Climate change integrated assessment methodology for cross-sectoral adaptation and vulnerability in Europe

1
Climate change integrated assessment methodology
for cross-sectoral adaptation and vulnerability
in Europe
Climate change scenarios incorporated into the
CLIMSAVE Integrated Assessment Platform
For further information contact Martin Dubrovsky
(email ma.du_at_seznam.cz) or visit the project
website (www.climsave.eu)
Funded under the European Commission Seventh
Framework Programme Contract Number 244031
2
Presentation structure

• 1. Introduction
• 2. Methodologies for preparing reduced-form
  ensembles of future climate scenarios (...focus
  on uncertainties)
• 2.1 GCM ensemble (CMIP3 data IPCC-AR4) for
  European case study
• 2.2 UKCP09 data for Scottish case study
• representativeness of the reduced-form
  ensembles
• 3. Comparison of GCM-based vs. UKCP09 scenarios
• 4. Summary Conclusion

3
Introduction CLIMSAVE project

• CLIMSAVE project (www.climsave.eu 2010-2013)
• coordinated by the Environmental Change
  Institute, University of Oxford
• 18 partners from 13 countries (incl. China and
  Australia)
• Aim integrated methodology to assess
  cross-sectoral climate change impacts, adaptation
  and vulnerability 
• The main product of CLIMSAVE a user-friendly,
  interactive web-based tool (Integrated Assessment
  Platform IAP) that will allow stakeholders to
  assess climate change impacts and vulnerabilities
  for a range of sectors
• IAP is based on an ensemble of meta-models, which
  are run with the user-selected climatic data
  representing present and future climates
• When creating an ensemble of climate change
  scenarios for the IAP, two requirements were
  followed
• 1. an ensemble of climate change scenarios is not
  large, and
• 2. it satisfactorily represents known
  uncertainties in future climate projections.

4
GCM-based scenarios(based on monthly GCM
outputsfrom IPCC-AR4 database /CMIP3/Europe)

5
GCMs in CMIP3 database
We use 16 SRES-A2 simulations of 24 GCMs x 6
emission scenarios (incomplete matrix).
6
Pattern scaling approach allows to reflect
multiple uncertainties - where several ?TG
values are used to multiply several GCM-based
patterns
Pattern scaling is used to create a set of
climate change scenarios
?TG change in global mean temperature ?XS
standardised scenario (related to ?TG 1K
derived from GCMs)
?X(t) ?XS x ?TG(t)
uncertainty in pattern ( modelling uncertainty)
uncertainty in ?TG (uncertainties in emissions
climate sensitivity)
X
3 sources of uncertainty
7
Reducing an ensemble of scenarios

• When using the above pattern-scaling approach
  (GCM-based standardised scenarios are scaled by
  MAGICC-modelled ?TGLOB values), we
• find a representative subset of GCMs, which
  satisfactorily represents the inter-GCM
  uncertainty,
• choose several ?TGLOB values, which account for
  uncertainties in emission scenarios and climate
  sensitivity.

8
Choosing a setof ?TGLOB values
?TGLOB (modelled by MAGICC for 6 SRES emissions
scenarios x 3 climate sensitivities)
Considering SRES emissions scenarios and 1.5-4.5K
interval for climate sensitivity 2050 effect
of uncertainty in climate sensitivity is
(slightly) larger 2100 both effects are about
the same CLIMSAVE employs 12 values of ?TGLOB (
4 emissions x 3 climate sensitivity) Reduced set
of 3 values
emissions clim.sensitivity high
scenario SRES-A1FI 4.5 K low scenario SRES-B1
1.5 K middle scen. SRES-A1b 3.0 K
9
Defining a representative subset of GCMs
Two approaches are used here to define a
representative GCM subset A. expert-based
judgement ? CLIMSAVE subset B. applying
objective criteria ? EU5a subset
10
CLIMSAVE subset (method expert choice)
Input
summer (JJA)
winter (DJF)
?TAVG

?PREC
Output (5 GCMS) MPEH5, HADGEM,
GFCM21, NCPCM, MIMR
11
Defining a EU5a subset(based on objective
criteria)

• Target size of the subset 5 GCMs
• The subsets will consist of
• best GCM Quality(GCM) ability to reproduce
  annual cycle of TEMP and PREC in a given 0.5x0.5
  gridbox
• central GCM (8D metrics changes in seasonal
  TEMP and PREC)
• 3 most diverse GCMs (maximising a sum of
  inter-GCM distances the same metrics)
• (prior to analysis, GCM outputs were regridded
  into 0.5x0.5 grid common with the CRU
  climatology)

