Characterizing the diurnal cycle of South Atlantic stratocumulus cloud properties from satellite retrievals

Abstract. Marine stratocumulus (Sc) clouds play an essential role in the earth radiation budget. Here, we compare liquid water path (LWP), optical thickness (COT), and effective radius (CER) retrievals from two years of collocated Spinning Enhanced Visible and InfraRed Imager (SEVIRI), MODerate resolution Imaging Spectroradiometer (MODIS), and Tropical Rainfall Measuring Mission Microwave Imager (TMI) observations, estimate the effect of biomass burning smoke on passive imager retrievals, as well as evaluate the diurnal cycle of South Atlantic marine Sc clouds. The effect of absorbing aerosols from biomass burning on the retrievals was investigated using aerosol index (AI) obtained from the Ozone Monitoring Instrument (OMI). SEVIRI and MODIS LWPs were found to decrease with increasing AI relative to TMI LWP, consistent with well-known negative visible/near-infrared retrieval biases in COT and CER. In the aerosol-affected months of July–August–September, SEVIRI LWP – based on the 1.6-µm CER – was biased low by 14 g m −2 (~ 16 %) compared to TMI in overcast scenes, while MODIS LWP showed a smaller low bias of 4 g m −2 (~ 5 %) for the 1.6-µm channel and a high bias of 8 g m −2 (~ 10 %) for the 3.7-µm channel compared to TMI. Neglecting aerosol-affected pixels reduced the mean SEVIRI-TMI LWP bias considerably. On a two-year data base, SEVIRI LWP had a correlation with TMI and MODIS LWP of about 0.86 and 0.94, respectively, and biases of only 4–8 g m −2 (5–10 %) for overcast cases. The SEVIRI LWP diurnal cycle was in good overall agreement with TMI except in the aerosol-affected months. Both TMI and SEVIRI LWP decreased from morning to late afternoon, after which a slight increase was observed. Terra and Aqua MODIS mean LWPs also suggested a similar diurnal variation. The relative amplitude of the two-year mean and seasonal mean LWP diurnal cycle varied between 35–40 % from morning to late afternoon for overcast cases. The diurnal variation in SEVIRI LWP was mainly due to changes in COT, while CER showed only little diurnal variability.


Introduction
Changes in marine boundary layer (MBL) clouds over eastern subtropical oceans and associated differences in cloud radiative forcing are thought to be the main source of uncertainty in climate feedback simulations (Bony and Dufresne, 2005;Meehl et al., 2007, Zelinka et al., 2017).Climate models do not yet adequately parameterize the physical and dynamical processes affecting the formation of these clouds and fail to represent their variability on (2002) showed that the amplitude of diurnal variations in cloud amount and LWP could exceed 20 % of the mean value.These studies, however, did not consider diurnal variations in cloud optical thickness (COT) or droplet effective radius (CER) and were usually based on limited measurements from a single instrument, the uncertainties of which were not well characterized.Painemal et al. (2012) evaluated the diurnal cycle of LWP, COT, and CER for 110 southeast Pacific Sc based on GOES-10 (Geostationary Operational Environmental Satellite-10) and microwave observations, but only for a period of two months.They noted that variations in COT drive the diurnal cycle of LWP mostly.
In this study, we investigate the diurnal variations of southeast Atlantic Sc clouds.The southeast Atlantic domain is notable for its unique feature that part of the year a smoke layer transported from the continent resides 115 above the Sc clouds, which poses a challenge to the retrieval of aerosol and cloud properties from space.In recent years several field campaigns have been initiated as described in Zuidema et al. (2016), to investigate aerosol-cloud interactions and their role in climate.The purpose of our study is three-fold.One, to compare the Meteosat Second Generation (MSG) Spinning Enhanced Visible and InfraRed Imager (SEVIRI) CLoud property dAtAset using  (Wentz 2018) and Collection 6 MODIS (Moderate Resolution Imaging Spectro-radiometer) retrievals (Platnick et al., 2017).Two, to quantify the effect of above-cloud aerosols on LWP retrievals from the SEVIRI and MODIS passive imagers.
Three, to study the diurnal cycle of Sc clouds in the South Atlantic, which is a somewhat neglected region as most previous studies focused on the North or South Pacific (west of California and Chile).The main strength of our 125 study is the use of an extensive two-year dataset, which allows us to investigate the seasonal variation of the diurnal cycle.SEVIRI's higher temporal resolution of 15 minutes allows examining the diurnal cycle with greater detail than offered by earlier GOES instruments.We only consider non-raining warm liquid clouds to avoid significant retrieval uncertainties associated with the presence of rain and ice clouds at higher altitudes.Retrieval artifacts related to absorbing aerosols (e.g., Haywood et al., 2004) have been evaluated and aerosol-affected grid boxes have 130 subsequently been removed from the analysis.
The paper is structured as follows.A description of our datasets including retrieval artifacts and uncertainties is provided in Section 2. The comparison methodology is described in Section 3. Section 4 discusses retrieval biases related to the presence of smoke from continental biomass burning over clouds and analyzes spatial distributions, comparison statistics, and diurnal variations of Sc properties from SEVIRI, TMI, and Terra and Aqua 135 MODIS on seasonal and two-year timescales.Finally, a summary is offered in Section 5.

