Geographic variables association detection model based on spatially weighted mutual information

• A Spatially Weighted Mutual Information (SWMI) is proposed to quantify association. • SWMI can handle continuous and categorical dependent variable. • SWMI is insensitive to noise. • SWMI can identify the independent variables group with the best explanatory power. Spatial association measurement is crucial for revealing the relationship between variables, and is widely used in spatial analysis, geographical simulation, and causal inference. Existing methods within the Geographic Detector (GD) framework rely on the variance of the dependent variable. When the dependent variable is categorical, it is difficult to assess the explanatory power and interaction effects of the independent variables. To address this issue, a Spatially Weighted Mutual Information (SWMI) model is developed to detect the association between geographic variables. The SWMI model first discretizes the geographic variables, then uses the spatial autocorrelation strength to adjust the probabilities of the variables, and finally calculates information entropy and mutual information based on the adjusted probabilities to quantify the spatial association between geographic variables. In the study, the SWMI model is applied to three scenarios: simulated data with varying noise levels, the identification of the influencing factors of NO 2 concentration in China (continuous dependent variable), and the detection of the driving factors of vegetation cover types in the Xizang Autonomous Region (categorical dependent variable).The results show that the SWMI model applies to both continuous and categorical dependent variables and demonstrates strong robustness to noise.