This presentation can be downloaded at: http:www'costruzioniidrauliche'ing'unibo'itpeoplealberto

1
European Geosciences Union General Assembly
2007 Vienna, 15 - 20 April 2007
Global uncertainty assessment for hydrological
model output based on the analysis of model
errors a multiple regression approach
Alberto Montanari Faculty of Engineering Universi
ty of Bologna alberto.montanari_at_unibo.it http//ww
w.costruzioni-idrauliche.ing.unibo.it/people/alber
to
2
(No Transcript)
3
The probability model we would like to use is a
multiregression of the model error against some
selected explanatory variables. This approach can
be considered a (relevant) extension of the
Meta-Gaussian model by Montanari and Brath (WRR,
2004).
4
(No Transcript)
5
(No Transcript)
6
(No Transcript)
7
(No Transcript)
8
(No Transcript)
9
(No Transcript)
10
Samoggia River Basin, in Northern Italy. Four
years of hourly rainfall, temperature and river
flows at the basin outlet are available. A
lumped rainfall-runoff model is calibrated by
using the data recorded during the flood event of
October 8, 1996. The model was validated by
simulating all the 1994-1997 flows. Model
efficiency in validation is 0.67.
11
Selection of the explanatory variables After a
series of trials, we decided to use as
explanatory variables s(t) ? the simulated river
flow s(t) - s(t-1) ? the first differences of the
simulated river flow Results of the
multiregression Ne(t) -0.02 Ns(t) 0.29
N(s(t) - s(t-1)) e(t) sNe(t)
0.95 Ne-(t) 0.04 Ns(t) 0.17 N(s(t) -
s(t-1)) e-(t) sNe-(t) 0.91
The multiregressive approach is
satisfactorilycalibrated. It can be used to
compute theuncertainty of the model simulation.
12
(No Transcript)
13
Conclusions
The analysis of the model error may be a valuable
tool to infer the uncertainty of hydrological
models. Statistically based uncertainty
estimation requires the introduction of suitable
assumptions. It is important to check whether
they are verified. The present work aims to
identify the probability distribution of the
model error through a multivariate regression in
the Gaussian domain. These techniques are very
simple to apply and the related assumptions are
clearly understandable and easy to check.
Please visit http//www.costruzioni-idrauliche.in
g.unibo.it/people/alberto The present
presentation and softwares for the application of
these uncertainty estimation methods can be
downloaded.
14
http//www.costruzioni-idrauliche.ing.unibo.it/peo
ple/alberto
Thanks for your attention!