Main

The topic of climate dynamics encapsulates the dynamical interaction of the atmosphere and ocean with each other as well as with other components of the Earth system, playing a leading-order role in regional climate, both in the response to external forcing and the background internal natural variability. As has been recognized for many decades, the role of the ocean and air–sea interaction is at the core of climate variability and change on seasonal to centennial timescales1.

Much research on climate dynamics has focused on statistical descriptions of variability and change. Empirical orthogonal function (EOF) analysis, or other techniques, are used to define indices of modes of variability, such as the North Atlantic Oscillation (NAO), the Pacific North American (PNA) pattern and the Southern and Northern Annular Modes (SAM and NAM). These descriptions offer compact ways of describing regional climate and its impacts but are less useful in providing insight into the dynamical and physical processes that drive variability and change. More importantly, they lack the ability to provide the basis for prediction beyond the use of emprical methods. We must advance from using simple descriptive indicators to quantitative theories that will lead to more reliable predictions and projections of climate to inform adaptation.

Variations in weather are controlled by large-scale dynamical processes in the atmosphere, for example extratropical storms, blocking, jet streams, and tropical waves, coupled with atmospheric convection, tropical convergence zones and monsoon flows. On timescales of days to a week, many of these phenomena can be predicted using initialized numerical models and there is a basic understanding of the dynamical processes involved in their variations (for example, storms are carried by westerly advection and the Rossby wave mechanism, and grow on horizontal temperature gradients). Weather forecasting up to the medium range is a relatively mature area of meteorology2.

On climate timescales, such as projections for the end of the century, we look to quantify the response of the climate system to external forcing (mainly increasing greenhouse gas concentrations), and measure this against the unpredictable background noise of natural internal climate variability. For more near-term or decadal predictions, it is both the forced response and the predictable component of the natural internal variability that is sought3. We make the distinction here between projections that are conditioned with a particular scenario of future emissions or concentrations, and predictions that rely on initial conditions and where the scenario is less important.

Both prediction and projection involve the understanding of dynamical motions of the fast-moving atmosphere, modulated by the slowly varying ocean, ice and land surface signals, or external factors, including both natural and anthropogenic forcing of climate. The ocean is a major player in climate dynamics — occupying seven tenths of the Earth’s surface. It is the main source of atmospheric water vapour, with far greater capacity to store heat than the atmosphere, and is dynamically varying in terms of its circulation and water properties. We seek to understand, and predict and project statistical measures of the behaviour of dynamical phenomena, such as the average and variability of the position of the jet stream and Inter-Tropical Convergence Zone (ITCZ), or the intensity and structural organization of storm tracks, and associated strength and frequency of storm events.

Here there is a dynamical gap in our understanding. While we have conceptual models of how weather systems form and can predict their evolution over days to weeks, we do not have theories that can adequately explain the reasons for an extreme cold or warm, or wet or dry, winter at continental scales. More importantly, we do not have the ability to credibly predict such states. Likewise, we can build and run complex models of the Earth system, but we do not have adequate enough understanding of the processes and mechanisms to be able to quantitatively evaluate the predictions and projections they produce, or to understand why different models give different answers.

For example, much of the understanding we have gained on spatial patterns of climate change has resulted from adopting an energetic framework in which radiative forcing is separated from radiative feedbacks in the climate system4. This paradigm has led to progress in quantifying feedbacks associated with, for instance, surface albedo, water vapour and atmospheric lapse rate and clouds. Despite shifts to regional approaches5, it is difficult to account for changes in horizontal transports of energy associated with dynamical processes using this approach. We require new ways of thinking.

Climate dynamics has traditionally been developed in studies of interannual variability. The coupled ocean–atmospheric perspective laid the foundation of seasonal forecasts routinely issued today. However, notwithstanding the many challenges still faced by the seasonal forecasting community, our hypothesis is that climate dynamics is insufficiently applied in studies of near-term and regional anthropogenic climate change in favour of global mean warming, climate feedback, or other robust thermodynamic mechanisms. Atmospheric circulation changes have been identified as the leading source of uncertainty in regional climate predictions on decadal timescales3 and projections on longer timescales6,7. The need to understand and reduce uncertainties in regional climate projections represents both a challenge and an opportunity to extend climate dynamics.

In this Perspective we identify a number of challenges in climate dynamics and suggest ways in which progress may be made. While not an exhaustive list, we identify the challenges as priority areas of research for the climate dynamics community.