12
Best GCM
Best GCM Q f RV(Temp), RV(Prec)
...based on RV(Prec)
MPEH5
GCM which is the best in the largest number of
gridboxes
Quality(GCM) ability to reproduce annual cycle
of TEMP and PREC in a given 0.5x0.5 gridbox
...based on RV(Temp)
13
Central GCM ( closest to Centroid)
GCM which is the Central GCM in the largest
number of gridboxes (metrics Euclidean(8D
seasonal changes in TEMP and PREC)

• note MPEH5 and HadGEM, which were found to be
  among the best GCMs, are also among the three
  most central GCMs

CSMK3
14
3 mutually most diverse GCMs
HADGEM, GFCM21, IPCM4
15
5 GCMs for Europe(3799 0.5x0.5 land grid boxes)
1 centroid
1 best
3 most diverse
3bests
? EU5a MPEH5, HADGEM, GFCM21, CSMK3,
IPCM4 vs. CLIMSAVE MPEH5, HADGEM, GFCM21,
NCPCM, MIMR
16
GCM subset validation(number of significant
differences in AVGs and STDs (subset vs. 16 GCMs)
EU5a vs. 16GCMs
CLIMSAVE vs. 16GCMs

• Whole Europe
• - the CLIMSAVEs problem significant
  underestimation of inter-GCM variability in ?TEMP
• - EU5a performs better
• both TEMP and PREC
• both AVG and STD
• UK
• - not such large differences between the two
  subsets

avg(?T)
std(?T)
avg(?P)
insignificant difference A16G-½S16G, lt
avgsubset lt A16G½S16G ?S16G, lt stdsubset lt
3/2.S16G
std(?P)
17
UKCP09-based climate scenarios

• UKCP09 future climate projection developed by
  UK Met. Office (http//ukclimateprojections.defra.
  gov.uk). It is based on
• PPE of HadSM3 simulations ( simplified HadCM3)
  (PPE Physically Perturbed Ensemble 31 key
  model parameters perturbed)
• downscaled by Hadley RCM,
• adjusted by outputs from 12 other GCMs,
• and disaggregated into 10000 values by a
  statistical emulator
• Probabilistic projections of climatic
  characteristics is given in terms of 10000
  possible values (realisations) for each 25x25 km
  grid box over UK
• the projection is available for 3 SRES emission
  scenarios (low B1, medium A1b, high A1FI)
• Aim Reduce 3 (emissions) x 10,000 realisations
  to reasonably large ensemble of scenarios
  (preserving the ensemble variability)

18
UKCP09 climate scenarios- creating the
reduced-form ensemble

• 3D space ?Tannual, ?Psummer, ?Pwinter
• 27 points relate to 3x3x3 combinations of low,
  med, high changes in the three variables median,
  10th and 90th percentiles along each of 13 lines
  going through the cubes center and defined by
  corners/centres of sides/centres of edges of the
  cube
• 27 scenarios the means of 10 neighbours closest
  to each of 27 points (in a 3D space)

?Ta
?Psummer
?Pwinter
27 climate change scenarios related to 3x3x3
combinations of (low, med, high) changes in
dTannual, dPsummer, dPwinter
19
UKCP09 (2050s) ?TEMPannual middle
WL-SL
WL-SM
WL-SH
WM-SL
WM-SM
WM-SH
WH-SL
WH-SM
WH-SH
?TEMPannual
?PRECONDJFM
?PRECAMJJAS
20
Same but for ?TEMPannual low
?TEMPannual
?PRECONDJFM
?PRECAMJJAS
slide 20
21
Same but for ?TEMPannual high
?TEMPannual
?PRECONDJFM
?PRECAMJJAS
22
UKCP09 full vs. reduced ensembles

• Q How does the reduced UKCP09 ensemble represent
  the original ensemble?
• input full database 30000 scenarios
• (3 emission scenarios) x (10000 realisations)
• for each grid, climate variable and 10 year
  timeslice)
• reduced-form scenarios 91 scenarios
• (3 emission scenarios) x (27 scenarios
  representing 3x3x3 combinations of
  low/medium/high values of ?Tannual, ?Psummer,
  ?Pwinter
• for each grid, climate variable, 2020s and 2050s
  timeslices
• maps avg(?std) from 10000 vs. 27 scenarios for
  2050s (this and following 2 slides)

3 emis.scen.
high (SRES-A1FI)
med (SRES-A1b)
low (SRES-B1)
JJA
DJF
JJA
DJF
JJA
DEC
JJA
DJF
10000 members
3x 10000 memb.
?PREC
full vs. reduced ensembles good fit between the
means
27 clusters
3x 27 clust.
JJA
DJF
JJA
DJF
JJA
DEC
JJA
DJF
23
UKCP09 full vs. reduced ensembles
3 emis.scen.
high (SRES-A1FI)
med (SRES-A1b)
low (SRES-B1)
JJA
DJF
JJA
DJF
JJA
DEC
JJA
DJF
10000 members
3x 10000 memb.
?TEMP
perfect fit
3x 27 clust.
27 clusters
3x 10000 memb.
10000 members
?PREC
perfect fit
3x 27 clust.
27 clusters
JJA
DJF
JJA
DJF
JJA
DEC
JJA
DJF
24
UKCP09 vs. GCM (only UK territory)

