The WeIzmann Supercooled Droplets Observation on a Microarray (WISDOM) and application for ambient dust

2.1 Droplets production and trapping

The WISDOM setup, shown in Fig. 1, is made of a microfluidic setup which include a pressure-controlled pump with four independent flow channels (OB1 MK3 by Elveflow), a stereoscope (SMZ-171 by Motic) that permits a full view of all channels and inlets, and a charge-coupled device (CCD) camera (GS3 by Point Grey) that enable real-time monitoring of the droplets production. The flows in the channels are continuous and controlled by the pressure pump. One channel is connected to the continuous (oil) phase, and a second channel contains the sample (aqueous solution that can contain INPs). The two phases meet in a narrow junction where monodisperse droplets are generated due to the pressure exerted by one phase over the other. The ratio between the flows determines the size of the emerging droplets; the volume increases with increasing flow rate of the sample. In this setup, droplets are suspended in an oil mixture, consisting of mineral oil (Sigma Aldrich) and a 2 weight percent (wt %) nonionic surfactant (span80^®, Sigma Aldrich), added for droplets stabilization (Riechers et al., 2013). Hence, an array of picoliter (micrometer-size) droplets is generated directly on a device.

Figure 1The WISDOM setup. (a) The design of the microfluidic device is based on Schmitz et al. (2009). Aqueous solutions (including the sample) and oil are connected through the inlets and merge in a junction to generate monodisperse droplets. Subsequently, droplets flow into a trap array and settle in them as the flow is stopped. The device is transferred into a cooling stage for subsequent freezing experiments. (b) Upper and (c) side views of the device, which is made of PDMS (polydimethylsiloxane), plasma glued to a microscope glass slide, placed over the cooling silver block.

Download

The principle of the Schmitz et al. (2009) design is that the droplets flow into round chambers that are connected by the constriction channel. At a certain flow, the droplets are squeezed through the constriction channel and the array fills up with droplets. When the flow is too weak or stopped, the constriction channel stops the droplets' movement and they are trapped in the chambers. The droplets are isolated and stable in the chambers and it is safe to move the device from the generation stage to the cold stage for the freezing experiments.

Devices are fabricated following the Schmitz et al. (2009) protocol. Briefly, the device pattern is imprinted on a polydimethylsiloxane (PDMS) polymer, later glued to a 1 mm thick microscope glass slide using air plasma treatment. After the plasma treatment the PDMS surfaces are hydrophilic (Eddings et al., 2008). Therefore, the devices were used only in the following day, after their surfaces became hydrophobic, following their exposure to the atmosphere, or after their annealing at 60 ^∘C for 30 min.

2.2 Freezing experiments and detection

The droplets array is placed in a commercial cryostage (Linkam, THMS600) coupled to an optical microscope (Olympus, BX-51 with 10x magnification, transmission mode). Experiments are monitored by a microscope-mounted CCD camera (Allied Vision Technologies, Oscar F-510C) for automatic identification of droplets and their freezing events. Both the device and the cooling stage are cleaned with 2-propanol. Then the device is placed over the stage together with a thin layer of oil on its bottom to provide good thermal conductivity. Each freezing experiment starts with dry N₂ purging to replace the moist atmosphere inside the cryostage to prevent condensation. During the experiment, N₂ flow prevents water condensation on the cryostage window. Freezing experiments are conducted with a cooling rate of 1 K min⁻¹, which is relevant for atmospheric conditions and also allows good thermalization of the droplets, as will be shown in the calibration section (Sect. 3.1). Each cooling cycle is followed by a heating cycle, where melting is observed. Analysis of the melting onset is then used to verify that the thermal conductivity is good and thus validate the measurement.

In-house LabVIEW software is used to record a freezing experiment movie file and analyze it offline. The temperature readings by the Linkam cryostage temperature sensor ($<\pm$0.25 K for the operated temperature range) and the movie frames are synchronized and integrated. In most cases, 1 s (or 0.017 K at 1 K min⁻¹) per frame is used. Currently, the WISDOM setup operates with two types of devices that differ in their droplets' trap diameter: 40 and 100 µm. Approximately 550 and 120 droplets can be monitored per experiment in the smaller and larger diameter devices, respectively. Statistically, for the same sample, larger droplets encompass more INP surface area within each droplet, which can be more sensitive for detecting rare active sites. The device can be reused for the same sample, if it is not clogged or destroyed during the experiment. However, because the channels of the 40 µm device are smaller they tend to clog faster (for instance by large particles).

