J. For. Sci., 2023, 69(10):415-426 | DOI: 10.17221/93/2022-JFS

Estimation of Fagus orientalis Lipsky height using nonlinear models in Hyrcanian forests, IranOriginal Paper

Mohammad Rasoul Nazari Sendi¹, Iraj Hassanzad Navroodi¹, Aman Mohammad Kalteh²

¹ Department of Forestry, Faculty of Natural Resources, University of Guilan, Sowmehe Sara, Iran

² Department of Range and Watershed Management, Faculty of Natural Resources, University of Guilan, Sowmehe Sara, Iran

Tree height is one of the most important variables in describing forest stand structure. However, due to difficulty in height measurement, especially in dense and mountainous forests, the common approach is to invoke the height-diameter (H-D) models. The oriental beech (Fagus orientalis Lipsky) is one of the most important species of Hyrcanian forests, over the mid to high-altitudes (400–1 800 m a.s.l.), in northern Iran. In this study, the H-D relationship of beech trees was investigated separately for mid-altitude and high-altitude in Shafaroud forests of Guilan using 14 nonlinear H-D models and an artificial neural network model (ANN). To collect data, a systematic random sampling method within a 100 × 100 m regular randomized grid was applied. In total, 3 243 individual trees in 255 circular plots with 0.1 ha were measured. For comparing the results, performance criteria including root mean square error (RMSE), R²_adj, Akaike's information criterion (AIC), and mean absolute error (MAE) were used. In high and mid altitudes, Meyer (1940) and Bates and Watts (1980) models had the best performance, while Watts (1983) model and Burkhart-Strub (1974) model had the worst performance in high-altitude and in mid-altitude, respectively. On the other hand, the ANN model had the best accuracy and performance in both sites. Since the performance of the ANN model is superior and consistent compared to the common nonlinear models, here it is preferred for both regions.

Keywords: elevation; height-diameter modelling; neural network; oriental beech; Shafaroud

Received: July 18, 2022; Accepted: July 17, 2023; Prepublished online: October 12, 2023; Published: October 30, 2023  Show citation