the systematic variability of soil water in the field. A topographic survey of the site was completed using an aerial light detection and ranging (LiDAR) survey of the study area at a 5-m ground resolution. The digital elevation map (DEM; Fig. 1) was prepared at that same resolution. The vegetation of the study site was mixed grass including Agropyron elongatum (Host) Beauvois, Agropyron intermedium (Host) Beauvois, Bromus biebersteinii Roem and Schult., Elymus dauricus Turcz. Exgriseb., Festuca rubra L.,

Onobrychis viciifolia Scop., Elymus canadensis L., Agropyron trachycaulum (Link) Malte, and Medicago sativa L., which was seeded in 2004 and allowed to grow every year. Before the grass was seeded, the study area was under cultivation in a crop-fallow sequence. Each sample location along the transect had a 50-mm diameter and 2.00-m long plastic (polyvinyl chloride; PVC) neutron moisture meter access tube installed. A truck-mounted hydraulic press was used to extract a soil core using a hollow tube sampler and then to carefully press the access tube into place. The open ends of the PVC tubes were kept closed with caps to prevent water entry from the top. A neutron moisture meter (model: CPN 501 DR Depthprobe; CPN International Inc., Martinez, CA, USA) was used to measure the soil water down to 1.40 m depth at 0.20-m intervals. Additional access tubes were installed using the same method to enable development of a site-specific calibration of the neutron probe. Neutron count was recorded at 0.10 m depth intervals to capture small variations in soil water content with depth. The standard neutron

140 2006 (489)

Precipitation (mm)

2007 (366) 2008 (331)

105

2009 (402) 90 yrs Normal (360)

70

35

0

Jan.

Feb. Mar. Apr.

May

Jun.

Jul.

Aug. Sept.

Oct.

Nov.

Dec.

Months Fig. 2. Monthly precipitation data for the years of 2006, 2007, 2008, and 2009 with 90 years average from Saskatoon International Airport (40 km west of study site).

A. Biswas / Catena 116 (2014) 38–50

count was also recorded. Soil cores (50 mm diameter, 2.00 m deep) were immediately collected adjacent to these access tubes, sectioned into 0.10 m lengths, wrapped with plastic film to prevent water loss, and transported to the laboratory. Care was taken to reduce the compaction during soil core collection (e.g., use of thin walled sampler, bevelled cutting edge). The length of collected soil cores and the depth of the holes were compared to check for compaction. Cores showing any signs of compaction were discarded. The volumetric water content of each 0.10-m segment of soil core was determined gravimetrically and dry bulk density determined. The above procedure was repeated several times at different moisture conditions over 3 years (2007–2009) at different topographic locations. Volumetric soil water content was compared with the neutron count ratio (neutron count/standard neutron count), and a calibration equation was developed (θv = 0.8523 P + 0.0612 with n = 101 and r2 = 0.86, where P is the neutron count ratio; Fig. 3). Although the measurements were completed at different depths, only one calibration equation was developed from the measurements at different depths. This is because no significant differences were observed between depth-wise calibration equations, and thus a single calibration equation was used for the simplicity. As neutron probe measurements near the soil surface are prone to error, the average soil water at 0–0.20 m was measured using time domain reflectometry (TDR) with a metallic cable tester (Model 1502B; Tektronix, Beaverton, OR, USA). The TDR probes were inserted each time of measurement into the soil as close as possible (within 10 cm radius from the tube) to neutron access tube during each time of measurement. A site-specific calibration was not p completed for TDR but a standard calibration ffiffiffiffiffi equation (θv ¼ 0:115 ka −0:176, where ka = [L2/L]2 is the dielectric constant, L2 is the distance between the arrival of signal reflected from the probe-to-soil interface and the signal reflected from the end of the probe curves (measured from waveform), and L is the length of the TDR probe) was used to calculate the soil water of surface soil following Topp and Reynolds (1998). Soil water was measured in the field 25 times over a 5-year period (17 July 2007, 7 August 2007, 1 September 2007, 12 October 2007, 2 May 2008, 31 May 2008, 21 June 2008, 16 July 2008, 23 August 2008, 17 September 2008, 22 October 2008, 20 April 2009, 7 May 2009, 27 May 2009, 21 July 2009, 27 August 2009, 27 October 2009, 6 April 2010, 19 May 2010, 14 June 2010, 28 September 2010, 13 May 2011, 6 June 2011, 29 June 2011, and 29 September 2011), covering different environmental conditions including spring snowmelt, summer, and fall. The first measurement of each year was completed within a few days of snowmelt to determine ‘initial’ soil water content. Further measurements were completed based on the environmental conditions (e.g., rainfall or a prolonged dry period) to gain representation of

41

different soil water conditions. The soil profile in the study area undergoes a freeze–thaw cycle. Measurements were timed to allow some redistribution of water after rainfall events. SWS at each sampling time was calculated for the surface layer (0–0.20 m depth), the root zone (0–0.60 m depth), and the total active soil profile (0–1.20 m depth). These depth ranges were used irrespective of soil horizons present at different landscape positions. 2.3. Data analysis (TS analysis) There are two groups of methods to characterize TS (Vanderlinden et al., 2012). Among the first group of methods, mean relative difference (MRD) uses all the observation made during the observation period. The second group of methods uses a pair of observation times and calculate the similarity in the spatial patterns between those times. Spearman rank correlation is one of the most commonly used within the second group of methods (Vachaud et al., 1985). More details of some other methods in identifying RMLs are given by Vanderlinden et al. (2012). In this study, the RMLs were identified following the above-mentioned methods as introduced by Vachaud et al. (1985). Briefly, if the SWS at ith location and tth date is θi,t and the spatial mean water content is θt at the same time, the difference (Δ) between individual determinations is Δi;t ¼ θi;t −θt

ð1Þ

The spatial mean water content was calculated from θt ¼

 X 1 n θ n i¼1 i;t

ð2Þ

where n is the number of measurement locations (for this study n = 128). Then the relative difference, δi,t, was calculated as δi;t ¼

Δi;t

ð3Þ

θt

The relative difference provides an estimation of difference that is unbiased to the magnitude of the mean value. The temporal MRD was calculated as δi ¼

m 1X δ m t¼1 i;t

ð4Þ

where m is the number of sampling days (for this study, m = 25). SDRD, the standard deviation σ(δi) of relative differences (δi,t) from the mean relative differences (δi ), was calculated as

σ ðδi Þ ¼

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 uX u δi;t −δi t m−1

ð5Þ

Therefore, each location will have one MRD value and its associated SDRD. The MRD values are then sorted form the smallest to the largest to identify the RML, which has the MRD closest to zero and smaller SDRD. The MRD values can be positive or negative. In this study, I have adopted the absolute value of the difference between individual determinations. Therefore, the Eq. (1) can be written as   Δi;t ¼ θi;t −θt 

Fig. 3. Site-specific Neutron probe calibration with the regression equation used for soil water calculation.

