🛰️ Wildfire Dynamics Forecasting & Simulation (ISRO BAH 2025) National Top 3 | Problem Statement: Leveraging AI/ML for Unprecedented Accuracy in Wildfire Dynamics.

📌 Project Overview This repository contains the implementation of a high-resolution spatiotemporal forecasting model designed to predict and simulate wildfire propagation. By integrating multi-modal satellite data, the system provides a predictive lead-time for emergency response and resource allocation.

🏗️ Model Architecture: FireUnetMultiRes The core of this project is a custom Dual-Encoder U-Net architecture designed to handle the complexities of wildfire data:

Static Encoder: Processes time-invariant features such as Elevation (DEM) and Landcover Classification.

Dynamic Encoder: Utilizes ConvLSTM (Convolutional Long Short-Term Memory) cells to capture temporal dependencies in weather (wind speed, rainfall) and historical fire perimeters.

Multi-Res Skip Connections: Bridges high-resolution spatial features directly to the decoder to prevent information loss during downsampling.

📊 Key Features & Data Spatiotemporal Imagery: Processing of high-resolution multi-spectral satellite data.

Feature Engineering: Integration of static (topographic) and dynamic (meteorological) variables.

Simulation Engine: Generates fire spread probability maps to visualize potential expansion paths.

🏆 Achievement Awarded National Top 3 in the ISRO Bhartiya-Antariksh-Hackathon (BAH) 2025.

Recognized for advancing the state-of-the-art in wildfire simulation using high-resolution Earth Observation (EO) data.