Content area
Abstract
Generations of climate models exhibit biases in their representation of North Atlantic storm tracks, which tend to be too far near the equator and too zonal. Additionally, models have difficulties simulating explosive cyclone growth. These biases are one of the reasons for the uncertainties in projections of future climate over Europe, and the underlying causes have yet to be determined. All three biases are shown to be related, and diabatic processes are pointed to as a likely cause. To demonstrate this, two hemispherically symmetric storm tracks forming downstream of an idealized sea surface temperature (SST) front on an aquaplanet are examined using the seamless ICOsahedral Non-hydrostatic weather and climate model (ICON) and its grid refinement capabilities. The analyzed perpetual boreal winter has a global grid spacing of 20 km, two bi-directionally interacting grid nests over the Northern Hemisphere that refine the grid to 10-km spacing over much of the stormtrack and further to 5-km spacing near the SST front. In contrast, no grid refinement is performed for the Southern Hemisphere. Feature-based cyclone tracking shows that the poleward propagation in the NH is enhanced, so the high-resolution storm track is less equatorward and less zonal; explosive deepening rates are more frequent and precipitation rates are amplified. The implication is that resolving diabatic processes on the storm scale improves all three intersecting biases in the representation of storm tracks. While new challenges arise at cloud resolving scales, much improvement for the representation of storm tracks will be gained once climate models resolve the meso-γ scale.
Full text
Climate models exhibit systematic circulation-induced biases in key large-scale atmospheric flow features in both the tropics and extratropics. Two of the best known examples are the bias of the double intertropical convergence zone, which leads to a double peak in the tropical precipitation distribution near the equator and unrealistically large precipitation south of it (Adam et al., 2017; Mechoso et al., 1995; Tian & Dong, 2020) and the “too-zonal and too-equatorial” bias in North Atlantic storm tracks that has existed for several CMIP generations (Doblas-Reyes et al., 1998; B. J. Harvey et al., 2020; Priestley et al., 2020; Ulbrich et al., 2008). As a result, the IPCC AR6 states that “there is high confidence that atmospheric circulation biases can degrade model representations of regional land surface climate” (WG1 Chapter 10). For the projection of future regional land precipitation changes downstream of the North Atlantic storm tracks, this implies increased uncertainty, since regional precipitation changes are mainly determined by large-scale circulation changes. As a result, the sign of, for example, changes in fall but also annual mean precipitation under the SSP5-8.5 scenario over western and central Europe remains unclear which can be conveniently verified using the new IPCC WGI Interactive Atlas accessible via
Climate modelers place great confidence in the next generation of kilometer-scale (or storm resolving) climate models (Neumann et al., 2019; Stevens et al., 2019), which will allow explicit representation of deep convection and more detailed representation of the bulk influence of diabatic processes on the adiabatic flow, even if structural details of individual clouds remain unresolved (Palmer, 2014; Prein et al., 2015; Schär et al., 2020). Regional kilometer-scale simulations have indeed improved the representation of, for example, the diurnal cycle of convection and precipitation (Panosetti et al., 2018, 2019; Stevens et al., 2020) as well as the structure of mesoscale eddies along cold fronts (B. Harvey et al., 2017; Leutwyler & Schär, 2019), tornadoes (Hanley et al., 2016), cloud cover (Hentgen et al., 2019), and the response of low-lying clouds over the tropics to warming (Schneider et al., 2019). It is unclear to what extent kilometer-scale global simulations can overcome the “too zonal and too equatorial” bias. Significant technical challenges remain to be overcome to enable long-term kilometer-scale global simulations with coupled Earth system models. These include efficient use of novel computer architectures, exploiting the power of machine learning and data assimilation to produce better parameterizations of subgrid-scale processes, and developing better numerical algorithms to solve the laws of physics (Bretherton et al., 2022; Neumann et al., 2019; Schneider et al., 2017; Schulthess et al., 2019). It seems necessary to clarify the added value of these efforts to overcome known circulation biases at an early stage.