Frontline problems in climate dynamics

The following three problems have climate dynamics at their core.

Response to external forcing of mid-latitude jets, storms and blocking

Assessment of the impact of forced climate change on mid-latitude weather systems indicates low confidence in changes in future projections for the end of the century8. There is also limited success in transferring skill in predicting extratropical ocean heat content to continental regions3. Our ability to quantify the sensitivity of storm tracks to external factors such as greenhouse gas increase or natural fluctuations such as ocean heat content anomalies and their role in modifying surface heat exchange is limited by a lack of quantitative theory of how storm tracks respond to changing boundary conditions on seasonal timescales and beyond, and by large random internal variability of the atmosphere.

Many complex climate models are now only just reaching a stage in which storms and storm tracks are simulated in the present day with reasonable fidelity. The structure of Atlantic and European winter blocking is also now represented reasonably well by some models, albeit with slightly reduced frequency in comparison with observations9. Additionally, excessive precipitation associated with storms is often found over the ocean, leading to inadequate precipitation extremes over land10 — a problem that may be alleviated by increasing model resolution11; see ‘High resolution coupled modelling’.

Under climate change there are competing influences on Northern Hemisphere storms and blocking. Polar amplification and sea-ice loss would tend to weaken the low-level baroclinicity (the energy source for storms) and ocean circulation changes associated with western boundary currents may exert a regional influence on temperature gradients12. Warming of the atmosphere in the tropics and subtropics enhances mid-latitude baroclinicity at upper levels, whereas dynamical warming of the stratosphere due to the increase in the Brewer–Dobson circulation counteracts the enhancement of upper-level baroclinicity13,14,15. Vertical stability of the atmosphere is increased but latent heat release in storms is enhanced due to enhanced moisture content. The thermodynamic component of moisture transport may rise under climate change due to enhanced column water vapour, but the total transport may decrease if the dynamical transport declines enough16. Regional complications are also likely.

Models tend to show only modest changes in storm tracks in climate change scenarios in the Northern Hemisphere, although a feature that is not fully understood is the extension of the Atlantic storm track over the UK and associated cyclones propagating into northern Europe17. Models generally project a reduction and eastward shift in blocking occurrence that appears to result from mean state changes18. While thermodynamic aspects of storms and storm tracks in the Northern Hemisphere seem relatively robust across models, there is little confidence in their projected changes in dynamical aspects19.

In the Southern Hemisphere, a poleward shift of a few degrees latitude is observed as low-level baroclinicity intensifies with reduced surface warming over the Southern Ocean20, albeit with some uncertainty in observations and reanalyses. Models in which the latitude of the Southern Hemisphere westerlies is biased towards the equator simulate the largest poleward shift as a result of climate change21. However, polar ozone recovery may oppose changes induced by greenhouse gas increases in the near term.

The impact of reductions in Arctic sea ice on storm tracks and blocking has been a topic of considerable recent research22,23, revealing a lack of quantitative understanding of the contribution of sea-ice loss. Models that simulate larger Arctic amplification under climate change tend to depict shifts towards the equator of the jet, whereas the jet shifts poleward in those with smaller Arctic amplification24,25. Theories of how sea-ice retreat might influence storms and blocking are incomplete26. One particular question is how an upstream perturbation to sea-ice and surface heat fluxes might influence a downstream storm development.

Basin-to-basin and tropical–extratropical teleconnections

Much research in climate dynamics has been focussed on understanding the dynamics of basin-scale modes of variability such as the El Niño Southern Oscillation (ENSO), the Pacific Decadal Oscillation and Interdecadal Pacific Oscillation (PDO and IPO) and Atlantic Multi-decadal Variability (AMV). While these are far from solved problems, it becomes increasingly apparent that interactions between ocean basins and interactions between the tropics and the extratropics (including polar regions — see ‘Response to external forcing of mid-latitude jets, storms and blocking’) can drive global-scale variability.

