As the initial check, CATS data from nearly the entire mission (March 2015–October 2017) were spatially (within 0.4∘ latitude and longitude) and temporally (±30 min) collocated against ground-based AERONET data. Note that one AERONET measurement may be associated with several CATS retrievals in both space and time, and vice versa. Thus, both CATS and AERONET data were further averaged spatially and temporally, which results in only one pair of collocated and averaged CATS and AERONET data for a given collocated incident. Also, only data pairs with AOD larger than 0 from both instruments are used for the analysis. This step is necessary to exclude CATS profiles with all retrieval fill values as discussed in Sect. 2 (Toth et al., 2018). Such profiles containing all retrieval fill values were found to make up approximately 5.3 % of all CATS profiles in the dataset. Note that the CATS-AERONET comparisons are for daytime only, and higher uncertainties are expected for CATS daytime than nighttime AODs.

As shown in Fig. 1a, without quality-assurance procedures, high spikes in CATS AOD of above 1 (1064 nm) can be found for collocated AERONET data with AOD less than 0.4 (1020 nm). Still, those high spikes in CATS AOD are much reduced compared to the V2-01 CATS aerosol products (e.g., a similar plot to Fig. 1 is included in Appendix A with the use of V2-01 CATS aerosol data). Upon completion of the QA steps as outlined in Sect. 2.1, a reasonable agreement is found between quality-assured CATS (1064 nm) vs. AERONET (1020 nm) AODs with a correlation of 0.65 (Fig. 1b). Comparing Fig. 1a with b, with the loss of only ∼1 %–2 % of collocated pairs due to the QA procedures, we have observed an overall improvement in correlation between CATS and AERONET AOD from 0.51 to 0.65; thus, only quality-assured CATS data are used hereafter. We also found that requiring the extinction QC flag to be equal to 0 and the extinction uncertainty to be less than 10 km-1 had the largest impacts on reducing the difference in mean and medians of the AERONET and CATS AOD. Still, this exercise highlights the need for careful quality checks of the CATS data before applying the CATS data for advanced applications to overcome cloud–aerosol discrimination uncertainties.

To examine over-ocean performance, column-integrated CATS AODs are intercompared with collocated MODIS C6.1 Terra and Aqua DT over-ocean AODs, interpolated to 1064 nm. Over-ocean MODIS C6.1 DT data are selected due to the fact that higher accuracies are reported for over-ocean vs. over-land MODIS DT AOD retrievals (Levy et al., 2013). In addition, compared to over-land MODIS DT data, which provide AOD retrievals at three discrete wavelengths (0.47, 0.55 and 0.65 µm), over-water AOD retrievals are available from seven wavelengths including the 0.86 and 1.24 µm spectral channels, allowing a comparison with CATS AOD at the same wavelength upon logarithmic interpolation, again, assuming the aerosol Ångström exponent value remains unchanged from 0.86 to 1.064 µm as well as from the 1.064 to 1.24 µm spectral channels.

MODIS and CATS AOD retrievals are collocated for the study period of March 2015–October 2017 (Fig. 2). Pairs of CATS and MODIS data were first selected for both retrievals that fall within ±30 min and 0.4∘ latitude and longitude of each other. Then, similar to the AERONET and CATS collocation procedures, collocated pairs were further averaged to construct one pair of collocated MODIS and CATS data for a given collocation incident. Shown in Fig. 2a, a correlation of 0.72 is found between collocated over-water MODIS C6.1 Terra DT and CATS AODs with a slope of 0.74. Similar results are found for the comparisons between over-water MODIS Aqua and CATS AODs with a correlation of 0.74 and a slope of 0.70.

In the previous two sections, AODs from CATS were intercompared with retrievals from passive-based sensors such as MODIS and AERONET. In this section, AOD data from CALIOP, which is an active sensor, are evaluated against AOD retrievals from CATS. Note that despite difference in instrumental designs, CALIOP and CATS are both elastic backscatter lidars. Again, for each collocation incident, pairs of CALIOP and CATS data are selected in which both retrievals fall within ±30 min temporally and 0.4∘ latitude and longitude spatially. There could be multiple CATS retrievals corresponding to one CALIOP data point, and vice versa. Thus, the collocated pairs are further averaged in such a way that only one pair of collocated CATS and CALIOP data is derived for each collocation incident.

