Volume 123, Issue 22 p. 12,762-12,777
Research Article
Free Access

The Role of Secondary Ice Processes in Midlatitude Continental Clouds

Assaf Zipori

Assaf Zipori

Department of Earth and Planetary Sciences, Weizmann Institute of Science, Rehovot, Israel

Fredy and Nadine Herrmann Institute of Earth Sciences, Hebrew University, Jerusalem, Israel

Contribution: Methodology, Formal analysis, ​Investigation, Data curation, Writing - original draft

Search for more papers by this author
Naama Reicher

Naama Reicher

Department of Earth and Planetary Sciences, Weizmann Institute of Science, Rehovot, Israel

Contribution: ​Investigation, Writing - original draft

Search for more papers by this author
Yigal Erel

Yigal Erel

Fredy and Nadine Herrmann Institute of Earth Sciences, Hebrew University, Jerusalem, Israel

Contribution: ​Investigation, Writing - original draft

Search for more papers by this author
Daniel Rosenfeld

Daniel Rosenfeld

Fredy and Nadine Herrmann Institute of Earth Sciences, Hebrew University, Jerusalem, Israel

Contribution: ​Investigation, Writing - original draft

Search for more papers by this author
Amir Sandler

Amir Sandler

Geological Survey of Israel, Jerusalem, Israel

Contribution: ​Investigation, Writing - original draft

Search for more papers by this author
Daniel A. Knopf

Daniel A. Knopf

Institute for Terrestrial and Planetary Atmospheres, School of Marine and Atmospheric Sciences, Stony Brook University, New York, NY, USA

Contribution: ​Investigation, Writing - original draft

Search for more papers by this author
Yinon Rudich

Corresponding Author

Yinon Rudich

Department of Earth and Planetary Sciences, Weizmann Institute of Science, Rehovot, Israel

Correspondence to: Y. Rudich,

[email protected]

Contribution: Conceptualization, ​Investigation, Writing - original draft, Supervision

Search for more papers by this author
First published: 09 October 2018
Citations: 12


Clouds contribute very large uncertainties to our understanding of Earth's climate system. This is partly attributed to the insufficient predictive abilities of ice formation processes in clouds and the ramifications for the hydrological cycle and climate. To improve predictions of ice particle concentrations in clouds, a better understanding of the relative contributions of ice nucleating particles and secondary ice processes (SIPs) is needed. To address this challenging question, we combine ice nucleation measurements via immersion freezing of particles filtered from rainwater, with satellite-retrieved cloud top glaciation temperatures (Tg) of the same clouds, while considering the chemical composition of the rainwater, the particles, and the particles' mass loads. In addition, laboratory-derived ice nucleation parameterization of K-feldspar was implemented in an ice nucleation model in order to reconstruct Tg considering primary ice nucleation only. We show that the observed Tg does not correlate with the median freezing temperature of the drops from the laboratory measurements froze (T50), and are significantly warmer than the model prediction. This suggests that SIP play a major role in glaciating the investigated clouds system. Furthermore, we show that the difference between Tg and T50 best correlates with the size of the cloud droplets at −5 °C, indicating that SIP is controlled by cloud droplet sizes. Hence, our results suggest that the effect of SIP on Tg, and therefore on Earth's radiation budget, may be significant.

Key Points

  • Ice nucleation efficiency of natural ice nucleation particles is controlled by K-feldspar
  • Secondary ice processes have a large influence on the ice content in midlatitude orographic-triggered convective clouds
  • The effect of secondary ice processes on Earth's radiative budget is possibly greater than previously assumed

1 Introduction

In mixed-phase clouds (where liquid water droplets and ice crystals coexist), ice forms via either primary ice nucleation or secondary ice processes (SIPs). Primary ice nucleation depends on the presence of ice nucleating particles (INPs; Boose et al., 2016; Demott et al., 2003, 2010; Hoose & Moehler, 2012; Knopf et al., 2018; Rogers & Yau, 1989) that promote heterogeneous freezing at warmer temperatures than the homogenous freezing temperature of pure water (−38 °C; Koop & Zobrist, 2009; Rosenfeld & Woodley, 2000). In SIP, ice crystals can form by four main mechanisms: rime splintering (Mossop, 1976), droplets shattering (Lawson et al., 2015; Mason & Maybank, 1960), break-up upon collisions of ice crystals (Vardiman, 1978), and sublimation fragmentation (Bacon et al., 1998). Ice formation in clouds greatly affects Earth radiative balance, as it influences clouds' reflectivity (Lohmann & Feichter, 2005; Seinfeld et al., 2016), lifetime (Chakraborty et al., 2016; Rosenfeld et al., 2001), precipitation initiation (Chakraborty et al., 2016; Lohmann & Feichter, 2005), and atmospheric electrification (Fuchs et al., 2015; Guo et al., 2016; Williams et al., 1991). Therefore, understanding the role of the primary ice processes and SIPs is essential for assessing the effects of clouds on climate.