Spinning Enhanced Visible and InfraRed Imager (SEVIRI)
140 SEVIRI is an optical radiometer onboard the MSG geostationary satellite series operated by the European Organization for the Exploitation of Meteorological Satellites (EUMETSAT).SEVIRI measures radiances in 12 spectral bands including 4 VIS/NIR channels (0.6-1.6 µm plus a broadband high-resolution VIS channel) and 8 IR channels (3.9-13.4µm).It has a spatial resolution of 3x3 km 2 at nadir and a repeat frequency of 15 minutes for fulldisk images covering Europe, Africa, and the Atlantic Ocean.

145
The CM SAF CLAAS-2 climate data record is described in Benas et al. (2017).Part of the cloud processing software is the CPP (cloud physical properties) algorithm, which retrieves cloud optical thickness and cloud particle effective radius based on measured reflectances in the 0.6-µm and 1.6-µm channels.The retrieval scheme is based on earlier bispectral methods (hereafter also referred to as visible/near-infrared or VIS/NIR technique) that retrieve cloud optical thickness and cloud particle size from satellite radiances at wavelengths in the 150 (for clouds) non-absorbing visible and the moderately absorbing solar infrared part of the spectrum (Nakajima and King 1990;Han et al. 1994;Nakajima and Nakajima, 1995;Watts et al., 1998;Roebeling et al., 2006].The liquid water path is computed from the retrieved optical thickness (τ or COT) and droplet effective radius (r e or CER) as where ρ l is the density of liquid water (Stephens 1978). (1) The SEVIRI retrievals are available only during daytime and are performed assuming plane parallel clouds.Because 155 r e is not well constrained by the measured 1.6-µm channel reflectance for thin clouds, it is weighted towards a climatological a priori value of 8 µm for pixels with τ ≤ 4 -similar to the handling of small optical thicknesses in optimal estimation methods.The relationship used to weight the CER retrieval is, where, w = 1 (1 + e (!!.!"(! !"# !! !,!"#$ )) ), r e,clim = 8 micron; τ w,clim = 2.5 160 In part of our analysis, a τ > 3 threshold is applied to minimize the impact of strongly weighted effective radii for thin clouds on the results.The SEVIRI shortwave channels were calibrated with Aqua-MODIS as described in Meirink et al. (2013).More details on the CPP retrieval algorithm are provided in CM SAF (2016).

165
MODIS is the flagship instrument aboard the Terra and Aqua polar orbiter satellites.Terra has a 10:30am COT and CER bins, redesigned cloud thermodynamic phase detection based on a variety of independent tests, and separate spectral retrievals of COT, CER, and derived LWP for channel combinations using the 1.6, 2.2, and 3.7-µm 180 bands.Differences in CER between C5 and C6 are evaluated in Rausch et al. (2017).Depending on a subpixel heterogeneity index, the properties of partly cloudy pixels are listed separately and the algorithm also provides retrieval failure metrics for pixels where the observed reflectances fall outside the LUT solution space.

185
While these datasets offer excellent resources for investigating warm, overcast single-layer clouds, they are subject to certain retrieval artifacts due to algorithm assumptions and complexities in the retrieval technique.The VIS/NIR cloud property retrievals rely on 1-D radiative transfer model-generated LUTs, which do not account for subpixel cloud heterogeneity and 3-D cloud structure, and that could lead to significant biases in retrieved cloud properties for inhomogeneous and partially cloudy scenes.Cloud vertical stratification is essential to consider when computing 190 LWP.Although MODIS retrieves effective radius at three separate water-absorbing channels, 1.6, 2.2, and 3.7-µm, all three are most sensitive to near cloud-top properties (Platnick 2000;Zhang and Platnick 2011).Hence, the LWP derived by combining retrieved τ and retrieved r e from any one of the near IR channels could potentially under-or overestimate the true value depending upon the actual cloud stratification.For stratocumulus that typically follows a sub-adiabatic r e profile, bigger droplets will be located near cloud top, and thus the derived LWP could be an 195 overestimate.As a first-order correction, an adiabatic model is proposed by Wood and Hartmann (2006), which results in a ~17% reduction from the standard vertically homogeneous LWP in eq. 1 (Bennartz 2007;Bennartz and Rausch, 2017).More details about the retrieval uncertainties of the VIS/NIR technique can be found in Horváth and Davies (2007), Seethala and Horváth (2010), Horváth et al. (2014), Zhang et al., (2012), Grosvenor et al., (2018)

250
Because microwave and optical techniques represent fully independent approaches, each having their own shortcomings, the analysis of retrieval discrepancies does not necessarily establish absolute accuracies.A considerable number of studies have investigated the differences between LWP retrievals based on passive microwave and VIS/NIR satellite observations (Bennartz, 2007;Borg and Bennartz, 2007;Horváth and Davies, 2007;Horváth and Gentemann, 2007;Wilcox et al., 2009;Greenwald, 2009;Seethala and Horváth, 2010;Horváth 255 et al., 2014, Cho et al., 2015;Greenwald et al., 2018).The major shortcomings of microwave measurements were found to be the uncertain retrieval of LWP in the presence of rain and a wet (positive) bias of 10-15 g m -2 in broken cloud fields.However, V7 TMI data now includes the small negative LWP values that were previously discarded, and thus the microwave wet-bias has been significantly reduced (Greenwald et al., 2018).1. 'all-sky': including all grid boxes in the identified Sc domain.
These criteria were imposed to minimize retrieval artifacts related to broken clouds as well as thin 280 clouds for which the CER retrieval in particular is relatively uncertain.

