Arctic amplification of climate change: a review of underlying mechanisms - IOPscience
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
TOPICAL REVIEW • OPEN ACCESS Arctic amplification of climate change: a review of underlying mechanisms To cite this article: Michael Previdi et al 2021 Environ. Res. Lett. 16 093003 View the article online for updates and enhancements. This content was downloaded from IP address 46.4.80.155 on 11/10/2021 at 13:36
Environ. Res. Lett. 16 (2021) 093003 https://doi.org/10.1088/1748-9326/ac1c29 TOPICAL REVIEW Arctic amplification of climate change: a review of underlying OPEN ACCESS mechanisms RECEIVED 6 February 2019 Michael Previdi1,∗, Karen L Smith1,2 and Lorenzo M Polvani1,3 REVISED 1 2 August 2021 Lamont-Doherty Earth Observatory of Columbia University, Palisades, NY 10964, United States of America 2 Department of Physical and Environmental Sciences, University of Toronto Scarborough, Toronto, Ontario M1C 1A4, Canada ACCEPTED FOR PUBLICATION 3 Department of Applied Physics and Applied Mathematics and Department of Earth and Environmental Sciences, Columbia University, 10 August 2021 New York, NY 10027, United States of America PUBLISHED ∗ Author to whom any correspondence should be addressed. 2 September 2021 E-mail: mprevidi@ldeo.columbia.edu Original content from Keywords: arctic amplification, polar amplification, climate change, climate feedbacks this work may be used under the terms of the Creative Commons Attribution 4.0 licence. Abstract Any further distribution of this work must Arctic amplification (AA)—referring to the enhancement of near-surface air temperature change maintain attribution to the author(s) and the title over the Arctic relative to lower latitudes—is a prominent feature of climate change with important of the work, journal impacts on human and natural systems. In this review, we synthesize current understanding of the citation and DOI. underlying physical mechanisms that can give rise to AA. These mechanisms include both local feedbacks and changes in poleward energy transport. Temperature and sea ice-related feedbacks are especially important for AA, since they are significantly more positive over the Arctic than at lower latitudes. Changes in energy transport by the atmosphere and ocean can also contribute to AA. These energy transport changes are tightly coupled with local feedbacks, and thus their respective contributions to AA should not be considered in isolation. It is here emphasized that the feedbacks and energy transport changes that give rise to AA are sensitively dependent on the state of the climate system itself. This implies that changes in the climate state will lead to changes in the strength of AA, with implications for past and future climate change. 1. Introduction simulate tropically-amplified warming in the upper troposphere that is not seen in observations (see also Arctic amplification (AA) of climate change is Po-Chedley and Fu 2012, Mitchell et al 2013, 2020, a prominent feature of the paleoclimatic record Santer et al 2013). We note, however, that the level of (Hoffert and Covey 1992, Miller et al 2010), present- agreement between modeled and observed tropical day observations (Serreze et al 2009, Bekryaev et al tropospheric temperature trends may depend on the 2010, Wang et al 2016, Richter-Menge et al 2019), choice of observational dataset. For example, Santer and model projections of future climate (Intergov- et al (2017) found that observed trends based on one ernmental Panel on Climate Change (IPCC) 2013, updated satellite dataset agreed well with the trends Davy and Outten 2020). AA refers to the enhance- simulated by CMIP5 models. ment of near-surface air temperature change in Whatever the cause(s) of model-data differences the Arctic relative to lower latitudes and the global (e.g. internal variability, model bias, observational mean, a feature that is readily apparent in obser- uncertainty), these features of recent climate change vations over recent decades (figure 1(a)). Climate are qualitatively consistent with the expected response models from the fifth Coupled Model Intercompar- to increases in atmospheric CO2 (figure 1(c)). As will ison Project (CMIP5) reproduce the observed pattern be discussed, however, AA is a robust response of the of polar-amplified warming (figure 1(b)), although Earth system to climate forcing4 , and other forcings the models on average simulate less warming in besides CO2 very likely played a role in determining the Arctic lower troposphere and more warming in the tropical lower troposphere (and thus weaker AA) compared to observations. CMIP5 models also 4 See section 3.1 for definitions of climate forcing and feedback. © 2021 The Author(s). Published by IOP Publishing Ltd
Environ. Res. Lett. 16 (2021) 093003 M Previdi et al Figure 1. (a) Latitude-pressure plot of 1979–2018 annual mean air temperature change (in K) based on linear trends (mean change of three atmospheric reanalyses: ERA5, JRA-55 and MERRA-2). (b) As in figure 1(a), but for the mean change of 12 CMIP5 models (bcc-csm1-1, CanESM2, CCSM4, CSIRO-Mk3-6-0, GFDL-CM3, inmcm4, IPSL-CM5A-LR, MIROC5, MPI-ESM-LR, MPI-ESM-MR, MRI-CGCM3 and NorESM1-M). For each model, 1979–2018 time series were created by concatenating the first ensemble member (r1i1p1) of the historical simulation with the first ensemble member of the RCP8.5 simulation. (c) Latitude-pressure plot of annual mean air temperature change (in K) based on the difference between the last 30 years of the CMIP5 abrupt 4 × CO2 and pre- industrial control simulations (mean change of the 12 models listed in figure 1(b)). Note the different color scale in figure 1(c) compared to figures 1(a) and (b). the magnitude of AA seen in recent decades (e.g. and other alteration to infrastructure (e.g. roadways, Najafi et al 2015, Polvani et al 2020). This polar- buildings), and by leading to potentially dangerous amplified warming has occurred in all seasons except conditions for hunting, recreation and other activities boreal summer, with the strongest warming in fall (e.g. Hovelsrud et al 2011, Schneider Von Deimling and winter (figures 2(a) and (b)). Once again, this is et al 2021). On the other hand, climate change offers in qualitative agreement with the expected response some benefits to local communities such as increased to CO2 forcing (figure 2(c)). opportunity for marine shipping, commercial fisher- AA has had and will have important impacts on ies, tourism, and access to resources as a result of the a range of human and natural systems, both within lengthening open water season in the Arctic Ocean and outside the Arctic. Within the Arctic, enhanced (e.g. Melia et al 2016). warming and the associated melting of snow and ice Natural systems within the Arctic are similarly present both challenges and opportunities to local affected by the pace and magnitude of climate warm- human populations. Many communities face risks ing. Changes in many animals’ feeding, mating and to food and water security as a result of changes in migration habits are altering species distribution and access to hunting areas and the distribution of tradi- abundance, introducing non-native species to the tional food sources, contamination of drinking water, Arctic and in some cases threatening native spe- changes in traditional food preservation methods, cies with extinction (CAFF 2013). Vegetation changes and increases in food contaminants (AMAP 2017, have also been observed in recent decades, includ- Osborne et al 2018). Additionally, the loss of snow ing an overall ‘greening’ of the Arctic tundra that and ice can create safety concerns as a result of damage reflects an increase in plant growth and productivity 2
