A icle h ps://doi.o g/10.1038/s41467-025-64091-w
Influence o indi idual models and s udies on
quan i a i e mi iga ion findings in he IPCC
Six h Assessmen Repo
Ida Sognnaes & Glen P. Pe e s
Quan i a i e mi iga ion findings based on emissions scena ios submi ed o
he In e go e nmen al Panel on Clima e Change (IPCC) play an au ho i a i e
ole in clima e policy and decision making. We analyse he impac o he
une en ep esen a ion o models and modelling s udies in he IPCC Six h
Assessmen Repo (AR6) on s a is ical alues ha a e used o p esen quan-
i a i e mi iga ion findings. We find ha se e al key AR6 findings a e influ-
enced conside ably by he model wi h he mos scena ios, including emissions
educ ions by 2030 and he decline in ossil uels consis en wi h 1.5 °C, and we
find ha he yea o ne -ze o g eenhouse gas emissions is influenced con-
side ablybybo h hemodeland hes udy wi h he mos scena ios. We find
ha weigh ing by model- o s udy does no p o ide a s aigh o wa d solu ion
and discuss h ee issues ela ed o he use o da abase s a is ics o p esen
emissions scena ios findings. In o med by he pu pose o he IPCC and he
kinds o insigh s ha can be ob ained om emissions scena ios, we sugges
imp o emen s o he assessmen o emissions scena ios.
Quan i a i e findings in he In e go e nmen al Panel on Clima e
Change (IPCC) Wo king G oup III (WGIII) epo s on clima e mi iga-
ion ely s ongly on scena ios in he IPCC scena ios da abases1–6.
Desc ip i e da abase s a is ics, including median alues and in e -
qua ile o 5 h-95 h pe cen ile anges, a e used o epo key findings,
including emissions educ ions o e ime, he yea o ne -ze o
emissions, and he educ ions in ossil uels consis en wi h di e en
empe a u e a ge s2. These findings play au ho i a i e oles and
ha e been used o in o m clima e nego ia ions and clima e policy a
in e na ional and na ional le els7. Inc easingly, hey a e also used by
o he ac o s, including local go e nmen s, p i a e companies, banks,
and financial egula o s o in o m ne -ze o s a egies8, e alua e
alignmen wi h clima e a ge s8, and o assess and disclose clima e-
ela ed financial isks8–10.
The emissions scena ios in he IPCC scena ios da abases a e
gene a ed almos exclusi ely by p ocess-based In eg a ed Assessmen
Models (IAMs). The collec ion is based on he olun a y submission o
scena ios ha ha e been published in indi idual modelling s udies o
in mul i-model s udies11,12. Because some modelling g oups publish and
submi mo e scena ios, he numbe o scena ios om di e en models
and s udies in he da abase is no e en12,13. Fo his eason, he IPCC
scena ios da abases a e e e ed o as “ensembles o oppo uni y”14.In
he Six h Assessmen Repo (AR6), he ou models wi h he mos
scena ios a e esponsible o wo- hi ds o he scena ios ha passed
e ing and ecei ed a clima e assessmen , and he s udy wi h he mos
scena ios is esponsible o almos hal o he scena ios (Fig. 1).
Because scena io ou comes depend on model and s udy
assump ions15–18, median alues and anges may be sensi i e o he
sampling o models and s udies in he da abase. Se e al ecen s udies
ha e ound model di e ences o be an impo an d i e o scena io
ou come a ia ion19,20, and ha e iden ified dis inc ‘model finge p in s’21.
Ye , he impac s o he une en ep esen a ion o models and s udies on
findings p esen ed in he IPCC epo s ha e no ye been quan ified.
Al hough i is no ed in Chap e 3 o he IPCC WGIII epo ha unco -
ec ed da abase s a is ics may be misleading11, such s a is ics a e s ill
used o headline AR6 findings.
We fi s analyse he influence o indi idual models and s u-
dies on da abase s a is ics ha a e used o epo key findings in
he AR6 WGIII Summa y o Policymake s22 (SPM). Secondly, we
assess heimpac o dominan modelsonmedianou comesin he
Recei ed: 23 Decembe 2024
Accep ed: 8 Sep embe 2025
Check o upda es
CICERO Cen e o In e na ional Clima e Resea ch, Oslo, No way. e-mail: [email p o ec ed]o.no
Na u e Communica ions | (2025) 16:8343 1
1234567890():,;
1234567890():,;
AR6 scena ios da abase o e all. Thi d, we compu e model- and
s udy-weigh ed medians and show ha he esul s depend on he
choice o weigh ing, in addi ion o he ep esen a ion o models
and scena ios. Finally, based on ou findings, we discuss mo e
undamen al issues wi h he use o da abase s a is ics o p esen
emissions scena ios findings. In o med by he pu pose o he
IPCC and o emissions scena ios, we sugges ways o wa d o he
IPCC au ho s o imp o e he assessmen and communica ion o
emissions scena ios findings.
Resul s
Influence o indi idual models and s udies on key AR6 mi iga-
ion findings
To assess he influence o indi idual models and s udies on key AR6
findings, we calcula e he impac on findings epo ed in he WGIII
SPM om emo ing each model and s udy one-by-one. WGIII SPM
findings a e epo ed using p ima ily median alues, wi h ei he
in e qua ile o 5-95 h pe cen ile anges in pa en hesis (see e.g., SPM
Table SPM.2, o SPM pa ag aphs C.1.2 o C.3.2 in he AR6 WGIII
epo ). Since he WGIII SPM is ocused on scena ios ha achie e
1.5 °C wi hou o e shoo (C1 ca ego y) and 2 °C (C3 ca ego y), we
ocus on 1.5 °C wi hou o e shoo and use 2 °C scena ios as a e e -
ence o assess he magni ude o he impac s. We analyse a o al o 27
WGIII SPM findings (see Me hods o he selec ion) and measu e he
impac o each indi idual model and s udy in wo di e en ways:
Fi s , we measu e he impac o emo ing he indi idual model and
s udy on he 1.5 °C median ela i e o he di e ence be ween he
1.5 °C and 2 °C medians, and second, we measu e how close he 1.5 °C
median is o he indi idual model and s udy ela i e o he median
o all he o he models and s udies. The fi s measu e cap u es he
impac o scena io sampling compa ed o he impac o swi ching
om one clima e a ge (1.5 °C) o ano he (2 °C). The second mea-
su e cap u es he influence o indi idual models and s udies ela i e
o all o he models and s udies wi hin he same clima e ca ego y (see
Me hods o de ails).
