A perspective on “Radar Hydrology”

based on space-time scales

Guy Delrieu*

LTHE, Grenoble, France

*… and colleagues + students at LTHE and elsewhere

Outline

• Hydrometeorological risk prediction in the

Mediterranean region

• Radar QPE: error models• Radar QPE: error models

• Hydrological observation and modelling

Outline

• Hydrometeorological risk prediction in the

Mediterranean region

• Radar QPE: error models• Radar QPE: error models

• Hydrological observation and modelling

Heavy rain events in France : location and occurrence Météo France 2001 (CD-rom Pluies extrêmes dans le sud de la France)

Number of days withdaily rain amounts > 200 mm between 1958 and 2000 in Southern France Occurrence :

2-3 events > 200 mm day-1 per year

1 major event every 4-5 years :

Nîmes october 1988Vaison la Romaine september 1992Aude november 1999Gard september 2002Rhône décember 2003Var june 2010….

A variety of convective systems

• Long lasting events

• Shallow / banded convection

• Orographic forcing

2-7 November 2011

• V-shaped Mesoscale

Convective Systems

• Deep convection

• Stationarity, dynamic forcing

(cold pool)

8-9 September 2002

Geomorphological features

of the Cévennes watersheds

Primary Era

formations

Sedimentary

formations

Po

nt d

'Arc

Ru

om

sP

rad

ons

Po

nt d

e B

ala

zuc

Vog

uë

Po

nt d

e S

t D

idie

r\A

ub

en

asP

on

t d' U

cel-

Au

ben

asP

on

t de

Val

sP

on

t de

Lab

eau

me

Bar

nasM

ayre

sS

ou

rce

de

l’Ard

èch

e

0

200

400

600

800

1000

1200

1400

1600

020406080100120140

Distance auRhône (km)

L'ArdècheLe ChassezacLa Beaume

Altitude (m NGF)

Po

nt d

e S

alav

as-V

allo

n P

on

t d'A

rc

St-

Mar

tin d

’Ard

èch

e

Gorges

steep karst gorges

slopes

10000,0

100000,0

Res

pons

e tim

e (m

in)

Hydrologic response time of Mediterranean watershed s

Marseilles urban watershedsHigh-mountain Italian basinsMedium-mountain Cévennes watershedsRhône and big tributaries

Hydrological response: time to peak

Frontal System

Hydrologic response time of Mediterranean watershedsas a function of watershed surface

2-6 days

« Scale resonance »

1,0

10,0

100,0

1000,0

1 10 100 1000 10000 100000

Res

pons

e tim

e (m

in

Watershed surface (km² )

Thunderstorm

MesoscaleConvectiveSystem

10-30 min

1-2 hr

10 km² 100 km² 10000 km²

Severity of the hydrological response

100,0

Max

imum

spe

cific

dis

char

ge

(m3s

-1km

-2)

Rodier & Roche (1984)Pardé (1961)Costa (1987)Gard 2002Aude 1999

Maximum specific discharge as a function of the watershed surface

11 casualties

in small watersheds

< 20 km²

1,0

10,0

10 100 1000 10000

Watershed surface (km2)

Max

imum

spe

cific

dis

char

ge

(m3s

-1km

-2)

2002 Gard case

12 casualties

in medium-size watersheds

500-2000 km²

Ruin et al. (2009)

We should be able to predict HPEs and the associated flash-floods

and floods over a wide range of spatial scales (local to regional)

and to communicate the warnings and alerts with appropriate

lead times to the end-users.lead times to the end-users.

« Prediction for Ungauged Basins »

« Change of scale problem »

Outline

• Hydrometeorological risk prediction in the

Mediterranean region

• Radar error models• Radar error models

• Hydrological observation and modelling

Required space-time resolution of rainfall measurements

Time to peak Rainfall decorrelation distance

Watershed

surface (km²)

Time

resolution

Spatial

resolution

1 1 min 1.6 km

10 6 min 3,7 km

100 35 min 6,4 km

1000 3,3 hrs 15 km

10000 18 hrs 36 km

Lebel et al. (1987); Berne et al., (2004)

