OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
Exploring nearly one-in-a-millennium scenariosof extreme rainfall through dynamicallydownscaling palaeoclimatic simulations
Juan José Gómez-Navarro, Christoph Raible,Sandro Blumer and Olivia Martius
September 1, 2015
Outline
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
1 Motivation and modelling framework
2 Downscaling results and bias correction
3 Caveats and outlook
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Outline
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
1 Motivation and modelling framework
2 Downscaling results and bias correction
3 Caveats and outlook
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Aim of the project
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
Studying extremes is important...
• Produce severe damages (relevant to insurancecompanies)
• The fingerprint of Climate Change in extremes isespecially relevant for policy makers
... but challengeling
• Infrequent by definition
• Requires having long series we do not have
• We try to do something in this respect (withmodels!)
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Aim of the project
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
Studying extremes is important...
• Produce severe damages (relevant to insurancecompanies)
• The fingerprint of Climate Change in extremes isespecially relevant for policy makers
... but challengeling
• Infrequent by definition
• Requires having long series we do not have
• We try to do something in this respect (withmodels!)
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Model chain
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
GCMCESM
0 15 30 45 60 75 90 105 120 135 150prec
mm
RCMWRF
BIAS CORRECTION
HydrologyPREVAH
1D-HydrailicBASEMENT-
ETH
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Model chain
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
GCMCESM
RCMWRF
BIAS CORRECTION
HydrologyPREVAH
1D-HydrailicBASEMENT-
ETH
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Model chain
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
GCMCESM
RCMWRF
BIAS CORRECTION
0
5
10
15
20
25
30
35
40
45
50
0 5 10 15 20 25 30
ER
AIN
CESM
QQ plot JJA 1-day
HydrologyPREVAH
1D-HydrailicBASEMENT-
ETH
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Model chain
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
GCMCESM
RCMWRF
BIAS CORRECTION
HydrologyPREVAH
1D-HydrailicBASEMENT-
ETH
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Model chain
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
GCMCESM
RCMWRF
BIAS CORRECTION
HydrologyPREVAH
1D-HydrailicBASEMENT-
ETH
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Details of the climate component
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
GCM
• CESM run in the University of Bern
• Transient simulation: fully coupled atmosphere &ocean driven by external forcings
• 850–2010 plus control run
RCM
• Weather Research and Forecasting model (WRF)
• Four nested domains down to 2 km
• Convecting-resolving resolution: no convectionparameterization
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Details of the climate component
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
GCM
• CESM run in the University of Bern
• Transient simulation: fully coupled atmosphere &ocean driven by external forcings
• 850–2010 plus control run
RCM
• Weather Research and Forecasting model (WRF)
• Four nested domains down to 2 km
• Convecting-resolving resolution: no convectionparameterization
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Outline
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
1 Motivation and modelling framework
2 Downscaling results and bias correction
3 Caveats and outlook
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Picking dates
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
The problem
• It is NOT possible to run 1000 years of 2km-resolution simulation!• Alternative: select "interesting dates" a priori withinthe GCM... but how?
Solution #0
• A region over Switzerland is selected (12 grid points inthe GCM)
• The precipitation over this region is used to selectevents
• Simple, but ignores the drivers of precipitation
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Picking dates
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
The problem
• It is NOT possible to run 1000 years of 2km-resolution simulation!• Alternative: select "interesting dates" a priori withinthe GCM... but how?