Effect of biomass burning smoke on SEVIRI and MODIS retrievals
This section presents the analysis of the effect of smoke and/or aerosols above marine Sc on passive VIS/NIR 285 imager retrievals of cloud properties.Our study domain, especially the Sc region located off the Namibia coast, is severely influenced by biomass burning on the African continent, as it produces episodic plumes of dark smoke that drift over the southeast Atlantic Ocean during the dry season JJASO (June-through-October).Beneath the elevated smoke layer, there is a persistent deck of bright marine Sc clouds.Previous research (Hobbs, 2002;McGill et al., 2003;Wilcox, 2010) has shown that the smoke is typically located in layers (at 2 to 4 km altitude) that are vertically 290 separated from the Sc clouds below (at ~1.5 km altitude) and, hence, direct microphysical interaction between the aerosols and the Sc is often inhibited by the strong temperature inversion above the cloud layer.However, more recent studies e.g., Rajapakshe et al. (2017) reported that smoke layers are closer to the cloud layer, and significantly enhance the brightness of stratocumulus over there (Lu et al. 2018).Recently, several studies evaluated the satellite and/or field campaign measurements (Adebiyi et al., 2015;Adebiyi and Zuidema, 2016;Zuidema et al., 2016;Das et al., 2017;Horowitz et al., 2017;Chang and Christopher, 2017;Lu et al., 2018;Kar et al., 2018).When smoke resides above low-level clouds, the observed visible channel (0.6-or 0.8-µm) reflectance is reduced due to absorption by smoke, which is not taken into account in the LUTs and can introduce a negative bias in the retrieved COT as well as CER, and hence in LWP.According to Haywood et al. (2004), this negative bias in the 1.6-µm CER 300 is significantly larger than that in the 2.1-µm CER (which is estimated to be less than 1 µm), while the bias in retrieved COT can be up to 30 %.Previous studies also noticed a domain-mean underestimation of ~3 to 6 g m -2 in MODIS LWP over the South Atlantic Sc region in the presence of absorbing aerosols (Bennartz and Harshvardhan, 2007;Wilcox et al., 2009;Seethala and Horváth, 2010).Therefore, we need to quantify the impact of absorbing aerosols on SEVIRI and MODIS VIS/NIR retrievals in our Sc domain for our study period.The presence of 305 absorbing aerosols can be diagnosed using the OMI Aerosol Index (AI), because large positive AIs correspond to absorbing aerosols, such as dust and smoke, and small positive or negative AIs correspond to non-absorbing aerosols and clouds.In Fig. 2, cloud properties from TMI, SEVIRI, and MODIS retrievals are binned into AI bins of 0.5 for the overcast Sc conditions.In the SEVIRI 1.6-µm CER retrievals, a steady and strong decrease from 11 to 6 µm is observed, while the COT decrease is weaker from 10.8 to 9 with AI increasing from 0 to 3.5.As a result, SEVIRI 320 LWP sharply decreases from 86 to 45 g m -2 over the same AI range.TMI LWP, in contrast, increases from 84 to 101 g m -2 between clean and increasingly polluted regions.For overcast grid boxes with little to no smoke absorption (AI < 0.5), SEVIRI LWP agrees well with TMI LWP, having only a 2 g m -2 high bias.However, SEVIRI has a low

