2.1 The LMDz–REPROBUS climate–chemistry model

Simulations are performed with the stratospheric chemistry–climate model developed in the framework of the IPSL Earth system model (IPSL-CM, climate model) development (Dufresne et al., 2013). The stratospheric chemistry is computed with the REPROBUS chemical model (Lefèvre et al., 1994, 1998; Jourdain et al., 2008) coupled with the LMDz atmospheric general circulation model (Hourdin et al., 2013). REPROBUS describes the chemistry of stratospheric source gases such as N₂O, CH₄, CH₃Cl and CH₃Br and the associated radical chemistry of hydrogen, nitrogen oxides, chlorine and bromine species. It computes the global distribution of trace gases, aerosols and clouds within the stratosphere considering gas-phase and heterogeneous chemistry. The heterogeneous chemistry component takes into account the reactions on sulfuric acid aerosols and liquid (ternary solution) and solid (nitric acid trihydrate particles, ice) polar stratospheric clouds (PSCs). The gravitational sedimentation of PSCs is simulated as well. The LMDz–REPROBUS chemistry–climate model allows an interactive coupling of ozone, shortwave heating rates and dynamics as recommended in Sassi et al. (2005). The resolution of the model is 3.75^∘ in longitude and 1.9^∘ in latitude, and it has 39 vertical levels, with around 15 levels above 20 km and around 24 above 10 km and a lid height at ∼70 km.

LMDz and LMDz–REPROBUS have been involved in a large range of studies, model intercomparisons and evaluations, notably through the participation of the LMDz model in the international Coupled Model Intercomparison Project (CMIP, phases 3, 5 and currently 6) and the participation of LMDz–REPROBUS in the Chemistry Climate Model Validation (SPARC CCMVal, 2010) and the Chemistry Climate Model Initiative (CCMI; Morgenstern et al., 2017). Results presented in the recent studies of, e.g., de la Cámara et al. (2016a, b), Thiéblemont et al. (2017) and Ayarzagüena et al. (2018) have shown that the stratospheric chemistry, dynamics and transport simulated by the LMDz model and its chemistry–climate model version are consistent with satellite observations, reanalysis and other models of the same kind.

2.2 Simulation setup

The setup of the four simulations performed in this study is summarized in Table 1. All the simulations consist of 30-year time slices, starting from atmospheric physical conditions and surface temperatures taken from very long coupled atmosphere–ocean simulations. For the analysis of our chemistry–climate simulations, a 5-year spin-up is considered. For all the simulations, the solar constant is set to 1366 W m⁻² and orbital parameters (obliquity, precession and eccentricity) are set to modern values as recommended in the DeepMIP protocol (model intercomparison project; Lunt et al., 2017). Oxygen variations are poorly constrained over pre-Quaternary timescales, and there is no consensus on the oxygen variations through the Cenozoic (see Fig. 1 of Wade et al., 2018). In view of these uncertainties, we use a present-day oxygen content to investigate Cenozoic past climates, as commonly done in climate models.

Table 1Setup of LMDz. AMIP is the Atmospheric Model Intercomparison Project.

Note: for the Eocene simulations, the REPROBUS chemical model considers a CH₄ concentration of 3614 ppb and an N₂O concentration of 323 ppb.

Download Print Version | Download XLSX

Figure 1Modeling setup for the Eocene simulations.

