As the global climate warms, the frequency and intensity of extreme events have increased, and events are occurring that are outwith the observational record . Examining past climates can help the understanding of the behaviour and response of a warmer climate system to different forcings, such as changes to Carbon Dioxide (CO2) and ice sheet configurations.

One time period well suited to this is the Late Pliocene, specifically Marine Isotope Stage KM5c, 3.205 Ma (within the mid-Piacenzian Warm Period) . During this time, both orbital forcing and CO2 levels (around 400 ppm) were similar to present day values . Reconstructing CO2 is challenging, so there is an uncertainty on this value, with many strands of proxy evidence. From Boron isotope analysis, report that CO2 values during the KM5c interval to be 391-29+30ppm with 95 % confidence. A value of 400 ppm was chosen for some climate modelling studies to account for the forcing derived from other greenhouse gases that have no proxy reconstructions available . Both the Greenland and the Antarctic ice sheets were reduced in size , leading to higher sea levels . Palaeogeography differs from modern day, including the closure of Northern Hemisphere gateways such as the Bering Strait and the Canadian Archipelago .

Through the Pliocene Model Intercomparison Project (PlioMIP), Late Pliocene climate was simulated by a number of climate models. PlioMIP phase 2 saw 17 climate models complete experiments to understand the behaviour of Late Pliocene climate, and the large-scale features of these simulations are described in , with the results of HadGEM3 presented separately in . The ensemble multi model mean (MMM) temperature increase from pre-industrial was 3.2 °C with a global increase in precipitation of 7 % . Some other significant features of Late Pliocene climate include polar amplification , a strengthened Atlantic Meridional Overturning Circulation (AMOC) and lower El-Niño variability .

To date, the majority of studies focused on the Late Pliocene have looked at changes to the mean state. As the Late Pliocene gains momentum as a potential analogue for the future , and to improve our understanding of the climate during this time, it becomes important to consider higher-frequency variability. Considering variability and extremes within a palaeoclimate context could provide important considerations for the interpretation of both proxy and climate model data. In climate projections, distributions of temperature and rainfall shift, with more extremes expected . It can be assumed that these distributions were also skewed, compared to pre-industrial, in past warmer climates like the Late Pliocene.

Obtaining well-dated proxy data at the temporal resolution needed to perform analysis on extreme events is challenging. It is therefore important to understand the drivers of extreme events and examine how they change, using climate models, before thinking about how they could occur in the proxy record. Furthermore, looking at extreme events in simulations of a warm climate where the mean state has been validated against proxy data , can help us understand how climate models represent large scale climate features. Understanding the large scale features, and how well a model captures them, is a necessary first step to understanding the variability, extremes and their drivers seen in the climate system.

One key feature in the mid-latitudes that can impact extreme events is the jet stream. Jet streams are fast-flowing, narrow bands of air in the upper troposphere. In each hemisphere, there are two main jet streams, the polar jet which forms at the polar front, and the sub-tropical jet which forms at the upper boundary of the Hadley Cell. These jet streams form in response to meridional temperature gradients and the Coriolis Force .

As observations and climate projections show an amplified rate of warming in Arctic surface temperatures , the surface meridional temperature gradient is reduced. This would suggest an equator-ward shift in the jet stream. Despite this surface change, the jet has been observed to be shifting polewards and becoming more wavy . link this trend to an increase in meridional temperature gradient in the upper troposphere, with changes to the tropopause height also being important for jet stream dynamics .

Many other factors can impact jet stream behaviour both globally and regionally. For example, the topography of the Greenland Ice Sheet can influence the North Atlantic jet . Sea ice cover has also been related to changes in mid-latitude circulation. Decreasing sea ice cover is linked to an equatorward shift in the jet stream in future projections, which could inhibit the poleward shift of the jet stream in future climate . Feedbacks in the climate system from sea ice loss are also an important factor to consider when thinking about Arctic influence on the jet stream. Observational studies show links between sea ice loss and changes to cold season weather in the mid-latitudes , although short observational records make it difficult to draw statistically significant conclusions from them. The same connection is debated in modelling studies with some studies finding no link and other studies finding weak, but significant links .

