Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Usefulness of radar QPE for
Mediterranean flash-flood
ensemble forecasting
B. Vincendon, V. Ducrocq, O. Nuissier, B. Vié
GAME/CNRM (Météo-France, CNRS)Toulouse, France
ERAD 2012 - Toulouse –28 June 2012
HYdrological cycle in the Mediterranean Experiment
� « Time-nested » observation stategy :
� Aim : understanding of thewater cyclewith emphases onintense events
ERAD 2012 ERAD 2012 --22--
� First HYMEX Special Observation Period (SOP) :� North-Western Mediterranean
� from 5 September to 6 November 2012
� dedicated to Heavy Precipitation Events and FF.
� Additional observing systems deployed
� Real time meteorological and hydrological forecasting � e.g.for FF : ISBA-TOPMODEL
Accumulated rainfall (radar QPE) from 3 Nov. 2011 to 8 Nov. 2011
0
100
200
300
400
500
600
700
800
(mm)
� Uncertainty affects QPF even for high resolution NWP
FF forecasting within HYMEX
Observation
Simulation driven by radar QPE
Forecast with deterministic QPF
METEOROLOGICAL FORCINGMETEOROLOGICAL FORCING
Precipitation
AROME* QPFPrecipitation
Radar QPE
ISBAISBA--TOPMODEL*TOPMODEL**Bouilloud et al, 2010; Vincendon et al, 2010
*(Seity et al, 2011)
ERAD 2012 ERAD 2012 --33--
StreamflowStreamflow
800
900
Forecast with deterministic QPF
Streamflow
Discharges from 3 Nov. 2011 @ 00UTC to 7 Nov. 2011 @ 00UTC
Ardèche riverGardons river
3 Nov. 4 Nov. 5 Nov. 6 Nov. 7 Nov. 3 Nov. 4 Nov. 5 Nov. 6 Nov. 7 Nov. 3 Nov. 4 Nov. 5 Nov. 6 Nov. 7 Nov. 3 Nov. 4 Nov. 5 Nov. 6 Nov. 7 Nov. 2011 2011 2011 2011 2011 2011 2011 2011 2011 2011 2011 2011 2011 2011 2011 2011 2011 2011 2011 2011
3 Nov. 4 Nov. 5 Nov. 6 Nov. 3 Nov. 4 Nov. 5 Nov. 6 Nov. 3 Nov. 4 Nov. 5 Nov. 6 Nov. 3 Nov. 4 Nov. 5 Nov. 6 Nov. 7 Nov. 7 Nov. 7 Nov. 7 Nov. 2011 2011 2011 2011 2011 2011 2011 2011 2011 2011 2011 2011 2011 2011 2011 2011 2011 2011 2011 2011
Assessment of high resolution QPF errors
� Climatology of differences between QPF and radar QPE
� Double penalty problem � object-oriented climatology of errors
� Objects defined according to thresholds in mm/h in the QPF and QPE fields
� PDF of amplitude and location errors
� Structure-Amplitude-Location* method
� measure of the QPF quality
QPE field
+
+
x qpe
ERAD 2012 ERAD 2012 --44--*(Wernli et al, 2008)
� no need to match objects one to one
QPF field
� S : difference in size , shape and gradientof the rainy objects
�A : normalized difference of averaged precipitation values
� L =L1 +L2
�L1 : global shift between QPE and QPF
�L2 : difference in relative positioning ofobjects in QPE and QPF
xqpf
+
SAL diagram
L component
A c
ompo
nent
Too much rain in QPF
Too small objects Too large objects Good forecast
False alarm
Convective rainfall
forecasted
Stratiformrainfallobserved
+2
ERAD 2012 ERAD 2012 --55--
A c
ompo
nent
S component
Not enough rain in QPF
and/or
too large a
gradient in QPF
and/or
too small a
gradient in QPF
Good forecast
No detection
Convectiverainfall
observed
Stratiform rainfall
forecasted
0
-20-2 +2
(Wernli et al, 2008)
Assessment of AROME QPF
L component
A c
ompo
nent
� A median value close to 0
� 1-h QPF vs 1-h QPE from Météo-France radar composite
� Days with significant rain from Sept. 2008 to Dec. 2011
ERAD 2012 ERAD 2012 --66--
A c
ompo
nent
S component
� A median value close to 0
⇒ No systematic bias
� L component
⇒ 70% of the cases with a location error < 50 km
� Valuable information in AROME QPF
SAL diagram for objects with 1-h rain over 2mm
Streamflow ensembles with ISBA-TOPMODELDeterministic AROME
