基于星载激光雷达数据的森林地上生物量估算方法比较.

Recently, spaceborne LiDAR data have been increasingly utilized for large-scale forest aboveground biomass (AGB) estimation. However, due to the discontinuity of laser spot sampling points, it is often necessary to integrate auxiliary data to estimate the continuous distribution of forest AGB, resulting in uncertainties in the estimation method. In this study, we focus on the Qilian Mountains National Park in northwestern China as a sample area. Three forest AGB estimation models (Ordinary Kriging (OK), Support Vector Regression (SVR), and Random Forest (RF)) were developed based on nonparametric algorithms by integrating ICESat/GLAS data, Landsat OLI imagery, and field inventory data. The accuracy of the estimation results generated by these three models was independently verified using forest inventory data. The results indicated that the root mean square error (RMSE) of the three models, from low to high, were SVR(19.053 t·hm-2), RF(21.074 t·hm-2), and OK(26.362 t·hm-2). The mean relative error (MRE), from low to high, were SVR (31.890%), RF (33.314%) and OK (55.398%). Moreover, except for the OK model, the total relative error (TRE) of the SVR and RF models falls within an acceptable range. Further verification of the accuracy of the spatial distribution of forest AGB was conducted for the SVR and RF model. It was found that the forest AGB spatial distribution generated by the SVR model was closer to the forest resource inventory data than that generated by the RF model. Therefore, we conclude that the SVR forest AGB estimation model demonstrates better quantitative accuracy and distribution accuracy. We also anticipate that these results can serve as a reference for forest AGB estimation based on spaceborne LiDAR data in future studies. [ABSTRACT FROM AUTHOR]