Detection and Attribution of Climate Change

1
Detection and Attribution of Climate Change
Ben Santer Program for Climate Model Diagnosis
and Intercomparison Lawrence Livermore National
Laboratory, Livermore, CA 94550 Email
santer1_at_llnl.gov American Statistical
Association A Statistical Consensus on Global
Warming Boulder, Colorado. Oct. 26th, 2007
2
Why is detection and attribution work important?

• It is another form of model evaluation
• Successful simulation of historical changes in
  climate enhances confidence in projections of
  future climate change
• In an environment where there is still political
  debate regarding the reality of a human effect on
  global climate, it is imperative to have sound
  science on the nature and causes of climate
  change

3
Structure of talk

• Introduction
• Charge to discussion leaders
• Where do you believe a consensus has formed?
• Where can consensus be expected in the near
  future?
• Where can statistical science provide further
  assistance to future research?
• Conclusions

4
Introduction Definition of consensus

• Con sen sus Noun.
• A view or stance reached by a group as a whole
  or by majority will.
• General agreement. (American Heritage College
  Dictionary)
• From the Latin consentire, to agree.

5
Introduction Do the IPCC AR4 findings constitute
a consensus of climate science experts?

• 152 Coordinating Lead Authors and Lead Authors
• Authors were from over 30 countries
• Drafts of Working Group I Report were subjected
  to two rounds of review and revision
• Report was reviewed by over 650 individual
  experts, as well as by governments and
  international organizations
• In total, over 30,000 written comments were
  received
• Summary for Policymakers was approved
  (line-by-line) by officials from 113 governments
• Report outline in Nov. 2003. Acceptance of SPM
  and underlying chapters in Feb. 2007

6
Structure of talk

• Introduction
• Charge to discussion leaders
• Where do you believe a consensus has formed?
• Where can consensus be expected in the near
  future?
• Where can statistical science provide further
  assistance to future research?
• Conclusions

7
Where do you believe a consensus has formed?

• Warming of the climate system is unequivocal
• There is compelling scientific evidence of a
  human fingerprint on global climate

8
Unequivocal warming of the climate system

• The oceans and land surface have warmed
• The troposphere has warmed
• Atmospheric water vapor has increased
• Sea level has risen
• Glaciers have retreated over most of the globe
• Snow and sea-ice extent have decreased in the
  Northern Hemisphere
• Individually, these changes are consistent with
  our scientific understanding of how the climate
  system should be responding to human influences

9
Where do you believe a consensus has formed?

• Warming of the climate system is unequivocal
• There is compelling scientific evidence of a
  human fingerprint on global climate

10
The scientific evidence for a human fingerprint
on global climate has strengthened over time
The balance of evidence suggests a discernible
human influence on global climate
Most of the observed increase in globally
averaged temperatures since the mid-20th century
is very likely due to the observed increase in
anthropogenic greenhouse gas concentrations
There is new and stronger evidence that most of
the warming observed over the last 50 years is
attributable to human activities
11
Natural factors alone cannot explain the recent
warming of the Earths surface
Observations
Red Model All forcing results
Blue Model Solarvolcanic results
Black Observed surface temperature changes
12
What is climate fingerprinting?

• Strategy Search for a computer model-predicted
  pattern of climate change (the fingerprint)
  in observed climate records
• Assumption Each factor that influences climate
  has a different characteristic signature in
  climate records
• Method Standard signal processing techniques
• Advantage Fingerprinting allows researchers to
  make rigorous tests of competing hypotheses
  regarding the causes of recent climate change

13
Human-caused fingerprints have been identified in
many different aspects of the climate system
Tropospheric temperatures
Stratospheric temperatures
Surface specific humidity
Water vapor over oceans
Tropopause height
Ocean temperatures
Sea-level pressure
Atmospheric temperature
Zonal-mean rainfall
Near-surface temperature
Continental runoff
14
We have made considerable progress in defining
the fingerprints of different forcings
Pressure (hPa)
Height (km)
1. Solar
2. Volcanoes
Pressure (hPa)
3. Well-mixed greenhouse gases
4. Ozone
Height (km)
5. Sulfate aerosol particles
6. 1st five factors combined
Pressure (hPa)
Height (km)
Santer et al., Climate Change Science and Policy,
2007
C/century
15
Fingerprint detection explained pictorially.
Time-varying observed patterns
Time-varying control run patterns
Model fingerprint
Projection onto model fingerprint
Projection onto model fingerprint
Signal and noise time series
Signal-to-noise ratios
16
Estimating the noise of natural internal
variability

• Model-based estimates of natural internal
  variability are an integral component of DA
  research
• Why do we rely on models for these estimates?
• They can be used to perform the control
  experiments that we cant conduct in the real
  world
• Why is it difficult to estimate natural internal
  variability from observations?
• We want to estimate noise on multi-decadal to
  century timescales
• Most observational records are too short for this
  purpose
• Signal and noise are convolved difficult to
  achieve unambiguous partitioning

17
Optimal fingerprinting A brief example
A) MODELS WATER VAPOR FINGERPRINT
B) MODELS LEADING PATTERN OF CLIMATE NOISE
C) OBSERVATIONS CHANGE IN WATER VAPOR
(1988-2006)
EOF loading (A-B) or total linear change in Wo in
kg/m2 (C)
18
DA in a multi-model framework Use of multiple
models to estimate fingerprints and noise
19
Estimating signal-to-noise ratios and detection
times
20
Estimating signal-to-noise ratios and detection
times
21
For water vapor, there is no evidence that
climate noise is systematically underestimated
in IPCC AR4 models
Average model water vapor variability is slightly
larger than in observations
22
Structure of talk