Since most ice nucleation in mixed-phase clouds occurs after liquid droplets have formed (Ansmann et al., 2009; de Boer et al., 2011), immersion freezing is thought to be the dominant mechanism of primary ice nucleation in these clouds (Hande & Hoose, 2017; Niehaus et al., 2014). Contact freezing (where freezing occurs following a collision of an INP and a water droplet), may be an important ice nucleating process (Hoose et al., 2008; Lohmann & Diehl, 2006; Lohmann & Hoose, 2009). However, experimental data is limited and indicates either enhancement (Hoffmann, Duft, et al., 2013; Hoffmann, Kiselev, et al., 2013; Moreno et al., 2013) or decrease in ice formation (Hande & Hoose, 2017; Nagare et al., 2016; Niehaus et al., 2014). In immersion freezing, ice forms on an INP immersed inside a supercooled cloud water droplet. The most abundant INPs in the atmosphere are thought to be desert dust (Demott et al., 2003), which plays key roles in affecting cloud properties and lifetimes (Boose et al., 2016; Murray et al., 2012; Rosenfeld et al., 2001). The most effective mineral component in desert dust is thought to be K-feldspar, which was suggested to determine the ice nucleation activity of dust (Atkinson et al., 2013; Augustin-Bauditz et al., 2014; O'Sullivan et al., 2014; Vergara-Temprado et al., 2017).

However, the contribution of primary nucleation to ice production in mixed-phase clouds is still uncertain. In situ studies often indicate that the measured ice concentrations are substantially higher than the measured INP concentration (Ackerman et al., 2015; Hobbs & Rangno, 1990; Hogan et al., 2002; Ladino et al., 2017; Lawson et al., 2015; Taylor et al., 2016), leading to the hypothesis that under suitable conditions, SIP dominate the ice content in mixed-phase clouds. Furthermore, when implementing the DeMott et al. (2010) parameterization to predict the INP concentration, numerical models cannot explain the observed ice particles size distribution and concentration (Farrington et al., 2016; Fridlind et al., 2012).

SIPs mainly depend on the ambient temperature and droplets' size distribution (Field et al., 2017; Lawson et al., 2015; Mossop, 1976; Sullivan et al., 2018). Several studies recognized the importance of SIP in glaciating clouds in pristine environments, such as the marine and polar atmospheres (Burrows et al., 2013; Engel et al., 2013; Hobbs & Rangno, 1990) and in convective clouds (Field et al., 2017; Ladino et al., 2017). Previous studies found high ice concentrations in continental clouds that cannot be attributed to primary ice nucleation (Blyth & Latham, 1993; Rangno & Hobbs, 1994). However, the measured high ice concentrations may also result from shattering ice artifacts (Farrington et al., 2016; Field et al., 2006; Jensen et al., 2009; Korolev et al., 2011). In general, the role of SIP in glaciating midlatitude orographic continental clouds is assumed to be minor, due to typical high aerosol and INP concentrations that affect these clouds (Freud et al., 2015; Lawson et al., 2017; Rogers & Yau, 1989; Sullivan et al., 2018; Zipori et al., 2015).

SIPs were first documented in laboratory experiments by Hallett and Mossop in the 1970s (Hallett & Mossop, 1974; Mossop, 1976). They showed that SIP through rime splintering is most efficient when the droplets' radii exceed 12 μm and at temperatures between −3 and −8 °C. This process is often called rime splintering or the Hallett-Mossop process (H-M process). Numerical models often use the Hallett-Mossop process to describe SIP (Field et al., 2017), using the same size threshold (droplets radius > 12 μm) and temperature range (−3 to −8 °C; Connolly et al., 2006; Crawford et al., 2012; Dearden et al., 2016; Fridlind et al., 2007; Scott et al., 1977). Another study suggested that the concentration of cloud condensation nuclei (CCN) that affects the cloud droplet size distribution (Andreae et al., 2004; Braga et al., 2017; Mossop, 1985) has a greater influence on ice content in clouds via SIP than the actual INP concentration (Sullivan et al., 2018). The effects of SIP on cloud glaciation were also suggested by Rosenfeld et al. (2011), who showed that the retrieved cloud top glaciation temperature (Tg) of convective clouds increases as the effective radius at the −5 °C isotherm (Re-5) increases.

The increase in ice crystals' concentration through SIP can further expedite cloud glaciation, as ice formation creates subsaturation with respect to water that leads to evaporation of water from existing droplets through the Wegener-Bergeron-Findeisen (WBF) process (Rogers & Yau, 1989). However, this effect can also decrease the rate of SIP through drop shattering and rime splintering (Crawford et al., 2012; Phillips et al., 2001), as it decreases the number and size of supercooled water droplets.

In this study, we used remote sensing to retrieve clouds' Tg (Rosenfeld et al., 2011; Rosenfeld & Lensky, 1998) and Re-5 (Rosenfeld et al., 2011) in continental orographic-triggered convective clouds that form in the eastern Mediterranean. We compared the satellite observations with laboratory measurements of immersion ice nucleation efficiency of insoluble atmospheric particles that were extracted from rain samples collected from the same cloud systems (Figure S1 in the supporting information). In addition, ice nucleation efficiencies of K-feldspar were used in a simple cloud model to predict the temperature for primary ice formation by immersion freezing in order to estimate the possible role of SIP in the investigated cloud systems.

INP freezing spectra from precipitation and cloud samples have been previously evaluated (Du et al., 2017; Petters & Wright, 2015), and satellite retrievals have been used for evaluating the importance of Re-5 with respect to SIP (Rosenfeld et al., 2011). However, the current study integrates, for the first time, aerosol characterization, INP freezing properties, and Tg estimates from remote sensing data, to derive a comprehensive picture of the importance of SIP to continental clouds.

