AI and Machine Learning-Based Spatial Modeling of Groundwater Quality Indices and Hydrogeochemistry for Accurate Prediction of Seawater Intrusion and Irrigation Sustainability in Coastal Agroecosystems

Abstract This study investigates the quality and spatial variability of groundwater in the coastal agricultural zone of Skhirat, Morocco, under growing environmental and anthropogenic stress. The main objectives were to assess hydrogeochemical characteristics, evaluate groundwater suitability for drinking and irrigation, quantify saltwater intrusion, and model quality indices using artificial intelligence. Groundwater (GW) samples were collected and analyzed for physico-chemical parameters. Hydrogeochemical characterization was performed using Piper, Gibbs, and Chadha diagrams. Water Quality Index (WQI), Irrigation Water Quality Index (IWQI), and Saltwater Mixing Index (SMI) were computed using standard equations. Statistical analyses included correlation matrices, Principal Component Analysis (PCA), and K-means clustering. Machine learning models (Random Forest (RF) and Artificial Neural Networks (ANN)) were applied to predict WQI, IWQI, and SMI, followed by spatial interpolation using GIS approach. Results revealed that WQI values ranged from 31.58 to 139.28, with 40% of samples falling in the "poor" to "very poor" categories. IWQI indicate that 43.3% of samples were classified as "good" and 6.7% as "very poor" for irrigation practices. SMI values >1, indicating seawater intrusion, were observed in 30% of samples. The ANN model achieved high predictive accuracy for IWQI (R²=0.81), while RF performed best for SMI (R²=0.74). Spatial analysis confirmed salinization patterns toward coastal zones. These findings highlight the value of integrated AI and geostatistical approaches for sustainable groundwater monitoring and management in vulnerable coastal aquifers.