Estimation of Annual Industrial Wood Production Level in Forestry Operations with the Artificial Neural Network

Numerous methods have been developed for estimating the production levels of forestry operations. Although these methods are classified in different ways in literature, they are fundamentally divided into two methods i.e. the quantitative and the qualitative method. In estimation studies, one of the alternative methods used instead of the traditional one is the Artificial Neural Network (ANN).

In this study, we attempted to determine the utility of the ANN method in predicting the industrial wood production yield in forestry operations according to the allowable cut. In this context, we utilized a set of variables described in the literature as influencing industrial wood production yield relative to allowable cut. These variables, which can all be measured on the basis of production units, were organized in 3 main groups; the general conditions of the stand, the natural structure of the production unit, and the production methods and tools. Using this set of variables and the Multi-Layer Perceptron (MLP) technique, various ANN models were developed for testing different degrees of learning and momentum coefficients.

Based on the estimations performed within the scope of this study with different ANN models, we were able to identify the model which provided predictive values closest to the real values during percentage of yield estimations. During the study, it was determined that the model which performed estimations for the industrial wood yield percentage at the Günye Forestry Operations Directorate that were the closest to the actual values, with a MAPE value of 5.3%, which was the multilayered ANN model with 26 input variables, 8 neurons, 0.4 degrees of learning, and a 0.8 momentum coefficient. The error rate for the total yield percentage of the 10 production units used for testing purposes in the model was 1.7%, which is a fairly low and acceptable value. Based on these results, it is possible to state that ANN applications and models have an important potential value for use in the estimation of the yield percentage relative to allowable cut forestry operations.