Abrupt Changes in the Subpolar North Atlantic and Their Impact on the Climate of the British Isles

1 Abrupt changes in the subpolar North Atlanc and their impact on the climate of the Brish Isles 2 Mahew B Menary1,*, Laura C Jackson1, Richard A. Wood1, Richard A. Bes1,2 3 1. Met Office Hadley Centre, FitzRoy Road, Exeter, EX1 3PB, UK 4 2. Global Systems Instute, University of Exeter, North Park Road, Exeter, EX4 4QE 5 *Now at: European Centre for Medium-Range Weather Forecasts, Shinfield Park, Reading, RG2 6 9AX, UK 7 8 Abstract 9 There has been increasing interest in the possibility of abrupt climac changes in the North Atlanc 10 and their impacts on northwestern Europe. Here, we invesgate such abrupt changes in a large 11 ensemble of CMIP6 climate models. We define two potenally observable metrics based on subpolar 12 sea surface temperatures (SSTs) or mixed layer depths (MLDs), to explore the link between 13 temperature changes and convecon collapse. The two metrics yield similar numbers of abrupt 14 events but suggest that several types of abrupt event are possible. Abrupt MLD changes appear 15 related to ongoing warming. Abrupt SST changes mostly consist of decadal cooling followed by 16 warming, apparently related to coupled dynamics involving the NAO. Models with more realisc NAO 17 variability show more such events. However, several more persistent SST events are also found. Both 18 cooling and warming phases have important implicaons for impacts and adaptaon, parcularly 19 over the Brish Isles. 20 Plain language summary 21 There is increasing interest in sudden climate changes in the North Atlanc and how these changes 22 might affect Europe. This study uses a large set of recent climate models to see if these sudden 23 changes are likely. We use two measures related to 1) sea surface temperatures (SSTs) and 2) mixed 24 layer depths (MLDs) to study the connecon between surface temperatures and the collapse of 25 oceanic convecon. Both measures show similar numbers of sudden changes, but they don’t seem to 26 be related. Sudden MLD changes mostly seem to be connected to ongoing global warming. Sudden 27 SST changes mostly follow a paern of cooling and warming over around 20 years, apparently linked 28 to the North Atlanc Oscillaon (NAO). Models which have beer NAO variability have more chance 29 of a sudden change in SST. A few SST events persist over several decades. These cooling and warming 30 phases could have significant effects, especially for the Brish Isles, and need to be considered when 31 planning climate change adaptaon. 32 Main points 33 1. The North Atlanc shows the potenal for abrupt changes in sea surface temperature linked 34 to decadal variability in coupled processes 35 2. The abrupt cooling and subsequent warming may have significant implicaons for impacts 36 and adapon, parcularly over the Brish Isles 37 3. Using observaonal constraints, the model-derived likelihood of an abrupt change in North 38 Atlanc temperatures approximately doubles. 39 1 Introducon 40 The subpolar North Atlanc (SPNA) region is a crical component of the global climate system, 41 influencing climate variability on annual to decadal mescales (Gasneau & Frankignoul, 2015; 42 Suon & Hodson, 2005; Yeager, 2015), with significant implicaons for temperature and salinity 43 distribuon (Holliday et al., 2018; Marzocchi et al., 2015). The SPNA gyre circulaon is also crucial in 44 driving decadal climate variability and potenally exhibits mulple circulaon modes (Born & 45 Mignot, 2012; Born & Stocker, 2014). 46 The North Atlanc Oscillaon (NAO) also plays a significant role in SPNA variability through 47 associated changes in heat fluxes and winds. In the mid-1990s, the SPNA underwent a rapid 48 warming, with SSTs increasing by around 1°C in just 2 years (Sarafanov et al., 2008). This warming 49 followed a prolonged posive phase of the NAO, followed by an unusually negave NAO index. This 50 rapid warming is consistent with a delayed response to the prolonged posive phase NAO, and not 51 simply an instantaneous response to the negave NAO (Robson et al., 2012). 52 Other notable historic shis in the SPNA climate system include the possibility that sea-ice to ocean 53 feedbacks sustained an inial cooling into the Lile Ice Age by weakening the SPNA gyre circulaon 54 (Moreno-Chamarro et al., 2017). Empirical evidence from annually resolved bivalve proxy records 55 from the North Icelandic shelf shows that the SPNA climate system destabilised during two episodes 56 prior to the Lile Ice Age, potenally indicang a system approaching an abrupt transion (Arellano-57 Nava et al., 2022). 