To estimate the global co-variability between mineral dust aerosol and cloud glaciation, we combined an aerosol model reanalysis with satellite retrievals of cloud thermodynamic phase. We used the CALIPSO-GOCCP product from the A-Train satellite constellation to assess whether clouds are composed of liquid or ice and the MACC reanalysis to estimate the dust mixing ratio in the atmosphere. Night-time retrievals within a temperature range from
Aerosol–cloud interactions affect the Earth’s climate through different mechanisms. These include impacts of aerosol particles on cloud glaciation that subsequently influence the clouds’ thermodynamic phase, albedo, lifetime, and precipitation. Specifically, there is growing evidence for a role of mineral dust aerosol (or of ice-nucleating particles correlated to dust aerosol) in influencing heterogeneous cloud ice formation on a global scale
The dust occurrence frequency retrieved from spaceborne instruments like the
Cloud-Aerosol LIdar with Orthogonal Polarization
Ice particles and cloud droplets may coexist in a so-called mixed-phase state
In this study, we use a global aerosol reanalysis together with the cloud thermodynamic phase retrievals of the CALIPSO-GOCCP
In Sect. 2, the datasets used for the study are presented. In Sect. 3, the processing of the datasets are described. In Sect. 4, the main findings are presented, including a case study, the distribution of cloud phase along temperature and latitude, and finally the day-to-day correlation between dust and cloud ice. In Sect. 5, the main overlaps and differences with respect to previous findings are discussed and put into context with the conceptual limitations of the approach.
This section presents an overview of the datasets used in this study. The cloud thermodynamic phase is obtained from the CALIPSO-GOCCP product, the aerosol information from the MACC reanalysis, and the large-scale meteorological conditions from the ERA-Interim reanalysis.
The CALIPSO-GOCCP v.3.0 product
The Monitoring Atmospheric Composition and Climate reanalysis
The averaged meteorological parameters (RH, large-scale updraught, and isotherm height) used in Sect. 5 were weighted by the cloud volume fraction retrieved by the CALIPSO-GOCCP product (see Sect.
The dust emission in the MACC model is parameterized as a function of the 10 m wind, vegetation, soil moisture, and surface albedo. The dust loadings are corrected by the assimilation of the total column AOT at 550 nm retrieved from the MODIS instrument on board NASA’s Aqua and Terra satellites. Dry and wet deposition of dust are simulated, as well as in-cloud and below-cloud removal. The freezing efficiency of INPs depends mainly on their surface area concentration
In the MACC reanalysis, dust aerosols are represented by three size bins, with size limits of 0.03, 0.55, 0.9, and 20
Flow chart showing the processing steps starting from the raw data (satellite retrievals and model reanalysis) to the dataset used for the analysis.
In this section, the different processing steps of the datasets presented in Sect. 2 are described. Figure
In order to exclude the effects of the scattering of sunlight on the cloud-phase detection from the CALIOP lidar signal, only night-time retrievals were used. Including convective clouds – as retrieved by the 2B-CLDCLASS product (see Appendix) – does not introduce a significant bias on the results. This low sensitivity to convective clouds is mainly due to the low area fraction represented by such clouds, especially in the mixed-phase regime at the mid-latitudes (less than 5 %). Similarly, precipitating clouds had little impact on the results.
The cloud thermodynamic phase is mainly a function of temperature. Therefore, temperature bins of 3 K each were used as a vertical coordinate throughout the study to constrain the variability of cloud phase. For the MACC and ERA-Interim reanalyses, we rebin the model levels into 3 K intervals to match the vertical resolution of the CALIPSO-GOCCP product.
For each product, the latitude
Dust aerosol can produce or be accompanied by changes in atmospheric stability and humidity. To disentangle such effects, we constrain the cloud environment using the air relative humidity with respect to liquid and the tropospheric static stability. Depending on the isotherm to be studied, we use the lower troposphere static stability (LTSS) or the upper troposphere static stability (UTSS). These parameters are defined as
In contrast to previous studies, in this work we want to isolate the day-to-day correlation between dust aerosol and cloud phase. In order to exclude the spatial component of the correlation, the complete time span 2007–2010 was used to assess the daily correlation between the MACC dust mixing ratio and the CALIPSO-GOCCP cloud phase. This correlation was done independently for each volume grid box – each constrained in latitude, longitude, and temperature.
