Estimating Crimean juniper tree height using nonlinear regression and artificial neural network models

Abstract

Artificial neural network models offer a number of advantages including the ability to implicitly detect complex nonlinear relationships between input and output variables, which are very helpful in tree height modeling. Back-propagation artificial neural network models were produced for individual-tree height estimation and the results were compared with the most used tree height estimation methods. Height diameter (h–d) measurements of 1163 Crimean juniper trees in 63 sample plots from southwestern region of Turkey were used. A calibrated basic h–d mixed model, a generalized h–d model and back-propagation artificial neural network h–d models were constructed and compared. When the variability of the h–d relationship from stand to stand can be incorporated into the model, then both mixed-effects nonlinear regression and back-propagation neural network modeling approaches can produce accurate results, reducing the root mean squared error by more than 20% as compared to a basic nonlinear regression model. The use of a generalized h–d model also showed reliable results (reduction of 13% in root mean squared error as compared to a nonlinear regression model). The back-propagation artificial neural network model seems a reliable alternative to the other methods examined possessing the best generalization ability. Further, from a practical point of view it has the advantage that no height measurements are needed for its implementation. On the contrary prior information is required for the mixed-effects model calibration which is a limiting factor according to its use.