An operational inflow forecasting system for Tehri Dam: system design, deployment and real-world performance

Abstract

Inflow forecasting is essential for optimising hydropower generation and reservoir safety. This study presents the design, deployment, and evaluation of an indigenous, real-time inflow forecasting system for the Tehri Dam (2,400 MW) in India. The system integrates a real-time hydro-meteorological network with a hybrid modelling framework: A Geomorphological Instantaneous Unit Hydrograph (GIUH)-based Nash model for 16 ungauged tributaries and stochastic Auto-Regressive (AR) and Auto-Regressive with Exogenous inputs (ARX) models for gauged sub-catchments. Evaluation of 2,618 operational forecasts from 2018 to 2026 demonstrates high predictive accuracy, with Nash-Sutcliffe Efficiency (NSE) values ranging from 0.78 to 0.97 across different seasons. Specifically, 84.53% of forecasts achieved a relative error within ±20%, and 89.58% maintained an absolute inflow difference within 60 m3/s. The system provides a parsimonious and cost-effective alternative to complex platforms like the Hydrologic Engineering Centre (HEC) or MIKE software suites, which often require more extensive data and computational resources. By enabling a 23.1% improvement in NSE over seven years through daily parameter updating, the system supports adaptive reservoir management under increasing hydrological variability. These results confirm that the integrated stochastic-geomorphological approach is a robust tool for real-time disaster management and sustainable hydropower operations in Himalayan basins.