Identification of potential dam sites for severe water crisis management in semi-arid fluoride contaminated region, India

Pressure on freshwater resources is tremendously increasing due to large-scale global population explosion, socio-economic development, climate change and infrastructural development worldwide. The study area faces severe water crisis, groundwater fluoride contamination, and high drought frequency. Thus, the principal objectives are i) to assess the recent surface and subsurface water dynamics in this plateau fringe using satellite datasets on Google Earth Engine (GEE) and ii) to demarcate the suitable sites for dam construction to manage the severe water crisis and substitute drinking water sources. Satellite datasets such as Sentinel 2 and Gravity Recovery and Climate Experiment (GRACE) have been used to access the surface and groundwater dynamics. Numerous criteria or influencing factors including geology, geomorphology, lineament, elevation, slope, rainfall, land use/land cover, soil, stream density, normalized vegetation index (NDVI), and distance from the river have been considered to demarcate the suitable sites for new dam site suitability. In this study, four advanced machine learning models namely support vector machine (SVM), random forest (RF), logistic regression (LR) and gradient boosting (XGBoost) have been applied to recommend suitable sites for dam construction. Average surface water changes from 157.375 km2 (2012–2016) to 156.185 km2(2017–2022). Estimated water thickness (EWT) values vary from 28.58 cm to −27.07 cm (2002–2017). In case of soil moisture (SM), the lowest value (2.4 cm) was in June 2009, and the highest (21.51 cm) was in September 2003. After the deduction of SM from EWS, it specifies that maximum groundwater storage (9.48 cm) occurred in July 2004 whereas a minimum (-30.21 cm) in March 2016. Dam site suitability results denote that 10% of areas come under the very high suitable for surface and subsurface dam construction. The area under curve (AUC) values of SVM, RF, LR, and XGBoost are 0.94, 0.95, 0.91, and 0.92 respectively. Therefore, the RF model has comparatively higher values regarding model performance. The output of this research will be advantageous to define suitable places for new dam construction and sustainable water resource management in semi-arid environment.