Reservoir Water Level Prediction and Gate Operation Optimization Using Machine Learning: A Case Study of Shetrunji Reservoir, Gujarat

Abstract

Control over reservoir water levels and gate settings at the spillways is key for preventing floods, protecting people living downstream, and efficient utilization of stored water especially when the reservoir is nearing its maximum capacity. This research project centers on the Shetrunji Reservoir in Gujarat, India, investigating the potential of machine learning to improve operations. Daily and event-based hydrological records for the reservoir covering the period 2020 to 2024 were compiled, digitized, and used to train and test predictive models. Precipitation behaviour, inflow characteristics, and the existing operating protocol were analysed, and three modelling frameworks were developed: computational automation of the current lookup-based practice; an Intermediate Machine Learning scheme that combines Quadratic Regression for the level-storage relationship with K-Nearest Neighbours regression for the gauge-discharge relationship; and an XGBoost regression model for direct water-level forecasting. On the held-out test partition, which comprises 20 percent of the observations, the Quadratic Regression model reproduced the level-storage curve with a coefficient of determination of 0.9999, the K-Nearest Neighbours model reproduced the gauge-discharge curve with a coefficient of determination of 0.9995, and the XGBoost model forecast the reservoir water level with a root mean squared error of 0.0419 m, a coefficient of determination of 0.9995, and a mean squared error of 0.00175. These accuracies are interpreted in the light of the smooth, near-deterministic nature of the underlying rating relationships. A tiered gate-operation rule is proposed that extends the simple full-reservoir-level threshold by incorporating water-level bands, gate-sequencing priority, and structural safety considerations. The results indicate that data-driven models can support proactive flood-risk reduction and more consistent, timely gate operation, contributing to safer and more resilient reservoir management.