Examining Spatially Varying Relationships Between Land Use and Water Quality Using Geographically Weighted Regression I: Model Design and Evaluation

Jun Tu, Kennesaw State University


Traditional regression techniques such as ordinary least squares (OLS) can hide important local variations in the model parameters, and are not able to deal with spatial autocorrelations existing in the variables. A recently developed technique, geographically weighted regression (GWR), is used to examine the relationships between land use and water quality in eastern Massachusetts, USA. GWR models make great improvements of model performance over OLS models, which is proved by F-test and comparisons of model R2 and corrected Akaike Information Criterion (AICc) from both GWR and OLS. GWR models also improve the reliabilities of the relationships by reducing spatial autocorrelations. The application of GWR models finds that the relationships between land use and water quality are not constant over space but show great spatial non-stationarity. GWR models are able to reveal the information previously ignored by OLS models on the local causes of water pollution, and so improve the model ability to explain local situation of water quality. The results of this study suggest that GWR technique has the potential to serve as a useful tool for environmental research and management at watershed, regional, national and even global scales.