The Cumulus And Stratocumulus CloudSat-CALIPSO Dataset (CASCCAD)

Abstract. Low clouds continue to contribute greatly to the uncertainty in cloud
feedback estimates. Depending on whether a region is dominated by cumulus
(Cu) or stratocumulus (Sc) clouds, the interannual low-cloud feedback is
somewhat different in both spaceborne and large-eddy simulation studies.
Therefore, simulating the correct amount and variation of the Cu and Sc
cloud distributions could be crucial to predict future cloud feedbacks. Here
we document spatial distributions and profiles of Sc and Cu clouds derived
from Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations
(CALIPSO) and CloudSat measurements. For this purpose, we create a new
dataset called the Cumulus And Stratocumulus CloudSat-CALIPSO Dataset
(CASCCAD), which identifies Sc, broken Sc, Cu under Sc, Cu with stratiform
outflow and Cu. To separate the Cu from Sc, we design an original method
based on the cloud height, horizontal extent, vertical variability and
horizontal continuity, which is separately applied to both CALIPSO and
combined CloudSat–CALIPSO observations. First, the choice of parameters used
in the discrimination algorithm is investigated and validated in selected
Cu, Sc and Sc–Cu transition case studies. Then, the global statistics are
compared against those from existing passive- and active-sensor satellite
observations. Our results indicate that the cloud optical thickness – as
used in passive-sensor observations – is not a sufficient parameter to
discriminate Cu from Sc clouds, in agreement with previous literature. Using
clustering-derived datasets shows better results although one cannot
completely separate cloud types with such an approach. On the contrary,
classifying Cu and Sc clouds and the transition between them based on their
geometrical shape and spatial heterogeneity leads to spatial distributions
consistent with prior knowledge of these clouds, from ground-based,
ship-based and field campaigns. Furthermore, we show that our method
improves existing Sc–Cu classifications by using additional information on
cloud height and vertical cloud fraction variation. Finally, the CASCCAD
datasets provide a basis to evaluate shallow convection and stratocumulus
clouds on a global scale in climate models and potentially improve our
understanding of low-level cloud feedbacks. The CASCCAD dataset (Cesana,
2019, https://doi.org/10.5281/zenodo.2667637) is available
on the Goddard Institute for Space Studies (GISS) website at https://data.giss.nasa.gov/clouds/casccad/ (last access: 5 November 2019) and on the zenodo website at
https://zenodo.org/record/2667637 (last access: 5 November 2019).


compute the lidar cloud fraction (CFlidar, from 0 to 1) based on the CALIPSO Vertical Feature Mask (VFM) of the level2 5km CALIPSO files (Vaughan et al., 2009), at the CPR resolution in the 2B-GEOPROF-LIDAR files. While the CFlidar can sometimes be lower than 0.5, we kept the 0.5 threshold to diagnose the presence of a cloud as in Mace and Zhang (2014) and Cesana et al. (2019b). Diagnosing a pixel as cloudy from values below 0.5 may result in an overestimate of the averaged cloud fraction when compared to ground-based measurements ( Fig. S1 and Marchand et al., 2010). 5

2B-CLDCLASS-LIDAR
The 2B-CLDCLASS-LIDAR product (Sassen and Wang, 2008; referred to as 2BCCL in the remainder of the manuscript) merges collocated observations from the CloudSat CPR, CALIPSO lidar and MODIS spectrometer to classify clouds into eight types based on several criteria: their vertical and horizontal extent, their precipitating state, their temperature, and their 10 radiance. Although eight cloud types are available in this dataset (Deep convective, Cirrus, Nimbostratus, Altostratus, Altocumulus, Cumulus including fair-weather and congestus, Stratus and Stratocumulus), we only focus on the Cu, stratus (St) and Sc cloud-types. The St and Sc cloud-types are combined into a single category referred to as Sc for consistency with the Sc-Cu discrimination algorithm, which does not differentiate these two categories. In addition, the Sc and St clouds are particularly difficult to distinguish in the 2BCCL product because of the ground clutter contamination in the radar signal 15 (Sassen and Wang, 2008) as shown by Huang et al. (2015). In 2BCCL, the cloud base and top are given for up to ten cloudy layers, which is why we re-project these cloudy layers onto a 480 m vertical grid from 0 to 19.2 km to be consistent with the GOCCP and RL-GeoProf datasets described in sections 2.1 and 2.2. Those are then accumulated into a 2.5˚x2.5˚ grid as for the two other products.