ð6Þ

The absolute value of difference overcomes the issues of positive and negative values of MRD, therefore making it easy to identify the smallest MRD. The SDRD indicates the uncertainty associated with the prediction of soil water content using the RML. For example, a large SDRD is an indication that the SWS at that location is not linearly related to the field average and therefore a poor predictor (Cosh et al., 2008). The

location with the smallest SDRD value can be used to predict the fieldaveraged soil water content. However, the location with the smallest MRD may not have the minimum SDRD. Sometimes the relatively smaller one is accepted for the site with the MRD close to zero (Hu et al., 2009), but it can introduce bias in identifying the RML. There is no definitive way to identify the best point. Therefore, to overcome the uncertainty, multiple points can be selected to represent the fieldaveraged SWS (Grayson and Western, 1998). In this study, I have also considered two more locations with the 2nd and 3rd smallest MRD in addition to the location with the smallest MRD to calculate the fieldaveraged SWS. The average SWS at these three points were compared with the field-averaged SWS. While the RMLs with the smallest MRD represent the field-averaged soil water content, the locations with the largest MRD represent the least temporally stable locations (the field extremes). In this study, the field extreme point(s) was also identified and used to predict the field-averaged SWS. The average SWS was calculated at multiple RMLs and at the field extremes, and these values were compared to the field-averaged SWS. The sum of squared error (SSE) between the field-averaged SWS and the SWS at RLM(s) or field extreme(s) was used as a relative measure of goodness of fit. The SSE was calculated as m  2 X SSE ¼ Y−Y^

ð7Þ

k¼1

where m is the number of measurements, Y^ is the field-averaged SWS, and Y is the SWS at the RML(s) or field extreme(s). The smaller the value of SSE, better the representation of field-averaged SWS. Similarities in the spatial pattern of the SWS at different measurement times were examined using Spearman rank correlation coefficients (rs). A value of rs = 1 identical rank for a location at two measurement times, i.e., TS between two measurements. The closer rs is to 1, the greater the similarity between locations in the changes in SWS through time. All the calculations were completed in Microsoft Office Excel (Microsoft Corporation Inc.), and the graphs were prepared in MATLAB (Mathworks Inc.) and SigmaPlot (Systat Inc.). 3. Results and discussion 3.1. SWS spatial pattern The spatial distribution of SWS with field average in the surface layer for selected dates is presented in Fig. 4. Relative elevation is in the top panel. The average SWS in the surface layer over the entire measurement period was 5.51 cm. SWS values at 100 to 140 m and 225 to 250 m from the origin of the transect, which were located within depressions (Pond 1 and Pond 2, respectively in Fig. 1), were very high compared to other locations during the spring and the early part of the summer seasons (Fig. 4). However, SWS values at around 450 to 550 m from the origin of the transect were not very high (Fig. 4). Although these locations look like a depression from the cross-sectional view of the transect (Fig. 4), they are actually at the mid-slope position of a large depression (Fig. 1). Generally, the strong wind in the prairie region redistributes snow in the landscape and within the depressions (Fang and Pomeroy, 2009). The depressions store more snow compared to surrounding knolls. During early spring, the frozen soil restricts the infiltration capacity (Gray et al., 1985), resulting in the redistribution of snowmelt water in the landscape. Depressions receive water from surrounding slopes and maintain a height of standing water. Conversely, knolls store less water due to drift accumulation in depressions and snowmelt runoff. A wide variation in SWS along the transect is the result. The spatial distribution of SWS in the surface layer almost mirrored (high topography low water storage and low topography high water storage) the spatial

RE (m)

A. Biswas / Catena 116 (2014) 38–50

Elevation

5.0 2.5 15 10

17 July 2007 5.65 cm

5

Soil Water Storage (cm)

42

12 Oct. 2007

12 8 4 12 8 4 12 8 4 12 8 4 12 8 4 12 8 4 12 8 4 12 8 4

5.04 cm

2 May 2008 6.28 cm

16 July 2008 4.03 cm

22 Oct. 2008 4.96 cm

7 May 2009 5.97 cm

21 July 2009 4.56 cm

27 Oct. 2009 5.30 cm

19 May 2010 6.04 cm

0

100

200

300

400

500

Location (m) Fig. 4. Spatial distribution of SWS for selected measurements at the surface layer (0–0.20 m) along the transect over 5 years with the relative elevation at top. X-axis indicates distance along the transect (m); Y-axis indicates SWS (cm); RE indicates relative elevation (m) r; dotted line indicates average SWS of each measurement with the average value in italics.