Download

3.1 Stratospheric ozone in Eocene conditions

We first investigate the impact of Eocene conditions on stratospheric ozone with respect to preindustrial conditions. Figure 2a and b show the latitude vs. pressure zonally averaged cross sections of temperature and ozone anomalies associated with the Eocene conditions. As expected, the CO₂ increase leads to a global radiative cooling of the stratosphere with decreases in temperatures exceeding 12 K above 10 hPa (∼32 km) and a warming of the troposphere (Fig. 2a). In the troposphere, we further notice a more pronounced Antarctic amplification of the temperature signals, which contrasts with present-day climate conditions where a more pronounced Arctic amplification of the global warming is observed (IPCC, 2013). This signal, consistently simulated by several models (Lunt et al., 2012), is linked to the absence of the Antarctic ice sheet in Eocene boundary conditions, which leads to lower surface topography and albedo. The cooling of the stratosphere slows down the ozone destruction, resulting in an increase in stratospheric ozone concentrations (Haigh and Pyle, 1982). This is consistent with the statistically significant positive ozone anomalies found above 50 hPa (∼20 km) over the polar regions and above 10 hPa in the tropical band (Fig. 2b). Note that this effect increases with altitude in the stratosphere as the photochemical control on ozone levels becomes prominent (Brasseur and Solomon, 2005). Similarly to the results of Chiodo et al. (2018), which investigate the effect of quadrupling CO₂ by starting from preindustrial climate, a maximum ozone increase of ∼40 % is found at about 2–3 hPa (∼40 km). Note that in our simulation, the stratospheric chemistry is also modified by the increase in N₂O and CH₄. However, their effect only reaches a maximum of 3 % in the equatorial upper stratosphere (∼5 hPa) (see Fig. S1 in the Supplement). Although this chemical effect on ozone is statistically significant, its impact appears to be small compared to the upper stratosphere 40 % increase in ozone due to increasing CO₂.

Figure 2Annual mean differences (EOCENE minus PREIND) of zonally averaged temperature (in K, a), ozone (in %, b) and age of air (in years, c). Color-filled contours in (a) and (b) indicate that anomalies are statistically different at the 1 % confidence level according to a t test. Black contours show the preindustrial climatology expressed in K (a), ppm (b) and years (c).

Download

In contrast, the lower tropical stratosphere (30^∘ S–30^∘ N, 100–30 hPa) exhibits a statistically significant ozone decrease of up to 40 %. In this region, the ozone concentration is mostly controlled by transport processes (Brasseur and Solomon, 2005), especially the strength of the Brewer–Dobson circulation ascending branch. Figure 2c shows the age of air (AoA) calculated after 20 years of simulations by taking as a reference entry point the equatorial lowermost stratosphere, slightly above the tropopause (i.e., pressure level corresponding to 74 hPa). Globally, the stratospheric AoA is younger in the Eocene experiment than in the preindustrial one, revealing an acceleration of the Brewer–Dobson circulation under Eocene conditions. This, in turn, is consistent with the reduced ozone concentration in the lower tropical stratosphere. Note also that the tropopause height is globally lifted up in the Eocene experiment (not shown). The rise of the tropopause is a robust feature of warmer climate conditions (Sausen and Santer, 2003) and contributes to the negative ozone anomaly found in the vicinity of the tropopause region (Fig. 2b) (Dietmüller et al., 2014).

Next, we examine anomalies of the annual total column ozone (TCO). Figure 3 shows the comparison of the latitudinal distribution of the annual TCO for Eocene and preindustrial conditions. In both simulations (Fig. 3a), the TCO shows a minimum in the tropical region (20^∘ S, 20^∘ N) of ∼270 Dobson units (DU) and maxima near 55^∘ N and 55^∘ S, followed by poleward decreases that are more pronounced in the Southern Hemisphere. The differences between Eocene and preindustrial conditions (Fig. 3b) reveal no changes in the tropical band 20^∘ S–20^∘ N but statistically significant positive anomalies at midlatitudes and in polar regions. The midlatitude maxima reach ∼390 DU for preindustrial conditions, whereas they exceed 430 DU for Eocene conditions. This latitudinal distribution of TCO anomalies is overall consistent with projections of TCO anomalies simulated in response to the 21st-century climate change and post-CFC (chlorofluorocarbon) era (Li et al., 2009) or to an abrupt 4×CO₂ increase from preindustrial climate conditions (Chiodo et al., 2018). The detailed comparison of our results with those of Chiodo et al. (2018) shows, however, noticeable differences at high latitudes. In our simulations, the TCO anomalies peak at 45^∘ S and 50^∘ N with maximum differences of 50 and 60 DU, respectively; the anomalies decrease from these maxima to about 30 DU at high latitudes (Fig. 3b). In Chiodo et al. (2018), hints of such a decrease were found in the Southern Hemisphere for only two out of the four models that are intercompared, and no such decrease was seen in the Northern Hemisphere. The negative poleward TCO gradient at high latitudes appears to strengthen markedly under Eocene conditions in the Northern Hemisphere (Fig. 3b). The seasonal dependence of the TCO high-latitude poleward gradient for the Eocene and preindustrial conditions in the Northern Hemisphere is explored in Fig. 4. Figure 4 reveals that, under Eocene conditions, the negative gradient is particularly pronounced during the winter season (from October to March), when the stratospheric polar vortex dominates the high-latitude circulation in the Northern Hemisphere. Hence, this indicates substantial changes in the stratospheric circulation associated with the Eocene conditions that we examine further in Sect. 3.2.

