Enhancing precision in quantification and spatial distribution of logging residues in plantation stands

Abstract

Forests, essential components of ecosystems, are managed for sustainable timber production in forest plantations to meet the growing demand for wood products. The intricate balance between sustainable forest management and logging residue management practices is crucial for ecological integrity and economic viability. Logging residues, byproducts of timber harvesting, significantly influence carbon and nutrient cycling, soil structure, and overall ecosystem health. Recent technological advancements, particularly the use of drones integrated with artificial intelligence, enable the processing of large datasets, providing meaningful insights into logging residues and forest dynamics. This study aims to evaluate the quantification and distribution of logging residues in forest plantations, utilizing machine learning classification models fed with drone-based images. The classification was performed using a Random Forest model fed with spectral and terrain variables, whereas the volume estimations were derived from field measurements and from the drone classification. Overall the classification achieved solid results (Overall Accuracy of 0.89), and the volume estimation resulting in solid comparison with field estimation (ratio 0.72–1.98), but poor correlation (R² of 0.26 and 0.36). We concluded that the proposed methodology is suitable for classifying and assessing residues distribution over recently harvested areas, but further improvement of the volume estimation methodology is necessary to ensure comprehensive and precise assessment of residue distribution over recently harvested areas.