Contents lists available atScienceDirect
Journal of Hydrology: Regional Studies
journal homepage:www.elsevier.com/locate/ejrh
Hydrological modeling of the pipestone creek watershed using the
Soil Water Assessment Tool (SWAT): Assessing impacts of wetland
drainage on hydrology
Cesar Perez-Valdivia
a,⁎, Barbara Cade-Menun
b, Dena W. McMartin
aaEnvironmental Systems Engineering, University of Regina, 3737 Wascana Parkway, Regina, SK S4S 0A2, Canada bAgriculture and Agri-Food Canada, Swift Current Research and Development Centre, Swift Current, SK S9H 3X2, Canada
A R T I C L E I N F O
Keywords:
Soil Water Assessment Tool (SWAT) Wetland drainage
Peakflow Annual volume Prairie Pothole Region
A B S T R A C T
Study region: Prairie Pothole Region of North America.
Study focus:The Prairie Pothole Region of North America has experienced extensive wetland drainage, potentially impacting peakflows and annualflow volumes. Some of this drainage has occurred in closed basins, possibly impacting lake water levels of these systems. In this study we investigated the potential impact of wetland drainage on peakflows and annual volumes in a 2242 km2 watershed located in southeastern Saskatchewan (Canada) using the Soil Water Assessment Tool (SWAT) model.
New hydrological insights:The SWAT model, which had been calibrated and validated at daily and monthly time steps for the 1997–2009 period, was used to assess the impact of wetland drainage using three hypothetical scenarios that drained 15, 30, and 50% of the non-contributing drainage area. Results of these simulations suggested that drainage increased spring peakflows by about 50, 79 and 113%, respectively while annualflow volumes increased by about 43, 68, and 98% in each scenario. Years that were wetter than normal presented increased peakflows and annual flow volumes below the average of the simulated period. Alternatively, summer peak flows presented smaller increases in terms of percentages during the simulated period.
1. Introduction
There are numerous applications of the Soil Water Assessment Tool (SWAT) model around the world. However, implementations of SWAT in the cold regions of North America, where snow processes have an important role in the overall annual hydrological cycle, are limited. Wang and Melesse (2005) evaluated the SWAT model's snowmelt hydrology in a 4340 km2 watershed located in northwestern Minnesota simulating daily, monthly, and seasonal streamflows. The model performed well in simulating monthly, seasonal, and annual meanflows. The performance of the model simulation for dailyflows was not as good but still had a satisfactory performance. Another study byAhl et al. (2008)calibrated and validated the SWAT model at daily and monthly time steps for a 22.5 km2watershed in central Montana. Results of this simulation suggested that the SWAT model performed acceptably for spring and early summer snowmelt but performed poorly for late summer and winter baseflow. The model performed better at monthly than daily time steps, and snowmelt parameters such as snowmelt temperature and snowpack temperature factor were the most sensitive parameters.Zhang et al. (2008)used the SWAT model to simulate runofffrom an ungauged 114,345 km2mountain river basin using three different snowmelt algorithms, based on a temperature index, a temperature index plus elevation band, and on an
https://doi.org/10.1016/j.ejrh.2017.10.004
Received 29 June 2017; Received in revised form 29 September 2017; Accepted 20 October 2017
⁎Corresponding author.
E-mail address:[email protected](C. Perez-Valdivia).
energy budget. Of these three algorithms, the one based on the temperature index plus elevation had a performance comparable to the simulation carried out with SWAT using the energy budget model. Simulation with the temperature algorithm had the worst performance due to its simplicity. The runoffsimulation using the temperature index plus elevation algorithm was almost as good as using the energy balance algorithms.
In Canada, the SWAT model and its snow module were also applied to a northern watershed in Quebec (Troin and Caya, 2014). The model was calibrated adjusting 17 parameters for a period offive years and validated for another two periods of 17 and 15 years, and showed a satisfactory performance. The model simulated the timing and the magnitude of the spring runoffwell and the results were comparable to other models such as the Streamflow Synthesis and Reservoir Regulation (U.S. Army Corps of Engineers, 1991) that had been implemented in that watershed. Also in Canada,Rahbeh et al. (2011)applied the SWAT model to a small watershed (∼34 km2) in the Canadian Prairies. This watershed was characterized by extensive irrigation, an upstream inlet, low runoffand only three years of records. Their approach was to calibrate the model using the more traditional one-way calibration method, in which the record was split into two and was used to calibrate and validate the model, and the less common two-way calibration method. This consists of splitting the recorded time series into two periods, A and B; the model is then calibrated and validated using periods A and B but is also calibrated and validated using the opposite period, B and A respectively. The SWAT model was successfully calibrated for the different scenarios used but could not be validated for any of the scenarios.
The Canadian Prairies form part of the Pothole Region of North America (PRNA) and its geography is characterized by thousands of shallow wetlands. These wetlands have been drained since thefirst Europeans started to cultivate in the PRNA, to maximize the land and have an easier access to it (Johnson et al., 2008), resulting in considerable wetland loss. In the USA, wetland loss has been severe; it was estimated that by 1991 the number of wetlands had been reduced by between 50 and 70% (Dahl and Johnson, 1991). In the Canadian Prairies the wetland loss is also significant and it is estimated that around 70% of wetlands have been lost (Dahl and Johnson, 1991).
Wetland drainage affects some hydrological processes such as storage, evaporation, and infiltration, as well as the timing in which the drainage area initially contributing to the wetlands contributesflow to the main channel. The watersheds within the PRNA are dominated by the principle offill and spill, and water levels in the wetlandsfluctuate mostly because of the runoff. The verticalflux of ground water in or out of wetlands is considered insignificant in the overall balance (van der Kamp and Hayashi, 2009). In these areas, wetlands might or might not contribute to runoffduring a determined event, depending on the magnitude of the event, the water level prior to the runoffevent, and the antecedent soil moisture conditions. This makes the contributing drainage area to runoff variable from year to year, season to season, and even from event to event. This behavior was initially observed during the 1950s (Stichling and Blackwell, 1957), and was refined in the mid-1970s when the concepts of Effective Drainage Area (EDA) and Gross Drainage Area (GDA) were introduced in the Canadian Prairies (Godwin and Martin, 1975). The EDA is the area that is expected to contribute to runoffduring a normal year, which has a return period of 2 years. The GDA is the topographic limit of the watershed, which in practical terms is expected to contribute to runoffonly during an extreme low-frequency event (e.g. a 1:1000 years event; SaskWater, 1993). The area that is found outside of the EDA but within the GDA is defined as the non-contributing drainage area (NCDA) and is usually characterized by a large number of wetlands with no or low connectivity; these are not expected to contribute runoffto the main channel byfilling and spilling during a normal runoffyear. Therefore, the simulation of the hydrological processes of watersheds with variable contributing drainage areas is not trivial.
The variable contributing drainage area issue was addressed byWang et al. (2008), who introduced the concept of the hydrologic equivalent wetland (HEW) to simulate daily and monthlyflows in a 4506 km2watershed in northwestern Minnesota. One of the objectives of that study was to incorporate the thousands of wetlands in the region into SWAT. The HEW concept is based on the maximum volume that a wetland can store, its surface area and the fraction of the sub-basin area that drains into the wetland. The HEW model was evaluated for a period of seven years (1969–1975) at daily, monthly, and seasonal time steps, and the performance of the model varied through the different time steps. However, when comparing this approach to other methodologies used to model wetland-dominated watersheds, such as the synthetic wetland approach, the HEW was found to be a superior method for in-corporating wetlands into the SWAT model.