The main goal of this study is to document spatial distributions and profiles of Sc and Cu clouds on a global scale, with the desire to further analyze long-term relationships between Sc-Cu clouds and environmental parameters in future studies. For this purpose, we need to i) distinguish the two cloud types based on observable cloud-properties and ii) use datasets that are available for a time-period sufficiently long (~ 10 years) to compute statistically significant relationships (e.g., using 4 years 25 of GOCCP rather than 10 may decrease the amplitude of the relationship between low clouds and SST anomalies by more than 15 %, Cesana et al., 2019a). Although both the Sc and Cu clouds form within the planetary boundary layer (PBL), they have relatively different shapes as they are controlled by different physical mechanisms. The Cu (Fig. 1, second to last column) can stretch up past the PBL into the lower free troposphere (z ~ 3 km) while they have typically a small horizontal extent (no more than a few km, e.g, Lamer et al., 2015;Nuijens et al., 2015b). On the contrary, the Sc (Fig. 1, first column) have a relatively 30 small vertical extent (no more than a few hundred meters) and a cloud-top height (CTH) controlled by the PBL depth but spread out over tens to hundreds of kilometers (Wood, 2012)  between, these two distinct regimes, various transitioning clouds may form (Albrecht et al., 2019;Teixeira et al., 2011) and the most frequent are: broken Sc and transitioning Sc-Cu, which is composed of Cu under Sc (Albrecht et al., 2019, Rauber et al., 2007 and Cu with stratiform outflow (Lamer et al., 2015, Nuijens et al., 2015b (Fig. 1, second to fourth column, respectively). Clouds with a cloud base and a cloud top within and outside the lower free troposphere, respectively, are classified as deep Cu (Fig. 1, last column). Bearing the above facts in mind, we design the CASCCAD DA based on cloud 5 height and vertical and horizontal cloud fraction, which can be applied to both GOCCP instantaneous profiles and the CloudSat-CALIPSO level 2 geometrical profile product (referred to as RL-GeoProf).
GOCCP instantaneous profiles satisfy the criteria i) and ii) mentioned above. They use all 70-m-large lidar shots every 333 m along-track without horizontal averaging, which allows the detection of the geometrically sparsest shallow Cu, besides the 10 more horizontally extended Sc, over a relatively extended time-period (from June 2006 to 2017, while CALIPSO is still operating as of April 2019). This decadal dataset makes it possible to analyze climatological values of Cu and Sc cloud fraction and their relationships to environmental parameters. However, as the lidar penetrates within cloudy layers, the signal eventually attenuates completely for optical thickness greater than 3 to 5. Therefore, it is not always possible to observe the full troposphere with a space-borne lidar, which may cause differences in satellite-based cloud climatologies obtained from 15 different instruments (Kikuchi et al., 2017;Thorsen et al., 2013). In these instances -i.e., in deep convective clouds or in the storm tracks-, the CPR capability complements cloud profiles beneath the height at which the lidar attenuates, although the CPR clutter prevents using CloudSat data below ~ 1000 m. Unfortunately, the RL-GeoProf product is only available for a short period of time (~ 4.5 years) due to the severe anomaly of April 2011, which is why CloudSat-CALIPSO observations satisfy i) but only partially ii). 20 3.2 First criterion of the discrimination algorithm: the cloud-top height As mentioned above, three main criteria -represented by different colors in Fig. 2-are used in the DA to separate Sc from Cu clouds and to characterize the various Sc-Cu transitioning clouds presented in Fig.2: the CTH, the Horizontal Cloud Fraction (HCF) and the Vertical Cloud Fraction (VCF). Note that the sensitivity to these criteria is later tested in section 4.1. The first 25 step of the DA depends on the height of the cloud top (Fig. 2, 1 st column, in grey). Because trade Cu and Sc are low clouds, their CTH must be within the lower free troposphere, which is defined as 3.36 km in GOCCP (approximately equivalent to the 680-hPa definition of Rossow and Schiffer, 1999). Furthermore, since Sc clouds cap the PBL, their CTH is typically lower than the PBL height over the main Sc deck areas, i.e., ~ 2 km (Albrecht et al., 1995;Bretherton et al., 2010;Garay et al., 2008;Wood, 2012;Zhou et al., 2015;Zuidema et al., 2009). Therefore, all cloud layers -i.e., a vertically-contiguous group of cloudy 30 480-m-pixels-with a cloud top higher than 1.92 km, which is the closest 480-m GOCCP level to 2 km, are diagnosed as Cu type. We remind the reader that the sensitivity of the algorithm to this criterion, as well as the other criteria, is tested in section 4.1.

