18
BAMOS
Dec 2019
Workshop report
Extreme Event Attribution in Australia
Co‑authored by the workshop attendees as listed.
Contact: Todd Lane, [email protected]
On 29–30 October 2019 a group of 15 scientists from Australian
Universities, the Australian Bureau of Meteorology, CSIRO and
NCAR came together in Melbourne for a two‑day workshop
on the science of event attribution. For the purposes of this
workshop, event attribution was defined as "the process of
determining the role of anthropogenic climate change in
modifying the likelihood, intensity, duration, or frequency of
occurrence of a particular extreme event".
Event attribution has emerged as a growing field in recent
years, with many journal articles, contributions to the media
and The Conversation, and an annual supplement in the Bulletin
of the American Meteorological Society (e.g. Herring et al., 2019).
This growth in activity has been driven in part by the scientific
challenge to understand the influence of anthropogenic climate
change on extreme events, and also motivated by the public
and stakeholder discourse and requests for information around
the role of climate change in particular high‑impact events. In
recent years there has been the development of a range of new
techniques and modelling tools to tackle this problem.
The goals of the workshop included answering the following
questions:
Question 1. What are the details of current event attribution
methods? What are their strengths and weaknesses and how do
these vary with phenomenon in the Australian region?
Question 2. What length of record is required to characterise
an extreme event? How well do our climate models need to
perform, in terms of statistics and key processes, to be useful for
event attribution studies?
In addition to these questions, there were detailed discussions
to help enhance collaboration across the weather/climate sector
and to identify future activities and opportunities amongst the
workshop participants, including those activities within the
ARC Centre of Excellence for Climate Extremes (CLEX), National
Environmental Science Program Earth Systems and Climate
Change Hub (NESP‑ESCC Hub) and the Bureau of Meteorology.
Many of the detailed answers to questions 1 and 2 above are
dependent on the particular extreme being considered and will
be described in more detail in a future publication.
Most methods for event attribution rely on a combination
of model experiments, observational analysis, and statistical
methods to make an ‘attribution statement’. Methods like
Fractional Attributable Risk (FAR) analysis (e.g. Allen, 2003) have
become common.
Workshop attendees:
Todd Lane 1,3 , Pandora Hope 2,4 , Andrew King 1,3 , Sarah
Perkins‑Kirkpatrick 1,5 , Lisa Alexander 1,5 , Julie Arblaster 1,6 ,
Nathan Bindoff 1,7 , Craig Bishop 1,3 , Mitchell Black 4 , Michael
Grose 2,8 , Neil Holbrook 1,7 , Greg Holland 9 , David Karoly 2,8 ,
Andrew Pitman 1,5 , and Michael Reeder 1,6 .
1.
2.
3.
4.
5.
6.
7.
8.
9.
ARC Centre of Excellence for Climate Extremes
National Environmental Science Program, Earth
Systems and Climate Change Hub
School of Earth Sciences, The University of Melbourne,
Parkville, Victoria
Bureau of Meteorology
Climate Change Research Centre, University of New
South Wales, Sydney, NSW
School of Earth, Atmosphere and Environment,
Monash University, Clayton, Victoria
Institute for Marine and Antarctic Studies, University of
Tasmania, Hobart, Tasmania
CSIRO
National Center for Atmospheric Research, Boulder,
Colorado, USA
One important element that emerged in the discussions
of Question 1 was that it has now become best practice for
studies to use multiple methods for event attribution, with
single‑method studies no longer considered state‑of‑the‑art.
Moreover, attempts are being made to incorporate additional
estimates of uncertainty in event attribution statements,
either through multiple methods or using a range of ensemble
techniques.
The group also had detailed discussions about question 2,
especially the aspects of how good a modelling system should
be to be useful for event attribution studies. Results from
cutting‑edge work on these processes were also described
at the workshop. There was agreement that for a model to
be useful for event attribution it should produce a good
representation of the statistics of the phenomenon in question
and must capture the key physical/dynamical processes that
lead to that phenomenon. When the modelled processes or
statistics are deficient, then appropriate and detailed caveats
and uncertainty bounds must be included in attribution
statements, papers or reports. For some events, attempting
attribution might be beyond our current skill. There was also
unanimous agreement that if a model reproduced the statistics
of the event but for the wrong physical/dynamical reason then
it should not be considered fit for purpose.