The impac on median alues is subs an ial o se e al key AR6
findings (Fig. 2,Table1). F om emo ing jus one model om he AR6
ensemble, median 2030 g eenhouse gas (GHG) educ ions in 1.5 °C
scena ios ( ela i e o 2019)—a widely ecognised a ge 23 w i en in o
he 2022 Sha m el-Sheikh Implemen a ion Plan24—shi s om 43% o
50%. Median 2030 CO
2
educ ions shi om 48% o 56%. Median coal
and gas educ ions in 2050 shi om 95% o 83% and 43% o 29%,
espec i ely. Ne -ze o GHG yea shi s om 2098 o 2086 when
emo ing one model, o 2084 when emo ing one s udy, and o a e
2100 ( he yea o ne -ze o GHG is no specified in hese scena ios)
when emo ing se e al o he models and s udies.
Fo mos o he assessed WGIII SPM findings, indi idual models
ha e a much la ge impac han indi idual s udies (Table 1,Fig.3and
Supplemen a y Figs. 1–3). One excep ion o his is he ne -ze o GHG
yea . Fo he ne -ze o GHG yea , he ENGAGE s udy, wi h 25 o 97
scena ios in he C1 ca ego y, has a la ge influence because many
ENGAGE scena ios do no each ne -ze o GHG emissions be o e 2100
by design (Fig. 2and Supplemen a y No e 1). The ne -ze o GHG yea
is also sensi i e o he emo al o models and s udies because ew
scena ios each ne -ze o a ound he median yea (2098) (Fig. 4). The
eason why median alues a e mo e sensi i e o indi idual models is
pa ly because he dominan model is esponsible o a la ge sha e
o he C1 scena ios (41 o 97 scena ios) han he dominan s udy (25
o 97 scena ios). Models wi h ewe han 25 scena ios also ha e a
smalle impac (Supplemen a y Figs. 1–3). In addi ion o his, models
ha e a la ge impac on median alues because some WGIII SPM
findings a e mo e model-dependen han hey a e s udy-dependen .
As seen in he cases o coal, oil, gas, CH
4
,N
2
O, and F-Gases, he le el
o disag eemen ac oss models is la ge han he le el o disag ee-
men ac oss s udies (Supplemen a y Figs. 1–3). This is consis en
wi h o he s udies ha ha e ound model di e ences o be la ge
han o he scena ios di e ences15,18,20,21.
The model wi h he la ges impac on mos median alues is he
model wi h he mos scena ios, which, o all he assessed findings, is
he REMIND model (Table 1). The s ong influence on median alues
o his model is explained in la ge pa due o i s subs an ial sha e o
he 1.5 °C scena ios in AR6 (41 o 97 scena ios). Because REMIND uses
less coal and mo e oil compa ed o o he models, mediancoalis highe
and median oil is lowe when REMIND is emo ed. And because nea -
e m emissions educ ions in REMIND a e lowe han in mos o he
models, median GHG and CO
2
educ ions in 2030 a e highe when
REMIND is no included. The model wi h he second la ges impac
a e REMIND is he model wi h he second mos scena ios, MESSAGE
(20 o 97 scena ios). Excep o he ne -ze o GHG yea , howe e , he
impac o emo ing MESSAGE is ela i ely small (Table 1). The magni-
ude o he shi when emo ing models depends on bo h he numbe
o scena ios and he posi ion o hose scena ios ela i e o o he sce-
na ios (Fig. 4).
The s udy wi h he la ges impac on median alues is also he
s udy wi h he mos scena ios, ENGAGE18 (Table 1). And he s udy wi h
he la ges influence a e ENGAGE is also he s udy wi h he second
mos scena ios, an Vuu en 202125,26. This s udy also happens o con-
ain all he IMAGE scena ios in he C1 ca ego y, and emo ing i is
he e o e equi alen o emo ing he IMAGE model. The impac o an
Vuu en 2021 on median alues is, howe e , ela i ely small (Table 1).
In e qua ile and 5-95 h pe cen ile anges a e also influenced by
he dominan model and he dominan s udy, wi h he in e qua ile
Fig. 1 | Rep esen a ion o models and s udies in he IPCC AR6 Scena ios
Da abase used o de i e quan i a i e emissions scena ios findings. a Numbe o
scena ios by model. bNumbe o scena ios by s udy. All he scena ios ha passed
e ing and ecei ed a clima e assessmen a e shown (1202 scena ios, ou o
2304 submi ed global scena ios12). Models a e g ouped in o ‘model amilies’
ep esen ing po en ially se e al e sions o a single model co e. The s udies co e
all scena ios in ha s udy, hough o en om single pape s. Scena io-based mi i-
ga ion findings in he AR6 WGIII epo a e based on hese scena ios (see “Me h-
ods”). Da a: IPCC AR6 Scena ios Da abase3.
A icle h ps://doi.o g/10.1038/s41467-025-64091-w
Na u e Communica ions | (2025) 16:8343 2
anges being mo e sensi i e han he 5-95 h pe cen ile anges, and wi h
he dominan model ha ing a la ge impac han he dominan
s udy (Fig. 3).
The median alues ha a e leas impac ed by indi idual models
and s udies ( ela i e o he di e ences in 1.5 °C and 2 °C medians)
include peak CO
2
and GHG emissions yea s and ne -ze o CO
2
yea
(Table 1). Because almos all scena ios ha e he same peak GHG and
CO
2
yea (2020), emo ing indi idual models and s udies has no
impac . Fo he ne -ze o CO
2
yea , in e -model a ia ion is s ill con-
side able, bu he wo models wi h he mos scena ios happen o bo h
be e y close o he ensemble median (Supplemen a y Table 1).