Second criterion: the horizontal cloud fraction
The remaining clouds are passed to the second step of the DA (Fig. 2, 2 nd column, in orange), which computes the HCF either centered around the lidar profiles (CHCF) or using forward (FHCF) or backward profiles (BHCF) as shown in Fig. 3. These three HCFs ensure capturing the full horizontal extent of the cloud layer regardless of whether the lidar probes toward the edge or the center of the layer while remaining more computationally efficient than treating clouds by contiguous horizontal-5 clusters. For example, the CHCF is 100 % in the specific case of Fig. 3a whereas FHCF and BHCF are about 50 %. Should the first lidar profile be at the beginning of the cloud, the centered, forward and backward HCF would be about 53, 100 and 7 % (Fig. 3b). Additionally, the HCFs are computed over four different length scales, 10, 20, 40 and 80 km, to characterize various cloud scenarios: open-cell and closed-cell Sc (Wood, 2012), Sc-Cu transitioning clouds (Albrecht et al., 2019;Teixeira et al., 2011) and different Cu organizations (Lamer et al., 2015;Rauber et al., 2007). The larger 40-80-km scales permit a clear 10 distinction between the two type of clouds since Sc clouds typically cover vast areas compared to more fractionated trade-Cu clouds. Figures 4c and 4d show the probability density function (PDF) of the 40-km and 80-km CHCFs, respectively, for typical Cu (light blue bars) and Sc (light red bars) cases extracted from day and night CALIPSO orbits over the tropics (35˚S/N, eight orbit segments in total). These results confirm a rather clear separation, marked by purple lines, between the two populations for CHCFs. Although some slight overlap is visible, it disappears when the 40 and 80-km CHCFs are run together 15 (not shown). However, in regions of open-cell Sc and Sc-Cu transition, the overlap may be larger (Fig. S2) and additional tests are needed to determine the type of clouds, i.e., Sc, broken Sc, transitioning Sc-Cu or Cu. In those instances, the 10 and 20km CHCFs help further distinguish Sc from the other clouds ( Fig. 4a and 4b). Finally, note that when all 480-m-pixels below 1.92 km are fully attenuated, the profile is excluded from the HCF computation. Sc clouds are relatively shallow -no more than a few hundred meters (Wood, 2012) -any cloud layer with a substantial VCF over 3 levels or more (vertical extension greater than 1.44 km) is diagnosed as transitioning Sc-Cu while the rest are diagnosed 25 as Sc. The VCF threshold (= 0.12) is defined as approximately two times the standard deviation of the PDF of the 40-km CHCF computed using each of the seven first levels (0 to 3.36 km) in typical Cu regions. Finally, this VCF test is also applied to any cloud layer that passes the 40 and 80-km HCFs thresholds, regardless of their 10 and 20-km HCFs. In these cases, the cloud type is diagnosed as transitioning Sc-Cu if the VCF threshold is met, otherwise the cloud layer is diagnosed as broken Sc. horizontally-contiguous clusters of clouds (without clear sky profiles). As a result, in Cu and transitioning Sc-Cu cases, the same horizontally-contiguous cloud layer may be diagnosed as both Sc and Cu (Fig. 5b), which is more likely a Cu with a stratiform outflow (Lamer et al., 2015). To avoid this "slicing" issue, we apply a horizontal continuity test, which first detects a horizontally-contiguous cluster of clouds and then turns it into a homogeneous Cu if one third of the cluster is diagnosed as Cu type. We chose this arbitrary threshold because, on average, the fraction of the Cu that expands further aloft (geometrically 5 thicker) is typically smaller than that near the lifting condensation level (Nuijens et al., 2015a) or that detrained near the tradewind inversion (Nuijens et al., 2015a, Lamer et al., 2015.

