Recall SWOT Hydro Data Assimilation Meeting Recommendations

Recall SWOT Hydro Data Assimilation
Meeting recommendations
All participants to the workshop
Meeting recommendations
• SWOT data and errors:
– Have access to realistic SWOT error to assess
impact of random and systematic errors.
– For some hydraulic model, assimilation of water
elevations in geoprojected point cloud format at
intrinsic KaRIn resolution.
• Need to prepare now long term perspectives
for hydrology (after SWOT).
• Scale issues: need to develop SWOT
down/upscalling methods for hydrodynamic
and hydrology models.
Meeting recommendations
• Open question: could DA be used to decrease
error on official floodplain and cross-section
topography SWOT product?
• Data exchange issues:
– Develop over the long term means to exchange
important amount of datasets (model codes,
input/output datasets).
– Access and exchange SWOT simulator outputs
among SDT members but also other research
teams involved in SWOT studies.
Meeting recommendations
• 2 working groups have been created:
– Hydraulic modeling (leads: J. Monnier, I. Gejadze)
– Large scale hydrology (leads: A. Getirana, S. Munier)
• Purposes:
– Promote scientific discussions among members;
– Discuss model benchmarkings, sensitivity studies
and assimilation schemes;
– Sharing tools of common interest;
– Propose roadmap of suggested studies to develop
SWOT-specific DA methods.
On going SWOT assimilation
Large/global scale hydrology
modeling and SWOT DA
Vanessa Pedinotti PhD (2009-2013)
Charlotte Emery PhD thesis (20132016)
Large/global scale: Niger River, OSSE with ISBA-TRIP
(V. Pedinotti’s PhD)
‘True’ parameter
Perturbed parameter
and orbit
Perturbed simulation
Simulateur SWOT
(Biancamaria et al., 2011)
SWOT virtual
water levels
EKF Assimilation
Corrected Manning
V. Pedinotti’s CNES/Région Midi-Pyrénées PhD grant + CNES/TOSCA
New model
Large/global scale: Niger River, OSSE with ISBA-TRIP
(V. Pedinotti’s PhD)
• Relative difference of Manning coefficient
averaged over the river vs assimilation cycles
The relative difference of WL is 30%
improved over the river.
Frequency of flooding
events classified by
intensity in the delta
over the assimilation
Relative difference of
water level averaged over
the river
DA-SWOT at the global scale
• Investigate SWOT potential for large/global scale
hydrology modeling, using SURFEX platform (ISBA-TRIP).
• How can distributed water elevations help to better
constrain the water budget estimates at regional to
global scales?
• Assimilate (test methods: EKF, EnKF?) virtual SWOT data
to correct parameters (Manning, bankfull depth) and
model outputs (water elevation).
• data downscaling issues (ISBA-TRIP spatial resolution
• Assimilation of water elevation or global discharge?
• Work done by C. EMERY’s (CNES/Region Midi-Pyrénées
PhD grant + TOSCA/CNES).
Régional hydrology modeling and
Vincent Häfliger PhD thesis (20122015)
DA-SWOT with MODCOU at the medium
range scale
I- Assimilation of data to modify prognostic variables
• Adaptation and comparison with the in-situ discharge assimilation
system developed by Thirel et al. (2010) with Extended Kalman
II- Assimilation of data to correct MODCOU parameters
• Mainly friction coefficients, to be coordinated with other works
Framework: V. Häfliger’s PhD (CNRM) + CNES/TOSCA propal on the
Garonne catchment.
DA-SWOT with MODCOU at the medium
range scale
• Issues to be treated :
- Adapt the system to treat spatialized data (SWOT/AirSWOT) instead
of local data (river gauges). Choice of the DA method ?
- Consider modifying other variables than the soil wetness (river
water, aquifer exchanges, snow cover, upstream dam releases,
- Scaling issues (average of SWOT data, consider anomalies vs
absolute data ?)
- Impact of the routing model (Muskingum, MCT, Manning)
comparison with hydrodynamic models. Consequences for
ISBA/TRIP [Manning]
Hydraulic modeling and SWOT DA
SWOT SDT – Toulouse, 17th of June 2014
- 4D-var for 2D SWE : DassFlow software – Garonne test case
- 1D effective river models
- Combination of multi-dimensional and hierarchical models and their inversions
Studies in progress by P.-A. Garambois (LEGOS, IMT, IMFT),
J. Monnier (IMT),
H. Roux, D. Dartus (IMFT)
S. Biancamaria, S. Calmant (LEGOS)
DassFlow software : sensitivity maps, identification – calibration, data
assimilation (adjoint, 4d-var) for 2D SWE
 Forward models: 2D SWE (streambeds, flood plains with wet/dry fronts).
Finite volume schemes 1st order / 2nd order, accurate-stable for flood plain dynamics.
 Adjoint code automatically generated (source-to-source, Tapenade INRIA + home-made scripts).
