Constraining estimates of climate

chirpskulkInternet και Εφαρμογές Web

3 Νοε 2013 (πριν από 3 χρόνια και 11 μήνες)

73 εμφανίσεις

1
/11

Constraining estimates of climate
sensitivity with present
-
day

observations

Daniel Klocke*
,1
, Robert Pincus
2
, Johannes Quaas
1











1
Max Planck Institute for Meteorology

*International Max Planck Research School on Earth System Modelling

2
University of Colorado/NOAA Earth System Research Lab



EUCLIPSE kick
-
off 28.09.2010

2
/11

Simplifying the problem


Two ensemble:


1.
capturing parametric uncertainty

-
> perturbed parameter ensemble

-
> focus on clouds

2.

encompassing structural uncertainty

-
> CMIP3 ensemble



Now we use the simple ensemble as an analogy to the complex

ensemble as the complex ensemble is used as an analogy to

nature.

3
/11

Perturbed parameter ensemble


ECHAM5 (T31L19)



500 one year simulation


Each with perturbed parameters


Parameters are varied simultaneously


Prescribed SST + sea ice cover



-
> Evaluate the models

Pincus et al., 2008

Targeted metric is justified, because

1.
Mainly cloud parameters are varied

2.
Change in the cloud radiative effect is
driving the spread in climate sensitivity.

3.
Broad error matrices have to deal with
compensating errors.

4
/11

Similarities between the ensembles


5
/11

Skill vs Climate Sensitivity

6
/11

Parametric sensitivity

7
/11

Targeted metric (tropical oceans)

8
/11

Stddev


Mean


0.56


3.41

0.38


3.80

PDF of climate sensitivity

9
/11

Breaks for a more complex ensemble

10
/11

Conclusions and implications

We reproduced the diversity in some important measures of the
CMIP3 ensemble.



with a surprisingly simple parametric sensitivity

-
> interpreting the results from the CMIP3 ensemble as the full
range of uncertainty is unfounded.



Weighting by simple skill measures or excluding models on
thresholds does not put higher constraints on climate sensitivity.



For a simpler ensemble we were able to identify a well
-
targeted
metric which related climate sensitivity to an observable in the
present
-
day climate.



Weighting by this measure narrowed the PDF of climate
sensitivity.



The fact that this did not generalize to a more complex ensemble,
we find no basis, that one could generalize from the complex
ensemble to the real world, even if one finds a relation.



11
/11

Targeted metric is justified, because

1.
Mainly cloud parameters are varied

2.
Change in the cloud radiative effect is driving the spread in climate sensitivity.

3.
Broad error matrices have to deal with compensating errors.

Skill metric

Present
-
day distribution of:


Short wave cloud radiative effect


Long wave cloud radiative effect


Cloud cover


Precipitation


Root
-
mean
-
square error of monthly
means over the seasonal cycle


Errors are standardized and aggregated

12
/11

Motivation

Reichler&Kim, 2008

1.5


4.5


1.5


4.5


2.0


4.5 (3.0)

Climate

Sensitivity

Present
-
day

performance

The approach: making models better to narrow the answers did not work so far.


Alternative approach: Weighting models by the likelihood that it is correct.
(Murphy et al. 2004)

13
/11

RMS for cloud radiative effects (sw, lw), precipitation and cloud cover
according to two observational data sets (climatologies) for each ‘model’

Skill measure

14
/11

Model ranking

15
/11

Climate Sensitivity

16
/11

What is coming

Ensemble Kalman filter data assimilation (DART) to initialize
the model with something close to reality and use the
assimilation increments as skill measures.


Does the error in the mean climate state manifest itself already
in the first few time steps?


Can we use data assimilation to estimate (some) parameters?


Or at least narrow the range for the parameters, get some
objective justification for parameter choices.