Global cloud distribution: Reconciling the active and passive points of view



bla, bla, bla...


Passive remote sensors have been monitoring the state of the atmosphere from space for several decades. Their measurements have led to reliable climatologies of clouds distribution on a global scale. Unfortunately, these retrievals were penalized by significant uncertainties above certain surface types (ice, snow, deserts, and more generally everything that is not ocean). The ability to detect clouds was moreover strongly dependent on their optical depth, with thinnest clouds often escaping detection. Along the vertical dimension, retrievals were often limited to the cloud top temperature, which required strong assumptions on cloud density that were rarely realistic and led to underestimations by several kilometers (Sherwood 2004). Considerable effort and ingenuity went into solving these problems and reducing uncertainties as much as possible, but the error bars remain significant. They mean that passive observations cannot answer specific questions about how the cloud cover reacts to changes in its atmospheric environment [Chepfer et al., in review].

bla bli bla bla

In the past few years, several active instruments put in orbit have led to significant insights regarding the spatial distribution of clouds. Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) (Winker 2010), bla bla. CALIOP cannot penetrate optically thick clouds, and thus cannot always document the entire vertical profile of cloud fraction. However, this does not impair its ability to retrieve whether a given column of atmosphere contain clouds or not, with a high horizontal resolution and a sensitivity to optically thin clouds unmatched by most passive sensors. CALIOP is therefore a reliable source for documenting the overall cloud fraction and it has been often used so in recent years [bla bla et al. 2012, 2013 etc].

Here we attempt to contrast how active and passive remote sensing observations see cloud amounts on a global scale. We present how passive sensors disagree on the retrieved cloud amounts in Sect. 2, and compare their averaged cloud amounts with those retrieved from three datasets derived from CALIOP data in Sect. 3. In Sect. 4, we focus on geographic areas where passive sensors disagree most among themselves or with active datasets.

Cloud cover from passive datasets

As our baseline, we used the most up to date and comprehensive cloud retrievals based on spaceborne passive remote sensing measurements. These were compiled, reconciled and made available in the context of the Global Energy and Water cycle Experiment (GEWEX) cloud assessment project (Stubenrauch et al. 2013) of the World Climate Research Program. In addition to retrievals from twelve passive sensors, this project included retrievals from CALIOP made compatible with passive ones between 2006 and 2009. 2007 is the first full year covered by CALIOP, however, as it began operation in June 2006. Apart from the TOVS IR sounder, stopped in 1994, passive retrievals span the 2006-2009 period. Three do not cover the entire period: HIRS (2006, 2008), ISCCP (2006-2007) and POLDER (2006-2008). In order to maximise the number of comparable datasets while having them measure the same atmospheric composition, the present study includes retrievals for 2007 only, when retrievals from ten passive sensors are available. TOVS and HIRS are excluded from the comparison. Moreover, some sensors only allow retrievals in absence of sunlight (ATSR, MISR), while others require it (POLDER). We considered separately daytime and nighttime retrievals so they document the same actual cloud cover. Table 1 describe how sensors contribute to the retrieval comparisons, along with their supporting satellite equator overpass local time. All data was obtained through

Fig. \ref{fig:fig1} shows the standard deviation of cloud amount (CA) across all retrievals based on nighttime (top) and daytime (bottom) measurements, considering one map of average 2007 CA per sensor. This standard deviation, referred to as inter-sensor standard deviation (ISSD) from now on, is symptomatic of the differences in cloud detection across all considered sensors.

Nighttime passive CA retrievals strongly disagree (ISSD > 0.2, reds in Fig. \ref{fig:fig1}, top) over iced and snowed surfaces (e.g. Greenland, Antarctica and regions partially covered by ice shelves in winter), high elevation (Himalayas, Andes and West U.S.), the mid-section of Africa directly south of the Sahara. Over the ocean passive sensors agree quite well (ISSD < 0.07), especially at mid-latitudes. Exceptions are tropical subsidence areas and directly west of Africa, South America and Australia (ISSD up to ~0.12). Disagreement is generally higher over continents. The same patterns appear in daytime passive CA retrievals (Fig. \ref{fig:fig1}, bottom), although sensors seem to agree more overall, especially over continents, high elevation, and west of continents.