Case studies
To assess our CASCCAD DA, we analyze a series of three typical case studies: trade cumulus, stratocumulus and 10 stratocumulus-cumulus transitioning clouds. First, we investigate the sensitivity of the DA to some of the criteria presented in Section 3 using GOCCP observations: the HCF (more or less conservative), CTH (one level higher) and VCF (smaller threshold, divided by 2) thresholds and the continuity test (turned off). We then compare the results of the standard DA applied to GOCCP and RL-GeoProf against the 2BCCL cloud types -for the same case studies-and utilize the collocated MODIS reflectance to provide a broader context of the cloud scene. Except for the lower VCF parameters, which reduce the along-track Sc CF (HCFSc) by 6.4 % (absolute value, Fig. 6b), the HCFSc is quite insensitive to the DA parameters (HCFSc = 67.4 +2.2/-1.4 %). Reducing the VCF (Fig. 6b) turns the edges of the Sc decks into transitioning Sc -Cu clouds (around 16˚N and 20˚N). However, changing the HCF thresholds have limited 25 effect on the HCFSc in this particular case ( Fig. 6f and 6g).
GOCCP and RL-GeoProf have a similar HCFSc while that of the 2BCCL product is larger along with its total HCF (Fig. 7).
The substantial difference between RL-GeoProf and 2BCCL total HCFs is mostly due to differences in lidar cloud fraction treatment. The lidar cloud fraction from RL-GeoProf comes from the CALIPSO 5km VFM mask, whereas that of 2BCCL comes from the Lidar-AUX product (Wang et al., 2013). As for the Sc case, the along-track Cu CF (HCFCu) is weakly sensitive to variations of the different DA parameters (Fig. 8, HCFCu = 23.6 +1.1/-1.5 %). The most sensitive parameter is the horizontal continuity (Fig. 8e). When activated, it captures most of the large Cu between 20˚S and 18˚S although its southernmost edge remains likely incorrectly diagnosed as Sc. Most of the other cloudy features are diagnosed as Cu except for the cloudy layer located between 31˚ and 29˚S, which seems to be stratiform judging from its geometrical thickness (Fig. 8a) and reflectance (Fig. 9d). 5 Here again, both GOCCP and RL-GeoProf diagnose similar Sc and Cu HCFs although the DA fully captures the aforementioned large Cu only when it is used with RL-GeoProf observations (Fig. 9b). Unlike the Sc case, GOCCP and RL-GeoProf disagree substantially with 2BCCL. For example, several cloud clusters are diagnosed as Sc by the 2BCCL algorithm although their geometrical thickness is larger than 1.5 km (Fig. 9c, around 34˚S, 19˚S and 12˚S), making it very unlikely that these clusters are actual Sc. As a result, the HCFSc (27.9 %) is larger than the HCFCu (22.6 %) and approximately five times 10 larger than that of GOCCP and RL-GeoProf. In addition, it is important to note that the 2BCCL total HCF is ~35 % larger than that of RL-GeoProf for two reasons. Firstly, the 2BCCL algorithm is derived from a different version of the collocated lidar product (Wang et al., 2013). Secondly, no CFlidar threshold is used, which may cause a slight overestimate of the cloud fraction, particularly in the low levels albeit to a smaller extent than the differences between the two lidar products. As a result, the lowlevel cloud fraction of the 2BCCL product may be largely overestimated over the tropical and subtropical oceans. This 15 overestimation is supported by a comparison with ground-based data and other satellite products over the Barbados (more than 2 times larger, Fig. S1), a region dominated by trade cumulus clouds (Nuijens et al., 2015b). An additional Cu case, in the trade-dominated NW Atlantic, confirms the ability of the DA to correctly diagnose a field of purely Cu clouds with no Sc ( Fig.   10a and 10b). As in Fig. 9c, the 2BCCL product classifies a non-negligible amount of the clouds as Sc (Fig. 10c, HCFSc = 6.8 %) and overestimates the HCFtot compared to GOCCP and RL-GeoProf. 20

Cumulus and open stratocumulus case
The last case study extends from the subtropics to the extra-tropics. Such location allows us to characterize transitioning Sc-Cu cases, which includes Sc, open-Sc and Cu clouds (Fig. 11). A visual inspection of the CTH variation from the CALIPSO VFM ( Fig. 11a) suggests that this orbit segment contains three distinct clusters of clouds: Sc from 55˚ to 43˚S and from 25˚ to 25 20˚S and Cu in between. These three distinct layers are quite well captured by the DA although the DA is more sensitive to changes in the parameters than in the other cases. The most sensitive parameters are the VCF and HCF thresholds. Reducing the VCF threshold (Fig. 11b) turns 7.9 % of the Sc into Cu (absolute value) mostly poleward of 43˚S because the BL height decreases, causing multiple levels to be cloudy and subsequently diagnosed as Cu. Choosing smaller HCF thresholds (Fig.   11g) increases the Sc amount by 5.3 % (absolute value). This converts the few Cu poleward 43˚S into Sc as well as some Cu 30 around 24˚S, which could very well be "true" Sc.
