A smoothing spline approach for integrating spatially misaligned water quality data

Abstract

Contamination is a looming threat to U.S. water resources, but the inconsistency of data collection across studies and agencies has made it difficult to identify effective treatments or resolutions. In many areas, the available water quality datasets are compatible with standard regression analysis; however, they cannot be integrated with other data sources to directly investigate those mechanisms. Here we develop an approach for connecting data with mismatched spatial resolutions and leverage it to integrate decades of Kansas ambient water monitoring with best management practice data. Our method builds upon the existing krige-and-regress framework by fitting a model of similar structure to spatio-temporal kriging; cubic splines fit via gamma generalized additive models to adjust for the non-negative and nonlinear characteristics present in ambient monitoring data. Applying this model to Kansas monitoring and best management practice data, we identify several candidate practices associated with nutrient reduction. These findings demonstrate the potential of spatially connected water data in advancing understanding of nonpoint source pollution and informing targeted best management practices at the subwatershed level.