2 Methods

2.1 Sample Collection and Handling

2.1.1 Rain Sampling, Handling, and Chemical Analysis

Rain samples for chemical analyses were collected in northern Israel (Nimrod, 1,100 m ASL; Figure S1) during three winters (2014–2017) in 50-ml tubes using a plastic funnel. Typically, rain clouds in this area are orographic-triggered convective clouds, with cloud base below 1,000 m ASL (Freud et al., 2015). Further details on the microphysical properties of the clouds in this area can be found in previous studies (Freud et al., 2015; Levin et al., 1996; Rosenfeld et al., 2001; Rosenfeld & Farbstein, 1992; Rosenfeld & Gutman, 1994; Zipori et al., 2015). The rain samples were acidified with distilled nitric acid for 1 to 2 weeks at a volumetric ratio of 1:100 to digest suspended particles. A total of 23 metals (Na, Mg, Ca, Al, K, V, Cr, Fe, Mn, Co, Ni, Cu, Zn, As, Rb, Sr, Mo, Ag, Cd, Ba, Tl, Pb, and U) were quantified using an ion coupled plasma mass spectrometer (ICPMS, Agilent, 7500cx). Calibration procedures, data quality control, and detection limits values were similar to those presented in Zipori et al. (2012, 2015) and hence are not detailed here. By applying principal components analysis and cluster analysis on this data set (Aldenderfer & Blashfield, 1984; Baxter, 1995; Dunteman, 1989), we identified the main aerosol types in the rainwater and the corresponding air mass origin. These analyses include also samples used by Zipori et al. (2015; 1,676 samples) plus additional 522 samples that were collected between 2013 and 2017. Further details and results of the statistical analysis are presented in the supporting information (Text S1, Table S1, and Figure S2).

Dayan et al. (1988) showed that during rain events in Israel, the mixed layer height is above 1,000 m ASL, which is the same height of the clouds base. In addition, the sampling station was at approximately the same altitude. This allows us to assume that the particles found in the rainwater can represent the aerosols in the clouds. In addition, studying aerosols properties and their effects on clouds is a common methodology that was used in several places before (Ault et al., 2011; Du et al., 2017; Petters & Wright, 2015; Zipori et al., 2015). These studies are based on the assumption that the particles that are scavenged by the raindrops represent the aerosol population in the same air mass and in the cloud.

2.1.2 INP Collection and Handling

In addition to the 50-ml samples, rainwater was also collected in 1-L bottles using two plastic funnels. The collected rainwater was filtered immediately after sampling through 0.2-μm filter (Whatman Nuclepore, Polycarbonate 47 mm), using a Teflon filter holder and a vacuum pump. Particles present in the rainwater were collected on the filter, while the soluble material and aerosols smaller than 0.2 μm passed through the filter. After filtration, the filters were sealed in clean petri plates. A blank filter was taken for each rain event, by filtering 1 L of Deionized water (DIW). Detailed sampling time of these samples can be found in Table S2.

Insoluble particles were recovered from the filters by sonication for 40 min (Delta DC200H) in 5-ml DIW in precleaned Teflon beakers. In cases where the particles' mass extracted was insufficient, a more intensive sonication was performed using Hielscher, UP200st sonicator. This was repeated three times, each time for 1 min, with 2 ml of the sample suspension (from the previous extraction). This suspension was used for the mineralogy and immersion mode ice nucleation activity analyses. Blank filters were extracted by the intensive sonication in 2 ml in DIW.

After each rain event, the filters were dried in a vacuum chamber (Struers, CitoVec) for at least 24 hr. The filters were weighed (Mettler Toledo MX5) before and after sampling to measure the net particulate mass collected and to calculate the particle mass density in the collected rainwater. In addition, in order to calculate the extraction efficiency, the filters were dried and weighed again after the extraction procedure. The extraction efficiency of the sampled filters ranged between 60% and 80% (Figure S3). The blank filters' mass changed by less than 2% (±0.2%) when weighed before sampling and after extractions. Weighing was performed under controlled conditions with relative humidity of 30%. Filters were kept at 4 °C until extraction and analysis.

2.2 Filter Analysis

2.2.1 Mineralogy

Insoluble particles recovered from the filters were analyzed for chemical composition using Electron Probe Micro Analyzer (JXA 8230 SUPERPROBE). Three 50-μl drops from each sample's suspension were placed on a gold-coated epoxy surface and were dried in a vacuum chamber (Struers, CitoVec). Prior to the analysis, the samples were coated with carbon. Samples were analyzed for Na, Cl, Al, Fe, Si, K, Mg, Ca, S, P, N, and Zn. We derived the mineralogy of the samples following the method described in Kalderon-Asael et al. (2009), where the mineralogy is retrieved from the chemical composition of the analyzed particles, using a conversion table. This method defines only natural minerals, so all elements not associated with any mineral were defined as excess. This excess is assumed to be anthropogenic and\or biogenic (Kalderon-Asael et al., 2009). Due to the small samples size, it was impossible to perform a more direct mineralogical analysis using X-ray diffraction. The method used here provides a crude mineralogical classification of each sample. Further information regarding this procedure and the conversion table are available in the supporting information (Text S2, Table S3 and Figure S4).

2.2.2 Immersion Mode Ice Nucleation Activity

Ice nucleation activity was examined using the WeIzmann Supercooled Drops Observation on Microchip (WISDOM; Edd et al., 2009; Reicher et al., 2018). All samples were analyzed at a 0.01 wt%. Drops were generated inside a microfluidic chip using pneumatic pumps (NE-500 Programmable OEM Syringe Pump), with 40-μm diameter droplets. Then, the chip was placed on a cooling stage (Linkam LTS420, cooling rate of 1 °C/min) under an optical microscope, and freezing was recorded by a charge-coupled device camera to provide a frozen fraction as a function of temperature. Freezing events were automatically detected by following changes in the brightness level of each drop. A schematic description of the microfluidic chip, detailed description of the droplet generation process, and the temperature calibration procedure can be found in Reicher et al. (2018), and in the Information (Text S3, Figures S5 and S6, and Table S4).

