Evaluation of RapidEye satellite data for estimation some quantitative structure variables in the Caspian forests of Gorgan region

Estimation of quantitative forest attributes is important for its applications in order to understand the forest condition and performance. The aim of this study was the estimation of some quantitative forest attributes (stand volume, basal area, and tree stem density) using the RapidEye satellite data (2011) and non-parametric algorithms in the part of Hyrcanian forests in the Gorgan region. For this purpose, 418 plots each with an area of 1000m2 were established using a simple random sampling method. In each plot, information including a position of plot center, diameter at breast height of all trees and height of selected trees were recorded. Based on which the standing volume and basal area per ha were derived. A RapidEye image was processed by different synthetic bands derived from rationing, principal component analysis, texture analysis, and Tasseledcap, and the pixel gray values corresponding to the ground samples were extracted from spectral bands. These were further considered as the independent variables to predict the Quantitative characteristics. Modeling was carried out based on 75% of sample plots as training set using K-Nearest Neighbor, support vector machine, and random forest methods. The predictions were cross-validated using the left-out 25% samples. The results showed Random forest comparatively returned the best estimates for stand volume, basal area and tree stem density with root mean square error of 39.83%, 29.71%, and 50.11% and relative bias of 0.01, 1.69 and 2.11 as well, respectively. The results of this study also showed that due to the heterogeneity and density of Caspian forests, RapidEye satellite spectral data have a moderate ability to estimate the quantitative forest attributes.