Environ. Res. Lett. 16 (2021) 093003 M Previdi et al Figure 2. (a) Observed 1979–2018 zonal mean surface air temperature (SAT) change (in K) for each season and the annual mean based on linear trends. Lines indicate the mean change of the three atmospheric reanalyses listed in figure 1(a), and shading denotes the ±1- sigma spread about this mean change (providing an estimate of observational uncertainty). (b) As in figure 2(a), but for the 12 CMIP5 models listed in the figure 1(b) caption (historical + RCP8.5 simulations). (c) As in figure 2(b), but for the SAT change based on the difference between the last 30 years of the abrupt 4 × CO2 and pre-industrial control simulations. Note the different scale on the ordinate compared to figures 2(a) and (b). (Osborne et al 2018). Arctic greening is an expec- and methane, and thus the global climate forcing ted response to higher temperatures, although other from these gases. Increases in global sea level due factors such as moisture availability and the occur- to melting of Arctic land ice are another important rence of extreme events (e.g. wildfires, insect out- impact of Arctic warming (e.g. Shepherd et al 2012, breaks) will continue to play an important role in tun- Box et al 2018, Bamber et al 2019). This ice melt, dra vegetation dynamics (e.g. Wu et al 2020). Higher along with projected increases in high-latitude pre- temperatures and the associated loss of sea ice are fur- cipitation and ocean warming, are expected to weaken ther driving increases in primary production by mar- the ocean’s meridional overturning circulation (IPCC ine algae, as well as an expansion of harmful algal 2013), with implications for regional climate change blooms that threaten human and ecosystem health and global ocean heat and carbon uptake. AA may (Osborne et al 2018). also influence the atmospheric jet streams (both in The impacts of AA are by no means limited to the troposphere and stratosphere) in ways that lead the Arctic. Sea-ice loss, permafrost thaw, and other to increased weather and climate extremes at mid- physical changes to the Arctic environment due to latitudes (e.g. Francis and Vavrus 2012, Cohen et al climate warming have the potential to significantly 2014, Barnes and Screen 2015, Coumou et al 2018). alter Arctic carbon cycling (e.g. McGuire et al 2009, However, the exact physical mechanisms involved, Parmentier et al 2013, Schuur et al 2015, Natali et al and the relative importance of Arctic warming com- 2019, Bowen et al 2020, Ito et al 2020, Bruhwiler pared to other influences (including internal variab- et al 2021). This can impact the atmospheric con- ility), remain uncertain (e.g. McCusker et al 2016, centrations of the well-mixed greenhouse gases CO2 Blackport and Screen 2020). 3
Environ. Res. Lett. 16 (2021) 093003 M Previdi et al Despite this wide array of impacts, a clear con- Articles that survived this elimination procedure sensus is still lacking within the scientific com- were used as evidence in our review. These articles munity as to which physical mechanisms are the most were supplemented with additional relevant articles important in causing AA. Previous review articles that were identified by the authors while writing the addressing this topic were published by Miller et al review. (2010), Serreze and Barry (2011), and Walsh (2014). More recently, Goosse et al (2018) reviewed some 3. Physical mechanisms of the key climate feedbacks in polar regions, and the methods for quantifying them. While AA was We now commence with the formal review of the briefly discussed, it was not the focus of the art- underlying physical mechanisms of AA. We begin icle per se. Thus, the goal of the current article is to with some notes on terminology (section 3.1), fol- provide an up-to-date, focused review of the phys- lowed by discussion of the roles of climate for- ical mechanisms that are likely responsible for AA: cing (section 3.2), climate feedbacks (section 3.3), our review draws on the wealth of new research that and changes in poleward energy transport (PET) has happened since the publication of earlier reviews. (section 3.4) in AA. The coupling between indi- This will serve to synthesize our current understand- vidual climate feedbacks, and between climate feed- ing, and to identify remaining knowledge gaps and backs and energy transport changes, is then discussed important paths for future research. (section 3.5). Finally, a synthesis is provided at the end of the section (section 3.6). The important ques- 2. Methodology tion of the dependence of forcing, feedbacks and energy transport on the state of the climate itself is Before beginning the formal review, we briefly considered in the next section (section 4). describe our approach for identifying articles that were used as evidence in the review. Article selection 3.1. Terminology began by searching for ‘Arctic amplification’ and then Throughout the discussion, we use the term climate ‘polar amplification’ using both Web of Science and forcing to refer to a change in an external driver of Scopus. For both databases, a document search for climate change, such as an increase in the atmo- these phrases within the article title, abstract, and/or spheric CO2 concentration (see section 3.2 for other keywords was performed. Documents identified in examples). The term radiative forcing is used to refer this manner were then screened using the following to the radiative flux change caused by a climate for- four-step procedure: cing. Radiative forcing without additional qualifiers refers to the radiative flux change at the top-of- (a) Irrelevant documents unrelated to polar amp- atmosphere (TOA; Intergovernmental Panel on Cli- lification as it pertains to climate change were mate Change (IPCC) 2013). However, we will also eliminated. For example, an initial search for consider the surface radiative forcing and atmospheric ‘polar amplification’ yielded multiple docu- radiative forcing, which denote, respectively, the radi- ments on topics as diverse as electronics and bio- ative flux change at the surface, and the change logy, which are clearly irrelevant to this review in radiative flux divergence within the atmospheric article. column (i.e. the difference between the TOA and sur- (b) Documents focused on polar amplification only face radiative forcing). In all cases, a positive (negat- in the Antarctic were eliminated. In addition to ive) forcing is one that induces warming (cooling). our review article being focused on the Arctic, In addition to the vertical level considered, radi- there is evidence that the magnitude of polar ative forcing can also be distinguished by whether or amplification—as well as its governing physical not the effects of rapid adjustments are included. A processes—are different in the two hemispheres. rapid adjustment is a response to a climate forcing (c) Opinion pieces and duplicate hits were elimin- that is driven directly by the forcing, independently ated. of any change in the global mean surface temperature (d) Documents not focused on the physical pro- (Intergovernmental Panel on Climate Change (IPCC) cesses driving AA were mostly eliminated. Two 2013). When rapid adjustments are not included, the especially prevalent examples of such documents resulting radiative forcing is referred to as the instant- are: a) articles identifying the occurrence of AA aneous radiative forcing. In this case, the radiative (either in observations or climate model simu- flux change is due solely to the change in the cli- lations) but not explicitly addressing its physical mate forcing agent (e.g. the increase in atmospheric causes; and b) articles discussing the impacts of CO2 ). Other types of radiative forcing include the AA (e.g. on midlatitude weather and climate). radiative effects of one or more rapid adjustments. While the main findings from some of these art- The stratosphere-adjusted radiative forcing includes icles were touched upon above in the Introduc- the effect of stratospheric temperature adjustment, tion, the primary focus of our review is the causes whereas the effective radiative forcing additionally of AA. includes the effects of other rapid adjustments (e.g. 4
Environ. Res. Lett. 16 (2021) 093003 M Previdi et al to tropospheric temperature, water vapor, and cloud and within the atmospheric column. At the TOA cover). Additional discussion of these different types (figure 3(a)), CO2 radiative forcing is largest at low of radiative forcing can be found in Hansen et al latitudes and decreases toward the poles6 . This meri- (1997), Shine et al (2003), Hansen et al (2005), IPCC dional structure is due to decreases in the clima- (2013), and Forster et al (2016), among others. tological surface temperature and tropospheric ver- While rapid adjustments represent the tical temperature gradient with latitude (Zhang and temperature-independent part of the response to Huang 2014, Merlis 2015, Smith et al 2018), leading a climate forcing (see above), climate feedbacks to the strongest enhancement of the greenhouse effect are the part of the response that is dependent on (and thus the largest radiative forcing) at low latitudes global mean surface temperature change (Inter- when CO2 is increased. governmental Panel on Climate Change (IPCC) A qualitatively different picture emerges when 2013). In the discussion that follows, we will con- one considers the CO2 radiative forcing at the surface sider feedbacks associated with changes in surface (figure 3(b)). In this case, the forcing is at a minimum and tropospheric temperatures5 (section 3.3.1), sur- at low latitudes and increases toward the poles. This face albedo (section 3.3.2), clouds and water vapor is due to the overlap between CO2 and water vapor (section 3.3.3), and other properties of the Earth sys- absorption bands in the 12–18 µm spectral region. tem (section 3.3.4). As for climate forcing, we will Because of this overlap, increases in the surface down- consider the impacts of climate feedbacks on the welling longwave radiation (DLR) are limited when energy budget at both the TOA and surface. Positive CO2 is increased, particularly at low latitudes where (negative) feedbacks are those that reinforce (oppose) water vapor concentrations are highest and the emis- the effects of a climate forcing. Thus, for a positive sion from this spectral region is already largely satur- forcing that induces warming, a positive feedback will ated (Kiehl and Ramanathan 1982, Huang et al 2017). lead to further warming. The very different spatial patterns of CO2 radi- Finally, we will consider changes in PET by the ative forcing at the TOA and at the surface result atmosphere (section 3.4.1) and ocean (section 3.4.2). in a strong latitudinal gradient in atmospheric radi- While it is unclear whether PET changes can be classi- ative forcing (figure 3(c)), with positive forcing at fied as feedbacks (Goosse et al 2018), they are part of low latitudes transitioning to negative forcing at the the response to climate forcing and are likely to con- poles. This pattern of atmospheric radiative forcing tribute to AA, as will be discussed. has major implications for the PET, as will be dis- cussed in section 3.4.1. To summarize, the impact of 3.2. Climate forcing direct CO2 radiative forcing on AA depends on per- The single most important climate forcing dur- spective. From the perspective of the TOA and atmo- ing the industrial era is increasing atmospheric spheric column energy budgets (figures 3(a) and (c)), CO2 (Intergovernmental Panel on Climate Change CO2 radiative forcing opposes AA by preferentially (IPCC) 2013). AA as a response to increasing CO2 heating the tropics. However, from the perspective of was initially predicted by Arrhenius (1896), and, the surface energy budget (figure 3(b)), CO2 radiative subsequently, much of the current understanding forcing contributes to AA by preferentially heating the of AA has been gleaned from CO2 -forced climate Arctic. model simulations (e.g. Manabe and Wetherald While the discussion thus far has focused on 1975, Manabe and Stouffer 1980, Holland and Bitz CO2 , other climate forcings have also been shown 2003, Winton 2006, Lu and Cai 2009a, Pithan and to produce AA. These include changes in solar irra- Mauritsen 2014, Previdi et al 2020). It seems appro- diance (Cai and Tung 2012, Stjern et al 2019), aer- priate then to begin our discussion of climate forcing osols and aerosol precursors (Lambert et al 2013, with CO2 . AA is a robust response to CO2 increases Yang et al 2014, Najafi et al 2015, Acosta Navarro in the studies cited above (and many others). The first et al 2016, Chylek et al 2016, Sagoo and Storelvmo question we wish to address, therefore, is this: is AA 2017, Conley et al 2018, Wang et al 2018, Stjern et al caused by the direct radiative forcing from CO2 , or 2019, Kim et al 2019b), ozone-depleting substances by subsequent responses within the climate system to (Polvani et al 2020), methane (Stjern et al 2019), this forcing? and land use/land cover (van der Molen et al 2011, Figure 3 shows the instantaneous radiative for- Lott et al 2020). The fact that AA occurs in response cing from doubling CO2 at the TOA, the surface, to these various forcings, with very different spa- tial, seasonal and spectral characteristics, suggests that 5 Stratospheric temperatures may also change significantly in it is a robust response to climate forcing that does response to an imposed climate forcing. For example, the strato- not depend—at least existentially—on the details of sphere cools when atmospheric CO2 is increased (figure 1(c)). Such the forcing. However, forcing details may impact the changes, however, are (largely) a direct response to the forcing (i.e., a rapid adjustment), rather than a feedback that is mediated by surface temperature change. The rapid adjustment of stratospheric 6 This strong latitudinal dependence is also present in the temperatures is included in the stratosphere-adjusted and effective stratosphere-adjusted (Hansen et al 1997) and effective radiative radiative forcing, as noted above. forcing (Forster et al 2016) from increasing CO2. 5