58 In climate model projecons, the SPNA is the epicentre of the so called “warming hole” (a relave 59 cooling), which has been linked to changes in the strength of the Atlanc Meridional Overturning 60 Circulaon (AMOC) due to weakening deep convecon (Menary & Wood, 2018). Anomalous cooling 61 in this region has been linked to changes in Atlanc hurricane frequency (Smith et al., 2010) and an 62 increase in summer precipitaon over Europe (Dong et al., 2013). In addion, some models exhibit 63 significant temperature drops over the North Atlanc that have been associated with a more 64 complete collapse of SPNA convecon (Sgubin et al., 2017; Swingedouw et al., 2021). Such 65 temperature drops may be a symptom of the region undergoing a rapid, sustained, and self-66 perpetuang change known as a pping point (Armstrong McKay et al., 2022) Weighng climate 67 models based on the fidelity of their ocean straficaon in this region significantly increases the 68 esmated risk of an abrupt cooling (Sgubin et al. 2017, Swingedouw et al., 2021). The associated 69 “cold waves” may have significant repercussions over Europe, including impacts on viculture 70 (Sgubin et al., 2019). 71 One mechanism proposed for abrupt changes in the SPNA involves feedbacks between surface 72 salinity, deep convecon, and subpolar gyre circulaon, with the laer hypothesised to feedback on 73 surface salinity (Born and Mignot 2012, Born and Stocker 2014). Born et al (2013) studied this 74 mechanism in CMIP3-generaon climate models, finding that while it operated in some models, it 75 was not universally present, and other processes may be at play for some abrupt SPNA events. Gu et 76 al (2024), analysing a large, single-model ensemble, proposed two further mechanisms leading to 77 alternave states of the SPNA, both sll involving convecon changes, while Born et al (2013) and 78 Sgubin et al (2017) found some events that did not appear to primarily involve a convecve 79 feedback. 80 Understanding potenal abrupt changes in the SPNA is crucial for developing effecve strategies to 81 migate and adapt to climate change risks. However, the current level of understanding is limited, 82 with more than one mechanism proposed. Interpretaon of the literature is further complicated by 83 differing choices of how to define abrupt events (e.g. in terms of gyre strength (Born et al 2013) or 84 sea surface temperature (Sgubin et al. 2017, Swingedouw et al. 2021)). In this paper we adopt two 85 pragmac definions, based on decadal surface temperature or mixed layer changes of a parcular 86 magnitude, that are not based on a parcular hypothesised mechanism. We use such events, 87 diagnosed across the CMIP6 model ensemble, to ask whether there is a robust connecon between 88 abrupt SST and mixing changes. We further assess regional climac anomalies associated with SPNA 89 events in the context of ongoing climate change, using the Brish Isles as an example of a region 90 likely to be strongly affected. 91 2 Methods 92 In this analysis we use data from all available CMIP6 climate models (Eyring et al., 2016) that 93 provided data for the historical simulaons or future projecons using scenarios SSP1-2.6, SSP2-4.5, 94 SSP3-7.0, or SSP5-8.5 (Supp. Table 1). We use all available ensemble members, analysing a total of 95 775 simulaons covering 90195 model years. Mixed layer depths (MLDs, defined as in Griffies et al, 96 2016), mean sea level pressure (MSLP), and the North Atlanc Oscillaon (NAO) index are calculated 97 as winterme (January to March inclusive) means, whilst other variables are calculated as annual 98 means. For comparisons to observaons, we use surface air temperature (SAT) reconstrucons from 99 the gridded BEST dataset (Rohde & Hausfather, 2020), staon-based sea level pressure observaons 100 from HadSLP2 (Allan & Ansell, 2006), and precipitaon reconstrucons from GPCC (Becker et al., 101 2013). 102 Previously, Swingedouw et al. (2021; hereaer SW21) defined abrupt events in the SPNA in terms of 103 decadal changes in SSTs. We use a similar method, defining an abrupt SST event as a 1 Kelvin 104 reducon in SPNA area-averaged SST from any given decade to the next, over at least 50% of this 105 region shown in Supp. Figure 1. A hypothesis when abrupt SPNA changes are defined in this way is 106 that they are associated with an abrupt change, or collapse, in deep convecon in the SPNA (e.g. 107 Born and Mignot 2012). To explore this link, we define a separate index of abrupt change in the SPNA 108 based on mixed layer depths (MLDs). Given the varied locaons and depths of simulated deep mixed 109 layers (Heuzé, 2021), for this index we use a model-based approach: we first calculate the winterme 110 mean MLDs from the first 20 years of the simulaon (or combined simulaons) and then use as a 111 search region the area where these MLDs are greater than 400 metres. We use the