Clouds and Aerosols

Methods.................................................................... 635 References .................................................................................. 636 Frequently Asked Questions FAQ 7.1 How Do Clouds Affect Climate and Climate Change?................................................................... 593 FAQ 7.2 How Do Aerosols Affect Climate and Climate Change?................................................................... 622 FAQ 7.3 Could Geoengineering Counteract Climate Change and What Side Effects Might Occur? ..... 632 Supplementary Material Supplementary Material is available in online versions of the report. 573 Clouds and Aerosols Chapter 7 7 1 In this Report, the following summary terms are used to describe the available evidence: limited, medium, or robust; and for the degree of agreement: low, medium, or high. A level of confidence is expressed using five qualifiers: very low, low, medium, high, and very high, and typeset in italics, e.g., medium confidence. For a given evidence and agreement statement, different confidence levels can be assigned, but increasing levels of evidence and degrees of agreement are correlated with increasing confidence (see Section 1.4 and Box TS.1 for more details). Executive Summary Clouds and aerosols continue to contribute the largest uncertainty to estimates and interpretations of the Earth’s changing energy budget. This chapter focuses on process understanding and considers observa- tions, theory and models to assess how clouds and aerosols contribute and respond to climate change. The following conclusions are drawn. Progress in Understanding Many of the cloudiness and humidity changes simulated by climate models in warmer climates are now understood as responses to large-scale circulation changes that do not appear to depend strongly on sub-grid scale model processes, increasing confidence in these changes. For example, multiple lines of evidence now indicate positive feedback contributions from circula- tion-driven changes in both the height of high clouds and the latitudi- nal distribution of clouds (medium to high confidence1). However, some aspects of the overall cloud response vary substantially among models, and these appear to depend strongly on sub-grid scale processes in which there is less confidence. {7.2.4, 7.2.5, 7.2.6, Figure 7.11} Climate-relevant aerosol processes are better understood, and climate-relevant aerosol properties better observed, than at the time of AR4. However, the representation of relevant processes varies greatly in global aerosol and climate models and it remains unclear what level of sophistication is required to model their effect on climate. Globally, between 20 and 40% of aerosol optical depth (medium confi- dence) and between one quarter and two thirds of cloud condensation nucleus concentrations (low confidence) are of anthropogenic origin. {7.3, Figures 7.12 to 7.15} Cosmic rays enhance new particle formation in the free tropo- sphere, but the effect on the concentration of cloud condensa- tion nuclei is too weak to have any detectable climatic influence during a solar cycle or over the last century (medium evidence, high agreement). No robust association between changes in cosmic rays and cloudiness has been identified. In the event that such an asso- ciation existed, a mechanism other than cosmic ray-induced nucleation of new aerosol particles would be needed to explain it. {7.4.6} Recent research has clarified the importance of distinguishing forcing (instantaneous change in the radiative budget) and rapid adjustments (which modify the radiative budget indirectly through fast atmospheric and surface changes) from feedbacks (which operate through changes in climate variables that are mediated by a change in surface temperature). Furthermore, one can distinguish between the traditional concept of radiative forcing (RF) and the relatively new concept of effective radiative forcing (ERF) that also includes rapid adjustments. For aerosols one can further dis- tinguish forcing processes arising from aerosol–radiation interactions (ari) and aerosol–cloud interactions (aci). {7.1, Figures 7.1 to 7.3} The quantification of cloud and convective effects in models, and of aerosol–cloud interactions, continues to be a challenge. Climate models are incorporating more of the relevant process- es than at the time of AR4, but confidence in the representation of these processes remains weak. Cloud and aerosol properties vary at scales significantly smaller than those resolved in climate models, and cloud-scale processes respond to aerosol in nuanced ways at these scales. Until sub-grid scale parameterizations of clouds and aerosol– cloud interactions are able to address these issues, model estimates of aerosol–cloud interactions and their radiative effects will carry large uncertainties. Satellite-based estimates of aerosol–cloud interactions remain sensitive to the treatment of meteorological influences on clouds and assumptions on what constitutes pre-industrial conditions. {7.3, 7.4, 7.5.3, 7.5.4, 7.6.4, Figures 7.8, 7.12, 7.16} Precipitation and evaporation are expected to increase on aver- age in a warmer climate, but also undergo global and regional adjustments to carbon dioxide (CO2) and other forcings that differ from their warming responses. Moreover, there is high confidence that, as climate warms, extreme precipitation rates on for example, daily time scales will increase faster than the time average. Changes in average precipitation must remain consis- tent with changes in the net rate of cooling of the troposphere, which is affected by its temperature but also by greenhouse gases (GHGs) and aerosols. Consequently, while the increase in global mean pre- cipitation would be 1.5 to 3.5% °C–1 due to surface temperature change alone, warming caused by CO2 or absorbing aerosols results in a smaller sensitivity, even more so if it is partially offset by albedo increases. The complexity of land surface and atmospheric process- es limits confidence in regional projections of precipitation change, especially over land, although there is a component of a ‘wet-get-wet- ter’ and ‘dry-get-drier’ response over oceans at the large scale. Chang- es in local extremes on daily and sub-daily time scales are strongly influenced by lower-tropospheric water vapour concentrations, and on average will increase by roughly 5 to 10% per degree Celsius of warm- ing (medium confidence). Aerosol–cloud interactions can influence the character of individual storms, but evidence for a systematic aerosol effect on storm or precipitation intensity is more limited and ambigu- ous. {7.2.4, 7.4, 7.6, Figures 7.20, 7.21} 574 Chapter 7 Clouds and Aerosols 7 Water Vapour, Cloud and Aerosol Feedbacks The net feedback from water vapour and lapse rate changes combined, as traditionally defined, is extremely likely2 positive (amplifying global climate changes). The sign of the net radia- tive feedback due to all cloud types is less certain but likely positive. Uncertainty in the sign and magnitude of the cloud feedback is due primarily to continuing uncertainty in the impact of warming on low clouds. We estimate the water vapour plus lapse rate feedback3 to be +1.1 (+0.9 to +1.3) W m−2 °C−1 and the cloud feedback from all cloud types to be +0.6 (−0.2 to +2.0) W m–2 °C–1. These ranges are broader than those of climate models to account for additional uncertainty associated with processes that may not have been accounted for in those models. The mean values and ranges in climate models are essentially unchanged since AR4, but are now supported by stronger indirect observational evidence and better process understanding, especially for water vapour. Low clouds con- tribute positive feedback in most models, but that behaviour is not well understood, nor effectively constrained by observations, so we are not confident that it is realistic. {7.2.4, 7.2.5, 7.2.6, Figures 7.9 to 7.11}. Aerosol–climate feedbacks occur mainly through changes in the source strength of natural aerosols or changes in the sink effi- ciency of natural and anthropogenic aerosols; a limited number of modelling studies have bracketed the feedback parameter within ±0.2 W m–2 °C–1 with low confidence. There is medium con- fidence for a weak dimethylsulphide–cloud condensation nuclei–cloud albedo feedback due to a weak sensitivity of cloud condensation nuclei population to changes in dimethylsulphide emissions. {7.3.5} Quantification of climate forcings4 due to aerosols and clouds The ERF due to aerosol–radiation interactions that takes rapid adjustments into account (ERFari) is assessed to be –0.45 (–0.95 to +0.05) W m–2. The RF from absorbing aerosol on snow and ice is assessed separately to be +0.04 (+0.02 to +0.09) W m–2. Prior to adjustments taking place, the RF due to aerosol–radiation interactions (RFari) is assessed to be –0.35 (–0.85 to +0.15) W m–2. The assessment for RFari is less negative than reported in AR4 because of a re-evaluation of aerosol absorption. The uncertainty estimate is wider but more robust, based on multiple lines of evidence from models, remotely sensed data, and ground-based measurements. Fossil fuel and biofuel emissions4 contribute to RFari via sulphate aerosol: –0.4 (–0.6 to –0.2) W m–2, black carbon (BC) aerosol: +0.4 (+0.05 to +0.8) W m–2, and primary and secondary organic aerosol: –0.12 (–0.4 to +0.1) W m–2. Additional RFari contributions occur via biomass burning 2 In this Report, the following terms have been used to indicate the assessed likelihood of an outcome or a result: Virtually certain 99–100% probability, Very likely 90–100%, Likely 66–100%, About as likely as not 33–66%, Unlikely 0–33%, Very unlikely 0–10%, Exceptionally unlikely 0–1%. Additional terms (Extremely likely: 95–100%, More likely than not >50–100%, and Extremely unlikely 0–5%) may also be used when appropriate. Assessed likelihood is typeset in italics, e.g., very likely (see Section 1.4 and Box TS.1 for more details). 3 This and all subsequent ranges given with this format are 90% uncertainty ranges unless otherwise specified. 4 All climate forcings (RFs and ERFs) are anthropogenic and relate to the period 1750–2010 unless otherwise specified. 