An Automated, Self-Supervised Machine Learning Framework for Near-Real-Time Wildfire Burned Area Mapping using Multi-Source Earth Observation

Launch GEE App »

Overview · BAM App · Features · Code Access · Citation · Contact

BAM (Burn Area Mapper) represents a paradigm shift in Earth observation for disaster management, moving from historical post-fire assessments to highly automated, near-real-time wildfire burned area mapping. By leveraging multi-source Earth Observation (EO) data within the Google Earth Engine (GEE) environment, BAM provides a scalable, self-supervised machine learning framework that operates without the need for extensive manual data annotation.

The framework integrates advanced atmospheric correction models (SREM) and state-of-the-art spectral indices with Gradient Tree Boosting algorithms. Through intelligent weak labeling driven by dynamic Otsu thresholding, BAM automates the generation of training data, achieving robust cross-validated classification accuracy across diverse biomes globally.

Project Status

The manuscript describing this methodology is currently under submission. The full dataset processing pipeline and source code will be made publicly available following the peer-review process and manuscript acceptance.

For early access to the codebase for validation or research purposes, please contact the author.

Visualize and interact with the BAM framework directly through our Google Earth Engine web application. The platform provides global coverage for monitoring recent and historical wildfire events using Landsat 8/9 imagery.

Click the image above to launch the BAM Web App.

• 🤖 Self-Supervised Machine Learning: Automated integration of Gradient Tree Boosting models trained on robust weak labels, eliminating the manual annotation bottleneck.
• 🛰️ Multi-Source Earth Observation: Harnesses the power of Landsat 8/9 data with advanced spectral indices designed specifically for vegetation burn severity.
• ☁️ Embedded Atmospheric Correction: Features built-in SREM (Simplified Robust Elevation Model) surface reflectance estimation for consistent multi-temporal analyses globally.
• 🏷️ Automated Weak Labeling: Employs dynamic Otsu-based thresholding for intelligent, biome-adaptive training label generation.

• 🌐 Global Operability: Designed and validated to function seamlessly across distinct fire regimes and disparate ecosystems worldwide.
• ⚡ High-Performance Deployment: Built natively on Google Earth Engine for planetary-scale computation and rapid inference.

This repository serves as the official code and documentation hub for the BAM framework.

BAM/
├── main_pipeline.py     # Orchestrates the end-to-end mapping pipeline
├── config.py            # Global configuration and hyperparameter settings
├── events.py            # Definitions and extents for validated fire events
├── modules/             # Core Python modules
│   ├── srem.py          # Atmospheric correction routines
│   ├── indices.py       # Computation of specialized spectral indices
│   ├── features.py      # Feature engineering and stack generation
│   ├── labeling.py      # Automated weak label generation algorithms
│   ├── ml.py            # Machine learning classification logic
│   └── visualization.py # Tools for rendering and exporting maps
├── notebooks/           # Jupyter notebooks for development and production
│   ├── BAM_Production.ipynb
│   └── Research_Lab.ipynb
├── gee_app/             # Google Earth Engine App deployment package
└── README.md            # Project documentation (this file)

Note: The underlying source code components in the modules/ and notebooks/ directories are presently maintained in an unpublished state to adhere to academic embargo policies prior to acceptance.

If you use the BAM framework or Web App in your research, please cite the foundational manuscript once published:

@article{waleed2026bam,
  title={BAM: An Automated, Self-Supervised Machine Learning Framework for Near-Real-Time Wildfire Burned Area Mapping using Multi-Source Earth Observation},
  author={Waleed, Mirza and Bilal, Muhammad},
  journal={Under Submission},
  year={2026}
}

Mirza Waleed (First Author & Developer)
Department of Geography, Hong Kong Baptist University
Hong Kong Special Administration Region of China

• Website: waleedgeo.com
• Email: [email protected]
• GitHub: @waleedgeo

Muhammad Bilal (Second & Corresponding Author)
Architecture and City Design Department, College of Design and Built Environment, King Fahd University of Petroleum & Minerals, Dhahran, Saudi Arabia Center for Aviation & Space Exploration, King Fahd University of Petroleum & Minerals, Dhahran, Saudi Arabia

• Email: [email protected]

The authors express their sincere gratitude to the United States Geological Survey (USGS) and the European Space Agency (ESA) for providing the Landsat 8/9 and Sentinel-2 imagery. We are deeply thankful to the Google Earth Engine team for providing the high-performance cloud computing infrastructure essential for this planetary-scale analysis.

We also acknowledge the open-access contributions of the FABDEM and WorldCover projects, which provided critical baseline datasets.

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.

Under this license, you are free to share and adapt the material, provided you give appropriate credit, do not use it for commercial purposes, and distribute any derivative works under the same license. For full license details, please visit the Creative Commons website.