Cloud formation in the Pi Convection–Cloud Chamber is achieved via ionization in humid conditions, without the injection of aerosol particles to serve as cloud condensation nuclei (CCN). Abundant ions, turbulence, and supersaturated water vapor combine to produce new particles, which grow to become CCN sized and eventually are activated to form clouds. Coupling between the new particle formation and cloud droplets causes predator-prey type oscillations in aerosol and droplet concentrations under turbulent conditions. Leading terms are identified in the budgets for Aitken and accumulation mode aerosols and for cloud droplets. The cloud coupling is proposed to be a result of cloud-induced runaway CCN production through aerosol scavenging. The experiments suggest potential applications to marine cloud brightening, in which ions rather than sea-salt aerosols are generated.
]]>Climate change is one of the defining challenges of our age, and the problem of understanding such a complex, high-dimensionality system has implications for predictability, assessing uncertainties, and evaluating potential mitigation approaches such as solar radiation management. One of the central components of the climate system, and ironically, a key part of mitigation approaches like marine cloud brightening, is the coupled aerosol-cloud-precipitation system and its interaction with boundary layer structure and turbulence. Aerosols, the precursor of clouds, are a critical dial that nature [1, 2] and humans [3,4] can turn to modulate clouds and adjust climate forcing. The problem of marine cloud brightening in particular begs for a better understanding of the aerosol-cloud interactions, including feedbacks to the sources and properties of those aerosols, in order to determine how and whether such a sensitive dial should be touched [5]. This article presents a process study of the impact of new particle formation through ionization in clouds and its implications for a potentially new approach to cloud modulation inspired by nature.
The cloud-topped atmospheric boundary layer is a system with constantly interacting ions, aerosols, and clouds. Examples of the interactions include cloud cleansing and cloud collapse through precipitation and scavenging [6][7][8]. Aerosols are continuously produced mechanically, by human-caused emissions, and through gas-to-particle nucleation. Furthermore, evidence continues to point to clouds and precipitation as playing an important role in new particle formation and eventual production of cloud condensation nuclei in cloud-cleansed environments [9][10][11]. This includes local production in a clean marine boundary layer [9,11] or new particle formations in the upper troposphere through detrainment of deep convection in tropics and their supply of CCN to the lower troposphere through descent [10]. An additional dimension to this problem is the role of ions in the atmosphere, and the coupling between ions and aerosol/cloud microphysics that occurs in multiple processes [12,13]. Ion-induced new particle formation in the atmosphere has long been speculated to explain sudden spikes of observed ultrafine new particles [14][15][16]. Recent controlled lab experiments also confirm such enhancement in aerosol nucleation in the presence of ions [17]. These studies suggest that in an environment where neutral clusters of molecules are unstable, charge is necessary to stabilize them. There also has been a suggestion of enhanced condensation growth of aerosol in the presence of ions [18]. In nature, galactic cosmic rays are one of the sources for ionizing air and producing new aerosols [12]. Such enhanced production of aerosols by nature, apart from other natural and anthropogenic sources, could be a critical but less constrained source of climate perturbation. Thus, a question arises-can ioninduced new particle formation impact clouds? In other words, can the ultrafine aerosol generated from ion-induced nucleation grow large enough to act as cloud condensation nuclei (CCN)? The production of CCN from ultrafine aerosols is still highly uncertain [19]. Since the chemistry of ion-induced new particle formation is extensively studied in recent laboratory experiments [17,20], the current study attempts to answer this question with the focus on physical aspects of the coupled ion-aerosol-cloud system.
We performed a controlled laboratory cloud-chamber investigation to inquire if the ion-induced new particle formation has the potential to generate enough CCN to impact clouds significantly. We use a convection-cloud chamber, which has the advantage of allowing the ion-aerosol-cloud system to be studied in a dynamic steady state, in which sources and sinks of all three are balanced, at least on average. The experiments are unique: abundant experimental studies of ion-induced nucleation and aerosol growth exist, as well as studies of cloud formation with aerosols as a controlled input. Here, we observe the fully-coupled system, with ionization leading to new particle formation, an evolving aerosol distribution, and finally activation to form cloud droplets. This allows us to explore the phenomenology of interacting ion, aerosol, and cloud populations, and therefore to answer several key questions. Can CCN generation through an ioninduced aerosol generation be fast enough to allow for cloud formation? How does ionization influence a pre-existing cloud field generated from efficient CCN (e.g. salt)? In this study, we focused on the physical aspect of the impact of new particle formation on clouds instead of the chemistry of particle formation, which requires a completely different setup. Thus, it complements previous laboratory studies of the chemistry of ion-induced new particle formation [17,20] and provides additional perspective on their implications for cloud physics.
