Download - Whatever happened to PalaeoQUMP?
Whatever happened to PalaeoQUMP?Multi-palaeo archive constraints on climate sensitivity
Tamsin Edwards, Open UniversityJonty Rougier, University of BristolMat Collins, University of Exeter
Multi-palaeo archive constraints on climate sensitivity
Workshop questions
• How strongly can observations (including palaeo) constrain sensitivities?
• How consistent are different constraints and what can be inferred from multiple constraints?
- Consistency across models
- Proxy types
- Palaeoclimate eras
PalaeoQUMP 2006-2010• Palaeoclimate version of QUMP
- QUMP: Quantifying Uncertainty in Model Projections (UKCP09; David Sexton, Mat Collins and others)
- PI Sandy Harrison; Mat Collins, Michel Crucifix, and others
• Bayesian update of perturbed parameter ensemble (PPE) with palaeoclimate reconstructions to constrain climate sensitivity
- HadCM3
• Palaeoclimates
- Mid-Holocene (MH: 6000 cal yr BP)
- Last Glacial Maximum (LGM: 21,000 cal yr BP)
• Magnitude of change in tropical SST
MMEs and PPEs at the LGM
Other PPEs and MMEUpdated from Edwards et al. (2007) PiPG
MIROC: Annan et al. (2005); Hargreaves et al. (2006)CLIMBER: Schneider von Deimling (2006)MME: Crucifix (2006)
HadCM3PPE
Clim
ate
sens
itivity
(deg
C)
Magnitude of Last Glacial Maximum tropical SST anomaly (degC)
• Magnitude of change in tropical SST
MMEs and PPEs at the LGM
Other PPEs and MMEUpdated from Edwards et al. (2007) PiPG
MIROC: Annan et al. (2005); Hargreaves et al. (2006)CLIMBER: Schneider von Deimling (2006)MME: Crucifix (2006)
HadCM3PPE
Clim
ate
sens
itivity
(deg
C)
Magnitude of Last Glacial Maximum tropical SST anomaly (degC)
MARGO reconstruction
(a)!
Temperature anomaly (degrees C)!(b)!
(c)!
• Magnitude of change in tropical SST
MMEs and PPEs at the LGM
Other PPEs and MMEUpdated from Edwards et al. (2007) PiPG
MIROC: Annan et al. (2005); Hargreaves et al. (2006)CLIMBER: Schneider von Deimling (2006)MME: Crucifix (2006)
HadCM3PPE
Clim
ate
sens
itivity
(deg
C)
Magnitude of Last Glacial Maximum tropical SST anomaly (degC)
MARGO reconstruction
(a)!
Temperature anomaly (degrees C)!(b)!
(c)!
State-dependent feedback parameter
(Wm-2K-1)!
(Wm-2K-1 )!
MIROC: Hargreaves et al. (2006) MME: Crucifix (2006)
HadCM3PPE
State-dependent feedback parameter
(Wm-2K-1)!
(Wm-2K-1 )!
MIROC: Hargreaves et al. (2006) MME: Crucifix (2006)
HadCM3PPE
Mid-Holocene Green Sahara
pollen reconstruction
HadCM3 PPE
Mea
n pr
ecip
itatio
n ch
ange
(m
m/y
r)
New statistical framework• Rougier, Goldstein, and House (2013), “Second-order
exchangeability analysis for multi-model ensembles”, Journal of the American Statistical Association
- See also Williamson et al. (2013) Climate Dyn.
• Update MME mean and variance with observations
• Exchangeability of
- ensemble members
- reality with ensemble
• Simplest possible judgements about structural uncertainty: how much does MME under-sample?
