Estimation of Forest Above-Ground Biomass in the Study Area of Greater Khingan Ecological Station with Integration of Airborne LiDAR, Landsat 8 OLI, and Hyperspectral Remote Sensing Data

Accurate estimation of forest above-ground biomass (AGB) is significant for understanding changes in global carbon storage and addressing climate change. This study focuses on 53 samples of natural forests at the Greater Khingan Ecological Station, exploring the potential of integrating Canopy Height Model (CHM) with multi-source remote sensing (RS) data—airborne LiDAR, Landsat 8 OLI, and hyperspectral data to estimate forest AGB. Firstly, RS features with strong horizontal and vertical correlation with the forests AGB are optimized by a partial least squares algorithm (PLSR). Then, multivariate linear stepwise regression (MLSR) and K-nearest neighbor with fast iterative features selection (KNN-FIFS) are applied to estimate forest AGB using seven different data combinations. Finally, the leave-one-out cross-validation method is selected for the validation of the estimation results. The results are as follows: (1) When forest AGB is estimated using a single data source, the inversion results of using LiDAR are better, with R2 = 0.76 and RMSE = 21.78 Mg/ha. (2) The estimation accuracy of two models showed obvious improvement after using fused CHM into RS information. The MLSR model showed the best performance, with R2 increased by 0.41 and RMSE decreased to 14.15 Mg/ha. (3) The estimation results based on the KNN-FIFS model using the combined data of LiDAR, CHM + Landsat 8 OLI, and CHM + Hyperspectral imaging were the best in this study, with R2 = 0.85 and RMSE = 18.17 Mg/ha. The results of the study show that fusing CHM into multi-spectral data and hyperspectral data can improve the estimation accuracy a lot; the forest AGB estimation accuracies of the multi-source RS data are better than the single data source. This study provides an effective method for estimating forest AGB using multi-source data integrated with CHM to improve estimation accuracy.