Emergent Behavior of Arctic Precipitation in Response to Enhanced Arctic Warming
Abstract
Amplified warming of the high latitudes in response to human-induced emissions of greenhouse gases has already been observed in the historical record and is a robust feature evident across a hierarchy of model systems, including the models of the Coupled Model Intercomparison Project Phase 5 (CMIP5). The main aims of this analysis are to quantify intermodel differences in the Arctic amplification (AA) of the global warming signal in CMIP5 RCP8.5 (Representative Concentration Pathway 8.5) simulations and to diagnose these differences in the context of the energy and water cycles of the region. This diagnosis reveals an emergent behavior between the energetic and hydrometeorological responses of the Arctic to warming: in particular, enhanced AA and its associated reduction in dry static energy convergence is balanced to first order by latent heating via enhanced precipitation. This balance necessitates increasing Arctic precipitation with increasing AA while at the same time constraining the magnitude of that precipitation increase. The sensitivity of the increase, ~1.25 (W/m2)/K (~240 (km3/yr)/K), is evident across a broad range of historical and projected AA values. Accounting for the energetic constraint on Arctic precipitation, as a function of AA, in turn informs understanding of both the sign and magnitude of hydrologic cycle changes that the Arctic may experience.
Key Points
- Intermodel differences in projected temperature increases are greatest in the polar regions
- Models with enhanced Arctic amplification (AA) of temperatures also show enhanced Arctic precipitation
- Enhanced Arctic precipitation's latent heat release serves to balance reduced dry static energy fluxes that accompany enhanced AA
1 Introduction
Although responses of the global climate system to anthropogenic forcing via carbon dioxide and other greenhouse gases are frequently quantified in terms of global mean temperature increases, the response is recognized to be rather spatially heterogeneous (e.g., Hegerl et al., 1996, and references therein). One of the most robust signatures of human-induced global warming is a polar amplification of the overall temperature increase, which was first hypothesized by Arrhenius (1896) and which has recently emerged as a leading indicator of human influence on the climate (e.g., ACIA, 2005; Bekryaev et al., 2010; Gillett et al., 2008; Jeffries et al., 2013; Serreze et al., 2009; Walsh et al., 2011). The presence, magnitude, and extent of this amplified warming of high latitudes in turn have important implications for chemical, biological, and physical systems across the region, particularly in the Arctic. Chemically, Arctic amplification (AA) of the global warming signal is of primary concern because of its potential for releasing vast amounts of organic carbon currently stored within permafrost and on the continental shelves (Archer et al., 2009; DeConto et al., 2012; Schuur et al., 2015). Ecologically, terrestrial and marine biota in the region are highly sensitive to changes in both geographic and temporal characteristics of the resultant climate (Diffenbaugh & Field, 2013; Post et al., 2013; Xu et al., 2013). And physically, polar amplification by definition directly influences meridional temperature gradients both near the surface and aloft (Chung & Räisänen, 2011; Laliberté & Kushner, 2013), which in turn influence the dynamic and thermodynamic state of the atmosphere and underlying ocean (e.g., Hassanzadeh et al., 2014; Held, 1993; Schneider et al., 2010; Shaw et al., 2016). While still a topic of active research (e.g., Cohen et al., 2014; Overland et al., 2016), potential responses include geographic shifts in the position and strength of the midlatitude jet stream and storm tracks (Barnes & Polvani, 2015; Butler et al., 2010; Cattiaux & Cassou, 2013; Rivière, 2011); changes in the frequency and intensity of quasi-stationary, long-lived atmospheric blocks that give rise to hot and cold extremes (Barnes, 2013; Cattiaux et al., 2016; Francis & Vavrus, 2012; Hassanzadeh et al., 2014; Screen & Simmonds, 2013); modifications of the extent and rate of cyclogenesis that determine storm intensity (Inoue & Hori, 2011; Overland & Wang, 2010; Terpstra et al., 2015); and adjustments of the thermal equator and its control on the low-latitude hydrologic cycle (Chiang & Bitz, 2005; Kang et al., 2009; Seo et al., 2014).
