Height prediction of tectona grandis trees by mixed effects modelling and artificial neural networks

This study evaluated the performance of artificial neural networks (ANNs), and a mixed linear regression model to predict the height of trees of Tectona grandis, teak, due to categorical and numerical variables. The data were originated from forest inventories carried out at 11 and 16 years of age. The prediction of the time by networks was simulated with different combinations of variables input and neurons numbers in the middle layer. We also evaluated different scenarios, by reducing the number of trees in training / adjustment. The two techniques were compared by statistical indicators and graphical analysis. Neural networks provide results similar to those of regression, with high predictive capacity, and the statistics variation was lower than 1% between the two techniques. The use of ANNs was superior in predicting the height of larger diameters trees when compared to the mixed regression.