Call Girls In Bloom Boutique | GK-1 ☎ 9990224454 High Class Delhi NCR 24 Hour...
IUKWC Workshop Nov16: Developing Hydro-climatic Services for Water Security – Session 5 – Item 1 M_Widmann
1. The challenge of providing defensible
downscaled and bias-corrected
climate simulations
Martin Widmann
School of Geography, Earth and Environmental Sciences
University of Birmingham
coworkers include
D. Hannah, S. Krause, A. van Loon (Univ. Birmingham)
D. Maraun (Univ. Graz), J. Gutierrez (Univ. Santander)
IUKWC workshop, Pune, 30. Nov. 2016
2. Multi-model mean precipitation projections from
CMIP5 and CMIP3 ensembles (relative to 1986-2005)
(Knutti and Sedlacek,
NCC 2013)
Stippling: high model agreement
Hatching: no significant change
- biased
- lack of spatial detail
- India: problems with monsoon representation
-> need for bias correction and downscaling
3. Statistical downscaling (Perfect Prog(nosis))
(courtesy D. Maraun)
derive statistical link
between large and
small scales
apply to GCM output
requires realistically
simulated predictors
(perfect prognosis)
4. RCMs often need bias correction
mean precipitation in ERA40-driven RCMs (from ENSEMBLES)
Kotlarski et al.,
2014
5. Perfect Prog(nosis) Downscaling vs.
Model Output Statistics (bias correction, quantile mapping)
Perfect Prog Model Output Statistics
6. There are many different downscaling methods
Perfect Prog
- deterministic (linear, non-linear, analog)
- probabilistic/stochastic (linear, non-linear, resampling)
- weather generators
Challenging predictor requirements
Bypasses complex synoptic- and mesoscale processes that may be
successfully simulated and describes them with simple statistical
models
Model Output Statistics
- deterministic (linear, non-linear (e.g. quantile mapping))
- probabilistic/stochastic
(e.g. Maraun et al., JGR 2010)
8. Validation indices and performance measures
Full list of indices available at www.value-cost.eu/indices
(Maraun, Widmann et al., Earth Futures 2015)
9. VALUE validation: setup and implementation
Experiment 1:
- perfect predictors from ERA-I
- Validation at 85 European stations
- implemented as web portal
- cross-validation
- approx. 40 methods evaluated
- upcoming special issue in IJC
Experiment 2 (yet to be done):
- use pseudo-reality to validate
low-frequency variability
12. VALUE validation: bias in correlation length
(Widmann, Bedia et al., IJC in preparation)
13. Bias correction: nonsense mapping (SH T onto NH precip)
T precip obs precip after quantile mapping
(SH) (Germany) (independent validation period)
(Maraun, Shepherd, Widmann et al.,
Nat. Clim. Change revised)
Bias correction can map
completely unrelated variables
and the lack of a link will not be
detected by distribution-based
cross-validation.
Timeseries-based validation
would detect the problem.
mean
T
95th precentile
14. Bias correction: temporal structure (precipitation over Peru)
(Maraun, Shepherd, Widmann et al.,
NCC revised)
15. Bias correction: climate change signal
(MAM temperatures in Sierra Nevada)
GCM (GFDL-CM3) GCM corrected RCM (WRF)
(Maraun, Shepherd, Widmann et al., NCC revised)
1981-2000
(2081-2100) – (1981-2000)
Analogous problem holds
for RCM bias correction:
If the relevant processes are
not simulated BC will
not help.
16. N
What do we need to redraw this
for specific climate information?
- precisely define target variable
(aspects of distribution,
temporal and spatial variability)
- understand which processes
are relevant for target variable
- validate GCM-RCM-BC chain
for the target to the extent
possible (low-frequency is the
problem)
- evaluate representation of the
relevant processes, in particular
for low-frequency variability in
the GCM-RCM-BC chain
Can only be done in collaboration between
global climate modelling, downscaling and
impact communities !
No general advice on the ideal regional
climate product possible.