Land subsidence along the northern coast of Java (Pantura) poses a significant threat to coastal infrastructure and flood resilience in densely populated areas. This study integrates 12 geo-environmental variables with Persistent Scatterer Interferometric Synthetic Aperture Radar (PS-InSAR) data to predict land subsidence in DKI Jakarta, Semarang City, and Pekalongan Regencyusing Random Forest (RF) and XGBoost regression models. Model performance was evaluated using mean absolute error (MAE), root mean square error (RMSE), and the coefficient of determination (R²) on an independent 20% holdout dataset. Across the three study areas, the optimized ensemble models demonstrated strong predictive performance. RF outperformed XGBoost in DKI Jakarta (image, image) and Semarang City (image, image), whereas both models showed comparable performance in Pekalongan Regency(image). Overall, RF consistently achieved equal or superior predictive accuracy, although the best performance in Pekalongan Regencywas obtained using RF without hyperparameter tuning. Feature importance analysis identified GEOLOGY_CODE as the dominant predictor in DKI Jakarta, while Land Surface Temperature (LST) was consistently important in Semarang City and Pekalongan Regency . Spatial analysis further identified Penjaringan, Genuk, and Kedungwuni as areas experiencing high levels of subsidence. The proposed framework provides a reproducible approach for district-scale land-subsidence assessment and supports risk-informed spatial planning across the Pantura coastal corridor.
Copyrights © 2026