The relevance of wetlands for biodiversity and hydrology in terms of water quantity and quality has been widely studied for decades (e.g.,Duffy, 1998; Landers and Knuth, 1991). In the PRNA,Vining (2002)simulated streamflow and wetland storage for the period 1981–1998 in the Starkweather Coulee watershed located in North Dakota. This study used a Digital Elevation Model (DEM) to estimate the spill elevation of wetlands previously identified by the National Wetlands Inventory and Geographic Information Systems to determine surface areas and depths. That watershed had a GDA of 803 km2, from which 544 km2were reported to contribute to runoffregularly (EDA). The drainage area was divided into 50 hydrological response units (HRU) for which the water balance was calculated. In order to account for and simplify the number of wetlands in each HRU, one equivalent wetland that was the sum of all the wetlands in the HRU, was included in the model. Wetlands in this watershed presented an average depth of 0.67 m and volume was estimated as function of the surface area. A modified version of the U.S. Geological Survey's Precipitation Runoff Modeling System was used to simulated dailyflows and wetlands levels. The results of this study suggest that the model under-estimated peakflows during spring runofffor all years but one; the poor performance capturing peakflows was attributed to the lack of a subroutine in the model for frozen soils.
increasedflood peaks in a considerable number of studies. During wet periods, nearly half of the headwater wetlands had a faster response to rainfall, increasing the volumes independent of increases in peaks. This was attributed to the antecedent conditions of the wetlands, with higher water levels and higher soil moisture than normal.
The effect of prairie wetlands onflow reduction has been shown to be greater during high-frequency runoffevents than during low frequency events (Padmanabhan and Bengtson, 1999; Vining, 2002), and it has been suggested that there is a threshold at which the contributing area increases rapidly during low frequency events (Shaw et al., 2011). This threshold is defined when wetlands and ponds reach their spill elevation and start connecting with other wetlands, spilling downstream until a large portion of the GDA is contributing to the main channel. In general, when storage is available in the wetlands the hydrographs will present an initial lag until the storage isfilled and then will rise quickly (Shaw et al., 2011). In addition,Acreman and Holden (2013)systematically described the hydrological processes taking place in upland andfloodplain wetlands and their impact onfloods, suggesting that the impact of wetlands onfloods is related to their location within the watershed. Within this context, upland wetlands likely generate floods because they are mostly saturated and near their spill elevation. On the other hand,floodplain wetlands were suggested to reducefloods.
In the Canadian Prairies, two studies recently addressed the impact of wetlands on hydrology in terms of water quantity.Yang et al. (2010)simulated a wetland restoration scenario for the Broughton’s Creek watershed, Manitoba, to the 1968 level to analyze the effects on streamflow and sediments. Results showed that wetland restoration decreased peak discharge by 24%. More recently, Pomeroy et al. (2014)investigated the effect of prairie wetlands on peakflows and annual volumes for 2007–2013 in the Smith Creek watershed within the PRNA. They simulated several wetland scenarios that included the historical maximum, an estimated wetland extent in 2000 and 2008, and total wetland drainage in the watershed. Their mainfinding was that wetland drainage increased annualflow volumes and peaks by 55% on average in a total wetland drainage scenario. Peakflows were most impacted pro-portionally during dry years. However, there were considerable impacts during wet years such as 2011.
In addition to the economical and biological implications, wetland drainage might affect significantly water levels of the lakes located in closed basins that are found throughout the PRNA. In the province of Saskatchewan, lakes such as Waldsea, Lenore, and Little Quill have presented increasing water levels from the 1960s (van der Kamp et al., 2003). In fact, Little Quill Lake merged with Mud and Big Quill, forming one lake that each year since 2008 has recorded a new historical maximum level (Water Survey of Canada, 2017). A study carried out by Chamulak (1999)suggested that drainage work in the Waldsea Lake watershed during 1965–1997 may have increased water levels by about 2.4 m. Understanding the impact of wetland drainage on the hydrology in Saskatchewan has become even more important now that the provincial Government approved a new wetland drainage regulation that takes into consideration gated structures to control spring runoffreleases equivalent to a median runoff(Water Security Agency, 2017). Within this context the successful implementation of the SWAT model in the Canadian Prairies, particularly in areas that have agriculture and mining as major economic activities, provides a new tool for the management of water resources with respect to water quantity and quality. The research reported here is part of the Saskatchewan Watershed Evaluation of Beneficial Management Practices project (Cade-Menun et al., 2013) and addresses relevant hydrological issues such as the impact of wetland drainage and climate change on the hydrology of the Pipestone Creek. Thefirst objective of this study was to calibrate and validate the SWAT model at daily and monthly time steps in a watershed with variable contributing drainage area and limited weather and hydrometric data over a period of 14 years (1997–2009). In order to do this, we used the concepts of EDA and GDA, which are widely used across the province of Saskatchewan for hydrological engineering. The successful calibration and validation of the model led to the second objective of this study, which was to assess the impact of wetland drainage on the hydrology of the Pipestone Creek watershed (Fig. 1), particularly in mean daily peakflows and annual volumes.
2. Study area
The Pipestone Creek is located in southwestern Saskatchewan within the Prairie Pothole Region of North America, which extends throughout three Canadian Provinces andfive states of the United States (Batt et al., 1989). The Pothole Region is the result of the last Pleistocene glaciation and it is estimated that there are between 5 and 8 million wetlands or small depressions, which have an important biological role and impact on the ecosystem (Johnson et al., 2010). It is estimated that the Pipestone Creek has a GDA and EDA of 2242 km2and 559 km2, respectively, and that near 11% of the watershed is covered by wetlands (Saskatchewan Watershed Authority, 2005).
The topography in the basin ranges from 547 to 804 m above sea level (m.a.s.l.) with a mean elevation of 650 ma.s.l. Annual precipitation for this area normally ranges from 450 to 500 mm and hydrological processes are typical of cold regions, in which land is covered by snow and soils are frozen during most parts of the winter (Fang et al., 2007). In this area and the Canadian Prairies in general, snow is the main driver of runoffand snowmelt can produce up to 80% of annual local surface runoff(Gray and Landine, 1988). Therefore, snow process such as snow accumulation and transport are an important part in the hydrological cycle.Fig. 1 illustrates the GDA of the Pipestone Creek and its spatial location within the province of Saskatchewan.
3. Data
3.1. Climate data
stations and data used in the initial configuration of the SWAT model.
Only one of these stations (Kipling) is located within the watershed. However, the Broadview and Whitewood stations provide reliable data due to their proximity, particularly for the northern area of the watershed. There are no major spatial changes in temperature and precipitation driven by topographic changes in the watershed. The summer months of June, July, and August record nearly 50% of the annual precipitation and the maximum temperatures of the year.
3.2. Hydrometric data
Hydrometric data for the watershed have been collected by Environment Canada (Water Survey of Canada) since the early 1960s (Pipestone Creek above Moosomin, 05NE003). However, the data for this hydrometric station present several gaps and for the purpose of this study only data from 1995 to 2009 were used. Daily and monthly hydrometric data for this station were obtained from Water Survey of Canada (https://www.ec.gc.ca/rhc-wsc/) for 1995–2009. The temporal distribution of the data is from March 1st through October 31st of each year, which is the period of time that Environment Canada operates this station.
As is typical of the Canadian Prairies, streamflows present a large inter-annual variability, which is driven by precipitation variability (Bonsal and Regier, 2007). The 15-year record illustrated inFig. 2includes remarkably highflows in 1996 and 2001, when the mean dailyflows exceeded 50 m3s−1. In contrast, mean dailyflows did not reach 6 m3s−1in 2002, 2004, and 2008. Frequency analysis of mean daily peakflows suggests aflow of∼9 m3s−1for a normal year, which has a probability of exceedance of 50%. Table 2summarizes the estimated mean daily peakflows by return period.
Peakflows during this time period were quite variable and ranged from as low as 1.1 m3s−1in 2004
–54 m3s−1during the spring runoffof 2001, which correspond to a return period of slightly higher than 25 years. Within the 15-year time period, three years presentedflows below normal and another three years show near normalflow; in other years, mean daily peakflows were above normal. Most peakflows during this period occurred during spring runoff; the exceptions were 2000 and 2007, when peakflows were
Fig. 1.Pipestone Creek Watershed and sub-basins used for the hydrological modeling.
Table 1
Weather stations and data used in the initial configuration of the SWAT model for the Pipestone Creek watershed.
Station Name Latitude Degrees Longitude Degrees Elevation (m) Period Variables
Kipling 50.200 −102.734 671.0 1990–2009 Precipitation/Temperature
Broadview 50.367 −102.567 599.8 1990–2009 Precipitation/Temperature
driven by summer rainfall.
