Study On Surface Reflectance Normalization Based On Multi-source Medium-to-high Resolution Remote Sensing Data

In recent years,the number and types of medium-to-high resolution remote sensing satellites are increasing year by year at home and abroad.It is a general trend to comprehensively use the surface reflectance products of multi-source remote sensing data to develop the high-frequency,high-precision,wide-range and long-sequence monitoring applications,but due to sensor performance differences,data product processing levels,and atmospheric conditions differences during imaging,the radiation inconsistency between multi-source data poses a huge challenge to the integrated applications.The normalization of surface reflectance products can not only maintain the continuity of data sources in time and space,but also make the quantitative research focus on the real change information of the surface and improve the accuracy of quantitative analysis based on multi-source data.In this paper,the necessity of normalization is expressed by comparing the different levels differences of multi-source remote sensing data.Then surface reflectance inversion algorithms based on radiation transfer model are compared,and the inversion algorithm selection strategies for different sensor data is proposed.On the basis of the surface reflectance product,the idea of pixel-by-pixel normalization based on spectral library is proposed.The main research content of this paper is as follows:(1)Performance and reflectance difference comparison of multi-source remote sensing data.By comparing nearly 20 performance parameters of Landsat-8 OLI,Sentinel-2A/B,GF-1,GF-2,HJ-1A/B mainstream medium-to-high resolution sensors at home and abroad,such as band setting,spectral response function,central wavelength,quantization grade and positioning accuracy,we select Landsat-8 OLI&Sentinel-2A,GF-1&HJ-1A/B to compare the spectral of typical objects in two groups of near-time images,and find that even at the same time,the apparent(surface)reflectance of multi-source remote sensing data is still different due to the different spectral response functions,so it needs to be normalized before application.(2)Proposed an algorithms selection strategy of multiple surface inversion for multi-source data.Comparing the characteristics and applicability of several atmospheric correction algorithms including DDV,VNIR,Deep Blue and MODIS-Based,this paper puts a selection strategy of surface reflectance inversion algorithm for multi-source data different bands setting.Taking HJ-1B CCD2 as an example,the corresponding lookup table is established by 6SV2.1,and the inversion accuracy of VNIR infrared iterative algorithms and MODIS-Based algorithms is compared.The results show that the inversion accuracy of the VNIR infrared iterative algorithm is higher,and the algorithm selection strategy provides the algorithm basis for the atmospheric parameters extraction and the surface reflectance precise inversion in flat areas.(3)AOD retrieval of Sentinel-2A/B highlighted surface area based on deep blue algorithm.Aerosol is the most important parameter in the process of surface reflectance inversion.High-precision inversion of AOD in highlight surface areas such as cities and sparse vegetation is always one of the difficulties in in the field of quantitative remote sensing.Based on Sentinel-2 data characteristics,a specific lookup table is established,and a deep blue band reflectance database is constructed using correction coefficients.Then taking Sentinel-2 images of deserts and urban areas as an example,the feasibility of deep blue algorithm in Sentinel-2 data is discussed.Finally,the absolute value accuracy and spatial distribution trend of inversion results are analyzed by using multi-source validation data including AERONET measured values,the results show that the inversion results have high inversion accuracy.(4)The surface reflectance is normalized pixel-by-pixel based on the equivalent spectral library and matching factor.According to the band range of multi-source sensors,the ground object spectral library is screened and integrated.And based on the spectral response function,the equivalent spectral library of different sensors is established.Then,according to the matching model of image and spectral library and the corresponding spectral matching factor,the matching conversion model between the similar bands of different sensors is constructed with the help of spectral matching factor,and then the multivariate linear regression analysis is used to realize the matching and conversion from the image spectral to the equivalent spectral library and to the reference image spectral.Finally,taking multi-source temporal image data as data source,the application of pixel-by-pixel normalization method in time series analysis of vegetation parameters is considered.