Using a multiscale, probabilistic approach to identify spatial-temporal wetland gradients

Abstract

Abstract Wetlands are highly dynamic ecosystem components that fluctuate dramatically in inundation and persistence of water both within and across years. However, these systems are commonly classified in a deterministic, discrete manner that does not reflect inherent spatial and temporal variation. Developing a methodology to identify gradients in water inundation is critical given the dynamic nature of wetlands. We present a methodology that applies probabilistic estimates, derived from a nonparametric model, to predict wetlands along a gradient in ephemerality, or degree of water inundation. We applied this model across four sampling areas in the Plains and Prairie Pothole Region (PPPR) in the U.S. Northern Great Plains. We developed a model relationship between high-resolution (RapidEye) and moderate resolution (Landsat) satellite sensor data. This allowed us leverage the benefits of high spatial resolution data and a long temporal series of freely available mid-resolution data to characterize water persistence in wetlands. To obtain measures of wetland inundation across a gradient of ephemerality, we estimated wetland probabilities across a temporal series reflecting large variation in moisture conditions. We found that a nonparametric statistical approach was highly effective in predicting wetlands of varying size and ephemerality. Our predictions were strongly supported with low error (RapidEye 3.1–15%, Landsat 0.3–1.5%). Probabilistic predictions of wetland ephemerality contribute valuable information needed for management and policy decisions, especially given potential alterations to wetland ephemerality and ecosystem services under climate change. Using predicted gradients in wetland ephemerality over time will enable researchers and land managers to more effectively capture nuance in ecosystem condition, function, and change.