Solution #0
• A region over Switzerland is selected (12 grid points inthe GCM)
• The precipitation over this region is used to selectevents
• Simple, but ignores the drivers of precipitation
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Criterion in detail
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
Algorithm
• The 4 most rainy days are selected
• The same is repeated for 2, 3, 5 and 10 windows
• Independently for each season (to minimise systematicerrors arising from GCM biases)
This results in 4× (1+ 2+ 3+ 5+ 10)× 4=336 simulated days,which is feasible
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Criterion in detail
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
Algorithm
• The 4 most rainy days are selected
• The same is repeated for 2, 3, 5 and 10 windows
• Independently for each season (to minimise systematicerrors arising from GCM biases)
This results in 4× (1+ 2+ 3+ 5+ 10)× 4=336 simulated days,which is feasible
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Criterion in detail
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
Algorithm
• The 4 most rainy days are selected
• The same is repeated for 2, 3, 5 and 10 windows
• Independently for each season (to minimise systematicerrors arising from GCM biases)
This results in 4× (1+ 2+ 3+ 5+ 10)× 4=336 simulated days,which is feasible
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Criterion in detail
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
Algorithm
• The 4 most rainy days are selected
• The same is repeated for 2, 3, 5 and 10 windows
• Independently for each season (to minimise systematicerrors arising from GCM biases)
This results in 4× (1+ 2+ 3+ 5+ 10)× 4=336 simulated days,which is feasible
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Downscaling Results
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
10 W
10 W
0
0
10 E
10 E
20 E
20 E 30 E
35 N 35 N
40 N 40 N
45 N 45 N
50 N 50 N
55 N 55 N
6 E
6 E
8 E
8 E
10 E
10 E
12 E
12 E
14 E
14 E
44 N 44 N
45 N 45 N
46 N 46 N
47 N 47 N
48 N
0
15
30
45
60
75
90
105
120
135
150
prec (mm)
0
0
10 E
10 E
20 E
20 E
40 N 40 N
45 N 45 N
50 N 50 N
3 E
3 E
6 E
6 E
9 E
9 E
12 E
12 E
15 E
15 E
42 N 42 N
45 N 45 N
48 N 48 N
51 N 51 N
10 W
10 W
0
0
10 E
10 E
20 E
20 E 30 E
35 N 35 N
40 N 40 N
45 N 45 N
50 N 50 N
55 N 55 N
GCM
Domain 1 (54 km) Domain 2 (18 km)
Domain 3 (6 km)
Domain 4 (2 km)
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Bias correction: generalities
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
In some applications, biases are not critical (e.g. delta approach),BUT as part of the model chain, biases are crucial (non linearprocesses)
WRF-ERAinWRF-CESM
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Bias correction: generalities
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
In some applications, biases are not critical (e.g. delta approach),BUT as part of the model chain, biases are crucial (non linearprocesses)
ClimatesimulationClimate
simulation climateObserved
WRF-ERAinWRF-CESM
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Bias correction: generalities
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
In some applications, biases are not critical (e.g. delta approach),BUT as part of the model chain, biases are crucial (non linearprocesses)
ClimatesimulationClimate
simulation climateObserved
Transferfunction
WRF-ERAinWRF-CESM
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Bias correction: generalities
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
In some applications, biases are not critical (e.g. delta approach),BUT as part of the model chain, biases are crucial (non linearprocesses)
ClimatesimulationClimate
simulation climateObserved
Transferfunction
Simulatedextreme
"Actual"extreme
WRF-ERAinWRF-CESM
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Bias correction: generalities
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
In some applications, biases are not critical (e.g. delta approach),BUT as part of the model chain, biases are crucial (non linearprocesses)
ClimatesimulationClimate
simulation climateObserved
Transferfunction
Simulatedextreme
"Actual"extreme
WRF-ERAinWRF-CESM
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Bias correction: examples
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
0.00
0.05
0.10
0.15
0.20
0.25
0 5 10 15 20 25 30 35 40 45 50
PD
F
BW
ERACESM
0
10
20
30
40
50
PDF JJA 1-day
prec (mm)
0.00
0.00
0.00
0.01
0.01
0.01
0.01
0.01
0.02
0 20 40 60 80 100 120 140 160
PD
F
BW
0
50
100
150
200
250
ERACESM
PDF DJF 10-day
prec (mm)
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Bias correction: examples
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
0.00
0.05
0.10
0.15
0.20
0.25
0 5 10 15 20 25 30 35 40 45 50
PD
F
BW
ERACESM
0
10
20
30
40
50
PDF JJA 1-day
prec (mm)
0.00
0.00
0.00
0.01
0.01
0.01
0.01
0.01
0.02
0 20 40 60 80 100 120 140 160
PD
F
BW
0
50
100
150
200
250
ERACESM
PDF DJF 10-day
prec (mm)
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Bias correction: fitting the CDF
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
0.0001
0.001
0.01
0.1
1
051015202530354045
PD
F
ERACESM
0
10
20
30
40
50
CDF JJA 1-day
PDF = AeB·prec
⇓y︷ ︸︸ ︷
ln(PDF) = lnA+B ·x︷︸︸︷
prec
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Bias correction: fitting the CDF
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
0.0001
0.001
0.01
0.1
1
051015202530354045
PD
F
ERACESM
0
10
20
30
40
50
CDF JJA 1-day
PDF = AeB·prec
⇓y︷ ︸︸ ︷
ln(PDF) = lnA+B ·x︷︸︸︷
prec
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Bias correction: fitting the CDF
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
0.0001
0.001
0.01
0.1
1
051015202530354045
PD
F
ERACESM
0
10
20
30
40
50
CDF JJA 1-day
PDF = AeB·prec
⇓y︷ ︸︸ ︷
ln(PDF) = lnA+B ·x︷︸︸︷
prec
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Bias correction: Quantile mapping
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
0
5
10
15
20
25
30
35
40
45
50
0 5 10 15 20 25 30
ER
AIN
CESM
QQ plot JJA 1-day
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Bias correction: Results
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
Lin. Fit Raw sim. Q. map
0 15 30 45 60 75 90 105 120 135 150
prec (mm)
DJF 10-dayaggregated
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Outline
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
1 Motivation and modelling framework
2 Downscaling results and bias correction
3 Caveats and outlook
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Limitations we are aware of
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
• The dataset used to remove biases is not itself unbiased ⇒use observations instead of model data
• The number of cases is limited ⇒ downscale more cases
• Bias not systematic, but seasonal- and event-dependent ⇒seasonal analysis
• We do not search for the physical mechanisms underlayingevents ⇒ split the analysis for each type of event (e.gVb-events,...)???
