Modeling wetland change along the United States Atlantic Coast

Abstract

As an aid in waterfowl habitat conservation planning, we predicted change in wetland area between the 1950s and 1970s and the 1970s to 1990s along the United States Atlantic Coast. We developed zero-inflation models using logistic regression to predict occurrence and linear regression to predict area for six wetland classes: estuarine emergent, lacustrine, and palustrine emergent, forested, scrub-shrub, and unconsolidated bottom. A prediction grid of 10.36 km 2 cells was established for the study area. We predicted wetland class area for cells for four decades: 1950s, 1970s, 1980s, and 1990s. Wetland occurrence and area measurements from the U.S. Fish and Wildlife Service’s Wetlands Status and Trends Study (WST) sample plots served as response variables. Spatial structure in the WST data was preserved by interpolating residuals at plot locations and summing predicted residuals and predicted wetland area for each cell. Wetland changes reflect conversion both to and from upland as well as to and from other wetland classes. We predicted a 30.7 thousand ha decline in estuarine emergents in the study area from the 1950s to 1970s, but an increase of 0.9 thousand ha between the 1970s and 1990s. Predicted lacustrine area increased 161.6 thousand ha from the 1950s to 1970s, but declined 5.4 thousand ha from the 1970s to the 1990s. A loss of 731.4 thousand and 189.9 thousand ha of palustrine emergents was predicted from the 1950s to 1970s and from the 1970s to 1990s, respectively. Palustrine forested was predicted to have declined by 536.5 thousand ha between the 1950s and 1970s and 985.8 thousand ha between the 1970s and 1990s. Palustrine scrub-shrub was predicted to have declined 562.6 thousand ha between the 1950s and 1970s, but increased 119.2 thousand ha between the 1970s and 1990s. Finally, palustrine unconsolidated bottom wetlands were predicted to have increased 159.2 and 142.8 thousand ha between the 1950s and 1970s and 1970s and 1990s, respectively. Predictions were most reliable for estuarine emergent, lacustrine, and palustrine forested classes, probably due to the restricted distribution and relative constancy of the estuarine emergent and lacustrine classes. Reliable palustrine forested predictions were likely due to the association of this class with watercourses and the relationship between the predictor variables and lotic habitats. Predictor variables were likely less relevant for palustrine emergent, scrub-shrub, and unconsolidated bottom, where model fit was poorer. Human activities such as agriculture and forestry probably contributed to the poorer model fit for these classes.