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

1
Ab up changes in he subpola No h A lanc and hei impac on he clima e o he B ish Isles 2
Mahew B Mena y1,*, Lau a C Jackson1, Richa d A. Wood1, Richa d A. Bes1,2
3
1. Me Office Hadley Cen e, Fi zRoy Road, Exe e , EX1 3PB, UK 4
2. Global Sys ems Ins u e, Uni e si y o Exe e , No h Pa k Road, Exe e , EX4 4QE 5
*Now a : Eu opean Cen e o Medium-Range Wea he Fo ecas s, Shinfield Pa k, Reading, RG2 6
9AX, UK 7
8
Abs ac 9
The e has been inc easing in e es in he possibili y o ab up climac changes in he No h A lanc 10
and hei impac s on no hwes e n Eu ope. He e, we in esga e such ab up changes in a la ge 11
ensemble o CMIP6 clima e models. We define wo po enally obse able me ics based on subpola 12
sea su ace empe a u es (SSTs) o mixed laye dep hs (MLDs), o explo e he link be ween 13
empe a u e changes and con econ collapse. The wo me ics yield simila numbe s o ab up 14
e en s bu sugges ha se e al ypes o ab up e en a e possible. Ab up MLD changes appea 15
ela ed o ongoing wa ming. Ab up SST changes mos ly consis o decadal cooling ollowed by 16
wa ming, appa en ly ela ed o coupled dynamics in ol ing he NAO. Models wi h mo e ealisc NAO 17
a iabili y show mo e such e en s. Howe e , se e al mo e pe sis en SST e en s a e also ound. Bo h 18
cooling and wa ming phases ha e impo an implicaons o impac s and adap aon, pa cula ly 19
o e he B ish Isles. 20
Plain language summa y 21
The e is inc easing in e es in sudden clima e changes in he No h A lanc and how hese changes 22
migh affec Eu ope. This s udy uses a la ge se o ecen clima e models o see i hese sudden 23
changes a e likely. We use wo measu es ela ed o 1) sea su ace empe a u es (SSTs) and 2) mixed 24
laye dep hs (MLDs) o s udy he connecon be ween su ace empe a u es and he collapse o 25
oceanic con econ. Bo h measu es show simila numbe s o sudden changes, bu hey don’ seem o 26
be ela ed. Sudden MLD changes mos ly seem o be connec ed o ongoing global wa ming. Sudden 27
SST changes mos ly ollow a pae n o cooling and wa ming o e a ound 20 yea s, appa en ly linked 28
o he No h A lanc Oscillaon (NAO). Models which ha e bee NAO a iabili y ha e mo e chance 29
o a sudden change in SST. A ew SST e en s pe sis o e se e al decades. These cooling and wa ming 30
phases could ha e significan effec s, especially o he B ish Isles, and need o be conside ed when 31
planning clima e change adap aon. 32
Main poin s 33
1. The No h A lanc shows he po enal o ab up changes in sea su ace empe a u e linked 34
o decadal a iabili y in coupled p ocesses 35
2. The ab up cooling and subsequen wa ming may ha e significan implicaons o impac s 36
and adapon, pa cula ly o e he B ish Isles 37
3. Using obse aonal cons ain s, he model-de i ed likelihood o an ab up change in No h 38
A lanc empe a u es app oxima ely doubles. 39
1 In oducon 40
The subpola No h A lanc (SPNA) egion is a c ical componen o he global clima e sys em, 41
influencing clima e a iabili y on annual o decadal mescales (Gasneau & F ankignoul, 2015; 42
Suon & Hodson, 2005; Yeage , 2015), wi h significan implicaons o empe a u e and salini y 43
dis ibuon (Holliday e al., 2018; Ma zocchi e al., 2015). The SPNA gy e ci culaon is also c ucial in 44
d i ing decadal clima e a iabili y and po enally exhibi s mulple ci culaon modes (Bo n & 45
Migno , 2012; Bo n & S ocke , 2014). 46