The e o e, emo ing hese wo models has almos no impac on he
median in his case. Models ha gi e significan ly di e en ne -ze o
yea s (such as GCAM, POLES, C-ROADS, and COFFEE) do no ha e
enough scena ios o influence he median ne -ze o yea on hei own.
This highligh s ha i is no only he numbe s o scena ios ha ma e ,
bu also hei posi ioning ela i e o he median.
O e all, he la ges influence on he assessed WGIII SPM findings
comes om he dominan model (Table 1). Compa ed o he dominan
model, he dominan s udy and o he models ha e much less influence
(Supplemen a y Figs. 1–3). Fo 16 o he 27 SPM findings, he epo ed
median is close o he median o he dominan model han o he
median o all he o he models combined (Table 1). Fo coal educ ions,
yea o ne -ze o GHG emissions, CH
4
and F-gas emissions, he impac
om he dominan model on he medianisasla geaso la ge han he
dis ance be ween epo ed median alues o 1.5 °C and 2 °C scena ios.
This shows ha he une en ep esen a ion o models can be as o mo e
impo an han he clima e ca ego y o median scena io ou comes.
Influence o dominan models on he AR6 scena ios da abase
o e all
We ha e so a ocused on he scena io a iables epo ed in he WGIII
SPM. To assess he influence o dominan models on he AR6 scena ios
da abase findings o e all, which is ele an o use s o he da abase
mo e b oadly7,10,13,27,11, we compu e he impac on he median alues o
all he a iables in he da abase (see “Me hods”). In each clima e
ca ego y, we compu e he numbe o a iables o which he median
(in 2050) is close o he median o he dominan model han o he
median o all he o he models combined. We do his o all scena io
a iables epo ed by a leas wo models.
Mos median alues in he AR6 da abase a e close o he median
o he dominan model han o he median o he o he models aken
oge he (Fig. 5). In scena ios ha limi global wa ming o 1.5 °C (C1
ca ego y), 79% o median alues a e close o he median o he
Fig. 2 | Impac o emo ing indi idual models and s udies on selec ed AR6
WGIII SPM findings. Changes in median alues o aYea o ne ze o GHG emis-
sions by model, bYea o ne ze o GHG emissions by s udy, cGHG emissions in
2030 by model, dCoal in 2050 by model, and eGas in 2050 by model. Boxes show
he minimum and maximum, he in e qua ile anges, and he median o each
model/s udy. The numbe o scena ios om each model/s udy is shown a he op
o each box, and he da a poin s a e shown o he le . Models and s udies a e
o de ed acco ding o he numbe o scena ios, wi h he model/s udy wi h he mos
scena ios u hes o he le . Long, solid ho izon al lines show median alues when
models (o ange) and s udies (yellow) a e emo ed one-by-one, wi h le e s a he
end o each line indica ing he model/s udy ha has been emo ed. Bolded and
s a ed le e s show he model/s udy whose emo alleads o he bigges shi in he
median alue. Dashed g ey ho izon al lines show ensemble medians. The findings
a e selec ed based on a combina ion o policy ele ance and impac (shown in
Table 1) o illus a e how indi idual models and s udies a ec median alues (mo e
findings a e shown in Supplemen a y Figs. 1–3). All a iables a e om scena ios ha
limi global wa ming o 1.5 °C (>50%) (C1 ca ego y). Values abo e 2100 on he y-axis
indica e ‘a e 2100’, which means ze o was no eached be o e 2100 (and may no
be eached). The model ac onyms a e R: REMIND, M: MESSAGE, W: WITCH, I:
IMAGE, GC: GCAM, A: AIM, P: POLES, CR: C-ROADS, GE: GEM-E3, CO: COFFEE and
he s udy ac onyms a e E: ENGAGE, V: an Vuu en 2021, E33: EMF33, N: NGFS2, C:
CD-LINKS, K21: Kiks a 2021, O21: Ou 2021, SSP: SSP, S18: S efle 2018, S21a: S efle
2021a, S21b: S efle 2021b, A: ADVANCE, B21: Baums a k 2021, H18: Holz 2018, K18:
K iegle 2018, So21: Soe gel 2021, L21: Lude e 2021, B18: Be am 2018, F20: Fuji-
mo i 2020, G18: G uble 2018, S21: Schul es 2021. Da a: IPCC AR6 Scena ios
Da abase3.