Rainfall observation system of the OHM-CV

32000 km²; 0-1800 m asl

1 hour : 250 raingauges

24 hours : 400 raingauges

2 S-band radars

2 C-band radars

2 disdrometers

Sembadel

Actual

spatial

resolution

Actual

time

resolution

Effective

time

resolution

Watershed

scale

properly

sampled

11 km 1 h 2,6 h 735 km² 2 disdrometers

11 meteo-GPS stations

Nîmes

Montclar

BollèneWatershed

surface (km²)

Time

resolution

Spatial

resolution

1 1 min 1.6 km

10 6 min 3,7 km

100 35 min 6,4 km

1000 3,3 hrs 15 km

10000 18 hrs 36 km

Error model: reference rainfall

derived from raingauge data

True unknown rainfall

Radar QPE

dvudvuRTA

txR

TA

AT ∫∫=,

),(11

),(

N

Reference rainfall:

Ordinary Kriging of raingauge values

Linear estimator:

Estimation variance:

∑=

=AN

iiT

AAT taR

NtxR

1

** ),(1

),(

))²,(),(( txRtxRE ATrefAT −

),(),(

1

txGtxR i

N

ii

refAT

g

∑=

= λ

Assessment of the OHM-CV rainfall re-analyses

Ra

da

r (m

m)

R

ad

ar

(mm

)

• Our best radar data processing:

4-radar mosaic/ GC / screening

and VPR…)

• Optimization of a single Z-R

relationship /event

Reference (mm) Reference (mm)

Ra

da

r (m

m)

R

ad

ar

(mm

)Conditional analyses

of the radar – reference rainfall

using the GAMLSS framework

(Rigby & Stanisopoulos 2004)

Kirstetter et al. (2010); Delrieu et al. (2012)

Conditional mean and standard deviation of the errors

1 h ; 5 km² 12 h; 100 km²

Toward probabilistic radar QPEs

010

2030

4050

E n s e m b l e P Q P E ( 0 8 / 0 9 / 2 0 0 2 ) M e s h 1 0

Ens

embl

e P

QP

E

R a d a r R a i n f a l lE n s e m b l e Q P E s i m 5 0R e f e r e n c e R a i n f a l l

100 km², 1 hour

Germann et al. (2009); Partheni and Delrieu, this conf.

0 5 1 0 1 5 2 0 2 5 3 0T i m e s t e p

0 5 1 0 1 5 2 0 2 5 3 0

010

2030

4050

C o n d i t io n e d E n s e m b le P Q P E G A M L S S ( 0 8 / 0 9 / 2 0 0 2 ) M e s h 1 0

T im e s t e p

Con

ditio

ned

Ens

embl

e P

QP

E

R a d a r R a in f a l lC o n d i t io n e d E n s e m b le Q P E s im 5 0R e fe r e n c e R a in f a l l

0 5 1 0 1 5 2 0 2 5 3 0

010

2030

4050

C o n d i t i o n e d E n s e m b l e P Q P E ( 0 8 / 0 9 / 2 0 0 2 ) M e s h 1 0

T i m e s t e p

Con

ditio

ned

Ens

embl

e P

QP

ER a d a r R a i n f a l lC o n d i t i o n e d E n s e m b l e Q P E s i m 5 0R e f e r e n c e R a i n f a l l

The radar error structure is complex

• rain-rate dependent

• space-time scales dependent

• location (range) dependent

• rain-type dependent

We are limited in our willing to provide radar error models

useful in hydrology by the reference rainfall problem

• micronets, microwave links?

• work on estimation variance of merged products through

geostatistical or Bayesian techniques; introduce some

physics

Outline

• Hydrometeorological risk prediction in the

Mediterranean region

• Radar QPE: error models• Radar QPE: error models

• Hydrological observation and modelling

0 5 1 0 1 5 2 0 2 5 3 0

010

2030

4050

C o n d i t i o n e d E n s e m b l e P Q P E ( 0 8 / 0 9 / 2 0 0 2 ) M e s h 1 0

T i m e s t e p

Con

ditio

ned

Ens

embl

e P

QP

E

R a d a r R a i n f a l lC o n d i t i o n e d E n s e m b l e Q P E s i m 5 0R e fe r e n c e R a i n f a l l

Hydrological modelling:

• Processes complexity

• Observability problem of the sub-surface

• Numerous landscape descriptors

• Spatial discretization

• processes representation (and selection?)