The No h A lanc Oscillaon (NAO) also plays a significan ole in SPNA a iabili y h ough 47
associa ed changes in hea fluxes and winds. In he mid-1990s, he SPNA unde wen a apid 48
wa ming, wi h SSTs inc easing by a ound 1°C in jus 2 yea s (Sa a ano e al., 2008). This wa ming 49
ollowed a p olonged posi e phase o he NAO, ollowed by an unusually nega e NAO index. This 50
apid wa ming is consis en wi h a delayed esponse o he p olonged posi e phase NAO, and no 51
simply an ins an aneous esponse o he nega e NAO (Robson e al., 2012). 52
O he no able his o ic shis in he SPNA clima e sys em include he possibili y ha sea-ice o ocean 53
eedbacks sus ained an inial cooling in o he Lile Ice Age by weakening he SPNA gy e ci culaon 54
(Mo eno-Chama o e al., 2017). Empi ical e idence om annually esol ed bi al e p oxy eco ds 55
om he No h Icelandic shel shows ha he SPNA clima e sys em des abilised du ing wo episodes 56
p io o he Lile Ice Age, po enally indicang a sys em app oaching an ab up ansion (A ellano-57
Na a e al., 2022). 58
In clima e model p ojecons, he SPNA is he epicen e o he so called “wa ming hole” (a ela e 59
cooling), which has been linked o changes in he s eng h o he A lanc Me idional O e u ning 60
Ci culaon (AMOC) due o weakening deep con econ (Mena y & Wood, 2018). Anomalous cooling 61
in his egion has been linked o changes in A lanc hu icane equency (Smi h e al., 2010) and an 62
inc ease in summe p ecipi aon o e Eu ope (Dong e al., 2013). In addion, some models exhibi 63
significan empe a u e d ops o e he No h A lanc ha ha e been associa ed wi h a mo e 64
comple e collapse o SPNA con econ (Sgubin e al., 2017; Swingedouw e al., 2021). Such 65
empe a u e d ops may be a symp om o he egion unde going a apid, sus ained, and sel -66
pe pe uang change known as a pping poin (A ms ong McKay e al., 2022) Weighng clima e 67
models based on he fideli y o hei ocean s aficaon in his egion significan ly inc eases he 68
esma ed isk o an ab up cooling (Sgubin e al. 2017, Swingedouw e al., 2021). The associa ed 69
“cold wa es” may ha e significan epe cussions o e Eu ope, including impac s on icul u e 70
(Sgubin e al., 2019). 71
One mechanism p oposed o ab up changes in he SPNA in ol es eedbacks be ween su ace 72
salini y, deep con econ, and subpola gy e ci culaon, wi h he lae hypo hesised o eedback on 73
su ace salini y (Bo n and Migno 2012, Bo n and S ocke 2014). Bo n e al (2013) s udied his 74
mechanism in CMIP3-gene aon clima e models, finding ha while i ope a ed in some models, i 75
was no uni e sally p esen , and o he p ocesses may be a play o some ab up SPNA e en s. Gu e 76
al (2024), analysing a la ge, single-model ensemble, p oposed wo u he mechanisms leading o 77
al e na e s a es o he SPNA, bo h sll in ol ing con econ changes, while Bo n e al (2013) and 78
Sgubin e al (2017) ound some e en s ha did no appea o p ima ily in ol e a con ec e 79
eedback. 80
Unde s anding po enal ab up changes in he SPNA is c ucial o de eloping effec e s a egies o 81
miga e and adap o clima e change isks. Howe e , he cu en le el o unde s anding is limi ed, 82
wi h mo e han one mechanism p oposed. In e p e aon o he li e a u e is u he complica ed by 83
diffe ing choices o how o define ab up e en s (e.g. in e ms o gy e s eng h (Bo n e al 2013) o 84
sea su ace empe a u e (Sgubin e al. 2017, Swingedouw e al. 2021)). In his pape we adop wo 85
p agmac definions, based on decadal su ace empe a u e o mixed laye changes o a pa cula 86