From the freezing profile, the number of nucleation sites per unit mass (Nm) was retrieved (Murray et al., 2012; Niemand et al., 2012; O'Sullivan et al., 2014). Although it is more common to normalize these parameters to the surface area, we used the mass concentration due to small sample size. The maximal amount that was collected on the filters was 85 mg, which did not enable Brunauer-Emmett-Teller (BET) specific surface area measurements. Other methods such as electron microscopy may have large uncertainties in surface area determination due to porosity, aggregation, and 3D shape assumptions of the particles (Bickmore et al., 2002; Macht et al., 2011). These uncertainties can lead to errors when calculating the number of nucleation sites and the freezing rate.

The blank filter analysis was used to exclude possible interference from contaminations that originate in the filters themselves. To do that, we defined the coldest temperature for analyzing the samples:
where, Tlim is the coldest temperature for analysis, urn:x-wiley:2169897X:media:jgrd54999:jgrd54999-math-0002 is the blank samples' mean temperature, where 10% of the drops froze, and σ10 is the standard deviation of the mean.

2.3 Tg Retrievals

The Tg was retrieved from the MeteoSat-09 geostationary satellite using the method detailed in Rosenfeld and Lensky (1998) and refined by Yuan et al. (2010). MeteSat-09 provides an image every 15 min with the spatial resolution of approximately 5 × 5 km in the analyzed area (Figure S1). This satellite is a passive satellite measuring reflected solar and infrared radiation. To retrieve Tg, the temperature and effective radii of the cloud tops (CTT and Re, respectively) that were within the analyzed area (Figure S1) were plotted against each other. The CTT of the clouds were divided into 1 °C intervals, from which the upper tenth percentile Re was taken. Since ice crystals have higher reflectivity than water droplet at wavelength of 3.9 μm (which is used to calculate the Re), Tg is determined when there is rapid increase in Re in the T-Re plot (Lensky & Rosenfeld, 2008). Some selected T-Re plots from each sampling time are presented in Figure S7.

In growing convective clouds, the glaciation level can spread to lower and warmer parts of the cloud with its maturation by precipitation of ice particles along as the WBF process progresses. The low percentile of the Re improves the detection of the initial freezing.

Because each of the rain samples was collected for several hours and the satellite images are provided every 15 min for each case, several satellite images were analyzed separately, and the mean Tg was taken. Error estimation on the mean Tg was calculated from its standard deviation. Uncertainty involving the detection of Tg is assumed to be minor. Because Re is examined in a relative perspective (i.e., Tg is detected where a sudden increase in Re is detected), such uncertainty can originate only from errors in the temperature. The CTT is calculated at 10.8 μm wavelength, which is within the atmospheric window and has minor interference from water vapor (Hunt, 1973; McClain, 1981). However, because the air above the clouds is very dry, detection of CTT should not be affected by such interference. Therefore, we assume that the uncertainty of Tg is very small (probably less than 1 °C; Lensky & Rosenfeld, 1997).

2.4 Immersion Freezing Model

The effect of primary ice processes on cloud glaciation was also examined by a model describing two approaches: a time-based approach that applies heterogeneous ice nucleation rate coefficient (Jhet; Alpert & Knopf, 2016; Knopf & Alpert, 2013) and a time-independent nucleation approach that uses the active sites density (Ns) parameterization (Atkinson et al., 2013; Broadley et al., 2012). Using these two approaches, the model calculated the frozen fraction (FF) in the cloud as a function of temperature assuming that immersion freezing is the sole freezing mechanism. The model was designed to support the formation of primary ice production via immersion freezing by assuming that all of the INPs had ice nucleation activity of K-feldspar. In addition, we estimated the INP concentration in the clouds to be higher by a factor ranging from 1 to 104 than the particle concentration found in the rainwater. To further support cloud glaciation due to primary ice nucleation, the calculated glaciation temperature by the model (Tgcalc) was determined when the calculated FF reached 10%. The WBF process was parameterized as a threshold process (Lohmann et al., 2007). When the calculated FF reached 0.05%, total glaciation occurred either within 5 min or when the calculated FF reached 10%. Although the WBF process does not increase ice concentration, it can increase FF by evaporating water droplets and decreasing their concentration. For the time-based model, we calculate the Jhet from the measured freezing temperatures by the following equation (Knopf & Alpert, 2013):
where m = −0.297, c = −3.54, and T is the temperature in Celsius. The values of m and c were retrieved by examining K-feldspar (FS04; Peckhaus et al., 2016)) solutions at different concentrations (Figure S8a).
For the time-independent model, Ns is described as a function of temperature by the following equation:
where m = 0.331, c = −2.48, and T is the temperature (Figure S8b).
The FF was calculated from Jhet following Pruppacher and Klett (1997):
where DC is the particles mass in the drop, BET is the surface area of K-feldspar, and t is the time step. The intervals in the model were set to be 0.1 °C, so t varied according to the selected updraft velocity.
The FF was calculated from Ns using equation 18 from Murray et al. (2012):
A flow chart of the model, validation test using the results presented by Knopf and Alpert (2013) for illite and sensitivity test for the updraft velocity are presented in Figures S9–S11.

