To fully isolate and understand the statistical properties of storm tracks, it is necessary to systematically add or subtract key processes to a nonlinear moist GCM , due to the pronounced role of latent heat release in storm track structure . In this study, we use the Model of an idealized Moist Atmosphere (MiMA), developed by , , and . Building on the moist aquaplanet models of and , introduced more realistic surface forcing to better represent a range of physical processes. The most important of these physical processes for the current paper is that MiMA also incorporates a physically consistent representation of moisture transport and latent heat release, within a parameterized convection scheme and resolved-scale evaporation, transport, and condensation scheme . Additionally, it features an idealized boundary layer scheme based on Monin–Obukhov similarity theory and a slab ocean. Futhermore, MiMA includes a full radiative transfer scheme – the GCM version of the Rapid Radiative Transfer Model (RRTM) . For further details on the model configuration, please see and . The net effect is an atmospheric general circulation model (GCM) of intermediate complexity, bridging the gap between more idealized dry GCMs and fully comprehensive models of the real atmosphere.

The first version of MiMA constructed by did not have realistic stationary waves, however stationary waves are known to affect both the climate and weather in the troposphere , including the extratropical storm tracks. and filled this gap and implemented a series of differences in the specification of the lower boundary to allow for a stationary wave pattern that closely resembles that in CMIP models, yet still retaining full flexibility to run in aquaplanet mode. These changes included (1) Earth's topography, (2) realistic horizontal gradients in ocean heat flux (which implicitly represents ocean heat transport that cannot be resolved by the slab ocean), and (3) land–sea contrast . The land–sea contrast includes different heat capacities, moisture availability, and roughness lengths, for land vs. ocean ; note that there is no representation of additional land processes such as land-use. We began our work with the configuration of , and have since implemented for this paper two main changes to their MiMA configuration. These changes are motivated by the fact that the storm tracks in the model configuration of and are weaker than ERA5 by 17 % (see Sect. ). The most important of these changes is to the albedo profile. The idealized albedo profile used in (red curve in Fig. e) does not match that observed in CERES (black curve in Fig. e).

We therefore use a revised albedo profile in this paper (magenta curve in Fig. ) that better matches the CERES data, while still using an analytical fit of the form of Eq. (A3) of , and with a similar global albedo to our previous MiMA work. See Appendix A for details, and Fig. S1 in the Supplement for a comparison of surface temperature in NH midlatitudes to ERA5. Note that the albedo in CERES and ECHAM refers to top of atmosphere albedo, while in MiMA it refers to surface albedo, as there are no shortwave cloud effects in MiMA. The albedo for the aquaplanet experiment is shown in Fig. a, and this profile is also used in runs without land–sea contrast.

The second major change made to the configuration of is an adjustment to the ocean heat flux to better account for the net transport of heat from the SH to the NH associated with the meridional overturning circulation in the Atlantic. This adjustment is necessary because the formulation in implies too little transport of heat from the SH to the NH within the ocean, as compared to the observationally derived flux of . The details of the perturbation are shown in Appendix A. The net effect of the perturbation (when A in Eq.  is set equal to 2.1 W m⁻²) is a more realistic implied net heat transport to the NH. Namely, the net heat transport from the SH (20° S-pole) to the NH (20° N-pole) is 0.54 PW, both here and in the observationally derived estimate from . For reference, the corresponding heat transport over these latitude bands in the configuration is just 0.24 PW, even as it captures the overall spatial pattern. inferred an alternative heat transport estimate of 0.4PW from the SH to the NH when integrating over the entire hemisphere, but with large interannual variability. Note that there are even larger differences in the ocean heat transport in other published estimates (compare , , , and ), with substantial sensitivity to methodology in the vicinity of the boundary currents .

The updated ocean heat flux is shown in Fig. c, while the configuration from (with A = 0 W m⁻² and 0.24 PW in interhemispheric transport) is in Fig. b. While the difference is difficult to detect by eye, this small change leads to a realistic NH vs. SH asymmetry in the surface energy flux term (as discussed in Sect. ). In addition to A = 2.1 W m⁻² (with 0.54 PW in interhemispheric transport), we perform sensitivity tests in Sect.  in which we set A to 3.5 W m⁻² to further increase the export of heat to the NH to 0.74 PW (Fig. d). This higher value of A is intended to explore sensitivity to uncertainty in this difficult-to-observe quantity. The ocean heat flux is steady (not monthly varying). The implications of this simplification are discussed in Sect. . Note that all configurations, including the aquaplanet version without any hemispheric asymmetries, have the tropical OHT of Fig. a.