The connection to marine cloud brightening is to suggest an alternative to the direct production of aerosols through pumping of seawater. The experiments reported here explore, in a highlyidealized system, whether ion-induced new particle formation could be used as an aerosol source. Producing ions is technologically straightforward and requires very low energy input. How the newly-formed ultrafine aerosol particles evolve in size to become large enough to serve as cloud condensation nuclei is a key question, and these laboratory observations suggest an efficient mechanism. Questions of what chemistry would be operating in the natural marine boundary layer for new particle formation, and how the laboratory results can scale to the full cloud-topped boundary layer are fascinatingly complex, and necessarily remain the focus of future work. For now, the processes investigated here simply provide a physically-robust but conceptual model for the role of ion-induced new particle formation in cloud formation in the natural environment, and finally its possible use in marine cloud brightening.
We begin the paper by describing the experimental setup, and then present results for the four scenarios that were explored: ion generation in dry, saturated, and supersaturated environments, the latter with and without external aerosol injection. We then consider budgets of ions, aerosols, and cloud droplets, informed by the observations to determine the dominant source and sink terms. Based on the results and the budgets, a proposed mechanism for rapid formation of CCN is proposed, and implications for atmospheric clouds and marine cloud brightening are discussed. We conclude by summarizing open questions and suggestions for next steps.
Experiments are conducted in the Pi Convection Cloud Chamber, a thermally controlled chamber operating based on Rayleigh-B� enard convection with cylindrical sidewalls. The details of the chamber and its operation have been extensively discussed in past literature [21,22]. For this study, the experiments are performed in four different configurations: (i) dry convection (RH < 75%) driven by 10K temperature difference between top and bottom boundaries with a steady aerosol injection, (ii) no convection with water saturated walls at 10 � C and no external aerosols, (iii) convection driven by the 10K temperature difference between water-saturated top and bottom boundaries (wet sidewalls at the mean temperature of the top and bottom boundaries) without external aerosols, and (iv) same as the previous configuration but with external aerosol injection after reaching a steady-state moist condition. The turbulent mixing with saturated boundaries at different temperatures produces supersaturated condition for droplet activation and cloud formation. The external aerosols (NaCl particles) for injection are formed by atomizing salt solution using an atomizer (from TSI Incorporated). In all these experiments, five ion generators (applying 7.5 Kv voltage with a minimal current of 35 mA at discharge needles) were turned on at different chamber locations at some point to ionize air inside the chamber. According to the specification, each ion generator produces 2 trillion ions/s. A scanning mobility particle sizer (SMPS) from TSI Incorporated is used for aerosol measurements, and a Phase Doppler Anemometry system from Dantec Dynamics is used for cloud droplet measurements.
In order to understand ion-aerosol-cloud interactions, four experimental scenarios of increasing complexity were explored. (i) Ions were generated within a dry chamber and it was confirmed that no new particle formation was observed. (ii) Ions were generated in a water-saturated chamber with no turbulent convection. (iii) Ions were generated in a water-supersaturated environment with turbulent convection. (iv) Ions were generated in a water-supersaturated environment with turbulent convection, and with a steady injection of background CCN to enable cloud formation. Please note that these experiments used laboratory air (after aerosol removal) without controlling or monitoring the gas composition.
Figure 1 shows the time evolution of the aerosol number and size distribution from an experiment without turbulent convection and external aerosol injection but in a moist environment (RH ≤ 100%). The initial condition of the chamber was relatively clean with aerosol concentration � 10 cm -3 when measured for an hour. Soon after starting the ion generators, the Aitken mode aerosol concentration shoots up and approaches a concentration of � 10 4 cm -3 . The timescale of this Aitken mode aerosol formation is less than 10 min. However, after the initial sharp rise in the aerosol concentration, the loss mechanisms, wall and coagulation losses, and increased competition for precursor gases suppress the increase in the aerosol concentration. The coagulation of Aitken mode aerosols gives slow rise to the accumulation mode aerosols. The rate of increase is extremely slow in this no cloud condition (from � 2 cm -3 to � 53 cm -3 in 193 min). An obvious question, which we cannot answer with confidence given the measurement capabilities, is what gases are contributing to the new particle formation. A likely candidate is ammonia [23], which is abundant in the atmosphere. Other trace secondary organic compounds from laboratory sources may be possible. This will be explored in future experiments, preferably by collaborating with a group with access to an aerosol mass spectrometer so we can identify the composition of the aerosols, and also possibly by injecting precursor gases with known composition.
