Sampling bias from satellite retrieval failure of cloud properties and its implications for aerosol-cloud interactions

Satellite sensors provide global monitoring of cloud properties, such as cloud effective radius and cloud optical thickness, which have been extensively used to quantify the radiative forcing due to aerosol-cloud interactions. These cloud properties are simultaneously retrieved from a pair of reflectance measurements using a bi-spectral retrieval algorithm. However, the algorithm’s solution space is limited, and retrievals often fail when observations fall outside this space. Upon analyzing five years of quality-constrained liquid-cloud pixels observed by MODIS aboard Aqua, we find that a significant 10% of cloudy pixels experience retrieval failure, primarily because the observations correspond to an effective radius exceeding MODIS’s upper retrieval limit of 30 µm. The omission of these cloudy pixels introduces a sampling bias in aggregated mean gridded cloud properties, affecting, among other things, radiative forcing calculations. To address this, we restore the failed cloud retrievals in MODIS using two reconstruction algorithms: (1) a conservative approach that assigns a fixed minimum effective radius to failed pixels, and (2) a realistic approach that uses extreme effective radius distributions from spaceborne radar measurements. Our findings reveal that MODIS-derived cloud droplet number concentration is positively biased, while liquid water path is negatively biased. Accounting for this bias increases the magnitude of cloud water adjustments, highlighting the crucial need to expand the solution space in MODIS and similar sensors.