Feature selection and modeling forest tree canopies using supervised and unsupervised neural network self-organizing maps (case study: District 2, Kacha, Rasht, Iran)

Canopy is a component of gross primary production, and the corresponding dimensions reflect tree health. There is a need to study canopies in the forests of northern Iran, in particular the Hyrcanian Forests, due to their unique biodiversity, endangered conditions, and their role in climate moderation. The sampling was executed using a systematic random method with grid dimensions of 150 × 200 meters. In these circular sample plots, each covering an area of 0.1 hectares, the sampling intensity was designated at 3.3%.. Within each plot, in addition to recording topographical attributes such as elevation, slope, aspect, and of trees greater than 7.5 centimeters(DBH) essential data was gathered. The current study aims to use the SSOM neural network to estimate forest tree canopies in the District 2, Kacha using self-organizing maps (SOM)-selected variables. The SOM neural network results reveal the significant role of the elevation, slope, aspect, and diameter at breast in the map structure. After selecting major features affecting tree canopies with the SOM neural network, elevation, slope, aspect, and diameter at breast variables were introduced to the supervised self-organizing maps (SSOM) neural network to estimate Fagus Orientalis Lipsky , Carpinus betulus L. , Diospyros lotus L. , Alnus subcordata CAM , and Parrotia persica (DC) CAM tree canopies. The result show that the SOM neural network focuses on key factors to increase modeling efficiency by removing unnecessary data and improving prediction accuracy by ensuring the use of selected variables. Further more, the strong performance of SSOM neural network in tree canopy estimation, particularly Fagus Orientalis trees, by utilizing SOM-selected features. It further highlighted the network's ability to use selected features for accurate and reliable estimations.