counterfactual thinking in climate...
TRANSCRIPT
COUNTERFACTUALTHINKINGIN CLIMATE SCIENCE
Dr. Gordon WooCatastrophist
Mathematics of Planet Earth Jamboree
21st March 2018
‘Science predicts that many different
kinds of universe will be spontaneously
created out of nothing.
It is a matter of chance which we are in’.
Stephen Hawking
Disaster: an unfavourable aspect of a star
The past cannot
be changed.
Hazard analysts
treat the past
as fixed and
determined,
like eclipses.
This is an
anthropocentric
bias.
Downward counterfactuals
If Things
Had
Gone
Wrong
If Things
Had
Turned
for the
Worse
Fritz
Leiber
Fritz
Leiber
‘Intelligence
without imagination:
a deadly combination’
Downward counterfactual
thought experiment
Consider a historical weather system state S[0] that led to
a major economic loss of L.
Construct an alternative weather system state S[k] that would
have led to a major economic loss of L + k* Δ*L,
where Δ = 5% , and k = 1.
Repeat for ever increasing integer values of k = 0, 1, 2,….n
Estimate the relative likelihoods of the alternative
weather system states S[k] for k = 0,1,2…n
State transition path to a Black Swan
S[0] S[1] S[2] S[n]
The Counterfactual Class of Black Swans:
The set of extreme events that can be explicitly constructed by
progressive downward counterfactual simulation of a historical event.
If there are M historical events, how complete is ? 1
M
j
j
B
Early on September 13, just 50 miles offshore, Ike shifted course.
The wall of water the storm was projected to push into the
Houston area was far smaller than predicted.
ALL NEIGHBORHOODS, AND POSSIBLY ENTIRE COASTAL COMMUNITIES,
WILL BE INUNDATED, PERSONS NOT HEEDING EVACUATION ORDERS
IN SINGLE FAMILY 1 OR 2 STORY HOMES WILL FACE CERTAIN DEATH.
Hurricane Ike: September 2008
$38 billion property damage
A 24ft storm surge along the
Houston Ship Channel would
have released about 90
million gallons of crude oil
and chemicals into Houston
neighbourhoods and
Galveston Bay.
It would have been the worst
environmental disaster in US
history, and caused massive
economic disruption.
‘Perfect storm’ tipping point in consequential loss
Cascading Economic Loss
Black Swan as an
economic black hole
Property Damage
Regional
Environmental
Pollution
Global
Economic
Disruption
Infrastructure
Dysfunction
Supply Chain Failure
Attribution of extreme weather events in the context of climate changeNational Academies Press (2016)
The likelihood
of specified events
resulting from a
given temperature
distribution
and its changes
What emergent
Black Swan
events might
lie in these tails?
How can emergent
Black Swan
events be
better imagined?
Central European flooding
S[0] S[1] S[2] S[n]
Central Europe flooding of August 2002
The severity of the flood was due to the
unusual confluence and stalling of two
continental weather systems.
Downward counterfactual:
Slowing of the Jet Stream
might enable weather patterns
to persist over several weeks.
January 2005 and December 2015 Carlisle floods
The Carlisle flood alleviation
scheme was completed in 2010.
What was the chance of rainfall
causing the river defences
to be overtopped in December 2015?
The January 2005 flood had an
estimated annual exceedance
probability of 1:170.
S[2005] S[2015]
Probabilistic definitions of causation
PS is the probability of sufficient causation.
It is defined to be the probability that Y would have occurred
in the presence of X, given that Y and X did not occur.
PNS is the probability of necessary and sufficient causation.
It is defined as the probability that Y would have occurred in the
presence of X, and that Y would not have occurred in the absence of X.
PN is the probability of necessary causation.
It is defined as the probability that the event Y
would not have occurred in the absence of the event X,
given that both events Y and X did in fact occur.
Hannart et al., 2016 DOI:10.1175/BAMS-D-14-00034.1
2003 European heat wave attribution
For the 2003 European heat wave,
PN = 0.9 whereas PS =0.0072
(Hannart, Pearl et al. 2016)
70,000 died in the European
heat wave.
Between 2030 and 2050, climate change is
projected to cause about 250,000 additional
deaths per year from heat stress, malnutrition
and the spread of infectious diseases.
