基于序贯高斯条件模拟的GEDI数据联合Landsat8反演森林地上生物量

Combined GEDI Data and Landsat 8 for Inversion of Forest Aboveground Biomass Based on Sequential Gaussian Condition Simulation

[目的]单一遥感技术估测森林生物量存在较大局限性,本研究旨在利用多源遥感协同技术互补激光雷达和光学遥感的优势,提高生物量估测精度。[方法]以星载激光雷达GEDI和光学遥感Landsat8数据为主要信息源,采用序贯高斯条件模拟方法实现GEDI光斑数据由“点”到“面”的空间扩展,结合地面138块生物量调查样地,利用随机森林回归方法估测云南省香格里拉云冷杉林的地上生物量。[结果](1)采用序贯高斯条件模拟方法对GEDI光斑点进行空间插值,模拟的12个生物物理指标在空间分布上呈现出随机性、破碎化的特征,这与森林的空间分布聚集性非常相似,参与建模的9个指标OEC均大于0.90;(2)利用单一Landsat8光学遥感数据和地形因子构建的随机森林模型精度为:R2=0.82,RMSE=35.51 t·hm^(-2),P=0.77;Landsat8数据协同星载激光雷达GEDI数据构建的随机森林模型精度为:R2=0.86,RMSE=32.11 t·hm^(-2),P=0.80,模型精度明显提升;(3)利用多源遥感技术估测的香格里拉2019年云冷杉林地上的生物量总量为37 042 605.68 t,平均生物量为123.28 t·hm^(-2)。[结论]基于地统计学的序贯高斯条件模拟方法考虑到研究对象的空间异质性、能克服一定的平滑效应,用于实现激光点由“点”到“面”的空间扩展是可行的。星载激光雷达GEDI与光学遥感Landsat8协同的多源遥感数据可有效填补单一遥感数据源的缺陷,提高森林生物量的估测精度,能为激光雷达联合光学遥感估测大范围、全覆盖的森林生物量提供参考。

[Objective]There are significant limitations in estimating forest biomass using a single remote sensing technology.The research aims to utilize the advantages of multi-source remote sensing collabora-tion technology to complement LiDAR and optical remote sensing,and to improve the accuracy of bio-mass estimation.[Methods]Based on the two main information sources,including spaceborne LiDAR GEDI and optical remote sensing Landsat8 data,Sequential Gaussian Conditional Simulation(SGCS)method was used to achieve spatial expansion of GEDI data from"points"to"polygons".Combing with aboveground biomass data of 138 plots,the random forest method was used to estimate the aboveground biomass of Picea asperata and Abies fabri forests in Shangri-La,Yunnan Province.[Result](1)The SGCS method was used to perform spatial interpolation on GEDI footprints,and the simulated spatial distribution maps of 12 biophysical indicators showed random and fragmented characteristics,which were very similar to the spatial distribution and clustering of forests,and the OEC of 9 indexes involved in modeling were greater than 0.90.(2)The accuracy of the random forest model constructed using a single optical remote sensing Landsat8 data was:R2=0.82,RMSE=35.51 t·hm^(-2),P=0.77;The accuracy of the random forest model constructed by combining two types of remote sensing data was:R^(2)=0.86,RMSE=32.11 t·hm^(-2),P=0.80.It could be obviously found that the accuracy of the model was improved.(3)The total above-ground biomass of Picea asperata and Abies fabri forests in Shangri-La in 2019 estimated by multi-source remote sensing technology was 37042605.68 t,and the average biomass was 123.28 t·hm-2.[Conclu-sion]The SGCS method based on geostatistics has some advantages,including considering the spatial heterogeneity of the research object and being able to overcome smoothingeffects I is feasible to achieve spatial expansion of GEDI footprints from"point"to"polygon".The multi-source remote sensing data based on the combination of spaceborne LiDAR GEDI and optical remote sensing Landsat8 can effect-ively fill the defects of a single remote sensing data source,improve the estimation accuracy of forest bio-mass,and provide a reference for the combination of LiDAR and optical remote sensing to estimate large-scale and fully covered forest biomass.