325
Considering that cloud amount happens to be spatially correlated with AI and that microwave retrievals are unaffected by absorbing aerosols, the increase in TMI LWP with increasing AI, that is closer to shore, seems plausible.Because the variability of LWP is mostly controlled by COT rather than CER in the absence of smokeinduced retrieval biases (Seethala and Horváth, 2010;Painemal et al., 2012), the microwave retrievals suggest that the true COT should also increase with AI.Taken together, the microwave and VIS/NIR retrievals imply that 330 SEVIRI COT is increasingly underestimated as AI increases.The low bias in SEVIRI LWP in smoke-affected areas arises from the combination of the negative COT and CER retrieval biases.A similar underestimation is reported in aircraft retrievals of COT and CER for a stratus deck residing below an absorbing aerosol layer (Coddington et al., 2010).
Interestingly, a systematic overall increase in LWP with AI as indicated by TMI LWP in Fig. 2a was also 335 noticed in previous observational and modeling studies, e.g., Johnson et al. (2004), Wilcox (2010), Randles and Ramaswamy (2010), Adebiyi et al. (2015), Adebiyi and Zuidema (2016).While this could partly be explained by the fortuitous spatial correlation between higher aerosol loads and thicker clouds in this Sc region, these studies argue that strong atmospheric absorption by the smoke warms the 700 hPa air temperature and increases upward motion.
This increased buoyancy inhibits cloud-top entrainment and promotes a stronger inversion, thereby helping to  The mean MODIS LWPs are 80 g m -2 , 87 g m -2 , and 90 g m -2 respectively for 1.6-, 2.1-, and 3.7-µm channel retrievals, while the corresponding mean SEVIRI LWP is 71 g m -2 .As shown in Fig. 2a, MODIS 1.6-µm retrieved LWP shows the largest decrease from 92 to 72 g m -2 with AI.In clean cases, MODIS 1.6-µm LWP is 10 % higher than the SEVIRI 1.6-µm LWP and the difference between MODIS and SEVIRI LWP is even larger for the 2.1-and 3.7-µm channel retrievals.
MODIS COT decreased slightly until AI < 1.5, increased steeply until AI = 2.5, and then leveled-off in all 365 three channels.However, SEVIRI and MODIS COTs differ by 1 with MODIS being higher even in grid-boxes unaffected by smoke.Taken together both COT and CER variations, MODIS LWPs show a decreasing trend with AI in all three channels, with the largest decrease of ~20 g m -2 seen in the 1.6-µm retrieval.The 2.1-and 3.7-µm MODIS LWPs show a reduction of only ~10 g m -2 .The SEVIRI minus MODIS differences in LWP, COT, and CER increased with AI even in the common 1.6-µm channel; although differences were the smallest in this channel, 370 especially for CER.This is somewhat surprising, considering that the CLAAS-2 SEVIRI and MODIS C6 COT-CER retrieval algorithms are rather similar, the SEVIRI 1.6-µm channel has been calibrated with the corresponding MODIS channel, and the comparison is done for the most favorable overcast condition.The finding that AI has a stronger impact on SEVIRI 1.6-µm LWP than on MODIS 1.6-µm LWP may partially be explained by the spectral difference that for SEVIRI retrievals the 0.6-µm channel is used as a non-absorbing channel in contrast to the 0.8-375 µm channel for MODIS, the latter of which is less affected by aerosol absorption.
Because the presence of absorbing aerosols above Sc clouds introduces a large negative bias in both SEVIRI and MODIS COT and CER retrievals, in the remainder of this work we will exclude grid-boxes with AI>0.1.The spatial distributions of two-year-mean SEVIRI cloud properties and TMI LWP for the overcast condition are shown in Fig. 3, whereas the results for the all-sky case are shown in Fig. S3.In the all-sky case, the spatial distribution of LWP indicates that over the marine Sc region the measurement techniques show good agreement with negligible bias, while the two-year mean SEVIRI LWP is much lower than the corresponding TMI 395 mean LWP in regions with generally lower cloud fractions.This is mostly due to a high bias in TMI LWP in broken scenes (Greenwald et. al., 2018).In the Sc region, SEVIRI COT varies from 6 to 11 and CER ranges between 8 and 14 µm.The two-year-mean liquid cloud fraction varies between 75 % and 100 %.The mean statistics also show robust skill in LWP retrieval for both SEVIRI and TMI with a high correlation of 0.89 for the Sc regime.Both TMI and SEVIRI show a mean LWP of ~53 g m -2 with negligible bias and standard deviation of 24 g m -2 for the study 400 period.
In the overcast case over the Sc regime, the two-year mean LWP increases to 84 g m -2 and 80 g m -2 respectively for SEVIRI and TMI, i.e., the mean SEVIRI LWP is about 5 % larger than the mean TMI LWP.In this case, applying an adiabatic correction to SEVIRI LWP would lead to a larger bias of -10 g m -2 (-12 %) and standard deviation of 28 g m -2 .The unbiased LWP observed in the all-sky Sc case could be associated with the cancellation 405 of errors between fully overcast and lower LCF grid-boxes within the domain.A higher mean COT of ~11 characterizes the overcast Sc case, whereas the mean COT is only about 7 in the all-sky case, suggesting the presence of optically thin clouds which are more prone to retrieval biases.Figure 4 shows a density scatterplot of TMI and SEVIRI LWPs in the overcast Sc region.Most data points are close to the one-to-one line, although at the lower end TMI LWP is slightly higher, while the reverse is true at the higher end -the same feature is also found in monthly and seasonal results.
The daytime-averaged two-year and seasonal statistics of SEVIRI and TMI LWP are listed in Table 1.
Seasonally, in the overcast Sc domain, the average LWP varies from 73 to 92 g m -2 in standard SEVIRI, 61 to 76 g m -2 in adiabatic SEVIRI, and 73 to 82 g m -2 in TMI retrievals.In the aerosol-free seasons of DJF and MAM, standard SEVIRI overestimates TMI LWP; applying an adiabatic correction to SEVIRI in these months brings the 415 LWP bias within 5 %, similar to previous studies.The standard SEVIRI likely overestimates the actual LWP in the overcast Sc regime due to the overestimation of CER, as the observed CER in the 1.6-µm channel corresponds to the top layer and is higher than the cloud layer-mean in sub-adiabatic stratocumulus.However, for JJA, when all three months were heavily affected by smoke aerosol, the standard SEVIRI already shows ~10 % lower LWP than TMI; therefore, applying the adiabatic correction would only enhance this negative bias.For SON, only September was 420 heavily affected by aerosol for the analysis years we considered.As a result, the mean standard SEVIRI LWP was ~5 % larger than TMI LWP and applying adiabatic correction would lead to a ~14 % underestimation in SEVIRI LWP.We found that SEVIRI underestimates LWP more during the aerosol-affected months, even after excluding grid-boxes with AI > 0.1 Applying a stricter criteria by excluding grid-boxes with AI > 0 did not improve the results, hinting at OMI AI retrieval biases.