The global warming ‘hiatus’ provides an example of a climate event potentially related to inter-basin teleconnections. While decadal climate variations are expected, the magnitude of the recent event was unforeseen. A decadal period of intensified trade winds in the Pacific and cooler sea surface temperatures (SSTs) has been identified as a leading candidate mechanism for the global slowdown in warming27,28,29. Forcing from a warm tropical Atlantic has been highlighted as a possible cause30,31,32, invoking feedbacks between SSTs and winds in the Pacific (Fig. 1). Other studies have noted interactions between the Pacific and the Indian Ocean as being important in affecting not only the strength of the trade winds over the tropical Pacific but also associated sea-level variations33. This, of course, raises the issue of the cause of decadal and multi-decadal variability in remote ocean basins that teleconnect to the Pacific. In the Atlantic, the large-scale ocean overturning is a prime candidate, although there has been some debate in the literature about a potential role for aerosols and simple mixed-layer ocean dynamics34,35,36,37. Observational estimates of surface turbulent heat fluxes seem to indicate that the atmosphere is responding to SST anomalies on timescales longer than 10 years, pointing to oceanic dynamical processes as the ultimate cause of the hiatus38. Other authors have noted the role of internally generated variability in the atmosphere39,40. It is not clear if models can capture the magnitude of Atlantic–Pacific connections as deduced from observations41.

Fig. 1: Sea surface temperature trends from observations for the period 1979–2012 indicating the concept of inter-ocean-basin teleconnections.
figure 1

If SSTs are relaxed to observations in the tropical and subtropical Atlantic in a model, the trade winds in the Pacific increase, resulting in a subsequent cooling trend there31,32. Nevertheless, current coupled climate models tend to underestimate the magnitude of the coupling41.

Full size image

Tropics to high-latitude teleconnections are also evident. The atmospheric circulation around Antarctica shows trends in the observations that are linked to a complicated pattern of reorganisation of sea ice, particularly around the minimum in austral Spring. Some regions show increasing sea-ice concentration trends and some show decreases, linked to variations in local winds42. This is in stark contrast to the Arctic, which displays a clear downward trend across the basin in boreal Autumn. The expansion of Antarctic sea ice in the early 2000s has been linked to trends in tropical SST anomalies43,44,45,46 with a notable role for both the IPO47 and Atlantic anomalies. It is also suggested that a hemispheric teleconnection exists between AMV and Antarctic sea-ice trends48. In the Southern Hemisphere, large-scale extratropical forcing has been shown to influence the strength of the South Atlantic Convergence Zone49.

An outstanding problem in quantifying the forced climate response to anthropogenic greenhouse gases is determining the spatial pattern of north–south asymmetry in warming: the Northern Hemisphere warms more than the Southern Hemisphere in models50. A simple explanation, based on the greater Northern Hemisphere landmass, is inadequate and it seems that the ability of the Southern Ocean to more efficiently uptake heat is important51. Nevertheless, the Southern Ocean is a region of known biases in climate models, with SSTs being generally too warm as a result of too much incoming shortwave (SW) radiation reaching the surface52. The relative energy balance of the two hemispheres, and the related cross-equatorial energy transport, is gaining prominence as a key determinate of many large-scale dynamical features53. Subtle changes in heat transport may manifest as quite large changes in, for instance, monsoon flows and associated rains53.

Developing predictive theories of climate dynamics

Simple theoretical models have been instrumental in understanding, for example, the basic dynamics of the ENSO cycle54. These models have been extended to diagnose the behaviour of coupled climate models and may be regarded as process-based ways of evaluating models. Concepts such as quasi-geostrophic theory or the Lorenz energy cycle have also been used in the past as advanced diagnostics for understanding the behaviour of models and the real world. ‘Theories’, or ‘predictive theories’, in this sense relate to robust aspects of the dynamics of simple models that can also be found in more complex models. This is closely related to prediction in the forecasting sense but may be subtly different in the case of climate change. An example would be quantitative predictions of the poleward extent of the Hadley circulation at different levels of global warming.

The challenges in developing theories of regional climate change come in dealing with non-linear interactions between processes and diverse timescales, from days to decades. One definition of climate is that it is the cumulative effects of weather, hence any theory must account for the feedbacks between weather events and their modulators such as variations in SST or ocean heat content. The number of processes and non-linear interactions that can be represented in conceptual models, which are formulated in terms of only a handful of differential equations, is clearly limited. We need to develop ways to derive predictive theories of climate dynamics that include the interactions between key dynamical and physical processes. As stated above, statistical descriptions such as EOFs provide compact ways of representing some aspects of the climate system, but fall short in providing predictive theories55,56.

The mid-latitude eddy-driven jets provide an example. They are sustained by non-linear eddy momentum fluxes which are a consequence of the baroclinic instability of the jet itself. The eddy-driven jet in both hemispheres varies in position and strength but there is currently no quantitative theory that can predict the magnitude of such variations. It remains a considerable challenge to advance our understanding of climate dynamics involving non-linear interactions across a range of space and timescales.