Perturbation generation*
(*Vincendon et al., 2011)
PDF of amplitude and location errors
ERAD 2012 ERAD 2012 --77--
Single streamflow forecast
8 members streamflowensemble
ISBAISBA--TOPMODELTOPMODEL
30 members streamflowensemble
30 members
(*Vié et al., 2011)
8 m
em
be
rs
Real time FF forecasting chain within HYMEX SOP� Daily forecast:
Deterministic AROME
|
Day D+1
à 00utc
AROME + Perturbation
Forecasts with
ISBA-TOPMODEL driven by :
T0= Day D
à 00utc
|
T0+36 h
AROME + Perturbation : RPSS ~0.23
3 Nov. 2011 4 Nov. 2011
ERAD 2012 ERAD 2012 --88--
30 members
AROME EPS
8 members
AROME EPS: RPSS ~ 0.35
3 Nov. 2011 4 Nov. 2011
Streamflow ensembles for Gardons river : 3 nov. 2011 @ 02UTC - 4 nov. 2011 @ 04UTC
Observation
Simulation driven by radar QPE
Interquartile range of the ensemble
Ensemble median
Forecast with deterministic QPF
Conclusions and future work� Usefulness of radar QPE for FF forecasting
⇒ To document the uncertainty on QPF
⇒ Method of perturbation of QPF to take benefit from valuable information of AROME deterministic forecast
⇒ Streamflow ensemble at reduced numerical cost
� Implemented in real-time for HYMEX SOP : http://sop.hymex.org
� Improvement of streamflow ensembles :
ERAD 2012 ERAD 2012 --99--
� Combining perturbation method with AROME EPS
� Considering other sources of uncertainty :
� Hydrological modelling
� Initial soil conditions
� Benefit from HYMEX SOP to : � Intercompare FF forecasts on HYMEX cases
� Assess ensemble forecasting on a satisfactory sample
� Improve initialisation and calibration of hydrological models
Bouilloud et al, 2010 :Bouilloud L., K. Chancibault, B. Vincendon, V. Ducrocq, F. Habets, G.M. Saulnier, S. Anquetin, E. Martin, J. Noilhan , 2009: An advanced coupling between the ISBA land surface model and the TOPMODEL hydrological model to simulate Mediterranean flash-floods, J. Hydrometeor., 11(2),315-333
Vié et al, 2011 :Vié, B., Nuissier, O., Ducrocq, V., 2010: Cloud-resolving ensemble simulations of heavy precipitating events : uncertainty on initial conditions and lateral boundary conditions, Mon. Weath. Rev.,139,403-423
Vincendon et al, 2010 :Vincendon B.,Ducrocq V., Saulnier G.M.; Bouilloud L., Chancibault K., Habets F., Noilhan J., 2010:
Thank you
ERAD 2012 ERAD 2012 --1010--
Vincendon B.,Ducrocq V., Saulnier G.M.; Bouilloud L., Chancibault K., Habets F., Noilhan J., 2010: Advantages of coupling the ISBA land surface model with a TOPMODEL hydrological model dedicated to Mediterranean flash floods, J. Hydrology, 394(1-2), 256-266
Vincendon et al, 2011 :Vincendon B., Ducrocq V., Nuissier O. et Vié B. , 2011:Introducing perturbation in rainfall fields for an ensemble forecasting of flash-flood, NHESS, 11, 1529-1544
Wernli et al, 2008 :Wernli H., Paulat M., Heigen M. and Frei C., 2008 : SAL A novel quality measure for the verification of quantitative precipitation forecast , Mon. Weath. Rev., 136,4470-4487
Point-to-point assessment of QPF values
D
� A : characterizes how different are domain-averaged precipitation values
QPE field
QPF field
ERAD 2012 ERAD 2012 --1111--*(Wernli et al, 2008)
Observed D-averaged precipitation
Modelled D-averaged precipitation
QPF field
Point-to-point assessment of QPF values
� L =L1 +L2
� L1 characterizes the global shift between QPE and QPF
Barycentre of the QPE field among D
Maximal distance among D
D
+
x(Rmod)+
x(Robs)
d
Object n
nx+
Rn
ERAD 2012 ERAD 2012 --1212--
� L2 characterizes the spatial precipitation distribution inside the domain.