Environ. Res. Lett. 16 (2021) 093003 M Previdi et al Figure 3. Instantaneous radiative forcing (in W m−2 ) from doubling CO2 as calculated by the rapid radiative transfer model (RRTM). The forcing is defined as the change in the net (i.e. longwave plus shortwave) downward radiative flux, and is shown for: (a) the top-of-atmosphere (TOA); (b) the surface (SFC); and (c) the atmospheric column (ATM, defined as TOA minus SFC). Adapted with permission from Huang et al (2017). © 2017. American Geophysical Union. All Rights Reserved. magnitude of AA. Modeling studies that have system- and summer. What is clear though is that the dir- atically varied the forcing latitude (Kang et al 2017, ect effects of climate forcing alone are insufficient Park et al 2018, Stuecker et al 2018, Semmler et al to fully understand AA (see also Virgin and Smith 2020) and season (Bintanja and Krikken 2016) indic- 2019). Such an understanding requires consideration ate that AA is likely to be strongest for high-latitude of the feedback mechanisms operating within the cli- forcings, and for forcings occurring during spring mate system that act to amplify temperature changes 6
Environ. Res. Lett. 16 (2021) 093003 M Previdi et al Figure 4. Warming contributions of individual feedback mechanisms calculated based on the difference between the last 30 years of the CMIP5 abrupt 4 × CO2 and pre-industrial control simulations. (a) Arctic (60◦ –90◦ N) versus tropical (30◦ S–30◦ N) warming from a TOA perspective. (b) Arctic winter (December–January–February) versus summer (June–July–August) warming. (c) Arctic versus tropical warming from a surface perspective. For figures 4(a) and (c) feedbacks above the 1:1 line contribute to AA, whereas feedbacks below the line oppose AA. Grey is the residual error of the decomposition. ‘Ocean’ includes the effect of ocean transport changes and ocean heat uptake. Reprinted by permission from Springer Nature Customer Service Centre GmbH: Springer Nature, Nature Geoscience, Arctic amplification dominated by temperature feedbacks in contemporary climate models, Pithan and Mauritsen (2014). at high northern latitudes. We discuss these feedbacks profile leads to a larger increase in OLR than would next. occur from the Planck response alone, representing a negative feedback on surface warming. However, the 3.3. Climate feedbacks opposite occurs over the Arctic, where the upper tro- 3.3.1. Temperature feedbacks posphere warms less than the lower troposphere and We begin our discussion of climate feedbacks with surface. This opposes the basic Planck response and the temperature feedbacks that are associated with represents a positive feedback on surface warming. changes in surface and tropospheric temperatures. Thus, both the Planck and lapse rate components of For ease of discussion, and for relevance to anthro- the temperature feedback are expected to contribute pogenic climate change, we consider a positive cli- to AA. mate forcing that induces surface and tropospheric Indeed, previous studies that assessed contribu- warming. The total temperature feedback at the TOA tions to Arctic warming and AA in both models in response to such a forcing can be decomposed (Langen et al 2012, Graversen et al 2014, Pithan into contributions from vertically-uniform warming7 and Mauritsen 2014, Payne et al 2015, Cronin and (Planck response) and changes in the tropospheric Jansen 2016, Previdi et al 2020, Zhang et al 2020) and lapse rate (e.g. Hansen et al 1984, Bony et al 2006, observations (Hwang et al 2018, Zhang et al 2018) Soden and Held 2006). The Planck response is the concluded that temperature feedbacks are import- most fundamental mechanism acting to stabilize the ant. Notably, Pithan and Mauritsen (2014) found that climate, since surface and tropospheric warming will these feedbacks make the single largest contribution increase the outgoing longwave radiation (OLR) and to AA in CMIP5 models subjected to an instantan- thus oppose the effects of a positive forcing. How- eous quadrupling of atmospheric CO2 (4 × CO2 ). ever, because of the nonlinear dependence of OLR Their analysis demonstrates a prominent role for on temperature (as given by the Stefan–Boltzmann temperature feedbacks (and in particular the lapse law), the Planck response is weaker (i.e. less stabiliz- rate feedback) both in causing AA in an annual mean ing) over the colder Arctic relative to lower latitudes sense (figure 4(a)), and also in producing the distinct and to the global mean. In other words, OLR increases seasonality in AA, with much stronger AA in winter less over the Arctic than at lower latitudes, per degree relative to summer (figure 4(b); see also figure 2). of local warming, and this contributes to AA (Henry The effects of temperature feedbacks on the sur- and Merlis 2019). face energy budget come from both surface warm- Globally averaged, changes in the tropospheric ing and atmospheric warming, each of these contrib- lapse rate are characterized by a reduction in the uting to AA (figure 4(c); see also Laı̂né et al 2016, rate of temperature decrease with height, since the Sejas and Cai 2016). The surface warming contribu- upper troposphere warms more than the lower tropo- tion arises because a larger increase in surface tem- sphere and surface. Such a change in the temperature perature is needed over the colder Arctic (relative to lower latitudes and to the global mean) to balance a 7 The magnitude of warming is typically assumed to be equal to given increase in downward energy flux (i.e. through that at the surface. enhanced surface emission of longwave radiation). 7
Environ. Res. Lett. 16 (2021) 093003 M Previdi et al This is fundamentally the same reason why latitud- climatological sense, this seasonal cycle is character- inal variations in the Planck response at the TOA ized by ocean heat uptake in the late spring and sum- contribute to AA. However, the effect is more pro- mer, followed by a loss of heat from the ocean to the nounced when considering the surface energy budget, atmosphere in fall and winter. In a warming climate, since the meridional temperature gradient at the sur- the amplitude of this seasonal cycle is expected to face is larger than that in the troposphere (Pithan and increase, meaning greater ocean heat uptake in spring Mauritsen 2014). The atmospheric warming contri- and summer and greater ocean heat loss in fall and bution to AA is associated with increases in the sur- winter (Carton et al 2015, figure 4(b)). This is a direct face downwelling longwave radiation (DLR). These result of the loss of Arctic sea ice. In summertime, the DLR increases are larger in the Arctic than at lower surface albedo feedback associated with sea ice loss latitudes as a result of the much greater warming of causes a greater amount of incoming solar radiation the Arctic near-surface atmosphere (see figure 1). to be absorbed by the Arctic Ocean8 (Pistone et al 2014). Additionally, with less sea ice present, a lar- ger fraction of the absorbed energy goes into increas- 3.3.2. Surface albedo feedbacks ing ocean temperatures rather than into melting ice As sea ice and snow cover retreat in a warming cli- (Carton et al 2015). The additional energy accumu- mate, the surface albedo (or reflectivity) decreases lated within the mixed layer in spring and summer is and a larger fraction of the incoming solar radiation subsequently released to the atmosphere in fall and is absorbed by the Earth system. Such surface albedo winter in the form of longwave radiation and latent feedbacks therefore act to amplify surface warming, and sensible heat. This process is facilitated by the loss especially