A icle h ps://doi.o g/10.1038/s41467-025-64091-w
Na u e Communica ions | (2025) 16:8343 3
Table. 1 | Impac o emo ing indi idual models and s udies on median alues in he AR6 WGIII SPM
Scena io findingsa1.5 °CbModel emo ed(1.5 °C) S udy emo ed(1.5 °C) 2 °Cc
SPM
Sou ce
Median New Median Model Impac mea-
su es (%)e
New
Median
S udy Impac measu es (%)eMedian
Wi hin Be ween Wi hin Be ween
F-gases, 2050 (%) C.1.2 88 79 R 99 207 88
88
V*
E
50
1
5
1
83
CH
4
,2050(%) C.1.2 51 59 R 66 182 49 E 22 47 46
Ne ze o GHG (y ) Table
SPM.2
2098 2086
>2100
M*
R
63
37
171
100
2084 E 67 200 >2100
Coal, 2050 (%) C.3.2 95 83 R 77 99 96
96
V*
E
10
58
7
4
83
N
2
O, 2050 (%) C.1.2 26 31 R 19 82 23 E 41 37 19
CH
4
,2050(%) C.1.2 44 48 R 69 66 43 E 14 10 37
Gas, 2050 (%) C.3.2 43 29 R 43 50 48
42
V*
E
7
69
18
3
16
Oil, 2050 (%) C.3.2 61 72 R 52 38 60
62
K21*
E
6
25
5
1
32
Oil wo. CCS in 2050 (%) C.3.2 61 71 R 51 36 57
61
V*
E
34
17
17
1
34
CO
2
,2030(%) C.1.2 48 56 R 65 31 46 E 23 7 22
GHG, 2030 (G CO
2
-eq/y ) Table
SPM.2
31 28 R 69 30 30
32
V*
E
42
13
8
4
44
GHG, 2030 (%) Table
SPM.2
43 50 R 65 29 46
43
V*
E
44
8
11
2
21
Coal wo. CCS in 2050 (%) C.3.2 98 97 R 84 29 99
98
V*
E
9
12
3
1
93
T anspo - ela ed CO
2
,
2050 (%)
C.8.1 59 53
62
M*
R
64
26
19
11
57
58
V*
E
9
8
6
0
29
Cumula i e ne -nega i e
CO
2
(G CO
2
)g
Table
SPM.2
−215 −188 R 57 16 −253 E 20 22 −40
Gas wo. CCS in 2050 (%) C.3.2 68 64 R 41 15 66
68
S21a*
E
15
10
7
1
38
Cumula i e CO
2
(G CO
2
)hTable
SPM.2
512 461 R 60 14 519
515
N*
E
10
7
2
1
888
GHG, 2040 (G CO
2
-eq/y ) Table
SPM.2
17 16 R 60 13 17
17
S21a*
E
21
69
3
3
29
GHG, 2040 (%) Table
SPM.2
69 72 R 51 12 70
70
S21a*
E
22
49
4
3
46
GHG, 2050 (%) Table
SPM.2
84 86 R 57 10 85 E 38 6 64
GHG, 2050 (G CO
2
-eq/y ) Table
SPM.2
98 R549 8 E335 20
CH
4
, 2030 (%) C.1.2 34 33
35
M*
R
43
46
9
8
33 E 41 9 19
CO
2
, 2040 (%) C.1.2 80 82 R 33 6 79
80
N*
E
20
11
3
0
51
Ne ze o CO
2
(y ) Table
SPM.2
2052 2051
2052
M*
R
100
33
5
3
2051 E 25 5 2071
Cumula i e CO
2
(G CO
2
)iTable
SPM.2
324 307
311
M*
R
76
47
4
3
259 E 52 14 797
Peak CO
2
(y ) Table
SPM.2
2020 2020 R 0 0 2020 E 0 0 2020
Peak GHG (y ) Table
SPM.2
2020 2020 R 0 0 2020 E 0 0 2020
The able is so ed acco ding o he ‘Be ween’measu e o he models. Numbe sa e ounded o henea es digi . Figu e 3shows he op 13 findings ha a e mos impac ed acco ding o hismeasu e
o ei he models o s udies ( alues >21%, bolded). See he expanded e sion o his able (Supplemen a y Table 1) o median alues o he indi idual models and s udies ha a e emo ed.
Models ac onyms a e R REMIND, M MESSAGE. S udy ac onyms a e E ENGAGE, V an Vuu en 2021, K21 Kiks a 2021, S21a S efle 2021a, N NGFS2.
aPe cen ages deno e educ ions om 2019 (nega i e numbe s a e inc eases). Coal, oil, and gas a e in p ima y ene gy.
b1.5 °C (>50%) wi h no o limi ed o e shoo (C1 ca ego y).
c2 °C (>67%) (C3 ca ego y).
dThe columns show he median and impac measu es when he indi idual model/s udy wi h he la ges impac is emo ed. When he model/s udy wi h he la ges impac is di e en om he model/
s udy wi h he mos scena ios, he impac om he la e is shown in he second ow. Models and s udies ha di e om he dominan model and s udy a e ma ked wi h an as e isk.
eThe impac s o emo ing indi idual models and s udies a e measu ed in wo di e en ways. ‘Be ween’isa uni less measu e o he change in median alue ela i e o hedi e ence be ween he 1.5 °C
and 2 °C medians and ‘Wi hin’is a uni less measu e o how close he epo ed median is o he median o he emo ed model/s udy e sus he median o all he o he models/s udies. Highe alues
deno e highe impac (see “Me hods” o de ails).
‘>2100’means a e 2100. Impac measu es may be o e - o unde es ima es because he yea can be ea lie o much la e ( he yea 2105 is used o calcula e he impac ).
gBe ween he yea o ne ze o CO
2
and 2100.
hBe ween 2020 and he yea o ne ze o CO
2
.
iBe ween 2020 and 2100.
A icle h ps://doi.o g/10.1038/s41467-025-64091-w
Na u e Communica ions | (2025) 16:8343 4
Fig. 3 | Impac o emo ing he indi idual model and s udy wi h he mos
scena ios on selec ed AR6 WGIII SPM findings. a Yea o ne ze o GHG emis-
sions, bCumula i e ne -nega i e CO
2
emissions be ween he yea o ne ze o and
2100, cCO
2
emissions in 2030, dGHG emissions in 2030, eCH
4
emissions in
2040, CH
4
emissions in 2050, gN
2
O emissions in 2050, hF-gas emissions in
2050, iCoal in 2050, jCoal wi hou CCS in 2050, kOil in 2050, lOil wi hou CCS
in 2050, mGas in 2050. The selec ion ep esen s he findings ha a e mos
impac ed acco ding o he ‘Be ween’measu e o ei he models o s udies, as
shown in Table 1. The di e en yea s and anges co espond o wha is epo ed
in he SPM: Fo ossil a iables, in e qua ile anges a e shown; o o he a i-
ables, 5 h-95 h pe cen ile anges a e shown. The g ey ba s show s a is ics o all
scena ios ha limi global wa ming o 1.5 °C (>50%) (ca ego y C1, including only
e ed scena ios ha ecei ed a clima e assessmen ). The o he ba s show s a-
is ics when he model wi h he mos scena ios is emo ed (o ange ba s) and
when he s udy wi h he mos scena ios is emo ed (yellow ba s). The blue dia-
monds show median alues o all scena ios ha limi global wa ming o 2 °C
( > 67%) (ca ego y C3, including only e ed scena ios ha ecei ed a clima e
assessmen ). The model wi h he mos scena ios is he REMIND model, and he
s udy wi h he mos scena ios is he ENGAGE s udy o all he findings. ‘>2100’
means ‘a e 2100’( he yea is no specified o scena io ha each ne ze o GHG
emissions a e 2100), ‘CCS’s ands o Ca bon Cap u e and S o age. Da a: IPCC
AR6 Scena ios Da abase3.
