A two-way calibration of the SWAT and OneLay/PolTra models using integrated modelling approach for the Lake Winnipeg Basin

Thumbnail Image
Booty, W.G.
Wong, I.W.S.
McCrimmon, R.C.
Leon, L.F.
Fong, P.
Richard, C.L.
Journal Title
Journal ISSN
Volume Title
Government of Canada, Environment Canada
Lake Winnipeg is Canada's sixth largest lake and the world's 10th largest freshwater lake with a surface area of 24,500 km2. The Lake experiences eutrophication and increasing algal blooms due likely to nutrient loadings from agricultural runoff. Lake Winnipeg receives surface water discharges from the Lake Winnipeg Basin which spans across 106 km2; including four provinces and four states. Major rivers that flow into Lake Winnipeg include the Saskatchewan, Winnipeg, Red and Dauphin Rivers. These four rivers account for more than 80% of the Lake inflow. The Winnipeg River has the largest inflow at 45% of the total Lake inflow but the Red River dominates nutrient loadings with 63% of the Lake’s phosphorus loading. An integrated modelling approach has been used to analyze and provide solutions to the Lake Winnipeg eutrophication problem. The Soil and Water Assessment Tool (SWAT) is the watershed runoff model being used to simulate sediment and nutrient loadings from the La Salle River watershed (2,400 km2), which will then be up scaled to represent the entire Red River watershed, and then these loadings will be used as input to the lake model, OneLay/PolTra (O/P), to simulate the Lake’s water quality. This integrated modelling approach has been found to provide a more flexible and efficient means to examine various remediation methods in the watershed. Only the Red River nutrient loads are being modelled due to its dominance of nutrient loads to the lake. The La Salle River empties into the Red River near Winnipeg, is 80% agricultural, and contributes approximately 1% of the nutrient load to the Red River watershed. The O/P model includes the 4 major inflowing rivers but only the Red River will vary during calibration and scenario analysis. The other 3 river loadings are based on observed data. Using artificial intelligence to determine the most optimal calibrations for loading provides for better calibration for SWAT and O/P through integrated modelling. A genetic algorithm is used to auto calibrate the integrated models. As a first attempt, model parameters being calibrated include those related to total suspended sediment (TSS) loadings from the SWAT model and in-lake suspended sediment concentrations from the O/P model. The objective functions used in the genetic algorithm is the normalized root mean square error (NRMSE). The integrated model calibration incorporates back calibration or two-way calibration to improve the overall objective.. Initial results from the two-way calibration of the integrated SWAT and O/P models show the NRMSE for the watershed and lake TSS are 0.18 and 0.12, respectively. With a Bayesian network, the uncertainty propagation of the integrated models is estimated to be 29.2%. This uncertainty is reasonable given the lack of precise input data, the simplicity of the Upscaler model from the LaSalle Watershed to Lake Winnipeg; and the significant spatial resolution change from the SWAT model to the O/P model. With the successful implementation of this integrated modelling approach, we plan to populate this approach to other watersheds in the Lake Winnipeg Basin and beyond where data is available.
19th International Congress on Modelling and Simulation, Perth, Australia, 12–16 December 2011 http://mssanz.org.au/modsim2011
Integrated modelling, two-way calibration, Lake Winnipeg