at high latitudes where sea ice and snow of sea ice (area and thickness) in the latter seasons, cover are most extensive. Changes in land ice (e.g. Hill and thus a weakening of the sea ice insulation effect. et al 2014, Stap et al 2014, 2017, 2018) and vegeta- The enhanced ocean-to-atmosphere energy tion (see also section 3.3.4) are additional sources of exchange in fall and winter, which acts to warm surface albedo change that may be important on long the near-surface atmosphere, is now recognized (multidecadal and longer) timescales. Here, however, to be of primary importance to AA (e.g. Screen we concentrate on the surface albedo (and related) and Simmonds 2010a, 2010b, Bintanja and van der feedbacks associated with the loss of sea ice and snow Linden 2013, Sejas et al 2014, Yoshimori et al 2014a, cover. Kim et al 2016, Franzke et al 2017, Boeke and Taylor Those feedbacks have been shown to contribute to 2018, Dai et al 2019, Chung et al 2021, Dai 2021). Its AA in both models (e.g. Budyko 1969, Sellers 1969, importance to observed AA in recent decades is sug- Manabe and Wetherald 1975, Manabe and Stouffer gested by the strong spatial correspondence between 1980, Hansen et al 1984, Holland and Bitz 2003, Hall near-surface warming and sea ice loss over the Arctic, 2004, Winton 2006, Crook et al 2011, Bintanja and as shown in figure 5. This strong spatial correspond- van der Linden 2013, Baidin and Meleshko 2014, ence further suggests a coupling between sea ice loss Graversen et al 2014, Pithan and Mauritsen 2014, and the Arctic lapse rate feedback (e.g. Feldl et al Song et al 2014, Lang et al 2017, Sun et al 2018, Dai 2020, Boeke et al 2021; see also section 3.5). The et al 2019, Duan et al 2019) and observations (e.g. warming and moistening of the Arctic lower atmo- Screen and Simmonds 2010a, 2010b, 2012, Pistone sphere increases the surface DLR, which hinders sea et al 2014, Hu et al 2017, Hwang et al 2018, Zhang ice growth and thus helps to sustain the enhanced et al 2018, 2019, Cai et al 2021, Yu et al 2021). For ocean-to-atmosphere energy flux (Burt et al 2016, example, in their analysis of the CMIP5 4 × CO2 Kim et al 2016, Alexeev et al 2017, Kim and Kim simulations, Pithan and Mauritsen (2014) found that 2017, Kim et al 2019a). surface albedo feedbacks make the second largest To summarize, there is considerable evidence contribution to AA, behind temperature feedbacks that surface albedo feedbacks—and other feedbacks (figures 4(a) and (c)). The former were found to be associated with the loss of sea ice and snow cover important both in causing AA in a multimodel mean in a warming climate—make significant contribu- sense, and also in explaining the intermodel spread in tions to AA. It is worth noting, however, that polar- the magnitude of AA (see also Boeke and Taylor 2018, amplified warming has still been simulated in cli- Dai et al 2019, Block et al 2020, Hu et al 2020). mate models even when surface albedo feedbacks It is important to point out that surface albedo (Hall 2004, Graversen and Wang 2009, Lu and Cai feedbacks are only active when sunlight is present, 2010, Graversen et al 2014) and all sea ice- and/or which is mainly in late spring and summer in the Arc- snow cover-related feedbacks (Schneider et al 1997, tic. AA itself, however, is absent in summer and is 1999, Alexeev 2003, Alexeev et al 2005, Duan et al most pronounced in fall and winter (figure 2). How 2019) have been eliminated (e.g. by fixing the surface then can surface albedo feedbacks contribute to AA? To answer this question, it is necessary to consider the 8 It is worth noting that this increase in surface solar radiation seasonal cycle of heat storage within the Arctic Ocean absorption would be significantly larger if not for the presence of mixed layer (e.g. Chung et al 2021, Dai 2021). In a extensive cloud cover over the Arctic Ocean (Alkama et al 2020b). 8
Environ. Res. Lett. 16 (2021) 093003 M Previdi et al Figure 5. Observed Arctic SAT and sea ice concentration (SIC) changes during October–November–December (OND) 1979–2018 based on linear trends. SAT changes (in K; shading) are the mean of the three reanalyses listed in figure 1(a), and SIC changes (in %; white contours) are taken from the HadISST dataset version 1.1. For SIC changes, only negative values (corresponding to sea ice loss) are plotted using a contour interval of 10%; positive SIC changes during this period (greater than +10%) are limited to a small area between northeastern Greenland and Svalbard, and are omitted here for clarity. albedo). This suggests that other processes are cap- water vapor feedbacks in more detail in the next able of producing AA, even in the absence of any section, and atmospheric PET changes will be con- changes in surface albedo and/or sea ice/snow cover. sidered in section 3.4.1. The importance of temperature feedbacks for AA was discussed above in section 3.3.1. However, other 3.3.3. Cloud and water vapor feedbacks processes may also be important. For example, Gra- Most climate models project that the Arctic will versen and Wang (2009) linked AA in their model become cloudier in a warmer climate (Vavrus 2004, to a stronger greenhouse effect over the Arctic res- Vavrus et al 2009, 2011, Taylor et al 2013), mainly ulting from increases in clouds and water vapor, and due to increases in low clouds in fall and winter. Alexeev et al (2005) emphasized the importance of This result is supported by analyses of observed cloud increased atmospheric PET. We discuss cloud and changes in recent decades (Wu and Lee 2012, Jun 9
Environ. Res. Lett. 16 (2021) 093003 M Previdi et al et al 2016, Cao et al 2017, Philipp et al 2020). Some recent decades are similarly uncertain, these estimates modeling studies also indicate that Arctic middle and depending on the choice of observational datasets, high clouds may increase as well (Vavrus et al 2009, the time period considered, and the method used to 2011). Increases in Arctic low clouds in fall and winter quantify cloud feedbacks (Zhang et al 2018). are a direct response to sea ice loss, which acts to Now we turn to water vapor feedbacks: these are destabilize the boundary layer and enhance mois- positive in the Arctic at both the TOA and surface, ture availability and boundary-layer convection (Kay with water vapor increases in the Arctic lower tropo- and Gettelman 2009, Vavrus et al 2009, Eastman and sphere being an important contributor to increases Warren 2010, Palm et al 2010, Vavrus et al 2011, Wu in surface DLR (Chen et al 2011, Ghatak and Miller and Lee 2012, Abe et al 2016, Jun et al 2016, Mor- 2013, Cao et al 2017). However, water vapor feed- rison et al 2019, Philipp et al 2020). In contrast, sea backs in the tropics (and in the global mean) tend to ice loss seems to have an insignificant impact on Arc- be stronger than they are in the Arctic, both at the tic cloud cover during summer (Kay and Gettelman TOA and surface, and thus the direct effect of these 2009, Morrison et al 2019, Choi et al 2020). feedbacks is to oppose AA (figures 4(a) and (c)). In Given that low cloud increases occur mainly dur- order to understand why water vapor feedbacks in the ing fall and winter when insolation is at a minimum tropics tend to be stronger, we decompose the differ- over the polar cap, the associated radiative effects ence in the tropical (30◦ S–30◦ N) and Arctic (60◦ – occur primarily in the longwave portion of the spec- 90◦ N) mean water vapor feedbacks in the CMIP5 trum. However, these effects differ considerably at the abrupt 4 × CO2 simulations. We express the water TOA and surface (Pithan and Mauritsen 2014). At vapor feedback dR as the product of two terms (see the TOA, low cloud increases have a relatively small Soden et al 2008): a radiative kernel K, describing the impact on OLR since the temperature of these clouds sensitivity of the TOA or surface radiation to incre- is not that different from the surface temperature. mental changes in specific humidity, and a climate In contrast, these cloud increases strongly enhance response dq, defined here as the difference in spe- the emissivity of the Arctic lower troposphere and cific humidity between the last 30 years of the CMIP5 thus the surface DLR (Vavrus et al 2011, Taylor et al abrupt 4 × CO2 and pre-industrial control simula- 2013, Abe et al 2016, Jun et al 2016, Cao et al 2017, tions, i.e. Morrison et al 2019, Philipp et al 2020). These DLR ∂R increases, which may be further enhanced by cloud dR = dq ≡ Kdq. (1) phase changes (i.e. a larger liquid-to-ice ratio; Tan ∂q and Storelvmo 2019, Huang et al 2021), contribute to The difference in water vapor feedback ∆dR between additional surface warming and sea ice melt (Vavrus the tropics and Arctic can then be written as: et al 2011, Cao et al 2017, Philipp et al 2020), and may also help to suppress the formation of Arctic ∆dR = dq∆K + K∆dq + ∆K∆dq (2) air masses over high-latitude continents (Cronin and Tziperman 2015, Cronin et al 2017). where dq∆K and K∆dq represent the contributions Understanding the effect of cloud feedbacks on to ∆dR from meridional differences in K and dq, AA requires consideration of the meridional struc- respectively, and ∆K∆dq is the nonlinear term rep- ture of these feedbacks. In other words, it is import- resenting the covariance between K and dq. ant to also consider cloud feedbacks outside of the Figure 6 shows the various terms in equation Arctic. Globally averaged, the net (i.e. longwave plus (2) for both the TOA (figure 6(a)) and surface shortwave) cloud feedback is positive in most cur- (figure 6(b)) water vapor feedbacks. We find that ∆dR rent climate models, but with large intermodel spread is 1.71 ± 0.63 K (multimodel mean ± 1σ) at the TOA, mainly due to differences in shortwave cloud feedback and 0.67 ± 0.24 K at the surface. This TOA value is (Zelinka et al 2020). This suggests that the effect of similar to the one found by Pithan and Mauritsen cloud feedbacks on AA is likely to be model depend- (2014) (compare figures 4(a) and 6(a)), although our ent. Additionally, this effect depends on whether one kernel (Previdi 2010, Previdi and Liepert 2012) yields considers changes to the TOA energy budget, or to a somewhat smaller value of ∆dR at the surface (com- the surface energy budget. From a TOA perspective, pare figures 4(c) and 6(b)). cloud feedbacks oppose AA in the CMIP5 multimodel The decomposition of ∆dR (equation (2)) pro- mean (figure 4(a)), whereas they contribute to AA duces qualitatively similar results at the TOA and sur- from a surface perspective (figure 4(c)). (Note that in face (figure 6). We find that in both cases, the stronger both cases, however, the magnitude of the cloud con- water vapor feedback in the tropics compared to the tribution is relatively small; see also Middlemas et al Arctic can be explained by the larger increase in spe- (2020).) This difference in the inferred cloud effect cific humidity in the former (i.e. by the K∆dq term). is partly due to the greater impact of Arctic cloud This larger increase in specific humidity is due to changes on the longwave radiation at the surface, as the strong (and nonlinear) dependence of the sat- discussed above. Cloud feedbacks (and thus the cloud uration vapor pressure on temperature—as given by contribution to AA) estimated from observations in the Clausius–Clapeyron relationship—which leads to 10
Environ. Res. Lett. 16 (2021) 093003 M Previdi et al Figure 6. Decomposition of the difference in tropical (30◦ S–30◦ N) and Arctic (60◦ –90◦ N) mean water vapor feedback at the (a) TOA and (b) surface in the CMIP5 abrupt 4 × CO2 simulations (blue bars represent the multimodel mean of the 12 CMIP5 models listed in the figure 1(b) caption, with black error bars bracketing ± 1σ about the mean). Labels along the abscissa are the terms in equation (2) (see discussion in text). Water vapor feedbacks are calculated using radiative kernels from ECHAM5 (Previdi 2010, Previdi and Liepert 2012), and are expressed as warming contributions (or partial temperature changes) as in figure 4. the largest increases in water vapor amounts where 3.3.4. Other feedbacks the climatological temperatures are highest (i.e. in To conclude this section, we briefly discuss a few the tropics). Note though that K∆dq is largely offset other feedbacks that have been suggested to contrib- by dq∆K and ∆K∆dq (particularly the latter). This ute to AA. One of these involves the northward shift offset reflects an overall lower radiative sensitivity to of boreal forest and overall greening of the Arctic water vapor perturbations (as measured by K) in the that are expected to occur with climate warming (see tropics compared to the Arctic, owing to the greater section 1). These vegetation changes enhance Arc- saturation of water vapor emission under moist trop- tic surface warming directly by reducing the surface ical conditions. The lower radiative sensitivity in the albedo of high latitude land areas, and thus increas- tropics leads to significantly smaller values of TOA ing the surface absorption of solar radiation. Fur- and surface ∆dR than would result from meridional ther enhancement of Arctic warming may also occur differences in dq alone. indirectly through interactions between vegetation To summarize, stronger tropical than Arctic water changes and changes in water vapor, clouds and sea vapor feedbacks in the CMIP5 abrupt 4 × CO2 simu- ice (O’ishi and Abe-Ouchi 2011, Jeong et al 2014, lations can be explained by larger increases in specific Willeit et al 2014, Chae et al 2015, Cho et al 2018, humidity in the tropics. The greater tropical moisten- Park et al 2020). Increases in phytoplankton biomass ing outweighs the opposing effect of lower tropical in the Arctic may produce an additional positive feed- radiative sensitivity to water vapor changes. In addi- back on surface warming, by enhancing solar radi- tion to these implications for water vapor feedback, ation absorption within the ocean surface layer (Park larger specific humidity increases in the tropics com- et al 2015). pared to the Arctic also lead to enhanced poleward These changes in the biosphere affect not only the moisture transport with climate warming. As will be surface solar radiation, but also other surface fluxes discussed below (see section 3.4.1), this enhanced such as the turbulent fluxes of latent and sensible heat moisture transport is now recognized to be import- (e.g. evapotranspiration changes due to vegetation ant for AA. change). More generally, it is important to consider 11
