Cloud Phase Changes Induced by CO2 Warming—a Powerful yet Poorly Constrained Cloud-Climate Feedback
We review a cloud feedback mechanism that has so far been considered of secondary importance, despite a body of research suggesting that it represents a powerful climate feedback that can control the sign of the overall cloud feedback simulated in global climate models (GCMs). The feedback mechanism is associated with phase changes in clouds triggered by a warming atmosphere, which in turn yields optically thicker clouds. Output from the latest generation of GCMs suggest that this is the dominant cloud feedback at high latitudes, with obvious implications for climatically sensitive regions such as the Arctic and the Southern Ocean. Here, we present an overview of the relatively few modeling studies that have investigated this particular feedback mechanism to date, along with new results suggesting that the cloud-climate feedback simulated by a GCM can change dramatically depending on its cloud phase partitioning.
KeywordsClouds Cloud phase Global climate modeling Satellite observations Cloud-climate feedback
Li and Le Treut were able to generate cloud feedbacks of opposing sign in their model simulations solely by changing the liquid-to-ice transition from −15 to 0 °C. They found that, in the latter case, the cloud coverage and thickness of low- to mid-level clouds increased everywhere except in the low-latitude lower troposphere (always a cooling effect, according to Fig. 1), while in the former case, the increased cloudiness or cloud optical depths occurred at mid- to high levels (cooling or warming, depending on the optical thickness and exact height, Fig. 1). Shifting the liquid-ice phase transition to colder temperatures has important implications for high latitudes. Since colder isotherms exist relatively lower in the atmosphere at high latitudes, shifting the liquid-ice phase transition to colder temperatures implies that phase changes in low clouds will be constrained to high latitudes, where their impact on the shortwave radiation budget is muted by the polar night. At high latitudes, clouds actually have a net warming effect during polar night, which is thought to be weak at the TOA but substantial at the surface, resulting from downward longwave radiation .
In situ and satellite observations have since these early studies demonstrated that the assumption of abrupt phase transitions at a set temperature is rather poor, as will be covered in more detail in the following section. Nevertheless, the finding that the overall cloud feedback simulated by a GCM can switch signs depending on the temperature at which phase changes are assumed to occur is one that is extremely important.
Given the gravity of both the above findings, relatively few follow-up studies have been dedicated to fully understanding this evidently important feedback mechanism. This can in part be explained by the complicated and poorly understood processes that govern cloud phase in the temperature range −35 °C to 0 °C, as well as the lack of observations of cloud phase with good spatial and temporal coverage .
The purposes of this paper are to review the limited work that has been done dedicated to the link between cloud phase and cloud-climate feedbacks in GCMs over the last two and a half decades (“Review of Literature”) and to summarize related research progress that has been made in laboratories, in the field and with numerical modeling, and satellite data recently (“Recent Relevant Research Progress—Satellite Observations, Laboratory and Field Measurements, and Model Parameterizations”). This progress has now brought us to a stage where we can revisit the feedback mechanism related to phase changes in clouds in a more thorough and comprehensive manner. New results from a recent modeling study dedicated entirely to this problem are presented in “Reexamining the Importance of the Cloud Phase Feedback” section. Finally, in the “Conclusion” section, we conclude that more research on this particular feedback mechanism is overdue and urgently needed, as it pertains to one of the most fundamental and controversial questions of our time, namely how sensitive Earth’s climate is to increased atmospheric levels of CO2.
Review of Literature
As much as the Mitchell et al. study was pioneering, the results were based on GCM simulations that were very crude in today’s standard. The horizontal resolution of 5 × 7.5° (latitude × longitude) and 11 vertical levels for the atmosphere is far coarser than the current standard GCM resolution of 1–2° in the horizontal and 30–40 levels in the vertical . Furthermore, model cloud parameterizations have undergone rapid development and generally now all include separate prognostic equations for cloud liquid and ice, with source and sink terms that represent our best knowledge of cloud microphysics, whether anchored in theory, laboratory experiments, field observations, or remote sensing. This is in stark contrast to the less sophisticated state-of-the-art GCMs used 25 years ago, when Mitchell et al. were among the first to include total cloud condensate (liquid and ice) as a prognostic variable. Until then, cloud amount had typically been prescribed and the clouds often had pre-specified radiative properties . It was in fact Mitchell et al.’s newly improved microphysics at the time that had allowed the cloud phase change feedback to surface for the very first time.