5 This species breakdown is less certain than the total RFari and does not sum to the total exactly. emissions5: +0.0 (–0.2 to +0.2) W m–2, nitrate aerosol: –0.11 (–0.3 to –0.03) W m–2, and mineral dust: –0.1 (–0.3 to +0.1) W m–2 although the latter may not be entirely of anthropogenic origin. While there is robust evidence for the existence of rapid adjustment of clouds in response to aerosol absorption, these effects are multiple and not well represented in climate models, leading to large uncertainty. Unlike in the last IPCC assessment, the RF from BC on snow and ice includes the effects on sea ice, accounts for more physical processes and incorpo- rates evidence from both models and observations. This RF has a 2 to 4 times larger global mean surface temperature change per unit forcing than a change in CO2. {7.3.4, 7.5.2, Figures 7.17, 7.18} The total ERF due to aerosols (ERFari+aci, excluding the effect of absorbing aerosol on snow and ice) is assessed to be –0.9 (–1.9 to –0.1) W m–2 with medium confidence. The ERFari+aci esti- mate includes rapid adjustments, such as changes to the cloud lifetime and aerosol microphysical effects on mixed-phase, ice and convective clouds. This range was obtained from expert judgement guided by cli- mate models that include aerosol effects on mixed-phase and convec- tive clouds in addition to liquid clouds, satellite studies and models that allow cloud-scale responses. This forcing can be much larger regionally but the global mean value is consistent with several new lines of evidence suggesting less negative estimates for the ERF due to aerosol–cloud interactions than in AR4. {7.4, 7.5.3, 7.5.4, Figure 7.19} Persistent contrails from aviation contribute a RF of +0.01 (+0.005 to +0.03) W m–2 for year 2011, and the combined con- trail and contrail-cirrus ERF from aviation is assessed to be +0.05 (+0.02 to +0.15) W m–2. This forcing can be much larger regionally but there is now medium confidence that it does not pro- duce observable regional effects on either the mean or diurnal range of surface temperature. {7.2.7} Geoengineering Using Solar Radiation Management Methods Theory, model studies and observations suggest that some Solar Radiation Management (SRM) methods, if practicable, could sub- stantially offset a global temperature rise and partially offset some other impacts of global warming, but the compensation for the climate change caused by GHGs would be imprecise (high confidence). SRM methods are unimplemented and untested. Research on SRM is in its infancy, though it leverages understanding of how the climate responds to forcing more generally. The efficacy of a number of SRM strategies was assessed, and there is medium con- fidence that stratospheric aerosol SRM is scalable to counter the RF from increasing GHGs at least up to approximately 4 W m–2; however, 575 Clouds and Aerosols Chapter 7 7 the required injection rate of aerosol precursors remains very uncertain. There is no consensus on whether a similarly large RF could be achieved from cloud brightening SRM owing to uncertainties in understanding and representation of aerosol–cloud interactions. It does not appear that land albedo change SRM can produce a large RF. Limited literature on other SRM methods precludes their assessment. Models consistently suggest that SRM would generally reduce climate differences compared to a world with elevated GHG concentrations and no SRM; however, there would also be residual regional differences in climate (e.g., tem- perature and rainfall) when compared to a climate without elevated GHGs. {7.4.3, 7.7} Numerous side effects, risks and shortcomings from SRM have been identified. Several lines of evidence indicate that SRM would produce a small but significant decrease in global precipitation (with larger differences on regional scales) if the global surface tempera- ture were maintained. A number of side effects have been identified. One that is relatively well characterized is the likelihood of modest polar stratospheric ozone depletion associated with stratospheric aerosol SRM. There could also be other as yet unanticipated conse- quences. As long as GHG concentrations continued to increase, the SRM would require commensurate increase, exacerbating side effects. In addition, scaling SRM to substantial levels would carry the risk that if the SRM were terminated for any reason, there is high confidence that surface temperatures would increase rapidly (within a decade or two) to values consistent with the GHG forcing, which would stress systems sensitive to the rate of climate change. Finally, SRM would not compensate for ocean acidification from increasing CO2. {7.6.3, 7.7, Figures 7.22 to 7.24} 576 Chapter 7 Clouds and Aerosols 7 7.1 Introduction 7.1.1 Clouds and Aerosols in the Atmosphere The atmosphere is composed mostly of gases, but also contains liquid and solid matter in the form of particles. It is usual to distinguish these particles according to their size, chemical composition, water content and fall velocity into atmospheric aerosol particles, cloud particles and falling hydrometeors. Despite their small mass or volume fraction, particles in the atmosphere strongly influence the transfer of radi- ant energy and the spatial distribution of latent heating through the atmosphere, thereby influencing the weather and climate. Cloud formation usually takes place in rising air, which expands and cools, thus permitting the activation of aerosol particles into cloud droplets and ice crystals in supersaturated air. Cloud particles are gen- erally larger than aerosol particles and composed mostly of liquid water or ice. The evolution of a cloud is governed by the balance between a number of dynamical, radiative and microphysical processes. Cloud particles of sufficient size become falling hydrometeors, which are cat- egorized as drizzle drops, raindrops, snow crystals, graupel and hail- stones. Precipitation is an important and complex climate variable that is influenced by the distribution of moisture and cloudiness, and to a lesser extent by the concentrations and properties of aerosol particles. Aerosol particles interact with solar radiation through absorption and scattering and, to a lesser extent with terrestrial radiation through absorption, scattering and emission. Aerosols6 can serve as cloud condensation nuclei (CCN) and ice nuclei (IN) upon which cloud drop- lets and ice crystals form. They also play a wider role in atmospheric chemistry and biogeochemical cycles in the Earth system, for instance, by carrying nutrients to ocean ecosystems. They can be of natural or anthropogenic origin. Cloud and aerosol amounts7 and properties are extremely variable in space and time. The short lifetime of cloud particles in subsaturated air creates relatively sharp cloud edges and fine-scale variations in cloud properties, which is less typical of aerosol layers. While the distinction between aerosols and clouds is generally appropriate and useful, it is not always unambiguous, which can cause interpretational difficulties (e.g., Charlson et al., 2007; Koren et al., 2007). 7.1.2 Rationale for Assessing Clouds, Aerosols and Their Interactions The representation of cloud processes in climate models has been rec- ognized for decades as a dominant source of uncertainty in our under- standing of changes in the climate system (e.g., Arakawa, 1975, 2004; Charney et al., 1979; Cess et al., 1989; Randall et al., 2003; Bony et al., 2006), but has never been systematically assessed by the IPCC before. Clouds respond to climate forcing mechanisms in multiple ways, and 6 For convenience the term ‘aerosol’, which includes both the particles and the suspending gas, is often used in its plural form to mean ‘aerosol particles’ both in this chapter and the rest of this Report. 7 In this chapter, we use ‘cloud amount’ as an inexact term to refer to the quantity of clouds, both in the horizontal and vertical directions. The term ‘cloud cover’ is used in its usual sense and refers to the horizontal cloud cover. inter-model differences in cloud feedbacks constitute by far the prima- ry source of spread of both equilibrium and transient climate responses simulated by climate models (Dufresne and Bony, 2008) despite the fact that most models agree that the feedback is positive (Randall et al., 2007; Section 7.2). Thus confidence in climate projections requires a thorough assessment of how cloud processes have been accounted for. Aerosols of anthropogenic origin are responsible for a radiative forcing (RF) of climate change through their interaction with radiation, and also as a result of their interaction with clouds. Quantification of this forcing is fraught with uncertainties (Haywood and Boucher, 2000; Lohmann and Feichter, 2005) and aerosols dominate the uncertain- ty in the total anthropogenic RF (Forster et al., 2007; Haywood and Schulz, 2007; Chapter 8). Furthermore, our inability to better quantify non-greenhouse gas RFs, and primarily those that result from aerosol– cloud interactions, underlie difficulties in constraining climate sensitiv- ity from observations even if we had a perfect knowledge of the tem- perature record (Andreae et al., 2005). Thus a complete understanding of past and future climate change requires a thorough assessment of aerosol–cloud–radiation interactions. 7.1.3 Forcing, Rapid Adjustments and Feedbacks Figure 7.1 illustrates key aspects of how clouds and aerosols contribute to climate change, and provides an overview of important terminolog- ical distinctions. Forcings associated with agents such as greenhouse gases (GHGs) and aerosols act on global mean surface temperature through the global radiative (energy) budget. Rapid adjustments (sometimes called rapid responses) arise when forcing agents, by alter- ing flows of energy internal to the system, affect cloud cover or other components of the climate system and thereby alter the global budget indirectly. Because these adjustments do not operate through changes in the global mean surface temperature (DT), which are slowed by the massive heat capacity of the oceans, they are generally rapid and most are thought to occur within a few weeks. Feedbacks are associated with changes in climate variables that are mediated by a change in global mean surface temperature; they contribute to amplify or damp global temperature changes via their impact on the radiative budget. In this report, following an emerging consensus in the literature, the traditional concept of radiative forcing (RF, defined as the instanta- neous radiative forcing with stratospheric adjustment only) is de-em- phasized in favour of an absolute measure of the radiative effects of all responses triggered by the forcing agent that are independent of surface temperature change (see also Section 8.1). This new measure of the forcing includes rapid adjustments and the net forcing with these adjustments included is termed the effective radiative forcing (ERF). The climate sensitivity to ERF will differ somewhat from tradi- tional equilibrium climate sensitivity, as the latter include adjustment effects. As shown in Figure 7.1, adjustments can occur through geo- graphic temperature variations, lapse rate changes, cloud changes 577 Clouds and Aerosols Chapter 7 7 Aerosol–Radiation Interactions (ari) Aerosol–Cloud Interactions (aci) Effective Radiative Forcing (ERF) and Feedbacks Aerosols Clouds and Precipitation Radiation Radiative Forcing Anthropogenic Sources Global Surface Temperature Adjustments Aerosol Feedbacks Cloud Feedbacks Other Feedbacks Moisture and Winds Temperature Profile Regional Variability Biosphere Additional state variables Greenhouse Gases Figure 7.1 | Overview of forcing and feedback pathways involving greenhouse gases, aerosols and clouds. Forcing agents are in the green and dark blue boxes, with forcing mechanisms indicated by the straight green and dark blue arrows. The forcing is modified by rapid adjustments whose pathways are independent of changes in the globally aver- aged surface temperature and are denoted by brown dashed arrows. Feedback loops, which are ultimately rooted in changes ensuing from changes in the surface temperature, are represented by curving arrows (blue denotes cloud feedbacks; green denotes aerosol feedbacks; and orange denotes other feedback loops such as those involving the lapse rate, water vapour and surface albedo). The final temperature response depends on the effective radiative forcing (ERF) that is felt by the system, that is, after accounting for rapid adjustments, and the feedbacks. and vegetation effects. Measures of ERF and rapid adjustments have existed in the literature for more than a decade, with a number of different terminologies and calculation methods adopted. These were principally aimed to help quantify the effects of aerosols on clouds (Rotstayn and Penner, 2001; Lohmann et al., 2010) and understand different forcing agent responses (Hansen et al., 2005), but it is now realized that there are rapid adjustments in response to