Eucalyptus diameter and volume prediction with deep neural networks: a long short-term memory model approach

Abstract

Effective management of forest resources is vital for forestry companies. Accurate information is essential for planning planted forests. Therefore, forest inventory is a necessary procedure for obtaining quantitative and qualitative information about a region. Total volume serves as a critical metric for evaluating the potential of a specific area. This study presents an approach for modeling and forecasting diameters and total volume in planted forests. Diameter data at different tree heights in three different forest sites were used, and five neural network models were developed to estimate diameters and calculate volumes of eucalyptus clones. The models underwent training and testing using cross-validation techniques and were compared to statistical models. The results showed that the models based on Long Short-Term Memory (LSTM) are superiors. The LSTM models generalized tree volumes from diverse forest sites with greater accuracy than statistical models, even without site-specific training data. The percentage differences between the cubed volumes at sites (I, II, and III) and the volumes obtained by LSTM were smaller (0.01%, 1.06%, and 0.42%) in comparison to statistical models (0.35%, 1.81%, and 0.17%). These results highlight the potential of LSTM networks to provide accurate, generalizable, and non-invasive solutions for forest volume estimation, offering significant benefits for forestry management and decision-making.