Quantifying the impact of subpixel reflectance variance on cloud property retrievals from high-resolution ASTER observations

Satellite-observations of cloud optical thickness (τ) and effective droplet radius (reff) are commonly retrieved by the bispectral method from a pair of cloud reflectance samples in a visible to near-infrared (VNIR) band and in a shortwave infrared (SWIR) band, respectively. Due to the assumption of horizontal heterogeneity of cloudy pixels, satellite observations with a lower spatial resolution cannot resolve heterogeneous cloud structures within a pixel. This introduces significant biases in retrieved τ and reff . A recent study by Zhang et al. (2016) introduced a theoretical framework for predicting the impact of subpixel reflectance variance on the retrievals of τ and reff . Here, we present experimental validation of this framework based on high-resolution reflectance observations by the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) aboard NASA's Terra satellite. The cloud property retrieval is provided by a new, research-level retrieval setup, which utilizes the operational MODIS Data Collection 6 (C6) retrieval algorithms. This yields retrievals of τ and reff at a horizontal resolution as high as 15m. A statistical comparison of the ASTER retrievals and the operational MODIS C6 results for 48 marine boundary layer (MBL) cloud scenes show high agreements and validate the feasibility of ASTER cloud property retrievals. For the same 48 MBL cloud scenes the high-resolution ASTER reflectances (γ) are aggregated from 15m to an increasingly coarse spatial resolution between 30-3840m. The difference between the retrievals of τ and reff from aggregated γ and the mean of the high-resolution cloud properties within the aggregated pixels yield the observed retrieval biases Δτ and Δreff . The changes in subpixel cloud cover, scene cloud cover, Δτ and Δreff with an increase in horizontal resolution of the aggregated pixels are analyzed. Finally, the observed Δτ and Δreff are compared to the predicted biases based on the subpixel γ variability following the framework presented in Zhang et al. (2016). It is shown that the high resolution of the native ASTER observations in combination with MODIS-like cloud property retrievals provides the means to not only predict, but also quantify the impacts of subpixel reflectance variability on retrieved cloud variables.