Given the rapid evolution of GCMs in recent decades, it is not at all clear that these earlier findings are valid for the latest generation of GCMs. An obvious place to look for cloud feedbacks generated by phase changes in modern GCM simulations is the Cloud Feedback Model Intercomparison Project (CFMIP) . Using model output from the CFMIP archive, Zelinka et al. [6, 12, 13] presented a decomposition of the cloud feedbacks simulated by 11 different GCMs into contributions from cloud height, cloud cover, and cloud optical depth. In response to CO2 doubling, robust features across models included the following: (i) at low latitudes, a reduction in low cloud cover and a decrease in the cloud top pressure of high clouds, both contributing to a positive cloud feedback (see Fig. 1); (ii) at mid- and high latitudes, an increase in mainly cloud optical depth but also cloud coverage, corresponding to a negative cloud feedback. The latter is particularly relevant to this review, as the aforementioned phase transition that accompanies a CO2 warming is a plausible, but not necessarily the sole explanation for this feature. This explanation is affirmed by Zelinka et al.’s findings of an ensemble mean increase in total water path (TWP, gm−2) at high latitudes, which is dominated by an increase in the liquid water path (LWP, gm−2). Observations consistent with a phase change feedback have also previously been reported from satellite data [14, 15], based on the International Satellite Cloud Climatology Project (ISCCP), and from several thousand in situ profiles of cloud water content and temperature . These measurements all found that cloud water content tends to decrease with temperature for warm stratus clouds (temperature T > 0 °C), while an increase in water content with temperature was reported for cold stratus clouds (−35 °C < T < 0 °C). In the early days of climate modeling, Somerville and Remer  used the in situ measurements compiled by Feigelson  to implement a relationship between temperature and cloud water content in a radiative-convective equilibrium model and found a strong negative cloud-climate feedback as a result. However, the model configuration did not allow for simulation of changes in cloud cover or height, a shortcoming acknowledged by the authors. In state-of-the-art GCMs, the observed relationship between temperature and cloud thickness is in fact reasonably reproduced, albeit with biases .
It is worth noting that the poleward shift of mid-latitude storm tracks has also been proposed as a possible explanation for the negative high-latitude cloud feedback [e.g., 19]. While the fact that the high-latitude cloud feedback is dominated by the change in cloud albedo as opposed to cloud amount does not support this hypothesis , the relative contributions of the different high-latitude cloud feedback mechanisms in GCMs remain unclear. However, a recent review of the cloud radiative response to mid-latitude jet shifts found that this mechanism can only explain a modest fraction of the mid-latitude cloud feedback in climate models and thus suggested a dominant role for thermodynamic effects .
With the introduction of prognostic equations for total cloud condensate, the cloud parameterization in GCMs became much more sophisticated and could begin to account for phase transitions in a warming climate, albeit in a crude manner. However, temperature was generally still the sole factor in determining cloud phase. A handful of GCMs from this generation of models was compared in terms of their cloud water content and implications for climate sensitivity . The study reported an intimate relationship between climate sensitivity and phase partitioning in clouds at temperatures between −35 and 0 °C (the mixed-phase layer). All models responded to a doubling of CO2 by producing more liquid in this temperature range, and in agreement with Zelinka et al., this increase was mainly constrained to mid- and high latitudes. Until very recently, this was, to our knowledge, the only study to attempt to follow up on the ideas put forth two decades earlier. The study attributed the stronger response at high latitudes to the presence of more cloud ice in the mixed-phase layer there. Because of the decrease in insolation with latitude, the resulting increase in cloud albedo, and thus its effect on climate sensitivity, becomes less powerful with increasing latitude. This was demonstrated in the study by focusing on cloud albedo over the Southern Ocean, a region considered to be particularly sensitive and important for global climate change . Cloud feedbacks over the Southern Ocean were also the focus of a recent study using several satellite data sets to estimate what albedo changes could be expected to accompany CO2-induced warming . Taking advantage of observed changes in cloud phase with the seasonal cycle, an increase in reflected shortwave radiation of 0.1–1 W m−2 per Kelvin of warming was deduced. These values are consistent with values reported from the CFMIP archive , but the negative feedback was estimated to extend further equatorward relative to the models.
