Remote Sensing-Driven Prediction of Groundwater Nitrate Risk: Insights from Machine Learning Applications in Taiwan

• DOI: 10.5194/egusphere-egu25-6502
• Publication Date: 2025-03-14T19:28:01Z
• Authors: Yu-Chun Hsu; Kai-Yun Li; Joel Podgorski; Michael Berg

Groundwater nitrate (NO3-) pollution is a pressing issue linked to agricultural practices, urbanization, and industrial activities. This study focuses on Taiwan’s groundwater nitrogen (NO3-N) contamination by integrating satellite remote sensing, monitoring, various environmental factors using GIS. Data from 451 monitoring stations, sampled quarterly 2020 2024, reveal that NO3-N concentrations generally range between 1–10 mg/L, while approximately 2% exceed Drinking Water Quality Standards of 10 mg/L for (equivalent 44.3 NO3-). In this study, machine learning models, including Random Forest (RF), Multilayer Perceptron, Support Vector Classifier, were employed predict risk at three ranges (10 mg/L) different feature combinations: (1) all features, (2) selective factors, (3) vegetation indices (VIs) alone. RF demonstrated the highest overall accuracy across combinations, achieving 87% in Feature Combination I. For III, which only used VIs derived achieved an OA 68%, highlighting its potential practical efficient application without ground-based survey data. Key findings highlight pivotal role variables, Sentinel-2 multispectral imagery, terrain parameters digital elevation meteorological data mapping hotspots. Future work should integrate higher-resolution imagery more advanced improve model performance decision-making accuracy.