Volume 56, Issue 12 e2020WR027984
Research Article
Free Access

Predicting Variable Contributing Areas, Hydrological Connectivity, and Solute Transport Pathways for a Canadian Prairie Basin

Diogo Costa

Corresponding Author

Diogo Costa

Environment and Climate Change Canada, Gatineau, Quebec, Canada

Department of Geography and Planning, University of Saskatchewan, Canada

Correspondence to:

D. Costa,

[email protected]

Search for more papers by this author
Kevin Shook

Kevin Shook

Department of Geography and Planning, University of Saskatchewan, Canada

Search for more papers by this author
Chris Spence

Chris Spence

Environment and Climate Change Canada, Gatineau, Quebec, Canada

Department of Geography and Planning, University of Saskatchewan, Canada

Search for more papers by this author
Jane Elliott

Jane Elliott

Environment and Climate Change Canada, Gatineau, Quebec, Canada

Department of Soil Science, University of Saskatchewan, Canada

Search for more papers by this author
Helen Baulch

Helen Baulch

School of Environment and Sustainability, University of Saskatchewan, Canada

Search for more papers by this author
Henry Wilson

Henry Wilson

Agriculture and Agri-Food Canada, Ottawa, Ontario, Canada

Search for more papers by this author
John W. Pomeroy

John W. Pomeroy

Department of Geography and Planning, University of Saskatchewan, Canada

Search for more papers by this author
First published: 30 October 2020
Citations: 18

Abstract

In cold agricultural regions, seasonal snowmelt over frozen soils provides the primary source of runoff and transports large nutrient loads downstream. The postglacial landscape of the Canadian Prairies and Northern Plains of the United States creates challenges for hydrological and water quality modeling. Here, the application of conventional hydrological models is problematic because of cold regions hydrological and chemical processes, the lack of fluvially eroded drainage systems, large noncontributing areas to streamflow and level topography. A new hydrodynamic model was developed to diagnose overland flow from snowmelt in this situation. The model was used to calculate the effect of variable contributing areas on (1) hydrological connectivity and the development of (2) tipping points in streamflow generation and (3) predominant chemical transport pathways. The agricultural Steppler Basin in Manitoba, Canada, was used to evaluate the model and diagnose snowmelt runoff. Relationships were established between contributing area and (1) snowmelt runoff intensity, (2) seasonal snowmelt volumes and duration, and (3) inundated, active and connected areas. Variations in the contributing area depended on terrain and snowmelt characteristics including wind redistribution of snow. Predictors of hydrological response and the size of the contributing area were developed which can be used in larger scale hydrological models of similar regions

Key Points

  • Seasonally frozen soils, level topography, and poorly defined basin drainage networks create challenges for modeling in prairie regions
  • Relationships were established between contributing area and (1) snowmelt intensity, volume, and duration and (2) inundated and connected area
  • A new model gives novel insights into the spatiotemporal patterns of overland flow, variable contributing areas, and transport pathways

1 Introduction

The control of overland flow and nutrient export from agricultural fields to rivers and lakes in Canada, such as Lake Winnipeg and Lake Erie, has been a major issue for many years, driven by concerns about increasingly frequent algae blooms. However, there are well-documented scientific challenges in understanding and modeling the complex interplay of different hydrochemical processes in cold regions, as well as their dependency on climate and land use (e.g., Baulch et al., 2019; Costa et al., 2020). Cold climate conditions have a strong impact on streamflow generation (Deelstra et al., 2009) and the transport of nutrients. In certain low-relief landscapes, such as the Canadian Prairies, Northern Great Plains of the United States, and Eurasian steppes, the hydrology and associated transport processes of cold regions are also unusual and complex, causing problems for conventional hydrological models (Costa et al., 2017; Pomeroy et al., 1998; Shook et al., 2013). These regions are dominated by snowmelt events, reduced baseflow, and extreme climatic and streamflow interseasonal and interannual variability (Fang & Pomeroy, 2007; Pomeroy et al., 2007). Snow and soil accumulate nutrients over most of the year, rapidly releasing them to overland flow during snowmelt and high-intensity rainfall-runoff events (Costa et al., 2020; Lilbaek, 2007; Lilbæk & Pomeroy, 2008; Pomeroy et al., 2005). The spring freshet has historically generated the largest annual overland flow events in these regions (Gray et al., 1970), transporting most of the nutrients exported annually (Baulch et al., 2019), although the dominance of snowmelt has been reduced due to climate change (Dumanski et al., 2015).

Landscape depressions can retain large fractions of these nutrients (Birgand et al., 2007; Crumpton & Isenhart, 1993; Neely & Baker, 1989). Erosion may be limited in early snowmelt due to frozen soils and low gradient (e.g., Cade-Menun et al., 2013; Granger et al., 1984). As a result, there is a high ratio of dissolved:particulate nutrients, particularly during early snowmelt (Cade-Menun et al., 2013; Tiessen et al., 2010). Overwinter freeze-thaw cycling can enhance soil erodibility, but this effect is generally only observed after the first spring thaw (Wall et al., 1988; Wilson et al., 2019).

The landscape and drainage basins of the Canadian Prairies and Northern Great Plains are characterized by a postglacial rather than fluvial geomorphology and low permeability glacial tills and/or seasonally frozen soils that restrict subsurface water flow (Granger et al., 1984; Hayashi & van der Kamp, 2000; Shook et al., 2013). These conditions cause most runoff to form as overland flow, which can accumulate in topographic depressions, sometimes called ponds, sloughs, lakes, or wetlands. These depressions sometimes connect and allow water and contaminants/nutrients to travel downstream to major rivers and lakes (Shook et al., 2013). While these interconnections increase as depressions fill and ponds expand, they also break as depressions empty and ponds retract, creating a nonlinear (collective) response of the wetland complex (Leibowitz et al., 2016). The contributing (drainage) areas of Canadian Prairie basins that have substantial depressional storage tends to be event dependent, impacted by the spatial arrangement of depressions in the basin and strongly affected by the existing water storage levels in the depressions (Fang et al., 2010; Gray, 1964; Pomeroy et al., 2010; Shaw et al., 2012; Shook et al., 2013; Stichling & Blackwell, 1957).