A icle h ps://doi.o g/10.1038/s41467-025-64091-w
Na u e Communica ions | (2025) 16:8343 5
dominan model han o he median o all he o he models. In sce-
na ios ha limi global wa ming o 2 °C (C3 ca ego y), 64% o median
alues a e close o he median o he dominan model han o he
median o all he o he models. Ac oss all clima e ca ego ies, 67% o
median alues in he AR6 scena ios da abase a e close o he median
o he dominan model han o he median o all he o he models.
Wi hin each clima e ca ego y, he dominan model depends on
he scena io a iable. This is because di e en models epo di e en
scena io a iables (Supplemen a y Fig. 4 and Supplemen a y Da a 1).
In scena ios ha limi global wa ming o 1.5 °C and 2 °C, he mos
common dominan model ac oss a iables is REMIND, ollowed by
MESSAGE and IMAGE.
Di e en models domina e ac oss a iables in di e en clima e
ca ego ies (Fig. 5). Whe eas REMIND is he mos common dominan
model in he C1, C3, C7, and C8 ca ego ies, MESSAGE is he mos
common dominan model in he C2, C4, C5, and C6 ca ego ies. This
means ha when compa ing median ou comes o a iables ac oss
clima e ca ego ies—which is o en done o show he implica ions o
di e en clima e a ge s— he di e ences may in some cases be mo e
eflec i e o di e ences in model sampling han o he clima e a ge .
Fo example, compa ing he implica ions o di e en le els o ‘o e -
shoo ’(C1 e sus C2) may be mo e abou di e ences in he REMIND
and MESSAGE models han abou o e shoo .
O e all, he models wi h he mos scena ios in he AR6 da abase
ha e a much la ge impac on median alues han he models wi h
ewe scena ios. The wo models wi h he mos scena ios (47% o all
e ed and clima e-assessed scena ios), REMIND and MESSAGE, a e
esponsible o 77% o he cases whe e he median is close o he
dominan model han o he median o all he o he models. Fo 96%
o he cases, he esponsible model is one o he ou models wi h
he mos scena ios in he AR6 scena ios da abase (REMIND, MES-
SAGE, IMAGE, and WITCH). The dis ibu ion o dominan models
(Fig. 5b) is hus e en mo e une en han he dis ibu ion o scena ios
pe model (Fig. 1a).
The esul s a e simila i we limi he analysis o include only Tie 1
and Tie 2 a iables, which a e conside ed he mo e impo an a i-
ables (see Me hods and Supplemen a y Da a 2). All models should
submi Tie 1 a iables (o which he e a e 82), sugges ing ha he
une en submissions ac oss a iables should be less common. In he C1
ca ego y, 63% o Tie 1 and 73% o Tie 2 a iables ha e median alues
ha a e close o he median o he dominan model han o he
median o all he o he models (Fig. 5). In he C3 ca ego y, 48% o Tie 1
and 59% o Tie 2 a iables ha e median alues ha a e close o he
median o he dominan model han o he median o all he o he
models. Ac oss all clima e ca ego ies, 47% o Tie 1 a iables and 65% o
Tie 2 a iables ha e median alues ha a e close o he median o he
dominan model han o he median o all he o he models. In mos
cases, he dominan model is one o he ou models wi h he mos
( e ed and clima e-assessed) scena ios in he AR6 scena ios da abase.
Models o he han REMIND, MESSAGE, IMAGE and WITCH a e he
dominan model in less han 5% o all cases.
We expec models ha epo mo e a iables (Supplemen a y
Fig. 4) obe he dominan model o mo e scena io a iables. Howe e ,
models wi h mo e scena ios end o also epo mo e a iables, which
means ha hese wo e ec s go in he same di ec ion (Supplemen a y
Fig. 5 and Supplemen a y Da a 1). O e all, he numbe o e ed sce-
na ios pe model (Fig. 1and Supplemen a y Fig. 6) is a be e p edic o
o model dominance (Fig. 5) han he numbe o a iables epo ed
(Supplemen a y Fig. 4 and Supplemen a y No e 2).
Median alues depend on he choice o weigh ing and sampling
The median alues p esen ed in he AR6 WGIII epo a e compu ed by
gi ing each scena io equal weigh . A simple way o coun e ac he
une en numbe o scena ios om di e en models and s udies is o
ins ead gi e each model o s udy equal weigh (see Me hods o
de ails). Fo he assessed WGIII SPM findings, he di ec ion o change in
median alues when each model o each s udy is gi en equal weigh is
usually he same as when he dominan model o s udy is emo ed
Fig. 4 | Dis ibu ion o ou comes and impac on median alues o emo ing he
model and s udy wi h he mos scena ios o selec ed AR6 WGIII SPM findings.
aYea o ne ze o GHG emissions by model, bYea o ne ze o GHG emissions by
s udy, cGHG emissions in 2030 by model, dCoal in 2050 by model, and eGas in
2050 by model. Dominan model/s udy scena ios a e shown in blue, and all o he
scena ios a e shown in o ange/yellow. The findings a e selec ed based on a com-
bina ion o policy ele ance and impac (shown in Table 1). All a iables a e om
scena ios ha limi global wa ming o 1.5 °C ( > 50%) (C1 ca ego y). The boxes a he
op o each his og am show he 5 h,25
h,50
h,75
h,and95
h pe cen iles o all sce-
na ios (in g ey), he median o he dominan model/s udy (in blue), and he median
wi hou he dominan model/s udy (in o ange/yellow). ‘>2100’means a e 2100
(ne ze o is no eached be o e 2100 and may no be eached in hese scena ios).
