Top-of-Atmosphere Albedo Estimation from Angular Distribution Models Using Scene Identification from Satellite Cloud Property Retrievals

The next generation of earth radiation budget satellite instruments will routinely merge estimates of global top-of-atmosphere radiative fluxes with cloud properties. This information will offer many new opportunities for validating radiative transfer models and cloud parameterizations in climate models. In this study, five months of Polarization and Directionality of the Earth’s Reflectances 670-nm radiance measurements are considered in order to examine how satellite cloud property retrievals can be used to define empirical angular distribution models (ADMs) for estimating top-of-atmosphere albedo. ADMs are defined for 19 scene types defined by satellite retrievals of cloud fraction and cloud optical depth. Two approaches are used to define the ADM scene types. The first assumes there are no biases in the retrieved cloud properties and defines ADMs for fixed discrete intervals of cloud fraction and cloud optical depth (fixed-t approach). The second approach involves the same cloud fraction intervals, but uses percentile intervals of cloud optical depth instead (percentile-t approach). Albedos generated using these methods are compared with albedos inferred directly from the mean observed reflectance field. Albedos based on ADMs that assume cloud properties are unbiased (fixed-t approach) show a strong systematic dependence on viewing geometry. This dependence becomes more pronounced with increasing solar zenith angle, reaching ø12% (relative) between near-nadir and oblique viewing zenith angles for solar zenith angles between 608 and 708. The cause for this bias is shown to be due to biases in the cloud optical depth retrievals. In contrast, albedos based on ADMs built using percentile intervals of cloud optical depth (percentile-t approach) show very little viewing zenith angle dependence and are in good agreement with albedos obtained by direct integration of the mean observed reflectance field (,1% relative error). When the ADMs are applied separately to populations consisting of only liquid water and ice clouds, significant biases in albedo with viewing geometry are observed (particularly at low sun elevations), highlighting the need to account for cloud phase both in cloud optical depth retrievals and in defining ADM scene types. ADM-derived monthly mean albedos determined for all 58 3 58 lat–long regions over ocean are in good agreement (regional rms relative errors ,2%) with those obtained by direct integration when ADM albedos inferred from specific angular bins are averaged together. Albedos inferred from near-nadir and oblique viewing zenith angles are the least accurate, with regional rms errors reaching ;5%–10% (relative). Compared to an earlier study involving Earth Radiation Budget Experiment ADMs, regional mean albedos based on the 19 scene types considered here show a factor-of-4 reduction in bias error and a factor-of-3 reduction in rms error.