Spatial Scale Effect And Correction Of Forest Aboveground Biomass Estimation Using Remote Sensing

Forest vegetation is an important part of global terrestrial ecosystem,which not only improves and maintains regional ecological environment,but also plays an important role in surface carbon cycle and energy balance.Forest biomass is closely related to carbon sources and sinks in forest ecosystems and reflects the material cycle of forest ecosystems.Accurate estimation of forest biomass helps to understand the changes of the entire ecosystem and climate change,and provides a basis for forest monitoring and a reference for assessing China’s contribution to global carbon source sinks.At present,the basic methods for estimating above-ground forest biomass include those based on forest inventory,combined with allometric growth equation or biomass expansion factor,physical model and remote sensing estimation.Remote sensing technology has the advantages of macro,comprehensive and efficient acquisition of surface feature information,which can provide spectral characteristics and forest coverage of forest vegetation,and become the main method of large-scale estimation of forest aboveground biomass.With the rapid development of remote sensing technology,remote sensing data products with different spatial and temporal resolutions have been used successively,which greatly enrich remote sensing data sources.However,the surface parameter information,analysis results and revealed rules and principles based on remote sensing data are often affected by different remote sensing sensors,surface spatial heterogeneity and algorithm nonlinear factors,so there are obvious differences in different observation scales,namely the so-called"scale effect".Among them,the most important is the spatial heterogeneity of the earth surface,which affects the variation of surface information in different spatial scales.For remote sensing images,pixels usually contain a mixture of multiple land use types,and pixel parameter extraction is a process of information averaging,which leads to a large error between surface parameters calculated by images with different spatial resolutions.There have been many researches and analyses on this issue.Many scholars have proved that,based on surface parameters,if there is spatial heterogeneity on the surface or nonlinear algorithm on the data of different spatial scales,the parameter results at different scales will be different.Therefore,it is important to find a suitable scale conversion method to solve the scale effect problem of remote sensing.In this study,three kinds of remote sensing data,GF-2,Sentinel-2 and Landsat-8,combined with multiple linear regression and random forest method,were used to establish forest aboveground biomass(AGB)estimation model.Next,the optimal scale estimating result was considered as a referred forest AGB for obtaining relative true forest AGB distribution at different scales based on the Law of Conservation of Mass,and the error of scale effect of forest AGB estimation at various spatial resolution were analyzed.Then,the information entropy of land use type was calculated to identify the heterogeneity of pixel.Finally,the scale conversion method of entropy-weighted index was developed to correct the scale error of forest AGB estimated results from coarse resolution remote sensing images.The results showed that the prediction accuracy of forest AGB was better by using the random forest model combined with GF-2(4 m),Sentinel-2(10 m)and landsat-8(30 m).The coefficients of determination of the predicted and measured values were 0.5711,0.4819 and 0.4321,respectively.After further scaling correction,the R~2 of Sentinel-2 and Landsat-8 increased from 0.6226 to 0.6725 and from0.5910 to 0.6704,respectively,which effectively corrected the scaling error.The scale error is corrected effectively.This study can provide a reference for forest AGB estimation considering the spatial heterogeneity of forest surface and reducing the upscaling error of forest AGB estimation.