Search

Research works on aboveground biomass (AGB) estimation have received attention for a long time. However, most existing research works were based on pixels at respective scale, and relatively few studies focused on estimation and multiscale transformation of AGB. We, therefore, developed an innovative object-oriented approach to estimate and transform AGB under multiple scales. First, AGB-correlated spectral, structural, and geographic indicators were derived from multisource data. Subsequently, multiresolution segmentation technology was performed to produce homogeneous geography and spectrum units (HGSUs) at different scales. Finally, AGB at each scale was retrieved based on HGSUs and Random Forest (RF) algorithm. Besides, the utilities of nonspectral variables in modeling were further evaluated. Results showed that the HGSU-oriented approach was effective and advantageous to achieve the AGB estimation and multiscale transformation based only on the same dataset with few user-defined parameters. Structural and geographic variables, especially soil type, vegetation species, and CHM, played important roles in modeling, while the contribution of spectral variables decreased with the increasing scale in general. The HGSUs combined multiple pieces of information such as spectra, texture, vegetation height, soil type, slope, elevation, and land use, and provided a more detailed segmentation, a faster stability speed with increasing regression trees, and a higher accuracy than those based on common image objects segmented only by spectral indicators. Results also evidenced that the RF regression model had the capability to ingest mixed data. This study supplemented the existing AGB estimation research works especially for shorter vegetation in coastal areas (relative to forests), and the proposed approach was promising in larger regional scales. [ABSTRACT FROM AUTHOR]