magni ude, ha a e no based on a pa cula hypo hesised mechanism. We use such e en s, 87
diagnosed ac oss he CMIP6 model ensemble, o ask whe he he e is a obus connecon be ween 88
ab up SST and mixing changes. We u he assess egional climac anomalies associa ed wi h SPNA 89
e en s in he con ex o ongoing clima e change, using he B ish Isles as an example o a egion 90
likely o be s ongly affec ed. 91
2 Me hods 92
In his analysis we use da a om all a ailable CMIP6 clima e models (Ey ing e al., 2016) ha 93
p o ided da a o he his o ical simulaons o u u e p ojecons using scena ios SSP1-2.6, SSP2-4.5, 94
SSP3-7.0, o SSP5-8.5 (Supp. Table 1). We use all a ailable ensemble membe s, analysing a o al o 95
775 simulaons co e ing 90195 model yea s. Mixed laye dep hs (MLDs, defined as in G iffies e al, 96
2016), mean sea le el p essu e (MSLP), and he No h A lanc Oscillaon (NAO) index a e calcula ed 97
as win e me (Janua y o Ma ch inclusi e) means, whils o he a iables a e calcula ed as annual 98
means. Fo compa isons o obse aons, we use su ace ai empe a u e (SAT) econs ucons om 99
he g idded BEST da ase (Rohde & Haus a he , 2020), s aon-based sea le el p essu e obse aons 100
om HadSLP2 (Allan & Ansell, 2006), and p ecipi aon econs ucons om GPCC (Becke e al., 101
2013). 102
P e iously, Swingedouw e al. (2021; he eae SW21) defined ab up e en s in he SPNA in e ms o 103
decadal changes in SSTs. We use a simila me hod, defining an ab up SST e en as a 1 Kel in 104
educon in SPNA a ea-a e aged SST om any gi en decade o he nex , o e a leas 50% o his 105
egion shown in Supp. Figu e 1. A hypo hesis when ab up SPNA changes a e defined in his way is 106
ha hey a e associa ed wi h an ab up change, o collapse, in deep con econ in he SPNA (e.g. 107
Bo n and Migno 2012). To explo e his link, we define a sepa a e index o ab up change in he SPNA 108
based on mixed laye dep hs (MLDs). Gi en he a ied locaons and dep hs o simula ed deep mixed 109
laye s (Heuzé, 2021), o his index we use a model-based app oach: we fi s calcula e he win e me 110
mean MLDs om he fi s 20 yea s o he simulaon (o combined simulaons) and hen use as a 111
sea ch egion he a ea whe e hese MLDs a e g ea e han 400 me es. We use he fi s 20 yea s, 112
a he han he p e-indus ial con ol un, o maximise da a a ailabili y. We de ec an ab up -MLD 113
change whe e he decline in MLDs in his sea ch egion is g ea e han 45% o he mean o e a leas 114
50% o his egion. While changes in some egions could be mo e significan han o he s, his 115
me hod gi es no p e e ence o he specific locaon o changes in MLDs. Ensemble membe s whe e 116
he me mean MLD a ea (g ea e han 400 me es) is less han 2x105 km2 a e discoun ed o exclude 117
ab up changes ha amoun o only a ew g id cells. In all cases, decadal changes a e calcula ed 118
using an annually mo ing window. To a oid double counng, when mulple, consecu e e en s 119
would be de ec ed – o e en s would o e lap – only he e en wi h g ea es magni ude is selec ed. 120
3 Resul s 121
3.1 Ab up e en s in SST and MLD 122
We begin by sea ching o e en s in he u u e p ojecons unde he 4 main scena ios (SSP1-2.6, 123
SSP2-4.5, SSP3-7.0, SSP5-8.5). Fo SSP1-2.6, we find 5 unique, ab up e en s based on ou SST index 124
and 5 e en s based on ou MLD index ou o 130 ensemble membe s (Supp. Tex 1, Supp. Table 1). 125