2.3 Automatic detection of phase transitions

The optical brightness of a droplet changes during a phase transition (freezing or melting) due to the different interaction of light with the liquid and the solids. For phase transition detection, an in-house image processing LabVIEW program monitors automatically the optical brightness change. The program detects the droplets using a spherical shape criterion and sets a square surrounding the droplet that defines an array of pixels that are attributed to that specific droplet. A change in the optical brightness is represented by the gray level value of the image's pixels, ranging from 0 to 255. Freezing is calculated per movie frame and is defined as the subtraction of the brightness mean value for each droplet in two consecutive frames (ΔGL), thus allowing derivation of freezing rates. At the beginning of the analysis, the first 15 frames are used to identify the noise level of the signal by calculating its standard deviation SD (ΔGL). The program then searches for the maximal freezing signal that is also greater than 5 times the noise level. The temperature associated with this freezing signal is assigned as the freezing temperature for that droplet.

In this algorithm, the program can distinguish successfully between a phase transition event and noise that arises from the camera signal, droplet movement or any other interruption. Figure 2 presents a spectral analysis for different types of phase transitions observed in WISDOM. Since WISDOM operates in transmission microscopy mode, the light is scattered more efficiently by ice crystals in comparison with a liquid droplet and a freezing event involves droplet darkening and a negative signal. The negative signal of a freezing event of a single droplet is shown in Fig. 2a. In comparison, during melting, the droplet becomes brighter until all the crystals melt and the signal is positive. In Fig. 2b and c the analysis of a melting signal and a eutectic melting signal are presented for the entire frame.

Figure 2Spectra of different phase transition events as observed in WISDOM. (a) Freezing, (b) eutectic melting, and (c) melting onset and clear point (liquefaction) are the mean of all sampled droplets in a single experiment. The phase transition is defined optically by the brightness information obtained by the gray level of the image pixels. SD (ΔGL) describes the standard deviation of the difference in mean GL for two consecutive frames. At the beginning of the experiment the noise level is studied and freezing or melting are detected only if SD (ΔGL) is as least 5 times greater than the noise standard deviation level. Freezing and melting examples are for pure water droplets and the eutectic melting example is for aqueous solution droplets of NaCl. The eutectic melting point of NaCl and pure water melting point are marked by the yellow and red lines in (b) and (c), respectively. In all cases the droplet diameter was 100 µm.

Download

3.1 Temperature calibration

Temperature accuracy is a most important parameter in ice nucleation experiments. An error propagation analysis by Riechers et al. (2013) demonstrated how the temperature uncertainty may lead to a distribution of temperatures between different instruments. Therefore, we performed a thorough temperature calibration using the known eutectic melting points and the melting points of several aqueous solutions as calibration reference points. Although ice nucleation experiments are performed while cooling, the calibration experiments were done while heating to improve the calibration precision and to avoid biases associated with supercooling of the liquids (Budke and Koop, 2015).

3.1.1 Droplets thermalization

The temperature of the Linkam stage was measured at the upper center part of the cooling stage and hence may differ from the actual temperature of the droplets in the device due to thermal effects such as temperature gradients and temperature lag. During cooling or heating, a vertical temperature gradient may develop between the top of the device, in contact with the inner ambient of the cryostage, and the bottom of the device, which is in contact with the cooling silver block. This gradient is expected to increase in magnitude, as the temperature of the stage decreases or increases below or above ambient temperature. Edd et al. (2009) used a similar setup and found a difference between the top temperature and the bottom temperature of about 2 K around 237 K and 3 K around 227 K. Stan et al. (2009) also reported a vertical gradient of 1–2 K, which was reduced to 0.5 K with a flow of cooled N₂ over their device. In addition, a thermal lag may arise during cooling or heating as the rate of temperature change is high and precludes proper temperature equilibration. Hence, a more accurate measurement of the droplet temperature is taken as a sum of the stage temperature with the contributions of both thermal gradient and lag.

