You are here:
Conductivity, Spatially Explicit Models, and the spmodel R Package
Citation:
Dumelle, M., J. Ver Hoef, A. Handler, R. Hill, M. Higham, AND Tony Olsen. Conductivity, Spatially Explicit Models, and the spmodel R Package. Society for Freshwater Science 2024 Conference, Philadelphia, PA, June 02 - 06, 2024.
Impact/Purpose:
This work is important to communicate three ideas. First, the nonspatial models are generally inferior to spatial models for spatial data problems. Second, the spmodel R package is a new resource that allows users to seamlessly transition from nonspatial to spatial models in R. Third, we identify nationwide drivers of lake conductivity and make predictions at all lakes in CONUS.
Description:
Conductivity is an important indicator of the health of aquatic ecosystems and is linked to anthropogenic activity. Understanding patterns and drivers of conductivity is important for effective management. We modeled lake conductivity data (n = 3,311) collected as part of the U.S. Environmental Protection Agency’s National Lakes Assessment (NLA) using spatial indexing, a flexible and efficient approach to fitting spatially explicit statistical models to big data sets. These spatial statistical models build spatial dependence directly into modeling and offer vast improvements over models that assume the data observations are independent (e.g., 45% reduction in mean-squared prediction error). We find lake conductivity is strongly related to calcium oxide rock content, crop production, human development, precipitation, and temperature. We used a final spatial model to predict lake conductivity at hundreds of thousands of lakes distributed throughout the contiguous United States. These maps revealed higher lake conductivities in the arid Southwest and several Midwestern states, such as the Dakotas. The combination of federal monitoring data with spatial modeling can offer important insights into the patterns and drivers of water quality nationally. Lake conductivity models fit using spatial indexing are nearly identical to lake conductivity models fit using traditional spatial analysis methods but are nearly 50 times faster. Spatial indexing and other tools for spatial statistical modeling and prediction are readily available in the spmodel R package. The views expressed in this article are those of the author(s) and do not necessarily represent the views or policies of the U.S. Environmental Protection Agency.