Around 275 min, a sharp jump in the Aitken mode aerosol formation (around four times increase) enhances the accumulation mode formation (a factor of sixty increase in 54 min). The increased formation of accumulation mode aerosols decreases the rate of Aitken mode aerosol formation, which decreases the Aitken and total aerosol concentration. This sudden increase in aerosol formation could be due to a slow mixing and availability of precursor gases, including water vapor, in this no-convection environment. Please note that in experiments in dry conditions (RH < 75 %), the ionization process does not increase aerosol concentration. Instead, they significantly accelerate their removal due to accumulated charges in aerosols through ionic diffusion and image charges when approaching a chamber wall [24] (see Fig. A1 in Appendix 1 for dry aerosol loss). Thus, the new aerosol formation due to ionization in the chamber depends on the availability of sufficient water vapor. The higher water vapor reduces ion loss through recombination (second term in Equation 1) due to ions clustering with water molecules and decreasing their mobility [25]. Moreover, water vapor is a key ingredient for watersulfuric acid binary nucleation.
An experiment with turbulent moist convection is performed to investigate if the new particle formation observed in the previous experiment without turbulence mixing can initiate and help sustain a cloudy condition. The turbulent mixing with saturated boundaries creates the supersaturated condition for the activation and growth of cloud droplets. The turbulent environment also helps homogenize ions and the precursor gases. In turbulent moist convection, supersaturation also fluctuates and causes multiple activation-deactivation cycles of aerosols depending on the mean and fluctuation magnitudes [26,27]. Figure 2 shows a time series of aerosol and cloud droplet concentration and aerosol size distribution from this experiment. Like the no-turbulence experiment, starting the ion generators in an initially clean chamber condition produces a burst of Aitken mode aerosols. These aerosols grow with time in a short time and produce a classical "banana event" of new particle formation and growth. It has a close resemblance to the new particle formation even in nature [16,28]. However, the growth rate is much faster (1.5 nm/min) than the atmospheric observations reported in past literature (e.g. 3.3 nm/h in [28]). A higher growth rate also indicates the influence of ions on condensation rate [18]. See Fig. A2 in Appendix 1 showing an example of the growth rate calculation. Shortly following that, the accumulation mode aerosol concentration increased, and cloud droplets formed and reached a detectable limit (�25 min from the start of the ion generators). Successive banana events following the first burst slowly build the concentration of accumulation mode and cloud droplets in the chamber. The cloud droplet concentration peak appears to follow the accumulation mode particles. There are oscillations in aerosol and droplet concentration, most likely due to the on all chamber surfaces. Due to the lack of temperature gradient, the humidity inside the chamber during the experiment was close to, but never exceeding 100%. Abundant production of Aitken particles occurs after the start of ion generation, through gas-to-particle nucleation. The Aitken particles slowly grow to populate the accumulation mode.
coupling of the populations of precursor ions, Aitken and accumulation mode aerosols, and cloud droplets through various loss mechanisms in this four-component predator-prey system. The above elements of the coupled system are interconnected through the nucleation of Aitken mode particles from precursor gases, the condensation of precursor ions and their clusters in Aitken and accumulation mode particles, coagulation of small particles to form accumulation mode, nucleation of cloud droplets, and diffusion of precursor gases and ultrafine aerosols into cloud droplets. There is also a loss of cloud droplets and aerosols to chamber surfaces due to inertia/gravity and diffusion.
After turning off the ion generation, the concentration of Aitken mode particles decreases. However, the accumulation mode particles increase for some time, most likely due to growth from Aitken mode through condensation, coagulation, and cloud processing. They also start decreasing after some time when activation loss to form new cloud droplets overwhelms the source of accumulation mode. Without sufficient accumulation mode aerosols, droplet concentration decreases sharply, further decreasing competition for water vapor, and droplets grow to larger sizes and fall faster. The rate of decay of the Aitken mode aerosols is the slowest due to the buffer of precursor ions that helps nucleate new aerosols in a progressively cleaner environment. The gap in the aerosol measurement is due to sampling cloud droplet residues through the PCVI (Pumped Counterflow Virtual Impactor) inlet to SMPS. The measured residual particles show a similar distribution shape as interstitial measured between the last concentration minimum to maximum but with suppressed Aitken mode aerosols (see Appendix 1). As expected, the accumulation mode aerosols primarily activate cloud droplets. But there is also a sign of activation of some smaller aerosols. Harrison and Ambaum [29] show that the critical supersaturation for activating cloud droplets decreases in the presence of charges in haze droplets. This helps account for the activation of some small Aitken mode aerosols in our experiments.