Mean monthlyflows are illustrated inFig. 3. The maximum monthlyflow occurred in April, was nearly 6 m3s−1, and was driven by spring snowmelt runoff. Flows and volumes in April were up to four times theflows and volumes observed in March, May, June and July. Monthlyflows during August decreased considerably and in general did not exceed 0.5 m3s−1. Meanflows during Sep-tember and October were minimal (Fig. 3).
3.3. Digital elevation model (DEM)
A 15 m DEM for the Pipestone Creek watershed, produced by the Saskatchewan Geospatial Imagery Collaborative (https://www. flysask2.ca/), was used to determine theflow direction. Initially this DEM was used to delineate the watershed sub-basins and the streamflow network. Large differences between this delineation and the real watershed andflow network were observed; therefore, a
Fig. 2.Mean dailyflow recorded at Pipestone Creek above Moosomin Reservoir (1990–2009).
Table 2
Frequency analysisaof mean daily peak
flows at Pipestone Creek above Moosomin (05NE003).
Probability of Exceedance in any Given Year (%) Return Period (Years) Mean Daily Peak (m3s−1)
50 2 9.4
20 5 21
10 10 32
4 25 49
2 50 66
1 100 87
0.5 200 112
0.2 500 157
aThe frequency analysis of peak
flows was carried out for 1960–2011 using the program HydroFreq v1.0 andfitting a Generalize Extreme Value distribution.
62.5 cm Orthophoto, also provided by the Saskatchewan Geospatial Imagery Collaborative, was used to refine the stream network and sub-basin boundaries.
3.4. Soil data
Soils in this area of Saskatchewan are dominated by Oxbow loam (Orthic Black Chernozems), which are grassland soils in the black soils zone that have a loamy texture (Saskatchewan Watershed Authority, 2005).
The soil data required to set up the SWAT model were obtained from the Saskatchewan Soil Information Database version 4 (SKSIDv4). Soil data can be imported to ArcSWAT by using the predefined soil classifications (for the U.S. only) that comes with the model or by using a customized soil database. Due to differences in soil classification between Canada and the USA, a customized soil database was built. This required formatting the database and calculating, from empirical equations, some soil parameters that were not sampled in Canada, but allowed the inclusion of up to 10 soil layers and their parameters.
Table 3summarizes the soil parameters required by SWAT.
Calculated soil parameters were the soil hydrologic group, the available water capacity, the soil albedo, and the universal soil loss equation (USLE) erodibility factor. The soil hydrologic group was estimated from the sand and clay content of thefirst horizon and was classified as the following (Table 4).
The soil available water capacity was defined byJones (1975)as the difference between the retained water capacity and the volume of water held at 15 bar suction (Eq.(1))
Av =ϴv (0.05)−ϴv(15) (1)
The soil albedo (Baumer, 1990) was estimated using the following equation:
SOL_ALB = 0.6/exp(0.4*SOL_OM) (4)
Where: SOL_OM: Percent organic matter of the surface soil
Table 3
Soil parameters required by the SWAT model.
Code Name Source
HYDGRP Hydrologic Group Table 5
SOL_ZMX Maximum rooting depth of soil profile(mm) SKIDv4 ANION_EXCL Fraction of porosity from which anions are excluded Default (0.5) SOL_CRK Potential or maximum crack volume of the soil profile SKIDv4 TEXTURE Texture of soil layers NA (optional) SOL_Z Depth to bottom offirst soil layer (mm) SKIDv4 SOL_BD Moist bulk density (Mg/m3) SKIDv4
SOL_AWC Available water capacity (mm/mm) Eq.(1)
USLE_K USLE equation soil erodibility (K) factor Williams, 1995
And organic matter can be calculated as: OM = 1.72 * organic Carbon
(5) The customized soil database for the Pipestone Creek consisted of 48 different types of soil profiles; each had information for four different layers to a maximum soil depth of 1 m.
3.5. Land cover and land use
Land cover and land use at a scale of 1:25,000 was obtained for the watershed from theBoychuk (2009)report. This inventory was produced with information collected from 1999 to 2001 and, for modeling purposes, was considered constant throughout the modeling period. The biophysical classes for land use and cover classification for the Pipestone Creek Watershed are summarized in Table 5.
These 16 biophysical classes were translated into classes within the SWAT software, resulting in 8 major SWAT land uses. More than 75% of the watershed is used for agriculture, including both cropland and pasture (Table 6).
*Includes UTRN and URBN
3.6. Pipestone reservoir
Unfortunately, there are no official operation records or water level information for the reservoir. Unofficial information suggests that the dam was operated based on complaints, whereby water would be released downstream in mid-October to reduce reservoir water levels. This occurred sporadically; thus, no operation rules were included in the model for the reservoir. In the simulation, waterflowed downstream when the water level reached the spillway elevation. The capacity of the reservoir to the spill elevation was estimated to be 1800 dam3with a surface area of 120 ha.
3.7. Non-contributing areas, wetland representation and internal storage
The Canadian Prairies hydrology is characterized by internal storage that occurs within basins (wetland areas and ponds), making the contributing areas to runoffvariable for every runoffevent. The areas that are internally drained will spill only after reaching
Table 5
Biophysical classes in the Lower Souris River Watershed, fromBoychuk (2009). Lower Souris River Watershed
Area covered for the different land uses in the watershed.
Table 7
Summary of calculated wetland surface areas and volumes. WET_FR: fraction of the surface area that contributes to the wetland; WET_MXSA: maximum wetland surface area in ha; WET_MXVOL: maximum wetland volume in 104m3; WET_VOL: volume of water in the wetland at the beginning of the simulation in 104m3. Normal
wetland surface area was assumed to be 90% of the maximum surface areas.
SUB-BASIN WET_FR WET_MXSA WET_MXVOL WET_VOL
1 0.68 603.9 429.8 171.9
2 0.94 891.4 633.9 253.6
3 0.95 1229.1 873.6 349.5
4 0.93 832.4 592.0 236.8
5 0.63 1208.8 859.2 343.7
6 0.96 520.1 370.3 148.1
7 0.84 403.5 287.5 115.0
8 0.00 11.0 5.3 2.1
9 0.44 34.6 21.5 8.6
10 0.00 0.0 0.0 0.0
11 0.00 1.1 0.3 0.1
12 0.72 453.8 323.2 129.3
13 0.75 1118.9 795.4 318.2
14 0.81 893.4 635.3 254.1
15 0.65 423.2 301.4 120.6
16 0.47 4.5 1.8 0.7
17 0.46 146.7 105.1 42.1
18 0.27 0.9 0.2 0.1
19 0.42 284.4 202.9 81.2
20 0.24 386.1 275.1 110.0
21 0.70 1382.9 982.9 393.1
22 0.99 3415.8 2426.2 970.5
23 0.38 518.2 368.9 147.6
24 0.86 1736.3 1233.8 493.5
25 0.97 428.8 305.4 122.2
26 0.68 1042.1 740.9 296.3
27 0.79 1891.8 1344.2 537.7
their maximum storage capacity. The SWAT model allows for the inclusion of only one wetland/pond per sub-basin; this certainly is not the case for the Pipestone Creek watershed, where there are thousands of wetlands. To represent these thousands of wetlands within SWAT we clustered all the wetlands in each sub-basin and estimated a HEW (Wang et al., 2008) by adding their surface area and volume.Table 7summarizes surface area, wetland volume, the fraction of watershed that is contributing to the wetland, and wetland initial volume, which is an arbitrary value. The contributing area to the wetland is estimated to be the area that does not contribute to runoffduring a normal year, i.e., the difference between the GDA and EDA. Thefinal EDAs delineated for the Pipestone Creek Watershed are illustrated inFig. 4.
The estimation of the contributing area to the HEW is based on the principle of the effective drainage areas, which assumes that the wetlands outside the effective drainage area do notfill and spill during a normal runoffyear. The effective and gross drainage areas were delineated by the now defunct Prairie Farm Rehabilitation Administration (PFRA) Hydrology Division (Martin, 2001). This initiative originated in the 1970s for the purposes of regionalflood and runoffstudies. This data base was last updated in 2008; however, it continues to be used for many hydrological studies in the Province of Saskatchewan.
