wildfire-prediction

This repository contains the code for a two-stage learning framework for wildfire forecasting under partial observability.

AI-powered wildfire detection and risk mapping using U-Net on satellite imagery with an interactive Streamlit dashboard.

Near-term wildfire risk forecasting platform for California using H3 geospatial indexing, NASA FIRMS, AlphaEarth embeddings, weather features, baseline ML models, and a map-ready dashboard.

Survival-probability model predicting wildfire threat to evacuation zones at 12/24/48/72h using only the first 5 hours of fire data — built for WiDS Datathon 2026.

Wildfire risk prediction models using ERA5 climate data, NDVI, and machine learning (XGBoost, Deep Learning). Final dissertation project at the University of Geneva.

Spatiotemporal wildfire risk prediction using NDVI + terrain features and a Genetic Algorithm–optimized ML pipeline.

Survival modeling based wildfire time to threat prediction using CV bagged Gradient Boosting Survival Analysis and IPCW weighted LightGBM. Hybrid metric optimization combining C index and weighted Brier scores with monotonic multi horizon probability enforcement. Public LB score 0.96841.

AI forest fire prediction using ResUNet-A deep learning and cellular automata. Real-time 30m resolution mapping for Uttarakhand. ISRO BAH Hackathon 2025.

Wildfire prediction using machine learning

Automated regional wildfire risk forecasting system using open data sources

Machine learning pipeline predicting 2021 Dixie Fire spread at 30m resolution using satellite NDVI, terrain data, and a spatially cross-validated Random Forest.

Hybrid CNN-LSTM + attention wildfire risk forecasting for Alaska (Sentinel-1/2, Landsat, MODIS, ERA5) — GSoC-style MVP

Socioeconomic drivers of wildfire probability and urban smoke exposure in East & Southern Africa (2018-2024)

Wildfire prediction using dual ML approaches: classical models (Logistic Regression, Random Forest, K-NN) on the WildfireDB tabular dataset, and transfer learning CNNs (VGG16, ResNet-50, EfficientNet-B3) on satellite imagery, with EDA, Grad-CAM visualisations, and full data pipelines.

Open-source wildfire prediction & early detection — free for firefighters and everyone at risk. By Infinite Space Technologies.

Project contribution to Omdena's initiative to develop an AI-based system for wildfire spread prediction and early warning in Uttarakhand. This contribution focuses on using Land Surface Temperature (LST) data, geospatial analysis with Google Earth Engine, and Python to predict wildfire behaviors and identify potential hotspots.

Sample Machine Learning project that predicts wildfires in real-time and alerts emergency services including police, ambulance, and fire response teams.

Wildfire ignition prediction using geospatial machine learning, extending FireCastRL with population density and land cover while demonstrating that feature importance alone does not imply predictive value.

Multi-task deep learning system predicting wildfire occurrence and intensity across India from NASA MODIS satellite data using a custom Spatio-Temporal Weighted architecture.

🔥 Model wildfire risk across the U.S. using a physics-informed Genetic Algorithm for efficient, autonomous scouting in constrained environments.