Comparison of artificial neural network and allometric equations associated with modeling bole biomass of maple trees (Acer velutinum Bioss.) in the Hyrcanian forests

Forest trees biomass estimation with the highest accuracy is the basis of forest management with respect to the sustainable development and that is also one of the most important fundamental issues of C sequestration for international communities to work the global warming out. Modeling bole biomass of maple trees through allometric equations and artificial neural network (ANN) was carried out in Sasunsi forests of Chamestan to achieve the highest accurate prediction of studied biomass. After felling of 20 individual trees from different diameter classes, each part of bole which had been converted was weighed and one disc was taken. Then, the samples with constant volume extracted from each disk were taken to lab and they were oven-dried at 105ú C for 24 hours. To develop the model, power function was basic allometric equation and transfer function of Log-sigmoid and of Tan-sigmoid was introduced in the various topologies of network of FFBP. The results showed that the exponential multiple regression including diameter and height with correction factor of CF = 1.04 was the optimal allometric model (S = 0.23). Pertaining to the least mean squared error of test associated with training and validation of data in the different epoch as well as average standard deviation was the main indicator for selection the best model in ANN. Furthermore, the results showed that the model having input layers of diameter and height with one-hidden layer and number of 10 neurons including tansig function is the best model (S = 0.1) for bole-mass prediction.