Since ou SST-based me ic is simila o ha used by SW21, his includes he e en s highligh ed in 126
SW21 (Supp. Tex 1). Howe e , wi h only 5 e en s he e is a lile s ascal powe in ou esul s. We 127
he e o e ex end ou analysis o include he his o ical simulaons om which he scena io 128
simulaons we e inia ed. 129
130
Figu e 1. A ea-a e aged subpola sea su ace empe a u e (SST; A) and Mixed Laye Dep h (MLD, 131
model-specific egions; B) anomalies in all models and ensemble membe s in he conca ena ed 132
his o ical and SSP1-2.6 scena io simulaons (g ey). Timese ies ha e a 10 yea unning mean 133
applied. In each panel, e en s in ol ing he a iable o in e es (SST/MLD) a e shown in ed, and 134
e en s in ol ing he o he a iable (MLD/SST) a e shown in blue. 135
Including he his o ical simulaons inc eases he numbe o esul an ab up e en s o 19 o he SST 136
me ic and 20 o he MLD me ic om a o al o 108 ensemble membe s (Figu e 1 and Supp. Tables 137
1 and 2). This includes p e iously unde ec ed e en s ha occu nea he s a o he scena io 138
simulaons. Figu e 1 confi ms ha SST and MLD e en s a e no co-loca ed in me. Mos o he MLD-139
based e en s occu a ound he yea 2020 (Figu e 1B), which is when he o e all o cing and 140
associa ed wa ming and mixed-laye shoaling ends a e highes . Hence, many o he “ab up ” MLD 141
e en s we see he e could also be conside ed a pa o an ongoing, sus ained decline, a he han an 142
indi idual e en . This is in compa ison o he SST-based e en s, many o which occu a mes when 143
he long- e m SST end is no s ong (Figu e 1A). 144
Fo SST-based e en s, he associa ed changes a e shown in Figu e 2 using composi e maps. These a e 145
defined as he a e age diffe ence om he p eceding decade ac oss all e en s, cen ed on he e en . 146
Along wi h a cooling o he SPNA (Figu e 2F, H) he e is a shallowing o he MLD in he Lab ado Sea 147
in he same and he p eceding decade (Figu e 2B, G). As such, he e is a elaonship be ween SST-148
based e en s and MLDs, al hough no wi h he ab up MLD e en s idenfied p e iously, and no 149
e e y SST e en is associa ed wi h MLD shallowing (Figu e 1B, Supp. Figu e 2). This may be because 150
ou MLD e en index picks up changes in he model’s p e e ed con econ si e, which is no he 151
Lab ado Sea in some models. The SST-based e en s also show a concomi an change in mean sea 152
le el p essu e (MSLP, Figu e 2J) consis en wi h an anomalously posi e NAO index and signals o a 153
nega e NAO in he p eceding decade (Figu e 2E). In he decade ollowing he e en he e is a 154
wa ming o he SPNA, suggesng ha he sys em is unde going empo a y decadal a iabili y, a he 155
han a sus ained change (Figu e 2K, M). Fo compa ison, his wa ming is simila in magni ude and 156
ex en o he wa ming seen in he ensemble mean du ing he pe iod o maximum wa ming (Supp. 157
Figu e 3). 158
159
In con as o he composi e pic u e, he ou ab up SST e en s ha occu in he u u e do no 160
appea o show decadal a iabili y, bu ins ead a mo e pe sis en empe a u e change. These e en s 161
appea o be associa ed wi h MLD shallowing (Supp. Figu e 2, e en s #4, #5, #16, #20. No e ha 162
e en s #4 and #5 co espond o he wo e en s highligh ed in SW21). 163
In he MLD-based e en s, he e is a gene al wa ming bo h be o e, du ing, and ae he e en s (Supp. 164
Figu e 4). We sugges ha he MLD-e en s a e no d i ing he wa ming seen in hose composi es. 165
Ins ead, long- e m wa ming may con ibu e o inc eased s aficaon o he SPNA and associa ed 166