Da a: IPCC AR6 Scena ios Da abase3.
A icle h ps://doi.o g/10.1038/s41467-025-64091-w
Na u e Communica ions | (2025) 16:8343 6
(Fig. 6). This is no su p ising, gi en he scena ios om he dominan
model and s udy a e gi en much less weigh in model- and s udy-
weigh ed medians. When each model is gi en equal weigh , median
coal educ ions in 2050 changes om 95% o 83% in 1.5 °C scena ios,
which is he same as when emo ing he dominan model, and median
gas educ ions changes om 43% o 33%, which is close o he change
om emo ing he dominan model (29%). Median GHG educ ions in
2030 changes om 43% o 45%, which is less han when emo ing he
dominan model (50%), because he models a e e enly dis ibu ed on
each side o he median o his a iable.
Weigh ed medians, howe e , do no o e a s aigh o wa d
solu ion o he p oblem o model and s udy ep esen a ion. The e a e
wo easons o his. Fi s , model- and s udy-weigh ed medians o en
mo e in opposi e di ec ions, leading o la ge di e ences o ce ain
key findings (Fig. 6and Supplemen a y Figs. 7–9). The median ne ze o
GHG yea a ies by almos wo decades o mo e, depending on whe-
he you gi e each scena io equal weigh (2098), each s udy equal
weigh (2085), o each model equal weigh (a e 2100). This is pa ly
because he dominan model and he dominan s udy si on opposi e
sides o he median. This shows ha he choice o weigh ing scheme
can be a key de e minan o median alues. Bu i is no clea wha
weigh ing scheme is mo e app op ia e. Scena io ou comes can
depend on model o s udy assump ions, o bo h, and i is no clea
whe he i is models o s udies (o scena ios) ha should be gi en
equal weigh in he calcula ion o da abase s a is ics. Second, model-
and s udy-weigh ed medians may be dependen on wha models and
s udies a e included and no included in he AR6 scena ios da abase.
While model- and s udy-weigh ed medians a e insensi i e o he
numbe o scena ios, hey a e no insensi i e o he ep esen a ion o
models and s udies. No all models submi ed scena ios o he AR6
da abase, and o he mo e han 50 models ha submi ed scena ios,
only 13 models submi ed scena ios ha passed e ing and ecei ed a
clima e assessmen 11 (Supplemen a y No e 1 and Supplemen a y
Table 2). Fu he mo e, because no all models epo all a iables in all
clima e ca ego ies, mos findings a e based on e en ewe han hose
13 models (Supplemen a y Fig. 10). Thus, he models ha a e used o
Fig. 5 | Scena io a iables wi h median alues close o he dominan model
han o all o he models. The le column shows he numbe o a iables wi h
median alues close o he dominan model han o he median o all he o he
models, by clima e ca ego y and by model (a,c,e). The igh column shows he
o al numbe o a iables ac oss all clima e ca ego ies wi h median alues close o
he dominan model han o he median o all he o he models, by model (b,d, ).
Resul s a e shown o all a iables (a,b), Tie 1 a iables (c,d), and Tie 2 a iables
(e, )(see“Me hods”). Pe cen ages abo e ba s in (a,c,e) show he p opo ion o
a iables ineachclima e ca ego y wi h median alues close o he dominan model
han o he median o all o he models. The colou s show wha model is he
dominan model. Pe cen ages abo e ba s in (b,d, ) show he p opo ion o a i-
ables o which he median is close o he dominan model han o all he o he
models, by he dominan model. Median alues a e in 2050. Only a iables epo ed
by a leas wo models in he AR6 Scena ios Da abase a e included (see Me hods).
Da a: IPCC AR6 Scena ios Da abase3.
A icle h ps://doi.o g/10.1038/s41467-025-64091-w
Na u e Communica ions | (2025) 16:8343 7
de i e AR6 findings ep esen only a small subse o he models in he
scena ios li e a u e.
The p oblem o une en sampling is no unique o he IPCC sce-
na ios da abase. Unbalanced samples a e common in social sciences,
and se e al weigh ing me hods ha e been de eloped o deal wi h his.
These me hods, howe e , ely on knowing he a ge popula ion (see
e.g.,28,29.). When he a ge popula ion is known, weigh s can be
assigned o ensu e ha he dis ibu ion o chosen cha ac e is ics in he
sample, such as age, gende , educa ion, and geog aphical loca ion,
eflec s he a ge popula ion. Fo he IPCC scena ios ensemble,
howe e , he a ge popula ion is unclea . Should i be all plausible
scena ios, including ones ha ha e ne e been modelled, o should i
be all scena ios ha ha e been published? And wha scena io cha -
ac e is ics should one be ensu ing a good ep esen a ion o ? While
median ossil uel educ ions depend mos ly on model ep esen a ion,
he ne -ze o yea depends on bo h model and s udy ep esen a ion.
This sugges s ha he ele an cha ac e is ics depend on he scena io
ou come in ques ion.
Weigh ing me hods ha e also been discussed ex ensi ely in ela-
ion o physical clima e modelling30–33. Mos likely clima e ou comes
(con ingen on o cing le els) a e mos o en compu ed in s uc u ed
expe imen s whe e each clima e model is gi en equal weigh 34.In he
li e a u e, i has been a gued ha g ea e weigh should be gi en o
clima e models ha ha e been shown o ha e g ea e skill and o
models ha a e mo e independen 31,33. Tha is, clima e models should
be weighed based on bo h pe o mance and model independence,
whe e dependencies may s em om he sha ing o ideas o pa a-
me iza ion o simplifica ions, o om sha ing o compu e code,
which may lead o simila model biases31. Because di e en clima e
models ha e g ea e skill a simula ing di e en clima e ou comes, and
because di e en ou comes depend on di e en assump ions, bo h
pe o mance and independence depend on he ou come in ques ion33.