Figure 3Latitudinal profile of the total column ozone (in Dobson units or DU, a) in the (red) EOCENE and (black) PREIND simulation. Total column ozone change (in DU, b) between the EOCENE and PREIND simulation. Dashed lines delimit the 2σ uncertainty envelop, which is represented by the standard error of the mean.

Download

Figure 4Zonally averaged seasonal evolution of the latitudinal gradient (computed as the difference between 84 and 57^∘ N) of the total column ozone for the EOCENE (red) and PREIND (black) simulations. Dashed lines delimit the 2σ uncertainty envelope, which is represented by the standard error of the mean.

Download

3.2 Seasonal evolution of the Northern Hemisphere stratospheric polar vortex in Eocene conditions

An overview of the annual average background zonal circulation in preindustrial conditions and its anomalies associated with Eocene conditions is shown in Fig. 5. In Eocene conditions, the high-latitude stratospheric westerlies maxima, indicative of the average location of the core of the stratospheric Southern and Northern Hemisphere polar night jets (near 60^∘ S and 60^∘ N), appear to be overall stronger and also shifted equatorward in comparison with the preindustrial climatology (black contour). These results hence suggest a strengthening and an extension of stratospheric polar vortices, which develop during winter in each Hemisphere. Note also that the upward extension of the subtropical upper-tropospheric jets in both hemispheres (centered near 35^∘ N/S around 200 hPa) are consistent with the tropopause rising associated with Eocene conditions.

Figure 5Annual mean differences (EOCENE minus PREIND) of zonally averaged zonal wind (in m s⁻¹). Color-filled contours indicate anomalies that are statistically different at the 1 % confidence level according to a t test. Black contours show the preindustrial climatology.

Download

To identify the processes leading to the strengthening of the stratospheric polar vortex under Eocene conditions, we explore the stratospheric dynamical wintertime evolution in the Northern Hemisphere, where the largest changes are found in our simulations (e.g., Fig. 5). Figure 6 shows the monthly evolution of the zonal-mean zonal wind from October to March in the Northern Hemisphere. Regardless of simulated climate conditions, the winter season in the stratosphere is characterized by the development of a mid-to-high-latitude strong westerly jet (or polar night jet – the center of which roughly corresponding to the edge of the polar vortex), which maximizes in midwinter (December–January). In early and midwinter (Fig. 6a–e), the polar night jet in Eocene conditions appears, however, to be twice as strong as in preindustrial conditions as shown, e.g., in January (Fig. 6d) where the maximum anomaly near 60^∘ N and 5 hPa (∼40 m s⁻¹) associated with Eocene conditions is larger than the preindustrial climatology (∼30 m s⁻¹). Such a strong boreal polar night jet was also found in Eocene simulations of Baatsen et al. (2018). In late winter (Fig. 6f), the differences in polar night jet strength between the two experiments are no longer statistically significant in the middle stratosphere and appear to even be reversed in the upper stratosphere; i.e., an easterly anomaly is found near the stratopause at midlatitudes. This indicates a very fast decay of the polar vortex in Eocene conditions in late winter (see also Fig. S1 in the Supplement). These differences in the seasonal evolution of the zonal-mean zonal wind are consistent with the seasonal evolution of the ozone gradient shown in Fig. 4. Indeed, under Eocene conditions, the stronger winter polar vortex is associated with a reinforcement of the mixing barrier at its edge. This leads in turn to a reduction in air exchanges between mid- and polar latitudes and hence to a steepening of the poleward ozone gradient. Similar (though less pronounced) differences between Eocene and preindustrial conditions are found in the Southern Hemisphere (not shown).

