A neuro-fuzzy model of error in directional felling operation using the subtractive clustering method

The study presents models of error estimation in trees’ directional felling according to several effective factors using the subtractive clustering in the Adaptive Neuro-Fuzzy Inference System. A total number of 95 trees in the compartment 207 of 2nd district of Nav watershed in Guilan province were felled by felling group and regardless to the group’s skill, using manual chainsaw. The difference between predicted and real falling direction of trees was measured as felling error. To generate models, twelve independent variables were assumed to be the effective factors, and the two types of learning algorithm (LA), two inference types (IT) and five types of membership function (MF) for input variables were applied through the subtractive clustering method in the ANFIS. Results indicated that the trapezoidal type of MF in combination with the first-order type of Sugeno IT and the back propagation LA had the best performance among all combinations of setting parameters. The sensitivity analysis of the optimal model showed that the model was very sensitive to the changes in terrain slope, the angles of backcut and undercut surfaces and DBH, respectively. Results also revealed that felling group properly predicted the fall direction and performed the directional felling in the steeper terrain. In addition, the increase of DBH and opening too much the undercut notch have accompanied with the increase of felling error.