Ten-year estimation of Fagus orientalis Lipsky increment using artificial neural networks model and multiple linear regression Ramsar forests

In this research, which was done in Ramsar forests in Mazandaran province, the forest increment was estimated using artificial neural network and compared with the actual increment of the forest which was directly measured in 20 sample plots of 1 hectare in 2002 and 2012. The annual volume growth of beech was 4.52 and 4.35 m3 in hectare for direct inventory and estimation by artificial neural networks. Then regression analysis was performed to compare the results, stepwise, and the best models were selected. After selecting the best model for forecasting growth, the sensitivity analysis of inputs was investigated.The results showed that the neural network with relatively good accuracy can estimate the annual growth and cutting. The value of R2, RMSE and MAE were 0.75, 17 and 13.6 (for 20 sample plot of one hectare) respectively in the multi-layer perceptron network. Results also showed that the MLP neural network had the highest accuracy in predicting and estimation. In the analysis of multiple linear regression, the coefficients of detection were 0.610 and 0.679, respectively, and RMSE was 1.5 and 1.42, respectively, for the first and second models respectively. The volume at the beginning of the period and the diameter at breast height had the greatest impact on the amount of wood produced. Comparison of the models showed the use of neural network can predict the growth rate with proper accuracy.