Improved Burned Area Detection Using Dual-Pol Sentinel-1 Data and Machine learning Approach: Random Forest and XGBoost

Paper: Geomatics

Dataset: Mendeley

🛠 Project Setup

Clone the repository:

git clone https://github.com/yourusername/burned-area-mapping.git

Move to the project directory

cd burned-area-mapping

Create conda environment

conda env create -f environment.yml

Activate environment

conda activate wildfire_burnet_env

Implementation

Machine Learning folders consists of notebooks for following tasks:

1. Data preparation

This notebook handles the initial data preparation steps, including generation of ground truth, cropping the data into same extent, and creating train and test tile selection images.

Splitting the dataset into training and testing sets.

2. Feature Extraction

Extract the SAR related features such as RVI, RBD, RBR etc.

3. Training models

This notebook performs hyperparameter tuning using techniques like grid search, evaluates model performance with metrics such as accuracy, precision, recall, and F1-score, and saves the best model for future predictions.

3. Prediction

his notebook loads the trained model, applies it to new SAR images to generate burned area predictions, and exports the results as GeoTIFF and PNG files for further GIS analysis.

4. Feature Importance

Visualize which input features contribute the most to the predictive performance of the trained machine learning model

Old Archives: Handout

Presentation Slides

Pre-processed Dataset