We evaluate the role of three core surface inhomogeneities – topography (TOPO), land–sea contrast (LSC), and ocean heat transport (OHT) – for storm tracks using the MiMA experiments listed in Table . These experiments are similar to those in which focused on stationary waves, but with updates in the specification of each of these inhomogeneities as discussed in and Sect. . Each experiment lasts 38 years after discarding at least 10 years of spinup to achieve a statistical steady state, and so oceanic and atmospheric energy storage do not contribute to differences in the energy budget among the experiments. In addition to these eight experiments, Sect.  describes sensitivity ALL3 experiments that explore the role of land–sea and meridional gradients in albedo, and the flux of additional heat to the NH from the SH. All integrations here were run at a horizontal resolution of triangular truncation 42 (T42), although the most realistic configuration was also run at T85 with similar results (see Sect. ). All integrations were run with 40 vertical levels with a model lid near 70 km.

These experiments can be used to infer the role of each surface inhomogeneity for storm tracks in two complementary ways. Adopting the terminology of , these are known as the isolated and full nonlinear response to a given surface inhomogeneity. First, we denote as M(I) the response to some source of inhomogeneity I in MiMA, when compared to a zonally symmetric aquaplanet; this M(I) is called the isolated nonlinear response to I. Next, we let T represent all three inhomogeneities in the most realistic configuration (ALL3), such that the response to T is M(T). As in , we can then compute the difference between M(T) and M(T − I), and this response is known as the full nonlinear response to I. If we consider adding the three different inhomogeneities in sequence, the isolated nonlinear response to I occurs when I is added first, while the full nonlinear response to I occurs when I is added last (or is the first to be removed).

Figure S2 demonstrates that the non-additivity noted in with regards to boreal winter stationary waves is still evident in these experiments which incorporate the changes of Sect. .

We adopt two measures of the storm track activity. The first is transient kinetic energy (TKE), which is computed by applying a 2 to 8 d band-pass 5th order Butterworth filter to u and v, and then computing uhi2+vhi2. We then vertically integrate from the surface (ps) to the top of atmosphere: 1TKE=-12g∫ps0uhi2+vhi2dp.

The second measure of storm track activity is motivated by the energy budget framework used in Sect.  to diagnose how the three surface inhomogeneities affect the NH vs. SH storm track asymmetry. For this energy budget framework, a more natural definition of storm tracks is the atmospheric moist static energy flux on sub-monthly timescales. We now introduce this energy budget framework. The total (atmosphere plus ocean) meridional heat transport (MHT) across a latitude circle by the coupled (ocean–atmosphere) climate system must, on long time scales, be balanced by the net top-of-atmosphere (TOA) radiative deficit spatially integrated over the polar cap bounded by that latitude circle (e.g., , Fig. ). This constraint helps motivate an energetic framework, in which MHT is equal to the net TOA radiative deficit integrated over the extratropics or, equivalently, the net radiative excess integrated over the tropics e.g.,. This statement can be expressed mathematically as 2MHT(ϕ)=-2πa2∫ϕpolecos⁡ϕ′ASRϕ′-OLRϕ′dϕ′, The hemispheric-scale radiative imbalance results from the equator-to-pole gradient of absorbed solar radiation (ASR; dark pink in Fig. ) being steeper than that of outgoing longwave radiation (OLR, light pink in Fig. ; ). On subyearly timescales, atmospheric and oceanic storage is a critical part of the energy budget , however we focus on 38-year averages after a long spinup and so storage is not important.

The MHT is typically decomposed into four terms, an ocean heat transport term (OHT; blue in Fig. ) and an atmospheric heat transport term (AHT) which is further decomposed into contributions from the mean meridional overturning circulation (MMC), stationary eddy flux (SE; red in Fig. ), and transient flux (TE; black in Fig. ). 3MHT=OHT+MMC+SE+TE︸AHT.