Figure 3a shows the comparison of CATS and CALIOP AODs for all collocated pairs including both day- and nighttime. A reasonable correlation of 0.74, with a slope of 0.73, is found for a total of 2762 collocated data pairs. Further breaking down the comparison into day and night cases, a much better agreement is found between the two datasets during nighttime with correlations of 0.83 and 0.81 for over-ocean and over-land cases, respectively. In comparison, a lower correlation of 0.64, with a slope of 0.49, is found between the two datasets, using over-land daytime data only, for a total of 170 collocated pairs. Correspondingly, a lower correlation of 0.55, with a slope of 0.57, is found between the two datasets, using over-ocean daytime data only, for a total of 1180 collocated pairs. This result is not surprising, as daytime data from both CALIOP and CATS are nosier due to solar contamination (e.g., Omar et al., 2013; Toth et al., 2016).

Note that based on the slopes of the regression lines shown in Figs. 1–3, AODs retrieved by CATS are less than AERONET, CALIOP and MODIS Aqua DT AOD retrievals. As shown in Table 1, however, for the 1-to-1 collocated datasets, mean CATS AODs (1064 nm) are ∼10 % higher than AERONET AODs (1020 nm). The CATS AODs are ∼3 % higher than CALIOP AOD (1064 nm) and are ∼5 %–10 % higher than DT MODIS AODs. One possible explanation for this discrepancy is because mean AODs are dominated by low AOD cases and the slopes of the regression relationships are strongly affected by a few high AOD cases. Thus, it is likely that CATS AODs are overestimated at the low AOD ranges and are underestimated at the high AOD ranges.

Also note that as suggested by Omar et al. (2013), the choices of spatial and temporal collocation windows have an effect on collocation results. Thus, we repeated the exercises in Figs. 1–3 by doubling the spatial and temporal collocation windows as well as reducing the collocation windows by half. The descriptive statistics of this sensitivity study are included in Table 2. While the number of collocated data pairs is drastically affected by the spatial and temporal collocation window sizes, less significant changes are found in descriptive statistics such as mean, median and standard deviations of AODs, as well as slopes and correlation values. The slope of MODIS Aqua DT and CATS AODs, however, seems sensitive to changes in collocation methods. Changes in slope of 0.61 to 0.78 are found for the change of temporal collocation window from 15 to 60 min with a fixed spatial collocation window of 0.4∘ latitude/longitude.

Still, larger discrepancies between CATS and CALIOP AODs during daytime indicate that both sensors are susceptible to solar contamination. To overcome solar contamination and more accurately detect aerosol layers, CALIOP and CATS data products are averaged up to 80 and 60 km, respectively. Noel et al. (2018) found that the feature type score can be used for cloud screening throughout the diurnal envelope of solar angles. To further evaluate impact of the solar-contamination-introduced bias in the diurnal analysis in aerosol detection or products, CATS AODs are evaluated as a function of local time. For each CATS observation of a given location and UTC time, the associated local time is computed by adding the UTC time by 1 h per 15∘ longitude away from the prime meridian in the east direction. Figure 4a shows the CATS AOD vs. local time for both global land and oceans, constructed using 6 h mean CATS AOD binned on a 5∘ by 5∘ grid globally. While the data have additional noise, no major deviations in AODs are found during either sunrise or sunset times, although we speculate that larger uncertainties in CATS AODs and extinctions may be present around day and night terminators. Figure 4b shows a similar plot to Fig. 4a but with the region restricted to 25–52∘ S. Here, we want to investigate the variations in CATS AODs as a function of local time, over relatively aerosol-free oceans. We picked 25∘ S as the cutoff line as CATS data are only available to 51.6∘ S (limited to the ISS inclination angle), and thus this threshold is used to ensure enough data samples in the analysis, although some land regions are also included. As indicated in Fig. 4b, again, no significant deviations in pattern are found for both sunrise and sunset times, plausibly indicating that solar contamination, as speculated, may not be as significant. Comparing the mean AOD at local midnight to the mean AOD at local noon by performing a Student's t test, the difference is not significant at the 95 % confidence level, with a p value of 0.16.

