1. Introduction
How the atmospheric circulation will change due to anthropogenic global warming is an important open research question. While model projections of the thermodynamic response to global warming are rather robust, there are large differences between the modeled dynamic atmospheric circulation response of different models (IPCC 2013; Shepherd 2014; Vallis et al. 2015). Understanding the large-scale circulation response is of central importance because it is inherently tied to possible changes in the frequency, amplitude and volatility of weather regimes and extremes (Hannachi et al. 2017; Cassou 2008; Hoskins and Woollings 2015; O’Kane et al. 2013; Franzke et al. 2011, 2015), as well as to changes in the geographical distribution and amounts of precipitation (Knutti and Sedláček 2013).
In general, we expect that warmer temperatures will lead to a larger holding capacity of moisture in the atmosphere, due to the Clausius–Clapeyron relationship. All else kept the same, this increase in moisture is expected to enhance the precipitation patterns by the same amount. Such an increase, however, cannot be sustained, because the latent heating of the atmosphere is strongly limited by energetic constraints on surface evaporation (Held and Soden 2000) and the ability of the atmosphere to cool radiatively (Previdi 2010; O’Gorman et al. 2012). Rather, atmospheric convection as well as large-scale circulation patterns will be modified, leading to changes in the paths and locations of precipitating weather systems (Schneider et al. 2010). Thus, understanding the patterns of precipitation change requires a better understanding of atmospheric circulation changes. The interrelation between temperature, moisture, and global circulation changes is complex and raises a few fundamental questions regarding cause and effect.
An example is polar amplification. Since the polar regions are warming at the surface faster than the equatorial regions, the meridional temperature gradient in the lower troposphere is becoming weaker. This meridional temperature gradient is a major contributor to baroclinic instability, which drives the development of midlatitude storms (Holton and Hakim 2013). Thus, we expect eddy activity and corresponding eddy heat fluxes to weaken. Evidence for this occurring during summer is provided by Coumou et al. (2015, 2018). This is easy to understand if polar amplification is locally driven, by positive feedbacks involving changes in absorbed solar radiation of latent and sensible heat flux exchanges with the surface, and is expected to be accompanied by sea ice melting (e.g., Serreze and Barry 2011). However, various studies have shown that increased warming in the Arctic results from, or is intensified by, an increase in downward longwave radiation, due to the warmer and moister polar conditions, driven by stronger poleward fluxes of moisture or sensible heat (Gong et al. 2017; S. Lee et al. 2017; Franzke et al. 2017; Wang et al. 2017). Whether eddies and related energy fluxes have strengthened or weakened in the Northern Hemisphere over the past half century is still not clear, due to complexities arising from strong internal variability, zonal asymmetries and differences between the Atlantic and Pacific storm tracks, and a sensitivity to the specific diagnostics used (e.g., Wang et al. 2017; Blackport and Screen 2020). Shaw et al. (2016) discuss competing effects, such as the circulation responses due to either sea ice loss, or tropical warming, or the fast and slow responses to increased greenhouse gas concentrations. Those competing effects can lead to muted responses, hampering the detection of systematic circulation responses. Overall, however, model studies (e.g., Wu et al. 2011; Hwang and Frierson 2010) suggest a strengthening of the poleward energy transport, while observations (Chemke and Polvani 2020; Wang et al. 2017) suggest a weakening, or a wavenumber-dependent response (Lembo et al. 2019; Sang et al. 2022; Chemke and Ming 2020). In addition, moist effects can lead to changes in eddy efficiency, allowing for weaker eddy energies alongside stronger poleward moisture fluxes (Solomon 2006).
A more robust change in the global circulation is the poleward shift of the jets and storm tracks. While here most models agree (e.g., Yin 2005; Hartmann and Ceppi 2014; O’Gorman 2010; Simpson et al. 2014), and observational evidence of a poleward shift in the SH is clear (Archer and Caldeira 2008), the causes, and hence the relation to changes in the eddies and the mean flow are less clear. Furthermore, phase 5 of the Coupled Model Intercomparison Project (CMIP5) projections for the SH show a poleward shift of the jet stream and storm track (O’Gorman 2010; Simpson et al. 2014), while in the NH the signal is less clear due to seasonality and competing effects (Shaw et al. 2016; Shaw and Voigt 2015). This is further highlighted in the review by Shaw (2019), which describes 24 different theories, categorized by the initial cause for the shift. The robust changes in the global circulation are a feature of model studies; most CMIP studies focus on the trends toward 2100, while here we examine historical trends for the period 1958–2018. While the projected atmospheric temperature change pattern is robust in model runs, though biased warm in the tropical upper troposphere (Mitchell et al. 2013), observations of the warming pattern above the surface, in particular over the oceans and in the tropics, are only starting to emerge (e.g., Bindoff et al. 2013; Scherllin-Pirscher et al. 2021; Steiner et al. 2020). Furthermore, interannual variability is quite large. While the overall warming of the atmosphere and cooling of the stratosphere are very clear, the enhanced warming in the upper-tropospheric tropics is not as robust, and is mainly seen when taking into account recent radio occultation measurements (Steiner et al. 2020) or wind shear measurements (Allen and Sherwood 2008).
Tropospheric humidity is projected to increase with temperature so as to keep the relative humidity fixed (Held and Soden 2000; Schneider et al. 2010). This increased humidity contributes to the circulation changes both through the release of latent heat, and by affecting radiation, either directly via longwave absorption or through clouds, which influence the radiative heating and cooling via absorption and reflection of solar radiation, and the absorption and emission of terrestrial radiation. Predicting the overall influence of the different moisture effects on storms and circulation changes is challenging due to opposing effects (Shaw et al. 2016; Shaw 2019). In addition, there are challenges involved with observing and simulating the effects of moisture on the microscale physics as well as on the convective processes, making the predictions of circulation changes dependent on model setup and physics (Slingo and Slingo 1988; Voigt and Shaw 2015; Shaw and Voigt 2015; Shaw et al. 2016; Li et al. 2015; O’Gorman 2010; Schneider et al. 2010).
To gain deeper insight into past changes, we compute trends in various zonally averaged atmospheric circulation quantities in a long reanalysis dataset. We aim to examine the relationship between trends in different zonally averaged global circulation features, and their consistency. Specifically, we examine the following: 1) How have eddy fluxes of heat, moisture, and moist static energy changed over the reanalysis period, and how are these changes related to the zonal-mean trends in moisture and temperature? 2) How do trends in the different eddy and mean flow quantities relate to their climatological patterns? Specifically, do we see evidence of a poleward shift of the storm tracks and jet streams, and have the storm tracks strengthened or weakened over the past six decades? 3) Are the trends in poleward moist static energy flux consistent with trends in different diabatic heating terms and cloud distributions?
By addressing these questions, and noting the relative contribution of moisture transport and moisture-related diabatic heating to the trends, we also hope to get a sense of the specific contribution of moisture to future climate trends. We note that previous studies have either concentrated on observational trends in a specific component of the circulation, or have mainly used model data such as from the CMIP (e.g., Yin 2005; Barnes and Polvani 2013). In this study we aim to get a consistent picture of the trends in different components affecting the global circulation, using a long reanalysis dataset. In section 2 we describe the used data and methods, in section 3 we present our results, and in section 4 we provide our conclusions.
2. Data and methods
We use the daily mean Japanese reanalysis dataset JRA-55 (Kobayashi et al. 2015) for the period 1958–2018. We focus on the troposphere (1000–200 hPa) and examine zonal means over the following domains: (i) the whole longitudinal domain 0°–360°E, (ii) the Atlantic sector 250°–360°E, and (iii) the Pacific sector 120°–250°E. The boreal winter season covers the months December–February (DJF), while the boreal summer season covers June–August (JJA). We use JRA-55 reanalysis data because of their rather long time series, and they also provide various diabatic heating rate fields which other reanalysis products do not offer.
As every data product and model simulation, JRA-55 has some biases which are discussed here: https://climatedataguide.ucar.edu/climate-data/jra-55?qt-climatedatasetmaintabs=1#qt-climatedatasetmaintabs. In particular, relevant biases for our study are that JRA-55 has a warm bias in the upper troposphere, and that its upward global mean net energy flux is not well balanced, both at the top of the atmosphere and at the surface. A comparison of the zonal mean outgoing longwave radiation from JRA-55 and CERES (data available only since 2000) show quite a good agreement of the climatologies over this period. The trend fields (see also Hartmann and Ceppi 2014), which are quite noisy and not significant in large regions, do not match, suggesting we cannot really draw any conclusions from them. We do however have more confidence in the dynamical fields, which are better constrained. For example, Chemke and Polvani (2019, 2020) compared ERA-Interim, CFSR, NCEP2, and JRA-55 with respect to
Since our study focuses on trends, we carefully checked their robustness (see appendix A). To further examine the robustness of our trends we examined select time series to visually verify that we see a trend (see appendix A). In addition, to gain more confidence in specific trends, we examined the correlations between different quantities and verified they are consistent with the expected physical relations. While the correlation analysis does not necessarily capture the forced response and could be affected by interannual variability, our correlation results are consistent with our physical understanding of the trend behavior. To further examine whether the whole JRA-55 period is suitable for a trend analysis, we performed two types of tests. 1) We carried out a changepoint analysis for mean and variance using the R package “changepoint.” This analysis reveals no systematic occurrence of changepoints around 1979, when the use of satellite data started (see appendix A). 2) We repeated the trend analysis for the satellite period 1979–2018 and got qualitatively similar results (not shown). These tests are useful since the observation network has undergone considerable changes over time; for instance, the start of satellite data in 1979 has increased the quantity and quality of observations. Our two tests provide evidence that this does not affect the atmospheric variables we use for our trend analysis. In the following, we discuss the trends which, based on the above tests, we believe to be robust.