We calculate the AHT as the zonally and vertically (mass weighted) integrated meridional transport of moist static energy MSE = CpT + Lq + gZ, where T is the atmospheric temperature, Cp is the specific heat of air at constant pressure, L is the latent heat of vaporization of water, q is the specific humidity, g is the acceleration of gravity, and Z is the geopotential height. The meridional velocity and MSE are subdivided into the zonal and stationary eddy components, and also time-mean and transient components (as in ). Upon designating zonal averages with square brackets [], time averages over each month of analysis with overbars ()‾, departures from the zonal average with asterisks (^*), and departures from the time average with primes (^′), the time-mean, zonal-mean, and vertically integrated total energy transport is 4AHT(ϕ)‾=2πacos⁡(ϕ)g∫0ps[V‾MSE‾︸MMC+V‾*MSE‾*︸SE+V′MSE′‾︸TE]dp, where V is the meridional velocity and the vertical integral is over pressure p from the TOA to the surface. We use 6-hourly instantaneous fields to calculate the energy transport for each month of model simulation, averaging the results over all years to define the climatological AHT. Note that the transient term TE includes all deviations from sub-monthly timescales (both zonal mean and zonally anomalous; V′ = V′* + V′), and deviations are defined relative to its monthly average, that is an average over a particular month not the climatological monthly average . For the MMC term, mass conservation is ensured by subtracting from [V]‾ the pressure weighted vertical integral of meridional velocity at each latitude.

This calculation is performed on the model's native sigma levels to avoid interpolation errors and uses the daily varying latitude versus longitude surface pressure. In MiMA, all of these terms can be computed directly from the model output (as opposed to CMIP model data where typically transients are inferred as the residual of the rest of the terms, ), and the residual is small as discussed in Sect. . This small residual is due to performing the calculation on sigma levels – residuals are a factor of 3 to 4 larger when the calculation is performed after interpolating to standard pressure levels. We apply a 0.65 rescaling factor to all terms in the MSE budget following to allow for a simpler comparison to the TKE perspective of storm tracks. In statistical steady state with no net changes in oceanic or atmospheric storage, the OHT must balance the surface energy exchange, and thus can be computed by summing the evaporation, sensible heat, longwave, and shortwave surface terms.

In MiMA, we dictate the ocean heat transport through our specification of the ocean heat flux. We are also able to control absorbed solar radiation nearly completely via our specification of the albedo (note that atmospheric scattering and absorption are similar in all experiments, and recall that MiMA has no radiatively active clouds). Furthermore, changing the land–sea contrast and topography directly affects surface and atmospheric temperatures, and hence has an immediate impact on longwave radiation emitted. These surface boundary conditions, together with the ocean heat flux, also directly affect the atmospheric stationary eddies. The net effect is that by changing the albedo, the ocean heat flux, the land–sea contrast, and the topography, we exercise control over all of the parameters in Fig.  but one: the transient eddies. This enables us to use this budget to causally explain how changes in transient eddies are driven by perturbations to the surface boundary conditions. 5ΔTE=-ΔOHT-ΔMMC-ΔSE-2πa2∫ϕpolecos⁡(ϕ′)ΔASR(ϕ′)-ΔOLR(ϕ′)dϕ′,

This ability to relate surface inhomogeneities to storm track strength allows for a causal interpretation of storm tracks . Specifically, showed the SH vs. NH asymmetry could be connected to TOA radiation, surface fluxes and stationary circulation energy transport. Unfortunately, frameworks based on mean temperature cannot be easily related to these inhomogeneities, however changes in temperature are nonetheless helpful in understanding localization of storm tracks and are included in Sect. .

The asymmetry in storm track strength between the SH and NH is summarized for each configuration in Fig. . Storm tracks are 48 % and 25 % stronger in the SH for 2–8 d filtered and sub-monthly filtered transient kinetic energy, respectively, in general agreement with . As surface inhomogeneities are removed, the asymmetry declines. Even in the presence of just land–sea contrast (LSC-only), however, the asymmetry still has not been eliminated (e.g., 9.1 % for submonthly). Part of the asymmetry is due simply to the albedo profile: there is an asymmetry of 6.6 % to 7.5 % (for a submonthly filter and 2–8 d bandpass filter, respectively) in the aquaplanet run with the updated albedo profile in Fig. , as compared to the albedo profile used for in which the asymmetry is less than 1.5 %. This asymmetry can be explained by the fact that while the hemispherically averaged albedos are similar in both hemispheres, the subpolar albedo is higher in the SH than in the NH in the updated configuration and in CERES (Sect. , Fig. ), and so the meridional gradient in net energy input is larger in the SH (an effect we will quantify shortly using the energy budget). For most experiments results are similar whether we define transient kinetic energy using a 2–8 d filter or a sub-monthly filter. However, for runs with land–sea contrast off but topography on (particularly topography-only, and to a lesser degree OHT+topography) some of the topographically forced quasi-stationary waves are included in the transient eddies category for a sub-monthly definition, leading to stronger storm tracks in the NH.

