Sensitivity analysis and performance evaluation of neural networks for predicting forest stand volume - A case study: District 2, Kacha, Guilan province, Iran

Sima Lotfi Asl, Iraj Hassanzad Navroodi and Aman Mohammad Kalteh
Additional contact information
Sima Lotfi Asl: Department of Forestry, University Campus, University of Guilan, Rasht, Iran
Iraj Hassanzad Navroodi: Department of Forestry, Faculty of Natural Resources, University of Guilan, Sowmehsara, Iran
Aman Mohammad Kalteh: Department of Rang and Watershed Management, Faculty of Natural Resources, University of Guilan, Sowmehsara, Iran

Journal of Forest Science, 2024, vol. 70, issue 5, 209-222

Abstract: Tree volume is a characteristic used in many cases, such as determining fertility, habitat quality, growth size, allowable harvesting, and the principles of forest trade. It is imperative to develop methods that predict forest stand volume to obtain this extensive information quickly and cost-effectively. This study used supervised self-organising map (SSOM), multi-layer perceptron (MLP), and radial basis function (RBF) neural networks to predict forest stand volume based on physiography, topography, soil, and human factors. A sensitivity analysis method called the importance of prediction was used to determine how input variables influenced network output. First, the map of homogeneous units was prepared with ArcMap (Version 10.3.1, 2015) by combining digital layers to measure the tree's volume per hectare. Then, separate tree species in different diameter classes were measured in a circular grid of 200 m × 150 m, 0.1 ha of coverage, 3.3% sampling intensity, and a diameter at breast height (DBH) greater than 7.5 cm using systematic sampling on a homogeneous unit map in a regular random method. The neural network modelling results showed that SSOM, MLP, and RBF predicted forest stand volume most accurately according to physiography, topography, soil, and human factors. Furthermore, the sensitivity analysis results found that altitude above sea level, soil depth, and slope are the most influential input variables. In contrast, soil texture variables are the least effective at predicting forest stand volume.

Keywords: best matching unit; error back-propagation; importance of prediction neighbour function; spread; supervised learning (search for similar items in EconPapers)
Date: 2024
References: View references in EconPapers View complete reference list from CitEc
Citations:

Downloads: (external link)
http://jfs.agriculturejournals.cz/doi/10.17221/111/2023-JFS.html (text/html)
http://jfs.agriculturejournals.cz/doi/10.17221/111/2023-JFS.pdf (application/pdf)
free of charge

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:caa:jnljfs:v:70:y:2024:i:5:id:111-2023-jfs

DOI: 10.17221/111/2023-JFS

Access Statistics for this article

Journal of Forest Science is currently edited by Mgr. Ilona Procházková

More articles in Journal of Forest Science from Czech Academy of Agricultural Sciences
Bibliographic data for series maintained by Ivo Andrle ().