Techniques and opportunities for making progress

The following techniques may be brought to bear on all the problems discussed above.

High-resolution coupled modelling

It is not obvious that continually increasing the resolution of climate models leads to more accurate predictions or projections of regional climate change, whatever the timescale. Better representation of features such as coastlines and mountains undoubtedly gives more regional information, but it is the large-scale drivers of regional change, a simple example being a change in the direction of the prevailing wind, which are the leading-order sources of uncertainty. In increasing resolution, we look to ‘unlock’ physical processes that are missing from low-resolution models, including interactions across multiple spatial scales, and to significantly reduce biases in the simulated present-day climate.

It has now become clear that anomalous conditions in western extratropical ocean basins can affect the atmospheric circulation. The zero-order effect of extratropical oceans is that of increasing the persistence of atmospheric anomalies through reduced heat-flux damping57. This process has shown to be important, for example, for increasing the predictability of surface temperature in south-eastern South America58. Moreover, recent studies have indicated that SST gradients and strong ocean-to-atmosphere heat and moisture fluxes associated with variations in western boundary currents (such as the Gulf Stream and Kuroshio) can have a significant local influence on atmospheric vertical velocities through sea–air energy exchanges, providing diabatic sources of heating and moistening of the troposphere (Fig. 2)59,60,61. A realistic Gulf Stream, accompanied by strong horizontal temperature gradient, is found to be important in producing realistic blocking and jet stream distribution in an atmospheric numerical model62,63. The mechanisms for this are likely to involve lower atmospheric meridional temperature gradients caused by strong SST gradients across the currents and/or latent heat release associated with the moisture supply from the currents64. Interactions between SST fronts, storms and storm tracks have been shown to impact storms in the Pacific65 and blocking events in the Atlantic62. A recent study suggests that mesoscale ocean–atmosphere coupling markedly affects ocean eddies and the Kuroshio Extension jet through eddy fluxes, with potential effects on large-scale storm tracks66. In higher-resolution models, coupling between the ocean and the atmosphere in the extratropics has the potential to influence how climate change affects storms and storm tracks, so as to alter our current understanding significantly.

Fig. 2: The influence of sharp SST gradients in the Gulf Stream on the hydrological cycle of individual storms and their rectification on the mean climate state.
figure 2

Figures are derived from atmosphere model simulations62, performed at 50 km horizontal resolution, in which the Gulf Stream is represented at equivalent 50 km resolution (CONTROL) and in which SST gradients are smoothed out (SMOOTH). a–c,The mean winter (November–March) precipitation in CONTROL (a), SMOOTH (b) and the difference (that is, CONTROL minus SMOOTH) (c). d–f, The mean winter evaporation rate is also shown for CONTROL (d), SMOOTH (e) and the difference (f). g–i, Cyclone composites the precipitation rate are shown for CONTROL (g), SMOOTH (h) and the difference (i). Climatological contours of SST in a, c, d, f, g and i are shown from CONTROL and in b, e and h for SMOOTH, with a contour interval of 3 °C and the 12 °C isotherm emboldened for reference. Composite SLP contours are shown in black for the cyclone composites in g and h, with a contour interval of 4 hPa and the 1,000 hPa isobar emboldened for reference. The cyclone composites were produced by selecting identifying peaks in a 6-hourly index of relative vorticity (at 850 hPa) averaged over the green box shown in g and h. Only events that are greater than 2 standard deviations above the mean and do not occur within 7 days of a larger event are included in the composite.

Full size image

In the tropics, higher horizontal resolution has been shown to improve the simulation of ENSO in terms of the amplitude, spatial pattern, and teleconnection patterns67. More frequent (sub-daily) coupling between the atmosphere and ocean also enhances ENSO amplitude68. Improvements come from the representation of small-scale tropical instability waves that have an atmospheric imprint, which rectifies on both the mean climate and the interannual variability. Projections of increasing frequency of extreme ENSO events69,70,71 are related to shifts in precipitation that may be related to changes in the mean climate72. Such mean changes may, in turn, be related to biases in mean SSTs in models73. Increasing resolution in coupled models is one way of testing such a hypothesis and improving our understanding of regional climate change in the tropics, including teleconnections. However, a ‘reliable’ projection of the impact of climate change on ENSO ultimately requires a model with much-reduced biases in the mean state and a representation of the ENSO cycle with the correct balance of positive amplifying feedbacks and negative damping feedbacks74. This depends not only on the resolved dynamics but also on the interaction with unresolved physical processes, such as atmospheric convection.