This study addresses the question of whether kilometer-scale simulations have the potential to overcome the critical “too zonal and too equatorial” circulation bias observed over the North Atlantic storm track in CMIP models. Among the potential mechanisms underlying the bias, sea-surface temperature anomalies (Keeley et al., 2012), underresolved orography (Berckmans et al., 2013; Davini et al., 2022) and low model resolution in general (Curtis et al., 2020; Zappa et al., 2013) are all discussed in the literature. To test the role of resolution and the question if processes on the storm scale and thus internal to storm tracks account for the bias, a 120-month perpetual boreal winter simulation (corresponding to 40 winter seasons) is conducted at a range of resolutions from a global resolution of about 20–10 km over the storm track to 5 km over an idealized sea surface temperature (SST) front using regional resolution refinements via bi-directional interacting grid nests. Much of the storm track forming downstream of the idealized SST front is simulated within the first nest at a 10-km grid spacing, thus resolving the meso-γ scale circulation within storms, while convection begins to be resolved in the innermost 5 km nest over the SST front, where diabatic processes are climatologically most frequent. A particular focus is on the orientation of the high-resolution storm track compared to its low-resolution counterpart, which is simulated on the other hemisphere of the aquaplanet without grid refining nests (thus allowing for a direct comparison of the resolution impact within the same simulation). Consideration is also given to extreme growth rates. The aquaplanet setup with fixed SSTs used here allows to isolate the role of model resolution in the absence of mountains or any potential biases in the coupling between atmosphere-land-ocean models. The main hypothesis is that an improved storm track representation results from processes internal to storm tracks (i.e., diabatic processes), which ultimately drive more poleward propagating and also stronger deepening extratropical cyclones.
Model and MethodsThe ICOsahedral Non-hydrostatic weather and climate model (ICON) (Zängl et al., 2015) is the new seamless weather and climate prediction model at MeteoSwiss National Weather Service (MeteoSwiss) and is developed by the ICON consortium centered around the German Weather Service (DWD) and the Max Planck Institute for Meteorology. In this study, ICON 2.6.4 is used and the set of parameterizations follow the DWD operational standard configuration and include a one-moment two-category microphysics scheme (Doms et al., 2011), non-orographic gravity wave drag (Orr et al., 2010), a prognostic TKE scheme for sub-gridscale turbulent transfer (Raschendorfer, 2001), and the radiation scheme ecRad developed at ECMWF (Hogand & Bozzo, 2018). Globally, the model runs with an approximate grid spacing of 20 km (R2B7), 180 s time step and parameterized deep convection (Tiedtke, 1989). A reduced scheme for shallow convection is used in the 10 km (R2B8) nest and no convection is parameterized in the 5 km (R2B9) nest. Shallow convection is non-precipitating and momentum flux is, in contrast to temperature and humidity tendencies, neglected (Doms & Förstner, 2004). The two grid refining nests also refine the time step by a factor of two starting from 180 to 90 s and to 45 s in the inner nest. It is the innermost nesting that makes the computation particularly resource intensive, as the global model must wait for the computation to complete in the 5 km nest. All analyses are performed on the 20 km domain coarse grained globally to a regular 0.5° grid, by doing so only the influence of the high-resolution nest on the global scale is assessed.
Figure 1 illustrates the nesting procedure for a single time step and the wind speed at the 300-hPa level. The two-way interaction between the smaller and the larger nests consists of three steps. First, the fields are upscaled at each time step by interpolation; second, the difference between the solution on the larger grid and the upscaled solution is calculated; third, the solution on the larger grid is relaxed toward the upscaled solution. The interaction between the larger and smaller grid is done using the solution on the large grid as boundary values for the smaller grid. More details can be found in Section 3.9 of the ICON manual (Prill et al., 2020). Both nests extend all the way to the model top, which is placed at approximately 70 km using 70 vertical levels.
