|dc.description.abstract||Saskatchewan’s surface and ground water sources are vital to life in the province, not only as the supply of safe drinking water for the residents, but also as a key driver of economic activity. The Qu'Appelle and Assiniboine River Basins are among the highly valued water resources in the province as they supply water for more than one-third of the population of Saskatchewan and contain a chain of eight lakes that are major recreational and economically valued resources in the region. The health of several watersheds within these highly valued river basins is being degraded by intensive agricultural and other developmental activities. The decision making processes for sustainable water management in these watersheds is stunted by limited observed field data. As a result, for Saskatchewan watersheds in general, and the Qu’Appelle and Assiniboine River Basins in particular, a better understanding is required of the type, extent and sources of pollutant loadings, and effects of potential alternative management practices may have to mitigate water quality problems.
Modeling approaches that have the capacity to analyze the quantity and quality of water resources, identify existing and potential watershed stressors, and the relative importance of best management options are therefore needed. With the intention of helping decision makers in the province, this thesis focuses on developing an eco-hydrological model, which is suitable for Canadian prairie watersheds and capable of simulating the long term effects of management practices. Following a review of several models, the Soil and Water Assessment Tool (SWAT) has been selected for this study.
In order achieve the objectives, the SWAT model has been modified to suit site specific characteristics of the Canadian prairies. The first such modification was to incorporate the numerous landscape depressions that vary in storage capacity into SWAT. This was done by representing depression storage heterogeneity using a probability distribution using an algorithm called “Probability Distributed Landscape Depressions (PDLD)”. The modified model, called SWAT-PDLD, was tested over two prairie watersheds: the Assiniboine and Moose Jaw watersheds. An improved simulation for streamflow was achieved for both case study watersheds as compared to the original SWAT lumped storage approach.
The other modification to SWAT was the incorporation of seasonally varying soil erodibility due to the cold climate conditions. This was done using a sediment module with a time variant soil erodibility factor that allows the value of soil erodibility to vary between seasons. The modified SWAT-PDLD along with seasonally varying soil erodibility was tested for sediment export simulation for the same two case study watersheds: the Assiniboine and Moose Jaw watersheds. Results show an improved sediment simulation for both case study watersheds when seasonally varying soil erodibility factors are considered as compared to the original SWAT model sediment module, which uses annual values of soil erodibility. The modified model was also used to simulate phosphorous and nitrogen export from the Assiniboine watershed and a satisfactorily model performance was obtained.
In addition, the developed model was used to assess the impacts of three different management practices on the export of pollutants for the Assiniboine watershed. The scenarios considered were conservation tillage, a cover crop, and filter strips. Model results show that both the filter strips and cover crops decreased sediment, phosphorous, and nitrogen export, while conservation tillage increased phosphorous export in the study watershed.
Finally, the study investigated the different sources of modeling uncertainty for the developed model. Parameter as well as precipitation, observed discharge, and model structure uncertainty of the SWAT-PDLD model was evaluated. Parameter uncertainty was quantified using three different techniques that include GLUE, ParaSol, and SUFI-2. Model structure uncertainty was assessed using a framework that combines the Bayesian Model Averaging (BMA) and Shuffled Complex Evolution (SCE). Results suggest that ignoring either input error or model structure uncertainty will lead to unrealistic model simulations and incorrect uncertainty bounds. The study also shows that prediction uncertainty bounds, posterior parameter distribution, and final parameter values vary between methods.||