Neuro-evolutionary knowledge extraction for forest health assessment using Sentinel-2 imagery: a small-scale study in Algeria

Forests play a critical role in climate regulation, carbon sequestration, biodiversity conservation, and the livelihoods of local communities. However, increasing droughts, wildfires, and climate-driven disturbances have accelerated forest degradation, particularly in vulnerable regions such as North Africa. Timely and spatially explicit detection of forest dieback is therefore essential for effective forest management. We aim to assess forest health by detecting forest dieback from satellite imagery and to identify the spectral indicators most strongly associated with ecological stress. To this end, we developed a neuro-evolutionary system that integrates convolutional neural networks and genetic algorithms to perform pixel-based, single-image classification of forest condition using Sentinel-2 multispectral imagery. We evaluated the proposed approach on two forest sites in Algeria, representative of contrasting North African ecosystems affected by recurrent droughts and episodic wildfires. Sentinel-2 spectral bands and vegetation indices were used as input features to characterize forest health conditions. Our neuro-evolutionary system achieved classification accuracies of 0.90 and 0.94 for the two study sites, demonstrating robust performance under heterogeneous ecological conditions. Compared with existing machine learning approaches reported in the literature, our proposed system improves forest dieback mapping by combining automated hyperparameter optimization, enhanced generalization capability, and interpretable outputs that relate model predictions to ecologically meaningful spectral indicators, including the normalized burn ratio, enhanced vegetation index, red-edge chlorophyll index, and normalized difference water index. These results highlight the potential of the proposed neuro-evolutionary system as a scalable, accurate, and interpretable framework for operational forest health monitoring, supporting early detection of forest degradation and informed ecosystem management in Algeria and similar environments.