Despi e aluable sugges ions o how o e alua e IAMs35 he e a e
no ag eed upon pe o mance me ics o IAMs. While clima e model
pe o mance is based on compa ing model p ojec ions wi h his o ical
obse a ions33, obse a ions ha can be used o assess he pe o -
mance o IAMs a e no a ailable in he same way (Supplemen a y
No e 3). And esea ch in o IAM dependencies is lacking.
In lieu o pe o mance and independence me ics, weigh ing by
model migh ep esen an imp o emen o he cu en app oach o
weigh ing by scena io, as i emo es duplica ion o model ou comes
ha may esul om known model finge p in s. Gi ing each model
equal weigh is also he mos common app oach in clima e modelling.
Bu unlike clima e model ensembles, whe e each model is un once o
each scena io in a s uc u ed expe imen 34 (suchasinCoupledModel
In e compa ison P ojec s (CMIPs)), he IPCC IAM ensembles con ain a
Fig. 6 | Model- and s udy weigh ed medians o selec ed AR6 WGIII SPM find-
ings. a Yea o ne ze o GHG emissions by model, bYea o ne ze o GHG emissions
by s udy, cGHG emissions in 2030 by model, dCoal in 2050 by model, and eGas in
2050 by model. Boxes show he minimum and maximum, he in e qua ile anges,
and he median o each model/s udy. The numbe o scena ios om each model/
s udy is shown a he op o each box, and he da a poin s a e shown o he le .
Models and s udies a e o de ed acco ding o he numbe o scena ios, wi h he
model/s udy wi h he mos scena ios u hes o he le . Long, solid ho izon al
lines show median alues when models (o ange) and s udies (yellow) a e emo ed
one-by-one, wi h le e s a he end o each line indica ing he model/s udy ha has
been emo ed. Bolded and s a ed le e s show he model/s udy whose emo al
leads o he bigges shi in he median alue. Dashed g ey ho izon al lines show
ensemble medians. Longe dashed ho izon al lines show medians weigh ed by
model (o ange) and by s udy (yellow). The findings a e selec ed based on a com-
bina ion o policy ele ance and impac (shown in Table 1) o illus a e how indi-
idual models and s udies a ec median alues (mo e findings a e shown in
Supplemen a y Figs. 1–3). All a iables a e om scena ios ha limi global wa ming
o 1.5 °C (>50%) (C1 ca ego y). Values abo e 2100 on he y-axis indica e ‘a e 2100’,
which means ze o was no eached be o e 2100 (and may no be eached). The
model ac onyms a e R: REMIND, M: MESSAGE, W: WITCH, I: IMAGE, GC: GCAM, A:
AIM, P: POLES, CR: C-ROADS, GE: GEM-E3, CO: COFFEE and he s udy ac onyms a e
E: ENGAGE, V: an Vuu en 2021, E33: EMF33, N: NGFS2, C: CD-LINKS, K21: Kiks a
2021, O21: Ou 2021, SSP: SSP, S18: S efle 2018, S21a: S efle 2021a, S21b: S efle
2021b, A: ADVANCE, B21: Baums a k 2021, H18: Holz 2018, K18: K iegle 2018, So21:
Soe gel 2021, L21: Lude e 2021, B18: Be am 2018, F20: Fujimo i 2020, G18:
G uble 2018, S21: Schul es 2021. Da a: IPCC AR6 Scena ios Da abase3.
A icle h ps://doi.o g/10.1038/s41467-025-64091-w
Na u e Communica ions | (2025) 16:8343 8
di e si y o di e en scena ios, un by di e en se s o models, unde
di e en assump ions, o answe di e en ques ions. In his case, di -
e ences in ou comes do no ep esen di e en answe s o he same
ques ion, which may be in e p e ed as he unce ain y o he answe ,
bu how he answe s change when he ques ions change. In his case,
weigh ing by model emains an a bi a y app oach, which may also
emo e impo an a ia ion cap u ed by di e en s udies un by he
same model.
Discussion
We ha e shown ha se e al WGIII SPM median alues a e influenced
conside ably by he model, and in some cases s udy, wi h he mos
scena ios in he AR6 da abase. Addi ionally, we ha e shown ha
weigh ing does no o e a s aigh o wa d solu ion o he une en
ep esen a ion o models and s udies. The median will, in any case,
depend on he weigh ing choice and he ep esen a ion o models and
s udies. This b ings up mo e undamen al ques ions ega ding he use
o da abase s a is ics o p esen emissions scena ios findings.
In o med by he pu pose o he IPCC assessmen , we discuss h ee
issues and make ecommenda ions based on ou findings and he
kinds o insigh s ha can be ob ained om IAM scena ios.
Fi s , median alues and pe cen iles do no con ey he le el o
ag eemen in findings, which is key o in o ming confidence and
obus ness. The IPCC is se up o “ ell policymake s wha we know and
don’ know”and “whe e he e is ag eemen in he scien ific commu-
ni y, whe e he e a e di e ences o opinion, and whe e u he
esea ch is needed”36. As pa o his, assessmen findings wi h s onge
ag eemen and mul iple lines o e idence can be assigned a highe
deg ee o confidence37. While in e qua ile and 5 h-95 h pe cen ile an-
ges show he sp ead in scena io ou comes, hey don’ p o idein o -
ma ion abou whe he di e en modelsands udiesag eeo disag ee.
The e is ela i ely low ag eemen , o example, on he p ecise educ-
ions o emissions, coal, and gas in specific yea s, and on he ne -ze o
GHG yea . This is because hese ou comes clea ly depend on choices
and assump ions ha di e ac oss models and s udies (Figs. 2,3and 5).
As ou findings show, he epo ing o desc ip i e s a is ics o he
nea es pe cen age poin o emissions educ ions, he nea es 5% o
coal and gas, and he closes fi e-yea in e al o he ne -ze o yea is no
obus o he sampling o models and s udies. The epo ing o median
peak CO
2
and GHG emissions yea s is (Table 1). I he poin is o show
di e ences in implica ions be ween di e en clima e a ge s, obus
scena io findings can be defined as hose ha a y mo e ac oss clima e
ca ego ies han hey do ac oss models and s udies20. This obus ness
could be e alua ed, o example, by assessing he sensi i i y o ou -
comes o emo ing single models and s udies o o gi ing models and
s udiesequalweigh ,asisdonein hispape .