Shifts in the jet stream may lead to changes in the occurrence of extreme events in the mid-latitudes. Jet waviness has been linked to changes in extreme weather, although this is regionally dependent . This is a similar result to a study finding that jet stream tilt produces different regional responses over the European region . The relationship between the jet and mid-latitude winter extremes can also be related to atmospheric blocking , with atmospheric blocking being associated with Rossby wave breaking .

It is clear that no one change in the climate system is solely responsible for changes in jet stream behaviour, and is in response to a combination of forcings, which palaeoclimatology may be well placed to investigate.

To date, a few studies have investigated the behaviour of the jet stream during the Late Pliocene. looked at mean state changes to the westerly winds across the mid-Piacenzian Warm Period with data from PlioMIP1 and found that the westerly winds were displaced poleward compared with the pre-industrial. A difference was noted in the response of the atmosphere-only and coupled ocean–atmosphere models, with ocean–atmosphere models exhibiting a smaller polar shift compared to the atmosphere only models. This highlights the importance of ocean heat transport and sea ice dynamics on mid-latitude circulation. As PlioMIP2 provided new boundary conditions, including the closure of Arctic Ocean gateways, leading to a better data-model comparison in the higher latitudes , the PlioMIP2 ensemble will likely provide more insights into circulation changes in the Late Pliocene, relative to PlioMIP1.

Work has also been carried out looking at the drivers of jet stream change and variability. Using CCSM4-UoT it was noted that CO2 had little impact on changes to the jet stream and that ice sheet and orography changes were more significant in the generation of a wave train , although the forcing decomposition employed was not full as suggested by , that may neglect nonlinearities. also investigated jet stream variability in the North Pacific and found that in the Pliocene the jet stream was weaker and more variable in position due to increased Rossby wave breaking. It was also found that non-CO2 boundary conditions created the majority of the differences. Both of these studies only used one climate model in their analysis and may not capture the true change in the jet stream, as the representation of the jet stream varies between models due to model resolution and how drag is parameterised .

One challenge in examining changes to jet stream behaviour in the Late Pliocene is the lack of a direct proxy for jet stream behaviour to perform data-model comparisons. However dust flux proxies can be used to reconstruct past atmospheric circulation e.g., and more proxy derived information about interannual variability in the Pliocene may be available in the future. For this study we note that mean Pliocene climate variables have been reconstructed from proxy data e.g., hence Pliocene model simulations have a level of data validation that is not possible for future studies. This means that understanding modelled jet stream behaviour in the Pliocene can indicate how the jet stream may have responded to a real world warm climate, and can contribute to our overall understanding of the jet stream and how it is modelled.

This study aims to examine Late Pliocene jet stream change, in the mean state and monthly variability, to understand more about the climate system from a multi-model perspective. We also present new HadCM3 simulations with an aim to understand the drivers of the change. This will provide a perspective on the usefulness of the Late Pliocene as a past analogue for future jet stream variability. Details of the simulations and analysis techniques used are in Sect. . Section  contains the results and an interpretation of the mean state jet, the variability in the jet stream and explores the roles of different forcing to the changes observed. In Sect. , we then provide a perspective on how future studies can build upon this work, enabling a deeper understanding on Late Pliocene climate, and how that can help inform future changes in the climate system.

For the analysis, we take the final 100 years of each simulation and use the zonal wind speed and temperature fields at different pressure levels in the atmosphere. The model data was bi-linearly regridded to a 1° by 1° grid to allow for comparison between models and to construct a MMM.

