<p>One of the major risks brought on by both natural and human forces is the problem of land subsidence. Along with causing major economic losses, it also threatens the ecology and lowers the elevation of the land surface. The goal of this study is to create a comprehensive strategy for long-term land subsidence monitoring and risk assessment. Sentinel-1 satellite radar data from 2015 to 2025 was used for monitoring. The Interferometric Synthetic Aperture Radar (InSAR) technology was used to map and extract precise spatiotemporal subsidence patterns. Then, using 13 environmental driving factors (hydrological, geological, topographical, and infrastructure), the performance of four machine learning and deep learning algorithms Random Forest (RF), XGBoost, Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM) was assessed in order to forecast the subsidence hazard. According to the modeling results, the traditional RF and XGBoost models outperformed deep learning models such as CNN (AUC = 0.96) and LSTM (AUC = 0.84), with the maximum accuracy (AUC = 0.99), in forecasting the hazard of land subsidence. This result highlights how boosting and bagging models are more stable and generalizable when handling complex geospatial data. Hydrological and geological parameters are the main causes of subsidence in the research region, according to an examination of the influencing elements. These results offer a way to choose the best prediction model while also validating the effectiveness of 10 years of InSAR monitoring as a basis for decision-making. Additionally, hazard-resilient urban design and the sustainable management of groundwater resources would directly benefit from the findings of this study.</p> Graphical Abstract <p></p> <p>Based on the graphic representation, this study was conducted to investigate the extent of subsidence and the influence of environmental factors on subsidence in Pol Dokhtar district. The study area is the Poldokhtar County, located in southwestern Lorestan Province. In this region, land subsidence risk was evaluated using Sentinel-1 radar data for the period from 2015 to 2025, integrated with a suite of 13 environmental and 13 hydrogeological factors. The vertical surface displacement rate maps were derived through Interferometric Synthetic Aperture Radar (InSAR) processing within the SNAP environment, followed by phase unwrapping. Subsequently, these maps were integrated with environmental layers using Python. Four methods were used to estimate the susceptibility to ground subsidence: Random Forest (RF), XGBoost, Convolutional Neural Networks (CNN), and Long Short-Term Memory (LSTM). Standard measures such as AUC, Precision, Recall, and F1-score were used to assess their performance. In terms of AUC and classification metrics, the results show that the Random Forest (RF) and XGBoost models perform better (AUC = 0.99). The resulting maps show the zonation of land subsidence susceptibility in five classifications, from “very low” to “very high.” Groundwater level was shown to be the most important element, followed by distance from faults and railroads, according to the RF model’s variable importance analysis. These results highlight the simultaneous impact of structural and hydrogeological factors on land subsidence. Subsidence in the studied region is mostly caused by hydrological and geological characteristics, according to additional examination of significant elements. Based on these findings, regulating groundwater extraction, enhancing irrigation efficiency, and implementing continuous InSAR monitoring are essential to stabilize the land surface and protect infrastructure. Failure to address these issues specifically the continuance of over-abstraction could worsen subsidence and lead to irreversible harm to the aquifer.</p>