Rainfall Prediction in Himalayan Region
This project develops an interactive dashboard for 3-hour ahead rainfall forecasting in the Northwest Himalayan (NWH) region using the SARITA (Spatio-temporal Attention driven Rainfall Inference using Transformative Architecture) model. The dashboard provides real-time visualization of predicted rainfall patterns, spatial correlation analysis, and extreme event detection.
Technologies Used
PythonPyTorchScikit-learnNumPyTensorFlow
Project Details
Built a 3-hour ahead rainfall forecasting system for the Northwestern Himalayan region using the SARITA (Spatio-temporal Attention Driven Rainfall Inference using Transformative Architecture) model.
Implemented spatio-temporal deep learning architecture combining Transformer attention, Deformable ConvLSTM, and Moran’s I spatial autocorrelation to capture rainfall propagation patterns.
Developed an interactive web dashboard using React.js, Leaflet.js, Plotly, and Django REST APIs to visualize real-time rainfall heatmaps and forecasts.
Designed automated ERA5 weather data ingestion pipeline using CDS API, CRON jobs, and NetCDF processing for hourly precipitation data.
Implemented extreme rainfall detection algorithms using percentile and z-score methods to identify high-risk rainfall events.
Built a Vulnerability Index (VI) combining extreme frequency, intensity, and maximum rainfall to highlight disaster-prone regions.
Conducted spatio-temporal correlation analysis of the 2025 Uttarkashi flash flood, identifying ~6-hour upstream rainfall lag from Dharali to Uttarkashi.
Enabled real-time monitoring, anomaly detection, and flood early-warning insights for mountainous terrain.