For the trend analysis we are using the modified Mann–Kendall test (Hamed and Rao 1998). This test takes into account the serial correlation of the data. The Mann–Kendall test (Mann 1945; Kendall 1948) is a nonparametric trend test. Trends are estimated using the Sen slopes (Sen 1968), which is a robust and nonparametric estimator for linear trends and can also be applied to non-Gaussian data. The trends in a given quantity are presented in terms of the units of that quantity per decade.
As will be shown later, the two dominant terms in the climatology (and also in the trends, not shown) are due to bandpass and low-pass filtered meridional flux divergences (terms χ9 and χ10, respectively), which represent the meridional flux divergence by the storm-track eddies and the quasi-stationary waves, respectively. We also examine the eddy kinetic energy derived from bandpass and low-pass filtered wind anomalies:
3. Results
a. Zonal background field trends
We start by examining the trends in zonal-mean temperature [T], the meridional temperature gradient, the specific humidity [q], and their combination with potential energy to yield the MSE trends. Figures 1a and 1b show the trends in [T] overlain on the climatological [T], for DJF and JJA, respectively. We see clear polar amplification—a peak warming over the polar lower troposphere during winter, more notable in the NH DJF, which is absent or very weak over the summer poles. While Arctic amplification started in earnest in the 1970s (England et al. 2021), analyzing the period 1958–2018 will lead to a slight underestimation of the magnitude of the warming trend but does not affect our conclusions. The NH winter polar amplification signal is similar to the NH annual mean trend projected by climate models (Collins et al. 2013). There are also negative temperature trends in the region of the upper troposphere and lower stratosphere over the poles. We also see a maximum warming trend in the upper tropical troposphere during NH winter, qualitatively similar to the annual mean signal in climate models (Collins et al. 2013). It is also consistent with recent observed trends, which only recently, with the inclusion of radio occultation observations, show a stronger trend in the upper tropical troposphere (Steiner et al. 2020). The SH trends show, in addition to the winter polar warming, a peak warming in the upper troposphere in midlatitudes during both seasons, and another peak in the subtropics during winter. This more complex SH pattern could partly be related to the ozone hole (Son et al. 2018).
Zonal-mean temperature trend (shading; K decade−1) and climatological zonal-mean temperature (contours; K) for the solstice seasons: (a) DJF and (b) JJA. Zonal mean specific humidity trend (shading; kg kg−1 decade−1) and climatological specific humidity (contours; kg kg−1) for the solstice seasons: (c) DJF and (d) JJA. Zonal mean moist static energy trend (shading; J kg−1 decade−1) and climatological fields (contours; J kg−1) for (e) DJF and (f) JJA. Only statistically significant trends at the 5% level are shown.
Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-21-0724.1
Looking at the second row of Fig. 1 we see a significant increase in moisture in the tropics, with a peak in the upper tropical troposphere and closer to the surface, alongside a strong drying in the subtropical lower troposphere above the boundary layer (800–500 hPa) during winter of both hemispheres. This subtropical drying, alongside warming results in a sub-Clausius–Clapeyron relation between global mean surface air temperature and near surface humidity of specific humidity with temperature of 3.3% K−1 in DJF and 5.1% K−1 in JJA (correlation values are 0.67 and 0.73, respectively, both are statistically significant at the 5% level). We also see a significant increase in upper-tropospheric (above 500 hPa)1 water vapor in the extratropics of both hemispheres, except above the winter subtropical drying centers, where the moistening is confined to above 300 hPa. This pattern is consistent with an overall moistening above the boundary layer due to the warming, alongside a strengthening of the moisture convergence–divergence patterns which accompany the tropical overturning circulation (cf. Held and Soden, 2006). We note however that the reanalysis products have been shown to have biases in the tropical and subtropical precipitation trends, when compared to satellite-based precipitation trends (Chemke and Polvani 2019). In the lower troposphere, we also see a moistening over the Arctic during NH winter, consistent with previous studies (Lee 2014; H. J. Lee et al. 2017; Yoo et al. 2012; Gong et al. 2017; Gimeno-Sotelo et al. 2018; Woods et al. 2013; Jakobson and Vihma 2010; Woods and Caballero 2016; Franzke et al. 2017).
The zonal-mean MSE field (Figs. 1e,f) shows positive trends almost everywhere, with the exception of the polar tropopause regions for NH summer and SH in both seasons, consistent with cooling in those regions. There is a very clear positive peak trend in the lower and midtroposphere over the winter poles of both hemispheres, with the NH peak clearly amplified near the surface. At lower latitudes, the positive trends are strongest in the summer tropics and subtropics of both hemispheres, with the trends being more evenly distributed with height compared to the temperature trends. During NH summer, the MSE increased significantly in middle and high latitudes, but not over the pole, consistent with the NH summer temperature trends being positive in middle and high latitudes, and decreasing to a weak positive value over the pole (Fig. 1b).
We note that the trends are positive even over the winter subtropical regions, despite the humidity trends being negative, implying the warming dominates in setting the trend sign there. This is consistent with the nonlinear relationship between specific humidity and MSE, shown clearly in the scatterplots of the monthly mean values of these two quantities at each grid point in the midlatitude troposphere (Figs. 2a,b). Increasing specific humidity is linked with increasing MSE. While for moderate values there seems to be an approximate linear relationship, for high specific humidity values the MSE increase seems to taper off, suggesting that specific humidity contributes to MSE only up to some threshold. This is particularly pronounced during winter season in the respective hemisphere. This effect is less visible during the respective summer season.
Scatterplots and regression lines for moist static energy (MSE; J kg−1) and specific humidity (kg kg−1) for (a) DJF and (b) JJA for all values at each of the data grids between 30°–60°N and 30°–60°S at 1000 hPa; bandpass filtered eddy kinetic energy EKE (m2 s−2) and bandpass filtered meridional moisture transport υBqB (m s−1 kg kg−1) for (c) DJF and (d) JJA for all values at each of the data grids between 30°–60°N and 30°–60°S at 1000 hPa.
Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-21-0724.1
We next examine the zonal-mean zonal wind trends, plotted alongside the climatology, shown in Fig. 3, for the global domain, and for the Atlantic and Pacific regions, respectively. The strongest positive zonal wind trends occur in the SH in both seasons. The trends indicate poleward shifts of the Southern Hemispheric jet stream, and are consistent with the shift toward a preference of the positive phase of the Southern Annular Mode (SAM) (Thompson and Wallace 2000; Kushner et al. 2001; Marshall 2003; O’Kane et al. 2013; Franzke et al. 2015). An exception is the South Pacific sector during austral winter, where the positive trend is leading to a more pronounced split jet structure since the poleward, eddy-driven jet stream is intensifying while the subtropical jet stream experiences no significant trend. Over the Antarctic a negative trend is visible, which suggests, in addition, a narrowing of the SH jet stream.
Climatological zonal-mean zonal wind (m s−1; lines) and trend (color shading; m s−1 decade−1) for the (left) NH winter season (DJF) and (right) NH summer (JJA), averaged over (a),(b) the whole globe, (c),(d) the Atlantic sector, and (e),(f) the Pacific sector. Only statistically significant trends at the 5% level are shown.
Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-21-0724.1
In the NH there is also a poleward jet shift visible which is mainly occurring in the Atlantic sector during DJF (Fig. 3c), consistent with the findings by Franzke and Woollings (2011) and Pena-Ortiz et al. (2013). We note that the dynamical nature of the jets is very different between the Atlantic which is eddy driven and the Pacific which is thermally and eddy driven (Li and Wettstein 2012), and indeed idealized studies have shown that the lack of poleward shifting over the Pacific is consistent with idealized model runs in which only an eddy-driven jet shows poleward shifting under increased GHGs, while a merged jet does not (Son and Lee 2005). CMIP simulations show consistent results for the lower troposphere (Simpson et al. 2014). In the NH during JJA, we see oppositely signed zonal mean zonal wind trends, which occur at different latitudes in the different basins. This is consistent with the meridional temperature gradients being very different during summer, with a peak warming in mid- to high latitudes rather than over the pole (Figs. 1a,b), consistent with the study by Coumou et al. (2015). In the tropical and subtropical regions negative trends are visible which manifest as an intensification and slight poleward shift of the zonal-mean easterly winds in the lower and middle troposphere in both hemispheres which is consistent for the SH with the findings by Lee and Feldstein (2013).