Accordingly, amplified warming in the Arctic has the potential to influence regional climates both locally and remotely (Barnes, 2013; Schneider et al., 2015; Screen & Simmonds, 2013; Walsh, 2014). Importantly, the climate response not only involves changes in temperature but also precipitation (Bintanja & Selten, 2014; Deser et al., 2015; Kopec et al., 2016; Park et al., 2013; Semmler et al., 2012). For this reason polar amplification of the global warming signal has been analyzed extensively using simple energy balance models (e.g., Alexeev & Jackson, 2013; Hwang et al., 2011; Lian & Cess, 1977; Roe et al., 2015), atmosphere-only numerical climate models (Deser et al., 2015; Screen et al., 2012), idealized climate models (e.g., aquaplanet simulations—Alexeev et al., 2005; Feldl & Roe, 2013), and fully coupled Earth system models (e.g., Graversen & Wang, 2009; Holland & Bitz, 2003; Pithan & Mauritsen, 2014). However, the magnitude and structure of the amplified warming remains uncertain and estimates differ even when model systems are provided with equivalent trajectories of radiative forcing (Bracegirdle & Stephenson, 2013; Laliberté & Kushner, 2013). Indeed, the leading pattern of intermodel differences in projected temperature increases, after accounting for overall global mean climate sensitivity, shows some of the largest discrepancies in the Arctic regions (cf. Figure 1 below).

Following from this recognition, our motivating interest here is to isolate intermodel differences in the estimated AA of the overall global mean temperature trend and analyze corresponding intermodel differences in thermodynamic properties of the atmosphere. While numerous studies have addressed this issue (e.g., Bengtsson et al., 2013; Hwang et al., 2011; Kay et al., 2012; Semmler et al., 2012; Serreze et al., 2007), our focus is on the linkages between changing atmospheric energetics and its connection to precipitation. Specifically, based on previous analyses demonstrating high sensitivity of precipitation to changing energetics in the Arctic region (Bintanja & Selten, 2014; Vihma et al., 2016) as well as in the tropics (Chiang & Bitz, 2005; Friedman et al., 2013; Kang et al., 2009) and globally (Allen & Ingram, 2002; Pendergrass & Hartmann, 2014), we hypothesize that intermodel differences in AA provide an energetic constraint on intermodel differences in projected precipitation trends over the Arctic. The rest of this paper is laid out as follows. Section 2 details the data and methods used in the analysis. Results are presented in section 3. A discussion and summary of these results are provided in sections 4 and 5.
2 Data and Methods
For this study, we use data from Coupled Model Intercomparison Project Phase 5 (CMIP5) simulations forced by the RCP8.5 scenario, which imposes an effective 8.5 W/m2 forcing by the year 2100 (Taylor et al., 2012). Output from one ensemble member from each of eighteen (18) models for the period 2006–2100 is used (listed in Table 1). For intercomparison all model data have been linearly interpolated to a common 5° × 5° grid. For all data we calculate grid point trends over the 95 year period 2006–2100 by first calculating running 20 year means (for a given variable) across the full time period. We then fit a linear time trend to the data using least squares regression. Next we take the difference of the first and last value of the linear trend, which represents the difference between the climatological characteristics during the first and last 20 year intervals of the period, after removing high-frequency fluctuations.
Modeling center (or group) | Institute ID | Model name |
---|---|---|
Commonwealth Scientific and Industrial Research Organization (CSIRO) and Bureau of Meteorology (BOM), Australia | CSIRO-BOM | ACCESS1.0 |
Commonwealth Scientific and Industrial Research Organization (CSIRO) and Bureau of Meteorology (BOM), Australia | CSIRO-BOM | ACCESS1.3 |
Beijing Climate Center, China Meteorological Administration | BCC | BCC-CSM1.1 |
Canadian Centre for Climate Modeling and Analysis | CCCMA | CanESM2 |
National Center for Atmospheric Research | NCAR | CCSM4 |
Community Earth System Model Contributors | NSF-DOE-NCAR | CESM1(CAM5) |
Centro Euro-Mediterraneo per I Cambiamenti Climatici | CMCC | CMCC-CM |
Centre National de Recherches Météorologiques/Centre Européen de Recherche et Formation Avancée en Calcul Scientifique | CNRM-CERFACS | CNRM-CM5 |
Commonwealth Scientific and Industrial Research Organization in collaboration with Queensland Climate Change Centre of Excellence | CSIRO-QCCCE | CSIRO-Mk3.6.0 |
NOAA Geophysical Fluid Dynamics Laboratory | NOAA GFDL | GFDL-CM3 |
NASA Goddard Institute for Space Studies | NASA GISS | GISS-E2-R |
Met Office Hadley Centre (additional HadGEM2-ES realizations contributed by Instituto Nacional de Pesquisas Espaciais) |
MOHC (additional realizations by INPE) |
HadGEM2-CC |
Institute for Numerical Mathematics | INM | INM-CM4 |
Institute Pierre-Simon Laplace | IPSL | IPSL-CM5A-MR |
Atmosphere and Ocean Research Institute (University of Tokyo), National Institute for Environmental Studies, and Japan Agency for Marine-Earth Science and Technology | MIROC | MIROC5 |
Max-Planck-Institut für Meteorologie (Max Planck Institute for Meteorology) | MPI-M | MPI-ESM-LR |
Meteorological Research Institute | MRI | MRI-CGCM3 |
Norwegian Climate Centre | NCC | NorESM1-M |
- Note. For complete “Terms of Use” please see http://cmip-pcmdi.llnl.gov/cmip5/terms.html.