Figure 4c shows the difference between AERONET (1020 nm) and CATS (1064 nm) AOD (ΔAOD) as a function of local time. Again, although data are rather noisy, no major pattern is found near sunrise or sunset times, further indicating that solar contamination during dawn or dusk times may have a less severe impact on CATS AOD retrievals from a long-term mean perspective. In summary, Sect. 3.1.1–3.1.3 suggest that with careful QA procedures, AOD retrievals from CATS are comparable to those from other existing sensors such as AERONET, MODIS and CALIOP at the same local times.

One advantage of CATS is its ability to retrieve both column-integrated AOD and vertical distributions of aerosol extinction. Therefore, in this section, extinction profiles from CATS are compared with that from CALIOP. Again, similar to Sect. 3.1.3, collocated profiles for CATS and CALIOP are first found for both retrievals that are close in space and time (within ±30 min and 0.4∘ latitude and longitude). However, different from Sect. 3.1.3, only one pair of collocated CATS and CALIOP profiles, which has the closest Euclidian distance on the Earth's surface, is retained for each collocated incident.

The CATS cloud–aerosol discrimination (CAD) algorithm is a multi-dimensional probability density function (PDF) technique that is based on the CALIPSO algorithm (Liu et al., 2009). The PDFs were developed based on cloud physics lidar (CPL) measurements obtained during over 11 field campaigns and 10 years. As shown in Fig. 5e, a reasonable agreement is found between CATS V3-00 aerosol extinction with CALIOP for over land. However, CATS overestimates aerosol extinction around 1 km compared to CALIOP over ocean (Fig. 5d). This can also be seen on a plot of the difference between CATS and CALIOP 1064 nm extinction for all collocated profiles, included in Fig. 5f, where there is an overall positive difference around 1 km.

Due to the precessing orbit of the ISS, the CATS sampling is irregular and very different compared to the Sun-synchronous orbits of the A-Train sensors. These orbital differences between CATS and CALIOP make comparing the data from these two sensors challenging since they are fundamentally observing different locations of the Earth at different times. Thus, we should not expect the extinction profiles and AOD from these two sensors to completely agree. Additionally, there are other algorithm and instrument differences that can lead to differences in extinction coefficients and AOD. Over land where dust is the dominant aerosol type, differences in lidar ratios between the two retrieval algorithms (CATS uses 40 sr while CALIOP uses 44 sr) can cause CATS extinction coefficients that are up to 10 % lower than CALIOP, potentially explaining the higher CALIOP extinction values in Fig. 5e. Over ocean, especially during daytime, differences in CATS and CALIOP lidar ratios for marine and smoke aerosols can introduce a difference between CATS and CALIOP extinction coefficients (Fig. 5d). These differences in over-ocean data (Fig. 5d) could also be attributed to differences in CATS and CALIOP 1064 nm backscatter calibration. For example, Pauly et al. (2019) reports that CATS attenuated total backscatter is about 19.7 % lower than PollyXT measurements in the free troposphere and 18.2 % lower than CALIOP opaque cirrus clouds due to calibration uncertainties for both sensors.

Also, differences in the lowest 250 m between CATS and CALIOP extinction profiles are observable, which are due to how the instrument algorithms detect the surface and near-surface aerosols. Both the CATS and CALIOP feature detection algorithms create a gap between the surface and near-surface aerosol base altitude, despite the possible presence of aerosols in this altitude region. CALIOP has an aerosol base extension algorithm that is designed to (1) detect scenarios when aerosols are present in the bins just above the surface and (2) extend the near-surface aerosol layer base down to the surface (Tackett et al., 2018). However, CATS does not use such an algorithm, so false regions of “clear air” exist between the surface and near-surface aerosol layers.