The temperature trend (Fig. 1) shows a strong strengthening of the climatological equatorward temperature gradients in the upper midlatitude troposphere near the tropopause break, which coincides with strong vertically concentrated positive vertical shear trends. The increase in upper-tropospheric vertical shear is mostly coming from the Atlantic sector, consistent with Lee et al. (2019). In the SH, we find a similarly located increase in upper-tropospheric vertical shear in midlatitudes and a decrease over the polar cap, which appear to be in thermal wind balance with the trends in meridional temperature gradients. The seasonality, however, is quite different from the NH. The midlatitude increase in vertical wind shear is stronger in summer, possibly due to the thermal response to ozone depletion (Son et al. 2018), while the decrease in vertical shear over the pole is stronger during winter, when polar amplification is stronger (Lu and Cai 2009).
b. Eddy trends
Next we examine trends in band- and low-pass eddies, which, as we will show later, dominate the meridional fluxes of moist and dry static energy. We start by examining the contribution of bandpassed eddies, which represent the midlatitude storm tracks. Figure 4 shows the trends, overlain on the climatology, of three bandpass eddy quantities—kinetic energy (EKE), meridional fluxes of heat (
Climatological zonal-mean fields (m s−1; lines) and trend fields (m s−1 decade−1; color shading) for the (left) NH winter season (DJF) and (right) NH summer (JJA) of the following eddy fields: (a),(b) bandpass filtered eddy kinetic energy (m2 s−2 decade−1), (c),(d) meridional heat flux (K m s−1 decade−1), and (e),(f) meridional flux of moisture in (g kg−1 s−1 decade−1). Only statistically significant trends at the 5% level are shown.
Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-21-0724.1
In the SH, in both seasons we see a clear strengthening of the climatological patterns of all three fields in midlatitudes, which is stronger on the poleward side of the climatological peak during summer (DJF). More specifically, EKE increases most strongly in the upper troposphere (above 400 hPa) where the EKE is strongest in the climatology, with the overall increase over the 61 seasons shown being about 45 m2 s−2, which is around 30%–35% of the peak climatological values of about 130–140 m2 s−2, depending on the season. In the lower troposphere the EKE trends are much weaker and have a more complex pattern, showing an overall strengthening, with a decrease in the subtropics during DJF. An examination of the trends in the Atlantic and Pacific regions suggests they constitute a strengthening and slight poleward shifting in the upper troposphere in the South Atlantic, a strengthening and slight equatorward shifting in the Atlantic during JJA, and a strengthening of the double-peaked storm-track structure in the South Pacific (not shown). The strengthening of midlatitude SH winter EKE is found in other reanalyses as well, and is much larger than that projected by models (Chemke et al. 2022). The heat fluxes also strengthen significantly, by about 20%–25% of the peak climatological values, and as with the EKE, during DJF the trends constitute a slight poleward shifting of the climatological pattern. This strengthening of the heat fluxes was recently suggested to be associated with the cooling of the Southern Ocean, and corresponding increase in meridional temperature gradients (Chemke and Polvani 2020), though in a later paper (Chemke et al. 2022) find a much stronger relation to the trends in the barotropic zonal mean zonal winds. The poleward bandpass filtered moisture fluxes strengthen by about 30% during both seasons, and also shift the climatological pattern slightly poleward, more so in JJA than in DJF.
In the NH, on the other hand, the bandpass eddy trend patterns are weaker, spatially more complex, and vary more strongly between the seasons. Overall, however, they show a weakening of the climatological patterns, opposite the SH trends. During DJF, at high latitudes (poleward of 55°N), we see a significant decrease in upper-tropospheric EKE (between 600 and 300 hPa) and lower-tropospheric heat flux (below 600 hPa), poleward of (and for EKE slightly below) the climatological peaks of these quantities. The moisture flux trends are largely not significant at these latitudes. Examination of the ocean basins separately shows the heat flux decrease is mostly coming from the Pacific (not shown), while the upper-tropospheric EKE decrease is mostly occurring in the Atlantic. We note that the lack of a significant trend in the lower troposphere is consistent with the general conclusion from past studies of storminess trends over the North Atlantic (e.g., Feser et al. 2015). Storms undergo very large interannual variations (von Storch and Weisse 2008), which lead to a rather low signal-to-noise ratio (Franzke 2012), hampering the detection of trends in storminess and EKE (Blackport and Screen 2020).
In the midlatitudes, between 30° and 50°N, both EKE and the heat flux show an increase in the upper troposphere (above 300 hPa), on the upper flank of the climatological EKE peak, and the moisture fluxes show a very weak but significant increase above 500 hPa. Unlike the EKE and eddy heat fluxes, the positive poleward moisture flux trends extend also to higher latitudes. While these trends are very weak, and occur in the upper troposphere which holds only a small fraction of the atmospheric moisture, this increase, of about 10% potentially can have a significant effect on the circulation through its effect on longwave radiation and cloud formation. The changes in midlatitude eddy fields come from the North Atlantic and the North Pacific, but they are associated with different changes in the climatological pattern. In the Atlantic (not shown) the significant positive heat flux trends are seen above 850 hPa in a narrow band between 40° and 60°N. Since both the EKE and heat flux trends occur poleward of the climatological fields, they will shift the area of maximum EKE also poleward. In the Pacific, on the other hand, the heat fluxes increase around 30°N, equatorward of the climatological peak, while the EKE trends occur where the climatological EKE is already strong and, thus, is strengthening the climatological peak and extend it upward. The poleward shifting in the Atlantic and strengthening in the Pacific are consistent with other observational based studies (Chang and Yau 2016) and future climate model simulations (Yin 2005; O’Gorman 2010) and are also theoretically expected to respond differently to external forcing (Son and Lee 2005), given that the Atlantic jet is eddy driven, while the Pacific jet is mixed eddy–thermally driven (Li and Wettstein 2012). In the lower-tropospheric subtropics (below 600 hPa) both EKE and moisture fluxes show a negative trend. The decrease of poleward moisture fluxes between 20° and 35°N is consistent with the observed subtropical drying and corresponding decrease in meridional moisture gradient.
During JJA, on the other hand, at high latitudes (poleward of 65°N) we see a slight but significant increase in upper-tropospheric (above 500 hPa) heat and moisture fluxes, which increase the climatological values by about 50% and 10%, respectively. An examination of the two ocean basins shows the increase in heat flux is mostly due to the Pacific storm track, while both basins contribute to the increase in moisture flux. In the midlatitudes (between 30° and 60°N), we see an overall decrease in EKE, and poleward heat and moisture fluxes, all occurring on the equatorward flank of the climatological peaks. The decrease in moisture fluxes is more pronounced, with the total decrease over the observational period reaching around 25% of the peak climatological fluxes. This is consistent with observations showing a decrease in NH JJA cyclone activity (e.g., Chang et al. 2016). In the subtropics (15°–30°N), the trend pattern is again complex, with a negative EKE trend below 500 hPa (coming mostly from the Pacific, not shown), and a positive trend above 300 hPa, alongside an increase in poleward moisture fluxes above 750 hPa. We note that the increase in poleward moisture fluxes in the Atlantic is significant at all latitudes above 500 hPa.
Further evidence for a close link between bandpass moisture flux and EKE is given by scatterplots and correlation analysis (Figs. 2c,d); here we use data aggregated over the midlatitudes and the troposphere. In the scatterplots it is visible that in magnitude increasing moisture fluxes are linked to increasing EKE. This is confirmed by correlation (all correlation coefficients are significant at the 5% level) and regression analysis. For this analysis we used a mass-weighted vertical average of the data over the midlatitudes (30°–60°N and 30°–60°S).
The pattern of strengthening in the SH and weakening during NH winter is also seen in the bandpass eddy temperature variance, shown in Fig. 5. In the SH the temperature fluctuations increase in the jet stream region, where also the climatological temperature fluctuations have their maximum amplitude, with the increase being stronger during DJF. In the NH, on the other hand, during DJF the temperature fluctuations strongly decrease on the poleward flank of their climatological peak (poleward of 50°N) and they increase slightly at their climatological peak in the midlatitude upper troposphere above 300 hPa, while during JJA there are no significant trends. The DJF changes in temperature variance in both hemispheres are similar to those found for ERAI by Tamarin-Brodsky et al. (2019) and for CMIP5 by Schneider et al. (2015).
Climatological zonal-mean fields (K; lines) and trend fields (K decade−1; color shading) of bandpass filtered temperature variance: (a) winter season (DJF) and (b) summer season (JJA). Only statistically significant trends at the 5% level are shown.
Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-21-0724.1
Figure 6 shows the trends in low-pass filtered EKE, heat, and moisture fluxes, plotted on top of their climatological fields. Compared to the bandpass eddies (Fig. 4), the climatological fields, and correspondingly the trends, have a more complex structure, but we also see a clear asymmetry between the hemispheres, with an overall strengthening of the climatological fields in the SH (with the exception of JJA moisture fluxes) and an overall weakening of the climatological fields in the NH. More explicitly, in the SH, during DJF we see a strengthening of the EKE and of the complex climatological meridional heat flux pattern, and an increase in poleward moisture fluxes in the upper troposphere (above 500 hPa). During JJA, the heat flux pattern strengthens, but unlike the DJF trends, the EKE strengthening is significant only above 300 hPa, and the poleward moisture fluxes are significantly decreased (i.e., [υLqL]L increases) between 25° and 55°S. Over the entire 61 JJA seasons, the peak trend amounts to about 30% of the climatological moisture flux value. This large decrease of poleward moisture fluxes is consistent with a response to the reduction in the mean meridional moisture gradient, implied by the drying in the subtropics alongside moistening at around 55°S observed during that time and location.
As in Fig. 4, but for low-pass fields.
Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-21-0724.1
In the NH, during DJF we see a significant decrease in low-pass EKE and in lower-tropospheric (below 500 hPa) poleward moisture fluxes, with no significant change in eddy heat fluxes. This weakening is in contrast to that found by Sang et al. (2022), suggesting there may be a difference between decadal variations and the long-term trends. During JJA, there are no significant changes in EKE or heat fluxes, while the poleward moisture fluxes decrease below 500 hPa and increase above. In the tropical region (30°S–30°N) the climatological heat fluxes are directed toward the winter hemisphere in the upper troposphere (above 300 hPa), and toward the summer pole in the lower troposphere (below 500 hPa), consistent with the monsoonal circulations. Looking at the trends, we see that during DJF, this pattern, which is shifted mostly into the NH, strengthens, while during JJA, when this pattern is shifted more to the SH, it is weakened on the NH side and strengthened on the SH side.