first 20 years, 112 rather than the pre-industrial control run, to maximise data availability. We detect an abrupt-MLD 113 change where the decline in MLDs in this search region is greater than 45% of the mean over at least 114 50% of this region. While changes in some regions could be more significant than others, this 115 method gives no preference to the specific locaon of changes in MLDs. Ensemble members where 116 the me mean MLD area (greater than 400 metres) is less than 2x105 km2 are discounted to exclude 117 abrupt changes that amount to only a few grid cells. In all cases, decadal changes are calculated 118 using an annually moving window. To avoid double counng, when mulple, consecuve events 119 would be detected – or events would overlap – only the event with greatest magnitude is selected. 120 3 Results 121 3.1 Abrupt events in SST and MLD 122 We begin by searching for events in the future projecons under the 4 main scenarios (SSP1-2.6, 123 SSP2-4.5, SSP3-7.0, SSP5-8.5). For SSP1-2.6, we find 5 unique, abrupt events based on our SST index 124 and 5 events based on our MLD index out of 130 ensemble members (Supp. Text 1, Supp. Table 1). 125 Since our SST-based metric is similar to that used by SW21, this includes the events highlighted in 126 SW21 (Supp. Text 1). However, with only 5 events there is a lile stascal power in our results. We 127 therefore extend our analysis to include the historical simulaons from which the scenario 128 simulaons were iniated. 129 130 Figure 1. Area-averaged subpolar sea surface temperature (SST; A) and Mixed Layer Depth (MLD, 131 model-specific regions; B) anomalies in all models and ensemble members in the concatenated 132 historical and SSP1-2.6 scenario simulaons (grey). Timeseries have a 10 year running mean 133 applied. In each panel, events involving the variable of interest (SST/MLD) are shown in red, and 134 events involving the other variable (MLD/SST) are shown in blue. 135 Including the historical simulaons increases the number of resultant abrupt events to 19 for the SST 136 metric and 20 for the MLD metric from a total of 108 ensemble members (Figure 1 and Supp. Tables 137 1 and 2). This includes previously undetected events that occur near the start of the scenario 138 simulaons. Figure 1 confirms that SST and MLD events are not co-located in me. Most of the MLD-139 based events occur around the year 2020 (Figure 1B), which is when the overall forcing and 140 associated warming and mixed-layer shoaling trends are highest. Hence, many of the “abrupt” MLD 141 events we see here could also be considered a part of an ongoing, sustained decline, rather than an 142 individual event. This is in comparison to the SST-based events, many of which occur at mes when 143 the long-term SST trend is not strong (Figure 1A). 144 For SST-based events, the associated changes are shown in Figure 2 using composite maps. These are 145 defined as the average difference from the preceding decade across all events, centred on the event. 146 Along with a cooling of the SPNA (Figure 2F, H) there is a shallowing of the MLD in the Labrador Sea 147 in the same and the preceding decade (Figure 2B, G). As such, there is a relaonship between SST-148 based events and MLDs, although not with the abrupt MLD events idenfied previously, and not 149 every SST event is associated with MLD shallowing (Figure 1B, Supp. Figure 2). This may be because 150 our MLD event index picks up changes in the model’s preferred convecon site, which is not the 151 Labrador Sea in some models. The SST-based events also show a concomitant change in mean sea 152 level pressure (MSLP, Figure 2J) consistent with an anomalously posive NAO index and signals of a 153 negave NAO in the preceding decade (Figure 2E). In the decade following the event there is a 154 warming of the SPNA, suggesng that the system is undergoing temporary decadal variability, rather 155 than a sustained change (Figure 2K, M). For comparison, this warming is similar in magnitude and 156 extent to the warming seen in the ensemble mean during the period of maximum warming (Supp. 157 Figure 3). 158 159 In contrast to the composite picture, the four abrupt SST events that occur in the future do not 160 appear to show decadal variability, but instead a more persistent temperature change. These events 161 appear to be associated with MLD shallowing (Supp. Figure 2, events #4, #5, #16, #20. Note that 162 events #4 and #5 correspond to the two events highlighted in SW21). 163 In the MLD-based events, there is a general warming both before, during, and aer the events (Supp. 164 Figure 4). We suggest that the MLD-events are not driving the warming seen in those composites. 