Fantastic Wetlands and Where to Find Them: Modeling Rich Fen Distribution in New York State with Maxent

• Published in: Wetlands (Wilmington, N.C.) Vol. 38; no. 1; pp. 81 - 93
• Main Authors: Raney, Patrick A., Leopold, Donald J.
• Format: Journal Article
• Language: English
• Published: Dordrecht Springer Netherlands 01-02-2018; Springer Nature B.V
• Subjects: Biodiversity; Biomedical and Life Sciences; Climate; Climate change; Climate prediction; Coastal Sciences; Conservation; Ecology; Environmental Management; Fens; Freshwater & Marine Ecology; Geology; Groundwater; Habitat loss; Habitats; Humidity; Hydrogeology; Hydrology; Introduced species; Invasive species; Landscape Ecology; Life Sciences; Maximum entropy; Modelling; Nutrient loss; Original Research; Precipitation; Rare species; Terrain; Thin plates; Topography; Wetlands; Wildlife conservation; New York; United States > US; Climate-downscaling; Maxent; Fens; Rare plants; Wetlands
• ISSN: 0277-5212; 1943-6246
• DOI: 10.1007/s13157-017-0958-5

Rich fens are groundwater dependent, mineral-rich wetlands that support diverse plant assemblages and high densities of rare species. Despite their conservation importance, fens are threatened by habitat loss, altered nutrient regimes, invasive species, and groundwater extraction. Fen conservation is hindered by limited inventory efforts. We used maximum entropy modeling (maxent) to estimate fen distribution throughout New York State (NYS), USA (125,936 sq. km). During predictor assembly, coarse-resolution climate variables (temperature, humidity, and precipitation) were statistically downscaled to match higher-resolution terrain variables (10 m
2
) for fen distribution modeling using linear regression for temperature and humidity variables (R
2
 = 0.92), and thin-plate spline regression for precipitation (mean absolute error = 10.14 mm). A nine-parameter maxent model of fen distribution incorporating geological characteristics, soils, landcover, climate (PC axis 1), terrain facets, and hydrology accurately predicted fen records (AUC = 0.97), with an 82% correct classification rate compared to 70% for a null model. Fen habitat spanned only 0.04% of NYS, and matched well with assertions that fens occupy narrowly-distributed, rare hydrogeologic settings. Our results suggest distribution modeling techniques may improve fen detection in large, poorly sampled regions.