• Introduction
• Charge to discussion leaders
• Where do you believe a consensus has formed?
• Where can consensus be expected in the near
  future?
• Where can statistical science provide further
  assistance to future research?
• Conclusions

23
Where can consensus be expected in the near
future?

• We will have some form of operational
  attribution capability
• DA studies will routinely use information from
  large, multi-model ensembles (and will make more
  intelligent use of this information)
• Structural uncertainties in observations will
  become an integral part of DA research
• We will have formally identified anthropogenic
  fingerprints
• At sub-continental spatial scales
• In variables more relevant to climate impacts
• In plant and animal distributions and abundances
• Fingerprinting will be feasible with increasingly
  shorter (lt 30-year) observational records

24
Structure of talk

• Introduction
• Charge to discussion leaders
• Where do you believe a consensus has formed?
• Where can consensus be expected in the near
  future?
• Where can statistical science provide further
  assistance to future research?
• Conclusions

25
Where can statistical science provide further
assistance to future DA research?

1. In assessing sensitivity of DA results to model
   quality
2. By contributing state-of-the-art space-time
   modeling approaches to fill in the gaps in
   observational datasets with sparse, space- and
   time-varying coverage
3. By helping to provide a better assessment of the
   trade-offs between ensemble size (for any
   individual model) and the number of models
   contributing to a multi-model average
4. By contributing improved methods for assessing
   whether human influences have modulated the
   statistical behavior of existing modes of natural
   variability
5. By bringing statistical rigor to regression-based
   predictions of hurricane activity
6. Better constraining the Transient Climate
   Response obtained from DA methods

26
Future research I Sensitivity of DA results to
model quality

• A number of recent studies have attempted to
  weight model projections of future climate
  change (generally by model performance in
  simulating present-day climatological means)
• Thus far, no attempt to use any form of weighting
  in multi-model DA work
• All multi-model DA studies to date are one
  model, one vote
• Are results from current multi-model DA studies
  biased by inclusion of information from models
  with noticeable deficiencies in simulation of
  variability?

27
Future research I Sensitivity of DA results to
model quality
28
Future research II Improvement of observational
datasets with sparse, space- and time-varying
coverage

• Concerns have been expressed about the
  reliability of model-based estimates of the
  natural variability of ocean temperatures (e.g.,
  Lyman et al., 2006)
• Casts doubt on reliability of DA results
  obtained with ocean temperatures
• How do we address these concerns?
• Better quantification of uncertainties in
  observed variability estimates (e.g., AchutaRao
  et al., 2007). Involves use of both physical
  models (ocean data assimilation products) and
  statistical models
• Identify and adjust for the effects of
  instrumental biases in different ocean observing
  systems (Church et al., in preparation)
• Revisit ocean DA studies with improved,
  bias-corrected observational data
• Use proxy data to obtain better constraints on
  model estimates of natural internal variability
  (e.g., multi-century SST reconstructions from
  corals)

29
Future research II The ocean observing network
has changed dramatically over time
Source AchutaRao et al., JGR (2006)
30
Future research II Do models systematically
underestimate observed ocean temperature
variability?
Models with volcanoes
Time-variability of sub-sampled ocean
temperature data (C)
Models without volcanoes
Time-variability of complete ocean temperature
data (C)
Source AchutaRao et al., PNAS (2007)
31
Future research II Sampling model data at
locations of ocean observations improves
model-data agreement
Models with volcanoes
Time-variability of sub-sampled ocean
temperature data (C)
Models without volcanoes
Time-variability of complete ocean temperature
data (C)
Source AchutaRao et al., PNAS (2007)
32
Future research II Implications of observational
uncertainty for DA research
33
Conclusions

• We have identified human fingerprints in a
  number of different aspects of the climate
  system
• Temperature (land and ocean surface stratosphere
  and troposphere zonal-mean profiles through the
  atmosphere upper 700 meters of the ocean ocean
  heat content height of thermal tropopause)
• Atmospheric circulation (mean sea-level pressure)
• Moisture-related variables (zonal-mean rainfall
  surface specific humidity total water vapor over
  oceans continental runoff)
• The climate system is telling us a physically-
  and internally-consistent story
• From my own biased personal perspective, the
  collaboration between statisticians and climate
  scientists in the area of DA research has been
  very successful. These interactions have been
  facilitated by
• IDAG (International Detection and Attribution
  Group)
• IMSC (International Meetings on Statistical
  Climatology)

34
A brief history of DA research Some important
milestones
Publication of first paper on the theory of
optimal detection
First work on S/N ratios in climate model data
First application of optimal detection method to
problem of detecting human influences on climate
1979
1980
1981
1982
1983
1984
1985
1986
Publication of IPCC First Assessment Report
Application of pattern correlations and
multi-variable methods to DA problem
1987
1988
1989
1990
1991
1992
1993
1994
Publication of IPCC SAR Fingerprinting with
atmospheric temperature and SAT
Publication of IPCC TAR Fingerprinting with
ocean heat content
First use of Bayesian methods in DA studies
Introduction of space-frequency DA approach
Detection of GS fingerprint in SAT
Recognition that Optimal detection is
regression First use of space-time DA methods
Introduction of multi-pattern fingerprinting
1995
1996
1997
1998
1999
2000
2001
2002
Publication of IPCC FAR Fingerprinting with
zonal-mean rainfall, water vapor, and surface
specific humidity
Fingerprinting with tropopause height, sea-level
pressure, MSU T4 and T2 temperatures
Fingerprinting with continental runoff CCSP
Report 1.1 resolves MSU problem
First assessment of fractional attributable
risk for an extreme event
All sins of omission or commission are
unambiguously attributable to Ben Santer
2003
2004
2005
2006
2007
2008