Figure 3 demonstrates the combined effects of the temperature change rate and device properties on the thermalization of pure water droplets (double deionized water, DDW; 18.2 MΩ⋅cm). Specifically, freezing and melting experiments at different rates were performed. The temperature difference (ΔT) is the difference between the measured values and the extrapolated temperature at equilibrium conditions (0 K min⁻¹). As expected, at slower temperature cooling (heating) rates, the droplets are more equilibrated with the stage temperature and ΔT is negligible. However, ΔT increases at higher temperature cooling (heating) rates (e.g., 10 K min⁻¹). We observed that during cooling (heating) the droplet is warmer (colder) than the stage and will freeze (melt) at colder (warmer) temperature at higher cooling (heating) rates. We also found that because ΔT is higher, in absolute value, for devices of thicker PDMS and/or in devices which hold larger droplets, it should be considered in the final temperature calibration for these scenarios. Furthermore, ΔT was found to be almost symmetric for higher temperature cooling (heating) rates. However, for 1 K min⁻¹, ΔT during cooling is higher than that for heating. Our conjecture is that this can be an effect of the higher thermal gradient that develops as the temperature decreases well below ambient (236 K).

Figure 3The temperature difference (ΔT), defined as the temperature difference between the stage temperature and the droplet extrapolated temperature at equilibrium conditions at different cooling (heating) rates. Freezing and melting points of pure water are represented by circles and squares (40 and 100 µm droplet diameter, respectively) for different PDMS thicknesses and are represented by different colors. C denotes cooling and H denotes heating. Droplets are close to equilibrium with the stage temperature at rates <0.1 K min⁻¹ and ΔT increases with increasing temperature change rate and with the PDMS height.

Download

3.1.2 Melting of aqueous solutions

Figure 4 presents the measured melting points of NaCl solutions with different water activities. Reported melting points represent the temperature in which all ice crystals in the droplets completely melted, in contrast with melting temperatures reported for pure liquids such as water, where the onset of melting is defined as the melting point. Melting temperature results were consistent with theoretical melting temperatures reported in Koop and Zobrist (2009). This provides support to our conclusion that droplets thermalize with the cooling stage when using a heating rate of 0.1–1 K min⁻¹. For faster heating rates (i.e., 10 K min⁻¹), the thermal lag was more pronounced, leading to a melting point shift of about 2–3 K. For more concentrated solutions, faster heating rates shifted the melting points more.

Figure 4Temperature calibration by melting points of eutectic solutions and pure water droplets for different heating rates. Calibration is presented for 100 µm droplets with 4 mm PDMS thickness. The onsets of pure water droplets are also considered. Eutectic melting is used for the colder temperature range (<253 K) while clear point (liquefaction) at various water activities is taken for the warmer temperature range. The upper panel presents the temperature difference between the reference value and the cooling stage temperature after calibration. Most of the differences are within the range ±0.2 K.

Download

3.2 Measurement reproducibility and device variability

The device's inter-variability was determined from 20 devices by comparing their corresponding homogeneous freezing temperatures of pure water. Specifically, each device was recycled three times with freshly prepared droplets. Our results showed high reproducibility in the median freezing temperature, where 50 % of the probed droplets froze (T₅₀), and high reproducibility in the melting point temperature. Variation within devices was always smaller than ±0.2 K at 1 and 0.1 K min⁻¹ (variation within the devices over the whole freezing range is presented in Appendix A).

3.3 Homogeneous freezing rates of pure water

Homogeneous nucleation in supercooled water occurs in WISDOM between 238 and 237 K for a cooling rate of 1 K min⁻¹ and droplet diameter of 100 µm. Figure 5 shows WISDOM nucleation rates in comparison with other similar instruments. It is seen that the slopes of the rate and temperatures are similar to the slopes reported for other instruments. The temperature where 50 % of droplets froze (T₅₀) is also in the expected range according to model results of Hoffer (1961). WISDOM rates are slightly slower, but within the uncertainty of the instruments used by Stan et al. (2009) and Riechers et al. (2013). Stöckel et al. (2005) show a higher nucleation rate. This discrepancy can be explained by a decrease in the number of surface nucleation events due to the oil phase surrounding our droplets, whereas in Stöckel et al. (2005) droplets are suspended in air which allows surface nucleation to occur.

Figure 5The volume-dependent homogeneous freezing of pure water, derived for 100 µm droplets with 4 mm PDMS height. WISDOM rates are compared to relevant literature data. The obtained fit from WISDOM is $J_{v\left(T\right)}=\exp (-\mathrm{3.4}T+\mathrm{817.6})$. Temperature uncertainty for WISDOM is ±0.3 K.

