Estimating Aboveground Biomass of Boreal Forests in Northern China Using Multiple Data sets

Accurate estimates of aboveground biomass (AGB) are valuable for monitoring forest degradation and carbon stocks on Earth. However, the validity of multiple data types and diverse data combinations for AGB estimation is unclear. In this study, recursive feature elimination (RFE) combined with machine-learning regression models for AGB were developed using field data and multi-source remote sensing data, which included Sentinel-1, Sentinel-2, PALSAR, and DEM. The spatial distribution of AGB was mapped for the Daxing'anling region in the northernmost part of China at 30m resolution. We compared the ability of multiple data combinations to perform AGB estimation and found that using all four types of data combinations resulted in the highest estimation accuracy with fewer predictors. The combination of diverse data sources substantiates enhancements in the precision of AGB estimation, surpassing the utilization of singular or dual sensor modalities. In addition to the optical remote sensing data sentinel-2, topographic data has a non-negligible role in the AGB estimation in this study, even more than microwave remote sensing data. Finally, the extreme gradient boosting model (R ² =0.67, RMSE=22.57 Mg/ha) based on the combination of all four data types had the highest accuracy and mapped the AGB of the study area. The results indicate that the AGB can be estimated with reasonable accuracy for the boreal forest region based on publicly available multi-source remote sensing data. This study proposes diverse data combinations as well as derived variables for AGB estimation, aiming to explore the possibilities of more remote sensing data in AGB studies.