12.08.2013 Views

final_program_abstracts[1]

final_program_abstracts[1]

final_program_abstracts[1]

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

11 IMSC Session Program<br />

Quantifying uncertainty in climate forecasts from<br />

climateprediction.net<br />

Thursday - Parallel Session 2<br />

Dan Rowlands 1 , Dave Frame 2 , Tolu Aina 3 , Nicolai Meinshausen 4 and Myles<br />

Allen 1<br />

1 AOPP, Oxford University, UK<br />

2 Smith School, Oxford University, UK<br />

3 OeRC, Oxford University, UK<br />

4 Department of Statistics, Oxford University, UK<br />

Perturbed physics ensembles have been run to explore the uncertainty in climate<br />

forecasts by systematically varying physics parameters in a single base model. Here<br />

we discuss some results from the transient HadCM3L ensemble run as part of the<br />

climateprediction.net project.<br />

In particular we detail the approach taken in evaluating the level of model-data<br />

discrepancy in each ensemble member, based on methods used in Detection &<br />

Attribution studies. In particular we advocate comparing model output and<br />

observations only at a level where the residual is indistinguishable from our best<br />

estimate of natural variability – this avoids ignoring the irreducible error or the need<br />

for a subjective “discrepancy” term.<br />

Further, we consider the question of what observations are relevant for making a<br />

particular climate forecast: should the set that we use depend on the question that we<br />

are asking, or does it make sense to have a broad set over which we evaluate the<br />

model performance? Can the inclusion of transient and climatological information<br />

together help us rule out particular future climates?<br />

We <strong>final</strong>ly discuss how one may incorporate all of this information together into a<br />

single probabilistic forecast, and advocate objective approaches based on testable<br />

information rather then subjective, often untestable, prior assumptions on climate<br />

model parameters.<br />

Abstracts 245

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!