Figure 6Monthly evolution (October to March) of the zonal-mean zonal wind differences (m s⁻¹) between the Eocene and preindustrial conditions in the Northern Hemisphere. Dotted regions indicate that anomalies are insignificantly different at the 5 % confidence level according to a t test. Black isolines shows the climatology derived from the preindustrial experiment.

Download

At first glance, the strengthening of the polar vortex under Eocene conditions may seem inconsistent with the global acceleration of the Brewer–Dobson circulation as diagnosed by the younger stratospheric age of air (Fig. 2c). Indeed, a faster Brewer–Dobson circulation is associated with a stronger planetary wave drag in the stratosphere (i.e., an enhanced wave breaking), which in turn should lead to a weaker polar vortex. In the following, we hence investigate the seasonality of the planetary stationary wave activity and its interaction with the mean flow by calculating the Eliassen–Palm flux (hereafter EP flux) divergence, here scaled to units of zonal acceleration (Andrews et al., 1987):

where F is the EP flux whose components are

f is the Coriolis parameter, a is the Earth's radius, θ is the potential temperature, ρ₀ is the density profile of the atmosphere and $\left(u,v,w\right)$ are the three-dimensional velocity components in spherical coordinates $\left(\mathit{\lambda },\mathit{\varphi },z\right)$, where z is the log pressure. Overbars indicate zonal mean and primes denote departure from the zonal mean. As shown by Edmon et al. (1980), the EP flux constitutes a measure of the Rossby wave propagation from one height (z) and latitude (ϕ) to another, and its divergence (divEP) gives information about the forcing of the mean flow by the eddies.

Figure 7 displays the monthly evolution in winter of the EP flux and its divergence for the preindustrial conditions experiment. This analysis shows that, throughout winter, the wave activity penetrates the stratosphere (as indicated by the vectors) near 55^∘ N, propagates upward and tends to be increasingly refracted toward the Equator with height. The dissipation of planetary waves exerts a westward-momentum forcing on the mean flow between 30 and 70^∘ N (as diagnosed by the Eliassen–Palm flux convergence), which maximizes along the equatorward flank of the polar night jet where planetary wave breaking is large. This contributes to eroding and weakening the polar vortex and to a warming of the polar stratosphere and drives a persistent poleward mass transport in order to conserve the angular momentum. By mass continuity, this induces an upward transport at low latitudes and an extratropical downwelling (hence driving the Brewer–Dobson circulation). Under preindustrial climate conditions, we note that the wave activity and its interaction with the mean flow peaks in December or January (Fig. 7c, d) but is already large in November (Fig. 7b). Therefore, this contributes to slow down the radiatively driven development of the polar vortex in early winter.

Figure 7Monthly evolution (October to March) of the Eliassen–Palm flux (vectors) and its divergence (contours, in m s ⁻¹ d⁻¹) under preindustrial conditions in the Northern Hemisphere.