Seasonal, day-to-day, and day-to-day decile concept as used in this study. For this example, the day-to-day analysis of May contains 124 daily datapoints. In step
We also need to exclude the seasonal component of the temporal correlation. For this purpose, we process each month of the year independently. This is done as a multiyear selection (e.g. January containing January 2007, January 2008, January 2009, and January 2010) (see Fig.
The dust mixing ratio density distribution is heavily skewed to the right, while the cloud phase follows mostly a binary distribution. Because of this non-normality, a typical correlation approach like the Pearson's correlation coefficient will not reflect the genuine relationship between both variables
The resulting field contains one extra dimension for each volume grid box (month, dust decile, temperature, latitude, longitude). Figure 2 presents a visualization of this process.
The day-to-day correlation approach relies strongly on the available sample size. For small sample sizes, only a few retrievals (daily means within a volume grid box) can be found for a given dust decile. In this case, the average FPR may still be non-normally distributed, introducing a larger standard deviation. Within a 12 K range, each zonally averaged latitude bin ( The satellite swaths (orbits) produce a different density of retrieved profiles at different latitudes. Using only night-time data, the sample size in the meteorological summertime (shorter nights) is lower. The cloud-phase retrievals are less frequent for seasons, regions, and heights with low cloud cover (see Fig. S8). At high latitudes, relatively warm temperatures (e.g.
The averaging order of the dimensions was defined – from first to last – as longitude, month, decile, latitude, and temperature. This choice prevents artefacts resulting from too many missing values. Latitude and temperature are averaged last because of the higher associated correlations with cloud phase
In Sect. 4.1, the adjusted ice volume fraction,
Sample size of cloud phase (CALIPSO-GOCCP) of each latitude band for
This section seeks a better understanding of the ice-to-liquid ratio retrieved in the CALIPSO-GOCCP product. We provide a detailed case study of a stratiform cloud scenario. In this scenario, four stratiform cloud types from the CloudSat classification are included: stratocumulus (low-level clouds), altostratus and altocumulus (mid-level clouds), and cirrus (high-level clouds). Although not present in the case study, nimbostratus are included in the analysis of cloud phase as well and are particularly important in the high latitudes. Stratus clouds are defined for temperatures above 0
Case study 09:50 UTC 14 December 2010 for temperatures between
The A-Train segment shown in Fig.
Temperature is the main factor controlling the thermodynamic phase of clouds. Mixed-phase clouds between 0 and
Global ice cloud occurrence frequency (2007–2010). The fine-mode dust mixing ratio from the MACC reanalysis corresponds to the range 0.03–0.55
Figure
Additionally, the average fine-mode dust mixing ratio is also shown in Fig. 5. At the height of the 0
For both temperature ranges shown in Fig. 6 the absolute maximum of FPR is located near the Equator (85 % at
At
Zonal mean of stratiform cloud ice occurrence frequency for
Probability histogram at
For the clouds studied, the time-averaged large-scale vertical velocity (from the MACC reanalysis, shown in Fig. The spatial correlation can be a result of an enhanced transport of water vapour to higher levels at temperatures below The updraughts are associated with higher availability of INPs at the cloud level (from below the cloud), and the effect is large enough to mask the enhanced droplet growth typically associated with updraughts. The updraughts enhance a certain type of heterogeneous nucleation requiring saturation over liquid water (e.g. immersion freezing). Updraughts generate a local adiabatic cooling, possibly activating INPs that may not have been active before at higher temperatures.
To the authors' knowledge, there is currently no observational constraint to the source of cloud ice in the mixed-phase regime. Namely, the frequency of ice clouds between 0 and
In the following sections, the temporal correlation between mineral dust mixing ratio and cloud ice occurrence frequency is referred to as the dust–cloud-phase relationship. To study this relationship, we classify the retrievals into different weather regimes to constrain the meteorological influence. The resulting dust–cloud-phase relationship for different regimes may offer a good insight into the processes underlying the dust–cloud-phase relationship. Particularly, how heterogeneous freezing by dust aerosol may affect the cloud thermodynamic phase on a day-to-day timescale.
