Floodcast
AI-Driven Forecasting for Extreme Rainfall and Flood Events
Floodcast is a research project focused on long-lead extreme precipitation cluster prediction using deep learning models and multi-source hydrometeorological datasets. The goal is to enhance the forecasting of extreme rainfall events to support early flood warning systems and climate risk mitigation.
Objectives Floodcast aims to:
- 📊 Predict extreme precipitation clusters using spatiotemporal deep learning models.
- 🛰 Integrate multiple datasets (satellite rainfall, river discharge, soil moisture) for better flood forecasting.
- 🧠 Develop a robust deep learning model for long-lead prediction of extreme weather events.
- 🌍 Support climate resilience efforts by improving early warning systems.
Key Features
- 🌧 Extreme Precipitation Forecasting: Uses GPM, CHIRPS, and ERA5 rainfall data to detect heavy rainfall clusters.
- 🔍 Deep Learning for Long-Lead Prediction: Leverages transformer-based and recurrent neural networks (LSTMs, ConvLSTMs, Mamba) to predict future precipitation patterns.
- 🏞 Hydrological and Climate Data Fusion: Incorporates soil moisture (NASA SMAP), river discharge (USGS, GRDC), and flood extent data.
- ⚡ Spatiotemporal Model Optimization: Uses attention mechanisms, ensemble learning, and generative models (Diffusion, GANs) for improved rainfall event forecasting.