Antevendo o impacto climático e de sanidade florestal no potencial produtivo de plantios de eucalipto: uma abordagem para índice de sítio baseada em machine learning

Abstract

Climate change has increasingly posed significant challenges to the forestry sector, directly affecting forest stand productivity due to prolonged droughts, increased climate variability, and changes in water regimes. In this context, this dissertation aimed to deepen the understanding of environmental effects on forest growth—particularly dominant height—through an integrated approach that combines a comprehensive literature review and the development of improved predictive models. The first part of the study critically reviewed key advances in forest growth modeling, emphasizing the potential of incorporating climatic variables, remote sensing data, and machine learning techniques to enhance estimation accuracy. The second part presented an applied study that compared different dominant height projection models for eucalyptus stands, including conventional models, models expanded with climatic variables, and hybrid models incorporating remote sensing indices. Variable selection was supported by Genetic Programming, enabling the identification of the carry-over effect, which reflects the delayed impact of past climatic events on current tree growth. The results showed that including environmental variables significantly reduced bias and improved model explanatory power, particularly under water deficit scenarios. The integration of remote sensing indices, such as SAVI, also proved valuable for representing forest health and capturing spatial variations in water availability. Therefore, this dissertation contributes methodologically by proposing a hybrid, scalable, and environmentally sensitive modeling approach, offering valuable insights to support more efficient, resilient, and climate-adaptive forest management strategies.