GIS-BASED PREDICTION OF WATER QUALITY INDEX USING MACHINE LEARNING: A CASE STUDY OF THE THA CHIN RIVER, THAILAND

Abstract

This study integrates field data, machine learning (ML), and Geographic Information System (GIS) techniques to compute and predict the Water Quality Index (WQI) along the Tha Chin River Basin, Thailand. Five key water quality parameters—DO, BOD, COD, pH, and Salinity—were collected from monitoring stations between 2019 and 2022. These were normalized into Q-values and used to compute WQI through weighted aggregation. Six supervised ML algorithms were tested, with the Random Forest model yielding the highest predictive performance (R² = 0.664, MAE = 0.855) at a 10-hour lead time. Incorporating spatial indicators such as urban land cover and population density significantly enhanced model accuracy. Computed WQI values were visualized using Inverse Distance Weighted (IDW) interpolation to assess spatial and temporal trends. Results indicated a consistent decline in water quality from upstream to downstream, with all zones classified as "moderately degraded" or "degraded." The lowest WQI values were observed in Banglen District, ranging from 39 to 42, linked to high urban density and pollutant accumulation. In contrast, upstream areas such as Mueang Suphan Buri slightly improved over time. The study confirms that urban expansion is a major contributor to river water degradation. The proposed ML-GIS framework supports proactive monitoring, spatial prioritization, and evidence-based water resource management in rapidly urbanizing river basins.