 MPI codes (direct + adjoint). Scotch library (U. Bordeaux).
 Interfaced with standard pre & post-processors + Telemac 2D input data
In progress : multi-dimensional observation operator SWOT like
Sensitivities Maps : a flood plain example (Lèze river)
elevation time
series at the 2
stations (with ~
% noise)
Sensitivity wrt
friction coefficient
Sensitivity wrt
Lèze River, Toulouse
In progress
These sensitivity maps have not really been exploited in applications yet.
They may be useful to the modeler – expert to understand better the topography-friction
uncertainties and correlation ; also the representation scales required.
Rich information before performing « blind » assimilation / fitting process.
Present example : Identification of friction coefficient and/or inflow discharge can be performed accurately (given
the 2 times series of h): the numerical data assimilation process does its job well
Effective representation of river sections/reaches.
Braided river case : Xingu river with altimetric data
Case of single multi-thread
sections, backwater curve
[Garambois-Monnier] submitted
[Garambois-Calmant-MonnierBiancamaria] under progress
Effective representation of sections (here: braided and fitted on ENVISAT data)
→ flow line curvature change between low and high flows (SWOT has to detect it !)
Towards improved control sections detection (cf. O'Loughlin et al 2014, Congo river)
From 1D to floodplain dynamics, Garonne River.
Direct modeling, sensitivities and variational data assimilation
Garonne (80km) downstream of Toulouse, (proposed study zone for AirSWOT)
160 m3/s,
Euler 1st orer
Qin~2000 m3/s,
[Garambois-Monnier-Roux-Chorda-Dartus] study in progress
Spatially distributed sensitivities to (bathymetry, roughness) help to define finely the
Toy tests: Assimilation (adjoint method) of synthetic data by combining 1D effective river
model and 2D SW model in progress
Assimilation of simulated SWOT data : next step with S. Biancamaria et al.
Summary – On-going studies
Reconstruction of effective river models for :
- 1D flows – SWOT like data (< 15% discharge error, Garambois-Monnier submitted)
- Braided rivers: same approach in progress (Garambois-Calmant-Monnier-Biancamaria)
- Flood plains – SWOT like data, Garonne river in progress
- Hierarchical 1D models, adequate with the observation scale; least-square inversions.
- 2D SWE with 4d-var sensitivity maps / optimization
(DassFlow, low-water, flood plain dynamics)
- Combination of these hierarchical models / inversion methods
Inland Niger Delta
(images Landsat (NASA/USGS).)
Garonne (DassFlow 2D)
2D hydrodynamic modeling
Garonne river and SWOT DA
Nabil El Mocayd PhD thesis (20132016)
Using SWOT Data Assimilation to correct bathymetry and
roughness parameters (Thesis – CNES -EDF- CNRS)
Using DA algorithm wih SWOT products to correct
Bathymetry and roughness' parameter.
Uncertainty Quantification on rougness parameter and
bathymetry with MASCARET (TELEMAC). (which DA
algorithm ?)
Estimate Discharge and Water level by correcting Hydraulic's
Development of the TL and ADJ with TAPENADE.
Studies over Garonde and Gironde Catchments.
Uncertainty Quantification (UQ)
Random values of Ks respecting
gaussian hypothesis.
The behavior of roughness
parameter is strongly non-linear.
Asymetric PDF for WLE for
Using analytical Manning's
Using PC for UQ over Manning's
UQ with Manning's equation
UQ with Polynomial Chaos
1D hydraulic modeling and SWOT
Irstea-CLS PhD (2014-2017)
• DA of SWOT data for reconstruction of
discharges, friction and bathymetry
• Full Saint Venant 1D hydrodyn model + 4D-var.
• Adjoint generated with Tapenade.
Irstea-CLS PhD (2014-2017)
• 4D-Var
• Example of Data Assimilation
for unknown tributary
inflows, using spline
interpolation with limited
• Possible with Bathymetry,
Boundary conditions,
Friction, Cross Device
Reservoir operation and SWOT
Funded by NASA/JPL
DA-SWOT data for operational water
resources management
Selingue dam used (namely) to maintain
environmental minimum streamflows in the
Niger Inner Delta
Upper Niger River Basin
meteorological forcings
water levels
Munier et al. (2014)
JPL, U. Washington
DA-SWOT data for operational water
resources management
Munier et al. (2014)
JPL, U. Washington
Large-scale data assimilation
Funded by NASA/JPL
• Data assimilation of SWOT observations to primarily
estimate river discharge
• Very large area ~700,000 km2
• Test-bed for
assimilation algorithms
• Use hydraulic
geometry in
constrained EnKF
• Uses simulator output
• Assess sensitivity to model and observation errors
• Examine scaling behavior of estimates
Impact of assimilation on
Funded by NASA/JPL
• Evaluate ability of SWOT observables to reduce
forecast errors
• Study area of Ohio River basin
Obs: WSE
Obs: Width
Longest forecast lead time when
impact Is still positive

similar documents