Predicting stem tapers using artificial neural networks in mixed beech-fir stands in Karabuk

Development of artificial neural network models to estimate stem taper of Oriental beech and Kazdağı fir growing in mixed stands distributed in Karabük Region and comparison of the developed models and stem taper functions are objectives of this study. Measurements obtained from 516 sample trees (238 beech and 278 fir) growing in mixed stands within the boundaries of Büyükdüz Forest Enterprise were used as a study material. These measurements include tree height, diameter at stump height, diameter at breast height, and diameters at intervals of 1 m along the stem. Totally 45 artificial neural network model structures with combination of transfer functions (hyperbolic tangent transfer function, sigmoid transfer function or linear transfer function) used in hidden and output layers and the number of neurons (2, 4, 6, 8 or 10) used in hidden layer, and 4 different stem taper functions were developed in the study. The comparison of estimation performances of artificial neural network models and stem taper functions were executed by using relative rankings according to seven goodness-of fit criteria. As a result of comparisons made, it's detected that artificial neural network models are more successful in estimation of stem taper for both tree species. The most successful artificial neural network model structures are (i) the model using sigmoid transfer function in hidden layer with 10 neurons, hyperbolic tangent transfer function in output layer for Oriental beech and (ii) the model using sigmoid transfer function in hidden layer with 10 neurons, linear transfer function in output layer for Kazdağı fir. Besides, the equation developed by Kozak (2004) had the most successful estimations of stem tapers among the stem taper equations.