Rocky Mountain Research Station Publications
RMRS Online Publication - Journal
Articles, External Publications, and Special Reports
Wetland features and landscape context predict the risk of wetland habitat loss
Gutzwiller, Kevin J.; Flather, Curtis H. 2011. Wetland features and landscape context predict the risk of wetland habitat loss. Ecological Applications. 21(3): 968-982.
Wetlands generally provide significant ecosystem services and function as important harbors of biodiversity. To ensure that these habitats are conserved, an efficient means of identifying wetlands at risk of conversion is needed, especially in the southern United States where the rate of wetland loss has been highest in recent decades. We used multivariate adaptive regression splines to develop a model to predict the risk of wetland habitat loss as a function of wetland features and landscape context. Fates of wetland habitats from 1992 to 1997 were obtained from the National Resources Inventory for the U.S. Forest Service's Southern Region, and land-cover data were obtained from the National Land Cover Data. We randomly selected 70% of our 40 617 observations to build the model (n = 28 432), and randomly divided the remaining 30% of the data into five Test data sets (n = 2437 each). The wetland and landscape variables that were important in the model, and their relative contributions to the model's predictive ability (100 = largest, 0 = smallest), were land-cover/ land-use of the surrounding landscape (100.0), size and proximity of development patches within 570 m (39.5), land ownership (39.1), road density within 570 m (37.5), percent woody and herbaceous wetland cover within 570 m (27.8), size and proximity of development patches within 5130 m (25.7), percent grasslands/herbaceous plants and pasture/hay cover within 5130 m (21.7), wetland type (21.2), and percent woody and herbaceous wetland cover within 1710 m (16.6). For the five Test data sets, Kappa statistics (0.40, 0.50, 0.52, 0.55, 0.56; P < 0.0001), area-under-the-receiver-operating-curve (AUC) statistics (0.78, 0.82, 0.83, 0.83, 0.84; P < 0.0001), and percent correct prediction of wetland habitat loss (69.1, 80.4, 81.7, 82.3, 83.1) indicated the model generally had substantial predictive ability across the South. Policy analysts and land-use planners can use the model and associated maps to prioritize at-risk wetlands for protection, evaluate wetland habitat connectivity, predict future conversion of wetland habitat based on projected land-use trends, and assess the effectiveness of wetland conservation programs.
Keywords: estuarine and palustrine wetlands, land conservation, landscape context, multivariate adaptive regression splines, National Resources Inventory, policy and planning, risk of wetland habitat conversion, southern United States, spatially explicit predictive model, wetland protection
About PDFs: For best results, do not open the PDF in your Web browser. Right-click on the PDF link to download the PDF file directly to your computer. Click here for more PDF help.
PDF File Size: 840 K
Title: RMRS Other
Publications: Wetland features and landscape context predict the
risk of wetland habitat loss
Electronic Publish Date: June 27, 2011
Last Update: June 27, 2011
RMRS Publications | Order a publication | Contact Us