425
The spatial distribution of SEVIRI and MODIS cloud properties averaged for the study period for the overcast condition is shown in Fig. S5.In general, over the overcast Sc regime, MODIS retrieves higher LWPs in all three channels compared to SEVIRI, but for more broken scenes MODIS values are lower than SEVIRI LWP.The daytime-averaged two-year and seasonal statistics of SEVIRI and MODIS LWP re listed in Table 2, whereas the respective mean COTs and CERs are listed in Table S1.For overcast marine Sc clouds the two-year 445 mean LWP is 80 g m -2 for SEVIRI, 84 g m -2 for MODIS 1.6-µm, 88 g m -2 for MODIS 2.1-µm, and 87 g m -2 for MODIS 3.7-µm channels.The differences in retrieved LWP values vary from 4 to 8 g m -2 (5-10 %), whereas the differences in root mean square deviation (RMSD) values vary between 16-20 g m -2 .The SEVIRI and MODIS LWP retrievals are highly correlated, with correlations > 0.9.In the aerosol-unaffected seasons of DJF and MAM, the difference between SEVIRI and MODIS LWPs is within 0-5 %.In the heavily polluted months of JJA, LWP 450 retrievals from SEVIRI are about 10 % lower than those from the MODIS 1.6-µm and 20 % lower than those from the MODIS 3.7-µm band.This suggests that SEVIRI retrievals are more strongly affected by the presence of absorbing aerosols in the Sc regime than the corresponding MODIS 1.6-µm retrievals and that these polluted scenes are not sufficiently filtered out by the OMI AI threshold.Indeed, unlike in other seasons, in JJA MODIS LWP 1.6-µm < MODIS LWP 2.1-µm < MODIS LWP 3.7-µm , hinting at the influence of absorbing aerosols on MODIS LWP retrievals as 455 the 3.7-µm channel is known to be the least affected by smoke.In SON, since September is the only month strongly affected by aerosols, the comparison of SEVIRI and MODIS LWPs is better, with SEVIRI low biases of 6-12 %. at -1 in the overcast case; however, in the all-sky case there is a broader peak between -1 and 0. Histograms of CER differences reveal wider distributions, especially when compared against the 2.1-and 3.7-µm channels, which peak 470 at -1 µm in the overcast case; however, in the all-sky case the peak is again broader between -2 and -1 µm.For the two-year means (Fig. 6), both TMI and SEVIRI indicate a maximum LWP at 06h LST in the morning before sunrise, followed by a decrease until about 16h LST and an increase afterwards.During the night LWP continues to increase until sunrise, as indicated by the TMI night retrievals.At around 06h LST the two-yearmean all-sky LWP values are ~75 g m -2 for both TMI and SEVIRI, but they decrease to ~40 g m -2 by ~14h LST.

485
This decrease in LWP is linked to a sharp decline in COT from 11.5 to 5.5.The relative variation in CER is much smaller over most of this time period, in agreement with Zuidema and Hartmann (1995) in our data.This is in agreement with Fairall et al. (1990) and Ciesielski et al. (2001), who observed that fractional cloudiness is maximum in the predawn hours and minimum in the mid-afternoon, which is accompanied by an opposite trend in the MBL moisture with a predawn drying and an afternoon moistening.
The seasonal mean diurnal cycles of Sc clouds are qualitatively similar to the two-year mean, except for the aerosol affected months of JJA (Figs. 7 and S7-S9).The maximum LWP tends to occur between 06h and 10h LST.

525
The largest diurnal variation is seen during SON, which is also the season with the greatest cloud cover.We found In the all-sky case, the diurnal variation of TMI and SEVIRI LWP is in good absolute agreement within ±5 g m -2 , for all seasons and the two-year mean, except JJA.In JJA, however, a ±10 g m -2 or even slightly larger mean difference is found between the techniques, despite the exclusion of aerosol-affected pixels with AI > 0.1.MODIS Terra and Aqua mean LWPs also show excellent agreement with the corresponding SEVIRI LWPs within ±5 g m -2 , 535 for all seasons and the two-year mean.
In the overcast case, SEVIRI LWPs are 10-20 g m -2 larger than TMI LWPs especially for the aerosol-free seasons of DJF and MAM.After applying the adiabatic correction, the biases become negligible between the datasets.For the aerosol-affected seasons of JJA and SON, the mean SEVIRI LWPs likely underestimate the actual values and hence applying adiabatic correction (reduction) worsens the comparison with TMI LWPs.In the overcast 540 case, MODIS Terra and Aqua LWPs deviate by 5-10 g m -2 from SEVIRI LWPs for the aerosol-free seasons, but by a larger amount of 5-20 g m -2 for the aerosol-affected seasons due to smoke-induced biases being larger in SEVIRI than MODIS retrievals (see section 4.2).
Seasonally COT varies between 4 and 16, typically showing a relative decrease of 40-50 % from early morning to late afternoon.Not surprisingly, the diurnal amplitude of COT is similar to that of LWP.Although the 545 absolute value of CER varies from 7 to 12 µm between different seasons, the relative diurnal variation is negligible.