distribution of elevation during the spring and early part of the summer season (Fig. 4) (Biswas and Si, 2011a, 2011b). A similar spatial pattern was observed in other measurements completed within the spring or early summer. Prior to vegetation growth in early spring, the pathways of water loss from the landscape were evaporation from the soil surface and standing water, and to a lesser extent, deep drainage to groundwater. The loss of water through deep drainage in the hummocky landscape of the same study area is as low as about 2 to 40 mm per year (Hayashi et al., 1998). The main process of water loss is evapotranspiration by growing vegetation. The large amount of water stored in the depressions allowed the lush growth of grasses (particularly the aquatic species growing over 2 m in height in later summer), whereas the small amount of water stored on knolls restricted plant growth and evapotranspiration. The high evapotranspiration in depressions and low evapotranspiration on knolls during the summer and fall reduced the difference between the water content on the knolls and in the depressions (Biswas and Si, 2011a). This variable water uptake weakened the topography-induced spatial pattern during the later summer and fall or in dry periods imposing a seasonal dependence in the spatial pattern of SWS in the surface layer (Fig. 4) (Biswas and Si, 2011b). Tallon and Si (2004) also reported a similar spatial pattern in soil water with elevation during a very dry year from a glaciated agricultural landscape from Alvena, Saskatchewan, Canada. Although it changed over time within a year, the spatial pattern of a particular season was similar from 1 year to the next. For example, the spatial distribution during spring 2008 (2 May) was very similar to the spatial distribution during spring 2009 (7 May) and spring 2010

A. Biswas / Catena 116 (2014) 38–50

RE (m)

(19 May) (Fig. 4). A similar trend in the spatial pattern was observed in other seasons of measurement (summer and fall). This may be because the hydrology of the study site is dominated by the distribution of snow (Fang and Pomeroy, 2009; Gray et al., 1985) as the snowmelt water determines the spring spatial pattern in SWS. Therefore, in every spring, a similar pattern in SWS is demonstrated and later modified by the vegetation as the growing season progresses. A similar trend in the spatial pattern of SWS was reported by Starr (2005) from a gently rolling commercial agricultural farm near Houlton, Maine, USA. The average water storage over 5 years in the root zone and the total active soil profile were 164.4 mm and 333.0 mm, respectively. Spatial patterns of water storage in the root zone (Fig. 5) and the total active soil profile were similar to the spatial pattern at the surface layer (Fig. 4) and thus mirrored the spatial pattern in elevation (Biswas and Si, 2011a, 2011b). Unlike the surface layer, the pattern in the root zone and total active soil profile persisted even in later summer and fall. This may be because plants generally take up more than 70% of the water they need from the top 50% of the root zone (Feddes et al., 1978). Besides the intense root activity (Cassel et al., 2000), the more disturbed soil structure (Pachepsky et al., 2005) and the exposure to solar radiation, wind, and rainfall (Hu et al., 2010a) make the surface layer more dynamic. The dynamic behaviour of the surface layer exhausted the readily available water and equalized stored water over the transect during the latter part of the year. Therefore, the topographically induced spatial pattern was weakened, while there were still patterns in the root zone and total active soil profile due to topography.

5.0

Elevation

2.5 50

17 July 2007

34.84 cm

25

12 Oct. 2007

50 31.69 cm

25

Soil Water Storage (cm)

50

2 May 2008

35.20 cm

25

16 July 2008

50

30.20 cm

25

22 Oct. 2008

50 28.83 cm

25

7 May 2009

50

33.30 cm

25

21 July 2009

50

30.21 cm

25

27 Oct. 2009

50 30.12 cm

25 50

19 May 2010

35.65 cm

25 0

100

200

300

400

500

Location (m) Fig. 5. Spatial distribution of SWS for selected measurements at the active root zone (0–1.20 m) along the transect over 5 years with the relative elevation at top. X-axis indicates distance along the transect (m); Y-axis indicates SWS (cm); RE indicates relative elevation (m) r; dotted line indicates average SWS of each measurement with the average value in italics.

43

3.2. Time stability of the spatial pattern Spearman's rank correlation analysis is a statistical tool that measures the degree of concordance between two rankings. The rank correlation analysis was used to examine the similarity of the ranks of individual locations and thus the overall spatial pattern. All of the rank correlation coefficients were significant at p b 0.0001. A high rank correlation coefficient between the SWS series at the surface layer (Fig. 6) indicated that a large number of measurement locations maintained their spatial ranks over time. In other words, the locations with high water storage relative to the field average at one location in any time maintained the high water storage relative to the field average in other times and vice versa. The greater the number of measurement locations that maintain the rank, the more similar the spatial pattern. The high rank correlation coefficients between the measurement series during 2007 clearly indicated strong similarity in the spatial pattern of SWS (Fig. 6). The TS of SWS in this landscape, as indicated by rank correlations, varied depending on the time lag between the measurements. For example, the correlation coefficient was 0.96 between 31 May 2008 and 21 June 2008 (measurements from early part of 2008) and was 0.97 between 23 August 2008 and 17 September 2008 (measurements from latter part of 2008) (Fig. 6). Therefore, there was strong intra-season TS. However, the change in the intensity of processes controlling SWS from one season to another might yield a weaker inter-season TS compared to the intra-season TS. The correlation coefficients between a measurement series of early 2008 and of late 2008 were relatively low (Fig. 6). For example, the correlation coefficient was 0.63 between 2 May 2008 and 22 October 2008 (Fig. 6). This may be because similar processes are operating within a season and have similar effects on SWS (Martinez-Fernandez and Ceballos, 2003). While evaporation is the dominant process of water loss during the spring and early summer, the strong transpirative demand of growing vegetation adds up during the later summer and fall and determines the SWS in the semiarid climate (Parsons et al., 2004). The TS depended on the time difference between two measurement series—the highest rank correlation coefficient was observed between the measurements taken close in time and the coefficient gradually decreased as time difference increased. For example, the correlation coefficient between 2 May 2008 and 31 May 2008 was 0.91, but it gradually decreased to 0.63 between 2 May 2008 and 22 October 2008 (Fig. 6). The changing evapotranspiration demand of developing vegetation may gradually change SWS through time. Similar trends in the rank correlation coefficients and thus the inter-season and intraseason TS were observed in other years of measurement. A strong correlation was also observed between the measurements from a particular season in 1 year and the measurements from the same season of other years (inter-annual TS). For example, the correlation coefficient was 0.91 between 2 May 2008 and 7 May 2009 (Fig. 6). Similar high correlation was also observed between the measurements from spring, summer, or fall of different years (Fig. 6). This is because every year, the redistribution of snow in winter and snowmelt water during spring creates a spatial pattern in SWS. Similar time-stable spatial pattern between years was reported by Starr (2005). The spatial pattern persisted from 1 year to the next year despite the different crops grown in different years (Starr, 2005). However, in our study, the spatial pattern gradually changes with time as vegetation develops (Fig. 6). In later summer and fall, water stored in the soil is used by plants. The more water that is stored in the soil, the better the plant growth and, in turn, the more soil water extraction. Therefore, the spatial patterns established in the spring due to topography are weakened or diminished by vegetation in the surface layer. The similarity of the spatial pattern over time or TS within a season, between seasons, and between years provides insight about various important ongoing hydrological and ecological processes in the landscape. The information on the similarity of the spatial pattern can help catchment managers to identify the

