Exploring the Non-Stationary Effects of Forests and Developed Land within Watersheds on Biological Indicators of Streams Using Geographically-Weighted Regression

This study examined the non-stationary relationship between the ecological condition of streams and the proportions of forest and developed land in watersheds using geographically-weighted regression (GWR). Most previous studies have adopted the ordinary least squares (OLS) method, which assumes stationarity of the relationship between land use and biological indicators. However, these conventional OLS models cannot provide any insight into local variations in the land use effects within watersheds. Here, we compared the performance of the OLS and GWR statistical models applied to benthic diatom, macroinvertebrate, and fish communities in sub-watershed management areas. We extracted land use datasets from the Ministry of Environment LULC map and data on biological indicators in Nakdong river systems from the National Aquatic Ecological Monitoring Program in Korea. We found that the GWR model had superior performance compared with the OLS model, as assessed based on R-2, Akaike's Information Criterion, and Moran's I values. Furthermore, GWR models revealed specific localized effects of land use on biological indicators, which we investigated further. The results of this study can be used to inform more effective policies on watershed management and to enhance ecological integrity by prioritizing sub-watershed management areas