The three clusters of clouds are also well captured in the RL-GeoProf dataset, which detects somewhat more Cu than GOCCP making the total HCF larger as well (Fig. 12). On the contrary, the 2BCCL product diagnoses nearly two times more Sc than the CASCCAD datasets and two to three times less Cu. In this section, we analyze climatological geographical distributions of Sc and Cu clouds for the three products presented before as well as for a subset of passive-sensor observations. These include the International Satellite Cloud Climatology 5 Project (ISCCP, Rossow and Schiffer, 1999), the Moderate Resolution Imaging Spectroradiometer (MODIS, King et al., 2013) and the Multi-Angle Imaging Spectroradiometer (MISR, Marchand et al., 2010) observations. Sc and Cu are separated using a cloud top pressure (CTP) -cloud optical thickness (COT) diagram introduced by Rossow and Schiffer (1999): CTP must be larger than 680hPa for each type and COT smaller or larger than 3.6 for Cu and Sc clouds, respectively. Note that this COTbased method has been shown to mis-classify Cu and Sc that have moderate optical thickness (e.g., Pincus et al., 1999). 10 Passive-sensor estimates of Sc and Cu provide a broader context and help us emphasize the added value of new Sc-Cu discrimination methods based on active-sensor satellites.
Overall, all products identify quite well the large cloud fraction in the tropical and subtropical stratocumulus areas, off the west coast of the continents (Fig. 13, top row). RL-GeoProf and 2BCCL products detect the largest low-level cloud fraction (Fig. 14, zonal and global mean, top row) although they might overestimate the fractionated clouds (e.g., Cu), in particular 15 2BCCL -as shown in the case study analysis-because it uses a different version of the collocated lidar product. Unlike the passive-sensor products -based solely on the CTP and COT-, the active-sensor products do not detect significant amount of Cu off the western coasts of the continents compared to the large Sc cloud fraction, which ranges from 50 % off the coast of Australia up to 85 % in the heart of the deck off the coast of Peru in both GOCCP and RL-GeoProf (Fig. 13, third row). These results are somewhat different from previous analysis in which the Sc cloud fraction ranges from 40 to 60 % over the Sc deck 20 areas (Wood, 2012). On the contrary, Additionally, the CASCCAD products place a substantial amount of Cu clouds west of Sc decks (up to 40 %) and in the trade-wind regions (between 20 and 30 %), similar to MISR observations (Fig. 13, third row), which is more sensitive to fractionated clouds than ISCCP and MODIS. Here again our findings somewhat contradict earlier results retrieving Sc clouds 20 % of the time in the trade-wind regions (Wood, 2012). Besides the Cu and Sc categories, the CASCCAD products have a third category referred to as transitional clouds (Fig. 13, fourth row), which is supposed to capture 25 regions of transition between Cu and Sc clouds. As expected, these clouds are located between Sc decks and trade-wind regions and in the extra-tropics, where one could expect the two types of clouds to co-exist. However, they represent a small part of the total low-cloud fraction (Fig. 14, fourth row). Finally, the ratio of Sc clouds to Sc and Cu clouds document the regions dominated by each type of clouds (Fig. 13, bottom row). Such information could be very useful to evaluate GCMs, which struggle to reproduce the Sc and Cu transition (e.g., Teixeira et al., 2011). The CASCCAD products robustly describe tropical 30 oceans being almost exclusively dominated by Cu clouds while the western coasts of the continents are mostly covered by Sc clouds. Such picture is consistent with previous results from field campaigns, e.g., along the Global Energy and Water Barbados (e.g., Nuijens et al., 2015). Finally, the CASCCAD products classify less Sc than the other products ( Fig. 13 and 14, second row) in the extra-tropics and polar regions (poleward of 35˚), where the cloud feedbacks due to a change in low-cloud cover is somewhat less important -yet non-negligible-than in the tropics (Zelinka et al., 2016). Unsurprisingly, the transitioning Sc-Cu cloud fraction is also the largest in the extratropics where both types have a similar and substantial cloud fraction (30 to 40 %). 5 In all the passive satellite products but MISR, most of the globe is dominated by Sc clouds, with higher frequency of large ratio (> 50%, Fig. 14 . This optical thickness is further used as a threshold to separate Sc (opaque clouds with COT > 3) from Cu clouds (thin clouds with COT < 3), which is about the same optical thickness used in the passive sensor (i.e, COT = 3.6) to distinguish Cu and Sc clouds (i.e., COT = 3.6). As for the passive-sensor satellite observations, this method does not allow a clear separation between the two cloud populations (Fig.  15 S4) although the ratio of Sc to Sc and Cu of derived from the two methods are well-correlated (~ 0.65). However, it confirms that trade-wind regions have smaller opacity and therefore have a different radiative impact on surface and TOA fluxes than more opaque Sc-dominated regions. Figure 15 shows global zonal profiles of cloud fraction for Cu, Sc, transitioning and all low-level clouds as observed by GOCCP, RL-GeoProf and 2BCCL, for the first time. Consistent with the case studies and map analysis, 2BCCL observations retrieve clouds in the low-levels more frequently than GOCCP and RL-GeoProf (Fig. 15, top row) although the difference with RL-GeoProf is rather small compared to that with GOCCP (approximately two times larger). The large difference between GOCCP and RL-GeoProf cloud fractions in the low levels mostly comes from mid and high-level topped clouds (e.g., frontal 25 clouds in the extra-tropics and cumulonimbus and congestus clouds in the tropics, Fig. S5), which typically obscure CALIPSO vision by attenuating the lidar beam before it reaches the low levels. When separated into cloud types, GOCCP and RL-GeoProf observations agree quite well for Sc and transitioning clouds globally and for Cu clouds in the deep tropics (15˚S/N) and down to 2 km in the subtropics and the extra-tropics (Fig. 15, 2 nd row). Between 2 km and 1 km, RL-GeoProf diagnoses more Cu than GOCCP. Such a difference is due to the different sensitivities of the lidar and radar instruments to clouds. The 30 lidar signal becomes quickly attenuated by the optically and geometrically thick Cu -besides the attenuation from overlapping mid or high-clouds-whereas CloudSat radar continues detecting clouds down to 1 km. Below 1 km the surface clutter and the lidar attenuation make it difficult to retrieve a reliable cloud fraction. Consistent with the case study and geographical analysis (Section 4.2.1), the 2BCCL product diagnoses far more (less) Sc (Cu) than the CASCCAD products, making the ratio of Sc to Sc-and-Cu clouds largely dominated by Sc (Fig. 15, bottom row).

