Impact of vector quantization compression on the surface reflectance retrieval: a case study

Abstract

This study is to investigate the effects of the vector quantization (VQ) compression on surface reflectance retrieval based on a CASI data set. The VQ compression with 4 different numbers of code vectors (256, 4096, 8192 and 16384) is evaluated. Based on the best-estimate input model parameters for atmospheric correction, the authors compare the mean and standard deviation of the surface reflectance over 9 specified zones retrieved from the original data cube and those retrieved from the reconstructed data cube. The authors also compare the uncertainty in the retrieved surface reflectance caused by the VQ compression and that caused by the uncertainties in the input model parameters for atmospheric correction. Results show that (1) as the number of code vectors increases, the VQ compression's effect decreases, and (2) the uncertainty caused by the VQ with 16384 code vectors is less than 5 percent and it is smaller than that caused by the uncertainties in the input model parameters under the assumption that the uncertainties in the input model parameters are independent for mote than half of the total CASI spectral bands.