3 Results

3.1 Air Mass Classification

The chemical composition of the collected unfiltered rainwater from the ICPMS analysis was used to determine the characteristics of air masses. Four different air mass types were characterized by the principal components analysis: Saharan, carrying high dust concentration from the Sahara Desert; Marine, with high fraction of sea salt; Anthropogenic, enriched with metals emitted by transportation and/or industry; and Mixed, indicating air masses with more than one main dominant source. Further information on the classification process and results can be found in supporting information S1, Table S1, and Figure S2. These results are also consistent with Zipori et al. (2015).

3.2 Mineralogical Classification

The elemental composition of the insoluble particles was determined using Electron Probe Micro Analyzer (Table S3 and Figure S4). Following that, the mineralogical composition was extrapolated using the method described in Kalderon-Asael et al. (2009). All four samples (collected at different times, see Table S2) from the Saharan air masses (Sah-1 to Sah-4) contained the highest fraction of mineral dust (mean = 84%, σ = 9%). The most abundant minerals in this group were illite-smectite, quartz, K-feldspar, and calcite. These samples also had the highest particle concentration in the collected rainwater (Table S2). The dust fraction in the Marine samples (Mar-1 and Mar-2) was substantially lower than in the Saharan samples (47% and 30%, respectively), resulting in a high excess fraction (elements that were not classified as natural minerals, see section 2.2.1 for further details). Due to its high solubility, small amounts (<5%) of halite were found in all samples, with highest fraction in the Marine samples (Figure 1).

Details are in the caption following the image
Calculated mineralogy of the insoluble particles from the collected rainwater samples following Kalderon-Asael et al. (2009). Sah, Mar, and Ant stand for samples collected during saharan, marine and anthropogenic dominated air masses, respectively (see supporting information S1, methods, and Zipori et al., 2015). The term illite-smectite represents total phyllosilicates in the dust as this group is the dominant phyllosilicate phase in dust in Israel (supporting information S2; Sandler, 2013). Error bars were estimated to be 30% based on Kalderon-Asael et al. (2009). Following the completion of the procedure described in Figure S4, the remaining elements' mass fraction sum was defined as Excess. This component represents mainly anthropogenic and organic aerosol components (Kalderon-Asael et al., 2009).

The mineralogical composition of Ant-2 had a higher mass fraction of dust than Ant-1 (72% and 39%, respectively; Figure 1). Despite the similar mineralogy of Ant-2 and the Saharan samples, Ant-2 was classified as Anthropogenic because the air mass classification was based on the unfiltered rainwater chemical composition, and the mineralogical analysis was based on the insoluble particles filtered from the rainwater. While the former is affected by the soluble fraction and the particle load in the rainwater, the latter is affected by the nonsoluble particles in the rainwater.

The anthropogenic influence on sample Ant-2 is seen in the chemical composition of the unfiltered rainwater (Figure 2). The Pb/Al ratio, indicative of anthropogenic influence (Harrison & Winchester, 1971), in the Ant-2 rainwater was significantly higher than in the Saharan samples (p value < 0.001). Moreover, the dust loads in the Saharan air masses were substantially higher than in the anthropogenic air masses (Zipori et al., 2015). The particle concentration in the Saharan samples varied between 10 and 300 mg/L,, while in sample Ant-2 it was substantially lower (1.35 mg/L; Table S2). Note that in order to avoid biases, we present the mineralogical composition as a weight fraction to eliminate the effect of particle loading in rainwater. These inferences are also supported by the air mass trajectories: the back trajectories of the Saharan samples passed above areas of high potential to produce dust (Figures 3a–3d), whereas the main region influencing sample Ant-2 was likely the Nile Delta (Figure 3h), contributing anthropogenic aerosols into the air mass.

Details are in the caption following the image
Distribution of log10(Pb/Al) values found in the rainwater from the 50-ml samples that were collected during Ant-2 sample (black dashed line) and during the Saharan air mass samples (brown continues line). The mean values of the Pb/Al ratios were 0.02 and 0.05 for the Saharan and Ant-2 samples, respectively.
Details are in the caption following the image
Air mass backward trajectories for the samples collection times. The trajectories were calculated by HYSPLIT model, at 1,000 m ASL, for 72 hr, using the GDAS (Global Data Assimilation System) and the reanalysis data sets (the mean trajectories are plotted). The red, green, and blue symbols represent the trajectories from the beginning, the middle, and the end of the sampling time, respectively. The shaded areas represent the uncertainty of the trajectories based on one standard deviation from the mean. The yellow symbols represent the main cities locations. The height profiles of the trajectories are presented in Figure S12.

3.3 Immersion Freezing Measurements

The immersion mode ice nucleation activity of the collected insoluble particles was studied in the WISDOM (Figures S5 and S6) (Reicher et al., 2018) and is expressed as active sites per mass (Nm) as a function of temperature (Figure 4).

Details are in the caption following the image
Laboratory results of immersion freezing experiments of nucleation site density per mass (Nm) representing ice nucleating particle (INP) activity in the immersion mode, as a function of temperature. Nm is plotted for various samples as given in the legend. All samples were diluted to 0.01 wt% with double distilled water. Error bars represent the uncertainty in Nm based on the error in frozen fraction and in INP mass in the droplets. The calculation of the error bars can be found in the supporting information S3. The error of the temperature is 0.3 °C. The gray area is the Atkinson et al. (2013) estimation for ice nucleation activity of natural dust based on the typical abundance of K-feldspar in soils. The hexagonal, triangle, and diamond light-blue symbols are for the 3.2, 1.8, and 1 μm size resolved samples from Reicher et al. (2018), respectively.