The new particle formation in a clean environment is evident in the previous experiment and can also be seen in nature. However, can ionization initiate new particle formation in preexisting clouds with extremely hygroscopic aerosols, e.g. sea salt? This question bears merit because of its implications for natural clouds and potential application for geoengineering. A controlled experiment where a steady-state laboratory cloud is created by constant injection of NaCl aerosols helps answer this question. Figure 3 shows results from this experiment. As evident, the cloud droplet and interstitial aerosol concentration (and the distribution) are approaching a nearly steady-state condition after the initial transient phase. After �220 min, ion generators are turned on. Immediately following this event, the aerosol concentration dropped suddenly, as observed in dry convection experiments (see Fig. A1 in Appendix 1). This short window of clean conditions provides a favorable opportunity for nucleating new aerosol particles from precursor ions. A burst of Aitken mode aerosols forms (a peak of �5505 cm -3 ) and grow to larger sizes. Following this burst of new particle formation, the cloud droplet concentration increases, and the interstitial accumulation mode concentration also rises following the rise in cloud droplet concentration. However, the Aitken mode aerosols start declining after cloud droplets and accumulation mode aerosols reach a peak. The cloud droplets and accumulation mode aerosols also decrease, creating aerosol and cloud droplet concentration minima. It gives rise to another clean environment window (relatively less clean than the first one). Consequently, another bust of new particle formation occurs, and this oscillation cycle in aerosols and cloud droplet concentrations continues. Therefore, the coupling between the new particle formation and cloud droplets exists even in the presence of external effective CCNs (salt particles).
The cloud droplet number concentration peaks during these new particle formation events are significantly larger than the condition before perturbing the cloud field through ionization. Thus, this ion-induced perturbation can significantly impact cloud microphysical and optical properties of preexisting clouds.
We now move beyond the qualitative discussion of results by considering phenomenological budget equations for ion-aerosol-cloud interactions in the four experiment configurations. In many cases, characteristic time scales for processes are not able to be determined from theory, and instead are obtained empirically from the observed transient response of the system. The motivation for this budget study is to determine what are the dominant terms in the production and loss of the various particle categories. Interactions between ions, Aitken aerosols, accumulation-mode aerosols (cloud condensation nuclei, CCN), and cloud droplets are considered. We note that ions are not equivalent to charge, in the sense that ions can be removed by scavenging to aerosols or droplets, but the charge still remains.
For the experiments with dry conditions and no turbulent mixing, we consider budgets of ions, of Aitken aerosols, and of accumulation-mode aerosols. Both positive and negative ions are produced by the corona source, and these in turn are lost through various processes: ion recombination, ion attachment to aerosol particles, and ions lost to the formation of aerosol particles through ion-mediated nucleation. The budget of ion concentration n ± is given by Ion-induced cloud modulation | 5
where, q, α, β j , n a ðj; r a Þ, and τ f ;ait are the ion source, ion recombination coefficient, ion attachment coefficient to a particle of radius r a and carrying charge j, joint concentration distribution of aerosol size and charge carried by the aerosol, and the aerosol formation timescale, respectively. Aitken mode particles are generated by ion-mediated nucleation and are lost by coagulation, by condensation growth to the accumulation-mode, and by diffusive losses to chamber walls. The budget of Aitken mode aerosol concentration n ait is therefore written as
where, n i , c 1 , τ coag , τ cond , and τ wall;ait are the total ion concentration, an average number of ions nucleating an aerosol particle, the timescales for coagulation loss, the timescale for condensation growth of the Aitken mode aerosols to the accumulation mode, and the timescales for loss to the surface, respectively.
