Cloud detection is the first step of complex satellite-based cloud retrievals. No instrument detects all clouds, and analyses using a given satellite climatology can only describe a specific subset of clouds. We attempt to clarify which subsets of clouds are detected in a robust way by passive sensors, and which require active sensors. To do so, we identify where retrievals of Cloud Amounts (CAs), based on numerous sensors and algorithms, differ the most. We investigate large uncertainties, and confront retrievals from the CALIOP lidar, which detects semitransparent clouds and directly measures their vertical distribution, whatever the surface below. We document the cloud vertical distribution, opacity and seasonal variability where CAs from passive sensors disagree most.

CALIOP CAs are larger than the passive average by +0.05 (AM) and +0.07 (PM). Over land, the +0.1 average difference rises to +0.2 over the African desert, Antarctica and Greenland, where large passive disagreements are traced to unfavorable surface conditions. Over oceans, CALIOP retrievals are closer to the average of passive retrievals except over the ITCZ (+0.1). Passive CAs disagree more in tropical areas associated with large-scale subsidence, where CALIOP observes a specific multi-layer cloud population: optically thin, high-level clouds and opaque, shallow boundary layer clouds (altitude 0.5-2km).

We evaluate the CA and cloud vertical distribution from 8 General Circulation Models where passive retrievals disagree and CALIOP provides new information. We find that modeled clouds are not more realistic where cloud detections from passive observations have long been robust, than where active sensors are required.