capping o deep con econ, which leads o a shallowing, and in some cases an ab up change, in 167
MLDs. The appa en lack o wa ming in he cen al SPNA would hen be consis en wi h a 168
compeon be ween la ge-scale wa ming and local cooling (due o a weakening AMOC), which is he 169
p oposed mechanism behind he SPNA wa ming hole (Mena y & Wood, 2018). 170
171
Figu e 2. Composi e maps o Sea Su ace Tempe a u e (SST; fi s column), Mixed Laye Dep h 172
(MLD; second column), Su ace Ai Tempe a u e (SAT; hi d column), p ecipi aon ( ou h column) 173
and mean sea le el p essu e (MSLP; fih column) o ab up SST e en s in he combined his o ical 174
plus SSP1-2.6 scena io simulaons p eceding he ab up e en (A-E), cen ed on he ab up e en 175
(F-J), and ollowing he ab up e en (K-O). Red  les deno e he a iable used o c ea e he 176
composi e. G id poin s whe e he da a is no significan a he 95% le el a e masked. This was 177
de e mined using boo s apping: he same numbe o samples (19 o he combined simulaons) 178
we e andomly selec ed om he en e da ase o c ea e a composi e, which was epea ed o 179
c ea e a dis ibuon om 10000 composi es. Also highligh ed a e he egions used o subsequen 180
lagged eg ession analysis (SST: F, MLD: G, NAO: c osses in J) and he egions defined as B ish Isles 181
and Eu ope (H). 182
183
3.2 P oposed mechanisms 184

Decadal o muldecadal a iabili y o SPNA SSTs is mainly d i en by changes in ocean ci culaon 185
(Buckley & Ma shall, 2016). Va iabili y wi h ~20-yea pe iodici y has been shown o be d i en by 186
in e nally gene a ed Rossby wa es which p opaga e om eas o wes (Sé ellec and Fedo o 2013, 187
Mui and Fedo o , 2017; Gasneau 2018). He e, mul-model composi es o SSTs do no show signals 188
o p opagaon (no shown) hough i is possible ha his mechanism is a play in indi idual models. 189
A diffe en mechanism has also been p oposed o muldecadal a iabili y in he SPNA, wi h a 190
p olonged posi e phase o he NAO d i ing inc eased con econ and deep-wa e o maon in he 191
no hwes SPNA, esulng in a s onge ocean ci culaon (Eden and Jung 2001, Robson 2012, Yeage 192
and Robson 2017). This d i es inc eased no hwa d ocean ad econ o hea and inc eased SSTs in 193
he SPNA. A p olonged nega e NAO esul s in coole SSTs h ough he same mechanism. To explo e 194
his mechanism, we show lead-lag co elaons using mese ies o he composi e e en s (Figu e 3). 195
We use annual SST, win e me MLD and NAO indices using he egions defined in Figu e 2 (F, G, J), 196
wi h MLD o e he no hwes SPNA used as a p oxy o con econ he e. 197
198
199
Figu e 3. Lagged co elaons be ween he No h A lanc Oscillaon (NAO) and no hwes e n SPNA 200
mixed laye dep hs (MLDs; A), sea su ace empe a u es (SSTs; B) and be ween MLDs and SSTs (C). 201
Solid black lines deno e co elaons equi ed o significance a he 90% le el. Significance was 202
esma ed ia a boo s ap app oach by conducng he same lagged eg essions on a bi a y 203
composi e me se ies. 204
Resul s show ha a nega e NAO p ecedes a educon in MLDs by a ound 7-9 yea s (Figu e 3A, α) 205
and a educon in SSTs by a ound 9-10 yea s (Figu e 3B, γ). This is b oadly consis en wi h he lag 206
be ween MLDs and SSTs, which is a ound 2-4 yea s (Figu e 3C, ε). 207
Al hough we canno assess he mechanisms in de ail, he composi e ab up changes s udied he e 208
show he same elaonships be ween he NAO, SSTs, and no hwes e n SPNA MLDs as ound in 209