To summarize the eddy trends, in the SH we see a strengthening of the band- and low-pass filtered eddies and their fluxes during both seasons, with a poleward shift which is stronger during JJA, with the exception that low-pass moisture fluxes during JJA are strongly decreased. In the NH, on the other hand, there is an overall decrease in EKE and eddy fluxes in the lower troposphere, while in the upper troposphere the trends are more complex, with the robust feature being an increase in upper-tropospheric eddy moisture fluxes during both seasons and an increase in upper-tropospheric eddy heat fluxes during JJA.
It is interesting to relate the trends in eddy fluxes to the trends in the mean flow, especially given the overall opposing signs of eddy-related trends between the two hemispheres.
A crude examination of the trends in bandpass poleward heat fluxes (Figs. 4c,d) and zonal mean temperature (Figs. 1a,b) suggests the trends in the magnitude of the poleward eddy heat fluxes are overall negatively correlated with the large-scale meridional gradients of the temperature trends. Specifically, high-latitude NH eddies and poleward [T] reduction are both weaker during DJF, but the relation is not as straight forward for the SH, where in both seasons, the temperature trends assume a complex structure with regions of positive and negative meridional gradients.2 This is consistent with Chemke et al. (2022), who found that eddy amplitudes during SH winter are correlated with the barotropic, not the baroclinic mean flow structure. To further quantify this relation, we correlate the lower-tropospheric (800–1000 hPa) midlatitude (30°–60°N and 30°–60°S) monthly mean eddy bandpass heat fluxes and meridional temperature gradients, and get a correlation of −0.34 in the NH and −0.39 in the SH in DJF (statistically significant at the 5% level), indicating that a stronger poleward heat flux indeed goes along with a stronger poleward mean temperature reduction. This extends the findings of Chemke and Polvani (2020) that annual mean lower-tropospheric poleward temperature gradients and poleward heat flux magnitudes are negatively correlated. This correlation is consistent with our understanding of eddies and their relation to the mean flow but does not imply causality.
Relating the trends in temperature bandpass variance to the zonal mean temperature gradients, both first averaged over the lower troposphere (800–1000 hPa) and midlatitudes (30°–60°N or 30°–60°S) provides the following correlation values −0.34 in the NH and 0.38 in the SH, suggesting temperature variance is stronger when the meridional poleward zonal mean temperature decrease is stronger. This relation is consistent with a notion that for adiabatic flow on pressure surfaces, the temperature fluctuations are proportional to the mean temperature gradients; however, the simple single-peaked structure of the SH temperature variance trends, alongside the much more complicated zonal mean SH temperature trends suggests further study is needed to explain the relation between the eddy and mean flow trends.
Finally, we examine the relation between the trends in eddy moisture fluxes (Figs. 4 and 6) and the mean meridional moisture gradient (as deduced from Figs. 1c,d). Again the two hemispheres show a different behavior. In the NH, eddy moisture fluxes have oppositely signed trends in the upper and lower troposphere. Consistently, we see oppositely signed meridional [q] gradients. In the lower troposphere, the poleward moisture fluxes poleward of the subtropical drying region decreased, while in the upper troposphere, both the midlatitude moisture fluxes and the equatorward mean moisture gradients have strengthened. In the SH, on the other hand, the low-pass and bandpass eddies show oppositely signed trends during JJA. Thus, the significant subtropical drying during JJA goes along with a weakening of the low-pass poleward eddy moisture flux (correlation coefficient −0.41,3 statistically significant at the 5% level), but a strengthening of the bandpass poleward eddy moisture flux (at a slightly more poleward location, correlation coefficient of 0.14, statistically significant at the 5% level). The weakening of the low-pass eddy moisture flux is consistent with a reduction of moisture supply from the dryer subtropics, while the drying of the subtropics is consistent with the bandpass eddies fluxing more moisture out of the region (e.g., Held and Soden 2006), though this drying is also consistent with the stronger Hadley circulation. We also see positive moisture trends at high latitudes (around 50°–60°S during JJA and around 60°–70°S during DJF), which are consistent with stronger bandpass eddy moisture fluxes during both solstice seasons.
The significant weakening of the lower-tropospheric heat and moisture fluxes in the NH during both seasons is interesting in the context of what causes polar amplification. While local amplification of surface feedbacks due to sea ice melting and local lapse rate changes have been shown to contribute (e.g., Screen and Simmonds 2010), various studies have shown that the stronger warming over the Arctic is associated, at least partly, with strong episodic eddy-driven fluxes of heat and moisture, which cause the downward longwave radiation at the surface to increase (Park et al. 2015; Liu et al. 2021; Yang et al. 2010; Serreze and Barry 2011). It is not clear how the strong and significant negative lower-tropospheric heat and moisture flux trends fit the above findings. Specifically, this raises the question of whether the weak but significant increases in upper-tropospheric moisture fluxes, and to some degree heat fluxes, can add to the driving of polar amplification in the lower troposphere. In addition, GCM studies predict a strengthening, rather than a weakening, of heat and moisture fluxes, based on trends in the meridional gradient of the TOA energy input (Hartmann and Ceppi 2014; Franzke et al. 2017; Zelinka and Hartmann 2012; Previdi et al. 2021). In the next section we will examine the meridionally varying moist static energy budget to examine these issues.
c. Moist static energy budget
To get a sense of the effect that the eddy trends have on the zonal mean moisture and temperature fields, we next examine the vertically and zonally averaged meridional convergence of the meridional MSE fluxes [a vertical average of Eq. (3)]. We start by examining the contribution of different terms in Eq. (4) to the climatological fluxes (Fig. 7). For comparison, we show both the dry and moist fluxes. We find that the meridional static energy budgets are dominated by the contributions of high-pass and low-pass eddies [terms χ9 and χ10 of Eq. (4), respectively]. Both χ9 and χ10 have sizeable amplitudes mainly poleward of 20° and act to decrease static energy in the subtropics, around 30°, and increase it at mid- and high latitudes, indicative of a poleward transport of static energy from the subtropics to high latitudes. Bandpass eddies dominate in the subtropics and midlatitudes, while low-pass eddies dominate at high latitudes, and to a lesser extent in the tropics. We note that the dominance at high latitudes of low-frequency planetary-scale waves over synoptic-scale eddies has also been found in previous studies (e.g., Nie et al. 2013; Lachmy and Harnik 2014, 2016). Comparing the dry and moist flux convergences suggests the relative contribution of moisture fluxes, in both hemispheres, is larger during summer compared to winter, larger for bandpass relative to low-pass eddies, and is larger in the SH. Overall, the contribution of bandpass moisture fluxes can reach around 30% (e.g., the DJF SH χ9 peak at 60°S increases from around 40 W m−2 to about 60 W m−2).
The climatological vertically and zonally averaged meridional convergence of the different terms (W m−2) contributing to the low-pass DSE and MSE fluxes [Eq. (4)]. Shown are (a) DJF DSE, (b) DJF MSE, (c) JJA DSE, and (d) JJA MSE.
Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-21-0724.1
Figure 8 shows the trends in the vertically averaged meridional MSE flux convergence, overlain on the climatological values. We note that the climatological MSE fluxes as well as their trends are quite constant with height, unlike, for example, the EKE fields and trends which are concentrated in the upper troposphere, or the eddy moisture fluxes which can change sign between the upper and lower troposphere. Consistent with the other eddy fields, in the SH, the bandpassed eddy-heating trends act to strengthen the climatological pattern during JJA, and to strengthen and shift it slightly poleward during DJF. An estimate of the total contribution4 over the analysis period yields trends which are about 25% and 40% of the climatology during JJA and DJF, respectively. A comparison to the corresponding DSE terms (not shown) shows that moisture flux convergence contributes significantly to these trends: during DJF, about a third of the high latitude heating and a half of the midlatitude cooling are due to moisture. During JJA moisture fluxes are responsible for about half of the trends, and they also act to widen the range of their influence (the subtropical cooling and high-latitude heating peaks due to DSE flux convergence are slightly closer together than the corresponding MSE-flux convergence peaks, not shown). The trends in the low-pass eddy heating are much less significant, with only a narrow region showing a strengthening of the heating at high latitudes (80°–90°S) during DJF.
Climatological zonal-mean fields (red lines; W m−2) and trend fields (black lines; W m−2 decade−1) of the vertically averaged low-passed meridional moist static energy flux convergence terms by bandpass eddies (χ9) for (a) DJF and (b) JJA, and by low-pass eddies χ10 (c) DJF and (d) JJA. Only statistically significant trends at the 5% level are shown.
Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-21-0724.1
In the NH, the trends are overall much less significant. The main significant signal is a strong negative trend of the low-passed eddy heating over the Arctic during DJF (with a hint of similar trends in the bandpassed eddies), consistent with a weakening of the poleward eddy MSE fluxes. Over the 61-yr observational period, this trend amounts to a weakening of about 25% of the climatological heating peak over the Arctic. A comparison to the DSE flux convergence suggests moisture is acting to reduce these trends (the DSE flux convergence peak has weakened by about 35%, not shown). This positive contribution of low-pass eddy moisture flux convergence is not expected from an examination of the low-pass moisture flux trends, which are weak and negative below 600 hPa and are weakly positive above 400 hPa. This suggests the changes in moisture fluxes are complex and need more thorough examination.