Ra he han ocusing on median alues and pe cen iles, which
mayno only be sensi i e o sampling bu can also be misin e p e ed as
p obabilis ic confidence in e als, he IPCC could epo he ull anges
o scena io ou comes and ocus mo e on how ou comes depend on
assump ions, including model and s udy assump ions. Many di e en
s a egies a e consis en wi h 1.5 °C and 2 °C, and scena io analysis is
la gely abou showing he di e en implica ions and ade-o s asso-
cia ed wi h di e en choices38. Scena io analysis is less sui ed o p o-
iding p ecise ou comes o specific a iables
38. In addi ion o his,
se e al models and s udies all en i ely ou side o he in e qua ile, and
in some cases, 5 h-95 h pe cen ile, anges ha a e used o p esen
findings (Supplemen a y Figs. 1–3). Bu gi en he non-p obabilis ic
na u e o he emissions scena io ensemble2and he ole o emissions
scena ios in explo ing di e en s a egies and ade-o s, hese esul s
a e no less impo an . An imp o ed assessmen o how scena io ou -
comes depend on choices and assump ions could be in o med by
se e al ecen s udies ha ha e analysed how he model used, he
scena io assump ions, and he clima e a ge a ec scena io
ou comes19–21. A ocus on such dependencies could p o ide policy-
ele an insigh s ega ding, o example, how he ne -ze o emissions
yea depends on key assump ions, including he clima e a ge defi-
ni ion, which is defined di e en ly in di e en s udies18, and he dis-
coun a e, which may be defined di e en ly in di e en models o
s udies39.
Second, he eliance on da abase s a is ics o p esen key sce-
na ios findings gi e a lo o weigh o he subse o scena ios ha a e
submi ed o he da abase, pass e ing, and ecei e a clima e assess-
men . This comes a he exclusion o findings ha a e cap u ed by
models and s udies ha a e no included in his subse bu s ill con-
ibu e o he li e a u e. This is an issue because he IPCC is mean o
assess he ull scena ios li e a u e, and he scena ios da abase is mean
o aid his assessmen , no eplace i . Wi h he cons uc ion and use o
he scena ios da abase o de i e mi iga ion findings, he IPCC is
essen ially conduc ing a me a-analysis o an “ensemble o
oppo uni y”2 ha includes only a subse o he scena ios li e a u e12.
Bu he AR6 is mean o40 “ ake all a ailable li e a u e on emissions
scena ios ully in o accoun independen ly o whe he unde lying
emissions scena ios a e submi ed o he AR6 scena io da abase”.To
se e his goal, he IPCC should assess whe he he subse o scena ios
used o de i e findings gi es a good ep esen a ion o he scena ios
li e a u e. Guidelines o he use o da abase s a is ics ha add ess he
issues o o e - and unde ep esen a ion could also be de eloped.
Making his an in eg al pa o he use o he IPCC scena ios da abase
may help educe biases in key findings ha a e based on he da abase.
Thi d, da abase s a is ics may no be e y meaning ul gi en he
a ied ep esen a ion o di e en esea ch ques ions and assump-
ions. Acco ding o AR6, “scena ios a e nei he p edic ions no o e-
cas s, bu a e used o p o ide a iew o he implica ions o
de elopmen s and ac ions”2. These iews a e p o ided h ough spe-
cific modelling s udies ha employ specific models and scena io
assump ions o answe specific esea ch ques ions, such as: Wha a e
he cos implica ions o mee ing s ingen clima e a ge s wi hou
o e shoo 18? Wha a e he impac s on elec ifica ion i enewable
ene gy cos s con inue o decline?41 How migh changes in ene gy
se ice p o ision a ec global ene gy demand and supply, and he
achie emen o clima e and de elopmen goals42? Al hough he sce-
na ios da abase has an ad an age in e ms o he numbe o models
and s udies and he e o e di e si y o assump ions, his di e si y also
means ha dis ibu ions o scena io ou comes a e di ficul o
in e p e 6and ha da abase s a is ics may no be e y meaning ul2,27.
Mo e eadily a ailable in o ma ion on model and scena io
assump ions1,13, g ea e openness and anspa ency in he cu a ion o
he IPCC scena ios da abase including on e ing12, and mo e esea ch
in o he di e en causes o scena io ou come a iabili y (e.g17.) would
enable use s o ge a be e unde s anding o he ep esen a ion o key
assump ions in he da abase, which could help hem make mo e
meaning ul compa isons and d aw mo e ele an insigh s sui ed o
hei needs. Bu his in o ma ion does no in i sel make he scena ios
in heda abasemo ecompa able
43 o s a is ical alues mo e mean-
ing ul. Modelling is o “insigh s no numbe s”44 and model ou comes
ca y less meaning when hey a e no in e p e ed wi h espec o he
choices and assump ions unde which hey we e gene a ed, and
he esea ch ques ions hey we e designed o answe . Because o his,
he insigh s ha IAM scena ios o e a e bes unde s ood and app e-
cia ed by assessing scena io pape s di ec ly.
The use o scena io da abase s a is ics o p esen key scena io
findings, and gi ing each scena io equal weigh , is an easy choice, bu i
is no a neu al choice as i gi es mo e weigh o choices and
assump ions embedded in models and s udies ha ha e a la ge
numbe o scena ios in he da abase. While he une en ep esen a ion
o s udies in he AR6 ensemble migh mean ha ce ain ques ions, and
he e o e ce ain answe s, a e o e ep esen ed, he une en ep e-
sen a ion o models migh mean ha ce ain mi iga ion s a egies a e
o e ep esen ed. Ou esul s confi m ha di e en models end o
A icle h ps://doi.o g/10.1038/s41467-025-64091-w
Na u e Communica ions | (2025) 16:8343 9