p e ious mechanisc s udies. We he e o e sugges ha many ab up SST cooling e en s in he SPNA 210
(using ou definion) could be he esul o decadal-muldecadal a iabili y in he coupled 211
a mosphe e-ocean sys em. These e en s may include a ole o MLD changes, bu no necessa ily 212
ab up ones. 213
3.3 Impac s 214
Ab up changes om he SST e en s a e associa ed wi h clima e anomalies in he B ish Isles ( he 215
highly popula ed egion po enally mos exposed o he SST e en s) and o Eu ope as a whole 216
(Figu e 2H). To pu impac s in con ex wi h p ojec ed long- e m wa ming, we show simula ed B ish 217
Isles SAT changes unde he his o ical plus SSP2-4.5 scena io as a plausible u u e pa hway (UN 218
En i onmen P og amme, 2023) and supe impose he composi e SST-e en mese ies a a bi a y 219
poin s in me o illus aon (Figu e 4). These composi e e en s e eal a po enally impo an 220
decadal cooling o e he B ish Isles, ollowed by a wa ming (Figu e 4A, blue). This cooling is 221
s ascally significan in compa ison o bo h simula ed and obse ed decadal mescale a iabili y 222
(Figu e 4B), in bo h summe and win e as well as he annual mean (Supp. Figu e 5A,B,C). The mean 223
decadal cooling in he composi es is -0.64°C, compa ed o -0.40 o 0.61°C (95% ange) o all yea s in 224
he simulaons. A e aged o e Eu ope, he decadal mescale changes in annual mean SAT a e less 225
significan , al hough subs anal cooling is seen in he annual mean (Figu e 4C, D) and pa cula ly 226
summe (Supp. Figu e 5E). 227
Whils he decadal mescale changes (associa ed wi h ab up SST e en s) may be la ge han he 228
obse ed decadal mescale a iabili y, hey would no be enough o “ ese ” SAT o e he B ish Isles 229
o 20 h cen u y le els. Ne e heless, since a cooling end is no cu en ly being planned o , his 230
could ha e implicaons o adap aon which is cu en ly ocussed on wa ming. Wi hin 15 yea s he 231
effec o he composi e ab up SST-e en on SAT is la gely undone. Howe e , i combined wi h 232
concomi an an h opogenic wa ming, his ebound could ep esen a significan and impacul local 233
decadal wa ming end o e he B ish Isles. Fo example, assuming he wa ming a es a e 234
independen and adding his ebound phase (0.39°C/decade) o he mean wa ming a e o he yea 235
2050 (0.16°C/decade; om he ull model ensemble using SSP2-4.5) yields a empo a y local 236
wa ming a e o 0.55°C/decade o he B ish Isles. This is g ea e han he decadal B ish Isles 237
wa ming a e in SSP5-8.5 o he same pe iod (0.39°C/decade) and mo e like he SSP5-8.5 a e 238
owa ds he end o he 21s cen u y (no shown). No e ha he mo e pe sis en SST e en s show 239
simila magni udes o cooling in he B ish Isles o he sho e e en s, albei longe lasng. (Supp. 240
Figu e 2) 241
242
Figu e 4. Su ace ai empe a u e (SAT) anomalies o e he B ish Isles (A) and Eu ope (C) in 243
obse aons ( ed) and CMIP6 models (his o ical plus SSP2-4.5 scena io simulaons, shown as he 244
ensemble mean and 1 and 2 s anda d de iaons, black). The SAT om he sea su ace empe a u e 245
(SST) composi es a e shown in blue and o e laid a a bi a y yea s ( end-adjus ed) o illus a e 246
hei magni ude ela e o long e m changes (shading ep esen s 2 s anda d de iaons). Also 247
shown a e he dis ibuons o decadal changes in SAT o e he B ish Isles (B) and Eu ope (D) in 248
obse aons ( ed), CMIP6 models (black) and ac oss he 19 membe composi es ( he maximum 249
decadal change, i.e. co esponding o Figu e 1; blue). 250
