ENSEMBLES RT2B Production of regional climate scenarios for impacts assessments RT leaders: Clare Go

1
ENSEMBLES RT2BProduction of regional climate
scenarios for impacts assessmentsRT leaders
Clare Goodess Daniela Jacob
Third General Assembly, 21 November 2006
2
Outline of RT2B approaches D2B.1 D2B.2
Preparing datasets RCM data server
D2B.3 Reanalysis D2B.13 Observed
D2B.15 GCM-based D2B.17
Developing/testing models Statistical D2B.5,
D2B.16 Dynamical D2B.9, D2B.10
Issues and methods Ensemble averaging
D2B.6 Pattern scaling D2B.7, D2B.25 Weighting
D2B.8 GCM-RCM matrix RCM quick-look D2B.21
Interactions with users (RT6) Web-based
downscaling service D2B.4, D2B.19,
D2B.23 Questionnaires Development of tools
D2B.18 Preliminary assessment D2B.20
s2d statistical downscaling D2B.12
s2d dynamical downscaling INM/RCA MARS
Modification of SDS methods for probabilistic
framework D2B.14
From month 30, the emphasis is on synthesis,
application and scenario construction
RT3
ERA_at_50/25 D3.1.4, D3.1.5 RCM weights
D3.2.2 RCM system D3.3.1
RT1 Grand PDFs RT2A stream 1 runs (s2d,
ACC) RT5 gridded data set
RT3
D2B.11 mo 31
25 km scenario runs D2B.22 mo 36
Final RCM system D3.3.2 GCM/RCM skill/biases
D3.4.1
Dynamical and statistical downscaling Probabilisti
c regional scenarios and tools
RCM quick look analysis D2B.24 mo 40
Mo 36 on
Questions issues Sources of uncertainty Reducing
uncertainty Robustness of SDS (D2B.27)
Synergistic use of SDS/DDS

• Applications to case studies
• Alps, Mediterranean (D2B.28)
• Storms, CWTs, blocking.
• Forestry, water.

Recommendations guidance on methods for the
construction of probabilistic regional climate
scenarios D2B.26 mo 42
3
Deliverable D2B.12/D6.10Recommendations for the
application of statistical downscaling methods
to seasonal-to-decadal hindcasts in ENSEMBLES
(Task 2B.2.6) Valentina Pavan

• 3rd ENSEMBLES General Assembly Lund, 20-24
  November 2006

4
A special thank to all those who contributed with
examples and ideas to this Deliverable!!

• ARPA-SIM Rodica Tomozeiu, Stefano Marchesi,
  Vittorio Marletto
• ECMWF Francisco Doblas-Reyes
• ICTP Filippo Giorgi
• INM Antonio Cofiño
• MeteoSwiss Mark Linger, Paul Della-Marta
• NMA Aristita Bisuioc
• UC Jose Manuel Gutiérez
• UEA Clare Goodess
• UNILIV Andy Morse
• University of Reading Caio Augusto Dos Stantos
  Coelho

5
Outline of D2B.12

• Introduction
• Specification of ENSEMBLES user requirements
• On the necessity and benefits of applying
  downscaling to GCM outputs
• Issues related with the application of
  statistical downscaling to seasonal and decadal
  hindcasts
• An ENSEMBLES test application
• Observational data for the ENSEMBLES study areas
• Conclusions, final recommendations and proposals
  for further work

6
4.4 Perfect prog vs MOS approach for
downscaling seasonal predictions

• The MOS (Model Output Statistics) approach
  consists in identifying the statistical relation
  between model generated large-scale variability
  and local climate variability.
• The Perfect Prog approach consists in identifying
  the statistical relation between the observed
  large-scale variability and the observed local
  climate variability

7
Advantages and disadvantages of the MOS approach

• directly deals with model biases and errors
  (also taking care of shifts in model generated
  patterns!!!)
• reduces the probability of overfitting
• requires long statistics of CGCM predictions
  obtained with the same model version (covering a
  period over which OBS are available)
• Applicable only to climate predictions for which
  a clear time correspondence is present between
  OBS and model (seasonal, AMIP,)
• NB If correctly formulated a perfect prog SD
  scheme should be equivalent to a MOS SD scheme of
  the same type (MLR, ACC using same
  predictors/predictands).

