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

DATA ASSIMILATION FOR ATMOSPHERIC CHEMISTRY
address the monitoring of air quality with MOCAGE-Valentina. The MOCAGE model has the advantage of
offering the possibility to work on four nested domains, from global to local scale. Working with the regional
domain is essential for the modelling of air quality in order to have a sufficient horizontal resolution. For
computational reasons it is not possible to increase the resolution of the global domain up to scales useful
for air quality studies. The Valentina system has thus been further developed to produce chemical analyses
on the limited area covered by the regional version of MOCAGE.
The first requirement of the MACC project was to produce analyses of the concentration at the surface
of several key atmospheric components. The analyses had to use the surface measurements from groundbased
stations available each day for the day before. As Valentina is able to assimilate only one chemical
component, we selected the ozone molecule. This choice was driven by the fact that ozone is one of the most
observed species and because ozone is actively involved in air pollution. We decided to use the full chemical
scheme of MOCAGE to obtain values for all the key atmospheric components of MACC. Because even if
only ozone is assimilated, modifying the ozone concentration produces modifications in the concentration
of other species due to the chemical links between them.
The version of MOCAGE-Valentina that assimilates the surface ozone measurements was delivered to
Météo-France/CNRM in mid-2010 for a pre-operational daily production of the ozone surface analysis
of the day before. We intensively used this pre-operational production to evaluate our analyses. In particular
we have developed a monitoring suite that produces each day and each week some plots that are displayed
on a web site to easily follow the evolution of the products. This monitoring allowed us to detect that we
had to periodically update the error statistics both for the forecast and for the observation. We have then
developed tools based on a posteriori diagnostics to automatically update the forecast and the observation
error covariances matrices. The monitoring was also useful to point out the stations where the measurements
were repeatedly not coherent with the model. This has lead to the creation of a blacklist for such stations,
the blacklist being also automatically updated.
(a) West-East length-scales
(b) North-South length-scales
FIG. 3.1: Diagnosed length-scales (in km) of the horizontal correlation of the forecast error, averaged on
the period between 1 st and 10 th July 2010 : West-East (left) and North-South (right) length-scales.
Among the work to ensure a good quality of the daily analysis of the surface ozone, we had begun a thorough
investigation on the background error covariance matrix (BECM). In the operation suite, the background
error variances comes from a posteriori diagnostics. The background error correlations are fixed in time and
had never changed. We first diagnosed the BECM using an ensemble method with a methodology similar
to the one we used for the global scale (see section 3). We have found that there are strong space and time
variations in the background error covariances (Fig. 3.1), the use of constant values for the correlations as
in the pre-operational suite is therefore not a good option [4]. We then derived several configurations of the
78 Jan. 2010 – Dec. 2011