Advanced time-series InSAR analysis to estimate surface deformation and utilization of hybrid deep learning for susceptibility mapping in the Jakarta metropolitan region

Excessive groundwater extraction in the Jakarta Metropolitan Region (JMR) has led to land subsidence, making the region more prone to flooding during heavy rain and at risk of being submerged by seawater during high tide. A reliable method for surface deformation measurements needs to be developed to address this issue. This article examines the use of multitemporal Sentinel-1 synthetic aperture radar (SAR) data from 2015 to 2023. It focuses on processing these data using the Improved Combined Scatterer Interferometry with Optimized Point Scatterers (ICOPS) method, integrated with the MintPy algorithm, to address instances of nonlinear deformation. The accuracy of the developed method was evaluated to that of the original ICOPS method using GPS data. According to the RMSE value, the ICOPS-MintPy (CBTU: 1.82 cm/year; CTGR: 0.93 cm/year; CJKT: 1.16 cm/year) outperformed the original ICOPS (CBTU: 1.99 cm/year; CTGR: 1.71 cm/year; CJKT: 1.93 cm/year) method. Thus, the mean deformation rate map obtained with the ICOPS-MintPy algorithm is suitable as an inventory map for identifying areas susceptible to future land subsidence around JMR. A convolutional neural network (CNN) and long short-term memory (LSTM) were utilized to generate susceptibility maps. In addition, metaheuristic algorithms were implemented to optimize the parameters of both CNN and LSTM. The metaheuristic algorithms are the gray wolf optimizer (GWO) and the imperialist competitive algorithm (ICA). We intend to employ a combination of deep learning and metaheuristic algorithms to create four hybrid models: CNN-GWO, CNN-ICA, LSTM-GWO, and LSTM-ICA. The performance of these hybrid models will be compared to that of standalone CNN and LSTM algorithms as the base model before parameter optimization. This process will be conducted using the area under the curve (AUC) value from the receiver operating characteristic (ROC) curve analysis. The susceptibility model performance of the LSTM-GWO has the highest AUC value of 0.976, followed by the CNN-GWO at 0.974, LSTM-ICA at 0.972, LSTM at 0.965, CNN-ICA at 0.960, and CNN at 0.951. Nevertheless, all of the models have excellent performance, as shown by the AUC values between 0.9 and 1.0. Finally, the ICOPS-MintPy algorithm for land subsidence monitoring and hybrid deep learning algorithms for susceptibility mapping resulted in more accurate results due to its improved accuracy.