Post on 05-Nov-2020
transcript
Oxford University
Oxford-Hadley pilot study for the design of an event-attribution system
Myles AllenDepartment of Physics, University of Oxford
myles.allen@physics.ox.ac.ukThanks to:
Pardeep Pall, Dáithí Stone, Peter Stott & many others
Oxford University
Motivation
South Oxford on January 5th, 2003
Phot
o: D
ave
Mitc
hell
Oxford University
The problem in October 2000 and January 2003: a consistently displaced Atlantic jet-stream
500hPa wind speed: Autumn climatology (colours) & Autumn 2000 (contours)
Blackburn & Hoskins, 2003
Oxford University
Meteorology provides the context, not the cause
Displaced jet-stream associated with anomalous geopotential-height wave train extending from the sub-tropical Atlantic through to Siberia
Low pressure near the UK and a strong Scandinavian ridge. “Scandinavia pattern”
Wet UK Autumns during the preceding 42 years also associated with this pattern.
Autumn 2000 geopotential-height anomaly (relative to ECMWF climatology) at 300hPa.
Blackburn & Hoskins (2001)
Oxford University
Fraction Attributable Risk
If, all other things being equal, human influence doubles the risk of a flood, and that flood occurs, there you can argue that human influence is “to blame” for half the risk.
Fraction Attributable Risk is defined asFAR = 1 – P0/P1
P0 = risk with human influence “removed”P1 = current risk, including human influence
Epidemiologists use “Relative Risk”:RR = P1/P0
P1 = risk in “diseased” group (less certain)P0 = risk in general population (more certain)
Oxford University
Schematic: Autumn 2000 (A2000) v. non-industrial Autumn 2000 (NIA2000) rainfall distributions
P1
P0
EVENT
Oxford University
Why we need a modelling approach
Estimating and removing the impact of human influence is simplest for variables whose variability is approximately unchanged.
This applies to very large scale temperatures: hence the Stott et al approach.
It doesn’t apply to small-scale temperatures because of soil-moisture feedbacks.
It cannot apply to precipitation: there are clear physical reasons why we expect the most obvious impact of climate change to be on precipitation extremes rather than mean precipitation.
Oxford University
The climateprediction.net seasonal attribution experiment (Pall, 2007)
Aim: to quantify the role of increased greenhouse gases in precipitation responsible for 2000 floods.
Challenge: relatively unlikely event even given 2000 climate drivers and sea surface temperatures (SSTs).
Approach: large (multi-thousand-member) ensemble simulation of April 2000 – March 2001 using forecast-resolution global model (90km resolution near UK).
Identical “non-industrial” ensemble removing the influence of increased greenhouse gases, including attributable SST change.
Use several coupled models for SST pattern to allow for uncertainty.
Oxford University
The climate that might have been, on your desktop PC: http://attribution.cpdn.org
Oxford University
Autumn 2000 in the ERA-40 reanalysis…
…and in one of the wetter members of our ensemble.
Oxford University
Observed v. simulated geopotential height pattern associated with Autumn rainfall
ECMWF re-analyses(300hPa; 40 seasons)
Bla
ckbu
rn &
Hos
kins
(2
001)
A2000 ensemble(500hPa; 1700 simulations)
Oxford University
Constructing the climate of Autumn 2000 without the influence of anthropogenic GHGs
Reduce GHG levels & subtract 4 different patterns of GHG attributable SST warming.
Amplitudes estimated by optimal detection.
Models used: HadCM3, GFDLR30, NCARPCM1, MIROC3.2
4 patterns, 10 possible amplitudes for each = 40 possible NIA2000 climates
Oxford University
Distributions of 90-day total rainfall: A2000 v. NIA2000 v. “observations”
Oxford University
Return-times versus total Sept-Nov precipitation in the “Autumn 2000” ensemble
Oxford University
Return-times removing greenhouse warming signal estimated using four coupled models
Oxford University
Influence of simulated meridional SST gradients
Oxford University
NCAR-PCM warming in the North Atlantic:evolution of S-N SST gradients in all-forcings runs
Oxford University
Combining results, excluding the PCM model
Oxford University
Combining results, excluding the PCM model
Oxford University
On shorter timescales, result is no longer model-dependent: so what is the “correct” diagnostic?
Return times for 5-day running mean rainfall
Oxford University
Convert rainfall into synthetic run-off using ARFIMA model to extract relevant timescales
Oxford University
Contribution of past greenhouse gas emissions to risk of extreme run-off
Oxford University
Fraction Attributable Risk
Oxford University
Conclusions from Pall et al
Results for 90-day rainfall totals (original focus) ambiguous: 3 GHG-SST patterns show increase in risk, 1 shows decrease.
Results for 5-day rainfall totals show consistent increase in risk.
Results using rainfall-runoff model to extract time-scale relevant to flooding are clearer: significant increase in risk independent of GHG-SST pattern removed.
Best estimate that influence of past GHG emissions contributed approximately half the risk of the rainfall responsible for the Autumn 2000 floods.
Oxford University
Increasing “drought” risk in Central France
Oxford University
A template for operational attribution
Simulate the year 2000 (or 2008) thousands of times with a forecast-resolution weather model.
Repeat after modifying ocean temperatures and atmospheric composition to remove a range of estimates of the influence of past greenhouse gases.
Extend to examine influence of anthropogenic aerosols, ENSO, volcanoes, solar forcing etc.
Repeat using other models, other years, mixed layer ocean (?), perturbed parameters etc.
Compare frequency of occurrence of interesting weather events between the ensembles.
Feed results into impact models.
Oxford University
Alternate approaches (contribution from IDAG)
Daithi Stone: analysing FAR for regional temperatures using CMIP-3 ensemble of opportunity– Accounts for model response uncertainty.– Coarse models appropriate for sub-continental scales only.
Francis Zwiers & Xuebin Zhang: analysing FAR using time-evolving extreme value distributions.– Less model-dependent than Pall et al approach.
Simon Tett (U. Edinburgh): Proposing climate of the 20th+ century studies with high-resolution models.
Oxford University
Global approach using CMIP-3 models (Daithi Stone, CSAG)
Oxford University
FAR for 2009 regional temperatures exceeding 10th percentile