the CO2 forcing itself (Section 7.2.5.6). In principle rapid adjustments are independent of DT, while feedbacks operate purely through DT. Thus, within this framework adjustments are not another type of ‘feedback’ but rather a non-feedback phenom- enon, required in the analysis by the fact that a single scalar DT cannot fully characterize the system. This framework brings most efficacies close to unity although they are not necessarily exactly 1 (Hansen et al., 2005; Bond et al., 2013). There is also no clean separation in time scale between rapid adjustments and warming. Although the former occur mostly within a few days of applying a forcing (Dong et al., 2009), some adjustments such as those that occur within the stratosphere and snowpack can take several months or longer. Meanwhile the land surface warms quickly so that a small part of DT occurs within days to weeks of an applied forcing. This makes the two phenomena difficult to isolate in model runs. Other drawbacks are that adjustments are diffi- cult to observe, and typically more model-dependent than RF. However, recent work is beginning to meet the challenges of quantifying the adjustments, and has noted advantages of the new framework (e.g., Vial et al., 2013; Zelinka et al., 2013). There is no perfect method to determine ERF. Two common meth- ods are to regress the net energy imbalance onto DT in a transient Figure 7.2 | Radiative forcing (RF) and effective radiative forcing (ERF) estimates derived by two methods, for the example of 4 × CO2 experiments in one climate model. N is the net energy imbalance at the top of the atmosphere and DT the global mean surface temperature change. The fixed sea surface temperature ERF estimate is from an atmosphere–land model averaged over 30 years. The regression estimate is from 150 years of a coupled model simulation after an instantaneous quadrupling of CO2, with the N from individual years in this regression shown as black diamonds. The strato- spherically adjusted RF is the tropopause energy imbalance from otherwise identical radiation calculations at 1 × and 4 × CO2 concentrations. (Figure follows Andrews et al., 2012.) See also Figure 8.1. ERF – fixed sea surface temperature ERF– regression RF – stratospherically adjusted N (W m -2 ) ΔT (ºC) 0 1 2 3 4 5 6 2 -2 0 4 6 8 578 Chapter 7 Clouds and Aerosols 7 warming simulation (Gregory et al., 2004; Figure 7.2), or to simulate the climate response with sea surface temperatures (SSTs) held fixed (Hansen et al., 2005). The former can be complicated by natural var- iability or time-varying feedbacks, while the non-zero DT from land warming complicates the latter. Both methods are used in this chapter. Figure 7.3 links the former terminology of aerosol direct, semi-direct and indirect effects with the new terminology used in this chapter and in Chapter 8. The RF from aerosol–radiation interactions (abbreviat- ed RFari) encompasses radiative effects from anthropogenic aerosols before any adjustment takes place and corresponds to what is usually referred to as the aerosol direct effect. Rapid adjustments induced by aerosol radiative effects on the surface energy budget, the atmospheric profile and cloudiness contribute to the ERF from aerosol–radiation interactions (abbreviated ERFari). They include what has earlier been referred to as the semi-direct effect. The RF from aerosol–cloud inter- actions (abbreviated RFaci) refers to the instantaneous effect on cloud albedo due to changing concentrations of cloud condensation and ice nuclei, also known as the Twomey effect. All subsequent changes to the cloud lifetime and thermodynamics are rapid adjustments, which contribute to the ERF from aerosol–cloud interactions (abbreviated ERFaci). RFaci is a theoretical construct that is not easy to separate from other aerosol–cloud interactions and is therefore not quantified in this chapter. 7.1.4 Chapter Roadmap For the first time in the IPCC WGI assessment reports, clouds and aer- osols are discussed together in a single chapter. Doing so allows us to assess, and place in context, recent developments in a large and growing area of climate change research. In addition to assessing cloud feedbacks and aerosol forcings, which were covered in previ- ous assessment reports in a less unified manner, it becomes possible to assess understanding of the multiple interactions among aerosols, Direct Effect Semi-Direct Effects Lifetime (including glaciation & thermodynamic) Effects Cloud Albedo Effect Radiative Forcing (RFari) Adjustments Effective Radiative Forcing (ERFari) AR4 AR5 Irradiance Changes from Aerosol-Radiation Interactions (ari) Adjustments Effective Radiative Forcing (ERFaci) Radiative Forcing (RFaci) Irradiance Changes from Aerosol-Cloud Interactions (aci) Figure 7.3 | Schematic of the new terminology used in this Assessment Report (AR5) for aerosol–radiation and aerosol–cloud interactions and how they relate to the terminology used in AR4. The blue arrows depict solar radiation, the grey arrows terrestrial radiation and the brown arrow symbolizes the importance of couplings between the surface and the cloud layer for rapid adjustments. See text for further details. clouds and precipitation and their relevance for climate and climate change. This chapter assesses the climatic roles and feedbacks of water vapour, lapse rate and clouds (Section 7.2), discusses aerosol–radiation (Section 7.3) and aerosol–cloud (Section 7.4) interactions and quanti- fies the resulting aerosol RF on climate (Section 7.5). It also introduc- es the physical basis for the precipitation responses to aerosols and climate changes (Section 7.6) noted later in the Report, and assesses geoengineering methods based on solar radiation management (Sec- tion 7.7). 7.2 Clouds This section summarizes our understanding of clouds in the current climate from observations and process models; advances in the rep- resentation of cloud processes in climate models since AR4; assessment of cloud, water vapour and lapse rate feedbacks and adjustments; and the RF due to clouds induced by moisture released by two anthropo- genic processes (air traffic and irrigation). Aerosol–cloud interactions are assessed in Section 7.4. The fidelity of climate model simulations of clouds in the current climate is assessed in Chapter 9. 7.2.1 Clouds in the Present-Day Climate System 7.2.1.1 Cloud Formation, Cloud Types and Cloud Climatology To form a cloud, air must cool or moisten until it is sufficiently super- saturated to activate some of the available condensation or freezing nuclei. Clouds may be composed of liquid water (possibly supercooled), ice or both (mixed phase). The nucleated cloud particles are initially very small, but grow by vapour deposition. Other microphysical mecha- nisms dependent on the cloud phase (e.g., droplet collision and coales- cence for liquid clouds, riming and Wegener–Bergeron–Findeisen pro- cesses for mixed-phase clouds and crystal aggregation in ice clouds) 579 Clouds and Aerosols Chapter 7 7 can produce a broader spectrum of particle sizes and types; turbulent mixing produces further variations in cloud properties on scales from kilometres to less than a centimetre (Davis et al., 1999; Bodenschatz et al., 2010). If and when some of the droplets or ice particles become large enough, these will fall out of the cloud as precipitation. Atmospheric flows often organize convection and associated clouds into coherent systems having scales from tens to thousands of kilo- metres, such as cyclones or frontal systems. These represent a signifi- cant modelling and theoretical challenge, as they are usually too large to represent within the limited domains of cloud-resolving models (Section 7.2.2.1), but are also not well resolved nor parameterized by most climate models; this gap, however, is beginning to close (Sec- tion 7.2.2.2). Finally, clouds and cloud systems are organized by larg- er-scale circulations into different regimes such as deep convection near the equator, subtropical marine stratocumulus, or mid-latitude storm tracks guided by the tropospheric westerly jets. Figure 7.4 shows a selection of widely occurring cloud regimes schematically and as they might appear in a typical geostationary satellite image. New satellite sensors and new analysis of previous data sets have given us a clearer picture of the Earth’s clouds since AR4. A notable example Deep Tropics Large-scale Subsidence Trade Winds Stratocumulus Land/Sea Circulation Shallow Cumulus Convective Anvils Thin Cirrus (b) (c) Subtropics C OLD Cirrus Altostratus Nimbostratus WARM Stratus Polar (mixed phase) Stratus High Latitudes Mid-Latitudes Melting Level WARM OCEAN C OLD OCEAN WARM SUBSIDING R EGIONS (a) 17 km 10 km (a) (a) ((((((((((((a a a a a a a a a a a a a a))))))))))) Figure 7.4 | Diverse cloud regimes reflect diverse meteorology. (a) A visible-wavelength geostationary satellite image shows (from top to bottom) expanses and long arcs of cloud associated with extratropical cyclones, subtropical coastal stratocumulus near Baja California breaking up into shallow cumulus clouds in the central Pacific and mesoscale convec- tive systems outlining the Pacific Intertropical Convergence Zone (ITCZ). (b) A schematic section along the dashed line from the orange star to the orange circle in (a), through a typical warm front of an extratropical cyclone. It shows (from right to left) multiple layers of upper-tropospheric ice (cirrus) and mid-tropospheric water (altostratus) cloud in the upper-tropospheric outflow from the frontal zone, an extensive region of nimbostratus associated with frontal uplift and turbulence-driven boundary layer cloud in the warm sector. (c) A schematic section along the dashed line from the red star to the red circle in (a), along the low-level trade wind flow from a subtropical west coast of a continent to the ITCZ. It shows (from right to left) typical low-latitude cloud mixtures, shallow stratocumulus trapped under a strong subsidence inversion above the cool waters of the oceanic upwelling zone near the coast and shallow cumulus over warmer waters further offshore transitioning to precipitating cumulonimbus cloud systems with extensive cirrus anvils associated with rising air motions in the ITCZ. is the launch in 2006 of two coordinated, active sensors, the Cloud Profiling Radar (CPR) on the CloudSat satellite (Stephens et al., 2002) and the Cloud–Aerosol Lidar with Orthogonal Polarization (CALIOP) on board the Cloud–Aerosol Lidar and Infrared Pathfinder Satellite Obser- vations (CALIPSO) satellite (Winker et al., 2009). These sensors have significantly improved our ability to quantify vertical profiles of cloud occurrence and water content (see Figures 7.5 and 7.6), and comple- ment the detection capabilities of passive multispectral sensors (e.g., Stubenrauch et al., 2010; Chan and Comiso, 2011). Satellite cloud-ob- serving capacities are reviewed by Stubenrauch et al. (2013). Clouds cover roughly two thirds of the globe (Figure 7.5a, c), with a more precise value depending on both the optical depth threshold used to define cloud and the spatial scale of measurement (Wielicki and Parker, 1992; Stubenrauch et al., 2013). The mid-latitude ocean- ic storm tracks and tropical precipitation belts are particularly cloudy, while continental desert regions and the central subtropical oceans are relatively cloud-free. Clouds are composed of liquid at temperatures above 0°C, ice below about –38°C (e.g., Koop et al., 2000), and either or both phases at intermediate temperatures (Figure 7.5b). Throughout most of the troposphere, temperatures at any given altitude are usually warmer in the