Predictive mapping of wetland types and associated soils through digital elevation model analyses in the Canadian Prairie Pothole Region

Effective management strategies are needed to control phosphorus loading of prairie watersheds that contribute to the eutrophication issues of Lake Winnipeg. Prairie Pothole Region (PPR) wetlands provide many ecosystems services including reducing nutrient mobility. Preferential conservation of PPR wetlands with calcium carbonate (CaCO3)-enriched soils may be a more effective strategy for controlling phosphorus loading, as these soils have greater potential to retain phosphorus from agricultural runoff. The spatial distribution of CaCO3-enriched wetland soils is controlled by hydrologic processes that may be modellable using high-resolution digital elevation models (DEMs). Two modelling approaches were tested to map spatial distributions of wetlands and wetland soils expected to be enriched with CaCO3. The models were trained and tested with wetland salinity and soil profile information collected at three Saskatchewan PPR sites, near Swift Current, St. Denis, and Smith Creek. The first model was developed to approximate landscape-scale hydrologic processes from high-resolution DEMs to predict the distributions of fresh and solute-rich wetlands; the solute-rich wetlands represent wetlands expected to have CaCO3-enriched soils. Spill channel connections between wetlands were modelled to characterize wetlands in terms of the runoff contributions they receive, their potential for contributing runoff downslope, and their relative position within the landscape; solute-richness predictions were based on these characteristics. This model was successful and achieved acceptable predictive accuracies based on external validation tests. Digital soil mapping (DSM) methodologies were tested for predicting the spatial distribution of wetland soil classes within PPR landscapes. Target soil classes were defined by hydropedological units that reflect differences in soil CaCO3 enrichment. Multiple machine-learning techniques were tested, which incorporated many topographic attributes derived from the DEMs as predictor variables, including knowledge-based topographic attributes developed specifically to characterize the PPR’s morphology. Certain DSM models achieved acceptable predictive accuracy based on external validation tests and mapped soils in expected distributions, but none predicted the occurrence of wetlands with CaCO3-enriched soils distributed throughout their basins. Both modelling approaches could potentially be used to 1) identify wetlands with CaCO3-enriched soils to target for conservation efforts to maximize phosphorus retention and 2) create upscaled estimates of phosphorus retention across the PPR