Jolo Island Surface Heat GeoAI: Reproducibility Package

Version 2.0.0 extends the original Jolo Island Surface Heat GeoAI reproducibility archive with a matched geographic validation framework for prediction and model-attribution transfer. The release adds fold-wise SHAP attribution stability under spatial blocking, municipality-held-out prediction and attribution transfer, environmental-support diagnostics, spatial-block origin sensitivity, cross-algorithm robustness using LightGBM, XGBoost and Random Forest, dominant-predictor ablation, seasonal matched-attribution analysis, seed-sensitivity analysis, and integrity verification of frozen manuscript-supporting outputs. The empirical study uses a 2013–2026 Landsat 8/9 surface-temperature climatology over Jolo Island, Philippines, sampled on a 300-m systematic terrestrial lattice. Version 2 reframes geographic model validation as a two-dimensional problem in which predictive transferability and attribution transferability are evaluated separately under matched geographic perturbations. Version 1.0.0 remains preserved as the original reproducibility archive (DOI: 10.5281/zenodo.22541245). Version 2.0.0 adds the new prediction–attribution transfer and robustness analyses without replacing the Version 1 scientific record.