Cloud-resolvingensemble simulationsandMediterranean HPEs
Benoît Vié1
Olivier Nuissier1
Véronique Ducrocq1
1Météo-France - CNRM
10 June 2010
Why do we need a cloud-resolving EPS?
Z500
925 hPa windθ′
w
10
15
20
25
30
40
55
75
100
150
250
300
400
24h accumulated precipitation, 12 UTC 2 Nov.2008, AROME forecast and observations
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D1
Why do we need a cloud-resolving EPS?
◮ Cloud-resolving, non-hydrostaticNWP models produce very realisticforecasts
◮ Realistic 6= Real
◮ Runoff forecasts are very sensitive tothe rainfall forecasts, especially forsmall and steep mountainouswatersheds
◮ EPSs are one method to evaluate the
forecast uncertainty
10
15
20
25
30
40
55
75
100
150
250
300
400
24h accumulated precipitation, 12 UTC 2 Nov.2008, AROME forecast and observations
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D1
Why do we need a cloud-resolving EPS?
◮ Cloud-resolving, non-hydrostaticNWP models produce very realisticforecasts
◮ Realistic 6= Real
◮ Runoff forecasts are very sensitive tothe rainfall forecasts, especially forsmall and steep mountainouswatersheds
◮ EPSs are one method to evaluate the
forecast uncertainty
10
15
20
25
30
40
55
75
100
150
250
300
400
24h accumulated precipitation, 12 UTC 2 Nov.2008, AROME forecast and observations
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D1
Why do we need a cloud-resolving EPS?
◮ Cloud-resolving, non-hydrostaticNWP models produce very realisticforecasts
◮ Realistic 6= Real
◮ Runoff forecasts are very sensitive tothe rainfall forecasts, especially forsmall and steep mountainouswatersheds
◮ EPSs are one method to evaluate the
forecast uncertainty
10
15
20
25
30
40
55
75
100
150
250
300
400
24h accumulated precipitation, 12 UTC 2 Nov.2008, AROME forecast and observations
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D1
Why do we need a cloud-resolving EPS?
◮ Cloud-resolving, non-hydrostaticNWP models produce very realisticforecasts
◮ Realistic 6= Real
◮ Runoff forecasts are very sensitive tothe rainfall forecasts, especially forsmall and steep mountainouswatersheds
◮ EPSs are one method to evaluate the
forecast uncertainty
10
15
20
25
30
40
55
75
100
150
250
300
400
24h accumulated precipitation, 12 UTC 2 Nov.2008, AROME forecast and observations
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D1
The AROME model
◮ 2.5 km horizontal grid spacing
◮ 41 vertical levels
◮ 3D-VAR data assimilation scheme
◮ Bulk microphysics parameterization,6 prognostic water variables: watervapour, cloud water, rainwater,primary ice, graupel and snow(Pinty and Jabouille, 1998, Caniaux,1994)
WMED
FRANCE
NWMED
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D2
AROME forecasts and uncertainty12 UTC06 UTC00 UTC 18 UTC 00 UTC 06 UTC 12 UTC
ALADIN
AROME
(10 km)
(2.5 km)
ALADIN & AROME data assimilation cycle
24-h forecasts
LBCs ICs
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D3
AROME forecasts and uncertainty12 UTC06 UTC00 UTC 18 UTC 00 UTC 06 UTC 12 UTC
ALADIN
AROME
(10 km)
(2.5 km)
ALADIN & AROME data assimilation cycle
24-h forecasts
LBCs ICs
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D3
AROME forecasts and uncertainty12 UTC06 UTC00 UTC 18 UTC 00 UTC 06 UTC 12 UTC
ALADIN
AROME
(10 km)
(2.5 km)
ALADIN & AROME data assimilation cycle
24-h forecasts
LBCs ICs
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D3
The Ensemble experiments
AROME-PEARP◮ Each AROME assimilation cycle uses LBCs from one
PEARP (global, short range ensemble) member
AROME-PERTOBS◮ Unique LBCs from the deterministic large scale forecast◮ Each AROME assimilation cycle uses randomly perturbed
observations
◮ The AROME-PEARP ensemble samples the uncertaintyon synoptic-scale LBCs and initial conditions
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D4
The Ensemble experiments
AROME-PEARP◮ Each AROME assimilation cycle uses LBCs from one
PEARP (global, short range ensemble) member
AROME-PERTOBS◮ Unique LBCs from the deterministic large scale forecast◮ Each AROME assimilation cycle uses randomly perturbed
observations
◮ The ensemble data assimilation technique is known tosample the analysis error quite well (Berre et al., 2006)
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D4
The Ensemble experiments
AROME-PEARP◮ Each AROME assimilation cycle uses LBCs from one
PEARP (global, short range ensemble) member