tropics, but clouds also extend higher there such that ice 580 Chapter 7 Clouds and Aerosols 7 cloud amounts are no less than those at high latitudes. At any given time, most clouds are not precipitating (Figure 7.5d). In this chapter cloud above the 440 hPa pressure level is considered ‘high’, that below the 680 hPa level ‘low’, and that in-between is con- sidered ‘mid-level’. Most high cloud (mainly cirrus and deep cumulus outflows) occurs near the equator and over tropical continents, but can also be seen in the mid-latitude storm track regions and over mid-lati- tude continents in summer (Figure 7.6a, e); it is produced by the storms generating most of the global rainfall in regions where tropospheric air motion is upward, such that dynamical, rainfall and high-cloud fields closely resemble one another (Figure 7.6d, h). Mid-level cloud (Figure 7.6b, f), comprising a variety of types, is prominent in the storm tracks and some occurs in the Intertropical Convergence Zone (ITCZ). Low cloud (Figure 7.6c, g), including shallow cumulus and stratiform cloud, occurs over essentially all oceans but is most prevalent over cooler subtropical oceans and in polar regions. It is less common over land, except at night and in winter. Overlap between cloud layers has long been an issue both for sat- ellite (or ground-based) detection and for calculating cloud radiative effects. Active sensors show more clearly that low clouds are preva- lent in nearly all types of convective systems, and are often under- estimated by models (Chepfer et al., 2008; Naud et al., 2010; Haynes et al., 2011). Cloud layers at different levels overlap less often than typically assumed in General Circulation Models (GCMs), especially Ice (kg m -2 ) 3 9 15 Height (km) 3 9 15 Height (km) 0.2 0.1 Ice Water Path Liquid Water Path c) d) b) 1 0 0.5 0.25 0.75 Fraction (or Occurrence Frequency) Precipitation Occurrence (x2) Cloud Occurrence Cloud Fraction Condensate Path Liquid 0 0 0 60ºS 60ºN 30º 30º Eq 60ºS 60ºN 30º 30º Eq a) 0ºC -38ºC over high-latitude continents and subtropical oceans (Naud et al., 2008; Mace et al., 2009), and the common assumption that the radi- ative effects of precipitating ice can be neglected is not necessarily warranted (Waliser et al., 2011). New observations have led to revised treatments of overlap in some models, which significantly affects cloud radiative effects (Pincus et al., 2006; Shonk et al., 2012). Active sensors have also been useful in detecting low-lying Arctic clouds over sea ice (Kay et al., 2008), improving our ability to test climate model simula- tions of the interaction between sea ice loss and cloud cover (Kay et al., 2011). 7.2.1.2 Effects of Clouds on the Earth’s Radiation Budget The effect of clouds on the Earth’s present-day top of the atmosphere (TOA) radiation budget, or cloud radiative effect (CRE), can be inferred from satellite data by comparing upwelling radiation in cloudy and non-cloudy conditions (Ramanathan et al., 1989). By enhancing the planetary albedo, cloudy conditions exert a global and annual short- wave cloud radiative effect (SWCRE) of approximately –50 W m–2 and, by contributing to the greenhouse effect, exert a mean longwave effect (LWCRE) of approximately +30 W m–2, with a range of 10% or less between published satellite estimates (Loeb et al., 2009). Some of the apparent LWCRE comes from the enhanced water vapour coinciding with the natural cloud fluctuations used to measure the effect, so the true cloud LWCRE is about 10% smaller (Sohn et al., 2010). The net global mean CRE of approximately –20 W m–2 implies a net cooling Figure 7.5 | (a) Annual mean cloud fractional occurrence (CloudSat/CALIPSO 2B-GEOPROF-LIDAR data set for 2006–2011; Mace et al., 2009). (b) Annual zonal mean liquid water path (blue shading, microwave radiometer data set for 1988–2005 from O’Dell et al. (2008)) and total water path (ice path shown with grey shading, from CloudSat 2C-ICE data set for 2006–2011 from Deng et al. (2010) over oceans). The 90% uncertainty ranges, assessed to be approximately 60 to 140% of the mean for the liquid and total water paths, are schematically indicated by the error bars. (c–d) latitude-height sections of annual zonal mean cloud (including precipitation falling from cloud) occurrence and precipitation (attenuation-corrected radar reflectivity >0 dBZ) occurrence; the latter has been doubled to make use of a common colour scale (2B-GEOPROF-LIDAR data set). The dashed curves show the annual mean 0°C and −38°C isotherms. 581 Clouds and Aerosols Chapter 7 7 Figure 7.6 | (a–d) December–January–February mean high, middle and low cloud cover from CloudSat/CALIPSO 2B-GEOPROF R04 and 2B-GEOPROF-LIDAR P1.R04 data sets for 2006–2011 (Mace et al., 2009), 500 hPa vertical pressure velocity (colours, from ERA-Interim for 1979–2010; Dee et al., 2011), and Global Precipitation Climatology Project (GPCP) version 2.2 precipitation rate (1981–2010, grey contours at 3 mm day–1 in dash and 7 mm day–1 in solid); (e–h) same as (a–d), except for June–July–August. For low clouds, the GCM-Oriented CALIPSO Cloud Product (GOCCP) data set for 2007–2010 (Chepfer et al., 2010) is used at locations where it indicates a larger fractional cloud cover, because the GEOPROF data set removes some clouds with tops at altitudes below 750 m. Low cloud amounts are probably underrepresented in regions of high cloud (Chepfer et al., 2008), although not as severely as with earlier satellite instruments. (a) (b) (c) (e) (f) (g) (d) (h) Fraction Mid-troposphere Vertical Pressure Velocity (hPa day -1 ) -50 50 -50 0 25 0 0.8 0.2 0.4 0.6 1 High Cloud Middle Cloud Low Cloud December–January–February June–July–August 582 Chapter 7 Clouds and Aerosols 7 effect of clouds on the current climate. Owing to the large magnitudes of the SWCRE and LWCRE, clouds have the potential to cause signifi- cant climate feedback (Section 7.2.5). The sign of this feedback on cli- mate change cannot be determined from the sign of CRE in the current climate, but depends instead on how climate-sensitive the properties are that govern the LWCRE and SWCRE. The regional patterns of annual-mean TOA CRE (Figure 7.7a, b) reflect those of the altitude-dependent cloud distributions. High clouds, which are cold compared to the clear-sky radiating temperature, dominate patterns of LWCRE, while the SWCRE is sensitive to optically thick clouds at all altitudes. SWCRE also depends on the available sunlight, so for example is sensitive to the diurnal and seasonal cycles of cloud- iness. Regions of deep, thick cloud with large positive LWCRE and large negative SWCRE tend to accompany precipitation (Figure 7.7d), showing their intimate connection with the hydrological cycle. The net CRE is negative over most of the globe and most negative in regions of very extensive low-lying reflective stratus and stratocumulus cloud such as the mid-latitude and eastern subtropical oceans, where SWCRE is strong but LWCRE is weak (Figure 7.7c). In these regions, the spatial distribution of net CRE on seasonal time scales correlates strongly with measures of low-level stability or inversion strength (Klein and Hart- mann, 1993; Williams et al., 2006; Wood and Bretherton, 2006; Zhang et al., 2010). Clouds also exert a CRE at the surface and within the troposphere, thus affecting the hydrological cycle and circulation (Section 7.6), though this aspect of CRE has received less attention. The net downward flux of radiation at the surface is sensitive to the vertical and horizontal distribution of clouds. It has been estimated more accurately through radiation budget measurements and cloud profiling (Kato et al., 2011). Based on these observations, the global mean surface downward long- wave flux is about 10 W m–2 larger than the average in climate models, probably due to insufficient model-simulated cloud cover or lower tropospheric moisture (Stephens et al., 2012). This is consistent with a global mean precipitation rate in the real world somewhat larger than current observational estimates. 7.2.2 Cloud Process Modelling Cloud formation processes span scales from the sub-micrometre scale of CCN, to cloud-system scales of up to thousands of kilometres. This range of scales is impossible to resolve with numerical simulations on computers, and this is not expected to change in the foreseeable future. Nonetheless progress has been made through a variety of mod- elling strategies, which are outlined briefly in this section, followed by a discussion in Section 7.2.3 of developments in representing clouds in global models. The implications of these discussions are synthesized in Section 7.2.3.5. 7.2.2.1 Explicit Simulations in Small Domains High-resolution models in small domains have been widely used to simulate interactions of turbulence with various types of clouds. The grid spacing is chosen to be small enough to resolve explicitly the dom- inant turbulent eddies that drive cloud heterogeneity, with the effects of smaller-scale phenomena parameterized. Such models can be run in (a) (b) (c) (d) (mm day -1 ) Cloud Radiative Effect (W m-2 ) 0 5 10 -100 0 100 -50 50 Shortwave (global mean = –47.3 W m -2 ) Longwave (global mean = 26.2 W m -2 ) Net (global mean = –21.1 W m -2 ) Precipitation (global mean = 2.7 mm day -1 ) Figure 7.7 | Distribution of annual-mean top of the atmosphere (a) shortwave, (b) longwave, (c) net cloud radiative effects averaged over the period 2001–2011 from the Clouds and the Earth’s Radiant Energy System (CERES) Energy Balanced and Filled (EBAF) Ed2.6r data set (Loeb et al., 2009) and (d) precipitation rate (1981–2000 aver- age from the GPCP version 2.2 data set; Adler et al., 2003). 583 Clouds and Aerosols Chapter 7 7 idealized settings, or with boundary conditions for specific observed cases. This strategy is typically called large-eddy simulation (LES) when boundary-layer eddies are resolved, and cloud-resolving model (CRM) when only deep cumulus motions are well resolved. It is useful not only in simulating cloud and precipitation characteristics, but also in understanding how turbulent circulations within clouds transport and process aerosols and chemical constituents. It can be applied to any type of cloud system, on any part of the Earth. Direct numerical simula- tion (DNS) can be used to study turbulence and cloud microphysics on scales of a few metres or less (e.g., Andrejczuk et al., 2006) but cannot span crucial meteorological scales and is not further considered here. Cloud microphysics, precipitation and aerosol interactions are treated with varying levels of sophistication, and remain a weak point in all models regardless of resolution. For example, recent comparisons to satellite data show that liquid water clouds in CRMs generally begin to rain too early in the day (Suzuki et al., 2011). Especially for ice clouds, and for interactions between aerosols and clouds, our understanding of the basic micro-scale physics is not yet adequate, although it is improv- ing. Moreover, microphysical effects are quite sensitive to co-variations of velocity and composition down to very small scales. High-resolution models, such as those used for LES, explicitly calculate most of these variations, and so provide much more of the information needed for microphysical calculations, whereas in a GCM they are not explicitly available. For these reasons, low-resolution (e.g., climate) models will have even more trouble representing local aerosol–cloud interactions than will high-resolution models. Parameterizations are under develop- ment that could account for the small-scale variations statistically (e.g., Larson and Golaz, 2005) but have not been