165 Instead, long-term warming may contribute to increased straficaon of the SPNA and associated 166 capping of deep convecon, which leads to a shallowing, and in some cases an abrupt change, in 167 MLDs. The apparent lack of warming in the central SPNA would then be consistent with a 168 compeon between large-scale warming and local cooling (due to a weakening AMOC), which is the 169 proposed mechanism behind the SPNA warming hole (Menary & Wood, 2018). 170 171 Figure 2. Composite maps of Sea Surface Temperature (SST; first column), Mixed Layer Depth 172 (MLD; second column), Surface Air Temperature (SAT; third column), precipitaon (fourth column) 173 and mean sea level pressure (MSLP; fih column) for abrupt SST events in the combined historical 174 plus SSP1-2.6 scenario simulaons preceding the abrupt event (A-E), centred on the abrupt event 175 (F-J), and following the abrupt event (K-O). Red tles denote the variable used to create the 176 composite. Grid points where the data is not significant at the 95% level are masked. This was 177 determined using bootstrapping: the same number of samples (19 for the combined simulaons) 178 were randomly selected from the enre dataset to create a composite, which was repeated to 179 create a distribuon from 10000 composites. Also highlighted are the regions used for subsequent 180 lagged regression analysis (SST: F, MLD: G, NAO: crosses in J) and the regions defined as Brish Isles 181 and Europe (H). 182 183 3.2 Proposed mechanisms 184 Decadal to muldecadal variability of SPNA SSTs is mainly driven by changes in ocean circulaon 185 (Buckley & Marshall, 2016). Variability with ~20-year periodicity has been shown to be driven by 186 internally generated Rossby waves which propagate from east to west (Sévellec and Fedorov 2013, 187 Muir and Fedorov, 2017; Gasneau 2018). Here, mul-model composites of SSTs do not show signals 188 of propagaon (not shown) though it is possible that this mechanism is at play in individual models. 189 A different mechanism has also been proposed for muldecadal variability in the SPNA, with a 190 prolonged posive phase of the NAO driving increased convecon and deep-water formaon in the 191 northwest SPNA, resulng in a stronger ocean circulaon (Eden and Jung 2001, Robson 2012, Yeager 192 and Robson 2017). This drives increased northward ocean advecon of heat and increased SSTs in 193 the SPNA. A prolonged negave NAO results in cooler SSTs through the same mechanism. To explore 194 this mechanism, we show lead-lag correlaons using meseries of the composite events (Figure 3). 195 We use annual SST, winterme MLD and NAO indices using the regions defined in Figure 2 (F, G, J), 196 with MLD over the northwest SPNA used as a proxy for convecon there. 197 198 199 Figure 3. Lagged correlaons between the North Atlanc Oscillaon (NAO) and northwestern SPNA 200 mixed layer depths (MLDs; A), sea surface temperatures (SSTs; B) and between MLDs and SSTs (C). 201 Solid black lines denote correlaons required for significance at the 90% level. Significance was 202 esmated via a bootstrap approach by conducng the same lagged regressions on arbitrary 203 composite me series. 204 Results show that a negave NAO precedes a reducon in MLDs by around 7-9 years (Figure 3A, α) 205 and a reducon in SSTs by around 9-10 years (Figure 3B, γ). This is broadly consistent with the lag 206 between MLDs and SSTs, which is around 2-4 years (Figure 3C, ε). 207 Although we cannot assess the mechanisms in detail, the composite abrupt changes studied here 208 show the same relaonships between the NAO, SSTs, and northwestern SPNA MLDs as found in 209 previous mechanisc studies. We therefore suggest that many abrupt SST cooling events in the SPNA 210 (using our definion) could be the result of decadal-muldecadal variability in the coupled 211 atmosphere-ocean system. These events may include a role for MLD changes, but not necessarily 212 abrupt ones. 213 3.3 Impacts 214 Abrupt changes from the SST events are associated with climate anomalies in the Brish Isles (the 215 highly populated region potenally most exposed to the SST events) and of Europe as a whole 216 (Figure 2H). To put impacts in context with projected long-term warming, we show simulated Brish 217 Isles SAT changes under the historical plus SSP2-4.5 scenario as a plausible future pathway (UN 218 Environment Programme, 2023) and superimpose the composite SST-event meseries at arbitrary 219 points in me for illustraon (Figure 4). These composite events reveal a potenally important 220 decadal cooling over the Brish Isles, followed by a warming (Figure 4A, blue). This cooling is 221 stascally significant in comparison to both simulated and observed decadal mescale variability 222 (Figure 4B), in both summer and winter as well as the annual mean (Supp. Figure 5A,B,C). The mean 223 decadal cooling in the composites is -0.64°C, compared to -0.40 to 0.61°C (95% range) for all years in 224 the simulaons. Averaged over Europe, the decadal mescale changes in annual mean SAT are less 225 significant, although substanal cooling is seen in the annual mean (Figure 4C, D) and parcularly 226 summer (Supp. Figure 5E). 