Understanding and modeling such dynamic behavior is extremely important to support flood and nutrient management in these areas but has remained a major scientific problem for decades. Conventional hydrological models assume basins with fixed (pre-determined) contributing areas (Shook et al., 2013), which is inadequate to capture the temporally variable contributing areas and hydrological regimes typically observed in prairie basins. The problem of variable contributing areas and their impact on streamflow generation and nutrient retention has long been recognized, and efforts to improve the representation of their dynamic nature in models have shown progress with the development of the spatially distributed Wetland Digital Elevation Ponding Model (WDPM) and the parameterized Pothole Cascade Model (PCM) (Shook & Pomeroy, 2011; Shook et al., 2013). These models have been successful at predicting inundation maps (WDPM) and streamflow based on hydrological response units (PCM; Pomeroy et al., 2014). However, neither WDPM (steady-state model) nor PCM (hydrological response unit, HRU, model) is physics based, limiting their use for more in-depth investigations of the transient hydraulic processes and hydrological response characteristics, causes, and tipping points that are critical to support management.

Herein, the FLUXOS hydrodynamic model (Costa, Burlando, Priadi, et al., 2016) was modified to address this need. The model, which was initially developed for rainfall-runoff flood events at river-reach scales, is based on the 2-D Shallow-Water Equations and was adapted for snowmelt and small basin-scale events. A Canadian Prairie basin was used as a test case. The main objective of this work is to provide insights into the incompletely understood runoff response of prairie basins to snowmelt, by identifying tipping points in hydrological response and predictors for the variable contributing area. Specifically, this paper examines the impact of variable contributing area on (1) hydrological connectivity and (2) the development of dominant chemical transport pathways. The methodology proposed is extendable to the entire northern prairie region of North America and other cold regions steppe environments, and the results inform the development of coarser resolution hydrological models in such environments.

2 Materials and Methods

2.1 The Original FLUXOS Model

FLUXOS was originally developed as a dynamic, spatially distributed river-aquifer finite-volume model. It was initially designed for river-reach hydrodynamic and water quality simulations (Costa, 2015; Costa, Burlando, & Liong, 2016; Costa, Burlando, Priadi, et al., 2016; Costa et al., 2016). The model solves the dynamic wave for surface water hydrodynamics, also known as the 2-D shallow-water partial differential equations (PDEs) derived from the depth integration of the Navier-Stokes PDEs—that accounts for inertial (local and convective), pressure, gravity, and friction forces (momentum balance) (Equation 1-3).

Accounting for inertial forces is an essential, innovative aspect of this model and model application. This is because, unlike with the commonly used kinematic or diffusion waves (Novak et al., 2018), it captures backflooding that is an important hydraulic phenomenon in low-relief terrains such as found in the Canadian Prairies (e.g., Cordeiro et al., 2017; Wilson et al., 2019). The original model also solves the 2-D surface and 3-D subsurface advection dispersion-reaction PDEs, in addition to the 3-D groundwater flow equations for saturated flow. The reader is referred to Costa, Burlando, Priadi, et al. (2016) and Shaad (2015) for model details about the original FLUXOS.

Hydrodynamics. The hydrodynamic model component is based on the 2-D (depth-averaged) shallow-water equations, which is a set of hyperbolic PDEs. Equations 1-3 describe how these PDEs are implemented in FLUXOS through a two-dimensional Cartesian (i.e., orthogonal) coordinate system.
urn:x-wiley:wrcr:media:wrcr24971:wrcr24971-math-0001(1)
urn:x-wiley:wrcr:media:wrcr24971:wrcr24971-math-0002(2)
urn:x-wiley:wrcr:media:wrcr24971:wrcr24971-math-0003(3)
where h is the water depth [L]; u and v are, respectively, the velocities in the x and y directions [L T−1]; ρ is the density of water [M L−3]; g is the gravitational acceleration [L T−2; = 9.81 m/s2]; zb is the bed surface level relative to a given datum [L] and τ refers to both the turbulent and bed shear stresses [M L−1 T−2].
The turbulent transfer of momentum by eddies giving rise to internal fluid friction/resistance is accounted for through dynamic calculation of local (i.e., each cell) turbulent Reynolds shear stresses (τxx, τyy, τxy). Turbulent stresses are obtained by augmenting the molecular/dynamic viscosity of water (vt = 1.793 mPa · s at 0°C to represent snowmelt conditions) with an eddy viscosity. The temperature dependence of the dynamic viscosity of water was ignored at this stage because of the focus on the snowmelt period, but it should be included for warmer season runoff calculations. The approach used in FLUXOS is based on typical eddy viscosity models that assume proportionality with the gradients of the mean velocity, as happens in Newtonian laminar flows (Costa, Burlando, Priadi, et al., 2016; Ruf, 2007; Shaad, 2015).
urn:x-wiley:wrcr:media:wrcr24971:wrcr24971-math-0004(4)
urn:x-wiley:wrcr:media:wrcr24971:wrcr24971-math-0005(5)
urn:x-wiley:wrcr:media:wrcr24971:wrcr24971-math-0006(6)

The bed shear stress is computed from the quadratic friction laws: urn:x-wiley:wrcr:media:wrcr24971:wrcr24971-math-0007 and urn:x-wiley:wrcr:media:wrcr24971:wrcr24971-math-0008, where cf is a friction coefficient.

FLUXOS uses the Roe approximate Riemen solver (linearization of the Jacobian) for the estimation of fluxes between computational elements to avoid the need for (even more) computationally expensive iterations to calculate the exact numerical solution (the Riemen problem). This approximation solves the Rieman problem using a linear hyperbolic system at each cell interface (Roe, 1981). The dynamic wave in variable wetting-drying domains (as found in flood simulations) can cause solutions to exhibit shocks, discontinuities, or large gradients. This is addressed using the shock-capturing MUSCL scheme (Monotonic Upwind Scheme for Conservation Laws; van Leer, 1979). The time step of the model is dynamic, changing at every numerical interaction to obey the Courant-Friedrichs-Lewy condition (CFL; Courant et al., 1967).

