Machine learning-based assessment of leaf-cutting ant infestation in Eucalyptus forest plantations

The forestry plantation sector in Brazil covers an area of 10 million hectares and holds significant socio-economic importance for the country. While Eucalyptus forests have high productivity, their growth can be limited by pests and diseases. Leaf-cutting ants (Atta spp. and Acromyrmex spp.), native to Brazil, are considered the main pests, causing huge annual losses and high cost of monitoring and control. It is crucial to identify the sites most prone to infestation by these ants to implement more effective management strategies. In this study, we utilized machine learning models to analyze response patterns using a comprehensive database of predictor variables. Our objective was to assess the impact of edaphoclimatic and landscape factors on the establishment and growth of leaf-cutting ant nests across five distinct regions in Brazil. We employed the Random Forest model, a powerful algorithm capable of capturing complex relationships within the data. By leveraging this approach, we aimed to gain insights into the key variables influencing leaf-cutting ant populations and their nest dynamics. The models were trained using 10 years of leaf-cutting ants monitoring data from 33,182 Eucalyptus forest stands, covering a total area of 728,708 hectares across the assessed regions. A total of 97 predictor variables were used in the models. Considering the models with and without subsampling (total of 24 models), the mean accuracy for reinfestation and large nests were, respectively, 83 and 78%. The proposed models demonstrated sensitivity in detecting reinfestation and large ant nests, indicating that the edaphoclimatic and landscape conditions have varying influences on different macro-regions.