The dis ibuon o decadal mescale p ecipi aon change associa ed wi h he SST-based e en s 251
shows smalle changes han SAT, wi h he B ish Isles seeing a shi owa ds la ge decadal inc eases 252
in win e . Eu ope sees a sligh shi owa ds la ge dec eases in summe (Supp. Figu e 5). 253
Fo ab up MLD changes, he e is some signal o inc eased SAT (Supp. Figu e 6), again consis en wi h 254
he o e all wa ming end in he ab up MLD composi es (Supp. Figu e 4). P ecipi aon o e he 255
B ish Isles shows a shi owa ds la ge decadal dec eases in win e and away om he la ges 256
decadal inc eases in summe . 257
Discussion 258
We ha e in esga ed he occu ence o ab up SST and MLD e en s in he SPNA in CMIP6 models in 259
his o ical and u u e scena ios. The la ge se o model uns s udied he e, and ou diffe en 260
definions used, ha e e ealed a di e si y o SST e en ypes in he CMIP6 ensemble, including he 261
small numbe o pe sis en changes discussed by SW21 alongside a la ge numbe o sho e -li ed 262
changes (c . Sgubin e al. 2017). Many ab up SST e en s a e associa ed wi h decadal mescale, 263
coupled a iabili y, in which con econ may play a ole bu does no change ab up ly. On he o he 264
hand, ab up MLD e en s appea o be ela ed o longe mescale an h opogenic wa ming. 265
To ully unde s and he isks om ab up SPNA change i will be necessa y o de elop a mo e 266
comple e documen aon o possible ypes o e en , and hei d i e s. I should also be no ed ha 267
dominan mechanisms o a iabili y may be dependen on he unde lying mean SPNA s a e, due o 268
ei he model biases (Mena y e al. 2015) o he g adually changing backg ound clima e (Gu e al. 269
2024). 270
In Sgubin e al (2017) and SW21, models wi h bee SPNA s aficaon we e mo e likely o show 271
ab up SST changes. Using ou model-based definions o mixing locaons, we find no clea link 272
be ween he locaon o mean a ea o he mixed laye s and whe he a gi en model is mo e o less 273
likely o show an ab up SST change (Supp. Figu e 7). Howe e , ano he impo an pa o he 274
p oposed mechanism d i ing ab up SST changes is he NAO, whose s ascs can be poo ly 275
simula ed in clima e models (Da ini & Cagnazzo, 2014). Supp. Figu e 8 shows ha ensemble 276
membe s wi h mo e ealisc win e me NAO a iabili y a e mo e likely o show bo h ab up SST and 277
MLD e en s. Applying a simple Gaussian weighng (Supp. Tex 2) inc eases he likelihoods o a gi en 278
ensemble membe showing an ab up SST (MLD) e en be o e he yea 2100 om 15% (18%) o 29% 279
(33%). Hence, as in SW21, cons aining he models using ele an obse aons yields an inc eased 280
likelihood o ab up changes. 281
Summa y and Conclusions 282
Ou analysis sugges s ha many SST e en s a e consis en wi h decadal mescale a iabili y in he 283
SPNA, linked o he NAO, and ha hey a e no associa ed wi h an ab up and sus ained change in 284
MLDs. This appa en diffe ence om SW21 is likely due o he diffe en definion o SST e en s used 285
he e, and possibly he la ge sample o CMIP uns analysed. SW21 (and Sgubin e al 2017) define 286
e en s in e ms o a 3 s anda d de iaon depa u e om a iabili y in a long con ol un, whe eas we 287
use he e a mo e obse able definion based on an SST change o a defined magni ude. Ou 288
algo i hm does pick ou he small numbe o pe sis en e en s ha we e he ocus in SW21, as well 289
as a la ge numbe o decadal-mescale, NAO- ela ed e en s. Fo ou MLD-based me ic, ab up 290