Download

Figure 6Homogeneous and heterogeneous ice nucleation temperatures for 40 and 100 µm aqueous solution droplets as a function of solution water activity. Freezing and melting curves are derived from Koop et al. (2000). Heterogeneous ice nucleation is performed with 0.1 wt % of Arizona test dust (ATD) particles immersed in the droplets.

Download

3.4 Homogeneous and heterogeneous freezing of aqueous solutions

The water-activity-based ice nucleation theory by Koop et al. (2000) describes the dependence of the freezing temperature depression on the water activity (a_w) of the solution, regardless of the solute nature. Figure 6 presents the theoretical freezing and melting temperature curves from Koop et al. (2000) with homogeneous ice nucleation results measured in WISDOM, for four solutions with atmospheric relevance. Water activities for NaCl, ammonium sulfate (AS), glucose and levoglucosan mixtures were derived from the AIM model and were corrected for glucose and levoglucosan, for which water activity is temperature dependent (Zobrist et al., 2008; Knopf and Lopez, 2009). The experiments were conducted at 1 K min⁻¹ for 40 and 100 µm droplet diameters. The results follow the theoretical curves of the water-activity-based ice nucleation, and the dependence of the homogeneous freezing on the droplet volume is as expected (Hoffer, 1961; Kuan-Ting and Wood, 2016) as the curve of the smaller diameter droplets (green curve) is slightly colder compared with the larger volume droplets (dark green curve).

Figure 7Accumulated active site density spectra (n_s) of K-feldspar and illite-NX particles as a function of temperature from validation experiments of immersion freezing in WISDOM. Frozen fraction values are represented by a color bar for a few surface area values that are exposed in 40 and 100 µm droplets. The dependence of the nucleation site density on the surface area is illustrated here. WISDOM uncertainties, propagated from surface area estimation and measured frozen fraction errors, are included within the size of the markers. For validation, previous immersion freezing measurements are also presented (Hiranuma et al., 2015 and Atkinson et al., 2013). T-binned data (1 ^∘C) normalized by the Brunauer–Emmett–Teller (BET) surface area from Hiranuma et al. (2015) are presented in the color squares only for wet suspension analysis Bielefeld Ice Nucleation ARraY (BINARY), red; Colorado State University Ice Spectrometer (CSU-IS), orange; Leeds Nucleation by Immersed Particles Instrument (Leeds-NIPI), purple; Mainz acoustic levitator (M-AL), green; Mainz vertical wind tunnel (M-WT), black; North Carolina State cold stage (NC State-CS), brown; and University of Colorado Raman microscope cold stage (CU-RMCS), blue. The Hiranuma et al. (2015) log fit and n_s (BET) parameterization are also presented.

Download

Similar experiments were conducted for 0.1 wt % of Arizona test dust particles (ATD, Powder Technology Inc.) immersed in glucose solution droplets. The ATD particles facilitate the ice nucleation at warmer temperatures, in agreement with similar studies (Niedermeier et al., 2010; Hartmann et al., 2011), and the freezing depression follows the water-activity-based ice nucleation curves. Here, the dependence of the freezing point on the droplet volume is more pronounced, as the surface area of the immersed particles is higher; hence, they contain higher number of nucleation sites (Marcolli et al., 2007) as will be shown in the next section for two more types of dust.

Below 223 K, ice nucleation occurs at slightly lower temperatures than expected by the theoretical freezing curve. As the WISDOM temperature calibration is not valid in this temperature range, we cannot conclude if this is due to a change in the thermal conductivity of the device or an effect of the high concentration of the solute in the water.

Table 1Summary of immersion freezing experiments performed for WISDOM validation.

Download Print Version | Download XLSX

Figure 8Accumulated active site density spectra (n_s) of ambient super-micron mineral dust particles collected in Israel during the Saharan dust event in 2017, for three different sampling stages of the micro-orifice uniform deposit impactor (MOUDI); D₅₀ of 1, 1.8 and 3.2 µm.

Download

3.5 Heterogeneous nucleation and n_s spectra of INP in pure water

3.5.1 Standard dust powder

Heterogeneous freezing efficiencies of suspended mineral dusts K feldspar and illite NX in supercooled water droplets are presented in Fig. 7 and summarized in Table 1 and are compared to recent published data. The particles are suspended at different wt % and the frozen fraction of each suspension is derived as a function of temperature as represented by the color bar. To examine the freezing efficiency and compare the different mineral dust types, the results are normalized to the surface area within each droplet. Experiments were performed at 1 K min⁻¹ for 40 and 100 µm droplet diameters. Suspension preparation and evaluation of the surface area are described in Appendix B.