Surface areas for the wetlands in the watershed were extracted fromBoychuk (2009)and were assumed to be the maximum surface areas for the wetlands. Wetland volumes were then calculated using the equations proposed byWiens (2001):
For wetlands with surface areas up to 70 ha:
V = 2.85*A1.22 (6)
For wetlands with surface areas greater than 70 ha:
V = 7.1*A + 9.97 (7)
Where: A: Surface area in hectares (ha) V: Full volume (dam3)
4. Development of drainage scenarios
A baseline scenario was required to assess the impact of wetland drainage for different scenarios, using the GDA and EDA during the calibration period. The three drainage scenarios considered a drainage percentage of the NCDA. Implementing these scenarios involved the following steps:
i) Estimating the area that will be drained according to the different scenarios (15, 30, and 50% of the NCDA of the sub-basin). ii) Identifying the area that will be drained, which is based on topography and visual criteria that take into consideration proximity
to the stream and avoids extremely large water bodies. iii) Delineating the“New”EDA.
iv) Estimating the NCDA for the scenario. v) Identifying the wetlands within the new NCDA.
vi) Calculating areas and volumes of the individual wetlands according to equations 1 and 2.
vii) Estimating the HEW for each sub-basin (surface area and volume), and the area contributing to the HEW. viii) Entering the new data into the model.
Delineating the new EDA (iii) also required indirectly defining the NCDA or the drainage areas that were contributing to the HEW. The visual criterion was fundamental when defining the new drainage areas, primarily because the DEM may present significant differences in elevation in relation to the actual photos, resulting in watershed boundaries going over water sloughs and wetlands. These issues are likely the result of veryflat areas and the“coarse”resolution of the DEM.
All scenarios were based on an increasing EDA that reduces the NCDA or the drainage area that contributes to the HEW; thus, the number of wetlands within this area decreases.Fig. 5illustrates sub-basin 12 of the watershed and the different drainage areas under the three different scenarios.
5. Model set up, sensitivity analysis, calibration and validation
The initial sub-basin and stream definition was done with the automatic delineation tool incorporated in the ArcSWAT model, which uses the DEM information to define basin boundaries and streamflow direction. However, significant differences were found when comparing streams and sub-basins defined with the automatic tool to an orthophoto of the whole basin. As such, sub-basin and stream definitions were refined using the 2.5 m orthophoto and the 15 m DEM. Twenty-seven sub-basins and reaches were delineated and 561 hydrological response units were created. The Pipestone Reservoir was added to the model as a reservoir and a wetland unit was added to each sub-basin to model sub-basin internal storage.
The remaining model parameterization, which included groundwater and snow components, was also achieved within ArcSWAT. Once all the parameters were set in the model according to the best knowledge of the watershed, thefirst run was executed at daily and monthly time steps.Fig. 6illustrates the simulated results with the initial parameterization.
hydrograph and a reduction of the curve number by itself will make the watershed less responsive to these events.
5.1. Sensitivity analysis
Calibrating the model requires identifying parameters that are more sensitive to changes in their values, such that a small change in the parameter would produce a significant change in thefinal result. Sensitivity analyses were conducted using default values for most of the model parameters. Potential evapotranspiration was calculated with the Hargreaves equation (Hargreaves et al., 1985) and surface runoffwas calculated by using the Curve Number (SCS, 1972) approach and variable storage routing.
Sensitivity analysis and calibration were done with the SWAT-CUP software and the Sequential Uncertainty Fitting ver. 2 (SUFI2) algorithm (Abbaspour et al., 2004). Thirty-two parameters were considered for the sensitivity analysis, which included global sen-sitivity and one variable at a time. For global sensen-sitivity analysis, the model was run after entering a possible range of values for each of the thirty-two parameters and then evaluating t-stat and p-values for each parameter. The t-stat expresses the sensitivity of the
Fig. 6.a) Modeling results of the daily simulation after the initial parameterization in ArcSWAT. Below (b) the results of monthly simulation after the initial parameterization are illustrated. Note that for illustration purposes the data is continuous and does not represent the hydrograph for the period of time simulated, i.e. the monthly time series have data from October and it is followed by data from March of the following year.
Table 8
Parameters used for the sensitivity analysis.
Parameter name Component Description
ALPHA_BF gw Baseflow recession constant 0–1 GW_DELAY gw Groundwater delay time (days
GWQMN gw Threshold depth in shallow aquifer for returnflow (mm) GW_REVAP gw Groundwater re-evaporation coefficient
RCHRG_DP gw Deep aquifer percolation fraction (%)
HRU_SLP hru Slope
SLSUBBSN hru average slope length (m)
OV_N hru Manning's“n”value for overlandflow WET_MXVOL pnd Wetland maximum volume WET_MXSA pnd Wetland maximum area
WET_FR pnd Fraction of surface area that contributes to the wetland WET_K pnd Wetland hydraulic conductivity
CN2 mgt Initial SCS curve number
ESCO bsn Soil evaporation compensation factor EPCO hru Plant uptake compensation factor TIMP bsn Snowpack temp lag factor SURLAG bsn Surface lag coefficient
SMTMP bsn Snowmelt temp
SFTMP bsn Snow fall temp
SMFMN bsn Minimum snowmelt rate
SMFMX bsn Maximum snowmelt rate
SOL_AWC sol Soil available water capacity SOL_K sol Soil hydraulic conductivity
SOL_BD sol Soil bulk density
CH_N2 rte Manning's‘n' for the main channel CH_K2 rte Hydrologic conductivity for the main channel ALPHA_BNK rte Baseflow alpha factor for bank storage
TLAPS sub Temp lapse rate
CANMX hru Maximum canopy storage
REVAPMN gw Threshold depth in shallow aquifer for re-evaporation
SOL_ALB sol Moist soil albedo
parameter, and a larger absolute value of t-stat means a more sensitive variable; the p-value represents the significance of the sensitivity. Sensitivity analysis for one variable at a time consisted of keeping all variables constant and changing only one within a certain range. The parameters for sensitivity analysis (Table 8) were parameters typically used to calibrate streamflow, including snow, groundwater, channel, and soil parameters (Arnold et al., 1998), plus wetland parameters such as the wetland fraction, surface area, and volume.
5.2. Model calibration and validation
Daily and monthly calibrations and validations were performed for the periods 1997–2005 and 2005–2009, respectively. The calibration process used semi-automated calibration combined with manual calibration, using 1995 and 1996 as warm up years for the model. Manual refinement was performed to ensure that parameter ranges chosen by the semi-automated process had a physical meaning according to watershed characteristics. For example, the deep aquifer percolation fraction (RCHRG_DP) was set to a high value (0.95) because there is little groundwater contributing to runoffin this watershed (van der Kamp et al., 2003).
These semi-automated calibration and validation processes were carried out with SUFI2 within the SWAT-CUP software (Abbaspour et al., 2007) to address uncertainties in the parameters as uniform distributions. Uncertainty in the model was addressed with the 95% prediction uncertainty (95PPU), which is the bandwidth between the 2.5% and 97.5% levels of the cumulative dis-tribution output, resulting from Latin Hypercube sampling (Abbaspour et al., 2007). The algorithm follows the principle that a single parameter produces a single model response while the propagation of the uncertainty of the parameter will result in the 95PPU; i.e., the greater the parameter uncertainty the greater the model output uncertainty. The practical idea of the 95PPU is that the output bandwidth should cover most of the observation. To quantify the model results, the P-factor give the percentage of data that is within the 95PPU, and the R-factor gives the thickness of the 95PPU (average thickness of the 95PPU band divided by the standard deviation of the observed data). Suggested but notfirm values for these two statistics are P-factor > 0.7 and R-factor < 1.5 (Abbaspour et al., 2015). More details on SWAT-CUP and the SUFI2 algorithm can be found inAbbaspour et al. (2007).
Other statistics to compare the best simulation with the observed data is the Nash-Sutcliffe efficiency coefficient (Nash and Sutcliffe, 1970), which is used to provide an idea of the performance of the calibration. The NS is calculated according to Eq.(8).
= − −
Values of E range from−∞to 1, with 1 indicating that the model represents perfectly the observed discharge, and 0 indicating that the model has the same accuracy as the mean of the observed data.