Particles in the accumulation mode are generated through the coagulation of Aitken particles and through the condensation growth of Aitken particles, and are lost by diffusion to cloud chamber walls. A budget of accumulation mode aerosols concentration n acc can be written as
where c 2 and τ wall;acc are the average number of Aitken mode aerosol forming an accumulation mode aerosol through coagulation and the loss timescale of the accumulation mode aerosols to surfaces, respectively. Together, the condensation and coagulation sources more than offset the wall loss, as indicated by the slow rise of the accumulation mode concentration in Figure 1.
In the experiment where cloud formation is possible, new terms appear in the budget equations. Specifically, for ions it is necessary to include the diffusive loss to charged cloud droplets. The budget of ion concentration becomes
(4) where β j ðr d Þ and n d ðj; r d Þ are the ion attachment coefficient to a droplet of radius r d and carrying charge j, and the joint cloud droplet number concentration distribution of droplet radius and charge carried by the droplets, respectively. Aitken mode particles can now be lost either by diffusion to charged cloud droplets, or by being activated to form cloud droplets, although the latter is unlikely to be significant when cloud droplet concentrations are large enough to suppress the supersaturation. The budget of Aitken mode aerosol concentration becomes
(5)
where τ diff and τ act are the timescales for the diffusion loss to cloud droplets and activation loss, respectively. Finally, accumulation mode particles can be lost by diffusion to charged cloud droplets, by activation to form cloud droplets, and can be generated through the deactivation of cloud droplets. The budget of accumulation mode aerosol concentration is therefore written as
where τ diff ;acc , τ act;acc , and τ deact are the timescales for the diffusion loss to cloud droplets, the loss by activation, and the source from droplet deactivation, respectively.
In addition, we now include a budget equation for cloud droplets, with activation as the source, and with deactivation and gravitational settling as the loss terms. The budget of cloud droplet concentration in the absence of drop coalescence can then be written as
where τ act , τ deact , and τ sett are the activation, deactivation, and settling timescales, respectively.
When external aerosols are included in the system the equations are similar to those for the Aitken and accumulation particles in the cloudy case, i.e., Equations ( 5) and (6). The only addition is to add aerosol injection sources I ait and I acc to the corresponding equations.
We now simplify and interpret the budget equations by estimating the various time scales associated with aerosol and cloud processes. The estimates, provided in Table 1 either come directly from the observations in Figs 1, 2, and 3, or from theory informed by the observations. The first column of the table describes the process, the second column provides the symbols for the time scales described in the previous section as well as a parenthetical note as to how the time scales were estimated. The remaining columns list the estimated time scales in units of minutes, unless otherwise specified. The first simplification is in the Aitken-mode aerosol budget given by Equation (2). It can be further simplified by neglecting the condensation and coagulation growth to accumulation mode due to their slower nature. A typical condensation growth rate of small aerosols in atmospheric conditions is 3 nm/h (τ cond � 1400 min) [28,[30][31][32] and the Brownian coagulation timescale τ coag ≥ 154 min (see Table 1) [33]. The presence of ions and charged aerosols can enhance the condensation (�70 min in the current experiments) [18] and coagulation rates [12,24]. However, they are still weaker relative to the source from the new particle formation (τ f ;ait � 7 min) and the wall loss term (τ wall;ait �30 min). Thus, in a quasi-steady-state, n ait � n i τ wall;ait ðc 1 τ f ;ait Þ -1 .
Following the discussion for the no-cloud case (Equation 2), the Aitken-mode budget equation can also be further simplified for the cloudy case (Equation 5) by neglecting the condensation and coagulation growth to accumulation mode due to their slower nature. The condensation (τ cond � 47 min) and coagulation (τ coag ≥ 116 min) timescales are shorter than the no-turbulence case but still much longer than the timescales for new particle formation (τ f ;ait � 8 min) and diffusion to droplets (τ diff < 13 min). In that case, the budget simplifies to
Similar to the Aitken mode, the equation for accumulationmode particles in a cloudy environment (Equation 6) can be simplified by neglecting the coagulation and condensation gain, the diffusion loss to droplets, and wall loss due to their longer timescales relative to the activation deactivation process. As discussed earlier, the condensation (τ cond � 47 min) and coagulation (τ coag ≥ 116 min) gain timescales are long. The diffusion timescale of accumulation model aerosols to droplets is estimated to be more than an order of magnitude larger than the Aitken mode estimate (τ diff (Aitken) � 13 min). The wall loss timescale is � 30 min. The cloud droplet activation and deactivation timescales [34] depend on the exposure history of supersaturated (for activation) and subsaturated (for deactivation) regions in turbulent convection. In a nearly saturated mean environment and �1% saturation ratio standard deviation, the droplet activation/ deactivation timescale for accumulation mode aerosols (<500 nm dry diameter) is predominately controlled by the largeeddy correlation timescale of saturation ratio variability (O (10 s)). Thus, they are the dominant source/sink for the accumulation mode aerosols. Equation (6) then simplifies to
At quasi-state-state after the initial transient phase, n acc � n d τ act;acc ðτ deact Þ -1 . It is important to note that in steady state the activation rate must be greater than the deactivation rate in order to have a cloud droplet source that balances the cloud droplet loss rate due to sedimentation. Thus, there must be a net Ion-induced cloud modulation | 7 source of CCN equal to the cloud droplet sedimentation loss term.