When considering the 2–8 d bandpass filtered kinetic energy, the most important surface inhomogeneity for the asymmetry is OHT for the isolated nonlinear response (i.e. when comparing to AQUA), while all three have roughly equal contribution for the full nonlinear response (when each is removed from ALL3). However there is substantial non-additivity: the sum of the full nonlinear responses is 30 % larger than the actual asymmetry of 48 % in All3. In contrast, the sum of the isolated nonlinear responses is only 25 %, which is only half of the actual asymmetry in All3 of 48 %. Hence, surface inhomogeneities have a bigger impact on the NH vs. SH asymmetry when they are subtracted from ALL3, and a weaker impact when they are added to AQUA. This effect is opposite to what we saw in Sect.  with regards to regional impacts: the surface inhomogeneities have a bigger impact on regional storm tracks when they are added to AQUA than when subtracted from ALL3.

The moist static energy budget introduced in Sect.  is now used to explain why the SH storm tracks are stronger than the NH storm tracks. We begin by showing the terms of the MSE budget for the aquaplanet and ALL3 configurations in Fig. . Figure a shows the net energy input, and as expected energy accumulates in the tropics while there is a deficit in the extratropics (e.g., Fig. ). For a given albedo profile, the TOA terms are similar between ALL3 and aquaplanet (Fig. b), while changing the albedo profile has a large effect on the TOA radiative terms (not shown). In contrast, changing the ocean heat flux between the idealized configuration for an aquaplanet (Fig. a) and any of the more realistic configuration leads to a large change in the surface energy input (Fig. c).

The integrated energy input from latitude ϕ to the pole must be balanced by the energy transport at that latitude (Eq. ), and these energy transport terms are shown in the right column of Fig. . The poleward energy transport is dominated by transient eddies (Fig. d), and these transient eddies are stronger in the NH for the aquaplanet configuration than for the ALL3 configuration (Fig. a, b). Conversely, the stationary eddy terms are stronger in the ALL3 configuration than in the aquaplanet configuration (Fig. f), such that there is a hand-off from transient to stationary eddies that keeps net MSE transport quantitatively similar to balance the similar energy input (Figs. a; a, b). Note that we compute all terms explicitly from the model output – none are inferred as residuals.

Next, we quantify how each of the surface inhomogeneities affects each of the terms in the MSE budget, and thus can explain the changes in the hemispheric-averaged transient MSE flux. To do this, we apply Eq. () to all latitudes from 20° to the pole in both hemispheres, and then take the area-weighted average from 20° to the pole. For context, applied this identical framework to reanalysis data and ECHAM, and showed that the Southern Hemisphere is 24 % stormier than the Northern Hemisphere across reanalysis data sets. Furthermore, the MSE budget based storm track asymmetry climatology (rescaled to match the EKE) in ERA5 is 23 % and can be diagnostically decomposed into a 12 % contribution from surface energy fluxes, a 14 % contribution from stationary circulation atmospheric energy flux, and a small −3 % contribution from TOA radiative fluxes. The net energy flux in each hemisphere in MiMA (SH filled and NH outline) is shown in Fig. a, with the percentage difference between the hemispheres in Fig. b. In ALL3, the SH is 34 % more stormy by this metric (black bar), and of this 15.6 % is due to the surface energy input (blue bar; that is, the ocean fluxes more heat poleward in the NH than in the SH) and 22.4 % to stationary eddies/MMC (red bar; that is, stationary eddies are stronger in the NH than in the SH). Note that this suggests MiMA has a different MSE budget based storm track asymmetry climatology than ERA5 as it places more weight on the stationary eddy contribution. The top of atmosphere term contributes -5.2 %, such that the TOA energy deficit is slightly larger over the NH than the SH and thus partially mitigates the effects of ocean and stationary eddy fluxes. The residual of the MSE budget (green bars) are a factor of 10 less than the smallest term, and so the energy budget is essentially closed. The residuals are similarly tiny (≤ 0.12 mPW per hemisphere) for all of the other experiments when we sum up the energy terms in each hemisphere.