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Limitations we are aware of
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
• The dataset used to remove biases is not itself unbiased ⇒use observations instead of model data
• The number of cases is limited ⇒ downscale more cases
• Bias not systematic, but seasonal- and event-dependent ⇒seasonal analysis
• We do not search for the physical mechanisms underlayingevents ⇒ split the analysis for each type of event (e.gVb-events,...)???
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Limitations we are aware of
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
• The dataset used to remove biases is not itself unbiased ⇒use observations instead of model data
• The number of cases is limited ⇒ downscale more cases
• Bias not systematic, but seasonal- and event-dependent ⇒seasonal analysis
• We do not search for the physical mechanisms underlayingevents ⇒ split the analysis for each type of event (e.gVb-events,...)???
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Limitations we are aware of
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
• The dataset used to remove biases is not itself unbiased ⇒use observations instead of model data
• The number of cases is limited ⇒ downscale more cases
• Bias not systematic, but seasonal- and event-dependent ⇒seasonal analysis
• We do not search for the physical mechanisms underlayingevents ⇒ split the analysis for each type of event (e.gVb-events,...)???
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
How extreme are the "extremes"we pick?
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0 10 20 30 40 50 60
PD
F
0102030405060
PDF DJF 1-day
prec (mm)
ERACESM
The "extremes" we previously selected within the GCM are notso extreme once downscaled
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
How extreme are the "extremes"we pick?
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0 10 20 30 40 50 60
PD
F
0102030405060
PDF DJF 1-day
prec (mm)
ERACESM
Statistical d
ownscaling?
The "extremes" we previously selected within the GCM are notso extreme once downscaled
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Conclusions
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
• Novel methodology that allows to simulate unprecedented,yet physically based extreme situations
• Although model data, the climate information is bias-free
• Most technical challenges (in the climate models, but andalso in the hydrological models) already addressed
• Further research is required to consistently remove biases ofunprecedented events
• The selection of the dates needs to be considered morecarefully, and more robust methodology should be developed
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Methodological uncertainties toaddress
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
• The CDFs not always line up
• Quantile mapping is ill-defined for extremely infrequentevents
• Should the transfer function be calibrated for each gridpoint? Separately for each region (regionalisation)?
0.0001
0.001
0.01
0.1
1
020406080100120140
CD
F
050100150200250
CDF DJF 10-day
ERACESM
prec. (mm)
0
10
20
30
40
50
60
70
80
90
100
0 10 20 30 40 50 60 70
QQplot JJA 5-day
?
ER
AIN
CESM
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Methodological uncertainties toaddress
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
• The CDFs not always line up
• Quantile mapping is ill-defined for extremely infrequentevents
• Should the transfer function be calibrated for each gridpoint? Separately for each region (regionalisation)?
0.0001
0.001
0.01
0.1
1
020406080100120140
CD
F
050100150200250
CDF DJF 10-day
ERACESM
prec. (mm)
0
10
20
30
40
50
60
70
80
90
100
0 10 20 30 40 50 60 70
QQplot JJA 5-day
?
ER
AIN
CESM
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern
Methodological uncertainties toaddress
OESCHGER CENTRECLIMATE CHANGE RESEARCH
UNIVERSITÄTBERN
• The CDFs not always line up
• Quantile mapping is ill-defined for extremely infrequentevents
• Should the transfer function be calibrated for each gridpoint? Separately for each region (regionalisation)?
0.0001
0.001
0.01
0.1
1
020406080100120140
CD
F
050100150200250
CDF DJF 10-day
ERACESM
prec. (mm)
0
10
20
30
40
50
60
70
80
90
100
0 10 20 30 40 50 60 70
QQplot JJA 5-day
?
ER
AIN
CESM
J.J. Gómez-Navarro 〉 5th European Windstorm Workshop 〉 31 Aug. – 2 Sep. 2015 Bern