Transcript Hideo Shiogama 1

Uncertainty propagation from climate change
projections to impacts assessments:
water resource assessments in South America
Hideo Shiogama1, Seita Emori1 , 2, Naota Hanasaki1,
Manabu Abe1, Yuji Masutomi3, Kiyoshi Takahashi1, and
Toru Nozawa1
1 National
Institute for Environmental Studies
2 Atmosphere and Ocean Research Institute, University of Tokyo
3 Center for Environmental Science in Saitama
AOGCMs
Biases of
current climate
Uncertainty of future
climate projections
Impact model
Uncertainty of
impact assessments
• Uncertainty of climate change projections propagates to impact
assessments.
• Impact researchers have often investigated relations between
regional impact assessments and regional climate changes.
• However large-scale climate changes can affect regional impacts.
• How to examine relations between large-scale climate changes
and regional impact assessments?
• How to constrain the uncertainty of impact assessments?
Toward more consistent analysis and communications
between climate scientists and impact researchers.
Parallel approach in the IPCC AR5
Moss et al. (2010, Nature)
We have developed a method to examine uncertainty
propagation from climate to impact and to determine metrics
relating to impact assessments.
Water resource impact assessments in South America
A global hydrological model (Hanasaki et al. 2008)
• Inputs: △T and △P from 14 AOGCMs of CMIP3.
• Outputs: 14 assessments of annual mean runoff changes
(△R).
• Changes from 1980-1999 to 2080-2099 (SRES A2).
• Normalized by the global mean △T of each AOGCM.
Uncertainties in annual mean runoff changes
• What kind of uncertainties in climate change projections did
affect △R?
• Is the ensemble mean assessment the best estimate?
How to examine relations between large-scale
climate change patterns and △R in SA?
Singular Value Decomposition Analysis
T0
P0
△T
△P
△R
• Covariance matrix:
C=Cov[△R/△Tgm, (△T /△Tgm, △P /△Tgm)]
• Singular value decomposition: C=UTΣV
• This statistical method tells us pairs of △R mode and
(△T, △P ) mode such that the covariance between
their expansion coefficients is maximized.
1st modes (about 50%)
upward
downward
2nd modes (about 20%)
upward
downward
How to examine patterns of present
climate simulations relating to the
uncertainties of impact assessment?
Regressions between the present climate
simulations and the expansion coefficients of
the runoff modes
T0
P0
△T
△P
△R
Present climate patterns relating to the 1st runoff mode
Vertical circulations in the present
upward
downward
Vertical circulations in the future
upward
downward
Present climate patterns relating to the 2nd runoff mode
Vertical circulations in the present
upward
downward
Vertical circulations in the future
upward
downward
How to determine metrics relating to the
uncertainties of impact assessments?
How to determine metrics?
Biases of surface air temperature
(from ERA40)
Present climate
patterns associated
with the leading
runoff modes.
Biases of precipitation
(from CMAP)
Inner products
Runoff modes vs. present climate biases
Constraining the uncertainty of runoff changes
Conclusions
• The ensemble mean is not always the best
estimate.
• A naive overreliance on consensus assessments
could lead to inappropriate adaptation policies.
• Our new approach could help find a targetoriented metric for a particular aspect of climate
change projections and impact assessments over
a particular region.
• This approach can help promote more
communications between climate scientists and
impact researchers.