Application of ensemble kalman filter to geophysical parameters retrieval in remote sensing: A case study of kernel-driven BRDF model inversion

The use of a priori knowledge in remote sensing inversion has great implications for ensuring the stability of inversion process and reducing uncertainties in retrieved results, especially under the condition of insufficient observations. Common optimization algorithms have difficulties in providing posterior distribution and thus cannot directly acquire uncertainties in inversion results, which is of no benefit to remote sensing application. In this article, ensemble Kalman filter (EnKF) has been introduced to retrieve surface geophysical parameters from remote sensing observations, which has the capability of not merely obtaining inversion results but also giving its posterior distribution. To show the advantage of EnKF, it is compared to standard MODIS AMBRALS algorithm and highly efficient global optimization method SCE-UA. The inversion abilities of kernel-driven BRDF models with different kernel combinations at several main cover types are emphatically discussed when observations are deficient and a priori knowledge is introduced into inversion.