A mul&'proxy model'tested NAO reconstruc&on for the last millennium Pablo Ortega (1,2), Flavio Lehner (3,4), Didier Swingedouw (5), Valerie MassonB DelmoDe (1), Christoph Raible (3,4), Mathieu Casado (1) and Pascal Yiou (1) (1) LSCE/IPSL, GifBsurBYveDe (France) (2) LOCEAN/IPSL, Paris (France) (3) KUP, University of Bern (Switzerland) (4) OCCR, University of Bern (Switzerland) (5) EPOC, Université Bordeaux I (France) Green Greenland AMA2015 / Workshop on storm-tracks, 21-23/01/2015
56
Embed
Amul&'proxymodel'testedNAO reconstruc&onforthelastmillennium · The"reconstruc&on"technique" It is based on a Principal Component Regression (PCR), that in essence represents a multivariate
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
34 from tree-rings 8 from ice cores 4 from lake sediments 2 from speleothems
Thanks to their annual resolution we can produce a yearly resolved NAO reconstruction in which to test the impact of the solar and volcanic forcings
The"reconstruc&on"technique"
It is based on a Principal Component Regression (PCR), that in essence represents a multivariate regression using the Principal Components of the standardised proxies as predictors
8&
The"reconstruc&on"technique"
It is based on a Principal Component Regression (PCR), that in essence represents a multivariate regression using the Principal Components of the standardised proxies as predictors To identify which features are robust, an ensemble of 200 reconstructions is produced, by considering: "
1) 50 different subsets of years for calibration/validation Calibration: in a random draw of 117 years from the common period to all proxies and the NAO: 1823-1969
Validation: in the remaining 30 years
8&
The"reconstruc&on"technique"
It is based on a Principal Component Regression (PCR), that in essence represents a multivariate regression using the Principal Components of the standardised proxies as predictors To identify which features are robust, an ensemble of 200 reconstructions is produced, by considering: "
1) 50 different subsets of years for calibration/validation Calibration: in a random draw of 117 years from the common period to all proxies and the NAO: 1823-1969
Validation: in the remaining 30 years 2) Four subsets of proxies depending on their type
All proxy types, ice cores excluded, tree-ring excluded, and tree rings plus ice cores
8&
Betaweight ensemble sum for each proxy
124 80-4-8-12-16
Calibration Constrained Reanalysis Constrained Model Constrained
Refining"the"proxy"selec&on"
Calibration Constrained
41 proxies
9&
Betaweight ensemble sum for each proxy
124 80-4-8-12-16
Calibration Constrained Reanalysis Constrained Model Constrained
Betaweight ensemble sum for each proxy
124 80-4-8-12-16
Calibration Constrained Reanalysis Constrained Model Constrained
Refining"the"proxy"selec&on"
9&
Betaweight ensemble sum for each proxy
124 80-4-8-12-16
Calibration Constrained Reanalysis Constrained Model ConstrainedCalibration Constrained
Reanalysis Constrained
41 proxies 10 proxies
Betaweight ensemble sum for each proxy
124 80-4-8-12-16
Calibration Constrained Reanalysis Constrained Model Constrained
Refining"the"proxy"selec&on"
Calibration Constrained (NAOcc)
Reanalysis Constrained (NAOrc)
Model Constrained(NAOmc)
41 proxies 10 proxies 7 proxies
9&
Cor
rela
tion
Coe
ffici
ents
0
0.2
0.4
0.6
0.8
●*
**
●●
NAOccNAOrcNAOmc
median
quartile 3
quartile 1
max
min
Ensemble
Ensemble Mean
First significant correlation
Refining"the"selec&on"of"ensemble"members"
Correlation with NAOVinther
1823-1969
10&
Refining"the"selec&on"of"ensemble"members"C
orre
latio
n C
oeffi
cien
ts
0
0.2
0.4
0.6
0.8
●*
**
●●
NAOccNAOrcNAOmc
median
quartile 3
quartile 1
max
min
Ensemble
Ensemble Mean
First significant correlation
Correlation with NAOVinther
1823-1969 Calibration Validation
We only keep the realizations whose correlation is in the upper quartile The calibration, reanalysis and model constrained NAO gives better validation scores