The vertically averaged zonal-mean MSE flux convergence should theoretically be balanced by the zonally averaged EIA, calculated by summing up the terms in Eq. (5). This is implied from the vertical average of Eq. (3), assuming the time derivative of vertically averaged [h] is negligible.5 Physically, we expect the poleward MSE flux to be stronger when the equatorward meridional gradient of EIA is stronger, implying a negative correlation between the meridional gradient of EIA and MSE flux. Mathematically, this correlation is obtained by taking a meridional gradient of the vertically averaged Eq. (3), which gives a balance between the meridional gradient of EIA and the second derivative of the meridional MSE transport, and noting that for a relatively smooth structure of a single midlatitude poleward flux peak, is roughly proportional to minus the flux itself. Given the overall strengthening of the heat and moisture fluxes in the SH, and their overall weakening in the NH, we expect the overall meridional gradients of EIA to increase in the SH and decrease in the NH.
We note that there are large biases in the JRA-55 surface (C. Liu et al. 2017) and top of the atmosphere (Kobayashi et al. 2015) energy fluxes, making it hard to draw significant conclusions from examining the EIA budget.
These discrepancies are indeed evident even in the climatology, for which the EIA (red curves in Figs. 9a and 10a) does not seem to match the climatological MSE flux divergence (minus the sum of the red curves of the top and bottom rows of Fig. 8). Nonetheless, it is interesting to see whether the overall structure of the EIA trends is consistent with the strengthening of the eddies in the SH and their overall weakening during NH winter, and if they are, which terms in the energy budget contribute to these trends in the reanalysis.
Zonal-mean climatology (red curves; W m−2) and trends (black curves; W m−2 decade−1) of different terms in the energy input into the atmosphere (EIA) budget for DJF. Shown are, top left to bottom right: (a) EIA, (b) OLR, (c) net atmospheric SW absorption, (d) latent heat flux, (e) sensible heat flux, and (f) upward LW radiation at the surface. Only statistically significant trends at the 5% level are shown.
Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-21-0724.1
Figures 9a and 10a (black curves) suggest this is indeed the case. In the SH, during both seasons, we see significant trends which tend to strengthen the climatological pattern of a net energy input into the atmosphere equatorward of 30°S and a net energy output poleward of 30°S. Over the observational period, the EIA trends strengthen the climatological pattern by about 25% in JJA and by about 15% during DJF, with a slight poleward shifting of the climatological pattern during DJF. This relation, however, is not quantitatively robust (the correlation coefficient between the meridional gradient of EIA and the vertically averaged MSE fluxes is 0.16, statistically significant at the 5% level; first averaged over 30°–90°S.
In the NH, on the other hand, the EIA trends during both seasons are much more constant in latitude compared to the climatology (except for the Arctic during DJF), indicating a weakening of the negative climatological meridional EIA gradient, consistent with the MSE fluxes weakening (Figs. 4 and 6). We also see a strong localized positive peak in the Arctic during DJF, which, over the observational period, consists of a decrease of the climatological atmospheric cooling by about 8% (an increase of about 10 W m−2 with a climatological cooling of about 125 W m−2), which is consistent with Arctic amplification of the temperature trends (Fig. 1).
Figures 9 and 10 also show the different terms contributing to the EIA (both the climatology and the trends), as listed in Eq. (5), for DJF and JJA seasons, respectively. Note that the OLR indicates an atmospheric cooling; thus, it is subtracted from all other terms to yield EIA. It is important to note that while the climatological EIA curve is the sum of the surface fluxes, SW heating, and minus OLR, the statistically significant trends in EIA cannot be deduced from the statistically significant trends in the individual terms, due to possible correlations between the different rather noisy terms, which cancel when summing up, resulting in the EIA trends being significant in a much wider latitudinal range than most other terms.
In the SH during JJA (Fig. 10), we see a significant reduction of the latent and sensible surface heat fluxes at midlatitudes, which peak between 60° and 70°S. A sharp significant cooling peak is also evident in the surface LW fluxes. This is consistent with a cooling of the ocean surface at these latitudes, noted by several studies (e.g., Armour et al. 2016; Zhang et al. 2019), and a strengthening of the SST gradients there, as pointed out by Magnusdottir and Saravanan (1999) and Woollings et al. (2010). We note, however, that an examination of the zonally averaged ERSSTv5 (Huang et al. 2017) SST trends for the period of our study (1958–2018) shows weak positive, or nonsignificant trends over the Southern Ocean, but there are significant positive SST trends to the north, which result in a significant sharpening of the meridional gradients (not shown).
The reduced net upward surface LW fluxes are strong enough to give a strong concentrated negative peak in the EIA of about 37 W m−2 over the 61-yr observational period (despite an opposing decrease in OLR). At the same time, we see significant and large positive trends in SW absorption in the tropics, and in the surface latent heat flux in the subtropics, alongside a strong reduction in OLR, all of which contribute to the positive EIA trend equatorward of 30°S. This subtropical atmospheric heating, alongside a midlatitude cooling, which sharpens the SH climatological meridional gradient in EIA, balances the strengthening of the SH eddy MSE fluxes. Interestingly, the temperature trends which accompany these changes in the MSE budget (Fig. 1b) are quite complex in structure, with significant subtropical cooling in the lower troposphere (below 600 hPa) and significant warming above.
The strong concentrated high-latitude negative surface sensible heat flux trends are not seen during DJF (Fig. 9). Rather, we see a significant increase of the latent heating at all latitudes, which peaks at 30°S, and strengthens the climatological meridional gradient. We note that the hemispheric increase in latent heating which peaks in the subtropics is similar to that found in Tan and Shaw (2020) in response to increased CO2. Using a series of model simulations, they showed that this increase was due to the increase in surface specific humidity, rather than surface wind changes, and was the second largest contributor, after cloud changes, to the corresponding poleward shifting of the jet. In the tropical region, both the SW absorption and the OLR trends contribute to the increased EIA there. This, combined with a strong minimum in the LH trends combine to give a relatively flat increase in EIA in the tropics. As we will see in the next section, these changes are consistent with an increase in LW cloud radiative effects and high cloud cover in the tropics. At higher latitudes, due to the low statistical significance of the individual terms, it is not clear from Fig. 9 what contributes to the reduction of EIA poleward of 40°S. There is a statistically significant reduction in sensible heat fluxes, but is not large enough to explain the significant reduction of EIA. A breakup of the surface LW flux trends into the upward and downward components (Fig. 11) suggests some of this reduction can be explained by the increase in downward surface LW flux, which shows significant increases at all latitudes, of about 0.6 W m−2 decade−1.
Zonal-mean climatology (red curves; W m−2) and trends (black curves; W m−2 decade−1) of the downward LW radiation at the surface, for (a) DJF and (b) JJA, and of the upward LW radiation at the surface, for (c) DJF and (d) JJA. Only statistically significant trends at the 5% level are shown.
Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-21-0724.1
In the NH, the trend patterns are very different. Mainly, we see strong positive trends in the surface fluxes over the Arctic during DJF, with a main peak at around 80°N in all surface fluxes (Figs. 9d–f), as well as in the upward surface LW flux (Fig. 11c), and a second peak in the latent and sensible heat fluxes between 50° and 70°N. These positive trends are counteracted by an increase in OLR (Fig. 9b) and surface downward LW fluxes (Fig. 11a) over high latitudes (poleward of 60°N). The sharp increases in surface fluxes over the Arctic region (poleward of 70°N) are likely caused by the observed retreat in sea ice, and explain the sharp significant increase in EIA over the Arctic during DJF. We note, however, that a retreat of sea ice does not directly explain the sharp increases of latent and sensible heat fluxes between 50° and 70°N (where EIA trends are not significant).
To further understand the relation between eddy fluxes and the Arctic EIA, we examine the correlation between the vertically averaged meridional moisture flux at 60°N and surface latent heat flux averaged over the polar cap north of 60°N. We find a correlation coefficient of −0.21 in DJF and −0.53 in JJA. Interestingly, we also find a significant correlation between the moisture flux and surface sensible heat flux, of −0.21 in DJF and −0.54 in JJA, and with EIA, of −0.11 in DJF and −0.55 in JJA (all correlations are statistically significant at the 5% level). While the correlations cannot indicate causality, the fact that we find the moisture fluxes to be significantly correlated not only with the latent heat fluxes, but also with the EIA and sensible heat fluxes, suggests that the heating of the Arctic results in more moisture, which reduces the meridional mean moisture gradient, and consistently, the moisture fluxes. This seems to contradict various studies that suggest an increase in poleward moisture fluxes contributes to Arctic amplification (Park et al. 2015; Alekseev et al. 2019; Hao et al. 2021). We note that if stronger moisture fluxes were behind a moister Arctic, via increased downward LW fluxes to the surface (which warm the surface causing sea ice melting and more sensible and latent heat fluxes), this would give a positive correlation between moisture fluxes and the latent and sensible heat fluxes. We discuss this further in the concluding section.
d. Longwave cloud radiative effect
To further understand the trends in the LW diabatic heating terms of Figs. 9–11, we examine trends in longwave cloud radiative forcing (Fig. 12). By this we mean the difference between all-sky and clear-sky longwave radiative input into the atmosphere; this will isolate the effects of clouds on longwave radiative energy input into the atmosphere. We note that the net LW heating is the upward minus downward surface LW fluxes, minus the OLR (subplot f minus b in Figs. 9 and 10). Cloud cover has also been shown to play a role in storm-track dynamics, mostly via the longwave cloud radiative effects, which influence vertical stability and baroclinicity, as well as the strength of the eddies (Voigt and Shaw 2015; Shaw et al. 2016; Ceppi et al. 2017), in particular at upper levels, and related eddy momentum fluxes (Li et al. 2015). Looking at Fig. 12, we see positive trends in the tropics and high latitudes in both hemispheres, while there are statistically significant negative trends in the midlatitudes. This is the case for both seasons, though JJA has less significant latitudes; hence the trends are more dominant during DJF. The trends in the mid- to high latitudes indicate poleward shifts of the climatological longwave cloud radiative effect in both hemispheres, consistent with a poleward shift of the jet stream and storm tracks in the midlatitudes (Voigt and Shaw 2015; Bender et al. 2012; Norris et al. 2016; Ceppi and Hartmann 2015).
