Explainable AI Optimizes Satellite Ground Station Siting with LiDAR Data

Current practices for satellite ground station siting and spectrum coordination often rely on fixed clutter heights assigned to land use classes, which can lead to inaccuracies and suboptimal site selection. A new interpretable machine learning framework has been developed to predict Representative Clutter Height (RCH), a critical parameter for radio propagation, using open geospatial data. The model, trained with LiDAR-derived labels and various global geospatial features, achieves a mean absolute error of 1.79m and an R^2 of 0.765, representing over a 60% reduction in error compared to the ITU baseline. Using SHAP analysis, the framework identifies tree canopy cover, land-cover semantics, and spectral reflectance as the most influential predictors. This approach offers a globally deployable solution that enhances clutter modeling accuracy at scale while maintaining interpretability, crucial for RF planning.