55 / 2025-08-10 02:57:16

Deep Learning for Interpreting Mechanisms and Predicting Fine-Scale Spatiotemporal Evolution of Regional Land Subsidence

Deep Learning,Land Subsidence,Interpreting Mechanisms,Fine-Scale Spatiotemporal Evolution

With the intensification of urban underground space development, regional land subsidence disasters exhibit complex characteristics of multi-factor coupling and cross-scale evolution, posing severe challenges to urban safety. This paper systematically reviews research advances in deep learning technologies for mechanistic interpretation and spatiotemporal prediction of land subsidence, focusing on three core directions: (1) explainable artificial intelligence (XAI)-empowered analytical techniques for disaster-inducing factors; (2) data-physics dual-driven intelligent inversion models for land subsidence; (3) hybrid architecture-based spatiotemporal prediction models for land subsidence. The study further analyzes multidimensional challenges faced by coastal soft-soil megacities under intensive human engineering activities: response lag to abrupt deformation, insufficient cross-geological-scenario transferability, and difficulties in quantifying multi-scale coupling mechanisms. Future efforts should prioritize developing prediction-regulation closed-loop systems and deepening coupled analysis of "engineering activities-subsidence response" mechanisms to advance an intelligent prevention system featuring mechanism-transparent, precision-oriented, and intelligently regulated capabilities. This research aims to provide theoretical paradigms and technical pathways for land subsidence risk prevention and control, promoting the transformation of geohazard mitigation toward proactive, systematic, and precision governance.