全球变化知识资源中心

Potential impacts of global climate change on the amount of low-marsh habitat in coastal wetlands of the Great Lakes are unknown, which may have important implications for the Great Lakes fish community that use such habitat. We developed a generalized linear model that uses only hydrogeomorphic (HGM) features and lake elevations to predict the extent of low marsh in coastal wetlands of eastern and northern Georgian Bay. The McMaster Coastal Wetland Inventory was used as a reference dataset to train the model, while best available data were assembled to create a digital elevation model that was used to derive all HGM features at a lake elevation of 176.17m (International Great Lakes Datum 1985). The best predictive model included depth, slope, and exposure as HGM variables, yielding an area under the curve (AUC) score of 0.83. We classified the model output into low-marsh and open-water habitat using a threshold value identified by maximizing the true skill statistic. The classified model output had sensitivity and specificity scores of 0.80 and 0.75, respectively, and correctly identified 81% of the low-marsh units present in the reference dataset with an average 60% areal overlap between the model prediction and reference dataset. We applied the model to two external datasets to check model performance, and found the lowest AUC to be 0.79, with associated sensitivity and specificity scores of 0.65 and 0.77, respectively. Applying this model with future water-level scenarios should provide a cost-effective alternative for forecasting changes in the amount of low marsh-habitat in Georgian Bay.