Forest Aboveground Biomass and Forest Height Estimation Over a Sub-tropical Forest Using Machine Learning Algorithm and Synthetic Aperture Radar Data

Forest aboveground biomass (AGB) is a key measurement in studying terrestrial carbon storage, carbon cycle, and climate change. Machine learning based algorithms can be applied to estimate forest AGB using remote sensing-based data. Our study utilized L-band ALOS-2/PALSAR-2 Synthetic Aperture Radar (SAR) data in combination with multi-parameter linear regression (LR) and Random forest regression (RF) for forest carbon estimation. Six L-band fully polarimetric acquisitions are used in this study. The input parameters to the RF algorithm are the backscatter, decomposition powers and species information. The multi-temporal backscatter (HH1 to HH6, HV1 to HV6, VV1 to VV6) and the temporal average are used. Furthermore, average decomposi-tion parameters from G4U decomposition—Double bounce (Dbl), Odd bounce (Odd), Volume scattering (Vol), and Helix scattering (Hlx) for all six dates. In the first case (1), the model is trained to estimate only the AGB. In the second case (2), the model is trained for forest height estimation. In the third case (3), the model is trained to predict both the AGB and height of the forest. In contrast to the LR method, there is a significant improvement in AGB estimation achieved with the RF algorithms. This study shows the potential of combined retrieval of AGB and forest height using time-series L-band backscatter data.