Assessment of the interactions among net primary productivity, leaf area index and stand parameters in pure Anatolian black pine stands: A case study from Türkiye

• BULUT, Sinan ^[1] ; GÜNLÜ, Alkan ^[1] ; KELES, Sedat ^[1]
  1. [1] Department of Forest Management and Planning, Faculty of Forestry, Çankırı Karatekin University, 18200 Çankırı, Türkiye
• Localización: Forest systems, ISSN 2171-5068, ISSN-e 2171-9845, Vol. 32, Nº. 1, 2023
• Idioma: inglés
• Enlaces
  • Texto completo (pdf)
• Resumen

  • Aim of study: To examine the relationships between net primary productivity (NPP) and leaf area index (LAI) and modeling these parameters with stand parameters such as stand median diameter (dg), dominant height (htop), number of trees (N), stand basal area (BA) and stand volume (V).

    Area of study: Pure Anatolian black pine (Pinus nigra J.F. Arnold) stands in semi-arid climatic conditions in the Black Sea backward region of Türkiye.

    Material and methods: In this study, the Carnegie-Ames-Stanford Approach model was used to calculate NPP; LAI, dg, htop, N, BA and V were calculated in 180 sample plots. The relations between NPP and LAI with stand parameters were modeled using multiple regression analysis, support vector machines (SVM) and deep learning (DL) techniques. Relationships between NPP and LAI were investigated according to stand developmental stages and crown closure classes.

    Main results: The highest level of relations was obtained in the stands containing the a-b developmental stages (r=0.84). The most successful model in modeling NPP with stand parameters was obtained by DL method (model R2=0.64, test R2=0.51). Although DL method had higher success in modeling LAI with stand parameters, SVM method was found to be more successful in terms of model-test fit, and modeling success in independent data set.

    Research highlights: Grouping parameters affecting NPP and LAI increased the level of correlation between them. In modeling NPP and LAI in relation to stand parameters, machine learning algorithms performed better than linear approach. The overfitting problem can be eliminated substantially by including arguments such as early stopping, network reduction and regularization in the network structure.