Download

As shown in Fig. 8, under Eocene conditions, it appears that the planetary wave activity penetrating the stratosphere in early winter (i.e., November–December; Fig. 8b, c) is significantly reduced and deflected equatorward as revealed by the downward and equatorward pointing of the EP flux vector in the lower midlatitude stratosphere. This is associated with an anomalous positive EP flux divergence (i.e., a reduced convergence) throughout the depth of the stratospheric polar night jet (near 60^∘ N), which indicates a substantially reduced westward-momentum forcing by planetary waves and hence allows a stronger development of the polar vortex in early winter in comparison with preindustrial conditions. In contrast, from January (Fig. 8d), the planetary wave activity becomes significantly larger under Eocene conditions, the westward forcing appears to be strongly amplified in the upper stratosphere and to progressively propagate downward in February (Fig. 8e). This is consistent with the reversal of the zonal-mean zonal wind anomaly in the upper stratosphere, but also with the overall extremely rapid deceleration of the polar-vortex strength noted previously (see Figs. 6 and S2). In addition, we analyzed the residual mass circulation (not shown) derived from the transformed Eulerian-mean formalism (Andrews et al., 1987). While no clear changes in the strength of the Brewer–Dobson circulation are found in early winter between Eocene and preindustrial conditions, late winter (February–March) reveals an important acceleration in Eocene conditions, which is consistent with the much stronger wave forcing found throughout the extratropical stratosphere (Fig. 8e, f). These results are consistent with a net acceleration of the Brewer–Dobson under Eocene conditions in comparison with preindustrial conditions as revealed by the younger age of air (Fig. 2c). Note that the Brewer–Dobson acceleration appears to be more pronounced in the Northern Hemisphere, where the mean flow and wave activity anomalies are found to be stronger than in the Southern Hemisphere (not shown).

Figure 8Monthly evolution (October to March) of the differences between the Eocene and preindustrial conditions of the Eliassen–Palm flux and its divergence. Dotted regions indicate that anomalies are insignificantly different at the 5 % confidence level according to a t test. Preindustrial climatology is shown with dashed contours.

Download

Although the large changes in the background state stratospheric circulation and its seasonal evolution under Eocene conditions in comparison with preindustrial climate conditions appear to be largely wave-driven, the origin of the identified changes in the planetary wave activity and its interaction with the mean flow remains to be determined. Note that this does not uniquely depend on changes in tropospheric wave sources but also on changes in the background flow itself, which modulates the wave propagation and the nature of wave-mean flow interactions. The planetary wave activity entering the stratosphere can be altered by numerous factors such as sea surface temperature changes (e.g., Hu et al., 2014), sea-ice changes (e.g., Kim et al., 2014), wind changes near the tropopause (e.g., Shepherd and McLandress, 2011; Karpechko and Manzini, 2017) or topography (Shi et al., 2014). Additional simulations of the Eocene and the preindustrial period with the atmospheric model (LMDz) without interactive chemistry and with a flat topography reveal that changes in the topography have first-order effects on planetary wave activity and hence on the stratospheric dynamics (not shown). Between the Eocene and the preindustrial conditions, beside large changes in the topography, important changes in air–sea thermal contrasts, sea-ice cover and sea surface temperature could all have a substantial influence on stratospheric circulation. The complexness of these effects and their possible interactions make an unambiguous attribution impossible in the absence of a dedicated experimental protocol that is out of the scope for the present study.

5.1 Impact on climate

The consideration of a stratospheric ozone compatible with the Eocene conditions perturbs the radiative balance compared to the use of a preindustrial climatology. The net radiative change (shortwave + longwave) between the simulation with interactive chemistry and the simulation with preindustrial climatology corresponds to a 1.7 W m⁻² effective radiative forcing (RF). This radiative forcing results from combined positive RFs in the longwave (LW) and shortwave (SW) simulated in the tropics. Beyond 50^∘ (north and south), the positive SW RF is partly counterbalanced by a negative longwave RF (see Table 3 and Fig. 12). This radiative forcing from the stratospheric ozone response represents a positive climate feedback, which is commonly ignored in Eocene climate simulations. In order to estimate the potential impact of an interactive ozone on surface temperature under Eocene conditions, we consider a large range of climate sensitivity from 0.4 to 1.2 K (W m⁻²)⁻¹ (Knutti et al., 2017). Given such a broad range, the surface temperature response to this stratospheric forcing could range from 0.7 to 2.0 K (assuming an effective radiative forcing of 1.7 W m⁻²). The surface temperature response to a specific radiative forcing depends on the considered climate conditions and on the nature of the climate forcer. Considering an interactive stratospheric ozone chemistry under a 4×CO₂ climate perturbation, Nowack et al. (2015) found a climate sensitivity of 1.05 K (W m⁻²)⁻¹ in their ESM. Applying this climate sensitivity, the surface temperature change associated with the ozone feedback would be about 1.8 K when considering interactive stratospheric chemistry (compared with the EOCENE_Oz1855 run). The climate sensitivity to ozone change can obviously vary from one ESM to another since, for example, the sensitivity of the Brewer–Dobson circulation to climate is highly model-dependant (SPARC CCMVal, 2010). Nonetheless, this estimation allows us to discuss the importance of considering this chemistry–climate feedback when attempting to simulate greenhouse paleoclimates. According to the IPCC AR5 report (IPCC, 2013), the global land surface air temperature anomaly is +12.7 K for the early Eocene climatic optimum (Masson-Delmotte et al., 2013). This estimation is based on simulations from several models analyzed by Lunt et al. (2012), for which there was no common modeling protocol (e.g., CO₂ being in the 2×[CO₂]_PI to 16×[CO₂]_PI range). One of these models, the HadCM model, estimates that the effect of changing non-CO₂ boundary conditions (topography, bathymetry, solar constant and vegetation) for Eocene conditions leads to a 1.8 K increase in the global mean surface air temperature (to be compared to a 3.3 K increase when doubling the CO₂). The feedback of stratospheric ozone on surface air temperature could thus represent about 15 % of the total temperature anomaly reported between Eocene and preindustrial conditions and be as important as the effect of external forcings.

