Predictive Analytics for Identifying Surface Water Sources for Domestic Water Supply in Phuentsholing, Bhutan

Abstract

Surface water is a primary source of drinking water in Bhutan. To ensure a reliable supply that
meets both quality and quantity requirements, it is crucial to identify suitable sources. This study
presents an integrated approach using Geographic Information Systems (GIS) and machine learning
to identify potential surface water sources. Satellite imagery from Landsat-8 and Sentinel-2 was
utilized to generate geospatial datasets. Five key variables influencing the spatio-temporal presence
of water—rainfall, temperature, soil type, Normalized Difference Vegetation Index (NDVI), and
topography were analyzed within a GIS environment. The Random Forest (RF) algorithm, known
for its robustness in handling nonlinear and high-dimensional data, was employed to predict
potential water sources. Model outputs were validated through field surveys and spectral analysis
using the Normalized Difference Water Index (NDWI). The study identified 50 viable water source
locations situated above 450 meters in elevation. The model achieved an area under the curve (AUC)
score of 0.99, indicating a strong correlation between predicted and actual water sources. These
results confirm that integrating machine learning with remote sensing and GIS is an effective
approach for surface water resource planning in Bhutan's hilly terrain.