used in the Coupled Model Intercomparison Project Phase 5 (CMIP5) simulations. High-resolution models have enhanced our understanding of cloud processes in several ways. First, they can help interpret in situ and high-resolution remote sensing observations (e.g., Stevens et al., 2005b; Blossey et al., 2007; Fridlind et al., 2007). Second, they have revealed important influences of small-scale interactions, turbulence, entrainment and precipitation on cloud dynamics that must eventu- ally be accounted for in parameterizations (e.g., Krueger et al., 1995; Derbyshire et al., 2004; Kuang and Bretherton, 2006; Ackerman et al., 2009). Third, they can be used to predict how cloud system properties (such as cloud cover, depth, or radiative effect) may respond to cli- mate changes (e.g., Tompkins and Craig, 1998; Bretherton et al., 2013). Fourth, they have become an important tool in testing and improv- ing parameterizations of cloud-controlling processes such as cumulus convection, turbulent mixing, small-scale horizontal cloud variability and aerosol–cloud interactions (Randall et al., 2003; Rio and Hourdin, 2008; Stevens and Seifert, 2008; Lock, 2009; Del Genio and Wu, 2010; Fletcher and Bretherton, 2010), as well as the interplay between con- vection and large-scale circulations (Kuang, 2008). Different aspects of clouds, and cloud types, require different grid reso- lutions. CRMs of deep convective cloud systems with horizontal resolu- tions of 2 km or finer (Bryan et al., 2003) can represent some statistical properties of the cloud system, including fractional area coverage of cloud (Xu et al., 2002), vertical thermodynamic structure (Blossey et al., 2007), the distribution of updraughts and downdraughts (Khair- outdinov et al., 2009) and organization into mesoscale convective systems (Grabowski et al., 1998). Modern high-order turbulence clo- sure schemes may allow some statistics of boundary-layer cloud distri- butions, including cloud fractions and fluxes of moisture and energy, to be reasonably simulated even at horizontal resolution of 1 km or larger (Cheng and Xu, 2006, 2008). Finer grids (down to hundreds of metres) better resolve individual storm characteristics such as vertical velocity or tracer transport. Some cloud ensemble properties remain sensitive to CRM microphysical parameterization assumptions regardless of res- olution, particularly the vertical distribution and optical depth of clouds containing ice. Because of these requirements, it is computationally demanding to run a CRM in a domain large enough to capture convective organisation or perform regional forecasts. Some studies have created smaller regions of CRM-like resolution within realistically forced regional-scale models (e.g., Zhu et al., 2010; Boutle and Abel, 2012; Zhu et al., 2012), a spe- cial case of the common ‘nesting’ approach for regional downscaling (see Section 9.6). One application has been to orographic precipitation, associated both with extratropical cyclones (e.g., Garvert et al., 2005) and with explicitly simulated cumulus convection (e.g., Hohenegger et al., 2008); better resolution of the orography improves the simula- tion of precipitation initiation and wind drift of falling rain and snow between watersheds. LES of shallow cumulus cloud fields with horizontal grid spacing of about 100 m and vertical grid spacing of about 40 m produces vertical profiles of cloud fraction, temperature, moisture and turbulent fluxes that agree well with available observations (Siebesma et al., 2003), though the simulated precipitation efficiency still shows some sensi- tivity to microphysical parameterizations (vanZanten et al., 2011). LES of stratocumulus-topped boundary layers reproduces the turbulence statistics and vertical thermodynamic structure well (e.g., Stevens et al., 2005b; Ackerman et al., 2009), and has been used to study the sensitivity of stratocumulus properties to aerosols (e.g., Savic-Jovcic and Stevens, 2008; Xue et al., 2008) and meteorological conditions. However, the simulated entrainment rate and cloud liquid water path are sensitive to the underlying numerical algorithms, even with vertical grid spacings as small as 5 m, due to poor resolution of the sharp cap- ping inversion (Stevens et al., 2005a). These grid requirements mean that low-cloud processes dominating the known uncertainty in cloud feedback cannot be explicitly simulat- ed except in very small domains. Thus, notwithstanding all of the above benefits of explicit cloud modeling, these models cannot on their own quantify global cloud feedbacks or aerosol–cloud interactions defini- tively. They are important, however, in suggesting and testing feedback and adjustment mechanisms (see Sections 7.2.5 and 7.4). 7.2.2.2 Global Models with Explicit Clouds Since AR4, increasing computer power has led to three types of devel- opments in global atmospheric models. First, models have been run with resolution that is higher than in the past, but not sufficiently high that cumulus clouds can be resolved explicitly. Second, models have been run with resolution high enough to resolve (or ‘permit’) large individual cumulus clouds over the entire globe. In a third approach, the parameterizations of global models have been replaced by 584 Chapter 7 Clouds and Aerosols 7 embedded CRMs. The first approach is assessed in Chapter 9. The other two approaches are discussed below. Global Cloud-Resolving Models (GCRMs) have been run with grid spac- ings as small as 3.5 km (Tomita et al., 2005; Putman and Suarez, 2011). At present GCRMs can be used only for relatively short simulations of a few simulated months to a year or two on the fastest supercomputers, but in the not-too distant future they may provide climate projections. GCRMs provide a consistent way to couple convective circulations to large-scale dynamics, but must still parameterize the effects of individ- ual clouds, microphysics and boundary-layer circulations. Because they avoid the use of uncertain cumulus parameterizations, GCRMs better simulate many properties of convective circulations that are very challenging for many current conventional GCMs, including the diurnal cycles of precipitation (Sato et al., 2009) and the Asian summer monsoon (Oouchi et al., 2009). Inoue et al. (2010) showed that the cloudiness simulated by a GCRM is in good agreement with observations from CloudSat and CALIPSO, but the results are sensitive to the parameterizations of turbulence and cloud microphysics (Satoh et al., 2010; Iga et al., 2011; Kodama et al., 2012). Heterogeneous multiscale methods, in which CRMs are embedded in each grid cell of a larger scale model (Grabowski and Smolarkiewicz, 1999), have also been further developed as a way to realize some of the advantages of GCRMs but at less cost. This approach has come to be known as super-parameterization, because the CRM effectively replaces some of the existing GCM parameterizations (e.g., Khairoutdi- nov and Randall, 2001; Tao et al., 2009). Super-parameterized models, which are sometimes called multiscale modeling frameworks, occupy a middle ground between high-resolution ‘process models’ and ‘climate models’ (see Figure 7.8), in terms of both advantages and cost. Like GCRMs, super-parameterized models give more realistic simula- tions of the diurnal cycle of precipitation (Khairoutdinov et al., 2005; Pritchard and Somerville, 2010) and the Madden-Julian Oscillation (Benedict and Randall, 2009) than most conventional GCMs; they can also improve aspects of the Asian monsoon and the El Niño–Southern Oscillation (ENSO; Stan et al., 2010; DeMott et al., 2011). Moreover, because they also begin to resolve cloud-scale circulations, both strat- egies provide a framework for studying aerosol–cloud interactions that conventional GCMs lack (Wang et al., 2011b). Thus both types of global model provide important insights, but because neither of them fully resolves cloud processes, especially for low clouds (see Section 7.2.2.1), their results must be treated with caution just as with con- ventional GCMs. 7.2.3 Parameterization of Clouds in Climate Models 7.2.3.1 Challenges of Parameterization The representation of cloud microphysical processes in climate models is particularly challenging, in part because some of the fundamen- tal details of these microphysical processes are poorly understood (particularly for ice- and mixed-phase clouds), and because spatial heterogeneity of key atmospheric properties occurs at scales signif- icantly smaller than a GCM grid box. Such representation, however, 10 1 10 5 10 4 10 3 10 2 10 1 10 2 10 3 10 4 10 6 10 5 10 7 General Circulation Model (GCM) Cloud-Resolving Model (CRM) & Large-Eddy Simulation (LES) Climate System Cloud Processes Spatial scale (m) Time scale (day) Super Parameterization (MMF) & Global Cloud-Resolving Model (GCRM) Figure 7.8 | Model and simulation strategy for representing the climate system and climate processes at different space and time scales. Also shown are the ranges of space and time scales usually associated with cloud processes (orange, lower left) and the climate system (blue, upper right). Classes of models are usually defined based on the range of spatial scales they represent, which in the figure is roughly spanned by the text for each model class. The temporal scales simulated by a particular type of model vary more widely. For instance, climate models are often run for a few time steps for diagnostic studies, or can simulate millennia. Hence the figure indicates the typical time scales for which a given model is used. Computational power prevents one model from covering all time and space scales. Since the AR4, the development of Global Cloud Resolving Models (GCRMs), and hybrid approaches such as General Circulation Models (GCMs) using the ‘super-parameterization’ approach (sometimes called the Multiscale Modelling Framework (MMF)), have helped fill the gap between climate system and cloud process models. affects many aspects of a model’s overall simulated climate including the Hadley circulation, precipitation patterns, and tropical variability. Therefore continuing weakness in these parameterizations affects not only modeled climate sensitivity, but also the fidelity with which these other variables can be simulated or projected. Most CMIP5 climate model simulations use horizontal resolutions of 100 to 200 km in the atmosphere, with vertical layers varying between 100 m near the surface to more than 1000 m aloft. Within regions of this size in the real world, there is usually enormous small-scale variability in cloud properties, associated with variability in humidity, temperature and vertical motion (Figure 7.16). This variability must be accounted for to accurately simulate cloud–radiation interaction, condensation, evaporation and precipitation and other cloud processes that crucially depend on how cloud condensate is distributed across each grid box (Cahalan et al., 1994; Pincus and Klein, 2000; Larson et al., 2001; Barker et al., 2003). The simulation of clouds in modern climate models involves several parameterizations that must work in unison. These include parame- terization of turbulence, cumulus convection, microphysical processes, radiative transfer and the resulting cloud amount (including the ver- tical overlap between different grid levels), as well as sub-grid scale transport of aerosol and chemical species. The system of parameter- izations must balance simplicity, realism, computational stability and efficiency. Many cloud processes are unrealistic in current GCMs, and as such their cloud response to climate change remains uncertain. Cloud processes and/or