Figure 1. Snapshot of wind speed at the 300-hPa level for the global grid and the two bidirectionally interacting grid nests, which refine the grid spacing from 20 to 10 km down to 5 km. The continents are shown only for illustrative purposes.
Initial conditions for wind, temperature and pressure are analytically prescribed following the Jablonowski-Williamson baroclinic wave test case (Jablonowski & Williamson, 2006). Prescribed SSTs, which are held constant, are following the “Qobs” distribution by Neale and Hoskins (2001). Both follow observed wintertime (DJF) zonal mean conditions in the Northern Hemisphere and are inverted and used to initialize the SH as well. In addition and in both Hemispheres, an idealized SST anomaly is superposed on the otherwise zonally symmetric background SST. It consists of two rotated ellipsoids of the same size but reversed sign with maximum amplitude at the centers of the ellipsoids (10 K) decaying exponentially with distance. Combined, the created SST anomaly mimics the Gulf Stream plus the land-sea contrasts along the east coast of North America. The design of the initial conditions, in particular the position of the jet and SST front position, are inspired by the zonal mean wintertime conditions over the North Atlantic seen in reanalysis data but it should be noted that the thereby created zonal asymmetry does not necessarily mimic the role of the Gulf Stream SST front alone. It is a convenient way to mimic a zonal asymmetry that locally perturbs the atmosphere through heat and momentum flux. The model simulates a total of 10 perpetual years, with the solar zenith angle held fixed over the equator at 90° corresponding to late winter (end of March) conditions. By doing so, both hemispheres are symmetric in terms of incoming radiation, which facilitates a comparison. The first month is considered as spin-up period. Apart from the two grid refining nests, both hemispheres are thus symmetric and are therefore directly comparable by inverting the Southern Hemisphere and removing it from the Northern Hemisphere. A very detailed analysis of the thereby generated storm tracks including their response to global warming, which mimic the response seen in CMIP6 data over the North Atlantic, is presented in Schemm et al. (2022).
Finally, the storm track is studied from an Eulerian and Lagrangian viewpoint. The advantage of using Lagrangian feature-based tracking is the ability to produce tendencies associated with the cyclones at different times and downstream locations within the storm track, and separate the roles of cyclones and anticyclones. The method to detect surface cyclones is described in Wernli and Schwierz (2006). It identifies regional minima in the mean sea level pressure (SLP) field, which are enclosed by a closed SLP contour. These minima are subsequently tracked in time. Further, cyclone masks are obtained by flagging all grid points inside the outermost closed contour with ones, and all other grid points with zeros. The time averaged cyclone mask yields a detection frequency at each grid point, which indicates the relative fraction of all time steps visited by a surface cyclone. A total of 41,300 cyclone tracks are analyzed based on the decade long simulation.
Result: Less Zonal, More Intense and More PolewardThe zonally asymmetric SSTs result in a localized storm track downstream of the SST front, reminiscent of the North Atlantic storm track that forms downstream of the Gulf Stream. The underlying reason is that differential heating by sensible and latent heat flux maintains a low-level zone of baroclinicity, which by virtue of thermal wind balance, induces vertical wind shear and facilitates the formation of locally enhanced jet stream (Brayshaw et al., 2011; Chang et al., 2002; Hotta & Nakamura, 2011; Sampe et al., 2010). The resulting storm tracks that intensify downstream of the SST front can be studied from an Eulerian perspective using eddy kinetic energy (EKE). EKE is a traditional variance-based measure of storm track intensity. On the aquaplanet, the EKE at the 300-hPa level, calculated here with a 6-day high-pass filter, has its maximum downstream of the SST front at about the same latitude as the center of the SST front (black solid contour in Figure 2). It is this region of maximum EKE where the storm growth climatologically is highest. The maximum EKE extends much farther downstream in the SH than in the NH, where it terminates at 90°E. Mean SLP (blue contours in Figure 2), which is another measure of storm track intensity, shows a localized region of lower SLP downstream and poleward of the SST front. The region with the lowest SLP is located poleward of the region with high EKE, consistent with reanalysis data (Schemm & Schneider, 2018). This is because EKE indicates the region with high growth, while SLP indicates the region with maximum intensity (defined as minimum SLP during life cycle); from a cyclone life cycle perspective, the former occurs earlier during the life cycle of cyclones and thus is located more equatorward (Schemm et al., 2021, see their Figure 7). Regions of low SLP values (below 995 hPa) extend far downstream in SH, whereas it terminates earlier in NH, similar to EKE.
