Trend analysis and learning-based groundwater level modelling over a tropical river basin

Groundwater trend analysis and modelling is challenging due to partially explicable factors and unexplained human influence. The Hurst index, sequential Mann–Kendall, and classical Mann–Kendall test offer a comprehensive groundwater trend analysis. A learning-based approach is developed to model groundwater levels using climatological variables of rainfall and temperature. Twenty-four locations were considered over Periyar river basin of Kerala, India, for the years 1996–2019, and during January, April, August, and November (JAAN) months. Significant trends were observed at 14 locations in at least one of the JAAN months, which is about 58%. Of these, eight locations exhibited positive trend, signalling a decline in groundwater supplies. The developed model yielded notable improvements in precision with 50%, 79%, 75%, and 83% of the locations in month-wise order. To gauge the model performance, observed and predicted location clusters obtained using k-means clustering are juxtaposed for the years 2017–2019, on both individual and average basis. This assessment indicated only one well transitioning in August, with the average approach resulting in a closer match to the original clustering for most of the wells. These findings will benefit future stakeholders and policymakers in optimising resource management strategies over the basin and wider.