turbulence parameterization are important not only for the GCMs used in climate projections but also for special- ized chemistry–aerosol–climate models (see review by Zhang, 2008), 585 Clouds and Aerosols Chapter 7 7 for regional climate models, and indeed for the cloud process models described in Section 7.2.2 which must still parameterize small-scale and microphysical effects. The nature of the parameterization problem, however, shifts as model scale decreases. Section 7.2.3.2 briefly assess- es recent developments relevant to GCMs. 7.2.3.2 Recent Advances in Representing Cloud Microphysical Processes 7.2.3.2.1 Liquid clouds Recent development efforts have been focused on the introduction of more complex representations of microphysical processes, with the dual goals of coupling them better to atmospheric aerosols and link- ing them more consistently to the sub-grid variability assumed by the model for other calculations. For example, most CMIP3 climate models predicted the average cloud and rain water mass in each grid cell only at a given time, diagnosing the droplet concentration using empiri- cal relationships based on aerosol mass (e.g., Boucher and Lohmann, 1995; Menon et al., 2002), or altitude and proximity to land. Many were forced to employ an arbitrary lower bound on droplet concentra- tion to reduce the aerosol RF (Hoose et al., 2009). Such formulations oversimplify microphysically mediated cloud variations. By contrast, more models participating in CMIP5 predict both mass and number mixing ratios for liquid stratiform cloud. Some determine rain and snow number concentrations and mixing ratios (e.g., Morrison and Gettelman, 2008; Salzmann et al., 2010), allowing treatment of aerosol scavenging and the radiative effect of snow. Some models explicitly treat sub-grid cloud water variability for calculating microphysical pro- cess rates (e.g., Morrison and Gettelman, 2008). Cloud droplet activa- tion schemes now account more realistically for particle composition, mixing and size (Abdul-Razzak and Ghan, 2000; Ghan et al., 2011; Liu et al., 2012). Despite such advances in internal consistency, a con- tinuing weakness in GCMs (and to a much lesser extent GCRMs and super-parameterized models) is their inability to fully represent turbu- lent motions to which microphysical processes are highly sensitive. 7.2.3.2.2 Mixed-phase and ice clouds Ice treatments are following a path similar to those for liquid water, and face similar but greater challenges because of the greater com- plexity of ice processes. Many CMIP3 models predicted the condensed water amount in just two categories—cloud and precipitation—with a temperature-dependent partitioning between liquid and ice within either category. Although supersaturation with respect to ice is com- monly observed at low temperatures, only one CMIP3 GCM (ECHAM) allowed ice supersaturation (Lohmann and Kärcher, 2002). Many climate models now include separate, physically based equations for cloud liquid versus cloud ice, and for rain versus snow, allowing a more realistic treatment of mixed-phase processes and ice supersatu- ration (Liu et al., 2007; Tompkins et al., 2007; Gettelman et al., 2010; Salzmann et al., 2010; see also Section 7.4.4). These new schemes are tested in a single-column model against cases observed in field cam- paigns (e.g., Klein et al., 2009) or against satellite observations (e.g., Kay et al., 2012), and provide superior simulations of cloud structure than typical CMIP3 parameterizations (Kay et al., 2012). However new observations reveal complexities not correctly captured by even relatively advanced schemes (Ma et al., 2012a). New representations of the Wegener–Bergeron–Findeisen process in mixed-phase clouds (Storelvmo et al., 2008b; Lohmann and Hoose, 2009) compare the rate at which the pre-existing ice crystals deplete the water vapour with the condensation rate for liquid water driven by vertical updraught speed (Korolev, 2007); these are not yet included in CMIP5 models. Climate models are increasingly representing detailed microphysics, including mixed-phase processes, inside convective clouds (Fowler and Randall, 2002; Lohmann, 2008; Song and Zhang, 2011). Such processes can influence storm characteristics like strength and electrification, and are crucial for fully representing aerosol–cloud interactions, but are still not included in most climate models; their representation is moreover subject to all the caveats noted in Section 7.2.3.1. 7.2.3.3 Recent Advances in Parameterizing Moist Turbulence and Convection Both the mean state and variability in climate models are sensitive to the parameterization of cumulus convection. Since AR4, the develop- ment of convective parameterization has been driven largely by rapidly growing use of process models, in particular LES and CRMs, to inform parameterization development (e.g., Hourdin et al., 2013). Accounting for greater or more state-dependent entrainment of air into deep cumulus updraughts has improved simulations of the Madden– Julian Oscillation, tropical convectively coupled waves and mean rain- fall patterns in some models (Bechtold et al., 2008; Song and Zhang, 2009; Chikira and Sugiyama, 2010; Hohenegger and Bretherton, 2011; Mapes and Neale, 2011; Del Genio et al., 2012; Kim et al., 2012) but usually at the expense of a degraded simulation of the mean state. In another model, revised criteria for convective initiation and parame- terizations of cumulus momentum fluxes improved ENSO and tropical vertical temperature profiles (Neale et al., 2008; Richter and Rasch, 2008). Since AR4, more climate models have adopted cumulus param- eterizations that diagnose the expected vertical velocity in cumulus updraughts (e.g., Del Genio et al., 2007; Park and Bretherton, 2009; Chikira and Sugiyama, 2010; Donner et al., 2011), in principle allowing more complete representations of aerosol activation, cloud microphys- ical evolution and gravity wave generation by the convection. Several new parameterizations couple shallow cumulus convection more closely to moist boundary layer turbulence (Siebesma et al., 2007; Neggers, 2009; Neggers et al., 2009; Couvreux et al., 2010) including cold pools generated by nearby deep convection (Grandpeix and Lafore, 2010). Many of these efforts have led to more accurate simulations of boundary-layer cloud radiative properties and vertical structure (e.g., Park and Bretherton, 2009; Köhler et al., 2011), and have ameliorated the common problem of premature deep convective initiation over land in one CMIP5 GCM (Rio et al., 2009). 7.2.3.4 Recent Advances in Parameterizing Cloud Radiative Effects Some models have improved representation of sub-grid scale cloud variability, which has important effects on grid-mean radiative fluxes 586 Chapter 7 Clouds and Aerosols 7 and precipitation fluxes, for example, based on the use of probability density functions of thermodynamic variables (Sommeria and Dear- dorff, 1977; Watanabe et al., 2009). Stochastic approaches for radi- ative transfer can account for this variability in a computationally efficient way (Barker et al., 2008). New treatments of cloud overlap have been motivated by new observations (Section 7.2.1.1). Despite these advances, the CMIP5 models continue to exhibit the ‘too few, too bright’ low-cloud problem (Nam et al., 2012), with a systematic over- estimation of cloud optical depth and underestimation of cloud cover. 7.2.3.5 Cloud Modelling Synthesis Global climate models used in CMIP5 have improved their represen- tation of cloud processes relative to CMIP3, but still face challenges and uncertainties, especially regarding details of small-scale variability that are crucial for aerosol–cloud interactions (see Section 7.4). Finer- scale LES and CRM models are much better able to represent this vari- ability and are an important research tool, but still suffer from imper- fect representations of aerosol and cloud microphysics and known biases. Most CRM and LES studies do not span the large space and time scales needed to fully determine the interactions among differ- ent cloud regimes and the resulting net planetary radiative effects. Thus our assessments in this chapter do not regard any model type on its own as definitive, but weigh the implications of process model studies in assessing the quantitative results of the global models. 7.2.4 Water Vapour and Lapse Rate Feedbacks Climate feedbacks determine the sensitivity of global surface temper- ature to external forcing agents. Water vapour, lapse rate and cloud feedbacks each involve moist atmospheric processes closely linked to clouds, and in combination, produce most of the simulated climate feedback and most of its inter-model spread (Section 9.7). The radia- tive feedback from a given constituent can be quantified as its impact (other constituents remaining equal) on the TOA net downward radi- ative flux per degree of global surface (or near-surface) temperature increase, and may be compared with the basic ‘black-body’ response of −3.4 W m−2 °C−1 (Hansen et al., 1984). This definition assigns posi- tive values to positive feedbacks, in keeping with the literature on this topic but contradictory to the conventions sometimes adopted in other climate research. 7.2.4.1 Water Vapour Response and Feedback As pointed out in previous reports (Section 8.6.3.1 in Randall et al., 2007), physical arguments and models of all types suggest global water vapour amounts increase in a warmer climate, leading to a positive feedback via its enhanced greenhouse effect. The saturated water vapour mixing ratio (WVMR) increases nearly exponentially and very rapidly with temperature, at 6 to 10% °C–1 near the surface, and even more steeply aloft (up to 17% °C–1) where air is colder. Mounting evidence indicates that any changes in relative humidity in warmer climates would have much less impact on specific humidity than the above increases, at least in a global and statistical sense. Hence the overall WVMR is expected to increase at a rate similar to the saturated WVMR. Because global temperatures have been rising, the above arguments imply WVMR should be rising accordingly, and multiple observing sys- tems indeed show this (Sections 2.5.4 and 2.5.5). A study challenging the water vapour increase (Paltridge et al., 2009) used an old reanalysis product, whose trends are contradicted by newer ones (Dessler and Davis, 2010) and by actual observations (Chapter 2). The study also reported decreasing relative humidity in data from Australian radio- sondes, but more complete studies show Australia to be exceptional in this respect (Dai et al., 2011). Thus data remain consistent with the expected global feedback. Some studies have proposed that the response of upper-level humid- ity to natural fluctuations in the global mean surface temperature is informative about the feedback. However, small changes to the global mean (primarily from ENSO) involve geographically heterogeneous temperature change patterns, the responses to which may be a poor analogue for global warming (Hurley and Galewsky, 2010a). Most climate models reproduce these natural responses reasonably well (Gettelman and Fu, 2008; Dessler and Wong, 2009), providing addi- tional evidence that they at least represent the key processes. The ‘last-saturation’ concept approximates the WVMR of air by its sat- uration value when it was last in a cloud (see Sherwood et al., 2010a for a review), which can be inferred from trajectory analysis. Studies since the AR4 using a variety of models and observations (including concentrations of water vapour isotopes) support this concept (Sher- wood and Meyer, 