Vertical profiles of collocated CATS and CALIOP extinction for daytime-only profiles and nighttime-only profiles are shown in Fig. 5b and c, respectively. Compared to a total collocated pair count of 2748 in the overall profile data, day and night profiles have 1311 and 1437 collocated pairs, respectively. Again, the shapes of the CATS and the CALIOP extinction vertical profiles are similar for all three cases, despite the abovementioned offsets in altitude. Figure 5d and e show the mean of those extinction profiles which occurred over water and over land, as defined by the CATS surface type flag. Again, in both cases, CATS and CALIOP have similar shapes in their vertical extinction profiles. The vertical structure of over-water extinction is also very similar to that of all profiles, day and night, which is perhaps not surprising as water profiles made up 2142 of 2748 (∼78 %) collocated pairs. The vertical structure of over land is more different than the other groups, as the extinction is higher throughout a larger depth of the atmosphere, tapering off much more slowly from the surface. Furthermore, the peak extinction from CATS is actually lower than CALIOP for over-land profiles, unlike all other categories.

Figure 6a–b show the spatial distributions of CATS AODs at the 1064 nm spectral channel for boreal winter–spring (December–May, DJFMAM) and boreal summer–fall (June–November, JJASON) seasons, for the period of March 2015–October 2017. To construct Fig. 6a and b, quality-assured CATS AODs are first binned on a 5∘ by 5∘ grid over the globe for the abovementioned two bi-seasons. For each 5∘ × 5∘ (latitude/longitude) bin, for a given season, CATS AODs are averaged on a pass basis first and then further averaged seasonally to represent AOD value of the given bin. Both daytime and nighttime retrievals are included in this figure, as well as in Figs. 7–9.

In the DJFMAM season, significant aerosol features are found over north Africa, the Middle East, India and eastern China. For the JJASON season, besides the abovementioned regions, aerosol plumes are also observable over south Africa, related to summer biomass burning of the region (e.g., Eck et al., 2013). The seasonal-based spatial distributions of AODs from CATS, although reported at the 1064 nm channel which is different from the 550 nm channel that is conventionally used, are similar to some published results (e.g., Lynch et al., 2016).

For comparison purposes, Fig. 6c–d show similar plots to Fig. 6a–b but with the use of CALIOP AOD at the 1064 nm spectral channel. Note that those are climatological means rather than pairwise comparisons. While patterns are similar in general, at regions with peak AODs of 0.4 or above for CALIOP, such as north Africa for the DJFMAM season and north Africa, the Middle East and India for the JJASON season, much lower AODs are found for CATS. However, in some other regions, such as over south Africa for the JJASON season, higher CATS AOD values are observed. A table of mean AOD across each of these regions as well as over the globe (within the latitude range where CATS has data) has been included for reference (Tables 3). Figure 6e and f show similar spatial plots to Fig. 6a and b but with the use of MODIS Aqua AODs from the DT products (using all available MODIS DT retrievals that passed QA steps as described in Sect. 2.3). For the MODIS Aqua DT products, aerosol retrievals at the shortwave infrared channels are only available over oceans, and thus Fig. 6e–f show only over-ocean retrievals. Again, while general AOD patterns look similar, discrepancies are also visible, such as over the coast of southwest Africa for the JJASON season and over the west coast of Africa for the DJFMAM season. Those discrepancies may result from biases in each product, but it is also possibly due to the differences in satellite overpass times, as CALIOP provides early morning and afternoon overpasses, and MODIS Aqua has an overpass time after local noon, while CATS is able to report atmospheric aerosol distributions at multiple times during a day.

Similar to Fig. 6a and b, Fig. 7a and b show the spatial distribution of CATS AODs but for CATS extinction values that are below 1 km above ground level (a.g.l.) only, for the DJFMAM and JJASON seasons, respectively. Figure 7c and d show the CATS mean AOD plots for extinction values from 1 to 2 km a.g.l., while Fig. 7e and f show CATS mean AOD for extinction values above 2 km a.g.l.. For the DJFMAM season, elevated aerosol plumes with altitude above 2 km a.g.l. are found over north Africa. For the JJASON season, elevated dust plumes (> 2 km a.g.l.) are found over the north African and the Middle Eastern regions, while elevated smoke plumes are found over the west coast of south Africa where above-cloud smoke plumes are often observed during the northern hemispheric summer season (e.g., Alfaro-Contreras et al., 2016).

