Study On The Pixel Scale Reference Truth Value Acquisition And Uncertainty Of High-resolution Surface Reflectance Products

The high-resolution surface reflectance products are the basic elements of quantitative inversion of relevant high-resolution surface parameters.Due to the influence of many factors,the accuracy of high-resolution surface reflectance products is reduced.By obtaining the reference truth value of pixel scale,quantitatively evaluating its accuracy compared with reflectance products,can it be applied to various fields.The reference truth value of pixel scale mainly comes from the ground measurement.Instruments are used to sample and observe the surface reflectance of several point scales on the ground,and obtaining the reference truth value.Due to the widespread heterogeneity of the surface and the difference between the spatial scale of the ground point scale and the pixel scale,there is uncertainty in the reference truth value of pixel scale.How to obtain high-precision reference truth and quantitatively estimate the uncertainty is an urgent problem to be solved.So far,scholars have studied the acquisition and uncertainty of pixel scale reference truth,but there is a lack of research on comparison of different scaling methods,heterogeneity index for selecting more homogeneous sites,and quantitative analysis of uncertainty.This paper aims to determine the appropriate scaling method,the best site heterogeneity evaluation index and reasonable sampling strategy in the acquisition of pixel scale reference truth value of high-resolution reflectance products,so as to reduce the uncertainty of reference truth value,so as to solve the problems of acquisition of high-precision reference truth value and quantitative estimation of uncertainty.This paper simulates and calculates the scaling uncertainty based on airborne image.Taking airborne hyperspectral images as the data source,represented by four scaling methods:arithmetic average,Kriging,trend surface and truth transfer,the uncertainty of different scaling methods is compared to determine the optimal scaling method;The scaling uncertainty of arithmetic average is simulated based on airborne image in a large number of sites,the statistical analysis method is used to screen the key spatial heterogeneity index with the best correlation with the scaling uncertainty among the seven spatial heterogeneity indexes of cv,Moran's I,sill,fractal dimension,CS,local variance and SHI,as the index for evaluation and selection of more homogeneous sites.And the correlation model between it and the scaling uncertainty is established;By comparing the correlation model differences between key spatial heterogeneity index and scaling uncertainty in different sampling methods and sampling points,the impact of sampling methods and the number of sampling points on uncertainty is analyzed,and the best sampling strategy considering low uncertainty and less sampling workload is determined;Taking Youyi farm of Heilongjiang Province as the experimental area,the ground measurement is carried out to obtain the reference truth value of pixel scale and estimate the uncertainty.Taking Sentinel-2A surface reflectance product as the benchmark,the rationality of uncertainty estimation and the accuracy of reference truth value are analyzed.The main conclusions of this paper are as follows:(1)When the site features are single,it is recommended to adopt arithmetic mean scaling,and multiple features adopt true value transmissionIn the acquisition of high-resolution pixel scale reflectance reference truth value,when the site feature type is single and systematic sampling is adopted,the uncertainty of trend surface is relatively unstable,and the uncertainty of arithmetic average,Kriging interpolation and truth transfer method is close.Among them,Kriging requires sufficient number of sampling points and increases the sampling time,which makes the difference between ground measurement time and satellite transit time and increases the uncertainty,the truth transfer method needs reference image,while the arithmetic average method does not need reference image and enough sampling points.When the site covers a variety of ground feature types,the uncertainty of truth transfer method is lower than that of other methods,so it is recommended to use truth transfer method.(2)The cv is the best index for evaluating site heterogeneityWhen the type of site features is single and the arithmetic average scaling is adopted,the cv is the best index for the evaluation of spatial heterogeneity.It is used for the selection of more homogeneous sites to reduce the uncertainty of scale conversion.It has a significant linear correlation with the scaling relative uncertainty in the range of 0.05-0.5,and the correlation coefficient and R~2 are basically above 0.9.The smaller the cv,the lower the scaling uncertainty.The linear model of cv and scaling uncertainty under different sampling strategies is established,which can estimate the scaling uncertainty from the cv of the site.(3)4-point or 5-point sampling system is recommended for sites with single feature typeWhen the site feature type is single and the arithmetic average scaling is adopted,the scaling uncertainty of system sampling is smaller than that of random sampling.The more the number of sampling points,the smaller the scaling uncertainty.However,too many sampling points will not effectively reduce the scaling uncertainty,but will increase the sampling time and the uncertainty caused by the change of solar altitude angle and solar irradiance.It is suggested that the 4-point or 5-point system should be adopted.