Model validation followed the calibration process to evaluate model accuracy. During the validation stage, the model was used to predict daily and monthlyflows for a period different from the calibration (Wilson, 2002), in this case 2005–2009, and used the optimal parameters values selected during the model calibration. The P-factor and R-factor were then estimated for this period of time and the predicted (best simulation) and observed values were compared using the NS coefficient and the coefficient of de-termination.
6. Results and discussion
6.1. Daily calibration
The SUFI2 algorithm within SWAT-CUP software was used for daily calibrations by realizing 500 simulations for the six most sensitive parameters. The parameters for daily and monthly calibrations and their initial andfinal parameter range are summarized in Table 9. Among the parameters analyzed for sensitivity, maximum snowmelt factor (SMFMX) and curve number (CN2) were found to be the most sensitive.
Despite the limited weather data available for the watershed, the lack of operational information for the existing dam (Pipestone
Table 9
Parameters used for the model calibration and validation.
Parameter Initial Range Lower Limit Upper Limit
Reservoir), and the complexity of the watershed with variable contributing areas, the SWAT model was successfully calibrated and validated at daily and monthly time steps for a period of 13 years. The model performed satisfactorily in terms of the P-factor and R-factor, although the P-factor was less than what is recommended byAbbaspour et al. (2015). The recommended values are not set rules and depend on data availability and the characteristics of the watershed. In our case the variable contributing drainage area added a unique complexity for simulating the watershed hydrology. Nevertheless, the P-factor and R-factor were in the order of other published studies (Abbaspour et al., 2015). Other calibration statistics such as the NS (0.72) and R2(0.72) that compared the best simulation with the observed data also suggested satisfactory results. However, during the validation both coefficients decreased (R2= 0.65, NS = 0.64). These results were similar to other watershed modeling published in the literature (Abbaspour et al., 2007; Abbaspour et al., 2015).Fig. 7illustrates the best simulation and the observed data. Overall, the model captured relatively well the inter-annual variability observed in recordedflows. Spring runoff(snowmelt driven) was in general well represented by the model, in particular the timing of peakflows, although there are some differences in their magnitude. Spring runoffpeakflows were over-estimated for 1997, 2002, 2004, and 2007, but were underover-estimated in 1998, 1999, 2003, and 2009. The model also captured some summer runoff(post snowmelt runoff), mostly in the timing of the event and not necessarily the magnitude of the peaks and volume. Sharp summer peaks after snowmelt were captured by the model, which tended to overestimate their magnitude (years 1998, 2000, and 2001) with the exception of 2007, where the summer peak was underestimated. Summer runoffis usually driven by intense precipitation events that might or might not cover large portions of the watershed. The overestimation of most of these peaks can arise because precipitation only covered the northern and western parts of the watershed where the weather stations were located (Fig. 1), but the model assigned this precipitation to the rest of the watershed.
During the validation period, the model captured the timing of the spring runoffwell but it was less consistent in capturing the magnitude. The significant summer runoffevent of 2007 was well simulated in the timing but underestimated in its magnitude. During the four years of validation there was one normal year (2006), one year with near-normal spring runoffbut a wetter than normal summer (2007), one dry year (2008), and another wet year (2009) in terms of peakflows; thus, the model did relatively well under the variable hydrologic conditions in such a short period of time.
Fig. 7.Measured and simulated daily discharge at Pipestone Creek above Moosomin Reservoir during calibration (1997–2005) and validation (2005–2009).
Table 10
Comparison between observed and simulated volume (dam3) for the calibration and validation periods.
Simulated Observed Sim/Obs
1997 34,410 43,417 0.79
1998 20,395 20,060 1.02
1999 31,204 47,610 0.66
2000 22,414 18,293 1.23
2001 56,795 67,087 0.85
2002 18,576 10,756 1.73
2003 13,706 24,822 0.55
2004 10,670 4357 2.45
2005 25,335 24,755 1.02
2006 12,360 17,092 0.72
2007 42,779 37,156 1.15
2008 4915 8384 0.59
2009 13,420 20,740 0.65
A comparison of the annual volumes suggests that the model did relatively well for most years during the calibration period (Table 10).
Total simulated volumes for the years 1998, 2001, 2005, and 2007 were within 15% of the observed volumes. The greatest difference between simulated and observed volumes occurred in 2004; spring runoffwas below normal, and the model overestimated the spring runoffpeak. In 2002, the simulated volume was just over 70% of the observed volume, indicating the model overestimated the spring runoffpeak for that year, too. The annual volume for 1998 was well estimated (within 2%), which is mostly due to the model showing some response to summer precipitation that year. On the other hand, all the volumes for the validation period were within 41% of the recorded volumes. Overall, the volumes simulated using SWAT over the whole-time period were 11% less than the recorded volumes.
6.2. Monthly calibration
The performance of the model was also considered satisfactory at monthly time steps, with P-factors and R-factors slightly greater than during the daily calibration. The coefficient of determination for the best simulation was greater than 0.8 during the calibration and validation period while the NS was also greater than 0.8 during the calibration and decreased slightly in the validation period. A visual comparison of the simulated (best simulation) and observed monthlyflows is presented inFig. 8. In general, the model performed very well, replicating the peaks and hydrograph shape. Summer runoffalso was well simulated by the model, with only one considerable exception during 1997. Sharp peaks were also well replicated; however, the model had more difficulties re-producing theflatter hydrographs such as the spring runoffof 1999 or 2002. Theflatter hydrographs represent a slow response of the watershed to snowmelt, in case of the springflows, or precipitation events during the summer. This slower response is also due to the routing effect that the wetlands have, which also reduces the peaks.
The performance of the model decreased during the validation period; however, the statistics for this period were still good. The
Fig. 8.Measured and simulated monthly discharge at Pipestone Creek above Moosomin Lake during calibration (1997–2001) and validation (2002–2005).
Table 11
Annual simulated and observed volumes (dam3).
Year Simulated Observed Simulated/Obs
1997 33,086 43,417 0.8
1998 23,687 20,060 1.2
1999 33,492 47,610 0.7
2000 22,668 18,293 1.2
2001 59,214 67,087 0.9
2002 17,163 10,756 1.6
2003 14,966 24,822 0.6
2004 11,398 4357 2.6
2005 25,234 24,755 1.0
2006 15,505 17,092 0.9
2007 53,509 37,156 1.4
2008 7108 8384 0.8
2009 18,054 20,740 0.9
visual comparison showed a difference during the spring runoffof 2007, with an underestimation of the monthly peak and annual volume. Total volumes of the simulation period were within 5% of the recorded volume.Table 11summarizes the simulated and recorded annual volumes and the annual ratio of simulated to observed volume. The greatest difference in annual volumes occurred in 2004 due to the overestimation of the peak. However, this different was not significant in the balance for the whole period, since 2004 was a drier than normal year.
Observed runoffevents such as in the summer of 2000 were replicated by the model at daily and monthly time steps; however, they were overestimated by both simulations. This can be logically attributed to local precipitation that was captured by the weather stations and extended across the watershed by the model. In cases like this a high-density weather network would provide valuable information for hydrological modeling and would reduce the uncertainty in the weather observation component. As an example, private records of precipitation such as the weather farm network might provide important precipitation data in areas where official weather stations are scarce. The negative side of using these sources of weather data is that generally they are not quality-approved and there are no long term records.
In order to represent the characteristics of the watershed and incorporate the wetlands within SWAT, some assumptions had to be made. The wetland surface areas were estimated to be the maximum areas for the wetland, and volumes were derived from empirical equations developed in other studies (Wiens, 2001). The drainage area contributing to the hydrologic equivalent wetland was as-sumed to be the area within the watershed that does not contribute to runoffduring a normal year, or simply the difference between the GDA and EDA. The results of the sensitivity analysis for the wetland parameters suggest that these assumptions were valid and had low impacts on the overall performance of the model. However, more refinement in estimating wetlands volumes could help to better reproduce the hydrograph shape, in particularflatter hydrographs that might be subject to routing effects. Other studies that have used the HEW (Wang et al., 2008; Yang et al., 2010) with the SWAT model have in general calibrated for the wetland para-meters.