Water quality (reactive-transport). The surface water quality component solves the 2-D advection-dispersion-reaction PDEs (Equation 7):
urn:x-wiley:wrcr:media:wrcr24971:wrcr24971-math-0009(7)
where cs is the concentration of a given dissolved substance [M L−3]; E is the horizontal dispersivity [L2−1]; and S is a source term [M L−2 T−1] that is linked to the WINTRA module (see section 2.2). The E term accounts for the combined chemical mixing effect of different phenomena:
urn:x-wiley:wrcr:media:wrcr24971:wrcr24971-math-0010(8)
where Eturb is turbulent or eddy dispersion, which is a complex flow-driven phenomenon characterized by multifractality that is caused by internal and external friction forces (Sreenivasan, 1991); Etsd is the so-called Taylor shear dispersion that arises from unresolved vertical variations in horizontal flow; Esgt is an additional dispersion term that accounts for unresolved subgrid eddy viscosity and mixing; and Ed is the background molecular diffusion resulting from the probabilistic Brownian motion concept occurring at particle scales. In FLUXOS, E is approximated to the dominant turbulent dispersion based on the eddy viscosity concept: urn:x-wiley:wrcr:media:wrcr24971:wrcr24971-math-0011, where σ is the Prandtl-Schmidt number. Turbulent viscosity (νt) depends on shear velocity (u) and on a turbulent length scale ( urn:x-wiley:wrcr:media:wrcr24971:wrcr24971-math-0012). In this model, this relationship is approximated by an algebraic expression: νt ≈ kult, where k is a user-defined scaling factor to account for sub-grid-scale eddies (Costa, Burlando, Priadi, et al., 2016).

2.2 Modifications to the Model

FLUXOS was repurposed in this study to allow for (1) field- to small basin-scale simulations, (2) hydrodynamic modeling of snowmelt events and the associated transport processes (advection and dispersion), and (3) solute mass balances in soil water in addition to runoff. 2—The new version of the model was named as FLUXOS-OVERFLOW, and only uses the modules for the surface domain at this stage. In addition to the changes described below, the programming language was also converted from Fortran to C++ to allow for more efficient simulations and better memory allocation. The program uses the OpenMP framework to improve computational efficiency through shared-memory and multithreading (Dagum & Menon, 1998).

Subgrid terrain roughness and storage. Vegetation, other obstacles/obstructions, and terrain roughness can affect the hydraulic structure of overland flow and channel flow. While DEMs should capture many of these features at the macro scale (i.e., grid scale), smaller features at subgrid scales are often unaccounted for. This problem is addressed via absolute roughness height values prescribed by the user, which in essence, characterize the subgrid storage volume that is retained at each computational grid cell (caused by terrain roughness and depressions) before flow can develop.

WINTRA module. The WINTRA module was originally developed for the Cold Regions Hydrological Model (CRHM Pomeroy et al., 2007) to investigate edge-of-field nutrient export during snowmelt (Costa et al., 2019, 2017). WINTRA solves nutrient mass balances for both the snow and surficial soil, dynamically switching between these two nutrient sources throughout the snowmelt process based on snowcovered area depletion curves (Essery & Pomeroy, 2004). The reader is referred to Costa et al. (2017) for more details about WINTRA.

The module was adapted here for integration into FLUXOS-OVERFLOW. This was necessary because CRHM and the original WINTRA module are based on spatially variable HRUs, while FLUXOS-OVERFLOW uses structured grids. In FLUXOS-OVERFLOW, the WINTRA module was adapted to solve nutrient mass balances for each computational grid cell. As FLUXOS-OVERFLOW is fully distributed, the release of snow and soil nutrients to overland flow (and their depletion from the snow and soil) is calculated locally (i.e., grid cell) and dynamically based on the wetting-drying process, instead of relying on averaged snow cover depletion curves as in the original WINTRA version. The rate [1/T] of soil-to-overland flow nutrient release (of the type: Flux [M/T] = Mass [M] × Rate [1/T]) is parameterized similarly to the original version.

2.3 Model Application

Case study. The model was applied to the Steppler Watershed within the South Tobacco Creek (STC) basin in Manitoba (Figure 1). The STC basin has a drainage basin of 205 ha and is of particular interest to the study of nutrient export to major Canadian lakes because it is intensely farmed, subject to long and cold winters, and drains to the increasingly eutrophic Lake Winnipeg (Schindler et al., 2012) via South Tobacco Creek and the Red River (Figure 1a). As in most of western Canada, snowmelt is responsible for the largest annual overland flow and nutrient transport events. The characteristics of the Steppler basin used in this study are representative of an important watershed classification in the prairies (Wolfe et al., 2019)—the Southern Manitoba Prairie Region.

Details are in the caption following the image
(a) Lake Winnipeg Watershed and (b) DEM of the case study region, the Steppler Watershed.

This basin has been subject to an intense hydrological, water quality, and agricultural practice monitoring program by Agriculture and Agri-Food Canada (AAFC), Environment and Climate Change Canada (ECCC), Fisheries and Oceans Canada (DFO), and local landowners since 1999. The reader is referred to Tierney et al. (2001), Liu et al. (2014), Mahmood et al. (2017), Costa et al. (2017) and Costa et al. (2019) for more information about the basin, monitoring program, experimental setup, and some of the previous modeling efforts.

Model application and validation. A three-step validation process was used to evaluate the robustness of FLUXOS-OVERLAND for prairie regions. First, seven spring snowmelt events were simulated and compared with observations for the spring snowmelts of 2005 and 2009–2014. These simulations tested the ability of the model to predict streamflow for different snowmelt scenarios. Second, the well-established WDPM (Shook et al., 2013) was used to validate the inundation maps simulated. The WDPM redistributes water applied to a digital elevation model (DEM), following the algorithm of Shapiro and Westervelt (1994). The program does not perform hydrodynamic simulations like FLUXOS-OVERLAND; water is merely distributed downstream until it is ponded or exits the region. Therefore, the inundation maps produced by WDPM correspond to the maximum storage capacity of the basin for a given total runoff volume, which can only be compared to FLUXOS-OVERLAND's dynamic/transient maps when streamflow ceases in the simulations (in late snowmelt). Third, simulated inundation maps were compared to landscape-vegetation features obtained from satellite imagery.

A subset of three simulation years was then selected for further analysis: 2010 (a low-intensity event), 2011 (a medium-intensity event), and 2009 (a high-intensity event). The use of FLUXOS-OVERLAND in this prairie basin is relevant because backflooding and turbulent flow are important hydraulic phenomena that can be best captured using detailed DEMs where flow-obstacle interactions can be adequately resolved. However, fine spatial resolutions also inherently result in small model time steps (e.g., often second or subsecond) due to numerical stability requirements.

