Research On Algorithm For Estimating Surface Albedo Of Remote Sensing Data Based On Prior Knowledge

Surface albedo reflects the ability of the surface to reflect solar radiation and has been widely used in the energy balance of surface,the prediction of medium and long-term weather,and the study of global change.At present,the albedo inversion of high-resolution remote sensing data is affected by factors such as cloud cover and operation period,which results in insufficient multi-angle observation data and inaccurate inversion.In order to solve this problem,we can take full advantage of the abundant historical surface BRDF product data to obtain prior knowledge,understand the anisotropic reflection characteristics of the surface deeply,and apply it to the inversion of high-resolution remote sensing data to make up for the lack of high-resolution remote sensing data.It simplifies the process of surface albedo inversion and can improve the inversion accuracy of surface albedo.In this study,a method of extracting prior knowledge from MODIS BRDF product(MCD43A1 data)to retrieve albedo was proposed,and the influence of the albedo obtained from the inversion in the zenith direction and any direction is evaluated based on the time series prior knowledge.Finally,based on the time series prior knowledge and LANDSAT surface reflectance data were inverted to obtain 30 m LANDASAT surface albedo data,and compared with the albedo is obtained by replacing the surface albedo algorithm with zenith reflectivity under the assumption of Lambert surface and the measured albedo data from the land surface station of the Surface Radiation Energy Budget Observation Network(SURFRAD)to verify the accuracy of the inversion results.The results show:(1)The anisotropic distribution of surface reflections at different times,land cover structures,and bands show a clumped distribution.The anisotropy of surface reflection has good consistency in time series.The prior knowledge extracted in this study can represent the reflectance anisotropy of most pixels.(2)The probability-weighted priori knowledge is located in the high-density region of reflectance anisotropy.The volumetric-dominated BRDF based on the random distribution of model parameters underestimates the surface albedo.The geometric-dominated BRDF based on the random distribution of model parameters overestimates the surface albedo.The probability-weighted priori knowledge is more suitable for BRDF to invert albedo.(3)When the high-quality pixels of MODIS BRDF are sufficient,more than 90% of pixels have an absolute difference between the simulated albedo of the zenith direction based on the time series prior knowledge and the actual albedo that are within-0.02 and 0.02.The simulated albedo of the zenith direction based on the time series prior knowledge is basically consistent with the actual albedo,and more than 90% of pixels have an absolute difference that are within-0.02 and 0.02.However,the absolute difference of the simulated albedo around the hotspot and large viewing zenith angle area is large.(4)During 2008 to 2010,comparing the LANDSAT shortwave surface albedo obtained from prior knowledge with the measured albedo of the ground station,it showed a bias of-0.019 and a root mean square error(RMSE)of 0.027.And comparing the LANDSAT shortwave surface reflectance with the measured albedo of the ground station,it showed a bias of-0.027 and a root mean square error(RMSE)of 0.037.It can be showed that the results of LANDSAT shortwave surface albedo obtained from inversion based on prior knowledge are close to the true albedo.The algorithm based on prior knowledge to retrieve albedo is more accurate than the algorithm that uses the zenith direction reflectance instead of the surface albedo to obtain the albedo under the assumption of the Lambert surface.