8
Downscaled operational summer predictions for
Northern Italy (ARPA-SIM)

• Type of SD scheme used MLR applied to
  multi-model operational seasonal predictions run
  at ECMWF (ECMWF operat. UKMO operat. Using same
  number of ensemble members)
• Two versions implemented MOS and Perfect Prog
  (PP)

9
2) The two approaches are equivalent in terms of
correlation between OBS and pretiction
Tmax JJA Corr MOS-OBS 0.72 Corr PP-OBS 0.64
Tmin JJA Corr MOS-OBS 0.61 Corr PP-OBS 0.59
10
The predictions of Tmax for summer 2003
OBS
PP
MOS
11
5. An ENSEMBLES test application

• The ENSEMBLES Statistical Downscaling Portal
  (SDP) is being developed by INM and UC (for more
  details see http//www.meteo.unican.es/ensembles).
  

12
7. Conclusions, final recommendations and
proposals for further work

• Downscaling is essential for the use of CGCM
  seasonal-to-decadal predictions in application
  studies
• A portal has been set up to help end-users to
  produce the downscaled prediction needed using
  their own observational data and ENSEMBLE
  large-scale ensemble predictions. To be extended
  to other SD schemes!!
• SD seasonal and decadal application are
  intrinsically different. A seamless system for
  producing predictions across all time-scales is
  not possible.

13
7. Conclusions (continued)

• SD schemes must always be optimised before used
  for each specific application
• Downscaled statistical ensemble predictions can
  be produced by
• Using multiple CGCM
• Using several SD schemes optimised with respect
  to different criteria
• The weighted multi-model ensemble approach is
  superior with respect to the plain multi-model at
  all time scales
• In order to produce robust predictions the
  observational data-set used to calibrate the SD
  scheme must have a sufficient resolution
• NB It is important that the discussions raised
  within this deliverable are continued in the
  continuation of the project!!!

14
Tailoring of ENSEMBLES regional climate scenario
outputs to user needs a questionnaire for users,
stakeholders and scenario developers

• Part 1 About you
• Part 2 Requirements of regional climate
  scenarios
• Part 3 Availability of observed climate data
• Part 4 Statistical downscaling and scenario
  generator tools

15
(No Transcript)
16
Outline of RT2B approaches D2B.1 D2B.2
Preparing datasets RCM data server
D2B.3 Reanalysis D2B.13 Observed
D2B.15 GCM-based D2B.17
Developing/testing models Statistical D2B.5,
D2B.16 Dynamical D2B.9, D2B.10
Issues and methods Ensemble averaging
D2B.6 Pattern scaling D2B.7, D2B.25 Weighting
D2B.8 GCM-RCM matrix RCM quick-look D2B.21
Interactions with users (RT6) Web-based
downscaling service D2B.4, D2B.19,
D2B.23 Questionnaires Development of tools
D2B.18 Preliminary assessment D2B.20
s2d statistical downscaling D2B.12
s2d dynamical downscaling INM/RCA MARS
Modification of SDS methods for probabilistic
framework D2B.14
From month 30, the emphasis is on synthesis,
application and scenario construction
RT3
ERA_at_50/25 D3.1.4, D3.1.5 RCM weights
D3.2.2 RCM system D3.3.1
RT1 Grand PDFs RT2A stream 1 runs (s2d,
ACC) RT5 gridded data set
RT3
D2B.11 mo 31
25 km scenario runs D2B.22 mo 36
Final RCM system D3.3.2 GCM/RCM skill/biases
D3.4.1
Dynamical and statistical downscaling Probabilisti
c regional scenarios and tools
RCM quick look analysis D2B.24 mo 40
Mo 36 on
Questions issues Sources of uncertainty Reducing
uncertainty Robustness of SDS (D2B.27)
Synergistic use of SDS/DDS

• Applications to case studies
• Alps, Mediterranean (D2B.28)
• Storms, CWTs, blocking.
• Forestry, water.

Recommendations guidance on methods for the
construction of probabilistic regional climate
scenarios D2B.26 mo 42