Figure 9Accumulated active site density spectra (n_s) of ambient super-micron mineral dust particles collected in Israel during the dust event in 2017 for three MOUDI stages that were analyzed with D₅₀ of 1, 1.8 and 3.2 µm. The fit $\ln \left(n_{\mathrm{s}}\right)=-\mathrm{005}{T}^{\mathrm{2}}+\mathrm{2468}T-\mathrm{293}$ (R²=0.98) is also presented. References of K-feldspar standard particles activated in WISDOM, Leeds-NIPI (Atkinson et al., 2013) and Leipzig Aerosol Cloud Interaction Simulator (LACIS; Niedermeier et al., 2015) instruments are presented, as well as ambient dust particles that were analyzed in AIDA and included Israeli dust (Niemand et al., 2012).

Download

The results demonstrate the effect of dust surface area immersed in the droplets on the freezing parameters. The freezing temperatures increase with increasing surface area and are also reflected in the warming of the median frozen fraction (T₅₀) colored in yellow. The spectra of the number of nucleation sites per unit surface area (n_s) also support surface area dependence because all spectra converge to a single line. The n_s results show the increase of nucleation sites at colder temperatures. Results from WISDOM are in good agreement with similar analyses from other instruments. In particular, n_s is in best agreement with the Leeds-NIPI (Murray et al., 2011; Broadley et al., 2012) results both for K-feldspar and for illite-NX particles. Results of illite-NX particles are also in good agreement with the BINARY instrument (Budke and Koop, 2015) and reside within the uncertainty of both instruments. The linear trend of a few weight percent values supports the assumption that particles in suspension are uniformly distributed and the droplets contain approximately the same surface area.

3.6 WISDOM in comparison to other cold stage instruments

The microfluidics technology used in WISDOM solves some substantial issues inherent in other currently used instruments. (1) It allows for good control of the size and number of monodisperse droplets. (2) Additionally, it also allows for the fast production of hundreds of nearly monodisperse droplets, which minimizes sample sedimentation or agglomeration that may occur in a suspension, leading to a good estimation of the surface area of the suspended material. Moreover, several droplet diameters can be employed in the same device without its modification, which allows for (3) good statistics achieved by the individual analysis of hundreds of droplets; (4) the individual analysis of monodisperse droplets, in contrast to some emulsion techniques (such as a differential scanning calorimeter, DSC), which allows for obtaining the frozen fraction at each temperature and acquiring detailed information about active sites and freezing rates. (5) The use of oil minimizes possible artifacts from the droplets' evaporation, neighbor seeding or vapor transfer due to the Wegener–Bergeron–Findeisen processes. (6) The small volumes decrease freezing artifacts by impurities, thus allowing us to reach the homogeneous freezing threshold (−37 ^∘C). (7) The static droplet array allows the recycling of the droplets for multiple freeze–thaw cycles. (8) The microfluidics method and the small droplet volumes enable working with small sample volumes which can be an advantage when working with atmospheric samples.

WISDOM has a very accurate temperature calibration that spans a wide temperature range, using the eutectic freezing method. WISDOM most resembles the instrument used by in Edd et al. (2009). However, it seems that issues with temperature calibration in Edd et al. (2009) led to a temperature offset and hence different freezing rates. Stan et al. (2009) achieved better temperature accuracy and high statistics. However, the freezing experiment was conducted in a flow mode, which is more complicated than in the WISDOM setup and requires complicated modeling. In addition, the cooling rates that were used were very fast, which induces additional errors. Riechers et al. (2013) had high temperature accuracy as they also used a DSC. However, they had to collect the droplets from the device as there was no static array option and this may add further complication and contamination.

The microfluidics technology also has disadvantages. These may include the following: (1) oil may interact with some of the analyzed particles, possibly leading to biased data; (2) the microchannels are susceptible to clogging; (3) it is not possible to perform any post analysis to the droplets' content after the experiment; (4) the small droplets' volumes reduce the sensitivity to rare active sites. This may be solved by performing many experiments or by using larger droplets with more surface area within the droplets.