For near-surface temperatures, we calculate two different grid point trend values. The first, Ts, is simply the grid point values as derived above. In addition, we calculate “structural surface temperature” (sTs) trend values, which are designed to represent the structure of grid point variations in surface temperature trends accounting for a given model's global mean temperature increase. To do so, at each grid point we linearly regress the intermodel Ts trend differences against intermodel differences in the models' global mean temperatures; for each model (and grid) we subsequently reconstruct the linear Ts trend map based upon the given model's global mean temperature increase and subtract it from the overall trend map (Anderson et al., 2015).
Following Anderson et al. (2015) and Langenbrunner et al. (2015), we then determine the leading contributors to intermodel differences in sTs trends by performing a “Principal Uncertainty Pattern” analysis. We do so here by applying Empirical Orthogonal Function (EOF) analysis across the “space model” domain using the grid point values of intermodel sTs trend differences. We perform the EOF analysis over both a global domain (as in Anderson et al., 2015) and over an Arctic-only domain, defined here as 70°–90°N. We then calculate weighted composite-mean trends (for a given variable) using the Principal Component (PC) weights for the leading mode; this is done separately for models with positive and negative PC weights. The model weights can also be regressed against intermodel trend differences for a given variable.
3 Results
We first illustrate the leading mode of global intermodel sTs variability, as determined via EOF analysis applied to the global sTs trends (Figure 1a). Overall, this mode is related to intermodel differences in interhemispheric sTs gradient trends principally over the extratropical regions and maximizing in the polar regions of both hemispheres. To further refine our analysis, we apply the same method to the sTs trends north of 70°N (Figure 1b). Importantly, the Northern Hemisphere structure of the leading mode of global sTs trend variability partially represents intermodel variability in the strength of AA of the global warming signal, as captured by the Arctic sTs trend variability (although the Southern Hemisphere structure in the latter contains somewhat enhanced intermodel variability in the tropics and reduced intermodel variability in the subpolar regions).
Considering just the Northern Hemisphere manifestation of the leading mode of Arctic sTs trend variability (Figure 2), the models with positive weightings (Figure 2a) manifest substantial amplification of the global mean signal, by up to 10 K, centered over the Arctic ocean. By contrast, for models with negative weightings (Figure 2b) the amplification of the global mean signal is weaker (~2–4 K) and centered over the Barents Sea. Differencing these two composite values (Figure 2c) suggests a difference in the AA signal by up to 5 K, with enhanced warming extending to the extrapolar regions of North America, Eurasia, and the North Pacific.

Given the mathematical construction of sTs, the values shown in Figure 2 represent temperature trends relative to the global mean and as such they reflect the structural signal of intermodel differences in AA. To confirm this, we reconstruct the model-specific sTs values using just the first PC of Arctic intermodel sTs trend differences. We then plot the area-averaged values from 70° to 90°N from each model against a more conventional measure of AA, namely, the difference in the area averaged Ts trend values from 70° to 90°N minus the global mean trend values (Figure 3). As expected, there is a robust relation between the two, although the two metrics do not align perfectly because the conventional AA metric only captures the magnitude of amplified warming in each model while the first PC of Arctic intermodel sTs trend differences also captures the structural representation of that amplification in each model. For this reason, in what follows we use the first PC of Arctic intermodel sTs trend differences as our metric for AA, although we note that sensitivity tests indicate that all results are robust if the analysis is repeated using the conventional AA metric.





Here SW represents shortwave fluxes, LW represents longwave fluxes, SH represents sensible heat transfer to the atmosphere, LH represents latent heat transfer to the atmosphere, and up (down) arrows represent upward (downward) fluxes of radiation.






