The uncertainty in the prediction of flash floods in the Northern Mediterranean environment; single site

approach and multi-catchment system approach

CIMA - Centro di Ricerca Interuniversitario in Monitoraggio Ambientale, Università di Genova, Università della Basilicata

R. Rudari

Thanks to: L. Ferraris, F. Siccardi, G. Boni, F. Giannoni

Civil Protection Authorities have to take decisions based on probability

assessment of the possible land effects

Scenarios and Civil Protection plans

Warnings: from 24 to 12 hours before the possible events

Scales of the basins of interest

La Barbera, P.; Lanza, L. G. 2001 On the cumulative area distribution of natural drainage basins along a coastal boundaryWater Resour. Res. Vol. 37 , No. 5 , p. 1503

Land effects

22/09/1992 - 10.45 22/09/1992 - 15.30

22/09/1992 - 15.40 22/09/1992 - 15.45

DETERMINISTIC Single Site

0

500

1000

1500

2000

2500

3000

1-ot

t-96

3-ot

t-96

5-ot

t-96

7-ot

t-96

9-ot

t-96

11-o

tt-96

13-o

tt-96

15-o

tt-96

17-o

tt-96

19-o

tt-96

21-o

tt-96

23-o

tt-96

25-o

tt-96

27-o

tt-96

29-o

tt-96

31-o

tt-96

2-no

v-96

Q (m

^3/s)

Forecast Hydrograph

Cumulative Distribution Function of peaks discharge in Tanaro at Belbo for 5 LAM-TEPS

0 100 200 300 400 500 600 700 800 900 1000

Discharge [m3s-1]

Observed from rainguages

Estimated from the LAM-TEPS 1

P[Q

>q* ]

.5

.2

.1

.05

.01

.005

.3

.4

PROBABILISTIC

Downscaling

CDF of peak discharges

GCM LAMEPS LEPS

DHM

Cumulative Distribution Function of peaks discharge in Tanaro at Montecastello for 5 LAM-TEPS

2000 3000 4000 5000 6000 7000 8000 9000 10000 11000 12000

Discharge [m3s-1]

Hydrometric hydrographEstimated from the

CTRL=RM1

P[Q

>q

* ]

.5

.2

.1

.05

.01

.005

.3

.4

.8

Example of single site

PLAM-TEPS1= 0.51 PLAM-TEPS2= 0.17PLAM-TEPS3= 0.12PLAM-TEPS4= 0.12PLAM-TEPS5= 0.08

PLAM-TEPS1= 0.51 PLAM-TEPS2= 0.17PLAM-TEPS3= 0.12PLAM-TEPS4= 0.12PLAM-TEPS5= 0.08

Conditioned CDFConditioned CDF

CDFCDF

1994 Piemonte flood1994 Piemonte flood

Ferraris, L., Rudari, R. and F. Siccardi, 2002 The uncertainty in the prediction of flash floods in the northern Mediterranean environment Journal of Hydrometeorology (AMS) in press.

Admission of ignorance

• Aggregate the rainfall volume to the scale where we have a reliable evaluation of its probability of occurrence on that area through the use of EPS.

• We downscale from the aggregate scale down to the scale of interest with a purely stochastic model

Single site & multi-catchment approach• The limit among single site and multi-catchment system approaches

is in the scale at which EPS can give a reliable estimation of the probability of the rainfall volume: The bigger the aggregation area, the smaller the stochastic ensemble, the bigger the catchment where I can apply the single site warning

n=1…N

Area Dimension Stochastic ensemble size Drainage Area Distribution

Different approach – multi-catchment distributed warning

Class A (< 10 km2)Class B (10 < Area < 150 km2)Class C (> 150 km2)

PROBABILISTIC Multi-catchment approach

Downscaling

EPS LEPS

DHM

Exceeding probability

curves

STATISTICALLY HOMOGENEOUS REGION Peak discharge values of the Liguria regionbelong to the same statistical distribution

Within a regional procedure it is possible to make dimensionless all the simulated discharge peak values (TCEV)

Bisagno at la Presa - 97 km2

-2

-1

0

1

2

3

4

5

6

7

0 100 200 300 400 500 600Discharge[m3s-1]

u [

-]

T=2.9

T=10

T=50

T=100

T=200

T=500

T=5

T=30

Local probability curve obtained by the expected

time series value

DRiFt Model

Historical data

Local probability obtained by the index discharge value (DRiFt T=2.9)

Through the Qindex is possible to specify the growth curve in each site of the region

Regional framework

0

0.2

0.4

0.6

0.8

1

0 10 20 30 40 50 60 70 80 90 100

Exc

eed

ence

Pro

bab

ilit

y

Q QQQQQQQQQ

Single site

Exceeding probability of a given discharge quantile in one catchment chosen randomly inside the multi-basin system

F1 F1

0

0.2

0.4

0.6

0.8

1

0 10 20 30 40 50 60 70 80 90 100

Ex

ce

ed

en

ce

Pro

ba

bili

ty

Q QQQQQQQQQ

Single site

Regional (indepedent

basins)

If catchments and simulations were independent it would be possible to compute exceedence probability of a given quantile

in at least one site of the multi-basin system directly from probability theory

NF1NF1

0

0.2

0.4

0.6

0.8

1

0 10 20 30 40 50 60 70 80 90 100

Exc

eed

ence

Pro

bab

ilit

y

Q QQQQQQQQQ

Single site

Regional

Regional (indepedent

basins)

Discharge realizations are not independent. Each disaggregation history is considered separately

Questions to be answered by:The EPS community:

• What is the scale where I have a reliable estimation of rainfall volumes with their probability?• How does it change when a different target region is concerned?• How does it change when a different synoptic condition is concerned?• What do I gain from LAMs?• What do I loose by clustering?• What are we calling probability?

The Downscaling community:• What is the best downscaling model (stochastic, physical, mixed approach)?• Is it possible to exploit better the Meteorological model info at different scales (Data fusion)?

The Hydrologic community:• How do I improve the I. C. of the hydrological models (remote measures, data assimilation)?

All the communities:• How do the uncertainty propagates through the chain?• How do we convey the Probabilistic Information?

Hydrologic scale inconsistency

0

700

20/06/96 18.00 21/06/96 6.00 21/06/96 18.00 22/06/96 6.00

Time

Dis

ch

arg

e [

m3 s-1

]

Recorded

5 MIN

15 MIN

60 MIN

180 MIN

Pebble creek June 20 1996

0

700

20/06/96 18.00 21/06/96 6.00 21/06/96 18.00 22/06/96 6.00

Time

Dis

ch

arg

e [

m3 s-1

]

Recorded

1 X 1 km

2 X 2 km

4 X 4 km

8 X 8 km

16 X 16 km

Pebble creek June 20 1996

Pebble creek

Space Time

Giannoni, 2001