Figure 2. Less zonally extended storm tracks in the Northern Hemisphere: Eddy kinetic energy (black contours; 215 and 245 J kg−1) at the 300-hPa level and mean sea-level pressure (blue solid contour; 990 and 995 hPa) indicate a reduced zonal extension of the high-resolution storm track in the Northern Hemisphere. Additionally shown are surface cyclone track frequencies (color shading; %). Further, the idealized sea surface temperature front is indicated in this and the following figures by its imprint in the 2-m temperature after removing the zonal mean (green solid contours indicate positive values and dashed negative values; starting at ±1.5, 2.0, and 2.5 K). The two bi-directional interacting grid refining nests in the Northern Hemisphere are indicated by black contours.
The EKE difference between the two hemispheres with the zonal mean removed is shown in Figure 3. The removal highlights the zonally-asymmetric component of the response, which pertains to changes in the direction of propagation of cyclones. Figure 3 shows that the EKE values in the NH are higher in a broad area encompassing the SST front (shaded red in Figure 3). EKE values are also higher northeast of the nest, suggesting a more tilted storm track in the NH, while the high EKE values in the SH extend zonally downstream (shaded blue in Figure 3). It is only a limited region along the eastern side of the 10-km nest where EKE in the SH exceeds EKE in NH, which is due to the downstream shift and more zonal orientation of maximum EKE in the SH. Overall, the differences suggest an enhanced deepening of the cyclones in the region around the SST front where EKE is enhanced, as well as a more northeastward oriented storm track in the NH. The removal of the zonal mean is justified because the enhanced poleward propagation of cyclones is best seen in the zonally asymmetric part of the response. In contrast, the full field difference (shown in Appendix A) highlights the zonally-symmetric part of the EKE response, which is dominated by a reduction of the zonal extension of the high-resolution storm track. A result that has already been discussed on the basis of Figure 2.
Figure 3. Stronger storm tracks in the high-resolution hemisphere: Shown is the difference in eddy kinetic energy (shading; J kg−1) between the two hemispheres at the 300-hPa level. The difference is computed between the hemisphere with the two interactive grid-refining nests (shown by black contours) and the hemisphere without the nests and the zonal mean removed to highlight the zonally-asymmetric change. The continental outlines are shown for illustrative purposes only. Stippling shows insignificant differences (below three times the standard deviation of a randomized ensemble of time series—see text for details).
Preferably, the simulation would span several decades to account for the internal variability of the extratropical circulation. Currently, performing a multidecadal kilometer scale simulation with a global or nested grid is still out of reach. Nevertheless, to reject the null hypothesis that both hemispheres do not differ significantly from each other, a randomized test is performed. If the null hypothesis would be true, a time step could equally likely occur in the other hemisphere as if the high-resolution nest would not affect the simulation. Therefore, the following test strategy is used. One hundred new time series are created, each of similar length to the original 10-year time series. A random number generator first selects the fraction of the time steps for which the NH and the SH are inverted. A second call to the random number generator selects the exact time steps to be inverted. For example, in the new randomized time series number 27 out of 100, 87% of all time steps have the hemispheres inverted. The randomized time series number 10 has only 25% inverted time steps. The exact time steps with inverted hemispheres are randomly selected in the second steps. Then, for each of the randomly inverted time series, the time average is calculated and the time average of both hemispheres is subtracted resulting in 100 hemispheric differences based on which a standard deviation is computed. In Figure 3, the stippling shows regions where the difference between the high- and low-resolution hemispheres (based on the original time series) is smaller than three times the standard deviation obtained from the ensemble of randomized time series. If it is larger, the null hypothesis is rejected and it is very likely that the high-resolution nesting produces a hemisphere that is distinct from the low-resolution hemisphere and both are not inevitable. The result of the test against randomization shows a high agreement with the significance obtained from a Kolmogorov-Smirnov test (not shown).
