Estimating the beech forest site productivity in Hyrcanian forest using classification and regression tree algorithm

In this study, the site form index which is the most reliable criterion for evaluation of forest site productivity in uneven-aged and mixed stands was used. For this purpose, random-systematic sampling method was used to locate 105 0.1 ha circular sample plots in beech dominated forests in Tarbiat Modares University research forest. The height and diameter of Fagus orientalis Lipsky trees within each sample plot was recorded along with elevation, azimuth and slope of the ground. Also, at the center of plot, soil samples from first layer (0-10 cm) were taken for analyzing several soil variables. Evaluation of forest site productivity by using classification and regression tree algorithm showed that after pruning the full tree, phosphorus, TRASP, clay and bulk density are effective variables, in order of relative importance, on site form and 62% variations in productivity can be explained by these variables. Using generalized linear model and evaluation criteria such as AIC, RMSE, R2 and adjusted R2, the performance of CART model was assessed. The results showed though CART techniques and the generalized linear model justify the same variability in forest productivity but decision tree technique in terms of AIC and BIC criteria is better than the GLM and as well as this technique is easier to interpret.