Unfortunately, the spatial and temporal resolution of existing hydrological models suitable for cold prairie regions such as CRHM (Pomeroy et al., 2007) and MESH (Pietroniro et al., 2007) that could be used to provide the necessary hydrological forcing, are too coarse for FLUXOS- OVERLAND. The spatially variable HRUs used in CRHM and the Grouped Response Units (GRUs) of MESH lack the spatiotemporal detail required for a meaningful validation of the model against field observations. Also, existing infiltration models such as the PrairieInfiltration module in CRHM based on the frozen soil infiltration routine of Gray et al. (2001) provide daily infiltration estimates only, which are inadequate for model validation purposes. For those reasons, an alternative strategy was needed to generate the fine-scale snowmelt input data needed to force FLUXOS-OVERLAND. Streamflow observations at the basin outlet (at 15-min intervals) were used as a proxy for the snowmelt that did not infiltrate, thus activating the surface hydrodynamics. This snowmelt proxy-data were forced dynamically and homogeneously into FLUXOS-OVERLAND.

After the model performance was evaluated via this three-step validation process, the model was used diagnostically to address the specific research questions of this research related to the impact of variable contributing areas on the hydrological response of prairie basins. Here, after the snowpack was generated using CRHM's PBSM blowing snow model, energy balance snowmelt simulations based on CRHM's SnobalCRHM module were coupled with infiltration estimates using CRHM's PrairieInfiltration module to generate snowmelt scenarios to run in FLUXOS-OVERLAND. The problem with the difference in temporal resolution between FLUXOS-OVERLAND and CRHM was less critical here because the model results were not used to look at temporal evolutions, but rather cumulative responses. These simulations were used to characterize the hydrological response of the Steppler basin to different snowmelt events (further details are provided in section 2.4).

The initial surface water storage conditions (IC) prior to each of the snowmelt events simulated were unknown and could not be generated through multiyear runs due to the computational cost of FLUXOS-OVERLAND (see section 4); thus, multiple scenarios had to be considered. The IC account for (1) antecedent (i.e., premelt) frozen or unfrozen water stored in the landscape that can be mobilized with snowmelt and (2) subgrid terrain roughness and depressions that may hold additional water. The range of IC values considered varied between 1 and 10 mm of water uniformly distributed across the basin, which corresponds to 2,104 and 21,044 m3 of water and represents between approximately 10% and 100% of the maximum depressional storage.

Model performance. Model performance was evaluated through comparisons between (1) observed and simulated streamflow at the basin outlet (section 3.1) and (2) simulated inundation maps and landscape-vegetation features obtained from satellite imagery (section 3.3). Streamflow simulations were verified using three objective functions: the Nash-Sutcliffe Efficiency coefficient (NSE), root-mean-square Error (RMSE), and model bias (MB) computed using observations and model outputs at hourly or subhourly resolution depending on the available observation data frequency.

2.4 Postprocessing Results: Shedding Light on Connectivity

Hydrological and hydrochemical connectivity. Understanding the impact that different snowmelt characteristics (e.g., peak melt rate, melt duration, snowmelt volume, and peak runoff) have on the basin's hydrochemical response within depression and wetland complexes is a subject of great interest to prairie basin water management. Since FLUXOS-OVERFLOW calculates the basin's 2-D hydrodynamics, it offers an opportunity for quantifying hydraulic and hydrochemical connectivity in a way that takes into account both spatial and temporal patterns of storage and fluxes. However, hydrological connectivity and hydrochemical connectivity should be treated differently because they are affected by different processes (Ali et al., 2018). While hydrological connectivity is related to the transfer of water across the landscape (Pringle, 2003), hydrochemical connectivity is related to chemical mobility between water bodies (Likens & Bormann, 1995), which requires considerations of soil-runoff interactions.

The following terms were used to support the analysis of hydrological connectivity. ACTIVE AREA comprises the fraction of the basin that produces surface runoff, which may or may not reach the outlet (Bracken & Croke, 2007). Hydrological connectivity among areas occurs when the magnitude of transmission of water among active areas is larger than the magnitude of losses as water moves from these areas at the frequency of observation (Ali et al., 2018), or in this case, the time step of the model. When the transmission of water exceeds losses, water can fill and spill (Shaw et al., 2012) and fill and merge (Leibowitz et al., 2016) among depressions within the catchment. This can form a CONNECTED AREA that establishes a contiguous path to the basin outlet (Shook & Pomeroy, 2011) within a model time step. This connected area includes a CONTRIBUTING AREA that is the runoff-generating area that provides enough water to overcome losses such that runoff manifests at the basin outlet at the time step of the model. The temporal dynamics of each of these areas can be important to characterize the hydrological response of basins to different snowmelt events. This was captured using time-cumulative values for each of the areas mentioned above (km2·hours). Chemical connectivity was quantified by the WINTRA module. A typical first-order nutrient transfer function was used to describe soil-to-runoff nutrient release. The results were used to provide insights into the basin's predominant chemical transport pathways that can inform the evaluation of BMPs in the basin.

Identifying predictors of streamflow generation and CONNECTED and CONTRIBUTING AREAs. The simulations used for validation of FLUXOS-OVERLAND are insufficient to provide a hydraulic response characterization that is wide ranging in magnitude, duration, and initial storage conditions of snowmelt events. For this reason, a CRHM model previously developed for this basin in Mahmood et al. (2017) and (Costa et al., 20172019) was used. Individual snowmelt events were identified from a continuous simulation of the 2005–2011 period, which has been previously validated against observations in those studies. This CRHM model used the SnobalCRHM and PrairieInfiltration modules for snowmelt and infiltration calculations, respectively. The snowmelt events selected covered a range of snowmelt conditions that are specific to this basin and varied in intensity (i.e., peak flow rates), duration and total runoff volume. The simulation results with FLUXOS-OVERLAND were used to develop power law models for the prediction of CONNECTED AREA and CONTRIBUTING AREA based on relevant (and measurable) characteristics of the snowmelt events.

3 Results

3.1 Model Validation: Flow and Inundation Maps

Channel flow (Step 1 of three-step validation process). Results show that the model can route the snowmelt runoff inputs in a way that reproduces much of the channel flow dynamics during snowmelt events of differing magnitudes (black line in Figure 2), including both the ascending and recessing limbs of multiple/compounded snowmelt hydrographs and three-step validation process (see section 2.3). However, underestimation of large channel flow peaks such as in 2009 and underestimation of low flows such as in 2010 can be observed. These may be caused by (1) the simplified snowmelt-spatiotemporal forcing used (that was a necessary simplification, see section 2.3), (2) IC and its spatial distribution in the landscape, (3) absence of a baseflow component, (4) accuracy of the DEM, and (5) presence of culverts and other hydraulic structures. Also, the roughness height parameter used to account for the unresolved subgrid water storage potential may not be static, nor spatially homogenous, as was assumed in these simulations. Snow dams in channels or icing in culverts, which are not represented in the model, may temporarily retain water until a more episodic discharge event can break the blockage.