Environ. Res. Lett. 16 (2021) 093003 M Previdi et al Figure 7. Changes in northward energy transports (in PW) in CMIP3 models (see legend) from 2001–2020 to 2081–2100 in the A2 scenario: (a) atmospheric energy transport; (b) moisture (solid) and dry static energy (DSE; dashed) transport; and (c) oceanic energy transport. Reproduced with permission from Hwang et al (2011). Copyright 2011 by the American Geophysical Union. the possible impact of changes in the surface tur- forcing within the atmosphere, with the forcing act- bulent energy fluxes on AA, in addition to the sur- ing to heat the atmosphere in the tropics and generally face radiative flux changes discussed in some detail cool it in polar regions (figure 3(c)). Such a forcing above. In a warming climate, the turbulent energy pattern contributes to an increase in the atmospheric (latent plus sensible heat) transfer from the surface to PET (e.g. Cai 2006, Huang et al 2017), which redis- the atmosphere is generally expected to increase (e.g. tributes the additional energy that is gained at low Cai and Lu 2007, Taylor et al 2013). Increases in the latitudes. In accordance with this, coupled climate surface latent heat flux (or evaporation) tend to be models forced with increased CO2 robustly simulate largest at low latitudes (particularly over oceans), a increases in atmospheric PET in the midlatitudes of consequence of the strong temperature dependence both hemispheres (Held and Soden 2006, Hwang and of the saturation vapor pressure (i.e. the Clausius– Frierson 2010, Wu et al 2011, Zelinka and Hartmann Clapeyron relationship). These evaporation increases 2012, Huang and Zhang 2014, Armour et al 2019). at low latitudes provide a strong negative feedback Over the Arctic, however, model responses are on surface warming, acting to stabilize tropical sea not so robust. While some models simulate increases surface temperatures, and thus contributing to AA in atmospheric PET into the Arctic, other models (Newell 1979, Hartmann and Michelsen 1993, Cai simulate decreases (figure 7(a); see also figure 3 and Lu 2007, Yoshimori et al 2014b). in Pithan and Mauritsen 2014). Decreases in atmo- spheric PET into the Arctic have also been reported 3.4. Changes in PET in observations over recent decades (Sorokina and 3.4.1. Atmospheric transport Esau 2011, Fan et al 2015), although the magnitude In section 3.2, we discussed the difference in the latit- and sign of the PET change are sensitive to the data- udinal structure of CO2 radiative forcing at the TOA set (Liu et al 2020), time period (Graversen 2006, and surface. Recall that, at the TOA, the forcing is Yang et al 2010), and Arctic sector (Mewes and Jacobi at a maximum in the tropics and decreases toward 2019) considered. From an energetic perspective, the poles (figure 3(a)). At the surface, however, this such PET decreases imply that the net effect of local latitudinal gradient is reversed, with the maximum feedbacks over the Arctic must be to heat the atmo- forcing occurring at high latitudes (figure 3(b)). This spheric column, thus balancing the cooling effects difference between the TOA and surface leads to an from reduced PET and CO2 radiative forcing. This even stronger latitudinal gradient in CO2 radiative further implies a coupling between local feedbacks 12
Environ. Res. Lett. 16 (2021) 093003 M Previdi et al and PET changes, which will be discussed in when the total PET decreases (Graversen and Burtu section 3.5. 2016, Graversen and Langen 2019). The total change in atmospheric PET can be Changes in PET are the means by which heat- decomposed into a change in dry static energy ing anomalies outside the Arctic can influence the (DSE) transport, and a change in moisture trans- Arctic climate. Previous studies have emphasized the port. While the total change in PET into the importance of heating anomalies in Northern Hemi- Arctic is not robust across model simulations—as sphere (NH) midlatitudes (Solomon 2006, Laliberté noted above—the changes in the DSE and mois- and Kushner 2013, Fajber et al 2018), and in the ture transports individually are robust. DSE trans- tropics (Rodgers et al 2003, Lee et al 2011, Yoo et al port into the Arctic decreases with climate warm- 2011, Lee 2014, Baggett and Lee 2015, Baggett et al ing, while moisture transport into the Arctic increases 2016, Baggett and Lee 2017, Yim et al 2017, Baxter (figure 7(b)). The sign of the total PET change et al 2019, McCrystall et al 2020, Dunn-Sigouin et al is primarily determined by the magnitude of the 2021). For instance, it has been suggested (Laliberté decrease in DSE transport. In models with strong and Kushner 2013, Fajber et al 2018) that near-surface decreases in DSE transport, the total PET into the temperature and moisture anomalies at midlatitudes Arctic also decreases. However, in models with weak are propagated by synoptic-scale eddies along moist decreases in DSE transport, the increase in moisture isentropes to the Arctic midtroposphere. This would transport (which has comparatively little intermodel imply a direct influence of midlatitude surface warm- spread; see figure 7(b)) results in an increase in the ing on the lapse rate and water vapor feedbacks over total PET. the Arctic, with implications for Arctic surface warm- Decreases in DSE transport into the Arctic are ing and AA. Alternatively, changes in atmospheric a response to reductions in the meridional temper- PET into the Arctic may be induced by heating anom- ature gradient due to polar-amplified warming (e.g. alies in the tropics. For instance, it has been suggested Langen and Alexeev 2007, Hwang et al 2011, Skific that enhanced localized convection over the Maritime and Francis 2013, Graversen and Burtu 2016, Audette Continent can trigger poleward-propagating Rossby et al 2021). The relationship between the DSE trans- waves that may enhance PET into the Arctic and con- port and the meridional temperature gradient leads to tribute to AA (Lee et al 2011, Yoo et al 2011, Lee 2014, a negative correlation between AA and the transport Baggett and Lee 2015, Baggett et al 2016, Baggett and change in climate models: models with the largest AA Lee 2017). PET into the Arctic could also be enhanced also tend to have the largest decrease in DSE (and by a northward shift of the Inter-tropical Conver- total) transport (Hwang et al 2011). We emphasize gence Zone, which many models project will occur that this correlation reflects the response of the atmo- with climate warming (Yim et al 2017). spheric PET to AA. However, PET changes are also The relative importance of heating anomalies in likely to be an important cause of AA, as will be dis- the tropics versus the midlatitudes has implications cussed. for how we view changes in atmospheric PET. If trop- Increases in moisture transport into the Arctic ical heating anomalies are more important, a bet- are a response to a stronger meridional gradient of ter understanding of changes in tropical convection specific humidity (e.g. Graversen and Burtu 2016; would be critical for understanding atmospheric PET see also section 3.3.3). Several studies have emphas- changes. On the other hand, if midlatitude heating ized the importance of this enhanced moisture trans- anomalies are more important, one might wish to port for AA (e.g. Flannery 1984, Cai 2005, Cai and place greater emphasis on better understanding the Lu 2007, Langen and Alexeev 2007, Graversen and dynamics of the eddy-driven jets and storm tracks. In Burtu 2016, Yoshimori et al 2017, Merlis and Henry addition to these considerations, it is also worth not- 2018, Graversen