Conducting multi-decadal simulations with high-resolution models has long been recognised as a challenge. Now the community is on the brink of being able to routinely run coordinated experiments in both atmosphere-only, and crucially, in coupled atmosphere–ocean configurations with a horizontal resolution of 25 km in the atmosphere and 0.25° in the ocean75. These High Resolution Model Intercomparion Project (HighResMIP) experiments will provide a very useful resource for the climate dynamics community, especially the ability to compare dynamical processes at low and high resolutions76. The HighResMIP experimental design also attempts to isolate the impact of resolution by running complementary low-resolution experiments without re-tuning the models. While not providing simulations at the resolution at which parameterisation schemes such as those associated with atmospheric convection may be switched off, potentially leading to better simulation of, for instance, summer convective rainfall77,78, nor being eddy resolving in the ocean, they nevertheless present a significant improvement when compared to the resolution of the previous generation of climate models. However, high resolution is not a panacea to solve all problems in climate dynamics and experiments require careful design and analysis. Progress may require years, if not decades, of coordinated effort (see ‘Complex diagnostics and simplified models’).

Partial coupling and pacemaker experiments

While atmospheric models forced by SSTs have long provided evidence for the impact of the ocean on the atmosphere, their use in understanding how extratropical SST anomalies influence the extratropical atmosphere is limited. SST-forced experiments produce the wrong sign of SST heat-flux correlations on daily timescales57. The development of experiments in which SST anomalies are nudged towards observed SST anomalies in some regions but left free to evolve in others (so-called partial coupling or pacemaker experiments) has provided insights in understanding the role of the Pacific in the global warming hiatus28, 79,80. The oceanic component need not be dynamical as even mixed-layer partial coupling experiments have shown to be useful in elucidating controls on South American precipitation58 and connections from the Atlantic to the Pacific32. A further type of experiment may prescribe heat-flux convergence anomalies in mixed-layer ocean models to drive ocean–atmosphere heat exchange81,82. Pacemaker experiments may also be performed in which the surface winds are nudged towards observed values27,83.

While pacemaker experiments are starting to be used more in the study of natural decadal variations in climate84 — the background from which the forced climate change signal emerges — they have not been fully exploited in the study of the dynamics of forced climate change. One example could be in the understanding of the north-south asymmetry in the temperature response, described above (see ‘Basin-to-basin and tropical–extratropical teleconnections’). Hemispheric differences in radiation balance have been shown to be related to persistent biases in models such as the ‘double ITCZ’, which may impact projections of regional climate change85,86. Pacemaker experiments could be used to artificially correct such biases or to control the level of hemispheric asymmetry in the climate change signal.

Global models with regional coupling may also be employed in understanding future changes in teleconnections arising from natural modes of variability, for example, by specifying observed SSTs in the Pacific associated with ENSO on top of different patterns of mean SST change. The advantages over atmosphere-only simulations would be in simulating the coupled aspects of teleconnections and the advantage over using fully coupled models would be in partially correcting SST biases such as the extended equatorial Pacific cold tongue. Another problem that would be amenable to such an approach would be the role of polar amplification and sea-ice decline in modifying mid-latitude weather (see ‘Response to external forcing of mid-latitude jets, storms and blocking’).

These are just a few examples of where pacemaker experiments could be useful in understanding the dynamics of regional climate change. There are many other possibilities.

Complex diagnostics and simplified models

The profile of metrics is growing within the climate modelling community with efforts to coordinate software and provide portals to calculate metrics for use in model evaluation and climate projections87. Basic metrics evaluate emergent aspects of climate, such as the spatial distribution of core variables of temperature and precipitation. More process-based metrics, for example, those used to evaluate the strength of different processes/feedbacks in the ENSO cycle54, are also now routinely used. The next level of diagnostics should address dynamical aspects of climate. Examples include the use of eddy-mean flow diagnostics, Lagrangian feature tracking and concepts such as moist static energy88, the use of potential vorticity budgets to assess the trajectories of storms89, or assessment of features such as stationary waves90,91. Development of more complex dynamical metrics should be encouraged. Also, we need to find ways of using these metrics to better inform the likelihood of projections of future climate change seen in models. This is particularly important in cases where simple emergent constraints92 have not been found.