Plotting trends in these quantities with respect to a given model's AA value (Figure 4) shows that trends in total AEC decrease with increasing AA, as has been reported elsewhere (e.g., Pithan & Mauritsen, 2014), such that low-AA models tend to be characterized by small to negligible increases in total AEC while high-AA models tend to be characterized by decreases in total AEC. Interestingly, there is very little relation between intermodel differences in AA and trends in SWatm; rather, the predominant balancing term for the decrease in AEC as a function of AA is an increase in
. This increase in surface heat flux to the atmosphere with increasing AA is partially offset by increasing
as a function of AA.








More germane to our primary findings are the contributors to the surface-generated AEC trends. In this case, all necessary flux values—including latent and sensible heat transfer along with longwave emissions to and from the surface—are directly available from the models. Plotting these against a given model's AA value (Figure 5) indicates that the decrease in surface-generated AEC as a function of AA results from increasing latent heat flux to the atmosphere with increasing AA, as well as from decreasing (negative) net longwave emissions to the atmosphere with increasing AA. For the latter, it is important to note that in all models the net transfer is positive, that is, from the surface to the atmosphere, and hence generates divergence of energy from the atmospheric column. In low-AA models, the net longwave transfer is reduced because the trend in emissions from the atmosphere to the surface exceeds the trend in emissions from the surface to the atmosphere (not shown); hence, the trend in net longwave emissions necessitates enhanced AEC in the low-AA models. As AA increases, however, the trend in emissions from the surface to the atmosphere comes back into equilibrium with the trend from the atmosphere to the surface (not shown) and the overall trend in net emissions tends toward 0 in high-AA models, similar to what occurs on synoptic time scales (Woods & Caballero, 2016). It is also important to note that intermodel spread in sensible heat transfer over the Arctic is not a function of AA.





The two terms on the left-hand side (LHS) of the equation represent heating of the atmospheric column via convergence of DSE and latent heat release via precipitation, respectively. The first term on the RHS of the equation represents the total net loss of longwave radiation by the atmosphere—which we will designate as LWatm—while the second and third terms represent heating of the atmosphere by sensible heat transfer and absorption of solar radiation, respectively. Plotting these five terms as a function of AA (Figure 7) reveals an emergent behavior over the Arctic. In particular, the heating of the atmosphere via sensible heat transfer and absorption of solar radiation are constant as a function of AA (as found earlier). In addition, while the emission of longwave radiation to space increases as a function of AA (cf. Figure 4), the net emission to the surface decreases (cf. Figure 5) and hence LWatm is also constant as a function of AA (in contrast to intermodel differences found at the global scale—Pendergrass & Hartmann, 2014). As such, the RHS of equation 11 is constant as a function of AA. By construction, then, the LHS of the equation is constant as a function of AA, which in turn prescribes that the decrease in DSE-generated AEC as a function of AA is balanced predominantly by increasing precipitation as a function of AA.

This emergent behavior can be rendered more explicit by plotting DSE-generated AEC as a function of Arctic precipitation (Figure 8a). This relation—which is one of the strongest of those shown here (r = −0.81)—suggests that Arctic precipitation increases are constrained by the magnitude of AA through its influence on DSE-generated AEC. Returning to Figure 7, we can understand this constraint by noting that for all models the heating of the atmosphere by sensible heat transfer and absorption of solar radiation, in addition to being invariant as a function of AA, are both relatively weak. Further, for all models the total net loss of longwave radiation by the atmosphere is positive definite such that it cannot balance the anomalous DSE-generated cooling. As such Arctic precipitation increases are not arbitrary but are in fact constrained by the necessity for column (latent) heating to balance the anomalous DSE-generated cooling that accompanies enhanced AA. Further confirmation of this emergent relation can be found by plotting the DSE-generated AEC against precipitation-generated AEC at all times for each of the 18 individual models (Figure 8b). While initial values of both quantities differ across the models, a significant linear relation between the two is found in all models, regardless of AA strength.