The Lagrangian perspective offers a convenient way for a detailed study of storm tracks and causes of the reduced bias. For example, feature-based surface cyclone frequencies, which is a measure for the fraction of time steps a region is affected by cyclones, show at first no clear difference between the two hemispheres. The picture becomes clearer when the SH is inverted and subtracted from the NH. Figure 4 shows the calculated difference between the cyclone frequencies of the NH and the SH and, for illustrative purposes only, the geophysical outline and grid refinement nests, and after removing the zonal mean to highlight the zonally-asymmetric response. The difference in the detection frequency, which is measured in percentage points, is in the order of one-third of the standard deviation of the month-to-month variability. In addition, as mentioned previously, the high-resolution storm track is less zonally extended and terminates earlier at about 75°E.
Figure 4. More poleward located storm tracks in the Northern Hemisphere: Shown is the difference in cyclone track frequency after 120 simulated months of perpetual winter (color shading; percentage points). The difference is computed between the hemisphere with the two interacting nest (shown by black contours) and the hemisphere without the nests and the zonal mean removed to highlight the zonally-asymmetric change. Additionally shown are difference in 500-hPa geopotential (red contours; 200, 225 and 250 m2 s−2) and the idealized sea surface temperature anomaly (black solid positive; negative dashed; −2.5, −2, −1.5, 1.5, 2, 2.5 K). The continental outlines are shown for illustrative purposes only.
Termination of the storm track is often favored by the occurrence of stationary waves that reduce the mean available potential through poleward energy transport (Kaspi & Schneider, 2011a, 2013). The storm track thereby terminates itself. The geopotential height at the 500-hPa level without the zonal mean serves as a proxy for stationary waves and it indeed shows the presence of an enhanced ridge far downstream of the high-resolution nest in the NH (red contours in Figure 4). Consequently, differences in cyclone frequency relative to the SH are greatest in this region, indicating the absence of the stationary ridge in the low-resolution SH and increased cyclone frequencies. The presence of the amplified stationary ridge at the exit of the storm track has also some important consequences for the regional climate in the accompanying region. Typically, regions affected by a stationary ridge (trough) are often warmer (colder) compared to other longitudes at similar latitudes (Kaspi & Schneider, 2011b), which is also the case on the aquaplanet.
The Lagrangian perspective provides further the opportunity to quantify changes in important life cycle properties that affect the climatology properties of storm tracks. First and foremost for this study, the meridional shift of extratropical cyclones is of great interest as it has been previously linked to the mean position and poleward shift of storm tracks (Tamarin-Brodsky & Kaspi, 2017). A physically plausible mechanism underlying the storm track bias is a too weak poleward propagation. Poleward propagation is known to driven by mutual interaction and amplification of upper- and lower level cyclonic flow anomalies, which correspond to the surface cyclone and the westward shifted upper-level trough, respectively (Besson et al., 2021; Coronel et al., 2015; Gilet et al., 2009; Oruba et al., 2013). The underlying non-linear dynamics can be described based on the interaction of two potential vorticity anomalies that are tilted westward with increasing height. Their associated cyclonic flow results in non-linear advection of the surface cyclone by the upper level anomaly (Coronel et al., 2015; Tamarin & Kaspi, 2016). A better representation of this interaction and/or increased diabatic strengthening of the PV anomalies will translate into enhanced poleward propagation.