There is no doubt that complex climate models have revolutionized the study of weather and climate. However, models that seek to represent all the complexities of dynamics, physics and, increasingly, biogeochemical cycles, are often as difficult to understand as the real world itself. Simplified or ‘stripped down’ numerical models, part of the model hierarchy ranging from very simple to very complex formulations, are growing in use and have been applied in understanding climate dynamics93. Examples include dry dynamical atmosphere models with simple Newtonian cooling representing radiation, and aquaplanet simulations with more complete moist physical processes, clouds and ‘grey’ radiation (Fig. 3)94,95,96.

Fig. 3: Schematic indicating the concept of the ‘hierarchies of models’.
figure 3

a–d, Configurations may range from a simple aquaplanet design with either fixed SSTs or a simple mixed-layer ocean (a), through simplified continental configurations (b), atmosphere only (c) and coupled atmosphere–ocean configurations (d). Coupling allows for a better representation of how the atmosphere and ocean interact but may also result in biases in models, as can be seen in differences in the shaded SST field in the bottom left (from observations) and bottom right (HadGEM2-AO coupled model).

Full size image

Studies of the basic dynamics of planetary atmospheres may provide insight and allow us to develop our theoretical understanding of climate dynamics in the complex Earth. The poleward migration of the eddy-driven jet on theoretical ‘ball bearing’ planets (that is, no mountains, continents or variations in the land-surface) at different rotation rates allows us to understand the basic properties of the governing macroturbulent scales in the planetary atmospheres, which are more difficult to separate under Earth conditions97. Similarly, idealised models have been useful in understanding the response of the storm track to the increase of upper level baroclincity versus the decrease in lower level baroclincity during climate change98,99. The poleward propagation of storms has been shown to be controlled by both the upper-level potential vorticity anomaly and by diabatic processes89. Both these processes are predicted to be enhanced during global warming, leading to a stronger Southwest–Northeast tilt of the storms in the Northern Hemisphere and an overall poleward shift100.

Idealised models are also being used to reshape the understanding of tropical large-scale circulations. The classical view of the monsoon as a planetary-scale sea breeze circulation is inadequate. Rather, monsoons can be viewed as the excursions of tropical convergence zones over land101. Monsoon onset is usually rapid and not adequately explained by the classical theory. An aquaplanet simulation with a simple mixed-layer ocean has been used to advance our understanding of monsoon onset or the rapid ‘jump’ in the location of maximum precipitation88. Both remotely forced stationary waves and local processes (for example, latent heating and land–atmosphere interaction) can influence regional monsoon dynamics in terms of timing and strength. A long-standing problem in many climate models is the inability to produce enough precipitation over land in the major monsoon systems. Rainfall may preferentially occur over the ocean, for example in the South Asian monsoon102. While land-surface and SST errors may be important103, the coupling between atmospheric convection and the dynamical flow is clearly of leading-order importance in setting the mean rainfall and variability. Under climate change, global monsoon rainfall generally increases but that increase does not scale with global mean temperature change, as there is a compensation between a weakening circulation and increased column water vapour104.

More dynamics please

Whether the goals of the Paris Agreement of “keeping a global temperature rise this century well below 2 °C above pre-industrial levels and to pursue efforts to limit the temperature increase even further to 1.5 °C” are realised or not, adaptation to climate change is essential. For this we need the best information about future changes in regional climate, much of which is controlled by the dynamical behaviour of the atmosphere, the ocean and their mutual interaction, as well as interaction with other components of the climate system such as the land surface. This understanding can also lead to improved near-term climate predictions.

The challenge for the climate dynamics community is to produce this information by exploiting hierarchies of models, including the new generation of high-resolution models, by developing metrics to evaluate dynamical processes to explain projections, and to design new targeted model experiments to isolate dynamical drivers of change. However, perhaps the biggest challenge is to produce theories of regional climate change on a par with, for example, theories of baroclinic instability, that can be rigorously tested using both observations and models.

This Perspective highlights three frontline problems in climate dynamics in which the ocean plays a key role: (1) the response to external forcing of storms, blocks and jet streams; (2) ocean-basin to ocean-basin and tropical–extratropical teleconnections; and (3) the development of predictive theories of climate dynamics. Other problems, such as those involving interactions between the troposphere and stratosphere, are also urgent. We have also highlighted some new techniques and capacity in the use of climate models to address these problems. We recommend that the climate dynamics community exploit these opportunities.