To further investigate this emergent relation, we next examine the time evolution of the weighted composite-mean Arctic AEC trends for low- and high-AA models, as determined by the model weightings from the leading mode of Arctic sTs trend variability. While the low-AA models manifest quasi-linear evolution of AEC trends across the full simulation period, the high-AA models—which initially track the low-AA evolution closely through the first half of the simulation—depart from this linear behavior with respect to both DSE-generated and precipitation-generated AEC, in concert with the nonlinear evolution in AA itself (not shown). Performing a similar analysis of the surface energy convergence terms indicates that this nonlinear time evolution is also found in the latent heating (equivalently evaporation) and the absorbed shortwave energy at the surface (not shown). To compare the time evolution of the high- and low-AA models, we difference the composite-mean values of DSE-generated and precipitation-generated AEC, absorbed shortwave energy at the surface, and the common term related to latent heat transfer to the atmosphere, along with the total Arctic AEC, and plot these as functions of time (Figure 9a). Doing so highlights that the initial (nonlinear) deviation in the evolution of the high-AA models vis-à-vis the low-AA models is associated with absorbed shortwave energy at the surface, followed approximately a decade later by simultaneous deviations of the four AEC terms. To better discern the relation between these four terms and the evolution of AA itself, the composite-mean differences of the energy budget terms are plotted against the composite-mean differences in AA over time (Figure 9b). We find that the difference in the composite-mean evolution of AA between the high- and low-AA models scales linearly with differences in the absorbed shortwave energy at the surface and hence is coincident with it. By contrast, differences in the AEC-related evolutions are delayed relative to the differences in AA itself (as evidenced by the near-zero AEC-related difference values at low but nonzero AA difference values), suggesting that enhanced precipitation-generated AEC is a response to, not a cause of, the enhanced AA in high-AA models.

How, then, should we interpret this full set of results? Considering only the DSE-generated and total AEC terms would suggest the conventional interpretation that as AA increases, DSE-generated AEC is reduced as a direct consequence of the reduced meridional gradient in temperature (Figure 6a). However, the added constraint that both precipitation-generated and evaporation-generated AEC scale with AA (Figures 5 and 7) suggests another possibility. Namely, we recognize that absent any other AA-related atmospheric heating processes, the first-order balance to the anomalous DSE-generated cooling must be latent heating associated with enhanced precipitation. Further, we recognize that as AA increases—most likely through reduction of sea ice leading to enhanced absorption of shortwave radiation at the surface—evaporation increases, increasing the moist static energy of the column. Results here suggest local surface-generated MSE leading to ascent culminates in enhanced precipitation and subsequent export of anomalous DSE from the region aloft. It is important to recognize that these terms are all deviations from the climatological behavior, and as such may not represent actual vertical ascent but a reduction in the strength of the descending branch of the polar cell, and by extension a reduction in the export of (low) DSE from the region. At the same time, these results implicate the role of moist processes (and enhanced precipitation) in mediating DSE-generated AEC associated with enhanced AA, a fact not fully appreciated until now.
The final question to be addressed here is how this enhanced precipitation (and export of DSE) is manifested in high-AA models, vis-à-vis low-AA models. As discussed elsewhere (Bengtsson et al., 2011; Bintanja & Selten, 2014; Kopec et al., 2016; Kug et al., 2010; Liu & Barnes, 2015; Woods et al., 2013) one option is that the excess precipitation is all generated “locally” by evaporation-supplied moisture. Alternatively, while evaporation may perturb the precipitation-producing processes (e.g., by reducing the static stability of the atmosphere and making it more conducive for generating precipitation), some of the moisture that precipitates out may arrive from outside the Arctic itself. While this question will not be answered conclusively here, composite-mean maps of precipitation and sTs trends for high- and low-AA models (Figure 10) suggest that Arctic precipitation trends for both low and high-AA models are coherent with extrapolar trends in the subpolar North Atlantic, eastern North America, and, particularly in high-AA models, the Asian continent. Differencing these fields highlights the strengthened connection between amplified Arctic precipitation and precipitation over the Asian continent, at the expense of its connection with precipitation over the North Atlantic. Further, although less robust, there is a geographic relation between enhanced extrapolar warming in the high-AA models with reduced precipitation, again most prominently over the North Atlantic but also apparent over the North Pacific as the well midlatitudes of North America and central Eurasia. In these regions, then, it may be that the local export of DSE enhances temperatures, thereby amplifying sTs trends but also stabilizing the atmosphere and hence reducing precipitation in these regions.

