Sampling Bias From Satellite Retrieval Failures of Cloud Properties and Its Implications for Aerosol-Cloud Interactions

Satellite radiometers like MODIS use a bi-spectral retrieval algorithm to simultaneously retrieve cloud optical thickness and cloud effective radius $\left(r_e\right)$. However, retrievals fail for liquid clouds when the $r_e$ observation exceeds the maximum threshold of 30 $\mu$m in MODIS's solution space, leading to a sampling bias. Here, we quantify this bias by reconstructing pixels with failed retrievals using two methods: a conservative approach assigning a fixed minimum $r_e$ threshold to failed pixels, and a representative approach modeling failed $r_e$ using CloudSat radar measurements. We show that MODIS overestimates cloud droplet number concentration by 8%–9% and underestimates liquid water path by 8%–11% globally. We demonstrate that this bias can introduce erroneous correlations between cloud properties that may be misinterpreted as causal processes. Accordingly, we show that accounting for this bias increases the cloud water adjustments by 24%–36%, highlighting the crucial need to expand the solution space in MODIS and similar sensors.