Tree height prediction approaches for uneven-aged beech forests in northwestern Spain

Artificial neural network methods appear to be a reliable alternative to traditional methods of tree height prediction in even-aged stands. However, this has not been demonstrated for uneven-aged forests. Two back-propagation artificial neural networks were constructed, and their performance in estimating the height of pure uneven-aged stands of common beech (Fagus sylvatica L.) in northwestern Spain was compared with that of the models most commonly used to estimate tree height (nonlinear calibrated local and generalized mixed-effects models and generalized fixed-effects models). All approaches produced accurate results, reducing the root mean squared error by more than 22% relative to basic nonlinear regression. Nonetheless, considering practical use of the models, the traditional approaches require measurement of several trees for calculation of stand-specific variables (generalized models) and for model calibration (mixed-effects models). Back-propagation artificial neural networks require less sampling effort because no height measurements are required for their implementation. However, this technique was not the best height predictor, because of the high degree of variability in site quality between stands. In this case, the local mixed-effects models yielded the best results and provided the best balance between the accuracy of the model and sampling effort.