HIGH-RESOLUTION MAPPING OF PHYSICAL AND CHEMICAL SOIL ATTRIBUTES IN GERMANY USING GEOSPATIAL MACHINE LEARNING TECHNIQUES

This study mapped physical and chemical soil attributes in Germany for the 0-20 cm layer at a spatial resolution of 90 meters. The geospatial mapping pipeline, entirely developed on the Google Colab platform and integrated with Google Earth Engine, utilized the Random Forest algorithm with hyperparameter tuning. The final model was validated using stratified cross-validation and spatial leave-one-out cross-validation (SLOO-CV) to mitigate spatial autocorrelation. Spatial predictions were carried out with 100 repetitions, generating prediction maps along with their confidence intervals (5% and 95%). The results demonstrate the spatial distribution of important soil attributes such as pH, bulk density, carbon, phosphorus, potassium, and CEC, highlighting significant patterns in Germany's agricultural regions.