Vegetation change detection and recovery assessment based on post-fire satellite imagery using deep learning

Area of Science:

• Ecology
• Remote Sensing
• Artificial Intelligence

Background:

• Wildfires significantly alter ecosystems and vegetation.
• Earth-observation data is crucial for monitoring post-fire vegetation changes.
• Existing methods may lack accuracy in assessing vegetation recovery.

Purpose of the Study:

• To introduce a novel methodology for evaluating post-fire vegetation effects.
• To accurately detect and classify vegetation changes after wildfires using AI.
• To map vegetation recovery in fire-affected areas.

Main Methods:

• Utilized Deep Embedded Clustering (DEC), an unsupervised method, for vegetation change detection.
• Employed Enhanced Vegetation Index (EVI) trend analysis to quantify greening and browning fractions.
• Applied Adaptive Generative Adversarial Neural Network (AdaptiGAN) for vegetation recovery mapping.

Main Results:

• Achieved 96.17% accuracy in classifying vegetation change.
• Quantified greening fractions (0.1–22.4 km²) and browning fractions (0.1–18.1 km²).
• AdaptiGAN demonstrated a low training error of 0.075 in vegetation recovery assessment.

Conclusions:

• The proposed methodology offers a significant advancement in post-fire vegetation analysis.
• The AI-driven approach provides accurate and detailed insights into ecosystem recovery.
• This study highlights the potential of advanced machine learning in ecological monitoring.