Hakim, W. L., Fadhillah, M. F., Lee, K. J., Lee, S. J., Chae, S. H., & Lee, C. W. (2023). Land Subsidence and Groundwater Storage Assessment using ICOPS, GRACE, and Susceptibility Mapping in Pekalongan, Indonesia. IEEE Transactions on Geoscience and Remote Sensing. Click (a) ICOPS-based mean vertical deformation rate map: Distribution of land subsidence in Pekalongan derived from Sentinel-1 data (2017-2022) and validation with GNSS observation points. (b) GRACE-based GWS and precipitation analysis: Long-term comparison of GWS anomalies derived from GRACE/GLDAS data against monthly precipitation patterns. (c) CNN-GWO based land subsidence susceptibility map and model performance: Vulnerable zones predicted by the optimized deep learning model using 12 conditioning factors, alongside the ROC curve demonstrating the highest AUC of 0.812. (d) ROC-AUC performance metrics: A curve demonstrating that the proposed CNN-GWO algorithm achieved the highest predictive accuracy with an AUC of 0.812, outperforming standalone models. Floods in Pekalongan, Indonesia, often occur due to river water overflowing during heavy monsoon rain, and the northern coastal area has been affected by coastal floods due to sea level rise. The flood conditions in this area were exacerbated by land subsidence, leading to coastal inundation. This study conducted a time-series InSAR analysis based on the ICOPS algorithm, utilizing 124 Sentinel-1 SAR datasets acquired from descending tracks between 2017 and 2022. By integrating a Convolutional Neural Network (CNN) and Optimized Hot Spot Analysis (OHSA), we effectively optimized the measurement points (MPs) by filtering out noise and enhancing spatial clustering. The resulting deformation patterns exhibited a good correlation with actual GNSS observation point measurements.To investigate the impact of excessive groundwater extraction, which is a primary driver of subsidence, Groundwater Storage (GWS) anomalies were calculated by combining GRACE satellite data with GLDAS parameters including SM, CWS, and QS. Comparing these GWS anomalies with local monthly precipitation data revealed a clear correlation accompanied by a time lag phase shift. Ultimately, a susceptibility mapping approach incorporating 12 conditioning factors and hybrid deep learning algorithms identified the CNN-GWO model as the most accurate predictor. Achieving an AUC of 0.812 and an RMSE of 0.305, the model successfully highlighted the extreme vulnerability of settlement areas situated on young alluvium soil. Continue reading...