Details are in the caption following the image
Comparison between observed and simulated streamflow for the spring snowmelt events of 2005 and 2009–2014 (a, c, e, g, h, i, and j), and peak inundation maps for three snowmelt events (2009, 2010, and 2011) representing different intensities (b, d, and f).

These uncertainties are important and need to be recognized when examining the detailed hydraulic features simulated and presented in the subsequent sections. Also, the model results indicate moderate sensitivity to the range of initial (premelt) water storage conditions considered (ICs, gray lines): 1 to 10 mm, that is, about 10% and 100% of the total depressional storage of the basin. It shows that this particular basin (unlike many other Prairie basins; Ehsanzadeh et al., 2012) can generate streamflow even during small snowmelt events (e.g., 2010). This is not necessarily the case for all Canadian Prairie basins. There is a high degree of variation in key features such as depressions across the prairies, from basins completely controlled by depressional storage (such as St. Denis, Saskatchewan, Canada) to those less dominated, such as Steppler (Wolfe et al., 2019).

Inundation maps (Steps 2–3 of the three-step validation process). Figure 3 shows the inundated maps simulated by FLUXOS-OVERLAND at three stages throughout the 2009 and 2011 snowmelt events: early, peak, and late snowmelt. These years were selected because they had a similar total snowmelt runoff volume of about 300 mm (see panel upper left of Figure 7d) but based on very different hydrographs. A parallel simulation with WDPM was performed using a similar forcing. Recall that the WDPM does not perform hydrodynamic simulations like FLUXOS-OVERLAND; water is merely distributed downstream, until it is ponded or exits the model domain. Therefore, WDPM's results correspond to the maximum storage capacity of the basin and can only be compared with FLUXOS-OVERLAND's dynamic inundation maps after streamflow has terminated at a later snowmelt stage (Step 2 of the three-step validation process, see section 2.3). Results show that water distribution is limited by water availability in the early stages of the melt. In the middle stage, near the snowmelt peak, the water distribution is affected by the flow dynamics, as it flows overland. At the end of the melt, the water is primarily ponded, as is shown by the very good agreement with the WDPM simulation.

Details are in the caption following the image
Comparing inundation maps simulated by WDPM and FLUXOS-OVERLAND. The panels in the lower row are comparable as WDPM's results correspond to the final state only those of FLUXOS-OVERLAND's are for late snowmelt. The panels in the middle and lower rows correspond to FLUXOS-OVERLAND simulation results at the snowmelt peak and early snowmelt period, respectively. The base map depicts topographic features.

Figure 4 overlays the simulated inundated areas (with a cumulative ACTIVE hours snowmelt runoff intensity than 10) on satellite imagery (Step 3 of the three-step validation process, see section 2.3). The simulation used is based on the 2009 flood event, for demonstration purposes, and examines the model's ability to capture the spatial patterns of surface flow through comparison with landscape and vegetation features indicative of major (and more prolonged) water interactions. Results show that the model captures the impact of landscape depressions (Views A and G), streams and riparian zones (Views A, B, and E), and vegetation (Views A, C, D, and F) on the spatial distribution of water storage and surface flow.

Details are in the caption following the image
(a–h) Comparison between landscape-vegetation features from satellite imagery (USDA FSA, GeoEye, Maxar) and the simulated inundated region (transparent white dotted areas) with ACTIVE hours snowmelt runoff intensity than 10 for the 2009 flood event.

3.2 Overland Flow and Velocity Fields

The simulated median velocities (solid red line) tend to oscillate around 0.05 m/s regardless of the intensity or duration of the snowmelt event (Figure 5), but the amplitude of the deviations from this average value increase with increasing snowmelt runoff intensity (i.e., 2009). The mean velocity values (dashed black line) tend to be smaller than the median values due to the presence of standing water in landscape depressions. Both the mean and median flow values are strongly affected by the snowmelt runoff intensity, with magnitudes halving between greater intensity (i.e., 2009) and smaller-intensity snowmelt events (i.e., 2010 or 2011).

Details are in the caption following the image
Simulated mean and median flow/runoff velocities (left axis) and runoff/flow (right axis) for three snowmelt events of different intensity (2009, 2010, and 2011).

As would be expected, the velocities tend to be greater along river channels and ditches than in the floodplains or landscape depressions (Figure 6). Small depressions in the terrain may connect and disconnect depending on the intensity of the snowmelt event, causing discontinuities in streamflow discharge (i.e., the gate-keeping effect Phillips et al., 2011) resulting in a network of preferential flow paths branches from the main channel for all the events simulated. The areal extent of this network and its ability to hydraulically connect the upstream, midstream and downstream regions of the basin (i.e., CONNECTED AREAS) depend on the intensity of the snowmelt event (greatest in 2009 and smallest in 2010). For example, the upstream and middle stream areas were well connected to the outlet during the 2009 spring snowmelt event but remained largely disconnected in the smaller-intensity events of 2010 and 2011 (see Figures 7a–7c).

Details are in the caption following the image
Computed velocity fields at the snowmelt runoff peaks of the snowmelt events of (a) 2009, (b) 2010, and (c) 2011. Panels (d)–(g) show in more detail the 2009 flow velocity distribution at the (d) lower, lower middle, upper lower, and upper reaches.
Details are in the caption following the image
(a–c) ACTIVE hours across the basin for the different snowmelt events; and (d) collective response of the basin: snowmelt runoff input (left upper panel), ACTIVE AREA and CONTRIBUTING AREA (left middle panel), and time-cumulative ACTIVE AREA and CONTRIBUTING AREA in km2-hours (left lower panel). Panel (d), right column, compares both the mean CONTRIBUTING AREA (upper-right-d panel) and the time-cumulative CONTRIBUTING AREA (lower right d panel) against the corresponding peak snowmelt runoff rates, total snowmelt runoff volumes, and total snowmelt duration for the three snowmelt events simulated.

3.3 Hydrological and Hydrochemical Connectivity

Hydrology, contributing areas, and contributing time. On the one hand, results show that the lower reaches of the basin actively contributed to the outlet in all snowmelt events (green regions in Figures 7a–7c), despite the varying intensities of the snowmelt peak rate (2009 > 2011 > 2010) affecting the extent of the inundated areas and duration of flooding. On the other hand, the contribution of the middle and upstream reaches strongly depended on the overall intensity of the snowmelt event and its influence on connectivity and contributing area, with the low-intensity 2010 event, for example, resulting in a near-complete hydrological disconnect between the upper and middle reaches (see green areas in panel Figure 7b).

