Wetlands provide a number of ecological goods and services, including wildlife habitat.  The loss of wetlands has been identified by the Prairie Habitat Joint Venture (PHJV) of the North American Waterfowl Management Plan as the single largest factor negatively impacting duck productivity in the Canadian prairie pothole region over the last 15 years. Accordingly, conservation agencies and non-government organizations have developed a number of policy and programmatic options to conserve these important habitats. Wetland drainage and degradation continues on the prairies, but the distribution of these activities is spatially heterogeneous. Therefore, to maximize the efficiency of conservation investments, it becomes imperative to understand where the risk of habitat loss is greatest. Thus, we used photogrammetrically derived wetland and drainage data for the Rural Municipality of Emerald in east-central Saskatchewan to develop a model that predicts the likelihood of basin drainage and degradation by agricultural activities. We hypothesized that variables such as topography, soil class, wetland size, distance to nearest stream, and market value of surrounding land would affect the likelihood of a given wetland being drained. We fit a series of generalized logit models and used information-theoretic techniques to select the best-approximating model. Likelihood of drainage was positively related to the proximity to an existing water conveyance. Similarly, drainage was more likely to occur in areas with low average slope. A number of soil attributes also helped predict wetland degradation. Cross-validation, using a one-out procedure for drainage, indicated that our model correctly classified ~73% of basins. We used the hypergeometric statistical distribution to contrast the performance of our model to the expected performance of conservation activities conducted in the absence of a predictive model. In an adaptive manner, we successively extended this test to 3 other landscapes in Saskatchewan, Manitoba, and Alberta, refining the model with each iteration.