Predicting wetland loss on private lands in the southeastern United States: Implications for land-use planning

Background/Question/Methods If land policy is to be efficient in conserving and protecting wetland habitats, it is important to identify, in a geographically explicit fashion, those areas where risks of future wetland conversion are highest.  Our objective was to predict the probability of wetland loss based on local characteristics of the wetland itself (e.g., type of wetland, whether it is subject to periodic cultivation) as well as the wetland’s landscape context (surrounding land uses and land covers).  We developed a predictive model of wetland loss using multivariate adaptive regression splines.  Spatial dependencies that can result from biotic processes or model misspecification were incorporated into the predictive surface for wetland loss using residual interpolation.  Wetland fate from 1992 to 1997 was obtained for >40,000 inventory points classified as wetland from the National Resources Inventory across the southeastern US.  Land use and land cover imagery in the vicinity of the wetland points was obtained from the 1992 National Land Cover Data at three spatial extents using circular buffers whose areas were linked to the footprint of the primary sample unit (fine scale = 102 ha; meso scale = 915 ha; macro scale = 8225 ha).  We randomly selected 70% of our observations to train the model, and randomly divided the held-out observations (30%) into five test data sets.

Results/Conclusions Overall prediction accuracy across the five test data sets was 75.3%, with nearly 80% accuracy on predicted wetland loss. Land use surrounding the wetland, wetland ownership class, and proximity to developed land (including roads) and other wetlands were important predictors of loss.  Landscape context was important across all three spatial extents.  We spatially interpolated the predicted probability of wetland conversion across the southeast US.  Low probabilities of wetland loss occurred throughout the Coastal Plain and Piedmont regions but were interspersed with areas of high probabilities of wetland loss around human settlement.  Land-use planners and policy analysts can use the model to identify wetlands in danger of loss, prioritize at-risk areas for protection, and identify in a spatially explicit manner factors that are likely to be responsible for wetland loss.