Indeed, Lagrangian tracking reveals that the poleward propagation of cyclones in the high-resolution storm track is enhanced compared to the low-resolution storm track but the increase is confined to the period of intense growth. The following analysis of the poleward propagation focuses on three key time periods in the life cycle of an extratropical cycles: the latitudinal difference between (a) genesis and maximum growth; (b) genesis and maximum intensity; and (c) genesis and lysis. Consideration is given to tracks with genesis in the 10 km box (see Figure 1) in the NH and the equivalent region of the SH. No clear difference is found between the latitudinal displacement during the genesis to maximum intensity period. Neither in the mean nor median or in the upper percentiles. The mean poleward displacement in both hemispheres is approximately 6.2°–6.4° and the median poleward displacement is approximately 4.5°. The same holds true for the period genesis to lysis. The mean displacement defined in this way in both hemispheres is approximately 8.3° and the median poleward displacement is approximately 6°. It is the earliest time period between genesis and maximum deepening, which is also by definition the period marked by strongest growth, for which a clear difference is found in the upper percentiles (Table 1). Within the 10 km nest in the NH, the 99th and 99.9th percentiles are larger by 3° and 4°, compared with the SH respectively. There is no change for the mean (4°) and median (2.5°) displacement, indicating a reduction for the lowest percentiles. Accordingly, the most strongly poleward propagating cyclones propagate even stronger poleward during the early phase of the life cycle, a phase that is also marked by highest growth rates. As a result of the early and strong poleward propagation, the storms have reached higher latitudes earlier during their life cycle and spend longer time periods at higher latitudes, which agrees with the poleward shifted cyclone frequencies (Figure 4).
Table 1 Poleward Propagation for Deepening Cyclones Defined as Latitude Difference Between Genesis Latitude in the 10-Km Nest (80°W–40°E) and Latitude of Maximum Deepening
| Percentile | NH | SH | Δ |
| 99.9 | 32.5° | 28.5° | 4° |
| 99.0 | 21.5° | 18.5° | 3° |
| Mean | 4.0° | 4.0° | 0° |
| Median | 2.5° | 2.5° | 0° |
It is known that the poleward propagation of extratropical cyclones is inherently connected to their intensification, since cyclones that intensify more rapidly tend to propagate more poleward (Besson et al., 2021; Tamarin & Kaspi, 2016). Thus, stronger poleward propagation during the early life cycle phase is expected to be associated with higher growth rates. In fact, higher growth rates, defined as the maximum of all 6-hourly SLP changes for each track, are found near the SST front (Table 2). Comparing the maximum growth rates downstream and near the NH SST front (20°–80°W) with those near the SH SST front, the 99th and 99.9th percentiles are larger by 2.5 and 5.7 hPa (6 h)−1, respectively, while the global mean is unchanged by the nest. The increase in the strongest growth rates is thus consistent with the increase in the strongest poleward propagation downstream of the SST front.
Table 2 Growth Rates (6-Hourly Sea-Level Pressure Change in hPa) in Region Downstream of the Sea Surface Temperature Front (20°–80°W)
| Percentile | NH | SH | NH (SST front) | SH (SST front) | Δ (SST front) |
| 99.9 | 20.8 | 20.8 | 24.3 | 18.6 | 5.7 |
| 99.0 | 15.4 | 14.8 | 17.7 | 15.2 | 2.5 |
| Mean | 4.5 | 4.5 | 5.2 | 5.0 | 0.2 |
| Median | 3.7 | 3.6 | 4.3 | 4.3 | 0 |
The growth of particularly strong cyclones is often driven by additional heating from moist diabatic processes (Binder et al., 2016; Kuo et al., 1991; Reed et al., 1988; Stoelinga, 1996; Wernli et al., 2002, and many others). Since precipitation is a simple but effective indicator for latent heat release by moist diabatic processes in an atmospheric column, a comparison between the NH and the SH in the nested regions serves as a proxy of enhanced diabatic heat release in particular over the key cyclone amplification region near the SST front. Indeed, peak precipitation values in the NH downstream of the SST front exceeds its counterpart in the SH by an average of 0.3 mm (3 h)−1 (or 1.4 mm (3 h)−1 on average compared to 1.1 mm (3 h)−1). Meridionally averaged between 30° and 45°, precipitation in the NH exceeds precipitation in the SH at all longitudes but in particular within the 10-km nest (Figure 5). As an unfortunate side effect of the nesting, precipitation clearly reduces at the transition from the 10-km to the 20-km nest, while there is no such transition between the 5 and the 10-km. Regions with low-resolution 20-km grid spacing on average receive reduced precipitation and conversion from latent to sensible heat that warms the atmosphere and promotes vertical motion. Thus, the increase in extreme growth rates near the SST front in the early life-cycle phase likely relates to additional diabatically driven growth, which is stronger in high-resolution NH than in the low-resolution SH, in particular within the 10-km nest. These results suggest that stronger moist diabatic processes in the high-resolution nests ultimately drives a stronger poleward propagation and higher growth rates during the early life-cycle phase and that diabatic processes contribute to all three circulation biases “too zonal, too equatorward and too weak for extreme cases.”