changes a e associa ed wi h longe mescale, sus ained inc eases in SST, and occu mos oen 291
du ing 2000-2040, when he subpola wa ming a e due o an h opogenic clima e change is la ges . 292
I is possible ha a diffe en ly defined me ic o SPNA con econ may iden y diffe en e en s. Gi en 293
he di e si y o e en s ound, i may he e o e be beneficial o communi y-defined me ics o be 294
discussed and ag eed upon. 295
Al hough he cooling associa ed wi h many SST-based e en s is empo a y, i could ne e heless 296
ep esen a significan change in SAT o e he B ish Isles, wi h he magni ude o he decadal change 297
compa able o he mos ex eme changes seen in obse aons o models. The e ha e so a been 298
e y ew s udies o he po enal socioeconomic o ecological impac s o ab up SPNA e en s (e.g. 299
Sgubin e al., 2019) al hough hei p esence in some CMIP5 and CMIP6 ensemble membe s means 300
ha hei effec s may ha e al eady been included in he ange o possible ou comes in impac s 301
s udies ha used hese ensembles (Hasegawa e al., 2022). Bo h a pe iod o cooling and a 302
subsequen ebound pe iod wi h ela ely apid wa ming could pose challenges o adap aon. 303
Rela ely cold empe a u es could di ec ly b ing isks i adap aon plans ha e been based on 304
assumpons o ongoing wa ming. A ela ely cool pe iod could also po enally delay u he 305
adap aon o ongoing wa ming by c eang a alse imp ession o a lack o a wa ming end. I he 306
subsequen ebound o empe a u es is mo e apid han he p e ious long- e m end, he as e 307
a e o wa ming may be mo e difficul o ecosys ems and socie y o adap o. Fu he analysis 308
specifically ocussing on impac s, ac oss mulple socio-economic sec o s, d i en by simulaons wi h 309
bo h empo a y and pe sis en ab up SPNA e en s, would be aluable. 310
We find ha hose models wi h a bee ep esen aon o NAO a iabili y a e mo e likely o show 311
ab up e en s. Fu he wo k should unde s and and quan y how he likelihood o ab up e en s 312
depends on he fideli y o he models and he g adually e ol ing mean s a e (Mena y e al 2015, 313
Sgubin e al. 2017, SW21, Gu e al. 2024). Finally, gi en he coupled na u e o ab up SST e en s and 314
hei link o p e ious wo k in a decadal clima e p edicon con ex (Msadek e al., 2014; Robson e al., 315
2012), i would be wo hwhile o unde s and o wha ex en ab up cooling is (o migh be) 316
p edic able. 317
Acknowledgemen s 318
MBM, RAW, LCJ and RAB we e suppo ed by he Me Office Hadley Cen e Clima e P og amme 319
unded by DSIT. RAW and LCJ we e addionally unded by ClimTip. This is ClimTip con ibuon #7; 320
he ClimTip p ojec has ecei ed unding om he Eu opean Union's Ho izon Eu ope esea ch and 321
inno aon p og amme unde g an ag eemen No. 101137601 322
Open Resea ch 323
The CMIP da a a e a ailable h ough ESGF's websi e a hps://esg -node.llnl.go /sea ch/cmip6/. The 324
BEST da a a e a ailable h ough he Be keley Ea h websi e a hps://be keleyea h.o g/da a/. The 325
HadSLP2 da a a e a ailable h ough he UK Me Office websi e a 326
hps://www.me office.go .uk/hadobs/hadslp2/. The GPCC da a a e a ailable h ough he Deu sche 327
Wee diens websi e a 328
hps://openda a.dwd.de/clima e_en i onmen /GPCC/h ml/download_ga e.h ml. 329
Re e ences 330
Allan, R., & Ansell, T. (2006). A new globally comple e mon hly his o ical g idded mean sea le el 331
p essu e da ase (HadSLP2): 1850-2004. Jou nal o Clima e, 19(22). 332
hps://doi.o g/10.1175/JCLI3937.1 333