44

A. Biswas / Catena 116 (2014) 38–50

Fig. 6. Spearman's rank correlation coefficients between the SWS measurements over 5 years in the surface soil layer (0–0.20 m). The color scale indicates the strength of correlation coefficients.

management units in a more cost effective manner for precision management of crop biomass production and the environment. For example, the temporally stable spatial pattern of soil water can also be used for precision irrigation management (Starr, 2005). The spatial patterns established in spring were weakened from wet state to dry state. Nevertheless, there were strong TS of the spatial pattern that existed over the whole period of measurement. However, Grayson and Western (1998) did not find any TS spatial pattern of soil water within the Tarrawarra catchment near Melbourne, Australia.

They argued that the topographically routed lateral redistribution of soil water created the spatial variability of hydrological processes and thus the time instability of the spatial pattern of soil water. Similarly, Gomez-Plaza et al. (2000) did not find time-stable spatial pattern along a transect in the Murcia region of southeast Spain. The vegetation and hillslope aspect of these transects modified the water loss behaviour through evapotranspiration and drying, thus yielding time unstable spatial pattern. In our study, a large amount of snowmelt water redistributes in the landscape along the topographic gradient. The

Fig. 7. Spearman's rank correlation coefficients between the SWS measurements over 5 years in the root zone (0–0.60 m). The color scale indicates the strength of correlation coefficients.

A. Biswas / Catena 116 (2014) 38–50

depressions store a large amount of water and create the spatial pattern. The stored water is more than sufficient for plants to grow during the year. Therefore, the spatial pattern persisted even during later fall and resulted in high Spearman rank correlation coefficients between measurements indicating strong TS. A similar trend in the TS of the spatial series was also observed in the root zone (Fig. 7) and total active soil profile (Fig. 8). The value of rank correlation coefficients between any two soil water measurement series was very high (significant at p b 0.0001), which suggested strong TS of the spatial pattern of the SWS in the root zone and total active soil profile. A similar trend was also observed in intra-season, inter-season, and inter-annual TS. However, the degree of similarity changed with depth as indicated by the change in the magnitude of the correlation coefficients (Figs. 7 and 8), which was generally in the order of the total active soil profile N the root zone N the surface layer. This suggested the increase in similarity of the spatial pattern with the increase in thickness of soil layer. For example, the correlation coefficients between 2 May 2008 and 22 October 2008 in surface layer, root zone, and total active soil profile were 0.63 (Fig. 6), 0.79 (Fig. 7), and 0.85 (Fig. 8), respectively. The value of rank correlation coefficient was consistently higher between any two soil water series of the total active soil profile compared to the surface layer. This may be attributed to lower root water uptake (Cassel et al., 2000) or less meteorological effect (Hu et al., 2010a) in deeper soil layers. The response of the thick soil layer is also buffered by the large volume of soil water in the profile as well as the difference in hydrological processes from the surface. In addition to this, measurement error associated with the surface soil layer was higher compared to deep layers. The sample volume for the TDR measurement is smaller compared to the neutron probe measurement. Therefore, the inherent variability of soil water may contribute more error to the surface point measurement compared to larger volume neutron probe measurement. Moreover, some effects of measurement techniques (TDR for the surface layer, TDR and neutron probe for the root zone, and total active soil profile) might have contributed to the variations in TS at different layer. The changing pattern in the water storage with depth was also reported by Martinez-Fernandez and Ceballos (2003), Guber et al. (2008), Tallon and Si (2004), Pachepsky et al. (2005), and Hu et al. (2010a). However, Grayson and Western (1998) did not observe any significant

45

difference in the spatial pattern of SWS between 0–30 cm (TDR data) and 0–60 cm depth (neutron probe data) at Tarrawarra catchment of Australia and concluded that there were no significant effects of measurement depth on soil water spatial pattern. 3.3. Identification of RML The stability of the spatial pattern of SWS aids us in identifying the RML. Fig. 9 shows the rank of mean relative difference (MRD) and the standard deviation associated with relative difference (SDRD) for (a) the surface layer, (b) the root zone, and (c) the total active soil profile. The location with the MRD closest to zero and small SDRD indicated that the value of SWS at this location was consistently close to the field average of SWS over time and was considered as RML. The 87th point on the transect had the smallest MRD and the SDRD (Fig. 10a) and was identified as the RML for the surface layer. Similarly, the RMLs for the root zone and total active soil profile were identified at the 70th (Fig. 10b) and 69th point (Fig. 10c) on the transect. The RML changed with the depth of measurement inferring monitoring locations for the field should be specific to a given depth range. Tallon and Si (2004) reported depth-specific RMLs, which were not adjacent in terms of spatial location. Martinez-Fernandez and Ceballos (2003), Guber et al. (2008), and Pachepsky et al. (2005) also reported different RMLs at different depths. However, (Cassel et al., 2000) found the same RMLs for different depths. In this study, the surface layer RML was different from the RMLs for the root zone and total active soil profile, which were located next to each other. The dynamic behaviour of the surface layer made the surface RML different from the RMLs of deep soil layers. This may be because the factors controlling the spatial pattern of SWS at the surface do not have as pronounced effect as at greater depths (Cassel et al., 2000; Hu et al., 2010a; Pachepsky et al., 2005). The RML identified for the surface layer was located at near neutral condition to the field slope variations. Generally, the surface layers are exposed to the environmental forcing, and the soil water is prone to change with time and environmental conditions. However, the nearneutral slope condition created the site most stable with time. The near-neutral position of RML for the surface soil was also reported by Grayson and Western (1998), Jacobs et al. (2004), and Vivoni et al.