A more recent modeling study presented model simulations that were dedicated specifically to the investigation of the cloud phase feedback and reported a higher climate sensitivity in simulations that had more liquid relative to ice in mixed-phase clouds . However, the difference was more subtle compared to the few previous comparable model experiments [1, 3, 21]. The study pointed out that compensating changes in climate feedbacks in the tropics and extratropics could explain the more modest global mean response.
Recent Relevant Research Progress—Satellite Observations, Laboratory and Field Measurements, and Model Parameterizations
Evident from Fig. 4 is the heterogeneity of the SCF at a fixed temperature, with high latitudes generally associated with high SCFs and low and mid-latitudes generally exhibiting low SCFs. In mixed-phase clouds (clouds that exist at temperatures below freezing but at temperatures warmer than approximately −35 °C), ice formation can only occur with the aid of IN, insoluble particles with the ability to lower the energy barrier associated with the liquid-ice phase transition at these temperatures, which would otherwise prohibit freezing. The most abundant and potent source of natural IN in the atmosphere is mineral dust, but primary organic particles and possibly black carbon have also been shown to have ice-nucleating ability . Evidence suggests that differences in IN abundance and efficiency between various aerosol species are partly responsible for the heterogeneity of SCF shown in Fig. 4 [20, 29]. While this underscores the powerful effect that perturbations in IN, whether natural or anthropogenic, could have on climate, this is not the focus of the present review. The scope of this paper is limited to fast climate feedback mechanisms involving warming-induced phase changes.
The CALIOP observations as well as in situ cloud phase profiles from around the world have demonstrated that diagnosing cloud phase as a function of temperature alone is a modeling approach that is oversimplified and therefore inappropriate. This practice is now gradually being abandoned and replaced by more sophisticated cloud microphysics schemes that carry both cloud liquid and ice as prognostic variables, both in terms of mass and hydrometeor number concentrations [34, 35, 36]. This was in turn made possible by a tremendous effort to develop parameterizations of cloud microphysical processes for use in GCMs, particularly when it comes to the complex microphysics of ice crystal nucleation and subsequent growth [37, 38, 39, 40].
In a recent study, the cloud phase of six GCMs that included state-of-the-art cloud microphysics schemes were compared to each other and to CALIOP observations . While these models had all abandoned the crude determination of cloud phase based on temperature only, the intercomparison revealed a systematic bias in the simulated cloud phase. All six models produced too much ice relative to liquid compared to the satellite observations, a finding that had previously been reported for comparisons between CALIOP measurements and individual models [42, 43] and has since been confirmed for the CMIP5 modeling archive . The biases were particularly large for the Southern Ocean, which stands out in Fig. 4 as a region of particularly high SCF. High SCF values are expected, because the pristine high Southern latitudes are expected to be essentially free of efficient IN. However, the models do not capture this behavior, and this could partly explain the large bias in cloudiness over the Southern Ocean in most GCMs . While necessary and important, the new and refined microphysics schemes increase the number of sub-grid scale processes that must be parameterized in models, which poses additional challenges. Continued efforts to test microphysics schemes, with a focus on ice crystal formation and subsequent growth, on a range of scales and for different cloud regimes will therefore be important for further progress. Specifically, the extent to which mixed-phase clouds are a homogeneous mixture of droplets and ice crystals as opposed to a non-uniform cloud volume with separate single-phase pockets must be addressed, and parameterizations with the ability to represent such sub-grid-scale features in GCMs should be developed.