The SH vs. NH asymmetry in TE drops to 24.5 % if we flatten topography but keep OHT and LSC (black in Fig. b2). This reduction is due to a reduction in the stationary eddy term (red in Fig. b2) as NH stationary waves weaken (red in Fig. a2), however the change in the NH vs. SH asymmetry of the TE term (black in Fig. b1, b2) is mitigated somewhat by changes in the TOA term (orange in Fig. b1, b2), whereby the TOA deficit is now nearly identical in the SH and NH (orange in Fig. b2).

If we remove realistic OHT but keep LSC and topography (Fig. b3), the SH vs. NH asymmetry in TE drops from 34 % to 22.4 %. This reduction is mostly due to zeroing out of the surface term (as expected, blue bar in Fig. b3), however the stationary eddy term also weakens (from 22.4 % to 15.5 %) as some of the stationary eddy energy transport is set up by wavetrains forced by sea surface temperature gradients arising from the ocean heat flux (Fig. ). The increase in TE in the NH vs. the SH is partially mitigated by changes in the TOA term (orange in Fig. b3): in ALL3 the TOA deficit was larger in the NH, but in TOPO+LSC the TOA deficit is larger in the SH (orange in Fig. a1 and a3). This change in the sign of the contribution of the TOA term is consistent with the ECHAM experiments of , though the effect is stronger here in MiMA. This change in the TOA term is almost entirely due to changes in longwave (light pink) rather than shortwave (dark pink). Namely, TOA shortwave energy input is essentially identical in all experiments (dark pink in Fig. a) due to the specified albedo profile, while TOA longwave is not (light pink in Fig. a).

If we remove LSC but keep OHT and topography (Fig. b4), the SH vs. NH asymmetry in TE drops from 34 % to 17.9 %. This reduction by half is not because of changes in the surface energy term (which are unchanged), and changes in the stationary eddy term are even more pronounced when we remove OHT but keep LSC and topography (red bar in Fig. b2 vs. Fig. b4). The dominant reason for this reduction is, instead, from the TOA asymmetry: the NH TOA deficit grows when LSC is removed – opposite to Fig. b2 and b3 – due to the outgoing longwave radiation term (orange and light pink bars in Fig. b4).

The remaining panels of Fig.  focus on the isolated nonlinear response when each building block is added onto an aquaplanet configuration. As for the full nonlinear response, the most important individual surface inhomogeneity for the NH vs. SH asymmetry in storm track strength is LSC. When LSC is added onto an aquaplanet (Fig. b5), NH TE weakens while stationary eddies strengthen, however changes in the TOA term (orange) are equally important to the changes in stationary eddies. Adding LSC leads to less TOA OLR in the NH (light pink), and hence less overall heat transport is needed in the NH, which accentuates the NH vs. SH asymmetry. For the other two isolated nonlinear responses, the TOA term acts as a negative feedback on the increase in the NH vs. SH asymmetry in TE, rather than contributing to it as it does for LSC. For example, adding OHT to an aquaplanet leads to an increase in extratropical NH TOA OLR, which subsequently increases the NH extratropical energy deficit and thus partially mitigates the reduced TE that is needed in response to increased OHT (Figs. a6, c).

Thus far, we have shown that the surface energy input and stationary eddy terms are the most important contributors to the SH vs. NH asymmetry in storm track strength in All3. Nonetheless, changes in the TOA energy output are not insignificant, and act to intensify the asymmetry when LSC is added yet reduce the asymmetry when topography or especially OHT are added. We now turn our attention to why the NH vs. SH asymmetry in the TOA term differs so strongly among these three surface inhomogeneities, and specifically we demonstrate its tight connection to hemispheric averaged surface temperature. Figure a shows the area-weighted OLR from 20° to the pole in each hemisphere in all experiments included in this paper, and we also include the corresponding core 8 simulations using the configuration to demonstrate statistical robustness. Consistent with Fig. , OLR is higher in the NH than in the SH in all simulations except for an aquaplanet configuration with completely symmetric albedo profiles (labeled “2- Aqua ac1”), but the magnitude of the SH vs. NH asymmetry in OLR differs across the experiments.

To explain this asymmetry in OLR, we next turn to the asymmetry in area-weighted surface temperature from 20° to the pole in each hemisphere (Fig. b). The NH is warmer than the SH in nearly all simulations, with the asymmetry smaller in aquaplanet runs or in runs with only land–sea contrast (runs 2 through 5). That is, both topography and ocean heat transport act to cool the SH more than the NH, while land–sea contrast cools both hemispheres nearly equally.

