CERFACS CERFACS Scientific Activity Report Jan. 2010 – Dec. 2011

DATA ASSIMILATION
FIG. 4.3: April 2006 event, Marne Vallage catchment. Two-hour forecasted water level at Joinville for the
Free run (black curve), the observation (blue curve) and the analysis of the Assim run with the two-step
assimilation (red curve).
is shorter downstream of the observation point. This approach enabled a realistic modeling of the spatial
error correlations for the data assimilation algorithm with MASCARET.
It was shown that the correction of the hydraulic state is not as predictive as the upstream flow correction
and is not sufficient to constrain the simulation over an interesting forecast period. This justifies the need
for the two-step data assimilation approach. This two-step procedure was applied to the Adour and Marne
catchments (France) and the results were interpreted for several events in each catchment. The two-hour
forecasted water levels in the Marne Vallage catchment, at Joinville, for the April 2006 event are presented
in Fig. 4.3 As the Free run (no assimilation) significantly overestimates the flood peak at Joinville, the
simulated discharge was improved by 35% at Joinville with the two-step data assimilation procedure. On
the Adour catchment, it was shown that, on average, the Nash-Sutcliffe criteria is improved by 60 % for
short time scale and 35% for 12 hours forecast. This work was accepted for publication in HESS in 2011
[DA16].
Complementary studies were carried out with the DAMP, on other catchments in France ([DA32]). It
has also been noted that the 2 approaches here do not allow for all necessary corrections and that the
extension of the control space could be useful as other sources of uncertainties or model errors (for
example the simplification of the flood plain representation) result in errors in the hydraulic state. Indeed,
the correction of the Strickler coefficients or of the bathymetry will be investigated in the framework
of J. Habert’s PhD at CERFACS/SCHAPI (started Oct. 2011). Finally, theoretical issues regarding the
data assimilation algorithm, especially the dynamics of the background error covariances matrices were
conducted at CERFACS in the framework of a Master2 internship ([DA31],[DA34]), in collaboration with
CNRM (Olivier Panneckouke). This study carried out on a diffusive flood wave propagation model opens
the way for the understanding and the implementation of advanced techniques such as Ensemble Kalman
Filter that will be applied to the DAMP in the framework of a PhD (S. Barthélémy) at CERFACS, founded
by SCHAPI and the Région Midi-Pyrénées.
CERFACS ACTIVITY REPORT 87