In the cloudy experiment the accumulation-mode particle budget is dominated by activation and deactivation sink and source terms. But that still leaves the question of where the accumulation-mode particles come from and how they are able to continue to feed the cloud-droplet population. Ultimately, the source must be the Aitken particles, so we consider that budget in more detail here. In a moist environment where the ion loss through recombination is lower due to ions clustering with water molecules, thereby decreasing their mobility [25], a sufficiently strong ionization source (e.g. �2 trillion ions/s in the current case) can accelerate the Aitken mode aerosol formation (τ f ;ait � 5-8 min) with enhanced condensation growth due to the presence of ions (τ cond �50-70 min). The newly formed aerosols also diffuse to the walls of the chamber. The attachment of ions to aerosols enhances their removal rate due to image charges when approaching chamber walls (estimated removal timescale from the dry experiments is τ wall;ait �30-40 min, see Fig. A1 in Appendix 1). These ultrafine Aitken mode aerosols coagulate to form accumulation mode aerosols. The minimum Brownian coagulation timescale for the moist convection cases is in the range of τ coag � 116-224 min (based on the maximum concentration). This coagulation timescale is significantly longer than the observed accumulation mode increase rate �20 min in cloudy convection, even with the activation and diffusion loss to droplets. Since the initial cloud droplet increase timescale is 30-50 min, but the settling loss timescale is �3 min, the net activation timescale would be at least �3 min for the case with no external aerosol injection. It also implies the timescale of activation loss of accumulation mode aerosols τ act ≤ 3 min. Thus, the accumulation mode formation timescale is much shorter than even the accumulation mode increase timescale of �20 min. This striking observation made us rethink-are we missing any critical CCN formation process?
The charge effects can enhance the coagulation of small aerosols by an order of magnitude (see Figure 9 in [12]), based on well established theory [24]. This charge-induced coagulation could be one of the crucial factors for enhanced coagulation growth. However, it still appears insufficient to justify the observed fast growth of accumulation mode aerosols and cloud droplet formation. We propose the "cloud-induced runaway CCN production" mechanism outlined in Fig. 4 that can provide a method to rapidly convert Aitken particles in accumulation-mode particles. The scavenging of aerosols by cloud droplets is one of the fastest processes in the current cloudy convection. The timescale of Brownian diffusion of aerosols to droplets is �12-13 min without charge effects. In the presence of charge, however, it is even faster. For example, the diffusion timescale of aerosols with 10e charge (10 nm size) to a cloud droplet population (the same mean concentration as the experiment) of diameter 6 µm and -100e charge would be �80 s [35]. Thus, it is one of the fastest aerosol processes in the system.