Results indicate that water tends to move through the basin via second- to fourth-order streams, which discharge to the main river channel at hot spot locations—this becomes evident by looking at the stream network that emerges in green (i.e., CONNECTED AREAS in Figures 7a–7c) that shows many small reaches jutting out from the main channel. This is apparent for all snowmelt events simulated, despite that noncontributing regions (in brown) tend to increase with decreasing snowmelt runoff intensity (see the 2010 panel). Many upslope areas surrounding the lower reaches of the basin and vicinity of the main river channel only connected temporarily to the outlet.

At the basin scale (Figure 7b), snowmelt runoff intensity affected the ACTIVE AREA and CONTRIBUTING AREA of the basin (left middle, Figure 7d). However, the duration of the event plays a critical role in the time-cumulative CONTRIBUTING AREA (left lower, Figure 7d)—the 2011 event shows the greatest time-cumulative CONTRIBUTING AREA despite that the highest snowmelt runoff intensity was observed in 2009. The mean CONTRIBUTING AREA increases with the peak snowmelt rate and total snowmelt volume (right upper, Figure 7d), but the time-cumulative CONTRIBUTING AREA (right lower, Figure 7d) depends not only on the total snowmelt volume but is also affected by the duration of the snowmelt event (although nonlinearly—note that the relationship is nonmonotonic). Both the mean CONTRIBUTING AREA (right upper, Figure 7d) and time-cumulative CONTRIBUTING AREA (right lower, Figure 7d) are close to 0 for the 2010 event. These results suggest the presence of tipping points for streamflow generation that are site specific and depend on gate-keeping effects that create discontinuities in connectivity between the different landscape depressions and channels (see section 4.1 for further discussions).

Hydrochemistry and predominant chemical transport pathways. As expected, the greatest hydrochemical connectivity is observed in the main river channel where predominant transport pathways develop (Figures 8a–8c). However, similarly to the hydraulic patterns, the results reveal the development of preferential chemical transport corridors (see Figures 6a–6c). Interestingly, these transport pathways tend to connect to the main channel at hot spot locations, suggesting that large portions of the riparian zone are bypassed or activated unevenly. This phenomenon can be observed for all the snowmelt events simulated but appears to become more pronounced for low-intensity events.

Details are in the caption following the image
(a–c) Biogeochemical connectivity (spatial distribution) as inferred from the predominant transport pathways for snowmelt events of different intensity. Biogeochemical connectivity was calculated based on a first-order soil-to-runoff nutrient transfer function in order to capture the likelihood of soil release and transport by runoff; (d) basin-lumped results: snowmelt runoff (d, upper left), hydrochemical connectivity (d, lower left), and relationship between hydrochemical connectivity, snowmelt duration and total volume (d, right).

Contrary to what was observed for hydrology, the snowmelt duration does not exert an important control on hydrochemical connectivity (d, left down). Instead, snowmelt runoff intensity (i.e., peak snowmelt rate) played the dominant role. This is also evident in Figure 8d (right), which shows that the duration of the snowmelt is not a good predictor for hydrochemical connectivity in this basin. A stronger correlation between hydrochemical connectivity and the total snowmelt runoff volume can be observed.

3.4 Identifying Predictors of Streamflow Generation and CONNECTED and CONTRIBUTING AREAs

Step 1: Running multiple snowmelt events with CRHM. Figure 9 shows the range of simulated snowmelt events with very distinct characteristics (black circles): snowmelt duration, total snowmelt volume, and peak snowmelt runoff. The subset of events selected for subsequent simulation with FLUXOS-OVERFLOW (“red” dots) was intended to represent the range of possible snowmelt scenarios in a way that could support the development of hydraulic response curves.

Details are in the caption following the image
Simulated snowmelt events using CRHM (black circles) and (2) subset of events simulated with FLUXOS-OVERLAND (red dots).

Step 2: Developing hydraulic response curves for CONNECTED AREA and CONTRIBUTING AREA. Figure 10 shows that, on the one hand, the mean CONNECTED AREA (panel a) appears to be inversely related to the total duration (“circles” in all panels) of the snowmelt event (see how the greater CONNECTED AREA in orange/yellow tends to be accompanied by smaller snowmelt durations that are represented by the white dots). That is likely because snowmelt events extending over more extended periods may be associated with low snowmelt intensities, more opportunity time for infiltration to frozen soils, greater evaporative losses and the slower flowing of water through channels, which are less likely to exceed their natural flow carrying capacity and cause flooding. On the other hand, high peak snowmelt rates (y axis of all panels) and high initial depressional storage (Sdinit) have a clear, direct positive effect on the mean CONNECTED AREA (Figure 10a), causing larger areas to flood.

Details are in the caption following the image
(a) mean CONNECTED AREA and (b) maximum CONTRIBUTING AREA as a function of initial average depressional storage (Sdinit, different panels), snowmelt runoff peak rate (y axis), total snowmelt runoff volume (x axis), and total duration of the snowmelt event (grayscale circles).

The total snowmelt overland flow depth (x axis of all panels) appears to have a small impact on the mean CONNECTED AREA, likely because snowmelt overland flow peaks (y axis of all panels) are the dominant factor causing overbank flow that increases the aerial extent of flooding, and, therefore, its impact on the CONNECTED AREA is not a direct one. The patterns observed for the maximum CONTRIBUTING AREA are very similar (Figure 10b), but their magnitude is much reduced. However, in this case, the initial depressional storage (as indicated by the shade of the points in all panels) does not appear to strongly affect the results.

Step 3: Developing hydraulic response curves for CONNECTED AREA and CONTRIBUTING AREA. Figure 11 indicates that both the duration (point shade) and total snowmelt overland flow volume (x axis of all panels) impact the time-cumulative CONNECTED AREA (Figure 11a). However, similar to what was observed in Figure 9b, the duration of the events (point shade) did not have a visible effect on the time-cumulative CONTRIBUTING AREA (Figure 11b). The impact of both the peak snowmelt overland flow rate (y axis of all panels) and the initial storage conditions (each panel corresponds to different ICs) was small for both CONNECTED and CONTRIBUTING AREAS.

Details are in the caption following the image
(a) time-cumulative CONNECTED AREA (b) time-cumulative CONTRIBUTING AREA as a function of Sdinit (initial average depressionl storage) and snowmelt peak rates and total volume. Information about the duration of the snowmelt event is also provided (circles).

