Simulating commercial biomass in the Hyrcanian mixed-beech stands

• Main Author: Ali Asghar Vahedi
• Format: Article
• Language: fas
• Published: Research Institute of Forests and Rangelands of Iran 2016-09-01
• Series: تحقیقات جنگل و صنوبر ایران
• Subjects: Artificial Intelligence; carbon sequestration; commercial bole; regression analysis
• Online Access: http://ijfpr.areeo.ac.ir/article_107376_595d90a441a2fc7e85d4eae7a08aabd5.pdf

The commercial bole of trees in the mixed-beech forests contributes the majority of biomass and of carbon pool, and is associated with the majority of monetary values in the Hyrcanian forests of Iran. This research aims to accurately predict commercial biomass compared to the allometric equations and field measurements in the third district of Glandroud forests in Noor. After harvesting of the trees, each part of the bole was weighed in the field and wood pieces were extracted from each part. The pieces were then oven-dried, on which the specific wood density was measured. Biomass was simulated by artificial neural network (ANN) including the FFBP network. Allometric equations (logarithmic multiple linear regressions and transformed power function models) with different parameters were examined to study the simulation uncertainty. Diameter at breast height, commercial height and specific wood density (WD) were inputs to the allometric functions and ANN simulation. Architectures of different topology of studied network including transfer functions of Log-sigmoid and Tan-sigmoid with variety of hidden layers and neuron members returned different error estimations of forest commercial biomass. Diameter was one of the most effective factors to predict biomass using ANN. Moreover, increasing height and WD in the ANN reduced the uncertainty of simulation outputs. Adding height and WD with the different combinations in the allometric models increased the accuracy of response variable prediction. The root mean squared errors (RMSE) showed that although there was slight differences in the estimation accuracies of ANN and allometric models, the optimal ANN outputs were of lower uncertainty to spatially predict the response