The samples can be divided into two main populations by their characteristic ice nucleation activity: The Saharan samples as well as the Mar-1 sample had similar ice nucleation activity, consistent with reported activity of K-feldspar by Augustin-Bauditz et al. (2014) and with the predicted ice nucleation activity of natural dust (Atkinson et al., 2013). This suggests that the active INP activity in these samples was dominated by K-feldspar, even though it was not the most abundant mineral in these samples (Figure 1). Our observations are consistent with the settled dust samples collected previously in Tel-Aviv (Niemand et al., 2012) and with recent measurements of size-resolved mineral dust by Reicher et al. (2018). Samples Mar-2 and Ant-1 contained less effective INPs. The dust and the K-feldspar mass fractions in these samples (30–39% and 2–4%, respectively) were lower compared to the group with the higher ice nucleation efficiency (47–90% and 7–14%, respectively). The high excess fraction, indicative of the presence of anthropogenic, organic, and black carbon aerosols, may explain the lower ice nucleation activity observed (Hoose et al., 2008; Lohmann & Diehl, 2006). This group also contains higher fraction of iron oxides, which may be internally mixed or coat aerosol particles (Erel & Pehkonen, 1993; Takahashi et al., 2011). The samples presented by Reicher et al. (2018) have a wide range of Nm values, due to the size separation. The INP properties of dust with the larger diameter (3.2 μm, light blue hexagonal) are consistent with the samples of this work that show high ice nucleation activity (Sah 1–4 and Mar-1), while the smaller diameters dust (1 μm, light blue diamond) behave similarly to the samples that show lower ice nucleation activity. This may be due to mixing of the dust with local pollution. However, since we did not perform size distribution analysis and Reicher et al. (2018) did not have a mineralogical analysis, it is difficult to directly compare these observations. At warm temperatures, sample Ant-2 followed the K-feldspar freezing profile. At colder temperatures, the freezing profile shifted toward the less efficient group (Figure 4). The K-feldspar fraction in this sample was 9% (Figure 1), which is consistent with the measured ice nucleation activity at the warmer temperatures. The shift toward the second INP group may be attributed to the anthropogenic effects that include physical and chemical aging and/or mixing with less efficient INPs, a process that reduces the overall ice nucleation efficiency.

3.4 Comparison of Tg With T50

For each sample, satellite images were analyzed from the same sampling period and from the area marked in Figure S1 in order to retrieve the mean Tg during the sampling time. More details on the procedure are in section 2.3 and in Figure S11. Tg was then compared with the temperatures at which 50% of the drops froze in the WISDOM experiments (T50) and their relation with several parameters (Figures 5a–5e). The retrieved Tg are substantially warmer than the measured T50, and the Tg values do not correlate with the corresponding T50 values. The lack of correlation can be attributed to the fact that while the WISDOM measurements represent only primary ice production via immersion freezing, Tg can be affected by other processes, such as WBF and SIP (Rogers & Yau, 1989). Additionally, the particle surface area used in the WISDOM experiments was constrained to the drops size (40 μm) at particle mass concentration of 0.01 wt%. About 1,000 droplets were analyzed (in three repetitions for each sample, to ensure statistical significance; Alpert & Knopf, 2016). In contrast, in atmospheric clouds, the particle concentration in the droplets and the droplet number can be substantially higher, increasing the probability to have active INPs that initiate freezing at relatively warm temperatures (Marcolli et al., 2007). Figure 5b suggests that the discrepancy between Tg and T50 may be a result of SIP, as Tg increases with the droplet's effective radius at the −5 °C isotherm (Re-5). This result is consistent with a previous study showing that higher Tg correlated with increasing Re-5 in convective clouds (Rosenfeld et al., 2011). However, in that study, this connection did not hold for the cases with high dust loads.

Details are in the caption following the image
Estimating factors that influence secondary ice process in the studied cloud systems. (a) Comparison of cloud top glaciation temperature (Tg) retrieved from satellite observations with the temperature at which 50% of the drops froze in the WeIzmann Supercooled Drops Observation on Microchip (WISDOM) experiments (T50). (b) The relation between Tg and the mean effective radius at −5 °C (Re-5). (c) The difference between Tg and T50T) versus K-feldspar content in the samples as calculated from the Electron Probe Micro Analyzer analysis using Kalderon-Asael method (Kalderon-Asael et al., 2009). Horizontal error bars represent the uncertainty of the K-feldspar fraction (taken to be 30%). (d) The dependence of ΔT on particle concentration in the sampled rainwater. (d) The effect of Re-5 on ΔT. Error bars for Re-5 and Tg are the confidence interval of the mean values (α = 0.05). Error bars for ΔT originate from the errors on Tg. Sah-1 and Mar-2 were collected during night time, and therefore, it was not possible to calculate Tg for these samples (Rosenfeld & Lensky, 1998).