AROME-PERTOBS◮ Unique LBCs from the deterministic large scale forecast◮ Each AROME assimilation cycle uses randomly perturbed
observations
AROME-COMB◮ LBCs from one PEARP member◮ Assimilation of randomly perturbed observations
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D4
Evaluation periods
31 (18) days◮ 6 October 2008 -> 5 November 2008 (31 days)◮ 15 October 2008 -> 1 November 2008 (18 days)
◮ 20 October 2008 ◮ 1-2 November 2008
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D5
Example on a case study: 1-2 Nov. 2008
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
AROME-COMB
24h accumulatedprecipitation,12 UTC 2 Nov. 2008
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D6
Precipitation: ROC and reliability diagram
Relative Operating Characteristics◮ Probability Of Detection against False Alarm Rate◮ The upper the curve is, the better the resolution of the
ensemble is
Reliability diagram◮ Observed frequency
againstforecast probability
0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8 1
PO
D
FAR
AROME-PEARPAROME-PERTOBS
AROME-COMB
RR24≥ 0.5 mm
0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8 1
PO
D
FAR
AROME-PEARPAROME-PERTOBS
AROME-COMB
RR24≥ 10 mm
0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8 1
Obs
erve
d fr
eque
ncy
Forecast probability
AROME-PEARPAROME-PERTOBS
AROME-COMB
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D7
Precipitation: Ensemble spread
0
2
4
6
8
10
12
14
0 2 4 6 8 10 12 14
RM
SE
Spread
AROME-PEARPAROME-PERTOBS
AROME-COMB
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D8
Ensemble spread for 925hPa wind speed
Rank histograms: a U-shaped histogram shows ensemble underdispersion
0
200000
400000
600000
800000
1e+06
1.2e+06
1.4e+06
1.6e+06
0 2 4 6 8 10 12
AROME-PEARPAROME-PERTOBS
AROME-COMB
3h
0
0.5
1
1.5
2
2.5
3
3.5
4
0 0.5 1 1.5 2 2.5 3 3.5 4
RM
SE
Spread
AROME-PEARPAROME-PERTOBS
AROME-COMB
0
200000
400000
600000
800000
1e+06
1.2e+06
1.4e+06
1.6e+06
0 2 4 6 8 10 12
AROME-PEARPAROME-PERTOBS
AROME-COMB
6h
0
0.5
1
1.5
2
2.5
3
3.5
4
0 0.5 1 1.5 2 2.5 3 3.5 4
RM
SE
Spread
AROME-PEARPAROME-PERTOBS
AROME-COMB
0
200000
400000
600000
800000
1e+06
1.2e+06
1.4e+06
1.6e+06
0 2 4 6 8 10 12
AROME-PEARPAROME-PERTOBS
AROME-COMB
12h
0
0.5
1
1.5
2
2.5
3
3.5
4
0 0.5 1 1.5 2 2.5 3 3.5 4
RM
SE
Spread
AROME-PEARPAROME-PERTOBS
AROME-COMB
0
200000
400000
600000
800000
1e+06
1.2e+06
1.4e+06
1.6e+06
0 2 4 6 8 10 12
AROME-PEARPAROME-PERTOBS
AROME-COMB
24h
0
0.5
1
1.5
2
2.5
3
3.5
4
0 0.5 1 1.5 2 2.5 3 3.5 4
RM
SE
Spread
AROME-PEARPAROME-PERTOBS
AROME-COMB
RMSE vs. Ensemble spread
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D9
Conclusions
Impact of uncertainty on LBCs and ICs◮ ICs: short forecast ranges◮ LBCs: grows with lead time, rapidly overcomes the
uncertainty on ICs◮ ICs and LBCs: depends on the atmospheric state
Ensemble evaluation◮ Promising precipitation scores◮ Underdispersive ensembles, especially for low-level
parameters to which the HPEs are very sensitive.
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D10
Conclusions
Impact of uncertainty on LBCs and ICs◮ ICs: short forecast ranges◮ LBCs: grows with lead time, rapidly overcomes the
uncertainty on ICs◮ ICs and LBCs: depends on the atmospheric state
Ensemble evaluation◮ Promising precipitation scores◮ Underdispersive ensembles, especially for low-level
parameters to which the HPEs are very sensitive.
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D10
Prospects
LBCs◮ How to select a few relevant forecasts from a global EPS?
ICs: perturbed observations method◮ What happens where few observations are available?
What else?◮ Model errors have to be investigated.◮ Other ensemble generation techniques (ETKF...)◮ Huge computing time and data volumes!◮ HyMeX SOP (Sept-Oct 2012, NW Med. Target Area):
a testbed for our cloud resolving EPS
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D11
Prospects
LBCs◮ How to select a few relevant forecasts from a global EPS?
ICs: perturbed observations method◮ What happens where few observations are available?
What else?◮ Model errors have to be investigated.◮ Other ensemble generation techniques (ETKF...)◮ Huge computing time and data volumes!◮ HyMeX SOP (Sept-Oct 2012, NW Med. Target Area):
a testbed for our cloud resolving EPS
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D11
Prospects
LBCs◮ How to select a few relevant forecasts from a global EPS?
