Generation Of Dense Time Series Of Medium-Resolution Shortwave Broadband Albedo Images Based On Kalman Filtering Algorithm

The surface albedo represents the quantized value of the radiation energy absorbed by the underlying surface. It contains all the shortwave energy information of the earth surface. Accurate retrieval of the surface albedo is a prerequisite for precise estimation of other land surface parameters. To meet the demands of remote sensing monitoring of large scale, high precision and fast changing surface conditions on high spatial and temporal resolution remote sensing data, it is necessary to obtain land surface albedo with high spatial and temporal resolution.In this paper, we propose a data assimilation method, which implements a Kalman filter (KF) algorithm that simulates sequence of medium-resolution (30m) synthetic images from existing medium-resolution (30m) imagery and time series of moderate resolution (500m) imagery. Firstly, the Landsat ETM+ reflectance images were transformed from narrow bands to broad bands so that shortwave albedo data at higher spatial resolution (30m) were generated. Meanwhile, the MCD43 A3 shortwave albedo products were interpolated by the weights of the sky light scattering factor, thereafter the shortwave broadband albedo with the spatial resolution of 500m could be acquired. Secondly, a general trajectory of the albedo values was used to characterize the temporal trend in the Kalman filter. The relevant parameters were determined by using the statistical regularity of the data in the study area. Finally, time series of surface shortwave albedo images at 30m and 8-day intervals from January 2010 to October 2010 were generated.The main results are as follows:(1) The average prediction residuals between the synthetic albedo images which predicted by Kalman Filter model and the Landsat ETM+ albedo images were mostly 2-3%. Except for few images in some dates which were strongly affected by the cloud, the R2 were greater than 0.7. It could reached the requirements of the land surface process ecological models for the input parameters.(2)The mean value of albedo products in the study area ranged from 13% to 20%.The albedo predicted by the Kalman Filter method captured the seasonal trend of MCD43A3 data very well.(3) Spatially, the surface albedo prediction image exhibited the similar detail features as those in the Landsat ETM+ images visually. By using forward and backward ETM+ images and time trends of MODIS data, some bad values which influence by the cloud were repaired.(4)Compared to the STARFM model, Kalman Filter method featured a faster prediction speed and can also forecast the data of a whole year at the same time. It considered the change rules of long time series albedo data and input the uncertainty factors. Spatially, Kalman Filter method could avoid the detailed patches that derived from the STARFM model. At the temporal scale, Kalman Filter method was less restricted by the lack of Landsat data.