Subsidence vulnerability assessment due to groundwater over-abstraction using AI-based multiple cluster analysis

Land subsidence triggered by excessive groundwater extraction is a topical research activity, and the ALPRIFT framework calculates the Subsidence Vulnerability Index (SVI). In this study, we employed Artificial Intelligence (AI) to reduce the inherent subjectivity in the ALPRIFT framework using Inclusive Multiple Modeling (IMM). IMM incorporates Random Forest (RF) and Support Vector Machine (SVM) to conduct cluster analysis at Level 1 and identify clusters fed into another RF model at Level 2. We applied this formulation to an unconfined aquifer, which was affected by water table decline. The study identified vulnerable areas in the central part of the aquifer, representing a maximum of 14 cm of subsidence detected by InSAR. The ratio of vulnerable areas to total areas are 5.5, 8.8 and 5.4 % for RF, SVM and IMM, respectively. Compared to the basic ALPRIFT framework, the AI models at both levels considerably improved the modeling performance from 0.7 to 96.5.