6.3. Results of the drainage scenarios
The proportional increment of the EDA based on the NCDA was done without analyzing the hydraulic and cost feasibility, and in some sub-basins expanded the EDA up to 37 times (i.e. sub-basin 22). The EDA in sub-basin 22 was estimated to be 4 km2and the NCDA 311 km2; therefore, according to the criteria used to define the drainage scenarios, a maximum EDA of 160 km2was assumed. On average, draining 15, 30, and 50% of the NCDA increased the EDA of the Pipestone Creek watershed by 2.2, 3.4, and 5 times, respectively.Table 12summarizes the EDA, GDA, and NCDA for the baseline conditions and the three drainage scenarios.
The surface area of the wetlands in each sub-basin decreased 9–53% for Scenario 1, 20–61% for Scenario 2, and between 35 and 66% for Scenario 3. The exceptions were sub-basins 8, 10, 11, 16, and 18, which are small and most of the drainage area contributes
Table 12
Estimated gross (GDA), effective (EDA), and non-contributing (NCDA) drainage areas by sub-basin. *Drainage area for sub-basin 10 is 0.2 km2.
Base Line Scenario 1 Scenario 2 Scenario 3
Table 13
3 103 1229 8736 0.95 87 962 6840 0.80 71 713 5074 0.66 52 469 3339 0.48
4 65 832 5920 0.93 54 613 4363 0.76 43 493 3511 0.61 31 369 2632 0.44
13 120 1119 7954 0.75 98 880 6256 0.61 83 764 5433 0.52 60 529 3769 0.38
14 69 893 6353 0.81 55 641 4564 0.65 46 546 3887 0.55 34 400 2853 0.40
21 108 1383 9829 0.70 91 1102 7832 0.59 75 972 6908 0.49 54 814 5792 0.35
22 311 3416 24262 0.99 261 3099 22016 0.83 220 2584 18358 0.70 155 2041 14498 0.49
23 32 518 3689 0.38 26 281 2005 0.32 19 201 1440 0.23 15 174 1244 0.19
24 130 1736 12338 0.86 108 1336 9495 0.72 88 1240 8811 0.58 65 1064 7562 0.43
25 83 429 3054 0.97 69 300 2139 0.81 57 249 1780 0.67 40 177 1267 0.46
26 82 1042 7409 0.68 72 890 6327 0.59 55 729 5187 0.45 42 614 4371 0.34
27 121 1892 13442 0.79 104 1700 12078 0.68 86 1456 10345 0.56 63 1009 7176 0.41
to runoffduring a normal year (Table 13). On average, the wetland surface areas were decreased by 26%, 38% and 52% for Scenarios 1, 2 and 3, respectively
6.4. Daily drainage scenarios
As described previously the performance of the daily model was considered satisfactory, particularly during the calibration period. Due to the unique characteristics of the model and the inclusion of the HEW, the timing of the peaks was not analyzed. In addition, the results presented here are not absolute numbers but they provide an idea of the impact of drainage on peakflows, particularly during the spring runoffperiod for which the model performs better. The results apply only to the watershed, for the period analyzed and it is not recommended to extrapolate the results because the watershed has unique characteristics and internal storage that might differ significantly from other watersheds.
Drainage of 15% of the NCDA in the Pipestone Creek increased mean daily spring runoffpeakflows by 50% during the simulation period (1997–2009). The modeling results of this scenario suggest that the largest impact on peakflows occurred during 2008; in that year, mean daily peakflows would have been 85% greater. That year was recorded as a dry year; therefore, the percentage increase would translate to aflow of∼5 m3s−1, which is still below a normal year. In other dry years such as 2000 and 2004 the impact on mean dailyflows was estimated to be 45%, although the model had some difficulties simulating these dry years.
For the purpose of this analysis the term“wet years”refers to years in which observed peakflow was greater than 20 m3s−1, which were well simulated by the model. According to this criterion, the wet years were 1997, 2001 and 2005. The increment of peak flows during the wet years ranged between 28 and 45%, with an average increment of 36%. This increment was smaller than the average increase for spring runoffpeaks. The smallest increase in mean daily peak for this scenario occurred for the spring runoffof 2001, for which the increment was 28%. The impact of drainage on summer peaks was less than for the spring runoff, 17% versus 50% during the snowmelt period.
The reduction of the NCDA by 30% in Scenario 2 suggests an increase of mean daily spring peakflow between 28 and 127%, with an average increase of 57%. As for the previous scenario, the largest spring runoffincrease (127%) occurred for 2008. Summer peaks also presented a smaller increase, with the summer runoffof 1998 presenting the lowest increase of all (17%). In general the increase was greater during spring runoff(average of 79%) than during the few summer events recorded in the simulated period (32%), which may occur because the whole watershed is wet and there are small losses during snowmelt. In contrast, the summer peaks simulated by the model were preceded by dry conditions in which there was little to no runoffin the watershed; thus, the response is smaller in magnitude.
The impacts of draining 50% of the NCDA on mean daily peakflows were much greater. On average, this scenario increased spring peakflows by 113% during the simulated period. Increments of peakflows ranged from 70 to 177% depending on the year. As observed for the previous scenarios, the low runoffyear of 2008 had the greatest increase in peakflow (177%) which in terms of magnitude meant a peakflow of∼7 m3s−1, still below normal peakflow. Spring runoffpeaks during the wet years would increase on average 81%, which is still below the mean for the whole period.Table 14summarizes the percentage increase of the peakflows during the simulated period. The summer peaks and wet years presented the smaller increases in terms of percentage.
Overall, for the three scenarios, summer peaks were impacted the least by drainage of the NCDA. During spring runoff, wet years had a smaller increase in terms of percentage than the other years (Table 14). It was also observed that the percentage of increment calculated for the years in which peakflows was less than 13 m3s−1was greater than for the years in where peakflows were greater than 13 m3s−1(56, 90 and 131% versus 42, 64, and 91% for Scenarios 1, 2, and 3, respectively).Fig. 9illustrates the percentage of change of peakflows for the whole period and for the scenarios.
6.5. Monthly drainage scenarios
The annual volumes of runoffunder the three different drainage scenarios were analyzed using monthly simulation because the model performed better at the monthly time steps. Monthlyflows were expected to increase for all three scenarios by about 43, 68 and 98%. The increments expected for Scenario 1 ranged from 25 to 69% depending on the year. This scenario did not present a clear tendency to increase or decrease monthlyflows based on the magnitude or return period of the runoffyear. However, wet years (1997, 2001 and 2007) in terms of the annual volume had lower volume increases (36%) relative to the average (43%). The greatest volume increase was for 2006, which had a slightly above average volume; the smallest volume increase occurred during 2000 (25%). No clear relationship was observed between the magnitude of the runoffyear and the magnitude of the impact caused by drainage.
Scenario 2 was expected to increase the annual volume between 43 and 98%. On average the watershed was expected to produce
Table 14
Percentage of increment of mean daily peakflows during the simulated period.
Spring Peaks (%) Summer Peaks (%) Wet Years (%) (1997,2001,2005)
Scenario 1 50 17 36
Scenario 2 79 32 57
Fig. 9.Percentage of change of peakflows by drainage scenarios.
68% more runoffwith 30% of the NCDA drained. The annual volume increase was less during the wet years relative to the mean of the whole period (56 vs 68%). As for the previous scenario, years with near normal or slightly above normal volume had the largest increment in annual volumes.
Scenario 3 produced the largest impact on annual volumes, which were expected to be 98% greater. The annual increments ranged between 65 and 138%, with 2008 again having the greatest increment. As with the analysis of the peakflows, 2008 had an annual runoffvolume that was slight less than normal (RP of 2 years) and considering the drainage impact of this scenario, the total runoffvolume corresponded to a return period of less than 3 years.Fig. 10illustrates the mean annual percentage of change by scenario as well as the whole range of change within the scenario through a Box-Whisker plot.
Monthlyflows from March to October were expected to increase 11–49% for Scenario 1, 28–77% for Scenario 2, and 48–115% for Scenario 3. The greatest increase inflows during the simulated period occurred in April for Scenario 1 and March for Scenarios 2 and 3. In terms of magnitude,flows in April had the greatest increments for all three scenarios. Monthlyflows were expected to be 49, 75, and 106% higher during the month of April. Summerflows (June-July-August) also showed a considerable increase, which likely resulted in the streams running longer during the summer or not drying out as has happened historically.Fig. 11illustrates the monthlyflow increments for the different scenarios. Runoffduring June presented the least impact of drainage during the summer months.