585
In the all-sky case, the diurnal variations of TMI and SEVIRI LWP were in good absolute agreement, being within ±5 g m -2 for all seasons and the two-year mean, except JJA.In JJA, the season most affected by biomass smoke, a larger mean difference was found between the techniques, although we eliminated aerosol-affected pixels with AI > 0.1.MODIS Terra and Aqua mean LWPs also showed excellent agreement with corresponding SEVIRI LWPs in the all-sky case, differences being within ±5 g m -2 for all seasons and two-year means.

590
In the overcast case, SEVIRI LWPs were 10-20 g m -2 larger than TMI LWPs especially in the smoke-free seasons of DJF and MAM.After applying an adiabatic correction to SEVIRI retrievals, however, the biases between the datasets became negligible.In the smoke-affected seasons of JJA and SON, the mean SEVIRI LWPs already underestimated the TMI values due to smoke-induced retrieval biases and hence applying the adiabatic correction (i.e.further reduction) worsened the comparison with TMI LWPs.In the overcast case, MODIS Terra and Aqua 595 LWPs differed by 5-10 g m -2 from SEVIRI LWPs in smoke-free seasons and by a larger amount of 5-20 g m -2 in smoke-affected seasons, due to the different magnitudes of smoke-induced biases in SEVIRI and MODIS retrievals.
Irrespective of season, both TMI and SEVIRI LWP decreased from morning to mid-afternoon, and after that a slight increase was observed.Prior to sunrise clouds are the thickest and as the day progresses the cloud layer thins due to the absorption of solar radiation and associated decoupling of the sub-cloud layer.We found that the 600 relative amplitude of the LWP diurnal cycle is typically 30-50%, which is close to but slightly larger than the diurnal amplitude reported in most previous studies.The temporal variation in SEVIRI LWP was mainly due to that in cloud optical thickness, while droplet effective radius showed relatively small diurnal variability.MODIS Terra      Rain-, ice-, and smoke-free conditions were applied.
confident liquid clouds with valid LWP retrieval, and cloud top temperature (CTT) > 275 K in SEVIRI and MODIS retrievals with ice fraction ≤ 0, and, rain rate ≤ 0 in TMI retrievals.Our study domain is also affected by continental biomass burning during austral winter and spring, which in turn affects VIS/NIR cloud retrievals, therefore, special attention is paid to the analysis of retrieval artifacts related to the presence of smoke over the Sc deck.270Wenoticed that the extent and location of South Atlantic Sc clouds vary from month to month; hence we opted to define the Sc domain dynamically, rather than selecting a fixed rectangular area to specify the study domain.Thresholding the spatial mean map of liquid cloud fraction (LCF) and the heterogeneity parameter (Hσ = reflectance standard deviation / mean reflectance) was found to delineate Sc regions in good agreement with visual observations.To precisely define the Sc domain, we used a region-growing algorithm to find adjacent, connected 275 grid-boxes with LCF > 80 %.The identified Sc regions were typically within 20 o W -20 o E and 5 o -35 o S. Cloud properties were separately evaluated for two cases: dynamical and climatological impacts of the presence of smoke above Sc clouds from both modeling as well as 295 Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-445Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 17 May 2018 c Author(s) 2018.CC BY 4.0 License.

Figure
Figure 1a depicts the spatial distribution of average OMI aerosol index during JAS for 2011 and 2012, with the black contour representing the Sc region.It is clear that absorption by smoke is highest near the Namibian coast Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-445Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 17 May 2018 c Author(s) 2018.CC BY 4.0 License.bias of 6-25 g m -2 for moderate AI between 1 and 2, and a large negative bias > 40 g m -2 for grid-boxes with AI > 2.5; the bias increases linearly with AI.
340preserve humidity and cloud cover in the MBL, resulting in increased cloud amount and LWP compared to a smokefree environment.Similar to our SEVIRI results,Bennartz and Harshvardhan (2007),Wilcox et al. (2009), andSeethala and Horváth (2010) also noted a systematic MODIS LWP underestimation in Sc off southern Africa during the biomass burning seasons.Painemal et al. (2014) also noted a CER decrease in MODIS data despite increased LWP north of 5 o S during the biomass burning season.345 Retrieval discrepancies due to the presence of absorbing aerosols above Sc clouds were also evaluated between SEVIRI and MODIS.Frequency histograms of SEVIRI minus MODIS LWP, COT, and CER biases, as well as, the biases relative to MODIS CPP for overcast conditions aggregated for JAS 2011 and JAS 2012 are shown in supplemental Fig. S1.SEVIRI COT appeared to be biased low by ~1 compared to MODIS.Compared to the 1.6-µm MODIS CERs, ~70 % of SEVIRI CERs have a mean bias of -1.5 µm.Although SEVIRI CERs are 350 biased low compared to all three MODIS CERs, the ~1 µm additional high bias relative to the 2.1-and 3.7-µm CERs likely indicates much smaller smoke-induced retrieval artifacts in these two channels.In general, the CER Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-445Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 17 May 2018 c Author(s) 2018.CC BY 4.0 License.retrievals from SEVIRI tend to be lower than corresponding retrievals from the three MODIS channels, with SEVIRI having about 1.5 µm to 2.5 µm lower CER values.The SEVIRI minus MODIS LWP distributions peak at about -10 g m -2 irrespective of the MODIS channel used for the retrieval. 355

