Towards an Hydrological Qualification of the Simulated Rainfall in Mountainous Areas Eddy Yates, Sandrine Anquetin, Jean-Dominique Creutin Laboratoire.

Towards an Hydrological Qualification of the Simulated Rainfall

in Mountainous Areas

Eddy Yates, Sandrine Anquetin, Jean-Dominique Creutin

Laboratoire d’étude des Transferts en Hydrologie et Environnement, Grenoble, France

Cévennes-Vivarais : a region prone to flash floods

• Objective : forecast of these flash floods.

• We focus here on the precipitation forecast.

Introduction Method Results Conclusions

• Watersheds– 100 to 1000 km2

– specific outflows of up to 5 m3s-1km-2

• Storms– 300-400 mm in 6-12 h. over some

100s km2

HYDRAM Water depth seen by the Nîmes radar (Météo-France)October 6, 2001 Vidourle, October 6 – 7, 2001

Q~ 100 Q mean

300 mm9 h

Hilly region between the Mediterranean sea and the Massif Central. Rainy autumns.

Precipitation forecast model

We use Meso-NH (Météo-France, CNRS) : a meso-scale non-hydrostatic model a nested configuration. The finest grid has a

2.5 km resolution which allows an explicit resolution of the convection

Introduction Method Results Conclusions

Reference observed rain fields

We use kriging : an exact interpolator it takes into account the statistical structure

of the rain-gauge data it gives an estimation of the reliability of the

interpolation (estimation variance)

Simulation and observation are observed for 1h and 11h cumulated rainfall.

Introduction Method Results Conclusions

Cases studied

Two simulations with very different qualities.

The point is : “how much better” is the better simulation ? is it better for hydrological purposes too ?

Introduction Method Results Conclusions

1995 : Gardon d’Anduze 2001 : Vidourle

Bad localisation Not enough precipitation simulated (maximum cumulated rainfall

of 160 mm vs. 260 mm)

Observations Simulation

2001

Observations SimulationQuite a good localisation

Not enough precipitation simulated (maximum cumulated rainfall of 100 mm vs. 170 mm)

1995

MethodIntroduction Method Results Conclusions

MethodIntroduction Method Results Conclusions

R²(area)

R²(area)

Observation

Forecast

MethodIntroduction Method Results Conclusions

0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1

10 100 1000 10000

Area (km²)

R²

0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1

10 100 1000 10000

Area (km²)

R²

estimation error limit

point to point correlation limit

0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1

10 100 1000 10000

Area (km²)

R²

Evolution of the correlation with the area

Introduction Method Results Conclusions

0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1

10 100 1000 10000

Area (km²)

R²

0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1

10 100 1000 10000

Area (km²)

R²

1995 2001

0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1

10 100 1000 10000

Area (km²)

R²

1995 11 h cumulated rainfall

Lower short-range accuracy for short time accumulation

1995 1 h cumulated rainfall

Limits of the methodIntroduction Method Results Conclusions

1

10

100

1000

10000

100000

10 100 1000 10000

Area (km²)

# pe

r cla

ss

1

10

100

1000

10000

100000

10 100 1000 10000

Area (km²)

# p

er c

lass

1

10

100

1000

10000

100000

10 100 1000 10000

Area (km²)

# p

er c

lass

2001 1995

Conclusions, perspectives

• The method can discriminate good forecasts from very bad forecasts

• We need other cases to test the method• The method must be tested with distributed data

too (radars)• Next step : use of TOPODYN (LTHE), a

hydrologic model from the TOPMODEL family. It considers several scales of the watersheds.

Introduction Method Results Conclusions

Thank you