Between February 2015 and November 2017, the CATS lidar reported profiles at a vertical resolution of 60 m every 350 m along-track, with an average repeat cycle of nearly 3 d (Yorks et al., 2016). CATS Level 2 Operational layer files (L2O files; Palm et al., 2016) describe altitudes where cloud layers were detected within profiles of backscatter coefficients measured at 1064 nm by the CATS lidar (Pauly et al., 2019), averaged 5 km along-track. We considered all such files over the CATS operation period and inspected each 5 km profile within. For profiles located in the Tropics (30∘ S–30∘ N), we inspected each atmospheric layer therein identified as a cloud layer according to the CATS layer type information. As in Noel et al. (2018), we considered layers with a feature type score above 6, to avoid any possibly mislabeled aerosol layers. We flagged the cloud layers with a top altitude above the tropopause. Since any CATS L2O layer entirely above the tropopause is labeled as an aerosol layer (like in CALIPSO; Pan and Munchak, 2011), our study will not include clouds with their base in the stratosphere.

Davis et al. (2010) noted that lidars in space may miss the thinnest subvisible cirrus clouds, but with enough spatial averaging cirrus clouds with optical depths near 0.001 can be detected (Martins et al., 2011). Lidar cloud detections also suffer from a lower sensitivity in the presence of sunlight, which induces significant additional noise in the lidar signal, but climatologies are still relevant (Noel et al., 2018).

To obtain the tropopause height, we considered profiles of temperature and pressure from the ERA5 (European Centre for Medium-Range Weather Forecasts – ECMWF – reanalysis) reanalysis dataset (Albergel et al., 2018). These profiles are available every 6 h, on 37 vertical levels and on a 0.25∘×0.25∘ horizontal grid. Such profiles in the ERA5 reanalysis agree well with observations in the high tropical troposphere (Podglajen et al., 2014). Using these profiles, we computed the vertical lapse rate profile (as in Reichler et al., 2003) and interpolated it on a 100 m vertical grid. We then applied the World Meteorological Organization (WMO) criteria defining the presence of a tropopause – i.e., the lowest altitude at which the lapse rate falls below 2 ∘C km-1, provided the lapse rate between this altitude and all higher altitudes within 2 km does not exceed 2 ∘C km-1 (WMO, 1957). Following the WMO definition, we also allowed for the possibility of a second tropopause if the lapse rate exceeds 3 ∘C km-1 at least 1 km above the first tropopause. In such a case, we started to look for another tropopause above. To limit computation overhead, we constrained the search below 22 km. Using the WMO tropopause definition further allows us to compare our results to previous efforts based on the CALIPSO database that used the same definition (Pan and Munchak, 2011).

For a given CATS 5 km profile (Sect. 2.1), we identified the ERA5 tropopause height (Sect. 2.2) closest in time and location. Given the 6 h time resolution of the ERA5 reanalysis, there is at most a 3 h difference between the observation time and the thermodynamic information used to retrieve the tropopause height. We used the cloud information contained in the 5 km profiles in two ways. First, in 2∘×5∘ latitude–longitude bins we counted how many profiles contained a cloud extending above the tropopause, compared to the total number of profiles in the bins. Aggregating such numbers observed in JJA (June–July–August) and DJF (December–January–February) over the CATS operation period produced seasonal maps of above-tropopause cloud amounts (Sect. 3.1). Second, from each CATS 5 km profile we built a vertical cloud mask, using the tropopause height as the vertical reference and considering clouds that extend above it. Within regions chosen based on the seasonal maps, we aggregated such cloud masks over the same periods as above, keeping also track of the local time of observation for the considered mask. This produced regional vertical cloud fraction profiles above the tropopause, with one profile for each local time of observation (Sect. 3.2).

Figure 1 shows the fraction of CATS profiles in which a cloud is detected above the tropopause, in all DJF (top) and JJA (bottom) months of CATS operation.