(Wernli et al, 2008)
Barycentre of object n
Mean rainfall on object n
Total number of objects
Point-to-point assessment of QPF values
� S: characterizes the size of the rainy objects as well as the gradient
D
Object n
ERAD 2012 ERAD 2012 --1313--(Wernli et al, 2008)
Maximal rainfall on object n
Usefullness of radar QPE for FF forecasting?
� Hydrological models dedicated to FF sensitive to rainfall (volume and spatial distribution)
⇒ Radar QPE = appropriate information
METEOROLOGICAL FORCINGMETEOROLOGICAL FORCING
Precipitation
Radar QPE
Coupled system ISBAISBA--TOPMODELTOPMODEL**
� FF simulation:
ERAD 2012 ERAD 2012 --1414--
Surface schemeISBA
Routing
Hydrological model TOPMODEL
Streamflow
Initial soil
conditions
*Bouilloud et al, 2010; Vincendon et al, 2010
�!
� Use of radar QPE for :
� hydrological simulation
� hydrological models calibration
Uncertainty on FF simulations due to QPE
• Paris
Rivers:
Ardèche at Saint Martin (2240 km²)
Cèze at Bagnols-sur-Cèze (1100 km²)
Gardons at Ners (1090km²)
ERAD 2012 ERAD 2012 --1515--
Bollène•Toulouse
Nîmes
Montpellier• •
•MontAigoual
0-4040-110110-190190-270270-370370-500500-650650-800800-980980-11801180-14301430-17301730-20502050-24002400-28302830-4760
Altitude (en m)
(C)
ME
TE
O-F
RA
NC
E /
IGN
Weather radars
Hydrological range (80 km)
Uncertainty on FF simulations due to QPE: November 2011 case
0
100
200
300
400
500
600
700
800
(mm)QPEs (mm) from 3rd nov. 2011 at 12UTC to 8 nov. 2011 at 00UTC
ERAD 2012 ERAD 2012 --1616--
900Bollène radar MosaicNîmes radar
Discharges (m3.s-1) from 3rd nov. 2011 at 12UTC to 8 nov. 2011 at 00UTC
Ardèche river Gardons river
Uncertainty on FF simulations due to QPE: November 2011 caseS
ever
al p
aram
eter
s
ERAD 2012 ERAD 2012 --1717--Discharges (m3.s-1) from 3rd nov. 2011 at 12UTC to 8 nov. 2011 at 00UTC
Ardèche river Gardons river
Sev
eral
rad
ar d
ata
Sensitivity to radar QPE time stepTemporal frequency:
1h
15 minutes
05 minutes
Catchments with an area < 200 km²
ERAD 2012 ERAD 2012 --1818--
area < 200 km²
Catchments with an area > 600 km²
Cumulated frequencies of nash efficiency (Prediflood project)
Quantification of AROME QPF error� PDF of amplitude and location errors for object Or and Oc
� Amplitude errors : no systematic bias
� Location errors : < 50 km in 70% of cases
ERAD 2012 ERAD 2012 --1919--PDF of location errors for Or objects along E/W axe(Vincendon et al., NHESS, in revision)
Quantification of AROME QPF error
�AROME hourly QPF vs Météo-France radar hourly QPE
� Significant rainy events
� Object–based climatology of AROME QPF errors :
� Rainy objects = Or
� Convective objects = Oc
Or
ERAD 2012 ERAD 2012 --2020--
1h-radar QPE the 01 Nov. 2008 at 20UTC 1h-AROME QPF the 01 Nov. 2008 at 20UTC(AROME run start = 01/11/2008 analyse)