Fig. 8. Spearman's rank correlation coefficients between the SWS measurements over 5 years in the total active soil profile (0–0.120 m). The color scale indicates the strength of correlation coefficients.

46

A. Biswas / Catena 116 (2014) 38–50

150

a) Surface layer (0-20 cm) 100

Mean Relative Difference (%)

50

0

b) Root zone (0-60 cm) 50

0

c) Total active soil profile (0-120 cm)

50

0

0

30

60

90

120

Rank Fig. 9. Ranked relative deviation (%) from the mean SWS for (a) the surface layer (0–0.20 m), (b) the root zone (0–0.60 m), and (c) the total active soil profile (0–1.20 m).

a) Surface Layer (0-20 cm)

160 120 80

87 34 72 109 70 47 22 35 88 110 38 46 69 60 45 106 105 112 36 21 37 71 84 103 73 108 61 7 93 49 6

0

75

b) Root Zone (0-60 cm)

50 25 0

45

70 38 36 105 109 110 7 61 112 22 107 64 69 35 47 66 46 37 104 45 21 44 6 43 111 62 8 3 34 114 108

Mean Relative Difference (%)

40

c) Total active soil profile (0-120 cm)

30 15 69 105 108 110 43 38 65 104 22 112 109 62 64 107 37 42 66 36 86 7 46 34 68 47 63 61 21 113 45

0

Sample Locations Fig. 10. Zoomed ranked relative deviation (%) from the mean SWS for (a) the surface layer (0–0.20 m), (b) the root zone (0–0.60 m), and (c) the total active soil profile (0–1.20 m). The error bars indicate the standard deviation of the mean relative difference.

A. Biswas / Catena 116 (2014) 38–50

47

years. Additionally, a positive correlation (correlation coefficient, r = 0.77, 0.75, and 0.67 for the surface layer, root zone, and total active soil profile, respectively) was observed between SWS and standard deviation of measurement. The wet extremes with high SWS had high variability in MRD compared to the dry extremes. In this study, without identifying wet and dry extremes separately, I adopted absolute values of MRD (Eq. (6)) to identify the extremes. The absolute value made me possible to identify locations with the smallest (RML) and largest (extremes) MRD and thus the identification of RML. The performance of the RML in representing field-averaged SWS was evaluated by comparing the SWS at the RML(s) with the average SWS at each time of measurement (Fig. 11). For the surface soil layer, the SWS at the 87th point was in good agreement with the fieldaveraged soil SWS (Fig. 11a). Similar agreement was also observed in the SWS of the 70th and 69th point at the root zone and total active soil profile, respectively. The relationship between the field-averaged SWS and the SWS at RMLs were examined using linear regression. The non-unity slope of the regression line with a non-zero intercept indicated the bias that is associated with the prediction of average SWS from the BM points. The coefficient of determination (r2) for the relationship was 0.90, 0.95, and 0.97 between the field-averaged SWS at the surface layer, root zone, and total active soil profile and SWS at the corresponding

(2008). For deeper soil layers (root zone and total active soil profile), the RMLs are situated at mid-slope of an average length north facing slope, which experience more average environmental conditions. The RML identified based on MRD ignores the fact that MRD are actually statistics with inherent errors characterized by SDRD. Guber et al. (2008) recommended RML as the location with the lowest MRD and SDRD. However, the lowest MRD may not have the smallest SDRD. Therefore, to reduce the uncertainty, I have selected two additional points at each soil layer to represent the field-averaged SWS. The 34th and 72nd points were selected for the surface layer. Similarly, the 38th and 36th and the 105th and 108th points were selected for the root zone and the total active soil profile, respectively. These points had the MRD very close to zero but more than the RMLs. Similar to the RMLs, the field extremes were also identified from the rank of the MRD for sampling points. The 6th, 108th, and 45th points were identified as the field extreme points for the surface layer, root zone, and total active soil profile. There are two types of extremes: wet extremes and dry extremes. Wet extremes are located in depressions and are generally flooded during spring and dried up during late summer or fall. The alternate drying and wetting contribute to high variations in SWS. Additionally, vigorous growth of aquatic plants within depressions also contributed to high variations, while the dry extremes are generally located on top of the knolls and had very small variations in SWS over the

a) Surface layer (0-20 cm) 7 6 5

0-20 cm Average 87th point 70th point 69th point

4

Soil Water Storage (cm)

b) Root zone (0-60 cm) 22 20 18 16 0-60 cm Average 87th point 70th point 69th point

14 12

c) Total active soil profile (0-120 cm) 0-120 cm Average 87th point 70th point 69th point

42 39 36 33 30

Sept 29, 2011

Jun 6, 2011

Jun 29, 2011

May 13, 2011

Jun 14, 2010

Sept 28, 2010

Apr 6, 2010

May 19, 2010

Oct 27, 2009

Jul 21, 2009

Aug 27, 2009

May 7, 2009

May 27, 2009

Oct 22 2008

Apr 21, 2009

Aug 23 2008

Sept 17 2008

Jun 21 2008

July 16 2008

May 2 2008

May 31 2008

Oct 12 2007

Aug 7 2007

Sept 1 2007

Jul 17 2007

27

Measurement date Fig. 11. SWS at the benchmark point and the field-averaged SWS of (a) the surface layer (0–0.20 m), (b) the root zone (0–0.60 m), and (c) the total active soil profile (0–1.20 m).