CATS has a non-Sun-synchronized orbit, which enables measurements at nearly all solar angles. Thus, we also constructed 5∘ × 5∘ (latitude/longitude) gridded seasonal averages (for DJFMAM and JJASON seasons) of CATS AODs at 00:00, 06:00, 12:00 and 18:00 UTC that represent four distinct times in a full diurnal cycle, as shown in Fig. 8. To construct the seasonal averages, observations within ±3 h of a given UTC time as mentioned above are averaged to represent AODs for the given UTC time. On a global average, the mean AODs are 0.090, 0.089, 0.088 and 0.089 for 00:00, 06:00, 12:00 and 18:00 UTC, respectively, for the JJASON season and are 0.099, 0.096, 0.093 and 0.093 for the DJFMAM season. Thus, no significant diurnal variations are found on a global scale.

Still, strong diurnal variations with the maximum averaged diurnal AOD changes of above 0.10 can be observed for regions with significant aerosol events such as north Africa, the Middle East and India for the DJFMAM season and north Africa, south Africa, the Middle East and India for the JJASON season, as illustrated in Fig. 9. Note that Fig. 9a shows the maximum minus minimum seasonal mean AODs for the four different times, as shown in Fig. 8a, c, e, g. Similarly, Fig. 9b shows the maximum minus minimum seasonal mean AODs for the four different times as shown in Fig. 8b, d, f, h. Interestingly but not unexpectedly, regions with maximum diurnal variations match well with locations of heavy aerosol plumes as shown in Figs. 6 and 8.

Using quality-assured CATS-derived aerosol vertical distributions, mean global CATS extinction vertical profiles are also generated as shown in Fig. 10. Similar to steps as described in Sect. 3.2.1, CATS extinction profiles are binned into 00:00, 06:00, 12:00 and 18:00 UTC times based on the closest match in time for the JJASON and DJFMAM seasons. Figure 10a shows the daily averaged CATS extinction profiles on a black line, and 00:00, 06:00, 12:00 and 18:00 UTC averaged on blue, green, yellow and red lines, respectively, for the DJFMAM season. A similar plot is shown in Fig. 10d for the JJASON season. CATS extinction profiles for the daily average as well averages for the four selected times are similar, suggesting that minor temporal variations in CATS extinctions can be expected for global averages.

Those global averages are dominated by CATS profiles from global oceans (Fig. 10b and e), which also have small diurnal variations, as ∼70 % of the globe is covered by water. In comparison, noticeable diurnal changes in aerosol vertical distributions are found over land as shown in Fig. 10c and f. For the DJFMAM season, at the 1 km altitude, the minimum and maximum aerosol extinctions are at 12:00 and 18:00 UTC, respectively. Similarly, the minimum and maximum aerosol extinctions are at 12:00 and 00:00 UTC below 400 m. For the JJASON season, the minimum aerosol extinction values are found at 12:00 UTC for the whole 0–2 km column, while the maximum aerosol extinction values are at 18:00 UTC for 1.5 km and 00:00 UTC for the 300–400 m altitude. Still, it should be noted that aerosol concentrations may be a function of local time, yet for a given UTC time, local times will vary by region. Also, due to solar contamination, nighttime retrievals from CATS are significantly and demonstrably less noisy than daytime retrievals, and this difference in sensor sensitivity between day and night may further affect the derived diurnal variations in CATS AOD and aerosol vertical profiles as shown in Fig. 3 for individual retrievals. Still, no apparent solar pattern is detectable from Fig. 8, and only minor diurnal variations are found for Fig. 10a and d, which indicate that such a solar contamination may introduce noise but not bias to daytime aerosol retrievals from a global mean perspective.