Climatological zonal-mean fields (W m−2; red line) and trend fields (W m−2 decade−1; black line) for longwave cloud radiative effect (difference between all-sky and clear-sky radiative energy input into the atmosphere). Global zonal mean for (a) DJF and (b) JJA. Only statistically significant trends at the 5% level are shown.
Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-21-0724.1
An increase in low cloud cover over the Arctic has been suggested as a possible positive feedback contributing to polar amplification, through the amplification of downward longwave radiation to the surface (e.g., Serreze and Barry 2011). An increased surface downward LW flux implies a net LW atmospheric cooling, suggesting further study is needed to understand the role of clouds.
4. Summary
In this study we examined long-term trends in the atmospheric circulation and its forcing terms based on the JRA-55 reanalysis data from the years 1958–2018. A changepoint analysis reveals no systematic shift at the start of the satellite era 1979, and a trend analysis over the satellite era 1979–2018 shows qualitatively similar results. These, along with a visual examination of select time series, raise our confidence that our trend results are robust.
We find marked differences between the hemispheres—while in the Southern Hemisphere there is an overall strengthening and poleward shifting of the jet stream, the eddies, and the meridional diabatic heating gradients, in the Northern Hemisphere there is an overall weakening of the eddy components and diabatic heating gradients.
More specifically, our main findings are as follows:
-
We find increased zonal-mean moisture during both seasons in the upper and lower-tropospheric tropics, during summer also in the upper-tropospheric midlatitudes, which in the NH also extends over the Arctic. In the SH there is a narrow region of moistening over the Southern Ocean during both seasons. In the NH during winter, we also see a moistening of the Arctic lower troposphere. There is also a significant drying during winter in the subtropical lower troposphere of both hemispheres.
-
We find significant poleward shifts of the jet streams, which is more pronounced in the SH compared to the NH, consistent with previous studies (e.g., Archer and Caldeira 2008; Woollings and Blackburn 2012; Pena-Ortiz et al. 2013; Butler et al. 2010; Barnes and Polvani 2013; Simpson et al. 2014). The poleward shift in the SH is consistent with the trend toward the positive phase of the Southern Annular Mode. It is more pronounced during summer, when it is accompanied by a poleward shifting of the various bandpassed eddy fields, latent heating, and diabatic heating. In the NH, the jet stream trend is mainly in the boreal winter season and in the Atlantic sector; in the NH Pacific sector we find neither a significant jet stream shift nor significant changes in jet stream intensity.
-
The SH storm tracks have on the whole strengthened. This is manifest as a strengthening of the EKE and temperature variance, and of the various eddy fluxes, consistent with the strong strengthening of the SH winter synoptic-time-scale eddies found by Chemke et al. (2022). Consistently, the meridional gradients of the total energy input into the atmosphere, and of the latent heating and shortwave absorption, have also strengthened. Over the 61-yr observational period, the vertically averaged meridional MSE flux convergence peaks have strengthened by about 40% (25%) and the bulk meridional EIA gradients have strengthened by about 25% (15%) during JJA (DJF). An exception to the strengthening of eddy fluxes is a significant weakening of the low-pass moisture fluxes during JJA in midlatitudes. We note that there is a drying trend over the subtropics alongside a moistening trend at higher latitudes, during this season. The strengthening of the bandpass eddy moisture fluxes alongside a weakening of low-pass eddy moisture fluxes when the equatorward meridional mean moisture gradient has decreased suggests the low-pass moisture fluxes weaken in response to the mean gradient, while bandpass fluxes contribute to the mean gradients.
-
In the NH, the trend patterns are quite complex and over large regions not significant. We see an increase of bandpassed EKE and temperature variance in the upper-troposphere subtropics (and in winter also midlatitudes). At the same time there is a decrease in winter EKE and temperature variance, over high latitudes for bandpass eddies and over the entire hemisphere for low-pass eddies, and a decrease in summer bandpass EKE in the subtropics and midlatitudes, consistent with Coumou et al. (2015). Overall, there is a reduction of poleward fluxes of temperature and moisture in the lower troposphere, most notably, a strong reduction of low-pass moisture fluxes during winter. At the same time, there is a weak but significant increase in poleward moisture fluxes above 500 hPa, and an increase in upper-tropospheric bandpass eddy heat fluxes during JJA. The vertically averaged meridional MSE flux convergence on the whole shows no significant trends, except for a strong reduction of the low-pass MSE flux convergence over the Arctic during winter, of about 25% of the climatological eddy heating in that region. Alongside the weakening of the eddy fluxes, over the winter Arctic we see a very strong increase in the surface fluxes (latent, sensible, and net upward longwave), in the total EIA, and in cloud LW heating (alongside an increase in low and medium cloud cover), consistent with local feedbacks amplifying Arctic amplification, and the corresponding reduced temperature gradients resulting in reduced eddy fluxes.
-
We find evidence for a close link between moisture and energy. An increased moisture flux is associated with increasing kinetic energy in both hemispheres. The increased moisture flux is also linked to increasing temperatures.
Furthermore, we see a clear polar amplification—a peak warming over the polar lower troposphere during winter, which is absent or very weak over the summer poles, consistent with previous studies (Screen and Simmonds 2010). The NH winter polar amplification signal is similar to that projected by climate models in its sign and spatial extend. We also see a maximum warming trend in the upper tropical troposphere during NH winter, consistent with but weaker than that projected by climate models (Collins et al. 2013), and similar to recent trends based on radio occultation observations (Steiner et al. 2020).
Examining the trends in eddies, diabatic heating, and the zonal-mean fields suggests an overall consistency between the hemispheric-mean changes in eddy fluxes, and the bulk hemispheric changes in zonal mean gradients of temperature, moisture, and EIA, with a more complex relation between the local details of these fields.
A caveat of our analysis is that the EIA and eddy trends do not correspond well. Kobayashi et al. (2015) already pointed out that the global mean energy balance at the surface and the top of the atmosphere are not well balanced. Another study by Schmeisser et al. (2018) showed biases in 5 reanalysis products (including JRA-55). Since clouds will affect the computation of EIA, clouds seem a likely cause of this discrepancy between the EIA and eddy trends. A further caveat of our analysis is that the discrepancy of the global surface fluxes is large, as concluded from a comparison between five atmospheric reanalysis datasets, with the zonal mean standard deviation estimated to be about 11 W m−2 (C. Liu et al. 2017). For these reasons we trust the eddy fluxes in our study more than the radiative fluxes. This calls for improvements in the radiative fluxes in reanalysis datasets (Liu et al. 2015).
A main question our results raise is why the SH eddies have strengthened. One suggested mechanism has been the strengthening of eddy growth due to sharper SST gradients. This mechanism requires some diabatic mechanism to communicate the meridional surface temperature gradients to the free troposphere. The significant reduction in surface fluxes over the cooler winter Southern Ocean (Figs. 10d–f), alongside the strengthening of the climatological patterns of diabatic heating are all consistent with such changes, but more study is needed to examine the specific processes at work. During summer, we do not observe the sharp reduction in surface fluxes over the Southern Ocean as we do during winter. Instead, we see an overall strengthening of the latent heat fluxes, which strengthen the meridional gradient of EIA. We note that Chemke et al. (2022) found that SST biases are not the main source of model biases in SH winter bandpass eddy trends. Rather, they found a strong correlation between the bandpass EKE and MSE fluxes, and the barotropic zonal mean zonal wind structure. This suggests a different physical mechanism behind the observed eddy trends, but the exact dynamical link between the eddies and the barotropic flow structure is still unclear.