2006; Galewsky and Hurley, 2010). The concept has clarified what determines relative humidity in the subtropical upper troposphere and placed the water vapour feedback on firmer theo- retical footing by directly linking actual and saturation WVMR values (Hurley and Galewsky, 2010b). CRMs show that convection can adopt varying degrees of self-aggregation (e.g., Muller and Held, 2012), which could modify the water vapour or other feedbacks if this were climate sensitive, although observations do not suggest aggregation changes have a large net radiative effect (Tobin et al., 2012). In a warmer climate, an upward shift of the tropopause and poleward shift of the jets and associated climate zones are expected (Sections 2.7.4 and 2.7.5) and simulated by most GCMs (Section 10.3.3). These changes account, at least qualitatively, for robust regional changes in the relative humidity simulated in warmer climate by GCMs, includ- ing decreases in the subtropical troposphere and tropical uppermost troposphere, and increases near the extratropical tropopause and high latitudes (Sherwood et al., 2010b). This pattern may be amplified, however, by non-uniform atmospheric temperature or wind changes (Hurley and Galewsky, 2010b). It is also the apparent cause of most model-predicted changes in mid- and upper-level cloudiness patterns (Wetherald and Manabe, 1980; Sherwood et al., 2010b; see also Sec- tion 7.2.5.2). Idealized CRM simulations of warming climates also show upward shifts of the humidity patterns with little change in the mean (e.g., Kuang and Hartmann, 2007; Romps, 2011). It remains unclear whether stratospheric water vapour contributes significantly to climate feedback. Observations have shown decadal variations in stratospheric water vapour, which may have affected the planetary radiation budget somewhat (Solomon et al., 2010) but are not clearly linked to global temperature (Section 3.4.2.4 in Trenberth et 587 Clouds and Aerosols Chapter 7 7 al., 2007). A strong positive feedback from stratospheric water vapour was reported in one GCM, but with parameter settings that produced an unrealistic present climate (Joshi et al., 2010). 7.2.4.2 Relationship Between Water Vapour and Lapse Rate Feedbacks The lapse rate (decrease of temperature with altitude) should, in the tropics, change roughly as predicted by a moist adiabat, due to the strong restoring influence of convective heating. This restoring influ- ence has now been directly inferred from satellite data (Lebsock et al., 2010), and the near-constancy of tropical atmospheric stability and deep-convective thresholds over recent decades is also now observ- able in SST and deep convective data (Johnson and Xie, 2010). The stronger warming of the atmosphere relative to the surface produces a negative feedback on global temperature because the warmed system radiates more thermal emission to space for a given increase in surface temperature than in the reference case where the lapse rate is fixed. This feedback varies somewhat among models because lapse rates in middle and high latitudes, which decrease less than in the tropics, do so differently among models (Dessler and Wong, 2009). As shown by Cess (1975) and discussed in the AR4 (Randall et al., 2007), models with a more negative lapse rate feedback tend to have a more positive water vapour feedback. Cancellation between these is close enough that their sum has a 90% range in CMIP3 models of only +0.96 to +1.22 W m−2 °C−1 (based on a Gaussian fit to the data of Held and Shell (2012), see Figure 7.9) with essentially the same range in CMIP5 (Section 9.7). The physical reason for this cancellation is that as long as water vapour infrared absorption bands are nearly saturat- ed, outgoing longwave radiation is determined by relative humidity (Ingram, 2010) which exhibits little global systematic change in any model (Section 7.2.4.1). In fact, Held and Shell (2012) and Ingram (2013a) argue that it makes more sense physically to redefine feed- backs in a different analysis framework in which relative humidity, -3 -2 -1 0 1 2 3 Total Planck Lapse WVMR Planck RH LapseRH RH Standard Decomposition Feedback (W m -2 ºC -1 ) RH-based Decomposition Figure 7.9 | Feedback parameters associated with water vapour or the lapse rate predicted by CMIP3 GCMs, with boxes showing interquartile range and whiskers show- ing extreme values. At left is shown the total radiative response including the Planck response. In the darker shaded region is shown the traditional breakdown of this into a Planck response and individual feedbacks from water vapour (labelled ‘WVMR’) and lapse rate (labelled ‘Lapse’). In the lighter-shaded region at right are the equivalent three parameters calculated in an alternative, relative humidity-based framework. In this framework all three components are both weaker and more consistent among the models. (Data are from Held and Shell, 2012.) rather than specific humidity, is the feedback variable. Analysed in that framework the inherent stabilization by the Planck response is weaker, but the water vapour and lapse rate feedbacks are also very small; thus the traditional view of large and partially compensating feedbacks has, arguably, arisen from arbitrary choices made when the analysis frame- work was originally set out, rather than being an intrinsic feature of climate or climate models. There is some observational evidence (Section 2.4.4) suggesting trop- ical lapse rates might have increased in recent decades in a way not simulated by models (Section 9.4.1.4.2). Because the combined lapse rate and water vapour feedback depends on relative humidity change, however, the imputed lapse rate variations would have little influence on the total feedback or climate sensitivity even if they were a real warming response (Ingram, 2013b). In summary, there is increased evidence for a strong, positive feedback (measured in the tradition- al framework) from the combination of water vapour and lapse rate changes since AR4, with no reliable contradictory evidence. 7.2.5 Cloud Feedbacks and Rapid Adjustments to Carbon Dioxide The dominant source of spread among GCM climate sensitivities in AR4 was due to diverging cloud feedbacks, particularly due to low clouds, and this continues to be true (Section 9.7). All global models continue to produce a near-zero to moderately strong positive net cloud feedback. Progress has been made since the AR4 in understanding the reasons for positive feedbacks in models and providing a stronger theoretical and observational basis for some mechanisms contributing to them. There has also been progress in quantifying feedbacks—including separating the effects of different cloud types, using radiative-kernel residual methods (Soden et al., 2008) and by computing cloud effects directly (e.g., Zelinka et al., 2012a)—and in distinguishing between feedback and adjustment responses (Section 7.2.5.6). Until very recently cloud feedbacks have been diagnosed in models by differencing cloud radiative effects in doubled CO2 and control cli- mates, normalized by the change in global mean surface temperature. Different diagnosis methods do not always agree, and some simple methods can make positive cloud feedbacks look negative by failing to account for the nonlinear interaction between cloud and water vapour (Soden and Held, 2006). Moreover, it is now recognized that some of the cloud changes are induced directly by the atmospheric radiative effects of CO2 independently of surface warming, and are therefore rapid adjustments rather than feedbacks (Section 7.2.5.6). Most of the published studies available for this assessment did not separate these effects, and only the total response is assessed here unless otherwise noted. It appears that the adjustments are sufficiently small in most models that general conclusions regarding feedbacks are not signifi- cantly affected. Cloud changes cause both longwave (greenhouse warming) and short- wave (reflective cooling) effects, which combine to give the overall cloud feedback or forcing adjustment. Cloud feedback studies point to five aspects of the cloud response to climate change which are distinguished here: changes in high-level cloud altitude, effects of hydrological cycle and storm track changes on cloud systems, changes 588 Chapter 7 Clouds and Aerosols 7 in low-level cloud amount, microphysically induced opacity (optical depth) changes and changes in high-latitude clouds. Finally, recent research on the rapid cloud adjustments to CO2 is assessed. Feedbacks involving aerosols (Section 7.3.5) are not considered here, and the discussion focuses only on mechanisms affecting the TOA radiation budget. 7.2.5.1 Feedback Mechanisms Involving the Altitude of High-Level Cloud A dominant contributor of positive cloud feedback in models is the increase in the height of deep convective outflows tentatively attribut- ed in AR4 to the so-called ‘fixed anvil-temperature’ mechanism (Hart- mann and Larson, 2002). According to this mechanism, the average outflow level from tropical deep convective systems is determined in steady state by the highest point at which water vapour cools the atmosphere significantly through infrared emission; this occurs at a particular water vapour partial pressure, therefore at a similar temper- ature (higher altitude) as climate warms. A positive feedback results because, since the cloud top temperature does not keep pace with that of the troposphere, its emission to space does not increase at the rate expected for the no-feedback system. This occurs at all latitudes and has long been noted in model simulations (Hansen et al., 1984; Cess et al., 1990). This mechanism, with a small modification to account for lapse rate changes, predicts roughly +0.5 W m–2 °C–1 of positive long- wave feedback in GCMs (Zelinka and Hartmann, 2010), compared to an overall cloud-height feedback of +0.35 (+0.09 to +0.58) W m–2 °C–1 (Figure 7.10). Importantly, CRMs also reproduce this increase in cloud height (Tompkins and Craig, 1998; Kuang and Hartmann, 2007; Romps, 2011; Harrop and Hartmann, 2012). On average, natural fluctuations in tropical high cloud amount exert little net TOA radiative effect in the current climate due to near- -1.0 -0.5 0.0 0.5 1.0 1.5 Total High Middle Low Amount Height Opacity CFMIP CMIP3 CMIP5 Net LW SW Feedback (W m -2 ºC -1 ) CFMIP Models (by cloud property) CFMIP & CMIP3 Models (by cloud level) Figure 7.10 | Cloud feedback parameters as predicted by GCMs for responses to CO2 increase including rapid adjustments. Total feedback shown at left, with centre light- shaded section showing components attributable to clouds in specific height ranges (see Section 7.2.1.1), and right dark-shaded panel those attributable to specific cloud property changes where available. The net feedback parameters are broken down in their longwave (LW) and shortwave (SW) components. Type attribution reported for CMIP3 does not conform exactly to the definition used in the Cloud Feedback Model Intercomparison Project (CFMIP) but is shown for comparison, with their ‘mixed’ cat- egory assigned to middle cloud. CFMIP data (original and CFMIP2) are from Zelinka et al. (2012a, 2012b; 2013); CMIP3 from Soden and Vecchi (2011); and CMIP5 from Tomassini et al. (2013). compensation between their longwave and shortwave cloud radiative effects (Harrison et al., 1990; Figure 7.7). Similar compensation can be seen in the opposing variations of these two components of the high-cloud feedback across GCMs (Figure 7.10). This might suggest that the altitude feedback could be similarly compensated. However, GCMs can reproduce the observed compensation in the present cli- mate (Sherwood et al., 1994) without producing one under global warming. In the above-noted cloud-resolving simulations, the entire cloud field (including the typical base) moved upward, in accord with a general upward shift of tropospheric fields (Singh and O’Gorman, 2012) and with drying at levels near cloud base (Minschwaner et al., 2006; Sherwood et al., 2010b). This supports the prediction of GCMs that the altitude feedback is not compensated by an increase in high- cloud thickness or albedo. The observational record offers limited further support for the altitude increase. The global tropopause is rising as expected (Section 2.7.4). Observed cloud heights change roughly as predicted with regional, seasonal and interannual changes in near-tropopause temperature structure (Xu et al., 2007; Eitzen et al., 2009; Chae and Sherwood, 2010; Zelinka and Hartmann, 2011), although these tests may not be good analogues for global warming. Davies and Molloy (2012) report an apparent recent downward mean cloud height trend but this is probably an artefact (Evan and Norris, 2012); observed cloud height trends do not appear sufficiently reliable to test this cloud-height feed- back mechanism (Section 2.5.6). In summary, the consistency of GCM responses, basic understanding, strong support from process models, and weak further support from observations give us high confidence in a positive feedback contribu- tion from increases in high-cloud altitude. 