angelil imsc
Comparing frequencies of regional precipitation and temperature extremes in climate model and reanalysis products Oliver Ang´elil1
Sarah Perkins-Kirkpatrick1 Lisa Alexander1 D´aith´ı Stone2 Michael Wehner2 1 CCRC, 2 LBNL,
UNSW, Sydney, Australia Berkeley, United States
9 June 2016
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Probabilistic Event Attribution Probability Ratio (PR) is our attribution statement Real World PR= PPNatural World
Likelihood
Threshold
If PR > 1, then “+” attribution statement If PR < 1, then “−” attribution statement
PReal World PNatural World
Climate Variable Oliver Ang´ elil
Extremes Comparison for Event Attribution
What if simulated tails are too short?
Likelihood
Threshold
Bias favouring overly strong attribution statements.
PReal World PReanalysis / Obs
Climate Variable
Oliver Ang´ elil
Extremes Comparison for Event Attribution
What if simulated tails are too long?
Likelihood
Threshold
Bias favouring overly weak attribution statements.
PReanalysis / Obs PReal World
Climate Variable
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Research Questions How do the probabilities of extremes in AGCMs and reanalyses compare to one another? Is there much disagreement amongst reanalyses products, i.e. is it okay to use one product? Are our attribution statements biased in favour of being overly strong or weak, and where do these biases occur?
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Data AGCMs Name: Period: Simulations: Forcings:
CAM5.1
MIROC5 HadGEM3-A January 1979 - December 2013 10 Observed values for GHG conc, tropospheric aerosols, volcanic aerosols, solar luminosity, ozone, SST, sea-ice conc.
Three models part of the C20C+ archive.
Reanalyses Name: Period:
ERAInterim
CFSR
MERRA
JRA-55
January 1979 - December 2013
NCEP1 & NCEP2 included in SI.
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Method Probability of exceeding 1-in-1-year and 1-in-10-year daily hot, cold, and wet anomalies in ERA-Interim, in distributions (corrected for mean bias), occurring over the 58 WRAF regions
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Return Periods (Hot – Region 24)
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Return Periods (Hot – Region 24)
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Results: Return Periods for 1-in-1-Year Hot Events
Over region 24, simulated tails are shorter than tails in ERA-Interim Bias favouring overly strong attribution statements Results vary significantly between regions/continents
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Results: Cold and Wet Events 1-in-1-Year Cold Events
1-in-1-Year Wet Events
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Summary Figures Proximity of reanalyses to ensemble spread (in each AGCM) can inform us about reanalyses credibility and biases in event attribution statements.
‘1111’ = ‘good’; ‘0000’ = ‘overestimated’; ’2222’ = ‘underestimated’ ‘0001’, ‘0011’, or ‘0111’ = ‘overestimated/good’ ‘1112’, ‘1122’, or ‘1222’, = ‘underestimated/good’ all remaining combinations = ‘reanalyses inadequate’ Oliver Ang´ elil
Extremes Comparison for Event Attribution
Summary Figures 1-in-1-Year Events
1-in-10-Year Events
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Conclusions Shapes of tails between reanalyses products vary considerably, so using one, or even two products for AGCM evaluation is poor science. Return periods in the AGCMs mostly agree with each other in the sense that they collectively fall on the same side of reanalyses. As the anomaly of the extreme increases, we see an increase in bias favouring overly strong attribution statements. If AGCM results fall on one side of reanalyses, it is necessary to correct for tail bias. If AGCM results fall within a large spread of reanalyses, how do we go about doing event attribution?
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Thank You! [email protected]
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Sarah Perkins-Kirkpatrick1 Lisa Alexander1 D´aith´ı Stone2 Michael Wehner2 1 CCRC, 2 LBNL,
UNSW, Sydney, Australia Berkeley, United States
9 June 2016
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Probabilistic Event Attribution Probability Ratio (PR) is our attribution statement Real World PR= PPNatural World
Likelihood
Threshold
If PR > 1, then “+” attribution statement If PR < 1, then “−” attribution statement
PReal World PNatural World
Climate Variable Oliver Ang´ elil
Extremes Comparison for Event Attribution
What if simulated tails are too short?
Likelihood
Threshold
Bias favouring overly strong attribution statements.
PReal World PReanalysis / Obs
Climate Variable
Oliver Ang´ elil
Extremes Comparison for Event Attribution
What if simulated tails are too long?
Likelihood
Threshold
Bias favouring overly weak attribution statements.
PReanalysis / Obs PReal World
Climate Variable
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Research Questions How do the probabilities of extremes in AGCMs and reanalyses compare to one another? Is there much disagreement amongst reanalyses products, i.e. is it okay to use one product? Are our attribution statements biased in favour of being overly strong or weak, and where do these biases occur?
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Data AGCMs Name: Period: Simulations: Forcings:
CAM5.1
MIROC5 HadGEM3-A January 1979 - December 2013 10 Observed values for GHG conc, tropospheric aerosols, volcanic aerosols, solar luminosity, ozone, SST, sea-ice conc.
Three models part of the C20C+ archive.
Reanalyses Name: Period:
ERAInterim
CFSR
MERRA
JRA-55
January 1979 - December 2013
NCEP1 & NCEP2 included in SI.
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Method Probability of exceeding 1-in-1-year and 1-in-10-year daily hot, cold, and wet anomalies in ERA-Interim, in distributions (corrected for mean bias), occurring over the 58 WRAF regions
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Return Periods (Hot – Region 24)
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Return Periods (Hot – Region 24)
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Results: Return Periods for 1-in-1-Year Hot Events
Over region 24, simulated tails are shorter than tails in ERA-Interim Bias favouring overly strong attribution statements Results vary significantly between regions/continents
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Results: Cold and Wet Events 1-in-1-Year Cold Events
1-in-1-Year Wet Events
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Summary Figures Proximity of reanalyses to ensemble spread (in each AGCM) can inform us about reanalyses credibility and biases in event attribution statements.
‘1111’ = ‘good’; ‘0000’ = ‘overestimated’; ’2222’ = ‘underestimated’ ‘0001’, ‘0011’, or ‘0111’ = ‘overestimated/good’ ‘1112’, ‘1122’, or ‘1222’, = ‘underestimated/good’ all remaining combinations = ‘reanalyses inadequate’ Oliver Ang´ elil
Extremes Comparison for Event Attribution
Summary Figures 1-in-1-Year Events
1-in-10-Year Events
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Conclusions Shapes of tails between reanalyses products vary considerably, so using one, or even two products for AGCM evaluation is poor science. Return periods in the AGCMs mostly agree with each other in the sense that they collectively fall on the same side of reanalyses. As the anomaly of the extreme increases, we see an increase in bias favouring overly strong attribution statements. If AGCM results fall on one side of reanalyses, it is necessary to correct for tail bias. If AGCM results fall within a large spread of reanalyses, how do we go about doing event attribution?
Oliver Ang´ elil
Extremes Comparison for Event Attribution
Thank You! [email protected]
Oliver Ang´ elil
Extremes Comparison for Event Attribution