In the NH, the weaker eddy fluxes are consistent with a diffusive response (Rose et al. 2014; Roe et al. 2015; X. Liu et al. 2017) to the weakening of the meridional temperature gradient at high latitudes which accompanies polar amplification. The strong increases in the surface fluxes and the LW cloud radiation effect over the Arctic during winter are consistent with the polar amplification being driven by local feedbacks involving sea ice melting and clouds (e.g., Serreze and Barry 2011). It is not clear how our findings fit studies which have shown poleward moisture fluxes result in a strong surface Arctic warming (Park et al. 2015; Liu et al. 2021; Yang et al. 2010; Previdi et al. 2021). The decrease in poleward moisture fluxes, both for bandpass and low pass eddies, is perhaps the most surprising result of our study. According to the recent review by Previdi et al. (2021), while models differ in the sign of the NH winter poleward energy transport trends, they all essentially project that moisture fluxes will increase while heat fluxes will decrease, but the relative magnitudes of the heat and moisture fluxes differ between the models, resulting in some models projecting an increase and some an overall decrease of total poleward energy fluxes. A decrease in poleward moisture fluxes is also opposed to projected model changes using idealized models and basic theoretical arguments (e.g., Held and Soden 2006; Schneider et al. 2010), which suggest the increase in moisture content will necessarily increase moisture fluxes if the circulation statistics do not change. An exception is the recent study by Bonan et al. (2023), which projects a decrease in poleward moisture fluxes in an energy balance model that accounts for the subtropical drying by the Hadley circulation. Indeed, we find a robust drying of the subtropics during winter, when the decrease in moisture fluxes is most robust. We also note that while we see a clear reduction in near surface moisture fluxes, we do see an increase of moisture and poleward moisture fluxes in the upper troposphere. Clark et al. (2022) is a recent study examining MSE transport trends in multiple reanalysis products. They find that the four reanalysis products used (including JRA-55), show largely agreement the eddy MSE transport trends but show disagreements in the zonal mean MSE flux. While they decompose the flow fields differently from our study, our results are largely consistent. Also, Shaw et al. (2022) use four reanalysis products (including JRA-55) to compute energetics, including MSE, they only compare the trends of kinetic energy among the reanalysis products and the trend of the energy flux as implied from the top-of-atmosphere radiative energy. While the kinetic energy trends are very similar, the energy flux trends show more marked differences and do not necessarily agree on the sign; however, they do not seem to have carried a trend test analysis. These trend disagreements suggest that this aspect of reanalysis data needs to be improved.
This raises some interesting questions. Francis and Hunter (2006) showed that increasing the downward surface longwave fluxes is a main mechanism by which enhanced eddy fluxes increase Arctic surface temperature (see also Gong et al. 2017; S. Lee et al. 2017; Franzke et al. 2017; Wang et al. 2017). We indeed observe such increases over the Arctic, especially during winter (Fig. 11). Part of this increase is likely due to increased cloud LW heating (Fig. 12), and low- and medium-level cloud cover (not shown) but our results raise the possibility that an increase in upper-tropospheric water vapor also contributes to the positive downward surface longwave trends. Woods and Caballero (2016) found that the moisture intrusions which warm the Arctic are bottom weighted, suggesting lower-level moisture is key; however, they examined the most extreme events, which are spatially localized, meaning their contribution to the time mean zonal mean could be small. Moreover, while they showed that the frequency of these strong moisture intrusions has been growing, they only consist of about 30% of the poleward moisture transport. It is possible that overall, lower-tropospheric poleward moisture fluxes have decreased, but the number of extreme moisture intrusions, which are likely also deeper than the weaker ones, result in an overall increase of upper-tropospheric moisture and moisture flux.
Overall, our trend analysis suggests a complex relation between the large-scale changes in the zonal means, the eddies and associated poleward MSE fluxes, and the diabatic heating fields. Understanding these relations from observations is challenging given the large biases and lack of long enough observations of the surface and top-of-atmosphere energy budgets. Specifically, the question of the underlying causes, or constraints, which shape the response, and in particular the asymmetry between the hemispheres, are still not clear. Our results indicate a role for high-latitude surface processes, but this does not exclude a fundamental role for changes in the tropical circulation and subtropical regions, especially in affecting poleward fluxes of moisture. To better understand how global warming, and specifically the corresponding increased moisture, affect the atmospheric circulation and storms, carefully designed model experiments are needed.
The significant moistening extends also to the lower troposphere over the Atlantic (not shown).
The Sen slope trend calculation is nonlinear, meaning the meridional gradient of the trends is not necessarily equal to the trends of the meridional temperature gradients. Both of these quantities are quite noisy and did not yield any clear significant results.
Correlations are performed for averages over 30°–60°N/S and 500–1000 hPa.
A bulk MSE flux contribution is taken to be the bulk meridional heating gradient, estimated by taking the difference between the high latitude positive peak and midlatitude negative peak, and comparing the six-decade trend value to the climatological value.
We note that we only applied the temporal filtering to the eddy flux terms, and have only used the low-passed MSE flux convergences [Eq. (4)] in our trend calculations, after removing a time mean. By definition of the trends, we expect them to be dominated by the low-pass part of the flow, suggesting trends in the low-passed eddy heating should be balanced by the trends in [EIA]. The low-passed MSE fluxes include both the contribution from low-pass eddies and from synoptic eddies (referred to as bandpassed here). By using the low-passed MSE flux convergence we are leaving out the contribution of these fluxes to changes on time scales less than 30 days.
Acknowledgments.
We thank three anonymous reviewers for their thorough comments that helped to improve this manuscript. We thank Dr. Q. Ma for help in processing the diabatic heating rate data and Drs. F. Lunkeit and V. Lembo for helpful discussions. CF was supported by the Institute for Basic Science (IBS), South Korea, under IBS-R028-D1. NH was supported by the Israeli Science Foundation Grant 1685/17.
Data availability statement.
The JRA-55 data are available at https://rda.ucar.edu/datasets/ds628.0/.
APPENDIX A
Changepoint Analysis
We perform a changepoint analysis using a binary segmentation algorithm of the R package changepoint (Killick and Eckley 2014; Scott and Knott 1974).
We display the results of the changepoint analysis here exemplarily for zonal-mean temperature, specific humidity, and MSE. As can be seen in Fig. A1, almost all identified changepoints occur away from winter 1979 (or months 66–68). For specific humidity no changepoints have been identified. The summer season and other variables show similar behavior.
Results of changepoint analysis for (a),(b) temperature, (c),(d) specific humidity, and (e),(f) MSE for DJF. Colors indicate number of months since 1958. Winter of 1979 would correspond to months 66–68. (left) Month of identified changepoint and (right) sample time series (black line) and changepoint line (red). The times series are from the lower-tropospheric Arctic (87°N, 1000 hPa).
Citation: Journal of Climate 36, 9; 10.1175/JCLI-D-21-0724.1
APPENDIX B
Meridional Flux Divergence Terms
The respective terms of Eq. (4) describe the following mechanisms: χ1 the linear meridional transport of zonal-mean low-frequency static energy by the time-mean zonal-mean flow, χ2 the linear meridional transport of zonally anomalous low-frequency static energy by the zonally anomalous time-mean flow, χ3 the linear meridional transport of zonal-mean time-mean static energy by the zonal-mean low-frequency flow, χ4 the linear meridional transport of zonally anomalous time-mean static energy by the zonally anomalous low-frequency flow, χ5 the low-frequency nonlinear transport of high-frequency zonal-mean static energy by the high-frequency zonal-mean flow, χ6 the low-frequency nonlinear transport of low-frequency zonal-mean static energy by the low-frequency zonal-mean flow, χ7 the low-frequency nonlinear transport of low-frequency zonal-mean static energy by the high-frequency zonal-mean flow, χ8 the low-frequency nonlinear transport of high-frequency zonal-mean static energy by the low-frequency zonal-mean flow, χ9 the low-frequency nonlinear transport of high-frequency zonally anomalous static energy by the high-frequency zonally anomalous flow, χ10 the low-frequency nonlinear transport of low-frequency zonally anomalous static energy by the low-frequency zonally anomalous flow, χ11 the low-frequency nonlinear transport of low-frequency zonally anomalous static energy by the high-frequency zonally anomalous flow, and χ12 the low-frequency nonlinear transport of high-frequency zonally anomalous static energy by the low-frequency zonally anomalous flow.
REFERENCES
Alekseev, G., S. Kuzmina, L. Bobylev, A. Urazgildeeva, and N. Gnatiuk, 2019: Impact of atmospheric heat and moisture transport on the Arctic warming. Int. J. Climatol., 39, 3582–3592, https://doi.org/10.1002/joc.6040.
Allen, R. J., and S. C. Sherwood, 2008: Warming maximum in the tropical upper troposphere deduced from thermal winds. Nat. Geosci., 1, 399–403, https://doi.org/10.1038/ngeo208.
Ambaum, M. H., 2020: Thermal Physics of the Atmosphere. 2nd ed. Elsevier, 268 pp.
Archer, C. L., and K. Caldeira, 2008: Historical trends in the jet streams. Geophys. Res. Lett., 35, L08803, https://doi.org/10.1029/2008GL033614.
Armour, K. C., J. Marshall, J. R. Scott, A. Donohoe, and E. R. Newsom, 2016: Southern Ocean warming delayed by circumpolar upwelling and equatorward transport. Nat. Geosci., 9, 549–554, https://doi.org/10.1038/ngeo2731.
Back, L. E., and C. S. Bretherton, 2006: Geographic variability in the export of moist static energy and vertical motion profiles in the tropical Pacific. Geophys. Res. Lett., 33, L17810, https://doi.org/10.1029/2006GL026672.
Barnes, E. A., and L. Polvani, 2013: Response of the midlatitude jets, and of their variability, to increased greenhouse gases in the CMIP5 models. J. Climate, 26, 7117–7135, https://doi.org/10.1175/JCLI-D-12-00536.1.
Barpanda, P., and T. Shaw, 2017: Using the moist static energy budget to understand storm-track shifts across a range of time scales. J. Atmos. Sci., 74, 2427–2446, https://doi.org/10.1175/JAS-D-17-0022.1.
