Biomass is a crucial component of the carbon cycle; thus, accurate evaluation is essential to manage the forest and understand its role in climate change. Biomass estimation also supports the international reduced emission from deforestation and forest degradation (REDD), including cases such as deforestation reduction, sustainable management of forests, protection and enhancement of forest carbon reserves). Today, using remote sensing techniques with the help of field data has revolutionized the estimation of forest biomass. Forest biomass estimation can be based on the processing of remote sensing data obtained from active sensors (for example, lidar and radar) and passive sensors (for example, optical sensors). In most previous studies, mainly vegetation indices (such as Normalized Difference Vegetation Index (NDVI), Ratio Vegetation Index (RVI), and Soil-Adjusted Vegetation Index (SAVI)) have been used to estimate biomass. Using terrain data, the amount of biomass is estimated using allometric equations, and the required pre-processing is done on optical and radar images. The attributes obtained from the scattering matrix and the ratios of the components of the scattering matrix and the attributes obtained from H/α decomposition are extracted from the radar image and the attributes of vegetation, soil and water are extracted from the optical image.
Results and discussion:
In this study, in order to improve the accuracy of estimating the biomass of forest areas, the features extracted from the optical images of Sentinel-2 sensor and Sentinel-1 radar data as well as field data of Noor forest areas, Mazandaran province, whose forest cover type, Carpinus betulus and Quercus Castaneifolia and also includes rare species such as Populus Caspica Bornm trees. In this study, we used the genetic optimization method in four classes of mixed vegetation, natural forest, degraded forest, and forest reserves were studied. In this regard, multivariate linear regression and support vector regression have been used to model between ground data and radar and optical features. Genetic algorithm (GA) is one of the most common evolutionary algorithms. This method finds potential solutions to optimize problems at the right time, especially when the search space is very wide. Also, a genetic algorithm has been used during the modeling process using multivariate linear regression to select the optimal features extracted from radar and optical images. Evaluation of the results showed that the use of the multivariate regression method led to more accurate results than the support vector regression method in the study area. Also, evaluation of the results showed that using features selected by a genetic algorithm led to an accurate R2 of 0.78, 0.87, 0.68, and 0.79 for first to fourth vegetation classes, respectively. Therefore, the results showed that the efficiency of the genetic algorithm in feature selection for biomass estimation from satellite images using the multivariate regression method.