Ai-Enhanced Fuzzy-Drastic Model For Groundwater Vulnerability Assessment And Dynamic Contamination Risk Prediction

Abstract

The groundwater contamination especially in the semi-arid region is a significant threat to public health, ecological balance and sustainable agriculture. Current groundwater vulnerability assessments are usually not temporally sensitive and do not represent complex spatial interactions in different regions and years. The goal of this work is to build an advanced, explainable and data driven framework to predict amount of groundwater contamination risk using geohydrological parameters, chemical water quality characteristics and temporal variations. Unlike conventional models, a hybrid framework composed of Fuzzy-DRASTIC model for uncertainty aware vulnerability indexing and Spatio-Temporal Graph Attention Network (ST-GAT) for intelligent risk classification is introduced in this research. Domain based fuzzification and spatial and temporal attention mechanisms are used in the approach to capture the real-world aquifer dynamic more accurately. Preprocessing of the groundwater quality datasets (2018–2020) by means of normalization and spatial-temporal merging are presented, and the methodology starts from there. Fuzzified geohydrological parameters are used in constructing a Fuzzy DRASTIC Vulnerability Index (FVI). Finally, the final graph structured input that combines chemical attributes, FVI and spatial temporal edges is integrated. This leads to an application of a ST-GAT architecture to model interactions between space and time, and to predict contamination risk categories (Safe, Marginal, Unsafe). Compared to traditional models the proposed Fuzzy-DRASTIC + ST-GAT model has overall accuracy of 98.5%. The model also achieved an excellent separation of the classes as shown by high AUC-ROC score of 0.961. A spatial risk map of high resolution was generated for targeting in high-risk areas. This research provides a high performance, interpretable, and scalable risk prediction solution for groundwater contamination, together in an integrated framework. It allows for proactive water resource management through spatio-temporal deep learning and fuzzy logic combination that enables policy level decision making and sustainable development planning.