The mechanism is as follows: In the case without external aerosols, some Aitken mode aerosols experience supersaturation more than the critical value of activation (S c ) in the turbulent moist convection where supersaturation fluctuates significantly, even if the mean supersaturation is lower than S c [36,37]. Harrison et al. [29] also point out that the critical supersaturation for activating cloud droplets decreases in the presence of charges in haze droplets. The measurement through the PCVI inlet shown in Fig. A3 in Appendix 1 confirms the activation of some Aitken mode aerosols. The activated cloud droplets then rapidly collect ultrafine Aitken aerosols, which are present in abundance. Thus, the solute mass in droplets grows. When they deactivate due to a Droplet number increase ðτ -1 act -τ -1 deact Þ -1 (observation) 30 47 Aitken removal (after turning off ion source)
Accumulation removal (after turning off ion source)
Droplet removal (after turning off ion source)
observation) 64 Mean droplet settling τ sett (theoretical estimate using observation) 3.1 2.5-34 Estimated growth of aerosols from Aitken mode to accumulation mode (10-80 nm) based on the estimated size growth rate of small aerosols (10-30 nm) τ cond (observation) 70 47 54 Minimum aerosol coagulation timescale (only Brownian diffusion of neutral particles) τ coag (theoretical estimate using observation) 154 (44 at the Aitken peak near the end) 116 224 Diffusion of aerosols (10 nm) to droplets (without charge effects) τ diff (theoretical estimate using observation) 13 12 Diffusion of aerosols (10 nm) with 10 e charge to 6 μm droplets with -100 e τ diff (theoretical estimate using observation) 80 s 80 s
Phase relaxation timescale of droplets τ c (theoretical estimate using observation) 5.3 s 5.2-6.0 s highly fluctuating supersaturation environment, they leave large residual aerosols that are more efficient CCN. Consequently, as the activation of a small fraction of the Aitken particles takes place, there is a continual source of CCN and therefore the concentration of cloud droplets increases with time. This, in turn, further enhances the collection of ultrafine aerosols; thus the runaway effect. The runaway CCN production is shut down when the droplet concentration becomes so large that supersaturation fluctuations are buffered out. Experiments show that such buffering does not occur under typical cloud conditions, suggesting that extreme droplet concentrations are required [38]. Furthermore, when the total surface area of cloud droplets becomes large enough, it shuts down the new particle formation. During that period of solitude, cloud droplet loss through sedimentation takes over the budget. Once the droplet and accumulation mode aerosol concentration falls below a critical threshold, the production of ultrafine particles is reinitiated, the activation of a small fraction of Aitken particles is renewed, and the process of "cloudinduced runaway CCN production" continues. The oscillations in concentrations of Aitken and accumulation modes and cloud droplets are evidence of the above-described process.
In this section we discuss various implications of the experiments and specifically focus on connections to the atmosphere as well as to marine cloud brightening. We conclude with some of the open questions that are prompted by this work.
The periodic bursts of new particle formation observed in these experiments are strikingly similar to those observed in the atmosphere [39], albeit with a different time scale. New particle formation is ubiquitous in the remote troposphere and, as is observed in our experiments, has a strong dependence on relative humidity [40]. Our observations also show a strong dependence of new particle formation on ion concentration, which is consistent with theory and observations that suggest they stabilize molecular clusters [41]. Indeed, ions enhance nucleation over the oceans, and it is thought that this can subsequently influence stratiform cloud properties [42]. There is longstanding evidence that new particle formation leads to CCN formation in remote environments [43]. And recent observations confirm that there are strong interactions between new particle formation and cloud/precipitation processes in the marine boundary layer [11].
An intriguing observation in the cloudy experiments is the existence of oscillations in aerosol and cloud droplet populations. Early observations and theoretical description of aerosol nucleation oscillations in a laboratory system were made by McGraw and Saunders [44]. Key components needed for oscillations are injection and depletion [45]. Recent work finds that three time scales govern the periodicity: that for nucleation, that for droplet growth and sedimentation removal, and that for restoration of the vapor profile through diffusion [46]. Our system is more complex because it involves coupling between Aitken and accumulation-mode aerosols, as well as with cloud droplets. Therefore, the time scale for conversion of newly formed particles to CCN is a crucial aspect. Our analysis of time scale suggests that none of the straightforward growth processes (e.g. condensation or coagulation) is sufficiently fast to explain the observations. We proposed that charged cloud droplets instead play a significant role in scavenging Aitken particles, thereby generating larger aerosol particles that can serve as efficient CCN in subsequent cloud cycles. The steady activation of a small fraction of Aitken particles due to supersaturation fluctuations therefore continuously feeds into a growing population of CCN available for cloud formation. An additional mechanism for CCN multiplication could be the scavenging of Aitken particles by charged cloud droplets and then their breakup into multiple fragments through the Rayleigh charge instability that can occur during evaporation [47]. One promising approach worth investigating is whether a predator-prey model similar in spirit of that of Koren and Feingold for precipitation-cloud oscillations [48] would be useful in describing the coupled ion-aerosolcloud system.
