Results
Preliminary Results of Data Assimilation Within Polyphemus
For detailed presentation, please click
here. We make brief resume as follows
- Thanks to object-oriented techniques, the developments of three
components of the data assimilation system, namely data (obsevation), model
(physics), and assimilation algorithms, are independent from one another.
- Classical sequential assimilation methods, such as OI, EnKF, RRSQRT, are
implemented. Adjoint model (obtained by automatic differentiation tool Odyssee)
for 4D-Var is ready.
- Premiminary results are obtained by perturbing model field data, such as
surface emissions, boundary conditions, etc.
- Still to come:
- Serious research on (model and background) error modeling for air quality
model;
- Investigations and implementations of new algorithms aiming at the
treatment of both nonlinearity and high dimension of the system, for
example particle
filter and maximum entropy methods.
First Results on Assimilation of Satellite Data
- A feasibility study has been led to assess the potential of IASI 0-6km
ozone column (level 2 data expected during 2007) for improving air quality
forecast at European scale.
- Conclusions:
- the 0-6km ozone columns mainly represents free troposphere ozone. Boundary
layer ozone contributes to this measure by a variable amount, depending
on the height of the boundary layer. This amount is small (mean 14%)
yet significant.
- most important, boundary layer ozone is sensitive to changes of concentrations
in the free troposphere (at least, as modelled by Polyphémus): numerical
experiments, where free troposphere ozone is perturbated, show a sensitivity
of about 25%, the maximum impact on boundary layer occurring about one
day after the perturbation. This allows to conclude that IASI 0-6km ozone
column can be used for controlling (by means of data assimilation technique)
a regional CTM.
- Data assimilation experiment: to confort these conclusions, assimilation
of 0-6km ozone columns has been performed, based on synthesized IASI data
(the real data being not available yet). The simulation of IASI data builds
on:
- Atmosphere model: Polyphemus concentration (July 2001 reference) from
0 to 5km high, MOZART concentrations above, ECMWF meteorological fields.
- Radiative transfer model: we have use the LBLRTM model from AER.
- IASI instrument model (thanks to C. Clerbaux, Service
d'aéronomie, IPSL).
- SA-NN inversion code (IPSL): a version of this code will be used by
EumetSat to provide Level2 data.
- Examples of simulated IASI data: raw radiance spectrum (left), 0-6km
(right): mean error compared to reference: 27%, instead of expected 20%
accuracy of the 0-6km column.
-
Simulated data have been assimilated using Polyphemus Optimal Interpolation
driver. The data have been assimilated into a perturbated model and compared
to the reference. The error of the forecast is 13% without assimilation,
9% with assimilation. Below an example of assimilation: mean ozone profile
at ground level: red curve represents the reference, black curve the
perturbated model (increase of NOx emission by 30%, of O3 boundary conditions
by
15%, decrease of O3 deposition by 15%), green curve the analysis.
- Current research: performance in case of missing observations (clouds),
impact of inversion averaging kernels, use of 4DVAR assimilation.