Chen et al. (2025) Regional Forest Wildfire Mapping Through Integration of Sentinel-2 and Landsat 8 Data in Google Earth Engine with Semi-Automatic Training Sample Generation

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Remote Sensing
• Year: 2025
• Date: 2025-12-16
• Authors: Yue Chen, Weili Kou, Xiong Yin, Rui Wang, Jiangxia Ye, Qiuhua Wang
• DOI: 10.3390/rs17244038

Research Groups

Not specified in the provided text.

Short Summary

This study developed an FS-SNIC-ML workflow integrating multi-source optical imagery fusion, semi-automatic sample generation, and object-based machine learning to accurately map burned forest areas in mountainous regions and identify wildfire driving factors. The workflow achieved high classification accuracies, with Random Forest performing best, and identified key environmental drivers of wildfire hotspot density.

Objective

• To develop an effective and transferable workflow (FS-SNIC-ML) for high-precision burned-area extraction in mountainous forest regions.
• To quantify the dominant driving factors of wildfire hotspot density in complex mountainous forest regions.

Study Configuration

• Spatial Scale: Mountainous forest regions (specific area not provided).
• Temporal Scale: Pre- and post-fire periods (specific dates not provided).

Methodology and Data

• Models used:
  • Pseudo-Invariant Feature (PIFS)-based image fusion
  • Semi-automatic SAM–GLCM–PCA–Otsu procedure for sample generation
  • Feature selection using Logistic Regression (LR), Random Forest (RF), and Boruta algorithms
  • Object-Based Image Analysis (GEOBIA) with Simple Non-Iterative Clustering (SNIC) superpixels
  • Classifiers: Random Forest (RF), Support Vector Machine (SVM), Classification and Regression Tree (CART), K-Nearest Neighbors (KNN)
  • K-means clustering
  • Geographical detector analysis
• Data sources:
  • Satellite imagery: Sentinel-2
  • Satellite imagery: Landsat 8

Main Results

• The FS-SNIC-ML workflow successfully integrated multi-source optical fusion, semi-automatic sample generation, feature selection, and object-based machine-learning classification for reliable burned-area mapping.
• Feature selection identified dNBR, dNDVI, and dEVI as the most discriminative variables for burned area mapping.
• Among the evaluated classifiers within the SNIC-supported GEOBIA framework, Random Forest (RF) performed best.
• RF achieved overall accuracies of 92.02% for burned areas and 94.04% for unburned areas, outperforming SVM, CART, and KNN.
• K-means clustering of dNBR revealed spatial variation in fire conditions.
• Geographical detector analysis showed that Normalized Difference Vegetation Index (NDVI), temperature, soil moisture, and their pairwise interactions were the dominant drivers of wildfire hotspot density.

Contributions

• Development of an effective and transferable FS-SNIC-ML workflow for high-precision burned-area extraction in complex mountainous forest regions.
• Integration of multi-source optical fusion (Sentinel-2 and Landsat 8) to generate cloud-free, gap-free, and spectrally consistent pre- and post-fire reflectance datasets.
• Introduction of a semi-automatic SAM–GLCM–PCA–Otsu procedure for robust and spatially representative sample generation.
• Comprehensive evaluation of multiple machine learning classifiers within an object-based framework for burned area mapping.
• Quantification of dominant wildfire-driving factors (NDVI, temperature, soil moisture, and their interactions) using geographical detector analysis.

Funding

Not specified in the provided text.

Citation

@article{Chen2025Regional,
  author = {Chen, Yue and Kou, Weili and Yin, Xiong and Wang, Rui and Ye, Jiangxia and Wang, Qiuhua},
  title = {Regional Forest Wildfire Mapping Through Integration of Sentinel-2 and Landsat 8 Data in Google Earth Engine with Semi-Automatic Training Sample Generation},
  journal = {Remote Sensing},
  year = {2025},
  doi = {10.3390/rs17244038},
  url = {https://doi.org/10.3390/rs17244038}
}