48

A. Biswas / Catena 116 (2014) 38–50

Table 1 Regression equation, r2 value, and the sum of squared errors between the field-averaged SWS and the SWS at single and multiple BM points (average) and the single and multiple points (average) of field extremes at the surface layer (0–20 cm), root zone (0–60 cm), and total active soil profile (0–120 cm). Soil layer

Measurement

Single point

SSEa

Regression (θv)b

r2

Multiple points (average of 3 points)

SSE

Regression (θv)

r2

0–20 cm

BM points

87th 70th 69th 6th 108th 45th 87th 70th 69th 6th 108th 45th 87th 70th 69th 6th 108th 45th

2.84 4.58 7.40 14.46 8.48 7.98 51.03 12.42 32.03 63.94 54.48 39.75 128.69 201.13 17.48 215.80 35.68 140.03

0.87θb 0.70θb 0.68θb 1.10θb 0.87θb 0.74θb 0.85θb 0.80θb 0.78θb 0.92θb 0.98θb 0.61θb 1.06θb 0.71θb 0.82θb 0.94θb 0.99θb 0.69θb

0.90 0.89 0.90 0.69 0.86 0.83 0.92 0.95 0.94 0.65 0.91 0.85 0.84 0.95 0.97 0.57 0.90 0.84

87th, 34th, 72nd 70th, 38th, 36th 69th, 105th, 108th 6th, 49th, 93rd 108th, 114th, 34th 45th, 113th, 21st 87th, 34th, 72nd 70th, 38th, 36th 69th, 105th, 108th 6th, 49th, 93rd 108th, 114th, 34th 45th, 113th, 21st 87th, 34th, 72nd 70th, 38th, 36th 69th, 105th, 108th 6th, 49th, 93rd 108th, 114th, 34th 45th, 113th, 21st

2.28 5.44 6.06 1.67 7.16 7.98 52.52 10.19 28.24 7.16 13.56 41.31 153.98 44.97 14.80 15.52 27.14 132.22

0.82θb 0.77θb 0.88θb 0.87θb 0.86θb 0.71θb 0.77θb 0.76θb 0.90θb 0.88θb 0.86θb 0.64θb 0.81θb 0.71θb 0.90θb 0.90θb 0.86θb 0.65θb

0.93 0.92 0.91 0.95 0.91 0.88 0.95 0.97 0.95 0.95 0.95 0.92 0.90 0.98 0.95 0.95 0.94 0.91

Field extreme

0–60 cm

BM points

Field extreme

0–120 cm

BM points

Field extreme

+ + + − + + + + + + − + − + + + − +

0.76 1.52 1.42 0.45 0.26 1.06 3.70 3.64 2.68 1.63 0.96 5.85 0.38 12.37 6.39 1.36 0.04 8.54

+ + + + + + + + + + + + + + + + + +

1.05 0.89 0.27 0.66 0.35 1.17 5.01 3.92 0.75 2.06 1.79 5.06 8.37 10.41 2.31 3.34 4.18 9.84

a

Sum of squared error; bpredictive water storage.

a) Surface layer (0-20 cm) 7 6 5 0-20 cm field Average Avg. of 87, 34, 72nd points Avg. of 70, 38, 36th points Avg. of 69, 105, 108th points

4 3

Soil Water Storage (cm)

b) Root zone (0-60 cm) 20 18 16 14

0-60 cm field Average Avg. of 87, 34, 72nd points Avg. of 70, 38, 36th points Avg. of 69, 105, 108th points

12 10

c) Total active soil profile (0-120 cm) 0-120 cm field Average Avg. of 87, 34, 72nd points Avg. of 70, 38, 36th points Avg. of 69, 105, 108th points

39 36 33 30

Sept 29, 2011

Jun 6, 2011

Jun 29, 2011

Sept 28, 2010

May 13, 2011

Jun 14, 2010

Apr 6, 2010

May 19, 2010

Oct 27, 2009

Jul 21, 2009

Aug 27, 2009

May 7, 2009

May 27, 2009

Oct 22 2008

Apr 21, 2009

Aug 23 2008

Sept 17 2008

Jun 21 2008

July 16 2008

May 2 2008

May 31 2008

Oct 12 2007

Aug 7 2007

Sept 1 2007

Jul 17 2007

27

Measurement date Fig. 12. Average SWS of three benchmark points and the field-averaged SWS of (a) the surface layer (0–0.20 m), (b) the root zone (0–0.60 m), and (c) the total active soil profile (0–1.20 m).

A. Biswas / Catena 116 (2014) 38–50

layer RMLs, respectively (Table 1). This means, that there would be around 10%, 5%, and 3% error, respectively, for the surface layer, root zone, and total active soil profile if we predict the average SWS from the measurement of RMLs. Not surprisingly, the dynamic behaviour of the surface layer increased error in prediction. The result was comparable or even superior to the results available in the literature. For example, Martinez-Fernandez and Ceballos (2005) reported an r2 value between the SWS at RML and the field-averaged SWS of 0.84 for REMEDHUS site and 0.92 for Rinconada site. The small SSE between the field-averaged SWS and the SWS at RMLs (Table 1) also indicated a superior performance of the RMLs. The performance of single BM point was evaluated against the average of multiple (3 for this study) BM points in predicting field-averaged SWS. The field-averaged SWS was in good agreement with the average SWS of multiple BM points (Fig. 12). Moreover, the r2 and the SSE values suggest greater accord between the field-averaged SWS and the average SWS of multiple RMLs as compared with the use of a single RML for a particular depth range. Choosing multiple RMLs yields a better result than use of a single RML in predicting the field-averaged SWS (Grayson and Western, 1998). Now I consider selection of only one RML for all three depths. If I choose the 87th point, the surface layer RML as RML for the deep layers, the prediction will include an additional error. The r2 value for the root zone and total active soil profile was 0.90 and 0.84 at the 87th point, which was either similar or smaller than that of surface layer. However, a high SSE was observed between the SWS at these points and the fieldaveraged SWS for the surface layer. Similarly, high SSE values between the SWS at the 70th point and the SWS at the surface layer or the total active soil profile and between the SWS at the 69th point and the SWS at the surface layer and the root zone indicated an additional error in predicting SWS. A similar trend was also observed for multiple RMLs. Therefore, the choice of RML(s) is depth specific. The performance of the field extreme point(s) were also evaluated using the regression relationship, the r2, and the SSE value. Non-zero intercept for the regression line indicated a bias in the prediction of SWS using extreme point(s). Although some high values of r2 were observed in certain situations, large SSE values indicated additional error in the prediction of SWS. Therefore, field extremes are the poor predictor of SWS (Cosh et al., 2008). This study was different from Tallon and Si (2004), which was based on soil water measurements taken from an undulating landscape during much drier years than the long-term average. In dry years, vegetation utilized most water that was available to plants, leaving no difference in depressions or on knolls. Therefore, topographically introduced patterns did not exist in the Tallon and Si (2004) study. However, the present study was conducted over years during which the total precipitation was near or higher than the long-term average. In these years, although the dynamic behaviour of the soil processes weakened the pattern in surface soil, the topographically induced spatial patterns still existed in deep soils. Thus, the topographically induced spatial pattern in SWS existed over the seasons and years. One of the important applications of the TS concept has been the identification of time-stable locations, which can considerably reduce the number of sampling locations in obtaining the mean soil water content for an area of interest (Vachaud et al., 1985) and holds considerable promise for minimizing costs. Monitoring soil water at time-stable locations or RMLs is useful in deciding various agronomic management practices such as irrigation scheduling, fertility monitoring and fertilizer recommendation, modeling soil water balance in a field, and understanding hydrological processes. Information on time-stable spatial patterns can be used in remote sensing of soil water in upscaling the water content from several or even single point measurement to the average soil water content across a footprint area (Cosh et al., 2004; Jacobs et al., 2004). It can also be used in establishing field or catchmentwide antecedent soil water condition for runoff simulations, relating spatiotemporal variation in soil water to the onset of subsurface flow