4. 2
Spatial distribution and mean statistics of SEVIRI, MODIS, and TMI cloud properties This section presents the results of the comparison of SEVIRI, MODIS, and TMI LWP retrievals, as well as the comparison of SEVIRI and MODIS COT and CER retrievals.Significant variation in the distribution and amount of clouds is observed over the Sc region from month to month.During SON, we observe frequent Sc clouds with large 385 spatial extent.During JJA there are relatively fewer clouds that are shifted slightly to the north.The lowest cloud fractions are seen during DJF and MAM.From a surface-based cloud climatology, Klein and Hartmann (1993) also showed that there is strong seasonal variability in the amount of Sc clouds, which is closely tied to the seasonal cycle of static stability.Over the South Atlantic Sc region, SON had the largest lower tropospheric stability (LTS), and DJF had the smallest.The strongest net cloud radiative effect also occurred during August through November, 390 which further motivates us to examine the seasonal variability of these clouds.
435 compared to MODIS, as SEVIRI COTs remain underestimated in these clouds similar to overcast clouds.This overestimation could be caused by, in thin clouds with COT < 4, the SEVIRI CPP algorithm weighing CER with an a priori (climatological) value of 8 µm, but MODIS providing smaller retrieved values.Also note that the SEVIRI CER overestimation in broken clouds systematically increases as the sampling height of the comparison MODIS Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-445Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 17 May 2018 c Author(s) 2018.CC BY 4.0 License.channel gets closer to cloud top, i.e. moving from the 1.6 µm to the 3.7 µm band.This could indicate a vertically 440 decreasing CER profile, sometimes seen in raining/drizzling small Cu clouds, in which case the cloud-top CER value, especially from the 3.7-µm band, would underestimate the cloud layer-mean, leading to a corresponding underestimation in LWP as well.

Figure 5
Figure5shows the density scatterplots of SEVIRI versus MODIS LWPs, COTs, and CERs in the overcast Sc region for the study period.Most data points are close to the one-to-one line, but with a SEVIRI low bias; the same feature is also found in monthly and seasonal results.A low COT bias of 1 compared to all three MODIS

