Integration Sentinel-1 SAR data and machine learning for land subsidence in-depth analysis in the North Coast of Central Java, Indonesia

Abstract

The escalating issue of land subsidence poses a critical threat to the economic prosperity of Indonesia’s North Coast in Central Java. This recurring phenomenon intensifies annual tidal floods, posing a severe threat to the infrastructure, buildings, coastal zones, land quality, and the livelihoods of local communities. Effective monitoring of land subsidence rates is essential to mitigate these impacts and implement pre-emptive measures. This study addresses this challenge by employing a two-pronged approach: measuring subsidence rates and assessing susceptibility. Over six years (2016–2021), the research utilizes SAR Sentinel-1 data coupled with machine learning algorithms to achieve these goals. The subsidence rates are generated by the time series InSAR SBAS method. Land subsidence susceptibility assessment uses algorithms such as Random Forest (RF), Gradient Boosted Trees (GTB), Classification and Regression Trees (CART), Support Vector Machine (SVM), Decision Trees with Bagging Method (MD), and K-Nearest Neighbours (KNN). An exhaustive assessment utilizing K-fold cross-validation, incorporating five folds with an 80% training and 20% validation split, effectively facilitates the identification of the model exhibiting the highest accuracy. The findings reveal significant spatial variations in land subsidence rates. Semarang, Pekalongan, and Jepara experienced the highest rates (ranging from − 13 cm/year to -5 cm/year) based on SAR Sentinel-1 data. Machine learning model evaluation yielded Overall Accuracy values of 0.761 (RF), 0.766 (GTB), 0.65 (CART), 0.456 (SVM), 0.359 (KNN), and 0.541 (MD). Based on this analysis, the RF and GTB algorithms are recommended for mapping land subsidence susceptibility. Additionally, the study identified influential factors, with distance from boreholes being the most significant influence. Other notable variables are distance to rivers, rainfall, wetness index, proximity to faults, and distance from residential areas. These valuable insights offer significant benefits to decision-makers and stakeholders, including local governments, urban planners, and disaster management agencies. These findings serve as a foundation for developing a comprehensive policy framework and strategic measures to address land subsidence in this critical region.