基于光学遥感联合GEDI反演森林蓄积量研究  

Forest Stock Volume Inversion based on Optical Remote Sensing Combined with GEDI

作　　者：冷鸿天 岳彩荣[1] LENG Hongtian;YUE Cairong(Southwest Forestry University,Kunming Yunnan 650224,P.R.China;Yunnan Forestry Survey and Planning Institute,Kunming Yunnan 650051,P.R.China)

机构地区：[1]西南林业大学,云南昆明650224 [2]云南省林业调查规划院,云南昆明650051

出　　处：《西部林业科学》2024年第4期145-153,共9页Journal of West China Forestry Science

基　　金：国家自然基金项目“星载SAR多频段极化干涉数据森林树高反演”(42061072);云南省科技厅重大科技专项“林业资源数字化-森林资源数字化子课题”(202002AA100007-015)。

摘　　要：传统的森林蓄积量调查方法强度大、成本高、耗时长。单纯利用光学遥感影像特征进行森林蓄积量反演时,由于缺乏大量的样地数据支持,加之在森林接近郁闭时光谱反射率易饱和,导致蓄积量反演蓄积精度有限。通过结合光学遥感数据和GEDI足迹点提供的冠层高度信息,可以极低的成本获取大量具有较高蓄积量精度的足迹点,从而通过加密训练样地提升研究区域使用光学遥感反演乔木森林蓄积量的精度。结果显示:(1)当森林高度百分位数(RH)为RH 80时,GEDI提取森林高度特征与机载LiDAR提取树高一致性最高,R^(2)=0.44,RMSE为6.04 m。(2)使用不包含树高特征的光学遥感回归模型预测森林每公顷蓄积量,R^(2)为0.29,RMSE为64.95 m^(3)·hm^(-2);加入树高特征的回归模型预测森林每公顷蓄积的R^(2)为0.59,RMSE为49.53 m^(3)·hm^(-2);样地加密后再使用无树高特征的回归模型预测森林每公顷蓄积的R^(2)为0.46,RMSE为56.55 m^(3)·hm^(-2)。因此,通过结合GEDI足迹点加密样地后可以明显提升研究区域遥感反演乔木森林蓄积量的精度。(3)2020年云南省普洱市思茅区公布的总活立木蓄积量为2629×10^(4)m^(3),使用GEDI足迹点加密训练样地后反演得到的总蓄积量为2289×10^(4)m^(3),整体预测精度为83%。其分布空间格局与森林资源二类调查结果基本相符。Traditional forest inventory methods are intensive,costly and time-consuming.When forest inventory is simply inferred from the characteristics of optical remote sensing images,the lack of large amounts of sample data and the saturation of spectral reflectance in forests with high tree density limit the accuracy of the inventory.By combining optical remote sensing data and canopy height information provided by GEDI footprint points,a large number of footprint points with high accuracy of the volume can be obtained at a very low cost,thereby improving the accuracy of optical remote sensing inversion of tree forest stock volume in the study area through denser training sample plots.The results show that:(1)When the relative height metrics at 80%,the consistency between the forest height characteristics extracted by GEDI and the tree height extracted by airborne LiDAR is the highest,with an R^(2)of 0.44 and an RMSE of 6.04 m.(2)Using an optical remote sensing regression model that does not include tree height characteristics to predict forest stock volume per hectare,the R^(2)was 0.29 and the RMSE was 64.95 m^(3)·hm^(-2);the R^(2)of the regression model that included tree height characteristics to predict forest stock volume per hectare was 0.59 and the RMSE was 49.53 m^(3)·hm^(-2);after the sample plot was densified and then using a regression model without tree height characteristics to predict the forest stock volume per hectare,the R^(2)was 0.46 and the RMSE was 56.55 m^(3)·hm^(-2).Therefore,by combining the GEDI footprint point with the sample plot,the accuracy of remote sensing inversion of the volume of tree forests in the study area can be significantly improved.(3)The total standing timber volume in Simao District,Pu'er City,Yunnan Province,announced in 2020 was 2,629×10^(4)m^(3).After the GEDI footprint point was used to densify training sample,the total volume obtained by remote sensing inversion was 2,289×10^(4)m^(3),and the overall prediction accuracy was 83%.The spatial pattern of its distribution

关 键 词：GEE GEDI 星载激光雷达 反演森林蓄积量 光学遥感 

分 类 号：S771.8[农业科学—森林工程]