Figures 6 -
Figures 6-7 and S7-S9 show the two-year mean and seasonal diurnal cycle of Sc cloud properties.The diurnal cycles shown here are limited to cases with AI values lower than 0.1, in order to minimize VIS/NIR retrieval biases due to . CER increased by 2 µm in the early hours between 06h and 10h LST, stayed around 11.0-11.5 µm most of the day, and decreased by ~1 µm in the late afternoon by 18h LST.As a result, the diurnal cycle of LWP was mainly driven by COT.Note that the allsky two-year-mean TMI (red solid line circles), SEVIRI (black solid line circles), and MODIS (colored circles) 490 LWPs exhibit excellent agreement not only in their relative diurnal variations but also in their absolute values -the curves almost completely overlap.For the overcast case, a ~30 % increase in COT and a slight <1 µm decrease in CER lead to an overall increase of 25-30 g m -2 (~40 %) in mean LWP compared to the all-sky case.Apart from that, the diurnal cycles of LWP, COT, and CER are very similar between the overcast and all-sky cases.The standard SEVIRI (black dash-dot 495 line, plus signs) and TMI day (red dash-dot line, plus signs) overcast LWPs also show very good quantitative agreement, with SEVIRI being biased high only about 5 g m -2 .Note that for the two-year means, adiabatic SEVIRI Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-445Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 17 May 2018 c Author(s) 2018.CC BY 4.0 License.LWPs (green dashed line, triangles) had larger and negative biases than standard SEVIRI retrievals.As shown later, this was the consequence of the significant smoke-induced negative biases in SEVIRI retrievals in the aerosolaffected seasons of JJA and SON.In the smoke-free seasons of DJF and MAM, adiabatic SEVIRI LWPs were in 500 better agreement with TMI microwave LWPs than were standard SEVIRI LWPs, echoing the findings of Bennartz (2007) and Seethala and Horváth (2010) for MODIS -AMSRE-E LWP comparison.Comparing MODIS Terra (10h LST) and Aqua (14h LST) LWPs, a similar decreasing diurnal trend can be observed, except that MODIS LWPs are 5-10 g m -2 larger than SEVIRI LWPs for the overcast case, probably due to the difference in pixel size (1 km vs. 3 km).The CM SAF (2016) validation report also suggests that the coarser 505 resolution of SEVIRI retrievals results in somewhat lower COT and LWP values compared to MODIS, due to nonlinear averaging effects (plane-parallel albedo bias).Our results are consistent withWood et al. (2002) andPainemal et al. (2012), who studied the diurnal variation of LWP over the southeast Atlantic and southeast Pacific Sc, based on microwave and near-infrared satellite data.Similar to our results,Painemal et al. (2012) also noted that COT rather than CER explains most of the 510 LWP variation.Blaskovic et al. (1990) associated the daytime decrease of LWP with the decrease of cloud thickness observed in their ground-based measurements, as the cloud base height increased from sunrise till mid-afternoon, while cloud top height decreased in the late afternoon.Duynkerke et al. (2004) found that the diurnal variation of Sc LWP is related to the transition from a decoupled MBL during daytime to a vertically well-mixed MBL during the night.The observed diurnal cycle of Sc is characterized by a cloud layer that gradually thickens during the night but 515 gets thinner during the day due to absorption of shortwave radiation and decoupling.The latter state exhibits slightly negative buoyancy fluxes and a minimum vertical velocity variance near cloud base.This implies that surfacedriven, moist thermals cannot penetrate the cloud layer, while entrainment maintains a steady supply of relatively warm and dry air from just above the inversion into the cloud layer, resulting in a distinct LWP diurnal cycle with minimum values during the day.The diurnal cycle of LWP also consistently follows the variation of cloud fraction 520 Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-445Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 17 May 2018 c Author(s) 2018.CC BY 4.0 License.that the relative amplitude of the two-year-and seasonal-mean LWP diurnal cycle is typically 35-40 %.Wood et al. (2002) reported diurnal amplitudes of 15-35 % in MBL clouds using TMI data and Zuidema and Hartmann (1995) obtained a 25 % variation in LWP over the North/South Pacific as well as South Atlantic stratus clouds using SSM/I data for the summer months.However, Fairall et al. (1990) found larger amplitudes of 60-70 % for Californian Sc 530 clouds using a 17-day period of near-continuous ground-based microwave radiometer data.
The diurnal reduction in COT is likely due to the reduction in cloud fraction and cloud physical thickness, while the variation in cloud-top CER is probably indicative of enhanced cloud-top entrainment of dry air and associated droplet evaporation.Although MODIS COTs are slightly higher in both Aqua and Terra data, the difference with SEVIRI is only about 1, while MODIS CER values are within 2 µm of the SEVIRI CER.5505.SummaryThe objective of this work was to compare LWP, COT, and CER retrievals from SEVIRI, MODIS, and TMI, in order to quantify the effect of biomass burning smoke on passive VIS/NIR imager retrievals as well as to evaluate the diurnal cycle of South Atlantic maritime stratocumulus clouds.In general, SEVIRI and TMI showed good Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-445Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 17 May 2018 c Author(s) 2018.CC BY 4.0 License.absorbing VIS channel used: the 0.8-µm channel used for MODIS retrievals is less affected by aerosol absorption than the 0.6-µm channel used for SEVIRI.

(
morning) and Aqua (afternoon) LWPs indicated a similar diurnal trend, but MODIS LWPs were 5-10 g m -2 larger than SEVIRI/TMI values in the overcast case.This maybe partly due to the plane-parallel albedo bias affecting the 605 larger SEVIRI pixels.While the discrepancies between microwave and VIS/NIR LWP retrievals in areas of broken clouds with low cloud fraction require further research to be fully resolved, our study has shown that there is a reasonable consensus between the techniques about the seasonal and diurnal cycles of LWP in nearly overcast stratocumulus fields.This lends some credibility to the VIS/NIR retrievals of the underlying cloud microphysical properties.In our 610 opinion, SEVIRI-derived CLAAS-2 cloud property observations provide a useful resource for the evaluation of stratocumulus cloud diurnal cycles in climate models.Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-445Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 17 May 2018 c Author(s) 2018.CC BY 4.0 License.Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-445Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 17 May 2018 c Author(s) 2018.CC BY 4.0 License.

Figure 4 .
Figure 4. Scatter density plot of SEVIRI versus TMI liquid water path for the overcast case (LCF ≥ 95% and COT > 3) in two

Figure 5 .
Figure5.Scatter density plot of SEVIRI versus MODIS liquid water path, cloud optical thickness, effective radius in the overcast case (LCF ≥ 95% and COT > 3) in two years of data.Rain-, ice-, and smoke-free conditions were applied.

Table 2 .
Two-year mean and seasonal statistics of collocated SEVIRI and MODIS retrievals in rain-free, ice-free, smoke-free (AI<0.1),COT>3, and overcast (LCF ≥95%) grid cells over the marine stratocumulus region.LWP means, biases (MODIS-SEVIRI), and Root Mean Square Differences (RMSD) are given in g m −2 .Corresponding COT means and CER means are 910 tabulated in TableS1.The values in brackets are statistics without filtering for LCF≥ 95% and COT >3, i.e., for the all-sky case.