A GIS-Based Logistic Regression Model for Groundwater Spring Potential Mapping in the Gamri Chhu Basin, Bhutan

Abstract

Groundwater springs constitute the predominant water supply for upland communities in Bhutan’s Hindu Kush Himalaya (HKH) region. However, 25.1% of springs in Bhutan are currently reported as drying, underscoring the need for accurate mapping of spring potential zones to support conservation and recharge planning. This study employed a logistic regression (LR) model to delineate groundwater spring potential zones in the Gamri Chhu Basin of eastern Bhutan. An inventory of 145 spring locations was compiled and paired with an equal number of pseudo-absence points generated using a 500 m exclusion buffer around known spring locations. Eight environmental conditioning factors altitude, slope, geology, drainage density, lineament density, land use, soil type, and precipitation were evaluated as spatial predictors. Correlation and multicollinearity analyses were performed to ensure model reliability. The LR model was trained using 70% of the dataset and validated with the remaining 30%. Results indicated that altitude, slope, geology, drainage density, precipitation, and soil type significantly influenced spring occurrence (p < 0.05). The model achieved good predictive performance, with AUC values of 0.86 (training) and 0.85 (validation). The resulting map was classified into low (81.34%), moderate (11.80%), and high (6.87%) potential zones. SDI values increased progressively from low- to high-potential zones, with the highest values observed in the high-potential class (6.00 for training and 5.08 for validation), indicating strong spatial agreement between predicted and observed spring locations. The resulting map serves as a preliminary, regional-scale screening tool to prioritize areas for field verification, recharge zone investigation, and informed groundwater assessment in data-limited mountainous regions.