Table 3Differences in shortwave and longwave radiative fluxes at different vertical levels between the EOCENE simulation and the climate-only simulation with the 1855 ozone climatology.

Download Print Version | Download XLSX

Figure 12Differences in radiative fluxes as a function of latitude between the EOCENE simulation and the climate-only simulation with the 1855 ozone climatology.

Download

In strong greenhouse climate, the terrestrial carbon and nitrogen cycles are intensified, releasing high CH₄ and N₂O in the atmosphere (Beerling et al., 2011). The effect of changing N₂O and CH₄ in the troposphere (including the H₂O increase in the stratosphere induced by the CH₄ increase) has been assessed by Beerling et al. (2011). These authors find, with the STOCHEM model, a 2.1 K increase in global surface temperature due solely to the tropospheric composition changes. Our results suggest that the effect of the stratospheric ozone feedback on surface temperatures is of similar importance.

5.2 Impact on tropospheric conditions

Using an ESM including chemistry, Unger and Yue (2014) found that under warm and high methane Pliocene conditions, the stratospheric ozone burden was 5 % higher than the preindustrial one. This stratospheric ozone increase resulted in a 20 % reduction in the tropospheric photolysis rate of ozone (${\mathrm{O}}_{\mathrm{3}}+\mathit{\text{hv}}-\mathrm{>}{\mathrm{O}}^{\mathrm{1}}\mathrm{D}+{\mathrm{O}}_{\mathrm{2}}$) that leads to the formation of OH, the hydroxyl radical. This radical is the main oxidant of the troposphere and its decrease (of 20 % to 25 % in the Unger and Yue simulations) impacts the lifetime of chemical species and in particular CH₄. Our simulations show a 7.2 % increase in the stratospheric ozone burden when comparing the EOCENE and the PREIND simulation (both including interactive chemistry) and an 8.8 % difference when comparing the EOCENE simulation to the EOCENE_Oz1855.

In addition, we estimate the change in surface UV radiations and ozone photolysis in the Eocene conditions. Using the radiation transfer model Quick TUV Calculator with a pseudo-spherical discrete ordinate four streams (http://cprm.acom.ucar.edu/Models/TUV/Interactive_TUV/, last access: 2 June 2019), we estimate the effect of the stratospheric ozone increase on the ozone photolysis, which controls the OH production. Considering a 65 DU change at a 50^∘ latitude (corresponding to the maximum of Fig. 10), the photolysis rate at the surface decreases by 25 %. A decrease in the ozone photolysis rate would induce a significant decrease in OH and hence in the tropospheric oxidizing capacity, thus making CH₄ longer lived and reinforcing its effect on climate. It would also impact the overall tropospheric chemistry.