Here we define the jet stream using the maximum zonal wind speeds. Following the methods employed by , the latitude of the jet stream is defined as the latitude of maximum zonally averaged zonal wind, and the speed is taken as the maximum speed of the zonal wind. Although taking a single maximum may remove any zonal variability in position, it is sufficient for this study examining monthly averages. As in , the zonal wind is examined on three pressure levels (850, 500 and 200 hPa). The analysis of the variability is concentrated at the 200 hPa level as all the models studies present data on this level, and the impact of the vertical movement of the jet is minimal (Fig. S1 in the Supplement). The latitude of maximum speed is found by performing a cubic spline interpolation around the points of maximum speed to achieve a latitude to 0.1°. As the jet stream and its impacts are regionally dependent, we study North Pacific and the North Atlantic separately to improve how. The North Atlantic region is defined as 15–75° N, 60° W–0° E following . We define the North Pacific region as 15–75° N, 160–220° E to be consistent with previous Pliocene studies of North Pacific jet stream variability . To understand the mechanisms of change we use the method described in Sect.  on the zonal wind speeds and upper level temperature change. We focus on Northern Hemisphere winter (December, January and February, DJF) for this analysis as the jet stream is strongest during this time due to an increased meridional temperature gradient. Similar analysis could be repeated for the Southern Hemisphere, but is not done so here to allow for more focused analysis on one region.

To measure the variability of the jet stream latitude (referred to as jet stream variability), we calculate the standard deviation of the latitude of maximum wind speed in the month of January at 200 hPa. Focusing on one pressure level will not account for the vertical displacement of the jet stream, but this vertical displacement was found to have little change on the results in HadCM3 (Fig. S1). As the surrounding pressure levels are not common to all the model output we chose to focus on one level to include as many models as possible in the variability analysis. Only one month was studied for the variability metrics to not smooth any changes in latitude across a season. The variability analysis is focused on the North Pacific, as a persistent wave breaking pattern in the North Atlantic and the presence of a double peak in some of the models complicated the analysis.

Unlike previous studies on the Late Pliocene jet stream that consider global zonal averages , the behaviour of the zonal winds is studied separately for different regions. In the North Pacific region, the pre-industrial control MMM captures the reanalysis well (Fig. ). In the simulations, the zonal wind speeds are reduced in the Late Pliocene compared to the pre-industrial with a slight poleward shift seen in the MMM (Fig. ). The MMM poleward shift of the latitude of maximum zonal wind speed is 0.6, 1.8 and 2.0° at 200, 500 and 850 hPa respectively. This is in line with the expected response due to weakening of the meridional temperature gradient reported in the PlioMIP2 ensemble .

Although the MMM shifts, the range of model response can not be ignored. As demonstrated in Figs.  and , the models feature a large range of possible jet stream changes. Eight of the models agree with a poleward shift at 200 hPa and 13 of the 15 models agree with a poleward shift in at least one pressure level (Fig. ). Each model in the ensemble captures the general feature of the jet stream when comparing the pre-industrial control simulation against ERA5 reanalysis data, although some models are a closer fit than others (Fig. S2).

In the North Atlantic region, a double peak in the zonal wind speeds can be seen, likely due to the presence of Rossby wave breaking (Fig. d), which can also be seen in Fig. . None of the models, with the exception of HadGEM3, capture the reanalysis well at this level (Fig. S3), with the MMM zoanl wind speed being higher than the reanalysis. This is expected as models often underperform in the North Atlantic region . When considering the MMM change, a poleward shift and a slight strengthening in the winds are observed. After removing the data points at 200 hPa from CCSM4-UoT and CESM1.2 (due to large shifts in the maximum from increased wave breaking), the North Atlantic jet stream shifts by 1.9, 0.4 and 1.1° at 200, 500 and 850 hPa respectively. There is also lower model spread with only two models exhibiting a large equatorward shift. The MMM, however, is driven by some models with larger changes, such as HadCM3 and MRI-CGCM2.3 (Fig. S3).

Considering the spatial variations, the change is seen as a strengthening on the poleward flank and a weakening at the equatorward flank at both the 200 and 850 hPa level (Fig. ). The MMM difference in Fig.  is muted compared to the differences seen when examining the individual model differences. This may be due to differences in the placement of the jet stream and the magnitude of poleward shift cancelling in the MMM (Figs. S4 and S5).

In the North Pacific, the weaker and poleward shifted westerly winds at 850 hPa are consistent with proxy data. examine changes in the strength and position of the jet stream using dust flux records in the warmer Pliocene climate compared to cooler glacial climates. They report that in warmer climates, like the Late Pliocene, the westerly winds are poleward and weaker compared to cooler climates matching what is found in this modelling study. More work is needed to fully assess the dust flux records in different ocean basins to understand the full proxy signal. However, it is difficult to obtain data at a high enough temporal resolution to support this analysis. It is also possible that changes in dust flux and aerosols may lead to changes in atmospheric circulation if there is a change in land surface type although argue that atmospheric circulation is the dominant driving factor of the observed flux changes.