7. Discussion and conclusions
The SWAT model was successfully applied to the Pipestone Creek watershed, which has limited weather information. Three weather stations provided observed precipitation and temperature data for the western and northern parts of watershed. However, the lack of observed historical weather data in the southern and southeastern parts of the watershed added some uncertainty to the performance of the model, which is represented by the P-factor and R-factor.
The snow module of the SWAT model was previously analyzed byWang and Melesse (2005); they tested a previous version of the model and found that it performed relatively well for the seasonal and annualflows and had a satisfactory performance for daily flows.Troin and Caya (2014)also concluded that the snow module within SWAT worked well in cold environments. The results of the current study suggest that the performance of the snow module in SWAT2012 performed well, reproducing daily and monthlyflows. In general, the model reproduced well the timing and the magnitude offlows with a few exceptions.
The physical characteristics of the watershed are typical of the Prairie Pothole Region, which is dominated by wetlands, making the contributing area to runoffvariable for every event. This, combined with the high inter-annual variability in precipitation and hydrology dominated by snowmelt events, suggested that traditional approaches might not successfully reproduce the hydrology of this watershed. The major challenges of watersheds like the Pipestone Creek are to determine the contributing area to runoffand to represent in the model the thousands of wetlands that can be found in this type of watershed. Within this context the HEW accounted for all the wetlands in the basin and their drainage area, which during normal years do not contribute to runoff. The implementation of this approach required detailed data about the watershed, wetland areas or volumes that in many cases may not be possible to obtain due to a large number of wetlands and the size of the watershed. In the Pipestone Creek watershed, the wetland identification and the calculation of the surface areas were done previously, which facilitated the execution of this approach. In many other cases this approach might not be feasible because of the costs and time required to identify and classify all the wetlands in the watershed. The use of EDA and GDA to determine the areas contributing to the HEW proved to be successful, because the variable that accounts for the area contributing to the HEW was not found to be sensitive during the daily simulation and had a very low sensitivity during the monthly calibration. This also supports traditional hydrological practices in the province of Saskatchewan by using the concepts of EDA and GDA. Other studies such asWang et al. (2008)andYang et al. (2010)that have used the HEW have calibrated wetland parameters.
The approach used to calibrate and validate the SWAT model was the traditional approach that consists of using two different
periods of time, one for calibration and the other for validation. This approach was successful for a period of 13 years (1997–2009) in whichflows were highly variable from one year to another. Studies such asRahbeh et al. (2011)andAbbaspour et al. (2010), which tried to apply the SWAT model in the Canadian Prairies at different scales, were not able to successfully calibrate and validate the model using the traditional approach. Less traditional approaches were used; however, the validation statistics were less than sa-tisfactory.
The uncertainty statistics (P-factor and R-factor) with the other standards statistics (R2and NS) presented for calibration and validation of the Pipestone Creek Watershed were comparable to the statistics presented for other calibrated and validated appli-cations of the SWAT model in Canada and around the world (Liu et al., 2008; Fontaine et al., 2002, and others).
The calibration and validation statistics for the daily and monthly simulation were considered satisfactory; however, calibration and validation at other hydrometric stations within the watershed would provide additional information about the performance of the model in this kind of watershed.
Even though, there is no general agreement on the literature about the impact of wetland drainage on hydrology, this study present similar results to other studies in the Canadian Prairies (Pomeroy et al., 2014; Yang et al., 2010), which suggest that wetland drainage tends to increase annual volumes and peaks. In terms of magnitude of peakflow and annual volume, spring peaks during wet years presented a below-average increase, because under natural conditions wetlands and ponds reach the thresholds of their spill elevations (Shaw et al., 2011) and then start connecting with other wetlands and contributing toflow downstream. In contrast, the impact of wetland drainage is greater in terms of percentage during high frequency events.
The generalization proposed byAcreman and Holden (2013)in which upland wetlands tend to producefloods is not directly applicable to the watersheds in the Canadian Prairies, because their impact will depend on wetland connectivity, which tends to increase during very low frequency events (wet conditions) or in years with runoffconditions wetter than normal. During those years evaporation losses tend to be below normal, resulting in water levels higher than normal that reduce the storage capacity for the following year.
The increment of peakflows during high frequency runoffdue to wetland drainage has to be interpreted with caution. In reality wetlands are drained by constructing ditches that have a maximumflow capacity and hydraulic conditions, which might or might not be near theflow expected during normal runoffconditions. Therefore, the impact of wetland drainage tends to be greater during high frequency runoffevents.
The successful implementation of the SWAT model in the Canadian Prairies provides a new tool to simulate the difficult hydrology typical of the Prairie Pothole Region, including the water quality component and the assessment of beneficial agricultural man-agement practices. The simulated period 1997–2009 was a relatively wet period where most years had higher peaks and volumes than normal and the wetlands within this region were subject to decadal precipitationfluctuations (Shabbar et al., 2011). The model at daily time steps had some difficulties reproducing the dry years, which suggests that the effects of drainage during normal periods with more dry than wet years might provide different results. During normal and dry years, the wetlandsfluctuate seasonally and inter-annually; many wetlands that dry out are cultivated later in the summer, resulting in plenty of available storage for spring runoff. In the same context, soils are dryer with low moisture conditions and are able to hold more water. Therefore, during wet period normal precipitation produces above normal runoffbecause of the antecedent moisture conditions. Thus, long-term simulation like the one presented in this study are important because consecutive years of runoffabove normal are simulated by the model, accounting for the antecedent moisture conditions.
Acknowledgements
The authors thank Agriculture & Agri-Food Canada for providing funding for this research through the Saskatchewan project of the Watershed Evaluation of Beneficial Management Practices program and the Lower Souris Watershed Committee Inc. for providing land use data and wetland inventory. A special thanks to colleagues at the Water Security Agency, Dr. Kangsheng Wu and Etienne Soulodre, and University of Regina, Dr. Kyle Hodder, for fruitful conversations about the hydrology of the Canadian Prairies. References
Abbaspour, K.C., Johnson, A., Van Genuchten, M.Th., 2004. Estimating uncertainflow and transport parameters using a sequential uncertaintyfitting procedure. Vadose Zone J. 3, 1340–1352.
Abbaspour, K.C., Yang, J., Maximov, I., Siber, R., Bogner, K., Mieleitner, J., Zobrist, J., Srinivasan, R., 2007. Modelling hydrology and water quality in the pre-alpine/ alpine Thur watershed using SWAT. J. Hydrol. 333 (2–4), 413–430.
Abbaspour, K., Faramarzi, M., Rouholahnejad, E., 2010. Hydrological modeling of Alberta using SWAT model. Preliminary Report. . Available onlinehttp:// albertawater.com/hydrological-modelling-of-albertaAccessed September 2017.
Abbaspour, K.C., Rouholahnejad, E., Vaghefi, S., Srinivasan, R., Yang, H., Klove, B., 2015. A continental-scale hydrology and water quality model for Europe: calibration and uncertainty of a high-resolution large-scale SWAT model. J. Hydrol. 524, 733–752.
Acreman, M., Holden, J., 2013. How wetlands affectfloods. Wetlands 33, 773–786.
Ahl, R., Woods, S., Zuuring, H., 2008. Hydrologic calibration and validation of SWAT in a snow-dominated rocky mountain watershed, Montana, U.S.A. J. Am. Water Resour. Assoc. 44 (6), 1411–1430.
Arnold, J.G., Srinivasan, R., Muttiah, R.S., Williams, J.R., 1998. Large-area hydrologic modeling and assessment: part I. Model development. J. Am. Water Res. Assoc. 34 (1), 73–89.
Batt, B.D.J., Anderson, M.G., Anderson, C.D., Caswell, F.D., 1989. The use of prairie potholes by North American ducks. In: van der Valke, A. (Ed.), Northern Prairie Wetlands. Iowa State University Press, Ames, Iowa, pp. 204–227.
Baumer, O.W., 1990. Predictions of Soil Hydraulic Parameters. WEPP Data Files for Indiana SCS National Soil Survey Lab. Lincoln, Nebraska, United States.