4 Discussion
In the preceding section, we use projections of future changes of the Arctic climate derived from fully coupled Earth system models to reveal an emergent behavior in which enhanced AA and its associated reduction in dry static energy convergence is balanced to first order by latent heating via enhanced precipitation. Here we attempt to determine whether the emergent behavior revealed through the CMIP5 analysis is consistent with the observed behavior over the historical record. The largest impediment to such a model evaluation is the dearth of robust, extensive, continuous, and accurate observations in this region, particularly with regard to the water cycle (e.g., ACIA, 2005; Hegerl et al., 2015; Vihma et al., 2016; Walsh et al., 2011). Further, reanalysis-based estimates show large discrepancies in both the energy and moisture budgets in this region (DuFour et al., 2016; Porter et al., 2010; Serreze et al., 2007). For this reason, we confine ourselves to the relatively direct, and simple, comparison of Arctic precipitation trends with AA of the overall global mean temperature trend. For observationally based estimates of precipitation, we use both the CPC Merged Analysis of Precipitation (CMAP) (Xie & Arkin, 1997) and Global Precipitation Climatology Project (GPCP) Version 2.3 combined precipitation (Adler et al., 2003) data, which are available at monthly time resolution from 1979 to 2016. The values are regridded from their native resolution to the same spatial resolution as the CMIP5 data. Grid point trends are calculated over the 36 year period 1979–2016 using the same method as described above. These trend values are then area averaged over the Arctic (70°–90°N).
For observationally based estimates of temperature, we find that commonly used gauge- and satellite-based records (including those found in HadCRUT4 (Morice et al., 2012), the NASA Goddard Institute for Space Studies (Hansen et al., 2010), and NOAA Global Surface Temperature V4.0.1 (Smith et al., 2008)) still have substantial missing data in the Arctic region. For that reason here we rely on observationally constrained reanalysis-based estimates, which are generally well representative of the observed state and show good agreement across data sets (Lindsay et al., 2014; Simmons & Poli, 2015). For our purposes we analyze data from the National Centers for Environmental Prediction-Department of Energy Reanalysis 2 (Kanamitsu et al., 2002) and ERA-Interim (Dee et al., 2011). As with the precipitation data, the near-surface (2 m) temperature values are regridded from their native resolution to the same spatial resolution as the CMIP5 data and grid point trends are calculated over the 36 year period 1979–2016 using the same method as described above. Since we are mainly interested in the magnitude of AA over the course of the observational record and not necessarily how its structure maps onto the intermodel difference pattern, we additionally revert to a conventional measure of AA, namely, the difference in the area averaged Ts trend values from 70° to 90°N minus the global mean trend values.
For comparison with the observationally based estimates of Arctic precipitation and amplification over the same time period, we concatentate the CMIP5 Historical simulation from a given model, which ends in 2005, with its continuation along the RCP8.5 trajectory, which commences in 2006. We then subselect data from the time period 1979–2016 and calculate the trends in Arctic precipitation and AA, as described above.
Two important results are obtained from this analysis (Figure 11). First, while there is large uncertainty in observed precipitation over this region (as documented elsewhere (ACIA, 2005; Vihma et al., 2016; Walsh et al., 2011) and as evident by the large difference in estimated trends from the CMAP and GPCP data), the emergent behavior revealed through the analysis of the CMIP5 data is qualitatively consistent with the observed behavior over the historical record. In particular, based upon the reanalysis data, the historical trajectory of observed AA is at the high end of the model projections for this time period. Correspondingly, the historical trajectory of observed Arctic precipitation spans the expected value derived from the CMIP5 results (i.e., the observations are not biased high or low with respect to the model estimate). Further, we find that the sensitivity of Arctic precipitation trends to the strength of AA within a given model (as represented by a least squares linear regression) during the historical period (1.24 W/m2/K) is nearly identical to that found over the course of the 21st Century under the RCP8.5 forcing (1.26 W/m2/K). We argue that the consistency of these two values across very different boundary forcing conditions, along with the fact that the observed record shows consistent changes to those in the models, is further evidence of a fundamental energetic constraint on Arctic precipitation as a function of AA.

5 Summary
In this paper we used the fully coupled atmosphere/ocean climate models of CMIP5 forced by similar trajectories of human-induced emissions of carbon dioxide and other greenhouse gases (RCP8.5) to quantify intermodel differences in the magnitude and structure of amplified warming projected to occur in the polar regions, with a specific focus on the Arctic. In fact, we explicitly demonstrate that some of the largest intermodel differences in surface warming, after accounting for differences in global mean temperature increases resulting from differing model climate sensitivities, occur over this region. Within the suite of CMIP5 models, enhanced Arctic amplification (AA) of the global temperature increase is accompanied by reduced atmospheric energy convergence (AEC).