There are several reasons for differences between models. First is the resolution of the model. Models with grater spatial resolution are able to capture more complex ocean–atmosphere interactions that play a large role in the representation of atmospheric circulation . Models with finer spatial resolution will also include a better representation of orographic features affecting both ocean and atmospheric circulation. Some models are obvious outliers from the reanalysis data, such as IPSLCM5A, IPSLCM5A2, COSMOS, HadCM3 and NorESM-L (Figs. S2 and S3). These are the models among the lowest spatial resolution of PlioMIP2 (Table ). This indicates that model resolution is an important factor to consider when examining changes in the jet stream. There is also added uncertainty in the representation of the jet stream within the ERA5 reanalysis, as observations higher in the atmosphere are sparse. Some models also better capture proxy reconstructions of sea surface temperatures, for example, CCSM4-UoT, CESM1.2, IPSLCM6A and MIROCm4m . These models may also be able to provide a better representation of jet stream changes during the Pliocene as they capture the general climate of the Pliocene well.

To examine possible causes of the change to the mean state jet, the change in jet stream latitude is compared to the ratio of the earth system sensitivity (ESS) and the equilibrium climate sensitivity (ECS) to assess the relative contribution of CO2 and non-CO2 forcings. However, no strong or significant relationship was detected (Fig. S6) suggesting that drivers of the change in the jet may not be consistent across the range of models. To fully understand the drivers of jet stream change, each model will need to be individually examined.

To understand which of the Pliocene boundary conditions contribute to the change in the mean state jet stream we use the HadCM3 forcing factorisation experiments described in Sect. . HadCM3 displays a small (0.7°) equator ward displaced jet in DJF at 200 hPa, but it does show poleward movement of 5.3° and 5.2° at 500 and 850 hPa respectively, and a weaker jet. Although considering MMMs is important, there is limited data to perform full forcing decomposition with the full range of the PlioMIP2 models. It should also be noted that it is impossible to fully disentangle the contribution of ice sheets and orography because the two are inherently interlinked per the boundary condition design in PlioMIP2. The ice sheet itself provides a change in orography, which is especially important for controlling changes in the North Atlantic jet stream with the position of Greenland being important for the pattern of the jet stream in that region .

The HadCM3 DJF surface temperature forcing decomposition can be seen in Fig. S7. This highlights the regional influences of changes to temperature from ice sheets and orography with the largest changes occurring where the changes to ice sheets and land sea mask were made. The Late Pliocene experiment has a stronger AMOC compared to the pre-industrial simulation, consistent with other models in the PlioMIP2 ensemble . The experiments employing Late Pliocene orography (including land–sea mask changes) show a stronger AMOC than the experiments with pre-industrial orography (Fig. S8). This is likely due to changes in the Northern Hemisphere ocean gateways. This is consistent with previous studies, which relate the increase in Late Pliocene AMOC to the closure of the Arctic Ocean gateways . This enhanced AMOC, due to ocean gateway changes, is one contributor to the reduced meridional temperature gradient, as more heat is transported northward. In the forcing factorisation experiments, vegetation changes are included in the orography boundary condition. A change in vegetation can also lead to a large change in temperature, due to different plant types having different impacts on temperature through albedo and evapotranspiration . Studies examining other time periods have reported that changes in vegetation in response to climate reinforce the signal, especially at high latitudes .

Figure  shows the total change in zonal wind speed between the pre-industrial and the Late Pliocene simulations in HadCM3 and the contribution of CO2, ice sheets and orography. In the North Pacific, the change in wind speed due to CO2 is small (and opposite in sign to the other forcings), with the largest contribution being orography (including land–sea mask and vegetation) and ice sheets following as the second leading cause in the change. Following the discussion about AMOC and vegetation impacts, this change could be due to changes to meridional temperature gradients. Over the North Atlantic, changes to the ice sheets have a larger effect, which is expected due to changes in the Greenland Ice sheet.