49

and upscaling of soil moisture information in irrigated and dry-land crops (Penna et al., 2009; Rolston et al., 1991). The use of temporally stable patterns to downscale remotely sensed data of soil water has also been suggested. Pachepsky et al. (2005) and (Guber et al., 2008) used TS of the spatial pattern to estimate the soil water contents at rarely sampled locations or at missing points. However, the BM point was identified based on 5 years of soil water measurement. Since the identification procedure requires a large amount of work, the next step for this research is to find the landscape characteristics for the RMLs. The identification of the minimum length of data required to identify the RML will also be an opportunity of future research. 4. Conclusion This analysis succeeded in identifying time-stable representative monitoring locations for field-averaged SWS in a hummocky landscape. The spatial pattern of SWS was stable over time, yet the degree of the stability of the spatial pattern changed between different times of measurement. The spatial pattern of intra-season SWS was more stable than the inter-season patterns. Furthermore, the inter-annual TS of SWS was weaker than the intra-season but stronger than the inter-season ones. The TS of the spatial pattern aids us in identifying RML(s) with SWS representing the field average with certain level of bias. The average of multiple RMLs predicted SWS better than that of single RML. The RMLs varied with the depth of measurement. The prediction of fieldaveraged SWS using RMLs was better compared to the prediction using field extremes. The RMLs can greatly improve the sampling efficiency by reducing the number of samples or observations for field-averaged SWS monitoring over conventional methods. Therefore, representative monitoring locations can provide useful information for environmental monitoring. Acknowledgements The project was funded by the National Sciences and Engineering Council (NSERC) of Canada. The help and guidance from Prof. Bing Si and Dr. Hamish Cresswell is highly appreciated. Assistance in the field from the summer and graduate students of the Department of Soil Science at the University of Saskatchewan is greatly appreciated. References Bell, K.R., Blanchard, B.J., Schmugge, T.J., Witczak, M.W., 1980. Analysis of surface moisture variations within large-field sites. Water Resour. Res. 16, 796–810. Biswas, A., Chau, H.W., Bedard-Haughn, A.K., Si, B.C., 2012. Factors controlling soil water storage in the hummocky landscape of the Prairie Pothole Region of North America. Can. J. Soil Sci. 92, 649–663. Biswas, A., Si, B.C., 2011a. Revealing the controls of soil water storage at different scales in a hummocky landscape. Soil Sci. Soc. Am. J. 75, 1295–1306. Biswas, A., Si, B.C., 2011b. Scales and locations of time stability of soil water storage in a hummocky landscape. J. Hydrol. 408, 100–112. Brocca, L., Melone, F., Moramarco, T., Morbidelli, R., 2009. Soil moisture temporal stability over experimental areas in Central Italy. Geoderma 148, 364–374. Brocca, L., Melone, F., Moramarco, T., Morbidelli, R., 2010. Spatial-temporal variability of soil moisture and its estimation across scales. Water Resour. Res. 46. Cassel, D.K., Wendroth, O., Nielsen, D.R., 2000. Assessing spatial variability in an agricultural experiment station field: opportunities arising from spatial dependence. Agron. J. 92, 706–714. Cosh, M.H., Jackson, T.J., Bindlish, R., Prueger, J.H., 2004. Watershed scale temporal and spatial stability of soil moisture and its role in validating satellite estimates. Remote Sens. Environ. 92, 427–435. Cosh, M.H., Jackson, T.J., Moran, S., Bindlish, R., 2008. Temporal persistence and stability of surface soil moisture in a semi-arid watershed. Remote Sens. Environ. 112, 304–313. Environment Canada, 2011b. Canada's Top Ten Weather Stories for 2010. (Available at http://www.ec.gc.ca/meteo-weather/default.asp?lang=En&n=7E58ECA3-1 (Verified on 5 July 2013)). Famiglietti, J.S., Rudnicki, J.W., Rodell, M., 1998. Variability in surface moisture content along a hillslope transect: Rattlesnake Hill, Texas. J. Hydrol. 210, 259–281. Famiglietti, J.S., Wood, E.F., 1994. Multiscale modeling of spatially-variable water and energy-balance processes. Water Resour. Res. 30, 3061–3078. Fang, X., Pomeroy, J.W., 2009. Modelling blowing snow redistribution to prairie wetlands. Hydrol. Process. 23, 2557–2569.

