Coimbatore Reservoir Level Predictor and Early Drought Warning System

The Coimbatore's water supply depends on rainfall during monsoon season as well as variations between Siruvani and Pillur reservoir levels. The current-level display on a dashboard helps report purposes; however, there isn't much opportunity available for decision-making during times of falling storage levels. The article describes an example model for predicting short term water levels of reservoirs along with its interpretability as a drought warning index. Daily precipitation data such as rainfalls amount to temperatures, water inflows/outflows levels are recorded while also providing lags and moving averages for analysis purposes along with predicting a 7 day variation of reservoir height. Linear Regression, Random Forest and XGBoost models will be compared on a 60-20-20 ratio. There are 120 synthetic daily records for each of Siruvani and Pillur from June 1st through July 30th, 2024. The Random Forest had the smallest validation error at 0.0190 m for Siruvani and 0.0237 m for Pillur. The performance was poor on the last few tiny samples with low R² scores for testing. The report is intentionally written to show how this project is just an example of software technology rather than a functioning water management program. Transparent 0-100 drought index is a measure of rainfall deficit, storage level, recent decline and future projection which can be converted to Normal, Watch, Warning or Severe Risk levels. The complete application uses a FastAPI backend, SQLite storage, role-based access, CSV ingestion, interactive charts and map-based reservoir views. This paper shows how to build a complete pipeline for detecting issues at their source stage of development including data collection requirements, calibration procedures and validation tests needed prior to release into production environments.