The leading contribution from orography is also apparent when considering the North Pacific vertical profile of the zonal wind speeds (Fig. ). The contributions from CO2 and ice sheets are also small here. Examining the change in the North Pacific vertical profiles of temperature (Fig. a), the decreased meridional surface temperature gradient is seen and is attributed to a change in the CO2. Orography contributes to warming in the mid-latitudes troposphere and cooling aloft in the high latitudes. This pattern of warming may be contributing to changes in the geopotential height (Fig. S9) with CO2 increasing the height globally. However, ice sheet and orography changes produce strong spatial variation, especially over the Allusion low region, consistent with previous studies . It is unclear what is causing this warming over the North Pacific region. There is a pattern of warming from orography in the surface temperature (Fig. S7), although it is not known if this is due to a change in ocean circulation or a response to the changes in atmospheric circulation. It is also acknowledged that the North Pacific jet stream is impacted by other areas of the world, including changes in the Tropics .

Two other models, CCSM4-UoT and COSMOS also provide the simulations needed for a full forcing factorisation. The results of the vertical profiles of the North Pacific zonal wind speeds are found in Figs. S10 and S11. In COSMOS (Fig. S10), each boundary condition change contributes small amounts to the total change and in CCSM4-UoT the orography contribution is similar to HadCM3, but also contains a pattern from the ice sheets resembling the orography pattern. The main differences in the setup of these experiments between models are the implementation of different land sea masks. COSMOS uses the standard boundary conditions , and CCSM4-UoT uses the enhanced boundary conditions but retains a Late Pliocene land sea mask in the Ei experiment. This suggested the change in the wind due to orography is driven by changes to the land sea mask used. To fully understand this an analysis of ocean circulation in each model and experiment would be needed, but is beyond the scope of this study.

Although jet stream variability has been investigated in similar manner before within CCSM4-Utrecht only , as discussed in Sect. , there are large differences in the mean state shift across the ensemble. This range in responses for the mean state may also suggest differences in simulated variability across the ensemble, highlighting the importance to examine variability in a multi-model setting. To assess the model performance against ERA5 reanalysis data we compare the mean latitude and variability in position in the pre-industrial simulations over the North Pacific against reanalysis data (Fig. ). The majority of models capture the jet latitude variability but there is a larger spread in the mean jet latitude. Models that are furtherest away from the reanalysis tend to have lower spatial resolutions (COSMOS, MRI-CGCM2.3 and MIROC4 m). The impact of multi-decadal variability cannot be ignored in this assessment and some longer modes of variability may be influencing the 100 year mean.

The response to Late Pliocene boundary conditions across models is diverse (Fig. ). The difference between models is larger than the difference caused by a change from pre-industrial to Late Pliocene boundary conditions, indicating that care should be taken in the interpretation of these results as the model uncertainty is large. However, some models may be better at capturing the change in the jet stream variability than others. For example, HadGEM3 and MRI-CGCM2.4 have an unchanged land–sea mask. Since we expect the differences observed in the jet to be partially caused by land–sea mask changes, we do not expect a large change in the jet stream in these models, as is seen in Fig. . The impact of model resolution is also clear from Fig.  with the lower resolution models displaying a more diffuse maximum (COMSOS, MRI-CGCM2.3 and NorEMS-L).

The jet stream latitude variability (σ(θ) in Fig. ) is higher in the Late Pliocene than in pre-industrial experiments in most models. However, a lot of this change is only small (Fig. ). HadCM3, CCSM4-UoT and MIROC4 m all show a larger change in the variability of the jet stream. Two models that show an increased speed and a decreased variability, COSMOS and NorESM-L, are the two models with the lowest resolution out of the PlioMIP2 ensemble. This reinforces that resolution does matter when considering jet stream dynamics. MRI-CGCM2.3 also shows an increase in speed and a decrease in variability which could be related to the unchanged land–sea mask in the Late Pliocene experiment in this model. The models with a Late Pliocene land–sea mask and higher spatial resolution indicate that the jet stream is weaker and more variable in the Late Pliocene. The change in the variability of the latitudinal position of the jet stream could be related to increased wave breaking in the Late Pliocene .