Implementing inventory in remote and hard-to-reach forests is rather challenging. The study aims to develop methods for identifying qualitative and quantitative characteristics of mixed forests using Sentinel-1 imagery. Research area covered the taiga zone of Russian plane. Various forest ecosystems (in term of species composition, age, growing class, forest type etc.) were examined.

Three options of radar satellite images were analyzed: images processed with incoherent accumulation (Multilooking) and Frost filter in SNAP software, and original images without processing. To determine the relationship between forest parameters (standing volume, forest density, age, number of trees) and radar survey indicators statistical methods of multiple regression were applied. Data processing was implemented using the Scikit-learn machine learning library in Python programming environment.

Determining forest characteristics applying various pre-processing methods showed similar efficiency (R=0.7-0.8). However, the highest correlations (up to 0.86) are obtained with Multilooking procedure. Imagery processing with no filter and using BFGS neural networks exhibited the possibility of determining dominant species with a correlation coefficient of 0.6 and higher. The most accurate determination of the standing volume and forest density was acquired using multiple factor regression models.

We revealed relationships between standing volume, forest density, age, and number of trees and following radar indicators: SRCS, GammaVV, GammaVH, GLCMMean, GLCMVariance, GLCMCCorrelation, GammaVH - GammaVV (Diff), and GammaVH + GammaVV (Sum). The results were compared with forest inventory materials for a part of the study area. For most stands there is a similarity in standing volume and forest density definition. The study results demonstrate that it is possible to identify quantitative and qualitative forests characteristics using radar survey data.