To further explore the causes for the discrepancies between Tg and T50, we define ΔT, the difference between Tg and T50. Smaller ΔT values are expected when the ice formation processes in the clouds are controlled by immersion freezing, while larger ΔT values imply that other processes such as SIP dominate cloud glaciation. Figure 5c shows that a higher K-feldspar fraction in the particles decreases ΔT, suggesting that K-feldspar is a key factor in primary ice nucleation. However, uncertainties associated with the K-feldspar fraction estimation result to a low significance level (p value = 0.094). Therefore, we consider other relationships between the particle concentration in the rainwater, which represent the particle load in the clouds, and ΔT. Figure 5d shows that ΔT reaches a minimal value at a concentration of approximately 20 mg/L (samples Sah-2 and Sah-4). At minimal ΔT values, ice-forming processes in the corresponding clouds are best represented by the WISDOM results, implying smaller contribution of SIP for the Sah-2 and Sah-4 events. The decrease in ΔT when the particle concentration is lower than 20 mg/L is attributed to the suppression of SIP with decreasing cloud droplets' effective radius, as was previously suggested (Hallett & Mossop, 1974; Lawson et al., 2015; Mossop, 1976; Sullivan et al., 2018). The high ΔT value of sample Sah-3 (concentration of ~300 mg/L) may be attributed to the more efficient ice nucleation processes that operated in these clouds due to the higher surface area of INPs immersed in cloud droplets compared to the concentrations employed in the WISDOM experiments. In addition, high dust concentrations also increase the probability for the presence of giant CCN, which can form large drops and increase the SIP efficiency (Alfaro & Gomes, 2001; Levin et al., 2003). Figures 5d and 5e provide a support for initiation of SIP due to giant CCN in sample Sah-3. The Re-5 retrieved for sample Sah-3 is larger than Re-5 retrieved for samples Sah-2 and Sah-4, although samples Sah-3 had much higher particle concentration in the rainwater. Note that the Re-5 values for samples Sah-2 and Sah-3 were larger than 12 μm, which is the threshold for Hallett-Mossop ice multiplication process (Field et al., 2017; Hallett & Mossop, 1974; Sullivan et al., 2018). This observation is consistent with Connolly et al. (2006) who found that higher INP concentrations increased SIP through the production of splinters in convective clouds. It is possible that this effect is also detected here. Figure 5e also demonstrates the importance of Re-5 to SIP, as the correlation coefficient and the significance level between ΔT and Re-5 are the highest among the presented relations.

Note that sample Sah-1 and Mar-2 were excluded from Figure 5 because they were taken during nighttime, where Re measurements are not available, and therefore, it was not possible to retrieve Tg. In addition, Tg values of Sah-3 and Sah-4 are based on one satellite image, unlike the rest of the samples where the reported Tg is the mean Tg of several images from the same sampling period and the error bars of Tg and ΔT are based on the confidence interval of the mean (α = 0.05). Because sample Sah-3 was collected mostly during nighttime, it was possible to retrieve Tg only from one satellite image. For sample Sah-4, the clouds' tops were too warm in most cases to detect glaciation. Because it was impossible to calculate the confidence interval for these samples, the Tg uncertainty for Sah-3 and Sah-4 was estimated to be ±3 °C. Further discussion on the uncertainty related to Tg retrieval is in section 2.3.

Atmospheric aging processes of the particles may modify the ice nucleation activity of dust particles and other INP (Brooks et al., 2014; Kanji et al., 2013, 2017; Knopf et al., 2018; Kulkarni et al., 2014; Sullivan, Miñambres et al., 2010; Sullivan, Petters et al., 2010). If such process occurred in the rainwater samples, they may have affected ΔT by altering T50. However, in our data set, we do not observe evidence for such processes as the freezing profiles of samples with relatively high K-feldspar fraction were similar to previously reported results of K-feldspar (Figure 4). Such wet aging may have played a role in samples Ant-2, but it is not possible to determine whether such aging occurred inside cloud droplets or during transport. In addition, a soluble material, potentially present as a coating, can decrease the ice nucleation efficiency of dust particles by either blocking ice active sites or decreasing water activity (Kanji et al., 2017; Knopf et al., 2018; Knopf & Alpert, 2013). Since during the filtration process such coating rinsed from the particles, our INP measurements ignore such effect and present an upper bound for the ice nucleation activity of the collected particles.

Biological particles or components may be important INPs, especially in the warm temperature regime (T > −15 °C; Morris et al., 2004; O'Sullivan et al., 2014). Such particles may be found as independent particles (externally mixed) or mixed within the dust particles (internally mixed). Since our Electron Probe Micro Analyzer analysis could not detect organic matter because the samples were coated by carbon, we cannot determine the absence or presence of such particles. However, our INP laboratory measurements show no evidence for the existence of such highly active INPs. Because we dried the filters at room temperature under vacuum (see section 2.1.1), destruction of such biological matter is not expected (O'Sullivan et al., 2014). In addition, studies have shown that the abundance of biological INPs in the atmosphere is typically an order of magnitude lower than mineral dust and soot (Broadley et al., 2012; Hoose et al., 2010). This ratio is expected to be even lower when the air mass contains high concentrations of dust from arid regions such as the Sahara Desert (O'Sullivan et al., 2014). If such biological INPs did exist in these cases, their concentration is expected to be too low to glaciate the clouds just by primary ice nucleation, and therefore, some SIP must have occurred.

3.5 Model Simulations

Figure 6 displays the model results using an assumed updraft velocity of 3 m/s. Agreement between Tgcalc and the observed Tg for samples Mar-1, Sah-2, and Sah-4 was achieved only after increasing the INP concentration by 104. Samples Sah-3 and Sah-4, which had the minimal ΔT values, reached their Tg values and uncertainty, respectively, when the modeled INP concentration in the cloud droplets was multiplied by 100. For samples Ant-1 and Ant-2, the model could not account for the observed Tg, even when the INP concentrations increased by a factor of 104. Previous studies that compared chemical concentrations of inorganic and organic species in cloud and rainwater showed that the in-cloud concentrations were higher by up to two orders of magnitude (Collett et al., 1993; Gioda et al., 2008, 2011, 2013). This suggests that simulating ice production with factors of 103 and 104 overestimate primary ice production by assuming unrealistic aerosols concentration in the cloud. In addition, the INP efficiency in the model is also enhanced by assuming that all INPs are pure K-feldspar particles. This further implies that primary ice production does not explain the observed Tg and that SIP must play an important role in these cloud systems.