Or
Oc
1h-radar QPE the 01 Nov. 2008 at 20UTC 1h-AROME QPF the 01 Nov. 2008 at 20UTC(AROME run start = 01/11/2008 analyse)
Quantification of AROME QPF error
ERAD 2012 ERAD 2012 --2121--
D
Perturbation generation method
PDF of location errors
(Vincendon et al., NHESS, in revision)
AROME déterministic forecast
Or
Oc
N members selected
PDF of
ERAD 2012 ERAD 2012 --2222--
D
Nnew fields
D
PDF of amplitude errors for Oc
fc
OcRainfall intensity of Or
X fPDF of amplitude errors for Or
f
Rainfall intensity of Oc X fc
Ardèche
Cèze
(m3 /
s)
(m3 /
s)
(Vincendon et al., NHESS, in revision)
21 October 2008 22 October 2008
Streamflow ensemble for 21-22 October 2008 event
ERAD 2012 ERAD 2012 --2323--
Gardons
(m3 /
s)
21 October 2008 22 October 200821 October 2008 22 October 2008
21 October 2008 22 October 2008
Ensemble median
Interquartiles range
Discharges simulated from :
50 members of the ensemble
Observed discharge
deterministic AROME
Ardèche
Cèze
(m3 /
s)
(m3 /
s)
(Vincendon et al., NHESS, in revision)
01 November 2008 02 November 2008
Streamflow ensemble for 01-02 November 2008 event
ERAD 2012 ERAD 2012 --2424--
Gardons
(m3 /
s)
01 November 2008 02 November 2008
01 November 2008 02 November 2008
01 November 2008 02 November 2008
Ensemble median
Interquartiles range
Discharges simulated from :
50 members of the ensemble
Observed discharge
deterministic AROME
Early verification
� RPSS ~ 0.35 mm/day on 24h-accumulated rainfall ⇒ Ensemble forecast > deterministic forecast
Gardons Perturbation method
(m3/s)� Comparison to a research AROME EPS (Vié et al., 2010)
⇒ Close results
ERAD 2012 ERAD 2012 --2525--
AROME EPS
�Advantage of the perturbation method:
� Reduced numerical cost
� Lot of members
How do we simulate Mediterranean FF?
d2
d3
∆xw2
w3
Runoff RDeep drainage
ISBA-3L :Management of water and energy at surface / atmosphere interface
WI,tW’
R
DR
Surface scheme resolution∆x = 1km / ∆t = 15min.
ERAD 2012 ERAD 2012 --2626--
TOPMODEL :Lateral water transfer within the catchments
ROUTING MODULE:Water transfer along slopes and
rivers
QtotCatchments resolution
δx = 50m / δt = 1h
δx
di,td’ i,t
Deep drainage DRdi,t= f( WI,t ) W’ I,t = f( d’i,t )
|
Day D-2
à 00utc
|
Day D-1
à 00utc
Radar QPE:
Spin-up period =>Simulation with
ISBA-TOPMODEL driven by :
Real time FF forecasting chain within HYMEX SOP
� Daily forecast :
Deterministic AROME
|
Day D+1
à 00utc
|
Day D+2
à 00utc
Forecasts with
ISBA-TOPMODEL driven by :
|
Day D
à 00utc
Climatology of amplitude and
ERAD 2012 ERAD 2012 --2727--
Initial soil conditions
Radar QPE:
RR1h AROME perturbated
30 members
AROME EPS
8 members
� Usefullness of radar QPE in this framework?� FF simulation � QPF assessment
amplitude and location errors