The importance of the surface temperature asymmetry for the OLR asymmetry is quantified in Fig. c. Across all 21 simulations, the correlation between the OLR and the surface temperature asymmetries is 0.99, and hence surface temperature asymmetries dictate OLR asymmetries in MiMA. The corresponding correlation with ocean heat transport or albedo is much lower (0.79 and 0.64). The asymmetry in surface temperature is particularly large in runs with OHT on but land–sea contrast off (runs 7 and 11), and also is enhanced using the updated OHT profile vs. the White et al 2021 profile, and even more-so when we amplify the OHT perturbation (ALL3 [extra SH→NH OHF], run 19). Of the three surface inhomogeneities, LSC tends to have a muted impact on the hemispheric asymmetries in surface temperatures and if anything preferentially cools the NH (runs 2 through 5 are near the origin in Fig. c), while the other two increase it and preferentially cool the SH (with the effect particularly strong for OHT, which warms the NH much more than the SH). Hence, the changes in the TOA term in Fig.  can be explained by the effect of each building block on surface temperature. These sensitivities appear to be larger than in the simulations of in which idealized orograpy is added to an aquaplanet.

Overall, the storm tracks in the SH are stronger than those in the NH. All three surface inhomogeneities contribute to this asymmetry, but the contributions are unequal. The most important contributor is land–sea contrast, the second most important is ocean heat flux, while topography is the least important contributor. This relatively small role for topography is particularly pronounced when using a sub-monthly definition for storm track intensity (Fig. ). Topography has a pronounced role only in the vicinity of the Tibetan Plateau, but in the zonal mean this effect is relatively weak. Most of the change in transient MSE flux as surface inhomogeneities are removed can be explained by the stationary eddies (for all three surface inhomogeneities) and surface terms (for ocean heat transport). These surface inhomogeneities also influence the hemispheric averaged surface temperature, and the subsequent effect on outgoing longwave radiation changes explains the rest of the change in transient MSE flux. Specifically, adding OHT acts to preferentially warm NH subpolar latitudes more than SH subpolar latitudes, which partially mitigates the effect of OHT on the SH vs. NH storm track asymmetry that would occur if just the stationary eddy and surface terms were present. In contrast, land–sea contrast acts to preferentially cool NH subpolar latitudes more than SH subpolar latitudes, which intensifies the effect of LSC on the SH vs. NH storm track asymmetry beyond what would occur if just the stationary eddy term was present.

Our results agree with previous work in terms of the importance of topography and OHT on the hemispheric asymmetry of the storm tracks (, cf. Figs. , ). Interestingly, in MiMA land–sea contrast is the most important of the three surface inhomogeneities for most storm track metrics, which clearly contrasts with the conclusions of and . There are a few possible reasons for this difference.

The albedo in MiMA is specified to be similar to the time mean state of CERES, however, there are no radiatively active clouds. In contrast, the ECHAM6 simulations of have a full cloud scheme with smaller biases as compared to many other CMIP6 generation models , however, there are nonetheless albedo biases in midlatitudes (Fig. ). We have performed a MiMA experiment designed to isolate the role of this albedo bias by mimicking the albedo bias in ECHAM6, and it indicates that this bias is not critical for storm track climatology in the hemispheric mean, though it does have regional impacts (Figs. c and c). More generally, MiMA does not have a realistic land surface (rather, the heat capacity, surface roughness, and moisture availability are modified to mimic land) and therefore might misrepresent the role of LSC.

, , and find that all three surface inhomogeneities act to localize storm tracks in the North Pacific and North Atlantic basin, in agreement with our results. These studies further documented non-additivity in the response to each of these inhomogeneities. For example, the land–sea contrast associated with the North American continent shapes the North Atlantic storm track more strongly when the orography is included than when it is absent . In contrast, our experiment adding land–sea contrast last leads to a weakening of the subpolar North Atlantic storm track, while adding it onto an aquaplanet allows for a strong localized North Atlantic storm track (Fig. h). We relate this response in the North Atlantic sector to differences in the subpolar surface temperature response and its impact on the local meridional temperature gradient (Fig. g, h): Greenland cools when land–sea contrast is added to an aquaplanet (Fig. h), however it does not cool when land–sea contrast is added onto a configuration that already has topography (Fig. g). did not include Greenland in their simulations, however Greenland is known to influence storm development . Future work should isolate its role for climatological North Atlantic storm track strength and location.

