Inversion exploration on forest aboveground biomass of optical and SAR data supported by parameter optimization method

  Objective  Forest is the largest storage of organic carbon in the whole terrestrial carbon cycle system. Accurate estimation of forest aboveground biomass (AGB) is essential for global carbon storage analysis and estimation. This paper aims to explore the potential of optical and synthetic aperture radar (SAR) data for forest AGB inversion. In this study, Landsat8 OLI, GF-1 data were selected as optical data and advanced land observing satellite (ALOS)-1 phased array type L-band synthetic aperture radar (PALSAR)-1 data was selected as SAR data.

  Method  Firstly, band ratio parameters, vegetation index parameters, and texture information were extracted from Landsat8OLI, gaofen-1 optical data and ALOS-1 PALSAR-1 SAR data, respectively. While polarization decomposition information was also extracted from ALOS-1 PALSAR-1 SAR data. Then these parameters extracted from different remote sensing data were sorted according to their importance by random forest (RF) algorithm. Finally, fast iterative feature selection method for k-nearest neighbor (KNN-FIFS) algorithm was used to analyze different feature combinations, and four models were constructed to estimate forest AGB and a cross-validation method was applied for result validation.

  Result  These remote sensing data were modeled to estimate forest AGB using four characteristic parameters: vegetation factor, band ratio, texture factor and polarization decomposition information. For parameters extracted from vegetation factor, band ratio, texture factor, Landsat8 OLI data performed best than GF-1 and PALSAR-1 data with R² = 0.50, RMSE = 33.34 t/ha. For GF-1 data, R² was 0.36, RMSE was 37.60 t/ha, R² was 0.45, RMSE was 35.40 t/ha for PALSAR-1 data. However, for parameters extracted from polarization decomposition, PALSAR-1data showed better performance with R² = 0.63 and RMSE = 28.84 t/ha.

  Conclusion  When the parameter extraction methods are the same, the forest AGB inversion potentials of the optical and SAR data are similar. However, when the parameter extraction method is different, for example, using polarization decomposition to extract parameters, the SAR data show better performance for forest AGB inversion.