Climate change as a necessary cause
of catastrophic societal losses
External
Weather
Hazard
Critical
Infrastructure
Failure
Catastrophic
Societal
Losses
Climate
ChangeHuman
Error
What is the most lethal
infrastructure catastrophe
of which climate change
would be a necessary
cause?
Taxonomy of emergent disasters
Complex system phenomena
e.g. trigger events, tipping points
Multi-hazard combinations e.g. natural and man-made hazards, cyber, pandemics etc.
Cascading disasters: e.g. drought; wildfire; mudslide
Historical event Compound disaster
Extreme windstorm impact on
nuclear plant in Massachusetts
On January 27, 2015, Winter Storm
Nor’easter Juno knocked out both of
the 345,000 volt transmission lines
connecting the ageing Pilgrim nuclear
plant in Plymouth, Mass..
The reactor shut down when the
second offsite power line was lost.
Equipment problems and operator
errors complicated the intended
response.
Nor’easter
12 March 2018
High sensitivity to
the coastal
Nor’easter track
shortens the
time horizon for
forecasting
severe weather,
(e.g. snow depth),
and consequent
economic impact.
On 7 February 2017,
water from powerful
winter storms rushed
under the spillway,
which forced up giant
slabs and ripped a
huge hole in
the structure.
The spillway was built
in the late 1960s.
The designer was
just two years out of
college with no prior
professional
experience designing
spillways.
Oroville dam,
Northern California
Greater fragility of critical infrastructure
Critical infrastructure is designed for external hazard levels
prescribed by civil engineers.
This infrastructure is vulnerable to external loading
exceeding these design levels.
These may give rise to tipping point effects.
Extreme November 2009
rainfall in Jeddah, Saudi Arabia
Standard Deviation
Average for November
mm
3.3 Standard Deviations
Just how harmful can
2.8 inches of rain be?
Flash flooding deaths in the desert kingdom
The level of rainfall over 6 hours
on 25 November 2009 in Jeddah
and surrounding areas reached
about 72mm.
Hundreds were swept away in one of
the most deadly floods that Saudi
Arabia has ever experienced. Many
were drowned in their cars.
The increased risk of flash flooding requires
reconsideration of the standards used in the
construction of roads, bridges, tunnels,
and drainage networks.
Saudi cities lack the infrastructure to deal
with heavy rain, such as having effective
rain sewers and drainage channels.
Climate change in Saudi Arabia
Stochastic modelling of the past
‘There is an infinitude of pasts,
all equally valid’André Maurois
Dynamic
Stochastic Process
Historical
Catalogue
Alternative
Catalogue
Alternative
Catalogue
The historical event loss catalogue
is just ONE realization of what
could have happened.
....Alternative catalogues might have been generated.
Stochastic
event sets
could be
benchmarked
against
these
alternative
catalogues.
Focusing more on the process than the events
Historical luck: zero event loss
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
zero $1bn $5bn $10bn $15bn $20bn
Loss Probability
Column2
Column1
Probability
Loss
Actual
loss
2016
The British navy
was fortunate:
the expected
result would have
been the
destruction of at
least one ship,
and possibly as
many as three.
The belief of military historians that this
was a squandered British victory
on January 24, 1915 - IS FALSE.
Results from Approximate Bayesian Computation
Gauging the volatility of historical losses
0
200
400
600
800
1000
Event A Event B Event C Event D
Extreme alternative
Plausible alternative
Actual historical
Unprecedented Simulated Extremes
using Ensembles
Thompson V. et al. , Nature Communications, July 2017: High risk of unprecedented UK rainfall
In the current climate. [Met Office Unified Model simulating 100x more winters than in 1981-2015.]
Southeast England January rainfall
D(t)
{ X(1), X(2),….X(n)}
Perturbation of a state into a more dangerous domain
System State
Major Disaster Domain
NEAR-MISS ZONE
East Bay Hills wildfire
October 1991
34
The Oakland Hills fire, killed 25
people, injured 150, and destroyed
more than 3,800 homes.
“It’s hard to get organized and run for
your life at the same time!”
The fire was contained only when
the Santa Ana wind changed.
Counterfactual wildfire loss distribution
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
$0.5bn $1.5bn $3bn $5bn $10bn $15bn
Loss Probability
Column2
Column1
Probability
Loss
Actual
loss
Climate change cliff-edge
"Climate change
is one of the
great dangers
we face, and it's
one we can
prevent.”