The zonal structure in the SH storm track is particularly pronounced in austral winter, and , , and focus on causes of this asymmetry. argue that zonal structure in SSTs is responsible for most of the large-scale storm track structure in the SH midlatitudes, while pinpoint the crucial role of Antarctic topography. Figure S7 isolates the role of each building block in JJAS, and demonstrates that both orography and OHT strengthen storm tracks in the Indian Ocean sector. However, topography appears to be slightly more important for this localization (Fig. S7e, f vs. Fig. S7c, d). The net effect is that the relatively strong SH storm track in the Indian Ocean sector in ALL3 is mostly due to the fact that East Antarctica has higher topography relatively close to the continental edge (consistent with , who additionally diagnose the relevant dynamics behind this effect), but with east-west gradients in OHT also playing a role . This effect is much stronger at T85 resolution than at T42 (Fig. ), which is not surprising given the nature of SH topography.

While all CMIP-class models used for climate projections have realistic topography and continental distribution, the strength of the ocean heat transport into the North Atlantic differs by more than a factor of two , even if attention is restricted to HighResMIP models. Heat transport is also too weak in nearly all coarse resolution models, though some higher resolution models are relatively better performing. Southern Ocean sea surface temperature biases are also prevalent and generally not improved at high resolution , and while some of the biases can be linked to biases in clouds or sea ice, this biased climatology affects storm track strength . These mean state biases contribute to intermodel spread in future climate projections , and there are large observational uncertainties in surface flux trends in the CERES era .

In our experiments, a relatively small change to regional OHT (well within observational uncertainty) causes large changes in regional storm track strength (Fig. a), specifically localized strengthening but hemispheric weakening (cf. Fig. k, l). These changes in regional OHT also influence stationary wave strength, and so it is oversimplified and incorrect to predict the changes in storm track strength as a simple compensation to the altered OHT (Fig. a), though other modeling studies have observed a higher degree of compensation than we find in response to idealized topography or Last Glacial Maximum conditions . Either way, the emerging discrepancy in storm track trends between CMIP and observations is likely linked to a misrepresentation of sea surface temperature trends in models and subsequent stationary Rossby wavetrains that spread biases remotely , due to the strong linkage between surface energy input and stationary eddy terms. Finally, because the net response to any surface inhomogeneity is dependent on the background state set up by the other inhomogeneities (as we demonstrate throughout this paper), differences in the background state across models due to differing biases in OHT or clouds can then lead to spread in how topography and land–sea contrast affect storm tracks. The net effect is that our study helps us understand why it is so difficult to remove lingering storm track biases from CMIP class models, while pinpointing physical processes that influence their behavior.

We used two distinct metrics to track storm tracks: moist static energy fluxes and transient kinetic energy. In many aspects the two metrics gave similar answers, however the role of OHT is somewhat larger for transient kinetic energy (Figs.  and ). While TKE is largest in the upper troposphere, transient MSE fluxes are influenced by moisture and therefore peak in the lower troposphere (Figs. S8 and S9). This might explain some of the disagreement regarding the impact of the different inhomogeneities for the NH-SH asymmetry. More generally, Figure S8 and S9 reveal non-additivity in the latitude vs. height distribution of TKE. Future work should consider the vertical structure of storm track and temperature changes to help understand the forcing mechanisms behind non-additivity, and could also consider a Lagrangian perspective on storm tracks. Future work could explore the relative roles of dry vs. moist processes for the results found in this paper, and also explore why some models show stronger compensation between stationary eddies, the mean meridional circulation, and transient eddies than we find here .

An assumption throughout this paper is that each of the three surface inhomogeneities is independent of the rest. This is because our goal is to represent each of the three surface inhomogeneities as closely as possible to current Earth-like conditions. Nonetheless, the ocean circulation would be drastically different if the current configuration of continents were changed or if topography was flattened , and MiMA's “water mountains” (i.e., topography without land–sea contrast) are highly idealized. If we allowed the ocean circulation to change, we expect that non-additivity would be even more pronounced. Future work should explore this possibility via coupling to a full ocean. Finally, future work should focus on how these localized storm tracks change in response to increased CO₂ by repeating the simulations here, but with elevated CO₂ concentrations.