In other words, to extract the specific influence of mineral dust on cloud glaciation, it is necessary to identify and constrain relevant meteorological confounding factors
The effect of humidity and static stability on ice production is not straightforward. In general, moist and unstable conditions are associated with enhanced lifting of air that likely causes nucleation of hydrometeors. Between 0 and
For the relative humidity, the bounds are defined at 60, 70, and 80 %; for the LTSS, they are defined at 10, 15, and 20 K; and for the UTSS, they are at 4, 6, and 8 K. The fraction of data inside each regime corresponds to the integral of the probability density within the regime bounds. For example, if the probability density between 4–6 K and 70 %–80 % is 0.01, then 20 % of the data is contained between these bounds. The magenta boxes in Fig.
Average cloud phase (GOCCP) for the mid-latitude and high-latitude bands averaged between
For dust mixing ratios between 0.1 and 2.0
At high LTSS in the high latitudes, the range of ice occurrence frequency values is higher than for the mid-latitudes and small increases in dust mixing ratio are associated with a strong increase in cloud ice occurrence frequency. For the high-LTSS regime, the ice occurrence frequency in the southern high latitudes increases by
Same as Fig.
At
Same as Fig.
At
At all temperatures studied, higher humidity values were associated with a higher cloud ice occurrence frequency. Additionally, for similar dust loadings, the cloud ice occurrence frequency was found to be higher at the mid-latitudes than at the high latitudes. However, against our expectations, for similar dust loadings the cloud ice occurrence frequency at
Some studies have already suggested that the lower occurrence frequency of cloud ice in the higher latitudes may be associated with lower INP concentrations
We have found that the ice occurrence frequency can vary at different latitudes even for similar mixing ratios of mineral dust. This variability could be explained by differences in the mineralogical composition of the mineral dust aerosol at the Southern Hemisphere and Northern Hemisphere. Clay minerals from the Northern Hemisphere are composed mostly of illite and smectite
For temperatures higher than
Nevertheless, at such high temperatures, other dust minerals like feldspar mineral are much more efficient as ice-nucleating particles than clay minerals
If feldspar minerals do dominate the heterogeneous freezing due to mineral dust above
Summary of the north–south differences in the cloud phase associated with mineral dust based on the day-to-day statistics for the middle and high latitudes.
Furthermore, such a depletion of highly efficient INPs during the transport of dust aerosol may also explain the higher ice occurrence frequency at the mid-latitudes compared to the high latitudes for similar mixing ratios of mineral dust, especially at higher temperatures. The ageing (e.g. internal mixing with sulfate or “coating”) of dust particles may also reduce the freezing efficiency of dust aerosol during the transport from low to high latitudes. The hypotheses explaining the differences in the freezing behaviour of dust between the Northern Hemisphere and Southern Hemisphere are summarized in Table .
In the analysis presented above, certain assumptions were made to assess the potential effect of mineral dust on cloud thermodynamic phase. In this section, these assumptions and the uncertainties that arise from them, as well as the subsequent limitations of the resulting interpretation, will be discussed.
Same as Fig.
Concerning the vertical resolutions of the different products, the choice of 3 K bins is based on the original 3 K bins of the CALIPSO-GOCCP product. Using a coarser vertical resolution (e.g. 6 K bins) would hinder the assessment of the role of dust as INPs. For example, a decrease of 3 K in temperature is roughly equivalent to a 5-fold increase in INP concentrations
As mentioned in Sect. 3, we excluded the seasonal component of the dust–cloud-phase correlation by calculating the deciles independently for each month of the year. However, shorter cycles (e.g. weather variability) may still influence the variability of dust and cloud phase. For example, below the
Despite the long period (2007–2010) used in the study, a significant fraction of the five-dimensional space used for our analysis (10 dust deciles, 12 months, 15 temperature bins, 96 latitudes, and 12 longitudes) is sparsely sampled or even contains missing values. In the high latitudes, a sampling bias exists towards the respective winter seasons, because very few night-time retrievals are available in summer. However, the seasonal variability was not found to be a dominating factor in the day-to-day impact of dust mixing ratio on the FPR (see Fig. S19). Furthermore, many factors may contribute to higher standard deviations for the ice occurrence, including
changes in dynamical forcing (e.g. updraughts) and cloud regimes; temperature changes after cloud glaciation (e.g. latent heat release); ice sedimentation from above (cloud seeding) and INPs other than dust; cloud vertical distribution within the studied temperature ranges; turbulence favouring aerosol mixing and sub-grid temperature fluctuations; differences in dust mineral composition, electric charge, or size; coatings (e.g. sulfate) affecting aerosol solubility and freezing efficiency; subsetting of the data (e.g. only night-time retrievals).