Step 4: Developing predictive models for CONNECTED AREA and CONTRIBUTING AREA. Based on the dominant factors identified previously, power law relationships were developed for prediction of CONNECTED AREA and CONTRIBUTING AREA (Figure 12). All possible parameter combinations were tested, but only the best performing cases are presented. The power law model fit was determined with a 95% confidence interval.

Details are in the caption following the image
Site-specific prediction models for (a, left) mean CONNECTED AREA, (a, right) time-cumulative CONNECTED AREA, and (b) maximum CONTRIBUTING AREA based on snowmelt runoff peaks (Qp), snowmelt runoff volume (Vt), snowmelt runoff duration (Dm), and snowmelt duration (Dm).

The plots show that both the CONNECTED AREA and CONTRIBUTING AREA (panels a, left, and b, left and right, respectively) can be predicted well by combining information about the snowmelt event. The CONNECTED AREA can be estimated based on the snowmelt overland flow peak and duration, as well as the initial (premelt) depressional storage conditions (a, left). The initial depressional storage conditions have a critical impact on the range of the CONNECTED AREAs that can be expected (see the spread of the different markers in Figure 12a, left). Interestingly, mean CONNECTED AREAs of similar magnitude can be achieved by snowmelt events with very different characteristics - notice how different markers overlap (Figure 12a, left). In the case of the time-cumulative CONNECTED AREA (Figure 12a, right), both the duration of the snowmelt event and the initial depressional storage have been identified as the key predictors. The prediction of maximum CONTRIBUTING AREA (Figure 12b, left and right) was more successful using snowmelt overland flow peaks (Qp) and snowmelt overland flow volumes (Vt) (b, right). However, the fitted model seems to be less accurate for large Qp·Vt values (b, right) and did not perform well when only Qp was considered (b, left). All other parameter combinations were tested but showed poorer prediction capability.

4 Discussion

4.1 Hydrological Response and Connectivity

Basin hydraulic flow structures and key flow controls: from microponding to large-scale wetland-complex interactions. The Canadian prairies are a rather diverse region. There is a high degree of variation in key features such as depressions, from basins completely controlled by depressional storage (such as St. Denis, Saskatchewan, Canada) to those less dominated, such as Steppler (Wolfe et al., 2019). The characteristics of the Steppler basin used in this study are representative of an important basin classification in the Canadian prairies—the Southern Manitoba Prairie Region (Wolfe et al., 2019). The Steppler basin is also within the basin of Lake Winnipeg, which is becoming increasingly eutrophic (Schindler et al., 2012), hence making the contributions of this research particularly relevant to understand nutrient export dynamics and hydrological connectivity in this region.

This study has shown that hydrodynamic modeling using FLUXOS-OVERFLOW can shed light on this problem by providing a detailed hydraulic characterization of mildtopography basins that explicitly accounts for both small (i.e., microponding) and large (i.e., fill-spill between depressions) scale flow-transport processes. This is possible because the model solves the 2-D dynamic wave that, unlike the commonly used kinematic or diffusion waves (Novak et al., 2018), can captures backflooding that is an important hydraulic phenomenon in low-relief terrains such as found in the Canadian Prairies (e.g., Cordeiro et al., 2017; Wilson et al., 2019). The Steppler Basin model results indicated that the magnitude of the snowmelt peak (2009 > 2011 > 2010) strongly affected the inundated area (Figure 7), which was to be expected as it directly relates to the natural water carrying capacity of the channels. It also showed that water tends to be discharged onto the main river channel at hotspot locations. The presence of “micro” ponding, affecting the hydrological connectivity of the basin, particularly in the upstream areas of this basin (see Figures 7a–7c), was identified as a leading cause for this phenomenon, an idea that has been supported by previous models (e.g., Shook & Pomeroy, 2011) and field observations in Prairie regions (e.g., Shaw et al., 2012; Wilson et al., 2019). These small ponds can quickly evaporate or infiltrate as soils thaw. The results also indicate that both microponding and large-scale wetland-complex interactions cause tipping points in streamflow generation—this is discussed in more detail below.

CONNECTED AREAs and CONTRIBUTING AREAs are dynamic: Can their spatiotemporal patterns be predicted and related to streamflow generation? The hydrology, hydraulics and associated transport processes of the Canadian Prairies cause major difficulties for conventional hydrological models (Costa et al., 2017; Shook et al., 2013). The impact of glaciation on geomorphology causes surface runoff drainage to accumulate in small depressions and wetlands that only connect episodically, acting as switches to the transport of water and contaminants/nutrients to major downstream rivers and lakes (Shook et al., 2013).

The FLUXOS-OVERLAND model was used to address this problem through the development of basin-specific hydraulic response curves that allowed determining simple parametric models for prediction of CONNECTED AREA and CONTRIBUTING AREA (Figure 12). These simple models can be used as model inputs to conventional hydrological modeling that does not account for the impact of variable contributing areas, to support WRM and can be useful to practitioners, managers, and scientists.

Results indicated that the instantaneous CONNECTED AREA and CONTRIBUTING AREA of the basin varied throughout the snowmelt event (see Figure 7d, left middle), which is in agreement with field studies that have emphasized the dynamic (and event dependent) nature of this phenomenon (e.g., Ehsanzadeh et al., 2012). As expected, they depended strongly on the instantaneous snowmelt rate (Figure 7d, left upper). However, when examining the process considering the effect over the entire duration of the snowmelt event, which was captured via the time-cumulative CONNECTED AREA and time-cumulative CONTRIBUTING AREA, the duration of the event emerged as a key factor. It was concluded that the non-static (time and space) nature of the basins' CONNECTED AREA and CONTRIBUTING AREA with such landforms are event dependent and are influenced not only by how the wetlands are arranged in the landscape and on their initial water storage (Gray, 1964; Pomeroy et al., 2010; Shaw et al., 2012; Shook et al., 2013; Stichling & Blackwell, 1957) but also on the spatiotemporal characteristics of the snowmelt event (Leibowitz et al., 2016).