227 Whilst the decadal mescale changes (associated with abrupt SST events) may be larger than the 228 observed decadal mescale variability, they would not be enough to “reset” SAT over the Brish Isles 229 to 20th century levels. Nevertheless, since a cooling trend is not currently being planned for, this 230 could have implicaons for adaptaon which is currently focussed on warming. Within 15 years the 231 effect of the composite abrupt SST-event on SAT is largely undone. However, if combined with 232 concomitant anthropogenic warming, this rebound could represent a significant and impacul local 233 decadal warming trend over the Brish Isles. For example, assuming the warming rates are 234 independent and adding this rebound phase (0.39°C/decade) to the mean warming rate for the year 235 2050 (0.16°C/decade; from the full model ensemble using SSP2-4.5) yields a temporary local 236 warming rate of 0.55°C/decade for the Brish Isles. This is greater than the decadal Brish Isles 237 warming rate in SSP5-8.5 for the same period (0.39°C/decade) and more like the SSP5-8.5 rate 238 towards the end of the 21st century (not shown). Note that the more persistent SST events show 239 similar magnitudes of cooling in the Brish Isles to the shorter events, albeit longer lasng. (Supp. 240 Figure 2) 241 242 Figure 4. Surface air temperature (SAT) anomalies over the Brish Isles (A) and Europe (C) in 243 observaons (red) and CMIP6 models (historical plus SSP2-4.5 scenario simulaons, shown as the 244 ensemble mean and 1 and 2 standard deviaons, black). The SAT from the sea surface temperature 245 (SST) composites are shown in blue and overlaid at arbitrary years (trend-adjusted) to illustrate 246 their magnitude relave to long term changes (shading represents 2 standard deviaons). Also 247 shown are the distribuons of decadal changes in SAT over the Brish Isles (B) and Europe (D) in 248 observaons (red), CMIP6 models (black) and across the 19 member composites (the maximum 249 decadal change, i.e. corresponding to Figure 1; blue). 250 The distribuon of decadal mescale precipitaon change associated with the SST-based events 251 shows smaller changes than SAT, with the Brish Isles seeing a shi towards larger decadal increases 252 in winter. Europe sees a slight shi towards larger decreases in summer (Supp. Figure 5). 253 For abrupt MLD changes, there is some signal of increased SAT (Supp. Figure 6), again consistent with 254 the overall warming trend in the abrupt MLD composites (Supp. Figure 4). Precipitaon over the 255 Brish Isles shows a shi towards larger decadal decreases in winter and away from the largest 256 decadal increases in summer. 257 Discussion 258 We have invesgated the occurrence of abrupt SST and MLD events in the SPNA in CMIP6 models in 259 historical and future scenarios. The larger set of model runs studied here, and our different 260 definions used, have revealed a diversity of SST event types in the CMIP6 ensemble, including the 261 small number of persistent changes discussed by SW21 alongside a larger number of shorter-lived 262 changes (cf. Sgubin et al. 2017). Many abrupt SST events are associated with decadal mescale, 263 coupled variability, in which convecon may play a role but does not change abruptly. On the other 264 hand, abrupt MLD events appear to be related to longer mescale anthropogenic warming. 265 To fully understand the risks from abrupt SPNA change it will be necessary to develop a more 266 complete documentaon of possible types of event, and their drivers. It should also be noted that 267 dominant mechanisms of variability may be dependent on the underlying mean SPNA state, due to 268 either model biases (Menary et al. 2015) or the gradually changing background climate (Gu et al. 269 2024). 270 In Sgubin et al (2017) and SW21, models with beer SPNA straficaon were more likely to show 271 abrupt SST changes. Using our model-based definions of mixing locaons, we find no clear link 272 between the locaon or mean area of the mixed layers and whether a given model is more or less 273 likely to show an abrupt SST change (Supp. Figure 7). However, another important part of the 274 proposed mechanism driving abrupt SST changes is the NAO, whose stascs can be poorly 275 simulated in climate models (Davini & Cagnazzo, 2014). Supp. Figure 8 shows that ensemble 276 members with more realisc winterme NAO variability are more likely to show both abrupt SST and 277 MLD events. Applying a simple Gaussian weighng (Supp. Text 2) increases the likelihoods of a given 278 ensemble member showing an abrupt SST (MLD) event before the year 2100 from 15% (18%) to 29% 279 (33%). Hence, as in SW21, constraining the models using relevant observaons yields an increased 280 likelihood of abrupt changes. 