1. inflation factor for variance: e.g. 1.5x
2. minimum variance: e.g. (0.5 degC)2
Palaeoclimate application• Joint update across MH and LGM
• MME:
- Palaeoclimate Model Intercomparison Project (PMIP)
- PMIP3/PMIP2V/PMIP2/most similar other model
• Reconstructions
- MH and LGM pollen-based growing degree days (GDD5) Bartlein et al. (2010)
- LGM multi-proxy sea surface temperatures (SST) MARGO et al. (2009); updates by Schmittner et al. (2011)
• Use for:
- critiquing judgements about structural uncertainty of MME mean
- Bayes Linear update of MME mean and estimated variance
Very preliminary update• mean = 3.0; > 95% credibility interval = [ 1.6 , 5.4 ]
• smaller if assume Gaussian distribution for reconstructions <->
Clim
ate
Sens
itivi
ty X
0
5
10
15
0
5
10
15MH GDD5
Initi
alU
pdat
ed
●3.2
● 6.1
LGM GDD5
Initi
alU
pdat
ed
●3.2● 2.9
LGM SST
Initi
alU
pdat
ed
●3.2 ● 3.1
All GDD5
Initi
alU
pdat
ed
●3.2 ● 3.0
All LGM
Initi
alU
pdat
ed
●3.2● 2.9
All
Initi
alU
pdat
ed
●3.2 ● 3.0●●●
●
●
●
●
●●
●●
●
●
Ense
mbl
e
PRIOR
Ensemble: 1.5-4.7
Very preliminary update• mean = 3.0; > 95% credibility interval = [ 1.6 , 5.4 ]
• smaller if assume Gaussian distribution for reconstructions <->
Clim
ate
Sens
itivi
ty X
0
5
10
15
0
5
10
15MH GDD5
Initi
alU
pdat
ed
●3.2
● 6.1
LGM GDD5
Initi
alU
pdat
ed
●3.2● 2.9
LGM SST
Initi
alU
pdat
ed
●3.2 ● 3.1
All GDD5
Initi
alU
pdat
ed
●3.2 ● 3.0
All LGM
Initi
alU
pdat
ed
●3.2● 2.9
All
Initi
alU
pdat
ed
●3.2 ● 3.0●●●
●
●
●
●
●●
●●
●
●
Ense
mbl
e
PRIOR
Update with each field
Ensemble: 1.5-4.7
Very preliminary update• mean = 3.0; > 95% credibility interval = [ 1.6 , 5.4 ]
• smaller if assume Gaussian distribution for reconstructions <->
Clim
ate
Sens
itivi
ty X
0
5
10
15
0
5
10
15MH GDD5
Initi
alU
pdat
ed
●3.2
● 6.1
LGM GDD5
Initi
alU
pdat
ed
●3.2● 2.9
LGM SST
Initi
alU
pdat
ed
●3.2 ● 3.1
All GDD5
Initi
alU
pdat
ed
●3.2 ● 3.0
All LGM
Initi
alU
pdat
ed
●3.2● 2.9
All
Initi
alU
pdat
ed
●3.2 ● 3.0●●●
●
●
●
●
●●
●●
●
●
Ense
mbl
e
LGM SST change mostly < 3sigma from zero (grey)
PRIOR
Update with each field
Ensemble: 1.5-4.7
Very preliminary update• mean = 3.0; > 95% credibility interval = [ 1.6 , 5.4 ]
• smaller if assume Gaussian distribution for reconstructions <->
Clim
ate
Sens
itivi
ty X
0
5
10
15
0
5
10
15MH GDD5
Initi
alU
pdat
ed
●3.2
● 6.1
LGM GDD5
Initi
alU
pdat
ed
●3.2● 2.9
LGM SST
Initi
alU
pdat
ed
●3.2 ● 3.1
All GDD5
Initi
alU
pdat
ed
●3.2 ● 3.0
All LGM
Initi
alU
pdat
ed
●3.2● 2.9
All
Initi
alU
pdat
ed
●3.2 ● 3.0●●●
●
●
●
●
●●
●●
●
●
Ense
mbl
e> 95% credibility interval
[1.6 , 5.4]Gaussian assumption:
approx. [2, 4.5]!
PRIOR
Update with each field
Ensemble: 1.5-4.7
MTWA MH anomalies in AR5
“Terrestrial MH summer-season temperatures were higher than modern in the mid-to-high latitudes of the NH, consistent with...PMIP2 and PMIP3/CMIP5 simulated responses to orbital forcing”
“Pollen-based records indicate positive MH temperature anomalies in southern South Africa that are not reproduced in the PMIP3/CMIP5 simulations.”
Tentative summary• Importance of using multiple datasets for robustness• State-dependent lambda, structural uncertainty & constraint• Is it coherent to combine different eras?- LGM temps (except E Antarctic) support lower ECS - MH land temp & Sahara precip support higher
• What would narrow the range in this study?- Better agreement between MH reconstructions and MME mean- Smaller LGM SST uncertainties- Smaller MME spread if better agreement of mean- New constraints e.g. historical?
• Why do GCMs systematically overestimate NH summer warming (underestimate monsoon response), esp low sens.?- What implications does this have for future precip projections?