7.2.5.2 Feedback Mechanisms Involving the Amount of Middle and High Cloud As noted in Section 7.2.5.1, models simulate a range of nearly compen- sating differences in shortwave and longwave high-cloud feedbacks, consistent with different changes in high-cloud amount, but also show a net positive offset consistent with higher cloud altitude (Figure 7.10). However, there is a tendency in most GCMs toward reduced middle and high cloud amount in warmer climates in low- and mid-latitudes, especially in the subtropics (Trenberth and Fasullo, 2009; Zelinka and Hartmann, 2010). This loss of cloud amount adds a positive shortwave and negative longwave feedback to the model average, which causes the average net positive feedback to appear to come from the short- wave part of the spectrum. The net effect of changes in amount of all cloud types averaged over models is a positive feedback of about +0.2 W m–2 °C–1, but this roughly matches the contribution from low clouds (see the following section), implying a near-cancellation of longwave and shortwave effects for the mid- and high-level amount changes. Changes in predicted cloud cover geographically correlate with sim- ulated subtropical drying (Meehl et al., 2007), suggesting that they are partly tied to large-scale circulation changes including the pole- ward shifts found in most models (Wetherald and Manabe, 1980; Sher- wood et al., 2010b; Section 2.7). Bender et al. (2012) and Eastman and Warren (2013) report poleward shifts in cloud since the 1970s 589 Clouds and Aerosols Chapter 7 7 consistent with those reported in other observables (Section 2.5.6) and simulated by most GCMs, albeit with weaker amplitude (Yin, 2005). This shift of clouds to latitudes of weaker sunlight decreases the plan- etary albedo and would imply a strong positive feedback if it were due to global warming (Bender et al., 2012), although it is probably partly driven by other factors (Section 10.3). The true amount of positive feed- back coming from poleward shifts therefore remains highly uncertain, but is underestimated by GCMs if, as suggested by observational com- parisons, the shifts are underestimated (Johanson and Fu, 2009; Allen et al., 2012). The upward mass flux in deep clouds should decrease in a warmer climate (Section 7.6.2), which might contribute to cloudiness decreases in storm tracks or the ITCZ (Chou and Neelin, 2004; Held and Soden, 2006). Tselioudis and Rossow (2006) predict this within the storm tracks based on observed present-day relationships with meteorologi- cal variables combined with model-simulated changes to those driving variables but do not infer a large feedback. Most CMIP3 GCMs produce too little storm-track cloud cover in the southern hemisphere compared to nearly overcast conditions in reality, but clouds are also too bright. Arguments have been advanced that such biases could imply either model overestimation or underestimation of feedbacks (Trenberth and Fasullo, 2010; Brient and Bony, 2012). The role of thin cirrus clouds for cloud feedback is not known and remains a source of possible systematic bias. Unlike high-cloud sys- tems overall, these particular clouds exert a clear net warming effect (Jensen et al., 1994; Chen et al., 2000), making a significant cloud- cover feedback possible in principle (e.g., Rondanelli and Lindzen, 2010). While this does not seem to be important in recent GCMs (Zelinka et al., 2012b), and no specific mechanism has been suggested, the representation of cirrus in GCMs appears to be poor (Eliasson et al., 2011) and such clouds are microphysically complex (Section 7.4.4). This implies significant feedback uncertainty in addition to that already evident from model spread. Model simulations, physical understanding and observations thus pro- vide medium confidence that poleward shifts of cloud distributions will contribute to positive feedback, but by an uncertain amount. Feed- backs from thin cirrus amount cannot be ruled out and are an impor- tant source of uncertainty. 7.2.5.3 Feedback Mechanisms Involving Low Cloud Differences in the response of low clouds to a warming are responsible for most of the spread in model-based estimates of equilibrium climate sensitivity (Randall et al., 2007). Since the AR4 this finding has with- stood further scrutiny (e.g., Soden and Vecchi, 2011; Webb et al., 2013), holds in CMIP5 models (Vial et al., 2013) and has been shown to apply also to the transient climate response (e.g., Dufresne and Bony, 2008). This discrepancy in responses occurs over most oceans and cannot be clearly confined to any single region (Trenberth and Fasullo, 2010; Webb et al., 2013), but is usually associated with the representation of shallow cumulus or stratocumulus clouds (Williams and Tselioudis, 2007; Williams and Webb, 2009; Xu et al., 2010). Because the spread of responses emerges in a variety of idealized model formulations (Medeiros et al., 2008; Zhang and Bretherton, 2008; Brient and Bony, 2013), or conditioned on a particular dynamical state (Bony et al., 2004), and is similar in equilibrium or transient simulations (Yokohata et al., 2008), it appears to be attributable to how cloud, convective and boundary layer processes are parameterized in GCMs. The modelled response of low clouds does not appear to be dominated by a single feedback mechanism, but rather the net effect of sever- al potentially competing mechanisms as elucidated in LES and GCM sensitivity studies (e.g., Zhang and Bretherton, 2008; Blossey et al., 2013; Bretherton et al., 2013). Starting with some proposed negative feedback mechanisms, it has been argued that in a warmer climate, low clouds will be: (1) horizontally more extensive, because changes in the lapse rate of temperature also modify the lower-tropospheric stability (Miller, 1997); (2) optically thicker, because adiabatic ascent is accompanied by a larger condensation rate (Somerville and Remer, 1984); and (3) vertically more extensive, in response to a weakening of the tropical overturning circulation (Caldwell and Bretherton, 2009). While these mechanisms may play some role in subtropical low cloud feedbacks, none of them appears dominant. Regarding (1), dry static stability alone is a misleading predictor with respect to climate chang- es, as models with comparably good simulations of the current region- al distribution and/or relationship to stability of low cloud can produce a broad range of cloud responses to climate perturbations (Wyant et al., 2006). Mechanism (2), discussed briefly in the next section, appears to have a small effect. Mechanism (3) cannot yet be ruled out but does not appear to be the dominant factor in determining subtropical cloud changes in GCMs (Bony and Dufresne, 2005; Zhang and Bretherton, 2008). Since the AR4, several new positive feedback mechanisms have been proposed, most associated with the marine boundary layer clouds thought to be at the core of the spread in responses. These include the ideas that: warming-induced changes in the absolute humidity lapse rate change the energetics of mixing in ways that demand a reduction in cloud amount or thickness (Webb and Lock, 2013; Bretherton et al., 2013; Brient and Bony, 2013); energetic constraints prevent the sur- face evaporation from increasing with warming at a rate sufficient to balance expected changes in dry air entrainment, thereby reducing the supply of moisture to form clouds (Rieck et al., 2012; Webb and Lock, 2013); and that increased concentrations of GHGs reduce the radia- tive cooling that drives stratiform cloud layers and thereby the cloud amount (Caldwell and Bretherton, 2009; Stevens and Brenguier, 2009; Bretherton et al., 2013). These mechanisms, crudely operating through parameterized representations of cloud processes, could explain why climate models consistently produce positive low-cloud feedbacks. Among CFMIP GCMs, the low-cloud feedback ranges from −0.09 to +0.63 W m–2 °C–1 (Figure 7.10), and is largely associated with a reduc- tion in low-cloud amount, albeit with considerable spatial variabili- ty (e.g., Webb et al., 2013). One ‘super-parameterized’ GCM (Section 7.2.2.2) simulates a negative low-cloud feedback (Wyant et al., 2006, 2009), but that model’s representation of low clouds was worse than some conventional GCMs. The tendency of both GCMs and process models to produce these positive feedback effects suggests that the feedback contribution from changes in low clouds is positive. However, deficient representa- tion of low clouds in GCMs, diverse model results, a lack of reliable 590 Chapter 7 Clouds and Aerosols 7 observational constraints, and the tentative nature of the suggested mechanisms leave us with low confidence in the sign of the low-cloud feedback contribution. 7.2.5.4 Feedbacks Involving Changes in Cloud Opacity It has long been suggested that cloud water content could increase in a warmer climate simply due to the availability of more vapour for con- densation in a warmer atmosphere, yielding a negative feedback (Pal- tridge, 1980; Somerville and Remer, 1984), but this argument ignores the physics of crucial cloud-regulating processes such as precipitation formation and turbulence. Observational evidence discounting a large effect of this kind was reported in AR4 (Randall et al., 2007). The global mean net feedback from cloud opacity changes in CFMIP models (Figure 7.10) is approximately zero. Optical depths tend to reduce slightly at low and middle latitudes, but increase poleward of 50°, yielding a positive longwave feedback that roughly offsets the negative shortwave feedback. These latitude-dependent opacity changes may be attributed to phase changes at high latitudes and greater poleward moisture transport (Vavrus et al., 2009), and possibly to poleward shifts of the circulation. Studies have reported warming-related changes in cloud opacity tied to cloud phase (e.g., Senior and Mitchell, 1993; Tsushima et al., 2006). This might be expected to cause negative feedback, because at mixed- phase temperatures of –38 to 0°C, cloud ice particles have typical diameters of 10 to 100 μm (e.g., Figure 8 in Donovan, 2003), sever- al-fold larger than cloud water drops, so a given mass of