281 Summary and Conclusions 282 Our analysis suggests that many SST events are consistent with decadal mescale variability in the 283 SPNA, linked to the NAO, and that they are not associated with an abrupt and sustained change in 284 MLDs. This apparent difference from SW21 is likely due to the different definion of SST events used 285 here, and possibly the larger sample of CMIP runs analysed. SW21 (and Sgubin et al 2017) define 286 events in terms of a 3 standard deviaon departure from variability in a long control run, whereas we 287 use here a more observable definion based on an SST change of a defined magnitude. Our 288 algorithm does pick out the small number of persistent events that were the focus in SW21, as well 289 as a larger number of decadal-mescale, NAO-related events. For our MLD-based metric, abrupt 290 changes are associated with longer mescale, sustained increases in SST, and occur most oen 291 during 2000-2040, when the subpolar warming rate due to anthropogenic climate change is largest. 292 It is possible that a differently defined metric of SPNA convecon may idenfy different events. Given 293 the diversity of events found, it may therefore be beneficial for community-defined metrics to be 294 discussed and agreed upon. 295 Although the cooling associated with many SST-based events is temporary, it could nevertheless 296 represent a significant change in SAT over the Brish Isles, with the magnitude of the decadal change 297 comparable to the most extreme changes seen in observaons or models. There have so far been 298 very few studies of the potenal socioeconomic or ecological impacts of abrupt SPNA events (e.g. 299 Sgubin et al., 2019) although their presence in some CMIP5 and CMIP6 ensemble members means 300 that their effects may have already been included in the range of possible outcomes in impacts 301 studies that used these ensembles (Hasegawa et al., 2022). Both a period of cooling and a 302 subsequent rebound period with relavely rapid warming could pose challenges for adaptaon. 303 Relavely cold temperatures could directly bring risks if adaptaon plans have been based on 304 assumpons of ongoing warming. A relavely cool period could also potenally delay further 305 adaptaon to ongoing warming by creang a false impression of a lack of a warming trend. If the 306 subsequent rebound of temperatures is more rapid than the previous long-term trend, the faster 307 rate of warming may be more difficult for ecosystems and society to adapt to. Further analysis 308 specifically focussing on impacts, across mulple socio-economic sectors, driven by simulaons with 309 both temporary and persistent abrupt SPNA events, would be valuable. 310 We find that those models with a beer representaon of NAO variability are more likely to show 311 abrupt events. Further work should understand and quanfy how the likelihood of abrupt events 312 depends on the fidelity of the models and the gradually evolving mean state (Menary et al 2015, 313 Sgubin et al. 2017, SW21, Gu et al. 2024). Finally, given the coupled nature of abrupt SST events and 314 their link to previous work in a decadal climate predicon context (Msadek et al., 2014; Robson et al., 315 2012), it would be worthwhile to understand to what extent abrupt cooling is (or might be) 316 predictable. 317 Acknowledgements 318 MBM, RAW, LCJ and RAB were supported by the Met Office Hadley Centre Climate Programme 319 funded by DSIT. RAW and LCJ were addionally funded by ClimTip. This is ClimTip contribuon #7; 320 the ClimTip project has received funding from the European Union's Horizon Europe research and 321 innovaon programme under grant agreement No. 101137601 322 Open Research 323 The CMIP data are available through ESGF's website at hps://esgf-node.llnl.gov/search/cmip6/. The 324 BEST data are available through the Berkeley Earth website at hps://berkeleyearth.org/data/. The 325 HadSLP2 data are available through the UK Met Office website at 326 hps://www.metoffice.gov.uk/hadobs/hadslp2/. The GPCC data are available through the Deutscher 327 Weerdienst website at 328 hps://opendata.dwd.de/climate_environment/GPCC/html/download_gate.html. 329 References 330 Allan, R., & Ansell, T. (2006). A new globally complete monthly historical gridded mean sea level 331 pressure dataset (HadSLP2): 1850-2004. Journal of Climate, 19(22). 332 hps://doi.org/10.1175/JCLI3937.1 333