• Input-output uncertainties• Input-output uncertainties

• Structural and parametric uncertainties

Monitoring of flooding rivers

Unfeasibility of conventional

techniques during floods

Need to develop « non-contact »

gauging techniques

Need to improve rating curves

(hydraulic modelling + gauging

uncertainty accounting)

Le Coz et al.

Streamgauge network density

45 stream gauges in the OHM-CV

0%

10%

20%

30%

40%

PD

F

points de coupure

bassins jaugés

River-road intersections

Gauged watersheds

• Post-event surveys

• A team/network of local observers?

0%

1 10 100 1 000 10 000

Taille de bassin (km²)Watershed size (km²)

O. Vannier

“topographic similarity”

Runoff production:

sub-surface lateral flow and saturation excess overland flow

Runoff transfer:

Geomorphological unit hydrograph

« Parcimony »: 4 parameters

1) nTopmodels 2) ISBA-TOPMODEL

Hydrological models, example 1: TOPMODEL

1) nTopmodels 2) ISBA-TOPMODEL

Hydrological meshes of ~constant area Coupled land surface – hydrological model

Le Lay and Saulnier (2009) Vincendon et al. (2009)

Spatial assessment of n-TopmodelsThe 2003 flood in the Ardèche catchment

Le Lay and Saulnier (2009); Bonnifait et al. (2009)

Limitations:

• Data-calibrated

• Initialization?

Qualities:

• uniform spatial parameteri-

zation works reasonably well

• sensitivity to the rainfall spatial

distribution

“Landscape similarity”

Landscape discretization (~0.5 km²)

Based on topography and soil maps

Spatialized forcings:

• Radar-raingauge rainfall

• Evapotranspiration (based on soil and

vegetation maps)

Hydrological models, example 2: CVN

vegetation maps)

Runoff production:

Vertical transfer of water in the soil

• use Solve Richards 1D

• use h(ſ) et K(ſ) relations (Brooks & Corey)

for the various layers

Runoff transfer:

• Ponding sent directly to the river network

• Transfer with kinematic wave equation

Braud, Anquetin, Vannier et al.

PhD thesis of O. Vannier

Qualities:

• Best use on spatialized data

• Physically based water transfer

• Continuous modelling

• No data-calibration

Limitations:

• Hydrological relevance of soil maps

• some process representations are

lacking so far (sub-surface lateral flow)

Geomorphological controls: Gardonnenque plain + Gardons gorges during the Gard 2002 flood

Bonnifait et al. (2009)

SOGREAH

� Coupling hydrologic and hydraulic models

Crieulon, Bge de la Rouvière (93 km²) :Marls and non karstic limestones

Vidourle, Bge de Conqueyrac (86 km²) :Karst

0

200

400

600

800

1000

1200

1400

1600

1800

2000

0:00 3:00 6:00 9:00 12:00 15:00 18:00 21:00 0:00 3:00 6:00 9:00 12:00 15:00 18:00 21:00 0:00

Time

Dis

char

ge (

m3 /s

)

0

100

200

300

400

500

600

700

800

900

1000

rain

fall

rate

(m

m/h

)

rainfall rates CN=50 CN=70 CN=100 Measured discharges

0

200

400

600

800

1000

1200

1400

1600

1800

2000

0:00 3:00 6:00 9:00 12:00 15:00 18:00 21:00 0:00 3:00 6:00 9:00 12:00 15:00 18:00 21:00 0:00

Time

Dis

char

ge (m

3/s

)

0

100

200

300

400

500

600

700

800

900

1000

Rai

nfal

l rat

es (

mm

/h)

rainfall rates CN=50 CN=70 CN=100 Measured discharges

Geological controls:

Gaume et al. (2004)

� Coupling hydrologic and hydraulic models

� Solution?

Coming work within HyMeX/FloodScale projectsInvestigate the « missing scales »:

• Reinforce the rainfall observation system

• Study hydrological processes for a series of Cévennes landscapes

Thank you!Thank you!