Profiles 20
Furthermore, the vertical distribution of Sc and Cu clouds substantially differs from that of the CASCCAD products. 2BCCL Sc clouds may extend up to 3 km while the most part of Cu clouds are concentrated around 1 km and almost exclusively in the tropics. This lack of Cu and excess of Sc clouds above 1 km in tropical subsidence regimes (Fig. 15, right column) is in 5 disagreement the CASCCAD products but also with previous studies focused on Sc regions (Cesana et al., 2019a, their  On the contrary, the CASCCAD products better match typical profiles of Sc and Cu clouds. Finally, these global scale profiles are consistent with the physical processes controlling each type of clouds and measurements 10 from previous literature. The Sc clouds -driven by radiative cooling-cap the PBL, a little higher than 1 km in the tropics and lower toward the poles, and their geometrical thickness is smaller than 1 km. Averaged over the tropical subsidence regimes (and in the extra-tropics, Fig. S6), the Sc vertical cloud fraction peaks between 9 and 12 %, depending on the dataset, whereas it is larger for Sc deck areas only (not shown). On the contrary, Cu cloud base -forced by surface fluxes-mostly form below 1 km, near the LCL, and vertically extend further aloft (around 2.5 km). Although the Cu map cloud fraction is smaller than 15 that of Sc in the tropics (Fig. 13, compare the second and third rows), their vertical cloud fraction is about the same, between 8 and 10 %, because they cover a larger domain (Fig. 15, right column). While the Sc vertical cloud fraction remains unchanged in the extra-tropics, its Cu counterpart appears slightly larger, up to 15 % for RL-GeoProf (Fig. S6).

Conclusion
In this paper, we document spatial distributions and profiles of stratocumulus (Sc) and cumulus (Cu) clouds on a global scale. 20 To this end, we design a discrimination algorithm (DA; Section 3) that distinguishes Sc and Cu based on three observable cloud-properties: cloud top height (CTH), horizontal cloud fraction (HCF) and vertical cloud fraction variability (VCF). These simple criteria are sufficient to characterize the distinctive shape of Cu, which have a limited horizontal extent and highly variable CTH as opposed to Sc, which cover larger areas and have a small and stable geometrical thickness. The DA is utilized The CASCCAD global-scale statistics (Section 4.2) are then compared to a subset of passive-sensor satellite datasets and to the only existing CloudSat-CALIPSO cloud-type climatology 2B-CLDCLASS-LIDAR (2BCCL, Sassen and Wang, 2008). In passive-sensor satellite observations, which distinguish Sc and Cu only based on their cloud optical thickness, Sc and Cu coexist everywhere and no region is fully dominated by a particular type of cloud (Fig. 12, bottom row). On the contrary, Sc clouds largely dominate the global statistics from the 2BCCL point of view, which may mis-diagnose a substantial portion of 5 Cu clouds as Sc clouds in the trade-wind and extra-tropical regions. Interestingly, the CASCCAD observations depict tropical oceans being almost exclusively dominated by Cu clouds (20 to 40 %) while the oceans off the west coasts of the continents are mostly covered by Sc clouds (50 to 85 %), with transitioning clouds in between (10 to 15 %). Our results provide a broader context to earlier findings from ground-based and field campaigns (Albrecht et al., 2019(Albrecht et al., , 1995Bretherton et al., 2010;Comstock et al., 2004;Garay et al., 2008;Wood, 2012;Zhou et al., 2015;Zuidema et al., 2009). For example, our globally-10 averaged profiles of Cu cloud fraction over the tropical oceans are almost identical to that found by Nuijens et al. (2015a) over the Barbados, in terms of shape (cloud base below 1 km and cloud top above 2 km) and frequency of occurrence (~ 10 %).
Another interesting result concerns the distribution and magnitude of Sc cloud fraction. Our results indicate that the Sc clouds occur up to 85 % of the time over Sc deck areas compared to 60 % in earlier studies (i.e., Wood, 2012) and that their presence in trade-wind regions is negligible as opposed to a 20 % cloud frequency (i.e., Wood, 2012). Furthermore, our analysis indicates 15 that the optical thickness, albeit useful, is not a sufficient parameter to discriminate Cu from Sc clouds, in agreement with previous literature (e.g., Pincus et al., 1999).
Finally, by documenting the global geographical distribution of Sc and Cu clouds for the first time, the CASCCAD datasets make it possible to evaluate the shallow convection (Cu type) and boundary layer (Sc type) clouds in state-of-the art climate models, which are typically generated by distinct parametrizations (i.e., Cesana et al., 2019a). By doing so, one could also 20 assess the radiative contribution of Sc and Cu clouds to climate and potentially improve our understanding of low-level cloud feedbacks.

Vertical CF and Continuity Test
3 or more levels with VCF > 0.12 è flagged as Cu   Stratocumulus off the coast of California      c.

Cumulus South East Pacific
d. e.