Additionally, some issues arise from the coarse spatial resolution used in our study. A high dust mixing ratio simulated in a volume grid box indicated as cloudy by the satellite observations does not ensure that the dust is actually mixed with the cloud. The sub-grid distribution of dust relative to the exact cloud position remains unresolved. Higher dust mixing ratios should be interpreted as an indicator or a higher probability that a significant amount of dust was mixed with a co-located cloud. This mixing may have happened during or before the observation by the satellite. However, we can assume that both cloud and dust aerosol followed a similar trajectory up to the moment of the observation. Overall, at coarse resolutions, the combination of modelled dust concentrations with satellite-retrieved cloud properties cannot guarantee the mixture of aerosol and clouds
As mentioned in Sect. 3, the total aerosol optical depth (AOD) from MODIS is assimilated in the MACC reanalysis. In general, we expect this assimilation to produce a fair estimation of the large-scale aerosol conditions on a day-to-day basis. At least for the Northern Hemisphere, this has been already validated with in situ measurements
The CALIPSO-GOCCP product relies on CALIOP to determine the presence of clouds. Nevertheless, the reader should be aware that several uncertainties remain. For example, the meteorology in the reanalysis and in the real atmosphere may differ, particularly on the sub-grid scale. In the worst case that the reanalyses are entirely inconsistent with the retrievals of cloud phase, we expect that the result would be the lack of correlation between dust and the ice occurrence (Figs. 8–10). We have included a reasonably large dataset for the study. Certainly, mismatches between reanalysis and cloud retrievals are possible. However, these would cause an underestimation – and not an overestimation – of the dust–cloud-phase correlation.
Concerning the interpretation of our results, it cannot be ruled out that the increase in ice cloud occurrence in the Southern Hemisphere for higher dust loading arises from other types of INPs such as biogenic aerosol
In general, meteorological parameters have a larger impact on cloud properties than aerosols do
Second, the significant positive correlation found between dust aerosol mixing ratio and the height of the isotherms (weighted by cloud volume fraction) points to a possible source of uncertainty (Fig. 11b). This correlation could be due to clouds being detected in a higher temperature bin after being glaciated at lower temperatures. Thus erroneously suggesting an enhanced glaciation occurrence frequency at higher temperatures. Therefore, future studies must take into account this possibility when studying the occurrence of ice clouds at a certain isotherm. More details on the spatio-temporal variability of the cloud height can be found in the Supplement (Fig. S12). Lastly, Fig. 11c shows a positive correlation between the fine-mode dust and the large-scale vertical velocity from the MACC reanalysis at
In summary, much of the co-variability between dust, humidity, updraughts, temperature, and cloud ice occurrence frequency is still poorly understood. However, we expect that the constrains on humidity and static stability minimized most of the biases discussed in this section.
For the first time, an aerosol reanalysis was combined with satellite retrievals of cloud thermodynamic phase to investigate the potential effect of mineral dust as INPs on cloud glaciation. We studied this effect on a day-to-day basis at a global scale for the period 2007–2010 focusing on stratiform clouds observed at night-time in the middle and high latitudes. Our main findings can be summarized as follows:
Between The response of cloud ice occurrence frequency to variations in the fine-mode dust mixing ratio was similar between the middle and high latitudes and between the Southern Hemisphere and Northern Hemisphere.
Even though dust aerosol is believed to play a minor role in cloud glaciation in the Antarctic region, increases in FPR from first to last dust decile were also present in both the northern and southern high latitudes. Using constraints on atmospheric humidity and static stability we could partly remove the confounding effects due to meteorological changes associated with dust aerosol. The results also suggest the existence of different sensitivities to mineral dust for different latitude bands. The north–south differences in ice occurrence frequency for similar mineral dust mixing ratios agree with previous studies on the mineralogical differences between the Southern Hemisphere and Northern Hemisphere. A larger fraction of feldspar in the Southern Hemisphere could explain the differences at
We believe these new findings may have an important influence on improving the understanding of heterogeneous freezing and the indirect radiative impact of aerosol–cloud interactions. The authors hope that the results of this work will also motivate further research, including field campaigns in remote regions, to study the day-to-day variability of cloud thermodynamic phase and the role of mineral dust in ice formation, satellite-based studies of associated changes in the radiative fluxes, and modelling studies to test the representation and relevance of specific processes involved in ice formation and mineral dust transport. Such studies could help to further improve our understanding of the influence of mineral dust or other aerosol types on cloud glaciation and the climate system.