ICs: perturbed observations method◮ What happens where few observations are available?
What else?◮ Model errors have to be investigated.◮ Other ensemble generation techniques (ETKF...)◮ Huge computing time and data volumes!◮ HyMeX SOP (Sept-Oct 2012, NW Med. Target Area):
a testbed for our cloud resolving EPS
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D11
- The END -
The AROME-PEARP experiment
(x11) (x11) (x11)
12 UTC06 UTC00 UTC 18 UTC 00 UTC 06 UTC 12 UTC
PEARP
ALADIN
AROME
(23 km)
(10 km)
(2.5 km)
(x11)
(x11)
AROME & ALADIN Data Assimilation Cycle
24-h forecasts
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D13
The AROME-PERTOBS experiment
12 UTC06 UTC00 UTC 18 UTC 00 UTC 06 UTC 12 UTC
ALADIN
AROME
(10 km)
(2.5 km)
AROME(2.5 km)
&
(x10)
ALADIN & AROME data assimilation cycle
24-h forecasts
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D14
Example on a case study: 1-2 Nov. 2008
AROME-PEARP AROME-PERTOBS AROME-COMB
h03 h06 h09 h12 h15 h18 h21 h24
12
34
56
7 P0P1P2P3P4P5P6P7P8P9P10Obs.
Forecast range (h)
Pre
cipi
tatio
n m
ean
(mm
)
FREQ
5
10
20
30
40
50
60
70
80
90
100
RR24≥ 50 mm probability
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D15
Example on a case study: 1-2 Nov. 2008
AROME-PEARP AROME-PERTOBS AROME-COMB
h03 h06 h09 h12 h15 h18 h21 h24
12
34
56
7 P0P1P2P3P4P5P6P7P8P9P10Obs.
Forecast range (h)
Pre
cipi
tatio
n m
ean
(mm
)
FREQ
5
10
20
30
40
50
60
70
80
90
100
RR24≥ 50 mm probability
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D15
Example on a case study: 1-2 Nov. 2008
AROME-PEARP AROME-PERTOBS AROME-COMB
h03 h06 h09 h12 h15 h18 h21 h24
12
34
56
7 P0P1P2P3P4P5P6P7P8P9P10Obs.
Forecast range (h)
Pre
cipi
tatio
n m
ean
(mm
)
FREQ
5
10
20
30
40
50
60
70
80
90
100
RR24≥ 50 mm probability
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D15
Example on a case study: 1-2 Nov. 2008
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
AROME-PEARP
24h accumulatedprecipitation,12 UTC 2 Nov. 2008
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D16
Example on a case study: 1-2 Nov. 2008
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
AROME-PERTOBS
24h accumulatedprecipitation,12 UTC 2 Nov. 2008
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D17
Example on a case study: 1-2 Nov. 2008
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
AROME-COMB
24h accumulatedprecipitation,12 UTC 2 Nov. 2008
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D18
Case study: 20 Oct. 2008
AROME-PEARP AROME-PERTOBS AROME-COMB
h03 h06 h09 h12 h15 h18 h21 h24
0.0
0.5
1.0
1.5
2.0 P0
P1P2P3P4P5P6P7P8P9P10Obs.
Forecast range (h)
Pre
cipi
tatio
n m
ean
(mm
)
FREQ
5
10
20
30
40
50
60
70
80
90
100
RR24≥ 50 mm probability
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D19
Case study: 20 Oct. 2008
AROME-PEARP AROME-PERTOBS AROME-COMB
h03 h06 h09 h12 h15 h18 h21 h24
0.0
0.5
1.0
1.5
2.0 P0
P1P2P3P4P5P6P7P8P9P10Obs.
Forecast range (h)
Pre
cipi
tatio
n m
ean
(mm
)
FREQ
5
10
20
30
40
50
60
70
80
90
100
RR24≥ 50 mm probability
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D19
Case study: 20 Oct. 2008
AROME-PEARP AROME-PERTOBS AROME-COMB
h03 h06 h09 h12 h15 h18 h21 h24
0.0
0.5
1.0
1.5
2.0 P0
P1P2P3P4P5P6P7P8P9P10Obs.
Forecast range (h)
Pre
cipi
tatio
n m
ean
(mm
)
FREQ
5
10
20
30
40
50
60
70
80
90
100
RR24≥ 50 mm probability
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D19
Case study: 20 Oct. 2008
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
AROME-PEARP
24h accumulatedprecipitation,12 UTC 2 Nov. 2008
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D20
Case study: 20 Oct. 2008
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
AROME-PERTOBS
24h accumulatedprecipitation,12 UTC 2 Nov. 2008
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D21
Case study: 20 Oct. 2008
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
10
15
20
25
30
40
55
75
100
150
250
300
400
AROME-COMB
24h accumulatedprecipitation,12 UTC 2 Nov. 2008
4th HyMeX Workshop, 10 June 2010, Bologna, Italy | B V, O N, V D22