Estimating quantitative forest attributes using Pleiades satellite data and non-parametric algorithms in Darabkola forests, Mazandaran

Knowledge on quantitative forest attributes is a prerequisite for forest stand management. The aim of this study was to evaluate high resolution Pleiades data in estimating the standing volume and basal area using non-parametric algorithms in Darabkola forest of Sari, Mazandaran province. A sampling design of 144 plots each with area of 1000 m2 was established using a systematic random sampling method. In each plot, information including as position of plot center, diameter at breast height of all trees within sample plot and height of selected trees were recorded, based on which the standing volume and basal area per ha were derived. The Pleiades data was preprocessed, and the pixel grey values corresponding to the ground samples were extracted from spectral bands. These were further considered as the independent variables to predict the standing volume and basal area per ha. Modeling was carried out based on 70% 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 30% samples. Support vector machine comparatively retuned the best estimates for stand basal area with root mean square error of 38.75% and relative bias of 3.12, while it predicted the stand volume with root mean square error of 45.13% and relative bias of -3.21 as well. The results of study proved the average spectral and spatial capability of Pleiades data to estimate these two main, where the caveats are concluded to be mainly due to the heterogeneity and the density of forest stands across the study area.