Validation of surface radiation fluxes is performed separately for different conditions: clear-sky and cloudy-sky, daytime and nighttime for four seasons. Differences between the CERES shortwave radiation flux and surface measurements have larger standard deviations in cloudy-sky conditions than in clear-sky conditions, indicating that cloud contamination increases uncertainty in the retrieval algorithm. Upward shortwave radiation flux (USW) is overestimated in cloudy conditions suggesting that the cloud parameters and surface scene type in the retrieval process are not optimal for northwestern China. The CERES downward longwave radiation fluxes (DLW) accurately follow the variation of surface measurements during daytime, but are slightly underestimated during nighttime due to the coarse sounding profile and undetected low clouds at nighttime. The CERES upwelling longwave radiation fluxes (ULW) are strongly underestimated during daytime but are slightly underestimated during nighttime regardless of cloud coverage. This large bias could be caused by an underestimate of surface skin temperature and/or surface emissivity, or spatial inhomogeneity around the site.
Meanwhile validation of cloud micro-physical properties is operated in single layer and overcast stratus condition during daytime. Both Terra and Aqua derived cloud optical depth (COD) and liquid water path (LWP) could generally track the variation of the surface counterparts with the modest correlation, while cloud effective radius (re) could not closely follow the surface retrievals' variation and with weak correlation. The mean differences between Terra and surface retrievals are -4.3±9.1, 1.1±3.4μm and 8.7±70.2gm-2 for COD, re and LWP, respectively. The corresponding differences for Aqua are 1.2±8.0, -0.3±3.3μm and 11.6±53.0gm-2, respectively. Possible causes for biases of satellite retrievals, such as cloud inhomogeneity, phase false, and large anglers, are discussed through statistical analysis and case studies.
Generally, except for diurnal ULW, other components of the surface radiative fluxes obtained from SSF datasets are close to meeting the accuracy requirements for climate research in this area. And the accuracy of COD meets the requirement for climate dataset, while re and LWP still need to be improved.