GIS-based spatial regression and prediction of water quality in river networks: a case study in Iowa

Nonpoint source pollution is the leading cause of the U.S.'s water quality problems. One important component of nonpoint source pollution control is an understanding of what and how watershed-scale conditions influence ambient water quality. This paper investigated the use of spatial regression to evaluate the impacts of watershed characteristics on stream NO(3)NO(2)-N concentration in the Cedar River Watershed, Iowa. An Arc Hydro geodatabase was constructed to organize various datasets on the watershed. Spatial regression models were developed to evaluate the impacts of watershed characteristics on stream NO(3)NO(2)-N concentration and predict NO(3)NO(2)-N concentration at unmonitored locations. Unlike the traditional ordinary least square (OLS) method, the spatial regression method incorporates the potential spatial correlation among the observations in its coefficient estimation. Study results show that NO(3)NO(2)-N observations in the Cedar River Watershed are spatially correlated, and by ignoring the spatial correlation, the OLS method tends to over-estimate the impacts of watershed characteristics on stream NO(3)NO(2)-N concentration. In conjunction with kriging, the spatial regression method not only makes better stream NO(3)NO(2)-N concentration predictions than the OLS method, but also gives estimates of the uncertainty of the predictions, which provides useful information for optimizing the design of stream monitoring network. It is a promising tool for better managing and controlling nonpoint source pollution.