Can tipping points of streamflow generation and hydrological and hydrochemical connectivity be identified? Canadian Prairie hydrology is dominated by snowmelt overland flow over frozen soils (Fang & Pomeroy, 2007; Pomeroy et al., 2007), which transports most of the nutrients exported annually to major rivers and lakes (Baulch et al., 2019; Corriveau et al., 2013; Costa et al., 2020; Gaynor & Bissonnette, 1992; Tiessen et al., 2010). However, the arrangement of depressions in the basin has a strong impact on (the timing and magnitude of) streamflow development (Leibowitz et al., 2016). Therefore, understanding the conditions and tipping points of streamflow generation and hydrological and hydrochemical connectivity has become a topic of significant interest to both scientists and practitioners.

The model allows the identification of some of these tipping points for the Steppler basin. For example, while an initial increase in the mean CONTRIBUTING AREA from 0 to 0.2 km2 produced a minimal increase in runoff peak flow (or total runoff volume) of 10 mm/h, a subsequent area increase of 0.10 km2 (to a total of 0.3 km3) caused the runoff peak flow (or total runoff volume) to increase by over tenfold (Figure 12b, right). Hydrochemical connectivity exhibited a very different pattern; it increased rapidly at the start of all snowmelt events regardless of their intensity (see Figure 8). This is because soil nutrients tend to be quickly mobilized with runoff during the early stages of snowmelt, depleting the soils (Baulch et al., 2019). Finally, we demonstrated that time-cumulative CONTRIBUTING AREAs, rather than mean or peak CONTRIBUTING AREAs, characterize better the basin's hydrological response because it accounts for the dynamic nature and total duration of the runoff generation process (see Figure 7).

4.2 Hydrochemical Response and Connectivity: Predominant Transport Pathways Form Naturally and can Impact the Efficacy of BMPs

The simulations indicate that chemicals are transported through the basin primarily along preferential transport pathways that emerge within floodplains and riparian zones (Figures 6 and 8). Snowmelt overland flow is discharged into the main river channel at specific locations bypassing large portions of the stream bank and hence the riparian zone or buffer strip. This may compromise the efficacy of BMPs as riparian zones may be activated unevenly, overloaded at particular hot spot locations and underutilized at other locations. While this problem has been widely recognized in the literature (Banaszuk et al., 2013; Baulch et al., 2019; Knight et al., 2010; Stewart et al., 2010), this model application demonstrated that FLUXOS-OVERFLOW can help to identify these critical regions and portions of the riparian zone in order to target conservation efforts. The model results indicated that hydrochemical connectivity is strongly affected by the intensity of the snowmelt event (greater in 2009), a problem that has been identified in the literature (Baulch et al., 2019; Satchithanantham et al., 2019). Typically, nutrients accumulated in snowpacks and soils are rapidly released to snowmelt overland flow and high-intensity rainfall-runoff events (Baulch et al., 2019; Pomeroy et al., 2005). The spring freshet is known to transport most of the nutrients exported annually (Baulch et al., 2019), but landscape depressions tend to retain a large portion of these nutrients (Birgand et al., 2007; Crumpton & Isenhart, 1993; Neely & Baker, 1989). The physics-based and fully-distributed nature of FLUXOS-OVERFLOW shows great potential for explicitly simulating these processes, and diagnosing key basin-specific controls that can be used to inform better design of BMPs.

5 Conclusions

The incompletely understood runoff response of Canadian Prairie basins to snowmelt over frozen soils has been studied using a hydrodynamic model purposely developed to address this problem. This research focussed on the impact of variable contributing areas on (1) hydrological and hydrochemical connectivity, and the development of (2) tipping points in streamflow generation and (3) predominant chemical transport pathways. The model addresses well-known issues with the application of conventional hydrological models in this region: (1) lack of well developed, fluvially eroded drainage systems and the preponderance of level topography in Prairie basins are addressed by solving the 2D Saint-Venant Equations that do not require the delineation of rivers and water bodies, and (2) slow-moving waters and backwater effects due to low-relief topography are addressed by accounting for inertial (local and convective), pressure, gravity, and friction forces in the momentum balance. The robustness of the model for simulation of dynamic inundation maps and streamflow in these regions was evaluated against observations using a three-step validation process.

The results indicated that the magnitude of the snowmelt peak strongly affected the inundated area, which was to be expected as it directly relates to the natural water carrying capacity of the channels. It also showed that water and chemicals tend to be discharged onto the main river channel at hotspot locations via preferential transport pathways that bypass large portions of the stream bank, riparian zone, or buffer strip—potentially compromising the efficacy of BMPs. It was concluded that the dynamic nature of the CONNECTED AREA and CONTRIBUTING AREA of basins with such landforms are event-dependent and affected by (1) how the wetlands are arranged in the landscape, (2) their initial water storage, and (3) the spatiotemporal characteristics of the snowmelt event. Basin-specific hydraulic response curves were computed for the Steppler basin to develop simple parametric models for prediction of CONNECTED AREA and CONTRIBUTING AREA that can be used in conventional models and support management. It allowed to identify tipping points and predictors of hydrological response. For example, while an initial increase in the mean CONTRIBUTING AREA from 0 to 0.2 km2 produced a minimal increase in runoff peak flow (or total runoff volume) of 10 mm/h, a subsequent 0.10 km2 area increment (to a total of 0.3 km3) caused a tenfold rise in runoff peak flow (or total runoff volume) as the hydrological connectivity of the basin increased. Time-cumulative CONTRIBUTING AREAs, rather than mean or peak CONTRIBUTING AREAs, are more representative of the highly dynamic runoff generation process ocurring in this basin. The methodology proposed is extendable to the entire Canadian Prairies and similar regions.

Terminology

  • Active area (km2): area of the basin that produces surface runoff, which may or may not reach the outlet;
  • Connected area (km2): area of the basin that establishes a contiguous path to the basin outlet, which may or may not be flowing; and
  • Connected area (km2): runoff-generating area that provides enough water to overcome losses such that runoff manifests at the basin outlet at the time step of the model.

Acknowledgments

The authors would like to thank Agriculture and Agri-Food Canada and Environment and Climate Change Canada for kindly providing the data used in this study. The research would not have been possible without the interest and cooperation of landowners in the South Tobacco Creek Basin and the Deerwood Soil and Water Management Association. The model runs presented in this study were performed using the Graham cluster of Compute Canada and the Plato cluster from the University of Saskatchewan. This research was supported by the Global Water Futures Program, NSERC Discovery Grants, and the Canada Excellence Research Chair in Water Security.

    Data Availability Statement

    Data are available through Tiessen et al. (2010), Mahmood et al. (2017), Costa et al. (2017), and Costa et al. (2019). These papers contain information about the basin, monitoring program, data collected, experimental setups, and some of the previous modeling efforts.