Severe land surface deformation caused by extreme drought, intense rainfall events, and groundwater fluctuations poses global challenges with significant environmental, hydrogeological, civil, and economic consequences. Qom Province in Iran has been identified as a hotspot for pronounced land surface deformation. Accordingly, this study aims to reveal the underlying mechanisms of land surface deformation using a multidisciplinary approach integrating remote sensing, neural networks, statistical analysis, and field reconnaissance with in situ sampling. Results from Small Baseline Subset Interferometric Synthetic Aperture Radar (SBAS-InSAR) analysis over a tenyear period show that more than 39% of the province is affected by land subsidence, reaching a maximum rate of -175 mm/yr, while the remaining areas exhibit surface swelling with magnitudes of up to 47 mm/yr. A fully connected neural network (FCNN) model, enhanced with the permutation feature importance method and trained using 30,000 sample points from ten influencing factors, indicates that aquifer conditions, precipitation patterns, and groundwater levels account for approximately 70% of the observed land surface deformation. Field investigations and laboratory testing further demonstrate that in saline environments near Namak Lake, precipitation can induce land subsidence through rainfall-driven soil dispersion, in addition to its role in groundwater recharge. Model uncertainty assessment using the Monte Carlo dropout method indicates that over 90% of the coefficient of variation values fall within -1 to 1, indicating stable model performance. Overall, the results highlight the necessity of jointly considering subsidence and swelling processes and provide an integrated methodology for analyzing land surface deformation in arid and semi-arid regions.