CONNECTING FLOODWATER, RAINFALL AND RESERVOIR DATA TO INTELLIGENT WATER REDISTRIBUTION FOR DROUGHT-RESILIENT RESOURCE MANAGEMENT - (AQUA RESCUE)

Abstract

Many regions of India face floods and droughts in the same year, sometimes within a few hundred kilometres of each other. During a heavy monsoon spell, reservoirs in one basin release large volumes of water to protect their dams, and much of this water flows to the sea, while districts in a neighbouring basin run short of drinking and irrigation water a few months later. The data that could help manage this imbalance, including rainfall records, reservoir storage, river gauge levels and flood extents, is collected by different agencies and is rarely analysed together. This paper presents Aqua Rescue, a data pipeline and decision support system that connects these sources to plan the redistribution of surplus water to drought-prone areas. The pipeline ingests daily rainfall from rain gauge networks and gridded products, reservoir levels and releases from water resources department bulletins, river discharge from gauging stations, groundwater levels from observation wells and flood extent maps derived from satellite radar imagery. Each source is cleaned, converted to common units and mapped to a shared spatial framework of basins, sub-basins, reservoirs and districts. A curated data warehouse keeps both the daily history and the current state of every reservoir and district, and data quality checks catch missing bulletins, unit errors and sudden unrealistic jumps in storage values. Machine learning models use this curated data in three ways. A Long Short-Term Memory network forecasts reservoir inflows for the next fourteen days from recent rainfall, upstream discharge and catchment soil moisture. A gradient boosting model estimates the water demand of each district for drinking, irrigation and industry using population, cropping pattern and temperature data. A drought risk score is computed from the Standardised Precipitation Index, groundwater decline and storage deficit. The forecast surplus and the estimated deficits are then passed to a linear programming optimiser that recommends how much water to move through existing canals, link channels and pumping schemes while respecting capacity, flood cushion and minimum environmental flow constraints.