Figure 5. Enhanced precipitation in the high-resolution Northern Hemisphere: Meridionally averaged precipitation between 30 and 45° latitude in the NH (solid black) and SH (solid gray), with the difference between the two shaded in brown when precipitation in the SH is below it in the NH (blue if vice versa). The sea surface temperature front is located near 60°; Vertical lines indicate the different grid nests in the NH. The transition between the 10-km and 20-km nests is clearly visible in the downstream precipitation.
Ten years of perpetual winter of meso-γ to deep-convection resolving simulations (5–10 km and 20 km grid spacing) of an idealized storm track were conducted using a grid nesting approach to study one of the most persistent climate model circulation biases known as the “too zonal and too equatorward” storm track bias and the added value of resolution at the meso-γ scale. It is the “storm resolving” resolution model developers target for the next generation of Earth system models. More specifically, the goal of this study is to investigate the high-resolution storm-track's location, orientation and intensity relative to a lower-resolution counterpart (20 km grid spacing) within the same simulation in a controllable and idealized environment. The hypothesis is that an enhanced poleward propagation of extratropical cyclones at storm resolving resolution reduces this decade old circulation bias in climate models and that the bias is interrelated with the tendency to underestimate rapid growth rates and ultimately results from insufficiently resolved diabatic processes.
Indeed, the simulation, which is conducted with the novel seamless weather and climate prediction model ICON, demonstrates the ability of the model to place a stronger storm track more poleward with less zonal extent once the meso-γ is resolved. Lagrangian tracking of surface cyclones indicates that a stronger poleward propagation is one cause of the improved storm track representation over the higher resolution hemisphere. It is noteworthy that the bias reduces despite the fact that the structural properties of individual clouds are not yet explicitly simulated at 5–10 km grid spacing. The enhanced poleward propagation is limited to the highest cyclone strength percentiles and occurring during the early and most intense growth phase of the life cycle of an extratropical cyclone. Since cyclone growth is inherently associated with poleward propagation (Besson et al., 2021; Gilet et al., 2009), it is not too surprising that growth rates near the SST front in the Northern Hemisphere also exceed those in the Southern Hemisphere, at least at the upper end of the distribution. Since precipitation rates downstream of the SST front are increased in the higher resolution hemisphere, there is evidence that one important underlying cause of both the stronger growth rates and the enhanced poleward spreading is increased diabatic heat release, especially for the very strong cases. In this respect, the “too zonal and too equatorward” bias is actually closely related to the underestimation of the frequency of high-intensity growth in many climate models (Priestley et al., 2020; Seiler & Zwiers, 2015), and thus all three biases are associated with insufficiently resolved diabatic processes. The bias is thus best summarized as “too zonal, too equatorward, too weak” and at least partly resulting from too small diabatic-growth contributions in addition to low-resolution SSTs (Lee et al., 2018; Woollings et al., 2010) and orography (Davini et al., 2022; Kanehama et al., 2019). Finally, it is worth emphasizing that the tropopause sharpness, as measured by the PV gradient, also benefits from higher resolution and a better storm track representation is thus to be expected even in a hypothetically dry atmosphere (Gray et al., 2014). In a moist atmosphere, the PV gradient is strongly modified by diabatic processes within extratropical cyclones (Gray et al., 2014; Pomroy & Thorpe, 2000; Schemm et al., 2013), hence, pointing again toward the importance of accurately represented diabatic processes. Future studies thus could conduct similar experiments but with coupled global climate models or steadily increase the realism of the setup by adding orography. Quantifying the performance of ICON in simulating properties of the North Atlantic storm strack in coupled mode at similar high-resolution, which has yet to be determined, is another reasonable next step.