Bender, F. A.-M., V. Ramanathan, and G. Tselioudis, 2012: Changes in extratropical storm track cloudiness 1983–2008: Observational support for a poleward shift. Climate Dyn., 38, 2037–2053, https://doi.org/10.1007/s00382-011-1065-6.
Bindoff, N. L., and Coauthors, 2013: Detection and attribution of climate change: From global to regional. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 867–952.
Bischoff, T., and T. Schneider, 2014: Energetic constraints on the position of the intertropical convergence zone. J. Climate, 27, 4937–4951, https://doi.org/10.1175/JCLI-D-13-00650.1.
Blackport, R., and J. A. Screen, 2020: Weakened evidence for mid-latitude impacts of Arctic warming. Nat. Climate Change, 10, 1065–1066, https://doi.org/10.1038/s41558-020-00954-y.
Bonan, D. B., N. Siler, G. H. Roe, and K. C. Armour, 2023: Energetic constraints on the pattern of changes to the hydrological cycle under global warming. J. Climate, https://doi.org/10.1175/JCLI-D-22-0337.1, in press.
Butler, A. H., D. W. J. Thompson, and R. Heikes, 2010: The steady-state atmospheric circulation response to climate change–like thermal forcings in a simple general circulation model. J. Climate, 23, 3474–3496, https://doi.org/10.1175/2010JCLI3228.1.
Cai, M., and M. Mak, 1990: Symbiotic relation between planetary and synoptic-scale waves. J. Atmos. Sci., 47, 2953–2968, https://doi.org/10.1175/1520-0469(1990)047<2953:SRBPAS>2.0.CO;2.
Cassou, C., 2008: Intraseasonal interaction between the Madden–Julian oscillation and the North Atlantic Oscillation. Nature, 455, 523–527, https://doi.org/10.1038/nature07286.
Ceppi, P., and D. L. Hartmann, 2015: Connections between clouds, radiation, and midlatitude dynamics: A review. Curr. Climate Change Rep., 1, 94–102, https://doi.org/10.1007/s40641-015-0010-x.
Ceppi, P., F. Brient, M. D. Zelinka, and D. L. Hartmann, 2017: Cloud feedback mechanisms and their representation in global climate models. Wiley Interdiscip. Rev.: Climate Change, 8, e465, https://doi.org/10.1002/wcc.465.
Chang, E. K. M., and A. M. W. Yau, 2016: Northern Hemisphere winter storm track trends since 1959 derived from multiple reanalysis datasets. Climate Dyn., 47, 1435–1454, https://doi.org/10.1007/s00382-015-2911-8.
Chang, E. K. M., C.-G. Ma, C. Zheng, and A. M. W. Yau, 2016: Observed and projected decrease in Northern Hemisphere extratropical cyclone activity in summer and its impacts on maximum temperature. Geophys. Res. Lett., 43, 2200–2208, https://doi.org/10.1002/2016GL068172.
Chemke, R., and L. M. Polvani, 2019: Opposite tropical circulation trends in climate models and in reanalyses. Nat. Geosci., 12, 528–532, https://doi.org/10.1038/s41561-019-0383-x.
Chemke, R., and Y. Ming, 2020: Large atmospheric waves will get stronger, while small waves will get weaker by the end of the 21st century. Geophys. Res. Lett., 47, e2020GL090441, https://doi.org/10.1029/2020GL090441.
Chemke, R., and L. M. Polvani, 2020: Linking midlatitudes eddy heat flux trends and polar amplification. Climate Atmos. Sci., 3, 8, https://doi.org/10.1038/s41612-020-0111-7.
Chemke, R., Y. Ming, and J. Yuval, 2022: The intensification of winter mid-latitude storm tracks in the Southern Hemisphere. Nat. Climate Change, 12, 553–557, https://doi.org/10.1038/s41558-022-01368-8.
Clark, J. P., S. B. Feldstein, and S. Lee, 2022: Moist static energy transport trends in four global reanalyses: Are they downgradient? Geophys. Res. Lett., 49, e2022GL098822, https://doi.org/10.1029/2022GL098822.
Collins, M., and Coauthors, 2013: Long-term climate change: Projections, commitments and irreversibility. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 1029–1136.
Coumou, D., J. Lehmann, and J. Beckmann, 2015: The weakening summer circulation in the Northern Hemisphere mid-latitudes. Science, 348, 324–327, https://doi.org/10.1126/science.1261768.
Coumou, D., G. Di Capua, S. Vavrus, L. Wang, and S. Wang, 2018: The influence of Arctic amplification on mid-latitude summer circulation. Nat. Commun., 9, 2959, https://doi.org/10.1038/s41467-018-05256-8.
England, M. R., I. Eisenman, N. J. Lutsko, and T. J. W. Wagner, 2021: The recent emergence of Arctic amplification. Geophys. Res. Lett., 48, e2021GL094086, https://doi.org/10.1029/2021GL094086.
Feldstein, S. B., 2002: Fundamental mechanisms of the growth and decay of the PNA teleconnection pattern. Quart. J. Roy. Meteor. Soc., 128, 775–796, https://doi.org/10.1256/0035900021643683.
Feldstein, S. B., 2003: The dynamics of NAO teleconnection pattern growth and decay. Quart. J. Roy. Meteor. Soc., 129, 901–924, https://doi.org/10.1256/qj.02.76.
Feldstein, S. B., and C. L. E. Franzke, 2017: Atmospheric teleconnection patterns. Nonlinear and Stochastic Climate Dynamics, C. L. E. Franzke and T. J. O’Kane, Eds., Cambridge University Press, 54–104.
Feser, F., M. Barcikowska, O. Krueger, F. Schenk, R. Weisse, and L. Xia, 2015: Storminess over the North Atlantic and northwestern Europe—A review. Quart. J. Roy. Meteor. Soc., 141, 350–382, https://doi.org/10.1002/qj.2364.
Francis, J. A., and E. Hunter, 2006: New insight into the disappearing Arctic Sea ice. Eos, Trans. Amer. Geophys. Union, 87, 509–511, https://doi.org/10.1029/2006EO460001.
Franzke, C. L. E., 2012: On the statistical significance of surface air temperature trends in the Eurasian Arctic region. Geophys. Res. Lett., 39, L23705, https://doi.org/10.1029/2012GL054244.
Franzke, C. L. E., and S. B. Feldstein, 2005: The continuum and dynamics of Northern Hemisphere teleconnection patterns. J. Atmos. Sci., 62, 3250–3267, https://doi.org/10.1175/JAS3536.1.
Franzke, C. L. E., and T. Woollings, 2011: On the persistence and predictability properties of North Atlantic climate variability. J. Climate, 24, 466–472, https://doi.org/10.1175/2010JCLI3739.1.
Franzke, C. L. E., T. Woollings, and O. Martius, 2011: Persistent circulation regimes and preferred regime transitions in the North Atlantic. J. Atmos. Sci., 68, 2809–2825, https://doi.org/10.1175/JAS-D-11-046.1.
Franzke, C. L. E., T. J. O’Kane, D. P. Monselesan, J. S. Risbey, and I. Horenko, 2015: Systematic attribution of observed Southern Hemisphere circulation trends to external forcing and internal variability. Nonlinear Processes Geophys., 22, 513–525, https://doi.org/10.5194/npg-22-513-2015.
Franzke, C. L. E., S. Lee, and S. B. Feldstein, 2017: Evaluating Arctic warming mechanisms in CMIP5 models. Climate Dyn., 48, 3247–3260, https://doi.org/10.1007/s00382-016-3262-9.
Gimeno-Sotelo, L., R. Nieto, M. Vázquez, and L. Gimeno, 2018: A new pattern of the moisture transport for precipitation related to the drastic decline in Arctic Sea ice extent. Earth Syst. Dyn., 9, 611–625, https://doi.org/10.5194/esd-9-611-2018.
Gong, T., S. Feldstein, and S. Lee, 2017: The role of downward infrared radiation in the recent Arctic winter warming trend. J. Climate, 30, 4937–4949, https://doi.org/10.1175/JCLI-D-16-0180.1.
Hamed, K. H., and A. R. Rao, 1998: A modified Mann-Kendall trend test for autocorrelated data. J. Hydrol., 204, 182–196, https://doi.org/10.1016/S0022-1694(97)00125-X.
Hannachi, A., D. M. Straus, C. L. E. Franzke, S. Corti, and T. Woollings, 2017: Low-frequency nonlinearity and regime behavior in the Northern Hemisphere extratropical atmosphere. Rev. Geophys., 55, 199–234, https://doi.org/10.1002/2015RG000509.
Hao, M., Y. Lin, Y. Luo, R. Nath, and Z. Zhao, 2021: The impact of atmospheric moisture transport on winter Arctic warming: Radiation versus latent heat release. Int. J. Climatol., 41, 3982–3993, https://doi.org/10.1002/joc.7054.
Hartmann, D. L., and P. Ceppi, 2014: Trends in the CERES dataset, 2000–13: The effects of sea ice and jet shifts and comparison to climate models. J. Climate, 27, 2444–2456, https://doi.org/10.1175/JCLI-D-13-00411.1.
Held, I. M., and B. J. Soden, 2000: Water vapor feedback and global warming. Annu. Rev. Energy Environ., 25, 441–475, https://doi.org/10.1146/annurev.energy.25.1.441.
Held, I. M., and B. J. Soden, 2006: Robust responses of the hydrological cycle to global warming. J. Climate, 19, 5686–5699, https://doi.org/10.1175/JCLI3990.1.
Holton, J. R., and G. L. Hakim, 2013: An Introdu