Details are in the caption following the image
Glaciation temperature calculated from models (Tgcalc) using either Jhet (full symbols) or Ns (empty symbols) compared with Tg retrieved by remote sensing of the clouds. The updraft velocity is assumed to be 3 m/s (Cotton & Anthes, 1989) and lapse rate = 6.5° C/km. The different shapes and colors represent different ice nucleating particle enrichments in the model input compared to the sampled rainwater, as detailed in the legend. The error bars are confidence intervals for the calculated value (α = 0.003). The black diamonds are the derived Tg from the satellite, with error bars as detailed in Figure 5. Sensitivity test of Tgcalc using Jhet on the updraft was performed for three updraft speeds (Figure S12). On the upper axis, the particle concentration in the rainwater is presented, and on the lower, the sample names are detailed.

Cloud seeding with silver iodide (AgI) took place during sampling as part of Israel 4 seeding experiment. However, we did not simulate in our model the AgI ice formation because its concentration was too low to cause glaciation through primary ice nucleation. The Ag concentration in the rainwater samples was on the order of 3 × 10−11 g/L, which is equivalent to 6.5 × 10−11 g/L of AgI. To get 1 L of water from drops with radius of approximately 15 μm (Table S3), ~7 × 107 drops are needed, following the model assumptions. If we assume that the AgI is distributed evenly between all of the drops, it would imply that each drop contained ~9 × 10−22 g of AgI (0.16 nm3; spherical particle with D = 0.34 nm). Such particle mass is far too low to act as an INP (DeMott et al., 2010). Furthermore, even if we assume that in order to achieve cloud glaciation the ice fraction in the clouds needs to be 10% (as was assumed in the model) and increase the AgI concentration by 106, the particles will still be too small to act as INP. This suggests that even if AgI played a part in cloud glaciation, some secondary ice production had to occur.

4 Conclusions

The findings from the combined field sampling, remote sensing, laboratory experiments, and modeling exercise suggest an important role for SIP in the glaciation of midlatitude continental clouds. By relating the mineralogical analysis of the samples to their ice nucleation activity, we showed that the ice nucleation efficiency is mainly controlled by the K-feldspar content. However, aging processes that have a minor effect on the mineralogical composition may change the ice nucleation efficiency.

We identified large discrepancies between the retrieved Tg from remote sensing and the measured T50 from the WISDOM ice nucleation experiments. By comparing the differences between Tg and T50 with the particle concentration in the rainwater and droplet effective radii, we showed that these discrepancies may be attributed to SIP. Although aerosol concentration has a great effect on SIP efficiency, we suggest that SIP can be better assessed by observing the droplet size distribution at the temperature range where SIPs are most active (Re-5). This is because SIP can be important even under high dust concentrations, possibly due to the presence of giant CCN that can form large droplets and/or increase in rime splintering process. Due to the proximity of the sampling site to the Sahara desert and other dust source regions, the aerosols and the dust concentration in the analyzed area is higher than the concentrations expected over many other parts of the world (Atkinson et al., 2013; Karydis et al., 2016; Tanaka & Chiba, 2006). This implies that the expected INP concentration may also be high (Atkinson et al., 2013; Demott et al., 2003; DeMott et al., 2010), thus suggesting that the SIP role in glaciating clouds in this study is possibly smaller compared to other locations. However, further studies are needed in order to conform and quantify this.

The results of this study demonstrate that SIP are not limited only to clouds that form in pristine environment such as marine, polar, and convective clouds (Field et al., 2017) but can also be important in orographic-triggered, convective midlatitude clouds. We propose that the effect of SIP on Earth's radiation budget and its influence on global climate are larger than commonly assumed. Quantification of INPs has generated much attention (Field et al., 2017), but it is very likely not sufficient for describing the aerosol indirect effect on cold clouds. Equally important is developing a fundamental understanding of the governing parameters that drive secondary ice production in order to improve our predictive understanding of cloud radiative properties and the hydrological cycle.


This work was partly funded by the Ice Nuclei Research Unit (INUIT) of the German DFG, the Water Authority of Israel (grant 4500962318,) and Israel Science Foundation (grant #213/16). We thank The Helen Kimmel Center for Planetary Sciences and the De Botton Center for Marine Science. We also acknowledge the support from the U.S. Department of Energy, Office of Science (BER), Atmospheric System Research (DE-SC0016370). Satellite images can be downloaded from the EUMETSAT Earth Observation Panel (https://www.eumetsat.int/website/home/Data/DataDelivery/DataRegistration/index.html). Air mass back trajectories are available online (https://ready.arl.noaa.gov/hypub-bin/trajtype.pl?runtype=archive). All of the chemical data, ice nucleation data, and analysis codes are available for download at: https://weizmann.alma.exlibrisgroup.com/view/delivery/972WIS_INST/1239256720003596. Further requests should be addressed to [email protected] and\or [email protected]. We also thank Tony Yovel for GIS services, to Susan K. Berhani for editing the paper, to Ofir Tirosh for chemical analysis of the rain samples, and to Hilla Blouch and Asi Fishman for helping with the sampling and lab work.