Modeling and Comparative Analysis of Spatial Distribution of SO2 Concentration Using MLR and GWR Models: A Case Study of Karabük, Türkiye

Abstract In this study, a comparative analysis of Multiple Linear Regression (MLR) and Geographically Weighted Regression (GWR) models was applied within the scope of Land Use Regression (LUR) analysis in order to reveal the spatial distribution of sulfur dioxide (SO 2 ) concentration in Karabük city. 5-fold cross-validation and external data validation methods were used to test the accuracy of the models. The MLR model successfully represented the distribution of SO2 concentrations in general. However, due to the application of fixed coefficients, spatial heterogeneity could not be fully revealed. In the GWR analysis, local differences were revealed more accurately thanks to the coefficient produced separately for each sample point. In the MLR model, the overall accuracy is R2:0.85, the 5-fold cross-validation method accuracy is R2:0.85, and the external data validation method accuracy is R2:0.92. In the GWR model, the overall accuracy is R2:0.95, the 5-fold cross-validation method accuracy is R2:0.95, and the external data validation method accuracy is R2:0.94. In both methods, high SO2 concentrations were observed in the southern part of the city where industrial areas are dense. In the local R2 distribution of the GWR model, high explanation values ​​were again obtained in the southern regions. In the residual analysis of the GWR model, it was observed that the prediction errors of the model were in the range of ± 1 µg/m³ and were randomly distributed except for a few small local regions. According to the results, the GWR model performed better than the MLR model in both predictive accuracy and spatial heterogeneity.