The experiments with a coupled ion-aerosol-cloud system are intriguing when considered in the context of marine cloud brightening. It has been suggested that even in the unperturbed environment, the ocean biome acts as a source of gases that can be converted to aerosols, thereby regulating climate [49]. Interactions between dimethyl sulfide (DMS), sulfate aerosols, CCN, and cloud properties in remote marine environment show some intriguing correlations, but many questions remain [50]. It can be stated with reasonable confidence, however, that the rate of producing CCN is a central, controlling aspect for cloud properties in the marine boundary layer. For example, the time scale for aerosol replenishment is observed to play a crucial role in cloud-precipitation cycles and the resulting formation of closed versus open cellular cloud patterns [51]. Even the Aitken particles produced through new particle formation can play an important role in buffering the cloud droplet concentration in clean environments [52,53]. In fact, the processes observed here have some similarities with the Aitken buffering mechanism, insofar as the activation of Aitken particles plays an important role, and there is a method for converting Aitken particles to CCN, i.e., through cloud processing. Key questions, therefore, are first, whether artificially enhancing the ion concentration will lead to higher aerosol concentrations through new particle formation; and second, whether the runaway CCN generation process postulated from the experiments reported here, will operate in the natural cloudtopped marine boundary layer. Obviously the time scales and rates will scale differently in the real atmosphere, but the system at its fundamental level consists of the same interacting particle populations as in the Pi Chamber experiments. The critical timescale that would differ significantly in nature from the laboratory experiment is the timescale associated with the dynamics/turbulence. It would affect the mixing and availability of precursor gases, ions, and freshly produced aerosols for cloud interactions. A typical eddy turnover timescale in the boundary layer is �30 min, comparable to other crucial processes in the coupled ion-aerosol-cloud droplet budget (e.g. τ f ;acc � 20 min). Therefore, it would be the primary controlling factor driving the marine cloud response to ionization in the presence of sufficient precursor gases.
The simplicity of the ionization method makes it appealing for cloud modulation. However, scaling it for marine cloud brightening would require future investigation on several fronts, including its feasibility, deployment techniques, cost, environmental safety, and relative advantages and disadvantages compared to standard technologies [54]. One of the challenges with the standard technologies for cloud brightening (e.g. effervescent Ion-induced cloud modulation | 9 nozzles) is pumping aerosols at an optimal size range, avoiding large CCN [4]. The current method can potentially overcome this limitation since it produces small Aitken aerosols that grow with time through ion-enhanced processing.
We conclude by listing several open questions that are prompted by the findings reported in this paper.
� What is the source of new particle formation in the Pi Chamber? Are there sufficiently abundant trace gases in the marine boundary layer that copious ion generation can lead to new particle formation even in regions with background aerosols? � There is a need to better quantify certain rate processes in the budget equations to verify the proposed runaway CCN generation mechanism. Can this mechanism operate in the cloud-topped marine boundary layer? � Does new particle formation increase monotonically with ion production rate, or is there a lower threshold, for example, due to ion recombination and scavenging, or is there a saturation effect at high rates? Future experiments will be dedicated to exploring the effect of ion source strength. � What strength of ion generation and possibly trace gas formation would be needed to accomplish meaningful marine cloud brightening in the remote marine boundary layer? Some of these questions will be addressed with further experiments in the Pi convection-cloud chamber. Others can be explored by implementing ion-aerosol-cloud interactions in detailed microphysics models, first compared with the Pi Chamber observations, and subsequently scaled up to the atmospheric context. Finally, some of these questions, particularly regarding the viability of generating significant concentrations of CCN in the clean marine boundary layer, will require on site experiments. An especially pertinent question is to what extent the abundance of precursor gases in the marine boundary layer is a controlling factor for the effectiveness of this mechanism in cloud brightening. In spite of the considerable questions that remain, we find the observations reported here to be compelling evidence supporting further investigation in the context of marine cloud brightening.
40 50 60 70 80 90 t [min] 10 15 20 25 30 35 40 45 50 D [nm] -0.5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 log 10 (dN dlogD -1 ) [X,Y] [51.3506 10.1874] Index 3.25696 [R,G,B] [1 0.278431 0] [X,Y] [65.3553 31.0701] Index 2.74544 [R,G,B] [1 0.670588 0] [X,Y] [74.6918 10.1874] Index 3.41284 [R,G,B] [1 0.145098 0] [X,Y] [91.0306 31.0701] Index 2.98003 [R,G,B] [1 0.47451 0] Growth rate ~ 1.5 nm min -1
(1.3 nm min -1 for the second event) Figure A2. An example showing the estimation of condensation growth rate for Aitken mode particles, using a visual method [28]. Results are shown for two observed new particle formation events during the moist convection experiment without external aerosol injection.