and Langen 2019). Graversen and ing that tropical and midlatitude heating anomalies Burtu (2016) demonstrated using reanalysis data that have different preferred teleconnection pathways to a given increase in moisture transport across 70◦ N the Arctic, meaning that their relative importance is results in a greater Arctic surface warming than an likely to be regionally dependent (Ye and Jung 2019, equivalent increase in the DSE transport. This occurs Hall et al 2021). because the increase in moisture transport acts to strengthen the greenhouse effect over the Arctic— 3.4.2. Oceanic transport both directly and by enhancing Arctic cloudiness— We now consider changes in PET by the ocean. leading to large increases in the surface DLR (see also In response to climate warming, models generally Rodgers et al 2003, Gong et al 2017, Kim and Kim simulate decreases in oceanic PET at NH midlatit- 2017, Praetorius et al 2018, Hao et al 2019, Clark udes, and increases in oceanic PET into the Arctic et al 2021, Dunn-Sigouin et al 2021). Thus, the warm- (figure 7(c); Holland and Bitz 2003, Bitz et al 2006, ing effect from enhanced moisture transport into the Held and Soden 2006, Hwang et al 2011, Kay et al Arctic with climate change can outweigh the cooling 2012, Graham and Vellinga 2013, Koenigk et al 2013, effect from reduced DSE transport, leading to a net Koenigk and Brodeau 2014, Koenigk and Brodeau Arctic warming from atmospheric PET changes even 2017, Nummelin et al 2017, Singh et al 2017, 2018, 13
Environ. Res. Lett. 16 (2021) 093003 M Previdi et al Docquier et al 2019, van der Linden et al 2019, Beer depth into the mixed layer (Graham and Vellinga et al 2020). Increased oceanic PET into the Arctic 2013). in recent decades has also been inferred from ocean While some of the additional energy delivered temperature reconstructions based on marine sedi- to the mixed layer through ocean transport changes ments off Western Svalbard (Spielhagen et al 2011). goes into warming the mixed layer, most of this addi- Modeling studies (Singh et al 2017, 2018) examining tional energy goes into melting sea ice and warming the seasonality of oceanic PET changes have found the near-surface atmosphere (Bitz et al 2006, Koenigk that enhanced oceanic PET into the Arctic in the and Brodeau 2014, Marshall et al 2015, Docquier et al annual mean is the result of PET increases during 2019, Alkama et al 2020a). This suggests a positive boreal winter, with summertime PET into the Arc- contribution from ocean transport changes to AA tic decreasing slightly. In addition to oceanic PET (Holland and Bitz 2003, Hwang et al 2011, Mahlstein increases, enhanced vertical ocean heat flux into the and Knutti 2011, Graham and Vellinga 2013, Koenigk Arctic Ocean mixed layer is also simulated by models and Brodeau 2014, Marshall et al 2014, 2015, in response to climate warming (Graham and Vellinga Nummelin et al 2017, Singh et al 2017, 2018, Beer et al 2013). 2020). It is worth emphasizing that this positive con- These changes in the oceanic energy transport tribution occurs during winter (figure 4(b)), the time are understood to varying degrees. The simulated of year when oceanic PET into the Arctic is enhanced decrease in oceanic PET at NH midlatitudes is mainly (Singh et al 2017, 2018), and when the ocean is a net the result of a weakening of the Atlantic meridi- source of heat for the atmosphere (section 3.3.2). The onal overturning circulation with climate warming contribution of ocean transport changes to Arctic (e.g. Gregory et al 2005, Cunningham and Marsh warming and sea ice melt exemplifies the coupling 2010, Hwang et al 2011, Nummelin et al 2017). between PET changes and local feedbacks, a topic At higher latitudes, however, the cause(s) of the that will be discussed further in the next section. increase in oceanic PET into the Arctic is less cer- tain. In particular, it is unclear to what extent this 3.5. Coupling between mechanisms increase is driven by circulation changes, versus the Thus far, we have considered the contributions advection of ocean heat anomalies by the back- of climate forcing (section 3.2), climate feedbacks ground flow. Bitz et al (2006) examined simula- (section 3.3), and PET changes (section 3.4) to AA. tions from a coupled atmosphere-ocean model forced While these contributions have been discussed separ- with increasing CO2 , and found that oceanic PET ately, it is now recognized that the physical mechan- into the Arctic was enhanced largely as a result of isms involved are closely coupled with one another. strengthened inflow driven by increased convection In this section, we discuss the possible couplings along the Siberian shelves. This increased convection, between mechanisms, and their implications for our in turn, was driven by enhanced sea ice production understanding of AA. and ocean-to-atmosphere heat flux in winter. How- Some examples of coupling have already been ever, other studies with coupled models have found mentioned. For instance, in sections 3.3.2 and 3.3.3, that the role of circulation changes is small, and that we noted that Arctic sea ice loss leads to enhanced oceanic PET increases are largely due to the advection ocean-to-atmosphere heat flux in the fall and winter. of warmer water into the Arctic by the background This is associated with a warming and moistening flow (Koenigk and Brodeau 2014, Nummelin et al of the boundary layer, and increases in low-level 2017). Nummelin et al (2017) linked this to reduced clouds, which implies an impact of sea ice loss on heat loss from the subpolar ocean to the atmosphere, the lapse rate, water vapor, and cloud feedbacks which increases the heat content of the water masses over the Arctic. The water vapor and cloud feed- that are advected into the Arctic. backs are further impacted by increases in poleward In addition to possible model dependence, the rel- moisture transport (section 3.4.1), and sea ice loss is ative importance of circulation changes versus pass- enhanced by increases in oceanic PET into the Arc- ive temperature advection for oceanic PET changes tic (section 3.4.2). Changes in atmospheric PET are is likely to vary with time, as patterns of ocean heat also known to play an important role in modulating uptake and other aspects of the climate response the Arctic lapse rate feedback (Graversen et al 2008, evolve. For example, strengthened oceanic inflow to Screen et al 2012, Laliberté and Kushner 2013, Cronin the Arctic driven by increased convection along the and Jansen 2016, Feldl et al 2017a, Fajber et al 2018, Siberian shelves is expected to eventually diminish in Kim et al 2018, Park et al 2018, Kim and Kim 2019, a much warmer climate as wintertime sea ice pro- Feldl et al 2020). duction declines (Bitz et al 2006). Sea ice declines are The interactions between these physical mechan- also key to understanding the enhanced vertical ocean isms work in both directions. For example, increases heat flux into the Arctic Ocean mixed layer noted in Arctic lower tropospheric temperature, moisture above. Specifically, as sea ice melts, the wind stress and cloud cover resulting from sea ice loss act to on the Arctic Ocean surface increases. This enhances strongly enhance the surface DLR, thus contributing upper ocean mixing and entrains warmer water from to even greater ice loss (sections 3.3.2 and 3.3.3). 14
You can also read