Refining"the"selec&on"of"ensemble"members"C
orre
latio
n C
oeffi
cien
ts
0
0.2
0.4
0.6
-0.2
Jones et al (1999)
[1781-1822]
Luterbacher et al (2002)
[1659-1822]
Kuettel et al (2010)
[1751-1822] Glueck & Stockton (2001)
[1429-1822]
Luterbacher et al (2001)
[1500-1822]
Cook et al (2002)
[1400-1822]
Trouet et al (2009)
[1049-1822]
●
●
●
●
●
●
●
●
●●
●●●
●
●●
●
●●
●●
Correlation with other NAO reconstructions
Jones et al (1999)
[1781-1822]
We only keep the realizations whose correlation is in the upper quartile The calibration, reanalysis and model constrained NAO gives better validation scores
Luterbacheret al (2002) [1659-1822]
Kuettel et al (2010)
[1751-1822] Glueck &
Stockton (2001) [1429-1822]
Luterbacher et al (2001) [1500-1822]
Cook et al (2001)
[1400-1822]
Trouet et al (2009)
[1049-1822]
The"final"ensemble"NAO"reconstruc&on"
1200 1400 1600 1800 1850 1900 1950
NA
O p
hase
-2
0
2
4
-4
Cor
rela
tion
��
0
���
Time (years)1200 1400 1600 1800 1850 1900 1950
NAOTrouet
NAOVinther
Calibration, reanalysis and model constrained NAO ensemble
The reconstruction does not support a persistent positive NAO in medieval times, although its suggest predominantly positive phases during the 13th century
Our mean ensemble reconstruction clearly outperforms the bi-proxy timeseries from Trouet et al (2009)
Impact"of"the"external"forcings"• No clear response to solar irradiance is identified
14&
Impact"of"the"external"forcings"• No clear response to solar irradiance is identified • Positive NAO phases appear 2 years after major volcanic
eruptions
Lag -1 In phase Lag 1 Lag 2 Lag 3 Lag 4 Lag 5
Gao et al (2008) Sigl et al (2014)Crowley and Unterman (2013)
95%
99%
Common-0.5
0
0.5
1.0
-1.0
-2.0
-3.0
0
1.0
2.0
3.0
Com
posi
te N
AO
NA
O In
dex
NAO response to 11 volcanoes with large radiative forcing
Common volcanoes
Useful to better constrain the magnitude and timing of this response
Impact"of"the"external"forcings"• No clear response to solar irradiance is identified • Positive NAO phases appear 2 years after major volcanic
eruptions
Lag -1 In phase Lag 1 Lag 2 Lag 3 Lag 4 Lag 5
Gao et al (2008) Sigl et al (2014)Crowley and Unterman (2013)
95%
99%
Common-0.5
0
0.5
1.0
-1.0
-2.0
-3.0
0
1.0
2.0
3.0
Com
posi
te N
AO
NA
O In
dex
NAO response to 11 volcanoes with large radiative forcing
Common volcanoes
Useful to better constrain the magnitude and timing of this response
Summary"and"conclusions"
• We have produced a new multiproxy yearly-resolved NAO-reconstruction for the last millennium, compatible with models and previous NAO reconstructions. The model-constrained proxy selection systematically improves the agreement with the reconstructions.
• Our reconstruction updates the previous results from Trouet et al (2009), and suggests that predominant, but not persistent, positive NAO conditions ocurred only during the second half of the Medieval Era.
• Positive NAO phases occur systematically two years after the 11 largest volcanic eruptions, offering a potential for multi-annual prediction
Thanks for your attention!!!!
And"we"validate"it"with"a"perfect"model"approach"
Validation scores over the whole millennium remain significant and close to the calibration range
We"produce"our"final"reconstruc&on"
1200 1400 1600 1800 1850 1900 1950
NA
O p
hase
-2
0
2
4
-4
Cor
rela
tion
��
0
���
Time (years)1200 1400 1600 1800 1850 1900 1950
NAOTrouet
NAOVinther
Calibration, reanalysis and model constrained NAO ensemble
50 year moving correlations with NAOTrouet
The reconstruction does not support a persistent positive NAO in medieval times, although its suggest predominantly positive phases during the 13th century
Common Period [1823-2008]
NAO
Proxy
A proxy may correlate significantly in the common period…
Calibration Period[1823-1969]
Common Period [1823-2008]
NAO
Proxy
Common Period [1823-2008]
NAO
Proxy
A proxy may correlate significantly in the common period…