Forecasting Urban Water Tanker Demand Using Hydrometeorological Signals: A Machine Learning Approach for Hyderabad, India

Groundwater scarcity has been a growing issue in many developing cities, especially during the era of rapid urbanization and as a result, the need for water supply to be carried in by tankers has risen. The seasonality of water stress is pronounced in Hyderabad in India, where almost half of the population uses private water tankers to supplement households' water needs. The present study proposes a machine learning (ML) model to predict the demand for water tankers within a month's time based on publicly available hydrometeorological and municipal data from the OpenCity Hyderabad and NASA POWER Agroclimatology API respectively. The data consisted of 23 monthly observations from January 2022 to February 2024. The performance of five supervised regression models were assessed by temporal hold-out method, training 2022 data and testing 2023 data. Among the evaluated models, Gradient Boosting had the lowest hold-out error (test R² = 0.7553, RMSE = 10,712 monthly tanker trips, MAPE = 12.16%), which is around 33% smaller than the best naive model, and significantly better than both regularized and unregularized linear models. A parsimonious model with the two predictors was also competitive, showing that there is a strong climatic signal in the demand. The best model was trained with seven monthly observations after lag construction and tested with eleven monthly observations so these results are offered as an indication of a proof of concept, and not to be taken as a deployment-ready system. The results of feature importance analysis indicated that seasonal timing and temperature variables were the most significant factors influencing the demand for tankers while direct rainfall played a minor role. These results indicate the viability of using open access data and ML methods to make forecasts of operational tanker demand. The proposed framework could be used as a support system for proactively managing fleets and allocating resources in urban water-constrained systems.