基于星载激光雷达数据的森林地上生物量估算方法比较  

Comparison of forest aboveground biomass estimation methods based on spaceborne LiDAR data

作　　者：宋洁[1,2] 刘学录[1,2] SONG Jie;LIU Xuelu(College of Natural Resources and Environment,Gansu Agricultural University,Lanzhou 730070,China;Land Use Research Institute,Gansu Agricultural University,Lanzhou 730070,China)

机构地区：[1]甘肃农业大学资源环境学院,兰州730070 [2]甘肃农业大学土地利用研究所,兰州730070

出　　处：《生态科学》2024年第5期52-62,共11页Ecological Science

基　　金：甘肃农业大学公招博士科研启动基金项目(GAU-KYQD-2021-46);甘肃省自然科学基金项目(23JRRA1413);甘肃省有色地勘局地质项目(GSAU-JSFW-2023-95)。

摘　　要：近年来,星载激光雷达数据已被广泛用于大尺度森林地上生物量估计,但由于其激光光斑采样点不连续,通常使其需要与辅助数据相结合来估算森林地上生物量的连续分布,且估算方法仍存在许多不确定性。研究以祁连山国家公园为样本,结合星载激光雷达ICESat/GLAS数据、Landsat OLI数据和样地调查数据建立了3种基于非参数化算法(普通克里金插值(Ordinary Kriging,OK),支持向量回归(Support Vector regression,SVR)和随机森林(Random forest,RF))的森林地上生物量估算模型,以森林资源清查数据独立验证各模型估计精度。结果发现:3种模型的均方根误差(RMSE)从低到高依次为SVR(19.053 t·hm^(-2))、RF(21.074 t·hm^(-2))和OK(26.362 t·hm^(-2)),平均相对误差(MRE)从低到高依次为SVR(31.890%)、RF(33.314%)和OK(55.398%),且除OK模型外,SVR与RF模型的总体相对误差(TRE)都在可接受的范围内。进一步对SVR与RF模型生成的森林地上生物量空间分布的准确性进行验证,发现相较RF模型,SVR模型生成的森林地上生物量空间分布与森林资源清查数据更为接近。SVR森林地上生物量估计模型在数量精度和分布精度上都表现更优。结果可为今后基于星载激光雷达数据的森林地上生物量估算提供借鉴。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.

关 键 词：森林地上生物量 星载激光雷达 普通克里金插值 支持向量回归 随机森林 

分 类 号：Q948.1[生物学—植物学]