Incorporation of Anisotropy through Azimuth for Machine Learning-Based Spatial Interpolation of Environmental Variables

This study proposes a methodology for spatial interpolation using machine learning algorithms, with an emphasis on incorporating spatial anisotropy through azimuth classification. Using the Meuse dataset, we compare the performance of machine learning models while testing the hypotheses that decision tree-based algorithms are more efficient than in predicting regionalized variables and that incorporating anisotropy improves predictive results when an anisotropic component is present in the data. The results demonstrate that the inclusion of the classified azimuth variable as a predictor enhances the models’ predictive capability, as evidenced by interpolated maps that capture the orientation of spatial patterns.