Turkish Journal of Agriculture and Forestry




Forest management and sustainable timber production rely on forest measurements that include tree height. However, the fieldwork needed for tree height measurements is time-consuming, and several times, hard to obtain. The solution to this problem is the construction of reliable height-diameter (h-d) models that can provide accurate tree height predictions. To this direction, a modified Gompertz model that included dominant height and diameter at breast height was used to predict tree height. In order to investigate the most promising modelling method, six alternative approaches were evaluated; these involved both regression and artificial intelligence techniques and produced the (i) fixed-effects (FE), (ii) mixed-effects (ME), (iii) three-quantile regression (3QR), (iv) five-quantile regression (5QR), (v) general regression neural network (GRNN), and (vi) support vector regression (SVR) models. The accuracy of the developed h-d models was studied in this work for mixed-oak stands (Quercus cerris L., Quercus petraea (Matt.) Liebl. and Quercus frainetto Ten.) in Türkiye. For this purpose, 1735 trees were measured in 52 sample plots in total. The tree variability among sampling plots was incorporated in the constructed models through the dominant heights and the tree height variance in each sample plot, while the models were localized using one to five oak trees, as calibration samples, per plot. The study showed that the ME constructed model with root mean square value equal to 1.8331 was constantly superior to nonlinear regression methods used, while the GRNN and the SVR models with root mean square values equal to 1.8330 and 1.8279, respectively, showed similar predictive ability with the later to prevail, as compared to the rest of the models tested. Finally, five trees per plot were found to be an acceptable trade-off between sampling cost and predictive capacity and reliability.


Mixed-stands, tree height, mixed model, generalized regression neural network, support vector machine, calibration

First Page


Last Page