As demonstrated by the studies summarized in the previous section, an underestimation of the SCF has implications for the magnitude and potentially even the sign of the cloud feedback simulated by these models. In the following section, we will briefly describe a modeling study specifically designed to address exactly this issue and thereby revisit the modeling study by Mitchell et al. 25 years later.
Reexamining the Importance of the Cloud Phase Feedback
Overview of the simulations conducted
Standard CESM1.0.5 and CAM5 coupled simulation
Same as CONTROL, except with negligible amounts of IN and revised treatment of ice nucleation and detrainment from convection
Same as CONTROL, except revised treatment of ice nucleation and IN are present in abundant concentrations (75 times the default)
Same as Low-IN, except that six microphysical parameters have been perturbed to achieve the best possible matcha to CALIOP observations to the best of the current ability of CESM
Same as CALIOP-1, except with different values of the six parameters which produced a roughly equally good matcha to CALIOP
Evident from Fig. 5 is the markedly lower (higher) SCF in High-IN (Low-IN) relative to CONTROL. However, CONTROL’s SCF is consistently lower than those obtained from CALIOP observations, particularly at the warmer mixed-phase cloud temperatures (T > −25 °C). Furthermore, the SCF at high latitudes in CONTROL is close to that of High-IN (not shown). CALIOP-1 and CALIOP-2 were both designed to produce similar SCFs to those obtained by CALIOP, and Fig. 5 shows that they do in fact match CALIOP’s SCFs relatively well.
A more thorough analysis of the simulations is presented elsewhere.1 Here, we present only a comparison of the cloud feedbacks in each experiment, calculated based on simulations that have radiation budgets balanced to within 0.3 W m−2 in the last 50 years of each simulation. Adopting the method of Zelinka et al. , the net cloud feedback has been decomposed into its three components (amount, optical depth, and altitude) using the cloud radiative kernels. The method has the distinct advantage of attributing TOA radiative budget changes to cloud tops the way ISCCP would view clouds. It reveals that the net cloud optical depth feedback parameter in the extratropics monotonically increases with mean state SCF, while the sum of the net cloud altitude and net cloud amount feedbacks in the extratropics remains constant to within 0.06 W m−2 K−1 in all simulations. This finding is consistent with the cloud phase feedback, where relatively higher SCFs in the initial state reduces the optical depth increase associated with ice-to-liquid transitions, thereby leading to a more positive net cloud feedback. These results imply that realistically constraining mixed-phase cloud SCFs results in a more positive net cloud feedback than would otherwise occur should SCFs not be constrained by observations.
The above results confirm that the powerful influence of cloud phase on climate sensitivity first detected in model simulations 25 years ago is still present and equally important in today’s much more sophisticated GCMs, here represented by CESM1.0.5-CAM5.
The aim of this review is to bring attention to a potentially crucial feedback mechanism that has so far been considered of secondary importance, namely that associated with phase changes in clouds in a warming climate. The relatively limited body of research carried out to date suggests that this phase change causes an increase in cloud albedo due to increased low cloud water content, particularly at mid- to high latitudes, and it thus represents a negative climate feedback. The strength of the cloud phase feedback is therefore likely to affect the degree of polar amplification of CO2-induced warming and will disproportionally affect sensitive climatic regions like the Arctic and the Southern Ocean. The research carried out to date also suggests that the strength of this feedback mechanism in GCMs is very sensitive to the way in which models partition cloud water into liquid versus ice. Three previous modeling studies [1, 3, 21] combined with the preliminary results presented in the “Reexamining the Importance of the Cloud Phase Feedback” section of this paper demonstrate that the climate feedbacks and sensitivity simulated by GCMs changes radically in response to changes in cloud phase partitioning. These modeling studies focused specifically on the link between cloud phase and climate feedbacks or sensitivity, and all suggest that an underestimate of liquid in mixed-phase clouds will produce an underestimate of climate sensitivity. It is therefore noteworthy that several recent GCM intercomparison studies have found models to severely underestimate the amount of supercooled liquid relative to satellite observations . This, combined with a recent study reporting that GCMs tend to overestimate the increase in water content with increasing temperature in cold clouds , suggests that the negative phase change feedback may be too strong in GCMs. All of the above calls for a serious evaluation of this particular cloud-climate feedback, which should include dedicated model simulations, extensive use of field measurements for testing of new cloud parameterizations, as well as both global and regional comparisons between GCMs and satellite data. Specifically for GCMs, one way to move forward would be to carry out an ambitious model inter-comparison project aimed at systematically testing the sensitivity of modeled ECSs to cloud phase. Such projects generally require the coordinated efforts of multiple modeling groups, and as such represent a tremendous challenge that we urge the modeling community to take on.
