Prediction of total tree height in Eucalyptus plantations in the Lower Amazon, Pará, using two artificial intelligence techniques

This study evaluated the efficiency of two artificial intelligence techniques, Artificial Neural Networks (ANN) and Random Forest (RF), for predicting the total height of Eucalyptus trees in clonal plantations of Eucalyptus urophylla × Eucalyptus grandis managed under a coppice regime in the Lower Amazon region of Pará, Brazil. Data from 3372 trees collected from 253 permanent plots measured between 2020 and 2022, located in the municipalities of Almeirim and Vitória do Jari, Pará, were used. The variables employed in the models included diameter at breast height (DBH), age, dominant height, clone, spacing, and area. The models were trained and evaluated using different combinations of input variables. Model performance was assessed using the correlation coefficient between observed and estimated values, root mean square error (RMSE), and graphical residual analysis, as well as the relative contribution of variables by Garson’s method for ANN and percent increase in mean square error (%IncMSE) for RF. The results showed high performance for both techniques, with correlation coefficients ranging from 0.93 to 0.99 and RMSE% below 10% in both the training and validation sets. RF showed slight superiority in some configurations, reaching RMSE% values below 7%. The relative importance analysis showed DBH and dominant height as the most influential variables in predicting total height, followed by age, while clone, spacing, and area made complementary contributions. It is concluded that ANN and RF are efficient techniques for estimating total height in the study region and may support forest management with gains in accuracy, time savings, and cost reduction.