Overall, the results of this study provide a promising look into the future capabilities of next-generation climate models to simulation key properties of the atmospheric circulation and regional climate patterns in midlatitudes. On regional scales, much of the future changes in the hydrological cycle are driven by change in the dynamics rather than thermodynamic properties of the atmosphere (Shepherd, 2014). During the past years, model development has focused intensively on convection-resolving resolution, arguably an important goal, while the meso-scale has never been appreciated as a desirable modeling target. However, the meso-γ or storm-resolving scale is, as shown here, itself a desirable goal in particular for the tilt of the storm track. A large fraction of high-impact weather events in the mid-latitudes are related to the dynamics of extratropical cyclones that steer precipitation and strong winds. Therefore, a more accurate simulation of their internal adiabatic and diabatic dynamics translates directly into reduced uncertainties in future projections of high-impact weather events and regional climate patterns, especially in regions located at the end and edges of major storm tracks such as Europe.
Figure A1 is similar to Figure 4 but without the removal of the zonal mean. It emphasizes the zonally-symmetric component of the EKE response, while Figure 4 emphasizes the zonally-asymmetric component of the response. The zonally-symmetric component pertains to the reduced zonal extension of the storm track in the high-resolution hemisphere; the zonally-asymmetric component pertains to the enhanced poleward propagation in the high-resolution hemisphere downstream of the SST front.
The author would like to thank two anonymous reviewers and Bjorn Steven for their detailed and constructive feedbacks, which helped to improve the manuscript. The Center for Climate Systems Modeling (C2SM) at ETH Zurich is acknowledged for providing technical ICON support. All simulations were carried out at ETH's Scientific and High Performance cluster EULER. The author is supported by funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program (grant agreement no. 848698).
Conflict of InterestThe authors declare no conflicts of interest relevant to this study.
Data Availability StatementThe used ICON run scripts and processed data underlying this study are made available via ETH Zurich's Research-Collection repository via the following
The originally published version of this article contained several typographical errors. In the Abstract, the sentence beginning “To demonstrate this . . . examined using the seamless weather and climate prediction model ICOsahedral Non-hydrostatic” has been corrected to “examined using the seamless weather and climate ICOsahedral Non-hydrostatic (ICON) model.” In the first paragraph of Section 2, the sentence beginning “The Icosahedral Nonhydrostatic Weather and Climate Model ICOsahedral Non-hydrostatic (ICON) (Zängl et al., 2015) is the new seamless weather and climate prediction model of at MeteoSwiss” has been corrected to “The ICOsahedral Non-hydrostatic weather and climate model (ICON) (Zängl et al., 2015) is the new seamless weather and climate prediction model at MeteoSwiss.” In the caption for Figure 4, the sentence beginning “Additionally shown . . . 500-hPa geopotential height at (red contours)” has been corrected to “500-hPa geopotential (red contours; 200, 225 and 250 m2 s−2).” These errors have been corrected, and this may be considered the authoritative version of record.
© 2023. This work is published under http://creativecommons.org/licenses/by-nc/4.0/ (the “License”). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.