• UKCP09
• original ensemble 3 emissions x 10000
  realisations 30000 scenarios
• reduced ensemble 3 emissions x 27 scenarios
  81 scenarios
• GCMs
• original ensemble 16 GCMs x 4 emissions x 3
  clim.sens. 192 scen.
• reduced ensemble 5 GCMs x 4 emissions x 3
  clim.sens. 60 scenarios
• UKCP09 vs GCMs
• ........................... UKCP09....... GCMs
• full datasets 30000 vs. 192 scenarios
• reduced dataset 81 vs. 60
  scenarios

25
UKCP09 vs GCMs avg(?PREC)
3 emis.scen.
high (SRES-A1FI)
med (SRES-A1b)
low (SRES-B1)
JJA
DEC
JJA
DEC
JJA
DEC
JJA
DEC
5GCMs x 3CS
reduced dataset
GCMs
16GCMs x 3CS
full dataset
? UKCP09 shows slightly larger reductions in PREC
10000 members
UKCP09
27 clusters
reduced dataset
JJA
DEC
JJA
DEC
JJA
DEC
JJA
DEC
26
UKCP09 vs GCMs avg(?TEMP)
3 emis.scen.
high (SRES-A1FI)
med (SRES-A1b)
low (SRES-B1)
JJA
DEC
JJA
DEC
JJA
DEC
JJA
DEC
5GCMs x 3CS
reduced dataset
GCMs
16GCMs x 3CS
? significant difference between GCM and UKCP09
full dataset
10000 memb.
UKCP09
reduced dataset
27 clusters
JJA
DEC
JJA
DEC
JJA
DEC
JJA
DEC
27
UKCP09 vs GCM std(?PREC)
3 emis.scen.
high (SRES-A1FI)
med (SRES-A1b)
low (SRES-B1)
JJA
DEC
JJA
DEC
JJA
DEC
JJA
DEC
5GCMs x 3CS
reduced dataset
GCMs
? GCMs the subset reproduces the internal
variability
16GCMs x 3CS
full dataset

• GCMs vs UKCP09 internal UKCP09 ensemble
  variability is larger
• (corresponds to larger avg(?TAVG) in UKCP
  scenarios)

10000 members
UKCP09
? UKCIP09 the reduced-form ensemble reduces
internal variability
27 clusters
reduced dataset
JJA
DEC
JJA
DEC
JJA
DEC
JJA
DEC
28
UKCP09 vs GCMs std(?TEMP)
3 emis.scen.
high (SRES-A1FI)
med (SRES-A1b)
low (SRES-B1)
JJA
JJA
JJA
DEC
DEC
DEC
JJA
DEC
5GCMs x 3CS
reduced dataset
GCMs
16GCMs x 3CS
? GCMs vs UKCP09 internal UKCP09 ensemble
variability is larger
full dataset
10000 memb.
UKCP09
reduced dataset
27 clusters
29
Summary Conclusions (1)

• Climate change impact studies require ensembles
  of climate change scenarios representing known
  uncertainties. Available scenario datasets were
  too large for CLIMSAVE, reductions were proposed.
  
• 2 case studies in CLIMSAVE 2 datasets to reduce
  in size
• GCMs (CMIP3 dataset of GCMs from various
  modelling groups)
• large ensemble 16 GCMs x 4 emissions x 3
  climate sensitivity 192 scenarios ( 3
  uncertainties)
• reduced-form ensemble 5 GCMs x 4 emissions x 3
  climate sensitivity (or 5 GCMs x
  3 dTglob) 60 (15) scenarios
• though the optimum subset varies across Europe,
  the single GCM subset still reasonably well
  represents the inter-GCM variability over
  majority of European territory
• UKCP09 PP(HadSM) HadRM statistical
  emulator
• large ensemble 10000 realisations x 3 emission
  scenarios 30000 scenarios (structural
  uncertainties within 10000 members also account
  for climate sensitivity uncertainty)
• reduced-form ensemble 27 scenarios x 3
  emissions 81 scenarios
• within-ensemble variability is lower (effect of
  natural climate variability is reduced)

30
Summary Conclusions (2)

• In both ensembles
• the reduced-form scenarios reasonably well
  represent means and variabilities of the original
  ensembles
• gt structural climate sensitivity emissions
  uncertainties are preserved
• GCMs vs UKCP09
• except for avg(?PREC), significant differences
  between the 2 ensembles were found
• these differences gtgt the differences related
  to reducing the original datasets