Tan, I., T. Storelvmo, and M. D. Zelinka, Manuscript in prep.
Compliance with Ethical Standards
Conflict of Interest
On behalf of all authors, the corresponding author states that there is no conflict of interest.
- 8.Boucher O, Randall D, Artaxo P, Bretherton C, Feingold G, Forster P, Kerminen VM, Kondo Y, Liao H, Lohmann U, Rasch P, Satheesh SK, Sherwood S, Stevens B, Zhang XY. Clouds and Aerosols, in Climate Change 2013: the physical science basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change, Stocker TF, Qin D, Plattner GK, Tignor M, Allen SK, Boschung J, Nauels A, Xia Y, Bex V, Midgley PM. Editor. Cambridge, United Kingdom and New York, NY, USA: Cambridge University Press. 2013p. 571–658.Google Scholar
- 11.McAvaney BJ, Le Treut H. The cloud feedback intercomparison project (CFMIP), in CLIVAR Exchanges. United Kingdom: International CLIVAR Project Office: Southampton; 2003. p. 1–4.Google Scholar
- 16.Feigelson E. Preliminary radiation model of a cloudy atmosphere. I—structure of clouds and solar radiation. Beitraege zur Physik der Atmosphaere. 1978;51(3):203–29.Google Scholar
- 18.Gordon ND, Klein SA. Low-cloud optical depth feedback in climate models. Journal of Geophysical Research: Atmospheres, 2014;11910:p. 2013JD021052.Google Scholar
- 23.McCoy DT, Hartmann DL, Grosvenor DP. Observed Southern Ocean Cloud Properties and Shortwave Reflection Part 2: phase changes and low cloud feedback. Journal of Climate. 2014.Google Scholar
- 28.Hu YX, et al, Occurrence, liquid water content, and fraction of supercooled water clouds from combined CALIOP/IIR/MODIS measurements. Journal of Geophysical Research-Atmospheres, 2010;115. Google Scholar
- 31.Doutriaux-Boucher M, Quaas J. Evaluation of cloud thermodynamic phase parametrizations in the LMDZ GCM by using POLDER satellite data. Geophysical Research Letters. 2004;316.Google Scholar
- 32.Tan I, Storelvmo T, Choi Y-S. A comparison of the ice nucleating efficiencies of clean dust, polluted dust, and smoke aerosols in mixed-phase clouds based on spaceborne lidar observations. J Geophys Res. 2014;119(11):6653–65.Google Scholar
- 40.Storelvmo T, et al., Modeling of the Wegener-Bergeron-Findeisen process-implications for aerosol indirect effects. Environmental Research Letters. 2008;34.Google Scholar
- 44.Cesana, G., et al., Multimodel evaluation of cloud phase transition using satellite and reanalysis data. Journal of Geophysical Research: Atmospheres. 2015;120:15: p. 2014JD022932.Google Scholar
- 48.Young KC. Conversion of a supercooled cloud to ice via contact nucleation and direct injection of ice crystals. Bull Am Meteorol Soc. 1974;55(6):679.Google Scholar