Improving aboveground biomass estimation by integrating ICESat-2 LiDAR sampling and tree species stratification strategies using multi-source datasets

Accurately estimating forest aboveground biomass (AGB) is essential for achieving suitable management of forests and reducing uncertainties in carbon dynamics assessment. Utilization of multi-source datasets provides extensive opportunities to estimate AGB. However, developing precise AGB models presents challenges under the limited field observations and environmental heterogeneity caused by various tree species and site conditions. We proposed an Upscaling and Tree Species Stratification framework (UTSS) by utilizing ICESat-2 LiDAR, Sentinel imagery, ALOS-2 PALSAR, and environmental data for improving AGB estimation. In UTSS, a tree species product was first generated by utilizing Sentinel-2 and topography data, subsequently, the AGB values from ICESat-2 LiDAR segments (ICESat-2-based AGB) were estimated, and ultimately ICESat-2-based AGB was used to establish AGB models based on different tree species. The performance of UTSS was compared with the upscaling model (UP), tree species stratification model (TSS), and conventional model (CT), which separately considered the LiDAR sampling strategy, tree species stratification strategy, and none of the strategies. The UTSS was examined in a boreal forest in Northeast China. Results showed that the UTSS (R2: 0.82, RMSE: 13.72 t/ha, rRMSE = 24.8%) outperformed the other three models (R2: 0.65–0.75, RMSE: 17.73–14.93 t/ha, rRMSE: 26.98–32.03%). The tree species map with an overall accuracy of 0.96 served as the base of stratification. Fire disturbance, geographical location, elevation, modified chlorophyll absorption ratio index, and sum average texture collected from HV band were important for AGB estimation. The AGB estimates exhibited a south-north gradient in all models, while UTSS performed best in both the low and high values. This study demonstrates that the UTSS framework incorporating ICESat-2 LiDAR sampling and tree species stratification strategies provides a promising approach to estimating AGB by integrating freely accessible datasets, algorithms, and platforms.