Figure 1 shows that clouds in the tropical stratosphere are mostly detected over continents (South America, Equatorial Africa and land masses in the South Warm Pool in DJF; Central America, Central Africa and land masses in the North Warm Pool in JJA). The cloud fraction in the lower stratosphere is largest in DJF, up to 24 % over central Amazonia and coastal areas in the South Warm Pool and up to 20 % over Equatorial Africa. It is significantly lower in JJA, up to 12 % over Africa and 16 % over the North Warm Pool, even though the lowermost stratosphere (380–420 K potential temperature) is moister in JJA than in DJF (cf. e.g., Fig. 8c in Fueglistaler et al., 2009). The seasonal variation of the stratospheric cloud fraction is neither due to changes in the tropopause height, as the tropopause is lower in JJA than in DJF, over all regions (0.2 to 0.9 km altitude difference; see Appendix A), in line with the zonally averaged tropopause heights presented by Fueglistaler et al. (2009) and Rieckh et al. (2014). Several factors may contribute to this seasonal variation: the density and strength of the convective systems (Liu and Zipser, 2005), their propensity to propagate or to be stationary (Houze et al., 2015), and the activity and efficiency of the in situ formation processes (Jensen et al., 2001; Jensen and Pfister, 2004).

The spatiotemporal distribution of the stratospheric clouds is in very good agreement with the 4-year climatology of Pan and Munchak (2011) from CALIPSO observations. The DJF distribution also matches the CALIPSO cirrus detection at 100 hPa reported by Wang et al. (2019) for January 2009 very well. We report lower cloud frequencies than Wang et al. (2019), which can be explained by the fact that we investigated slightly higher altitudes. Both CATS and CALIPSO datasets find (1) significantly weaker stratospheric cloud fraction in JJA than in DJF and (2) near-zero stratospheric clouds in the subtropics. These results are also consistent with the CALIPSO cloud fractions near 16 km reported by Schoeberl et al. (2019). Since those studies consider cloud detections derived from a spaceborne lidar instrument, over several years for most, their good agreement suggests that the CATS stratospheric cloud detections at 1064 nm are as reliable as the CALIPSO ones at 532 nm. A first conclusion of our results is therefore that CATS measurements strongly support the findings of all other studies using detections of high clouds from CALIPSO data.

Our CATS results are also in very good agreement with the distributions of clouds near the tropopause from other space instruments: 2006–2007 HIRDLS (High-resolution dynamics limb sounder) observations reported by Massie et al. (2010), 2006–2014 CloudSat observations (Kim et al., 2018) and the pioneering 1989 passive Stratospheric Aerosol and Gas Experiment (SAGE) II observations (Jensen et al., 1996). Besides the specificity in the cloud detection method employed by each instrument (occultation for HIRDLS and SAGE II and radar backscattering for CloudSat), the small differences between the distributions mostly come from the year-to-year variability. Larger differences can be found with the distributions of clouds penetrating the tropical tropopause derived from the 1998–2000 and 2002–2003 observations by the Tropical Rainfall Measuring Mission (TRMM) Precipitation Radar (Liu and Zipser, 2005). The densities of overshooting systems with tops in the lower stratosphere (on which Liu and Zipser, 2005, focused rather than all stratospheric clouds) are remarkably larger in Central America and Central Africa than over the Warm Pool. Since TRMM precipitation radar reflectivities are less sensitive to thin ice particles than CATS and CALIPSO lidars, we can interpret this difference by the fact that the American and African systems, though frequently overshooting the stratosphere, produce less thin stratospheric clouds than the Asian systems or other in situ processes (like gravity wave cooling) are more efficient to produce stratospheric clouds over Asia than America and Africa.

In contrast with the previous studies, the CATS dataset allows us to analyze the diurnal cycle of the cloud fraction in stratosphere. The cloud fraction at a regional scale shows a consistent diurnal cycle, robust over the different regions identified in the previous section (Fig. 2). In particular and in contrast to the diurnal cycle of surface precipitation, there is no land–ocean difference. All exhibit a pronounced minimum of about 2 %–4 % during the daytime, from 07:00 to 16:00 LT. They all present a first maximum at 19:00 or 20:00 LT (early-night peak), up to 16.5 % over Equatorial Africa. For all regions except South America and the North Warm Pool, this maximum is the largest cloud fraction of the day. All regions also present a second peak (late-night peak) at 00:00 or 01:00 LT (23:00 LT for the West Pacific and 02:00 LT for Central Africa), up to 16.5 % over South America. The midnight peak over Equatorial Africa is less clear than over the other regions because of the large variations between 23:00 and 03:00 LT. The capability of a longer dataset to produce a clearer signal is to be investigated.