Disaggregating the heating and cooling processes associated with AEC, we find relatively little relation between a model's AA strength and changes in atmospheric absorption of solar radiation within the model. Further, we find that models with enhanced AA tend to experience greater radiative cooling to space, which subsequently would need to be balanced by enhanced AEC rather than reduced AEC. However, the enhanced radiative loss to space is balanced by reduced (net) longwave radiative loss to the surface such that the change in net longwave radiative cooling (of the atmosphere) tends to be invariant with AA (which stands in contrast to intermodel differences at the global scale—Pendergrass & Hartmann, 2014).
Instead, reduced AEC within high-AA models is tied to increased surface turbulent (sensible and latent) heat fluxes. For all models, the change in sensible heat flux is small and effectively invariant with AA. Instead, reduced AEC in high-AA models is balanced principally by enhanced latent fluxes resulting from enhanced evaporation. However, the overall moisture flux convergence, and by extension atmospheric latent heat convergence, into the Arctic is not a function of AA, although we note that it is a linearly increasing function of overall global mean temperature change. This result has two important ramifications: (1) reduced AEC within high-AA models is primarily achieved through reduced dry static energy (DSE) fluxes into the Arctic and (2) enhanced Arctic evaporation within high-AA models is balanced by enhanced Arctic precipitation.
Together, these characteristics of high-AA models suggest an emergent relationship between the amplification of Arctic temperature increases (i.e., AA) within a given model and the amplification of Arctic precipitation increases in that model. Namely, within high-AA models, reduced AEC associated with reduced DSE convergence is balanced to first order by latent heat release via enhanced precipitation, that is, amplification of Arctic warming leads to amplification of Arctic precipitation. At the same time, the magnitude of these precipitation increases in the Arctic is constrained by the strength of AA via its influence on DSE fluxes into the Arctic.
More generally, the emergent behavior revealed through the analysis of the CMIP5 data—which is present across various boundary forcing conditions and is consistent with the observed behavior over the historical record—suggests underlying linkages between the energetics and hydrometeorology of the Arctic that may be relevant even outside the context of amplified warming of the Arctic in response to anthropogenic forcing. For instance, we show that within a given model changes in 20 year annual mean averages of DSE-generated AEC over time are linearly related to changes in Arctic precipitation. As such it may be that similar balances hold on interannual time scales as well (as found in other regions of the globe—e.g., Su & Neelin, 2003), although results are expected to be sensitive to how annual mean values are calculated with respect to the seasonal cycle. Further, we note that while this analysis does not necessarily clarify the mechanistic processes linking enhanced Arctic precipitation to enhanced DSE-generated AEC, nor to enhanced evaporation within high-AA models, recognition of the linkages between AEC and precipitation over the Arctic, as a function of AA, may help constrain various hypothesized processes. Indeed, multiple dynamic and thermodynamic processes could generate enhanced precipitation in response to enhanced AA (e.g., Abbot et al., 2009; Woods et al., 2013). In addition enhanced precipitation could draw from multiple local and/or remote sources of moisture (Bengtsson et al., 2011; Bintanja & Selten, 2014; Kopec et al., 2016; Kug et al., 2010; Singh et al., 2016). We argue that consistency with the emergent behavior of Arctic precipitation in response to enhanced Arctic warming as identified here should be a key criteria for the evaluation of these hypotheses.
Acknowledgments
B. T. A. acknowledges support of Department of Energy grant DE-SC0004975. We acknowledge the World Climate Research Programme's Working Group on Coupled Modeling, which is responsible for CMIP, and we thank the climate modeling groups (listed in Table 1) for producing and making available their model output. For CMIP the U.S. Department of Energy's Program for Climate Model Diagnosis and Intercomparison provides coordinating support and led development of software infrastructure in partnership with the Global Organization for Earth System Science Portals. To obtain model simulations of historical and projected temperatures from the Coupled Model Intercomparison Project 5 (CMIP5) multimodel ensemble used in this study, please see http://pcmdi9.llnl.gov/esgf-web-fe/. CMAP and GPCP Precipitation data provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, from their Web site at http://www.esrl.noaa.gov/psd/. NCEP_Reanalysis 2 data provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, from their Web site at http://www.esrl.noaa.gov/psd/