Accuracy of Deep Learning-Based Satellite Image Analysis in Early Detection of Insect Infestation-Induced Tree Mortality: A Comparative Analysis with Conventional Remote Sensing Methods
A comparative analysis finds deep learning outperforms conventional methods in detecting tree mortality due to bark beetles, indicating a need for updated monitoring systems.
Key Points
The research investigates the effectiveness of deep learning-based satellite image analysis in detecting tree mortality caused by bark beetle infestations compared to traditional methods.
Conducted in the Bohemian Forest using multitemporal Sentinel-2 imagery from 2019-2022.
Compared U-Net model with conventional techniques like NDVI/NDMI thresholding and machine learning methods.
Employed a five-fold spatial cross-validation framework for accuracy assessment.
U-Net model achieved 91.4% overall accuracy and a Kappa coefficient of 0.88.
The F1-score difference between U-Net and Random Forest increased significantly from 3 to 21 percentage points between infestation stages.
Model focused primarily on SWIR and red-edge bands for effective detection during the early green-attack stage.