However, if we examine the mean global CATS extinction vertical profiles with respect to local time, as shown in Fig. 11, some distinct features appear. For example, Fig. 11a and d suggest that, on global average, the minimum aerosol extinction below 1 km is found for 18:00 local time (LT), for both JJASON and DJFMAM seasons. Similar patterns are also observed for over global oceans. However, for over-land cases, for both seasons, the minimum and maximum aerosol extinction below 500 m is found for 12:00 and 00:00/06:00 LT.

In this section, the diurnal variations of aerosol vertical distributions are studied as a function of local solar time for selected regions with high mean AODs as highlighted in Fig. 6. Note a near-1-to-1 transformation can be achieved between UTC and local solar time. Also, as learned from the previous section, aerosol features are likely to have a local time dependency. A total of four regions, including north Africa, the Middle East, India and northeast China, which show significant seasonal mean AODs in Fig. 6, are selected for the DJFMAM season (Fig. 12). For the JJASON season (Fig. 13), in addition to the abovementioned four regions, the south African region is also included due to biomass burning in the region during Northern Hemisphere summertime. The latitude/longitude boundary of each selected region is described in Table 4. Regional-based analyses are also conducted for four selected regions for the DJFMAM season and five selected regions for the JJASON season at four local times: 00:00 (midnight), 06:00, 12:00 and 18:00 LT, using quality-assured CATS profiles. Generally, the maximum diurnal change in aerosol extinction is found at the altitude of below 1 km for all regions as well for both seasons. Also, larger diurnal variations in vertical distributions of aerosol extinction are found for the JJASON season, in comparison to the DJFMAM season, while regional-based differences are apparent.

For the north African region, the dominant aerosol types are dust and smoke aerosol for the DJFMAM season and dust for the JJASON season (e.g., Remer et al., 2008). Interestingly, the maximum aerosol extinction below 500 m is found at 06:00 LT for the DJFMAM season. While for the JJASON season, the maximum aerosol extinctions are found at 0:00/06:00 LT for the 100–500 m layer, with a significant ∼10 %–20 % higher aerosol extinction from the daily mean. Note that 06:00 LT in the north African region corresponds to early morning, which has been identified in several studies (Fiedler et al., 2013; Ryder et al., 2015) as the time of day when the nocturnal low-level jet breakdown causes large amounts of dust emission in this region. Thus, we suspect that this 06:00 LT peak in maximum aerosol extinctions may be the signal resulting from the low-level jet ejection mechanism captured on a regional scale. As the day progresses into the afternoon and early evening, we find the aerosol heights shifting upwards, likely related to the boundary layer's mixed layer development.

For the Middle Eastern region, for the JJASON season, a daily maximum in aerosol extinction of ∼0.15 km-1 is found at midnight (00:00 LT), with a daily minimum of ∼0.08 km-1 found at local noon (12:00 LT), for the peak aerosol extinction layer that has a daily mean aerosol extinction of ∼0.12 km-1. This translates to a ∼±20 %–30 % daily variation for aerosol extinction for the peak aerosol extinction layer. Smaller daily variation in aerosol extinction, however, is found for the same region for the DJFMAM season.

For the India region, for the JJASON season, a large peak in aerosol extinction of up to 10 % higher than daily mean is found at 06:00 LT below 500 m. The minimum aerosol extinction is found at 12:00/18:00 LT for the layer below 500 m and is overall ∼10 % lower than the peak daily mean aerosol extinction value. For the DJFMAM season, minimum aerosol extinctions are found at 12:00 LT for near the whole 0–2 km column, while for the layer below 500 m, the maximum aerosol extinction values are found at midnight (00:00 LT).

For the northeast China region, a significant peak is found at the 500 m–1 km layer for local afternoon (18:00 LT) for the DJFMAM season. A similar feature is also found for the JJASON season, while the peak extinction for the JJASON season happens at 06:00 LT for the aerosol layer below 500 m. Lastly, for the south African region, biomass burning aerosols are prevalent during the summertime, and thus only the JJASON season is analyzed. As shown in Fig. 13b, below 500 m in altitude, lower extinction values are found for local afternoon (18:00 LT) and higher extinction values are found for local morning or early morning (00:00 and 06:00 LT).