Enhancing active fire detection in Sentinel 2 imagery using GLCM texture features in random forest models.

Area of Science:

• Environmental Science
• Remote Sensing
• Forestry

Background:

• Human activities are a major cause of wildfires, leading to significant atmospheric pollution and risks to public safety.
• Accurate detection of active fire spots is crucial for mitigating wildfire impacts.

Purpose of the Study:

• To develop and optimize a Random Forest (RF) model for enhanced active fire spot detection using Sentinel-2 satellite data.
• To assess the contribution of spectral and texture features for improving fire detection accuracy.

Main Methods:

• Utilized Sentinel-2 satellite data for active fire detection across various Chinese land cover types.
• Employed spectral index methods, thresholding, and the Random Forest (RF) model.
• Assessed feature importance using the Gini coefficient and incorporated Grey Level Co-occurrence Matrix (GLCM) texture features.

Main Results:

• Grey Level Co-occurrence Matrix (GLCM) texture features were identified as crucial, forming 40% of the final feature set and significantly enhancing detection accuracy.
• The optimized RF model achieved an overall accuracy of 86.1% for active fire detection.
• The study demonstrated the effectiveness of the bespoke RF model across diverse land cover environments in China.

Conclusions:

• The optimized Random Forest (RF) model, incorporating GLCM texture features, provides a robust and accurate method for active fire detection.
• This approach is suitable for application in various land cover types across China, aiding in wildfire management and atmospheric pollution monitoring.