Multissensores na detecção e quantificação de desfolha por lepidópteros desfolhadores em eucaliptais

Abstract

Eucalyptus cultivation plays an important role in the Brazilian forestry sector, being the main source of raw material for paper, pulp, and charcoal. However, productivity can be compromised by insect pest, such as defoliating lepidopterans, making the monitoring of these infestations a fundamental step for sustainable management. In this context, remote sensing stands out as a promising alternative to estimate the damage caused by defoliating caterpillars.In this study, high-resolution reference data were obtained from drone RGB images and hemispherical canopy photographs, while scenes from Sentinel-1 (SAR) and Sentinel-2 (optical) satellites provided broad spatial coverage to assess areas infested by caterpillars. Machine learning algorithms were applied to predict the extent of pest impact. Among the evaluated algorithms—Support Vector Machine (SVM) with different kernels, Random Forest (RF) in multiple configurations, and Artificial Neural Networks (ANN) with various layer and neuron structures—SVM and RF demonstrated the best performance.The lowest mean absolute error (MAE) values were achieved by the radial, polynomial, and linear versions of SVM, as well as by various parameter combinations of Random Forest. Conversely, models based on ANN showed higher error values, indicating a lower predictive capacity for canopy coverage in this context. The smallest MAE values recorded were approximately 0.10 for SVM radial, 0.11 for SVM linear, and between 0.10 and 0.12 for various RF configurations, indicating higher accuracy in predicting canopy coverage. However, ANN-based models exhibited MAE values greater than 0.30, suggesting lower performance in this scenario.This approach enables the detection and monitoring of infested areas by caterpillars, facilitating early decision-making in pest control, optimizing forest management, reducing production losses, and minimizing operational costs and environmental impacts. It also aligns with sustainability guidelines of environmental certifications such as the Forest Stewardship Council (FSC) and the Brazilian Forest Certification Program (Cerflor).