Anadolu Orman Araştırmaları Dergisi

Öz

The use of diameter distributions in forestry allows for more detailed estimates of the number of trees, basal area, volume, biomass, and carbon amounts in stands. In this study, the diameter distributions of Crimean pine stands in Mengen (Bolu) Forest Management Directorate were first modeled using equations based on certain percentiles based on the 3-parameter Weibull probability density function, the prediction successes were compared and then network training was carried out using Artificial Neural Networks using the most successful percentile values as input variables. In the modeling with the 3-parameter Weibull probability density function used in the first stage, the R2adj. the diameter distribution of the 25%-50%-95% equation, which was found to be the most successful, was 0.727, and the mean absolute error and root mean square error values were 202.91 and 102.95, respectively. Then, by comparing the success of various artificial neural networks (27 different models) based on activation function, number of intermediate layers, and number of neurons in predicting diameter distributions, the mean error and root mean square of error squares values of 145.34 and 78.1, respectively, and R2adj. of 0.843 were obtained with the Artificial Neural Network structure including hyperbolic tangent (tanH) activation function, 3 hidden layers, and 50 neurons. As a result, the artificial neural network modeling resulted in a 28.4% reduction in the mean error a 24% reduction in the root mean square of the error squares, and a 16% improvement in the R2adj. In addition, the success of the 25%-50%-95% method in modeling the diameter distribution in Mengen Crimean pine stands was tested with Kolmogorov-Simirnov (K-S) test, and the diameter distribution modelling was found to be statistically appropriate in 107 of 108 sample plots.