The cirrus clouds observed over Amazonia by ground-based lidar (Gouveia et al., 2017) shows a very similar diurnal cycle: a first peak in the early night (at 18:00–19:00 LT) and a second peak later in the night (at 02:00–03:00 LT). Although Gouveia et al. (2017) do not consider the cloud above the tropopause only, their distinction between subvisible, thin and opaque cirrus clouds indicates that the opaque cirrus clouds are predominant during the early night (18:00–21:00 LT) and that the thin cirrus (and subvisible ones during the dry season) dominate during the later night (from 00:00 and 02:00 LT onward, in wet and dry seasons, respectively).

The very deep convection transports cloudy air masses beyond the tropopause via overshoots and then directly contributes to the stratospheric cloud fraction (Dauhut et al., 2016, 2018). The diurnal cycle of the stratospheric cloud fraction observed by CATS can at the first order be explained by the diurnal cycle of very deep convection over land (Liu and Zipser, 2005), especially (i) the minimal value during daytime and (ii) the first peak in the early evening. This first peak occurs with a delay of 3 to 4 h compared to the very-deep-convection maximum. As the dataset used by Liu and Zipser (2005) is more sensitive to overshoots freshly developed into the stratosphere, this delay can be explained by the subsequent horizontal expansion of the overshoots and their spread by the winds (Dauhut et al., 2018; Lee et al., 2019). The convective generation of gravity waves, which produce transient cooling off the convective centers and in some conditions trigger cloud formation, can also contribute to the increase of the stratospheric cloud fraction after the maximum of the very deep convection and then explain the delay of the first peak and potentially the second peak. It may also explain the similar diurnal cycle over the ocean regions, either close (South Warm Pool) or remote (West Pacific) from land masses. This process remains to be investigated.

Figure 3 shows how far above the tropopause the clouds extend, depending on the local time in each tropical region (Sect. 3.1). Some regions are considered in DJF, while others are considered in JJA, because the stratospheric cloud distribution changes throughout the year (Fig. 1), following the ITCZ (Intertropical Convergence Zone) position. Patterns appear very consistent in all the regions considered. In all regions the largest cloud fractions are found near the tropopause, with few clouds extending higher. Cloud fractions extend relatively high (up to 1 km above the tropopause) during the early night. The first peak of cloud fraction, near 19:00–20:00 LT (Fig. 2), is associated with the all-day maximum of cloud vertical extent, with clouds in 5 % of profiles reaching 1 km above the tropopause in DJF regions. During the rest of the night (after 00:00 LT) clouds are still present but extend less high (except over South America). During daytime (06:00–18:00) clouds appear very close to the tropopause. Cloud fractions are overall much smaller in JJA (maximum of 5 %–10 %; Fig. 3d, f) than in DJF (max 10 %–12 %, Fig. 3,a, b, c, e), even if the tropopause is lower in JJA than in DJF (Appendix A).

In addition to describing the evolution of the stratospheric cloud cover at hourly timescales, these observations help interpret observations with limited temporal sampling (Noel et al., 2018). The Microwave Limb Sounder (MLS), like CALIPSO and all other instruments on board platforms of the A-train (Afternoon Train), samples the atmosphere at 01:30 and 13:30 LT, providing one single night and one single day observation. Some authors (e.g., Dion et al., 2019) attempt to retrieve the diurnal cycle of the observed water contents in the tropopause region, combining MLS observations with higher-temporal-resolution observations of convective activity based on TRMM observations of precipitation. Dion et al. (2019) assumed an in-phase relationship between precipitation and ice water content in the upper troposphere and at the tropopause level. For the stratospheric ice water content, MLS data still provide a signal-to-noise ratio that is too low. For future investigations, our results indicate that the stratospheric cloud fraction at 13:30 LT is, whatever the region, close to the minimal value of its diurnal cycle, whereas at 01:30 LT it is more typical of the second maximum. Carminati et al. (2014) investigated, from MLS measurements between 2005 and 2012, the differences between day and night ice water contents in the upper troposphere and the tropopause level. Unlike the stratospheric cloud fraction, tropopause ice water contents are larger at 13:30 LT than at 01:30 LT over Equatorial Africa during DJF, over Central Africa during JJA and over South America during both seasons. A possible explanation to reconcile our results is that tropopause ice water content is more sensitive to fresh convective activity (very-deep-convection occurrence), whereas the stratospheric cloud cover is more sensitive to the diffusion of the injected ice in the stratosphere.