Although in our study we used the cloud-phase classification from the CALIPSO-GOCCP product, other products are also available. Therefore, we include in the following appendix a detailed comparison between the CALIPSO-GOCCP and the DARDAR-MASK product, which is commonly used in the literature as well.
The CloudSat cloud scenario classification (2B-CLDCLASS) was used in Sect.
The DARDAR-MASK v1.1.4 product available at the ICARE Data and Services Center combines the attenuated backscatter from CALIOP (at 532 nm; sensible to small droplets), the reflectivity from the CPR (at 94 GHz; sensible to larger particles), and the temperature from the ECMWF-AUX product to assess cloud thermodynamic phase. The radar voxels have a horizontal resolution of 1.4 km (cross track)
To assess the differences between the cloud phase from the DARDAR-MASK and CALIPSO-GOCCP products, we defined a new phase ratio based on the DARDAR-MASK classification. In this alternative definition, which we call ALT-DARDAR, only grid boxes (
Same as Fig.
Same as Fig.
Some major differences can be observed between the three FPR* variables in Fig.
The detected ice virgae below the liquid cloud top suggest that the cloud top did not fully attenuate the lidar signal (not optically thick enough). The number or size of the ice particles near the cloud top probably was not enough to increase the depolarization ratio above the threshold value for the GOCCP algorithm, and it was therefore classified as liquid.
In the decision tree of the DARDAR algorithm, there are multiple alternatives for a mixture of cloud droplets and ice particles (e.g. at cloud top) to be classified as ice only.
If the lidar backscatter signal is lower than If not (a), it is weakly attenuated (less than 10 times) or not rapidly attenuated (at a depth larger than 480 m). If not (b), the layer thickness of the cloud is larger than 300 m. This is equivalent to five voxels with a lidar vertical resolution of 60 m.
Therefore, there are many cases where a mixed-phase cloud can be misclassified as ice only in the DARDAR product and consequently in the FPR
In the case of droplets and ice particles coexisting at cloud top, we expect that at some location the cloud droplets will be enough in number for one of the voxels to be classified as liquid (strong attenuation) in the DARDAR-MASK algorithm. If this is the case, the entire volume grid box value of FPR
Summary of the different variables used to assess the frequency phase ratio (FPR).
Same as Fig.
In summary, the GOCCP algorithm is unable to detect ice in mixed-phase clouds, and the DARDAR algorithm tends to classify mixed-phase clouds as ice. Therefore, we avoid using the frequency of cloud ice (FPR) to compare the GOCCP and DARDAR products. Instead, we use the FPR
For temperatures between
In contrast, FPR
As shown in Fig.
All datasets used in the analysis are freely available at (last access: 13 February 2019):
The supplement related to this article is available online at:
IT, BH, PS, and DV contributed to the design of the study. DV processed the datasets, performed the analysis, designed the figures, and drafted the paper. All authors contributed valuable feedback throughout the process. All authors helped with the discussion of the results and contributed to the final paper.
The authors declare that they have no conflict of interest.
We thank the GOCCP project for providing access to the CALIPSO-GOCCP gridded cloud-phase profiles. We thank the NASA CloudSat project and the CloudSat Data Processing Center for providing access to the 2B-CLDCLASS product. We thank the ICARE Data and Services Center for providing access to the DARDAR and CloudSat data. We thank the MACC project and the ERA-Interim science team for providing access to the reanalysis data. We thank Albert Ansmann, Johannes Mülmenstädt, and Julien Delanoë for helpful discussions. The authors would like to thank the editor and anonymous referee no. 4 from the first version of the paper for suggesting the inclusion of constraints for humidity and static stability, which greatly improved the accuracy of the results.
The publication of this article was funded by the Open Access Fund of the Leibniz Association.
This paper was edited by Matthias Tesche and reviewed by two anonymous referees.