50

A. Biswas / Catena 116 (2014) 38–50

Feddes, R.A., Kowalik, P.J., Zaradny, H., 1978. Simulation of field water use and crop yield. Halsted Press, John Wiley and Sons Inc, NY. Gomez-Plaza, A., Alvarez-Rogel, J., Albaladejo, J., Castillo, V.M., 2000. Spatial patterns and temporal stability of soil moisture across a range of scales in a semi-arid environment. Hydrol. Process. 14, 1261–1277. Gray, D.M., Landine, P.G., Granger, R.J., 1985. Simulating infiltration into frozen prairie soils in streamflow models. Can. J. Earth Sci. 22, 464–472. Grayson, R.B., Western, A.W., 1998. Towards areal estimation of soil water content from point measurements: time and space stability of mean response. J. Hydrol. 207, 68–82. Grayson, R.B., Western, A.W., Chiew, F.H.S., Bloschl, G., 1997. Preferred states in spatial soil moisture patterns: local and nonlocal controls. Water Resour. Res. 33, 2897–2908. Guber, A.K., Gish, T.J., Pachepsky, Y.A., van Genuchten, M.T., Daughtry, C.S.T., Nicholson, T.J., Cady, R.E., 2008. Temporal stability of estimated soil water flux patterns across agricultural fields. Int. Agrophysics 22, 209–214. Hayashi, M., van der Kamp, G., Rudolph, D.L., 1998. Water and solute transfer between a prairie wetland and adjacent uplands, 2. Chloride cycle. J. Hydrol. 207, 56–67. Hu, W., Shao, M., Han, F., Reichardt, K., Tan, J., 2010a. Watershed scale temporal stability of soil water content. Geoderma 158, 181–198. Hu, W., Shao, M., Reichardt, K., 2010b. Using a new criterion to identify sites for mean soil water storage evaluation. Soil Sci. Soc. Am. J. 74, 762–773. Hu, W., Shao, M., Wang, Q., Reichardt, K., 2009. Time stability of soil water storage measured by neutron probe and the effects of calibration procedures in a small watershed. Catena 79, 72–82. Jacobs, J.M., Mohanty, B.P., Hsu, E.C., Miller, D., 2004. SMEX02: field scale variability, time stability and similarity of soil moisture. Remote Sens. Environ. 92, 436–446. Kachanoski, R.G., Dejong, E., 1988. Scale dependence and the temporal persistence of spatial patterns of soil–water storage. Water Resour. Res. 24, 85–91. Martinez-Fernandez, J., Ceballos, A., 2003. Temporal stability of soil moisture in a largefield experiment in Spain. Soil Sci. Soc. Am. J. 67, 1647–1656. Martinez-Fernandez, J., Ceballos, A., 2005. Mean soil moisture estimation using temporal stability analysis. J. Hydrol. 312, 28–38. National Wetlands Working Group, 1997. The Canadian wetland classification system. In: Warner, B.G., Rubec, C.D.A. (Eds.), Wetlands Research Centre. University of Waterloo, ON. Pachepsky, Y.A., Guber, A.K., Jacques, D., 2005. Temporal persistence in vertical distributions of soil moisture contents. Soil Sci. Soc. Am. J. 69, 347–352.

Parsons, D.F., Hayashi, M., van der Kamp, G., 2004. Infiltration and solute transport under a seasonal wetland: bromide tracer experiments in Saskatoon, Canada. Hydrol. Process. 18, 2011–2027. Penna, D., Borga, M., Norbiato, D., Fontana, G.D., 2009. Hillslope scale soil moisture variability in a steep alpine terrain. J. Hydrol. 364, 311–327. Pomeroy, J.W., de Boer, D., Martz, L.W., 2007. Hydrology and water resources. In: Thraves, B. (Ed.), Saskatchewan: Geographic Perspectives. CRRC, Regina, SK, Canada. Rolston, D.E., Biggar, J.W., Nightingale, H.I., 1991. Temporal persistence of spatial soil– water patterns under trickle irrigation. Irrig. Sci. 12, 181–186. Saskatchewan Centre for Soil Research, 1989. Rural municipality of Grant. Number 372: Preliminary soil map and report. Publication# SK372. Saskatchewan Centre Soil Research, University of Saskatchewan, Saskatoon, SK. Seyfried, M., 1998. Spatial variability constraints to modeling soil water at different scales. Geoderma 85, 231–254. Starr, G.C., 2005. Assessing temporal stability and spatial variability of soil water patterns with implications for precision water management. Agric. Water Manag. 72, 223–243. Sun, M., 1986. Groundwater ills—many diagnoses, few remedies. Science 232, 1490–1492. Tallon, L.K., Si, B.C., 2004. Representative soil water benchmarking for environmental monitoring. J. Environ. Inform. 4, 28–36. Topp, G.C., Reynolds, W.D., 1998. Time domain reflectometry: a seminal technique for measuring mass and energy in soil. Soil Tillage Res. 47, 125–132. Vachaud, G., Desilans, A.P., Balabanis, P., Vauclin, M., 1985. Temporal stability of spatially measured soil–water probability density-function. Soil Sci. Soc. Am. J. 49, 822–828. Vanderlinden, K., Vereecken, H., Hardelauf, H., Herbst, M., Martínez, G., Cosh, M.H., Pachepsky, Y.A., 2012. Temporal stability of soil water contents: a review of data and analyses. Vadose Zone J. 11. http://dx.doi.org/10.2136/vzj2011.0178. Vivoni, E.R., Gebremichael, M., Watts, C.J., Bindlish, R., Jackson, T.J., 2008. Comparison of ground-based and remotely-sensed surface soil moisture estimates over complex terrain during SMEX04. Remote Sens. Environ. 112, 314–325. 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 terrain indices. Water Resour. Res. 35, 797–810. Whigham, D.F., Jordan, T.E., 2003. Isolated wetlands and water quality. Wetlands 23, 541–549. Zhao, Y., Peth, S., Wang, X.Y., Lin, H., Horn, R., 2010. Controls of surface soil moisture spatial patterns and their temporal stability in a semi-arid steppe. Hydrol. Process. 24, 2507–2519.