Bonsal, B., Regier, M., 2007. Historical comparison of the 2001/2002 drought in the Canadian Prairies. Clim. Res. 33, 229–242.
Bullock, A., Acreman, M., 2003. The role of wetlands in the hydrological cycle. Hydrol. Earth Syst. Sci. 7 (3), 358–389.
Cade-Menun, B.J., Bell, G., Baker Ismail, S., Fouli, Y., Hodder, K., McMartin, D.W., Perez-Valdivia, C., Wu, K., 2013. Nutrient loss from Saskatchewan cropland and pasture in spring snowmelt runoff. Can. J. Soil Sci. 93, 445–458.
Chamulak, T.H., 1999. Waldsea Lake Hydrology Study. Saskatchewan Watershed Authority Report. September.
Dahl, T.E., Johnson, G.E., 1991. Status and Trends of Wetlands in the Conterminous United States, Mid-1970s to Mid-1980s. Department of the Interior, Fish and Wildlife Service, Washington D.C 28pp.
Duffy, W.G., 1998. Population dynamics, production, and prey consumption of fathead minnows (Pimephalespromelas) in prairie wetlands: a bioenergetics approach. Can. J. Fish. Aquat. Sci. 55, 15–27.
Fang, X., Minke, A., Pomeroy, J., Brown, T., Westbrook, C., Guo, X., Guangul, S., 2007. A Review of Canadian Prairie Hydrology: Principle, Modelling and Response to Land Use and Drainage Change. Center for Hydrology. University of Saskatchewan.
Fontaine, T.A., Cruickshank, T.S., Arnold, J.G., Hotchkiss, R.H., 2002. Development of a snowfall-snowmelt routine for mountainous terrain for the Soil and Water Assessment Tool (SWAT). J. Hydrol. 262 (1–4), 209–223.
Godwin, R.B., Martin, F.R.J., 1975. Calculation of gross and effective drainage areas for the Prairie Provinces. In: Canadian Hydrology Symposium–1975 Proceedings, 11–14 August 1975. Winnipeg, Manitoba. Associate Committee on Hydrology, National Research Council of Canada. pp. 219–223.
Gray, D.M., Landine, P.G., 1988. An energy-budget snowmelt model for the Canadian Prairies. Can. J. Earth Sci. 25, 1292–1303.
Hargreaves, G.L., Hargreaves, G.H., Riley, J.P., 1985. Agricultural benefits for Senegal River basin. J. Irrig. Drain. Eng. 108 (3), 225–230.
Johnson, R.R., Oslund, F.T., Hertel, D.R., 2008. The past: present and future of prairie potholes in the United States. J. Soil Water Cons. 63, 84A–87A.
Johnson, C., Werne, G., Guntenspergen, G., Voldseth, R., Millet, B., Naugle, D., Tulbure, M., Carrol, R., Tracy, J., Olawsky, C., 2010. Prairie wetland complexes as landscape functional units in a changing climate. Bioscience 60, 128–140.
Jones, R.J.A., 1975. Soils in Staffordshire II. Sheet J 82 (Eccleshall). Soil Survey Record No. 31. Harperden.
Landers, J.C., Knuth, B.A., 1991. Use of wetlands for water quality improvement under U.S. EPA, Region V, Clean Lakes Program. Environ. Manage. 15 (2), 151–162.
Liu, B.M., Collick, A.S., Zeleke, G., Adgo, E., Easton, Z.M., Steenhuis, T.S., 2008. Rainfall discharge relationships for a monsoonal climate in the Ethiopian highlands. Hydrol. Process. 22 (7), 1059–1067.
Martin F.R.J., 2001. Addendum No. 8 to Hydrology Report #104, Agriculture and Agri-Food Canada PFRA Technical Service: Regina, Saskatchewan, 109 pp. PFRA Hydrology Division, 1983, The determination of gross and effective drainage areas in the Prairie Provinces, Hydrology Report # 104, Agriculture Canada, PFRA Engineering Branch, Regina,S askatchewan, 22 pp. Addendum No. 8 to Hydrology Report #104.
Nash, J.E., Sutcliffe, J.V., 1970. Riverflow forecasting through conceptual models: part I. A discussion of principles. J. Hydrol. 10 (3), 282–290.
Padmanabhan, G., Bengtson, M.L., 1999. A Review of Models for Investigating the Influence of Wetlands on Flooding. North Dakota State University, Fargo, North Dakota, pp. 33.
Pomeroy, J., Shook, K., Fang, X., Dumanski, S., Westbrook, C., Brown, T., 2014. Improving and testing the prairie hydrological model at smith creek research basin. Centre for Hydrology Report No. 14.
Rahbeh, M., Chanasyk, D., Miller, J., 2011. Modelling the effect of irrigation on the hydrological output from a small prairie watershed. Can. Water Resour. J. 38 (4), 280–295.
SCS, 1972. National Engineering Handbook, Section 4. Hydrology, U.S. Department of Agriculture, U. S. Government Printing Office, Washington, DC.
SaskWater, 1993. Magnitude and frequency of peakflows andflow volumes in Saskatchewan. Summary Report and User’s Manual. SaskWater Corporation.
Saskatchewan Watershed Authority (SWA), 2005. Background Report Lower Souris River Watershed. Saskatchewan Watershed Authority Report. November.
Shabbar, A., Bonsal, B.R., Szeto, K., 2011. Atmospheric and oceanic variability associated with growing season droughts and pluvials on the Canadian Prairies. Atmos. Ocean 49, 339–355.
Shaw, D.A., van der Kamp, G., Conly, F.M., Pietroniro, A., Martz, L., 2011. Thefill–spill hydrology of prairie wetland complexes during drought and deluge. Hydrol. Process. 26, 3147–3156.
Stichling, W., Blackwell, S.R., 1957. Drainage area as an hydrologic factor on the Canadian Prairies. IUGG Proc. 111, 365–376.
Troin, M., Caya, D., 2014. Evaluation the SWAT’s snow hydrology over a Northern Quebec watershed. Hydrol. Process. 28, 1858–1873.
U.S Army Corps of Engineers (USACE), 1991. 199. SSARR Users’Manual. North Pacific Division, Portland, USA.
van der Kamp, G., Hayashi, M., 2009. Groundwater-wetland ecosystem interaction in the semiarid glaciated plains of North America. Hydrogeol. J. 17, 203–214.
van der Kamp, G., Hayashi, M., Gallen, D., 2003. Comparing the hydrology of a grassed and cultivated catchments in the semi-arid Canadian prairies. Hydrol. Process. 17, 559–575.
Vining, K.C., 2002. Simulation of streamflow and wetland storage, starkweather coulee subbasin, north dakota, water years 1981–1998. Water-Resources Investigations Report 02-4113. Department of the Interior and U.S. Geological Survey, Washington DC.
Wang, X., Melesse, A.M., 2005. Evaluation of the SWAT model’s snowmelt hydrology in a northwestern Minnesota watershed. Trans. ASAE 48 (4), 1359–1376.
Wang, X., Yang, W., Melesse, A.M., 2008. Using hydrologic equivalent wetland concept within SWAT to estimate streamflow in watersheds with numerous wetlands. Trans. ASAE 51 (1), 55–72.
Water Security Agency, 2017, Available at https://www.wsask.ca/About-WSA/News-Releases/2017/February/Water-Security-Agency-Issues-Historic-Drainage-Approval/ Accessed 20 September 2017.
Water Survey of Canada, 2017. HYDAT Database. Available athttp://ec.gc.ca/rhc-wsc/default.asp?lang=En&n=9018B5EC-1Accessed 20 September 2017.
Wiens, L.H., 2001. A surface area-volume relationship for prairie wetlands in the upper Assiniboine River basin, Saskatchewan. Can. Water Resour. J. 26, 503–513.
Williams, J.R., 1995. Chapter 25: the epic model. In: Singh, V.P. (Ed.), Computer Models of Watershed Hydrology. Water Resources Publications.
Wilson, B.N., 2002. Good calibration and validation practices for hydrologic and water quality models. ASAE Paper No. 022294.
Yang, W., Wang, X., Liu, Y., 2010. Simulated environmental effects of wetland restoration scenarios in a typical Canadian prairie watershed. Wetl. Ecol. Manage. 18, 269–279.
