Atmospheric chemistry data assimilation
General context and motivations
The assimilation of minor atmospheric trace species is a very promising technique to obtain global and regional datasets for the monitoring and the forecasting of the atmospheric composition. The challenge lies in the optimal combination of measurements having very different resolutions in space and time, with for examples in-situ data from ground-based stations, aircrafts, low Earth orbiting and geosynchronous satellite measurements. Obtaining datasets that combine in an optimal manner all those types of data is a necessity for the validation of 3D predictive chemical models and the GMES program of the European Union and ESA.
The birth of the Mocage-Palm assimilation system
To combine all the measurement types within the models in an optimal manner using data assimilation techniques, we developed at CERFACS the data assimilation suite named Mocage-Palm (Massart et al., 2005). Mocage-Palm was developed in close collaboration with the Météo-France/CNRM, as it combines the comprehensive Météo-France three-dimensional Mocage CTM and a variational assimilation technique, through the CERFACS Palm software.
Developments were supported by the EC-FP5 ASSET European project that spanned from January 2003 to June 2006 (Lahoz et al., 2007). Our first main objective was then to analyze ozone concentrations at the global scale from the vertical profiles of the MIPAS instrument on board of the ENVISAT satellite. In this configuration, Mocage-Palm participated to the systems intercomparison part of the ASSET project. This intercomparison proved that this first version of Mocage-Palm suite provided good quality ozone fields compared with independent measurements, and that the errors were of the same order as those of several other state-of-the-art assimilation systems.
The ADOMOCA project
In 2005, the Mocage-Palm assimilation system was proposed as a common collaborative tool for the whole community involves in the ADOMOCA project of the PNCA (latter replaced by the LEFE program). We had therefore added a lot of flexibility to Mocage-Palm, so it could satisfies the different applications of the ADOMOCA project. This development work was supported by the INSU and the CNES. It made Mocage-Palm a system able to assimilate several sorts of level 2 satellite measurements: profiles and partial or total integrated column amounts, and data requiring averaging kernel information. It also allowed Mocage-Palm to assimilate simultaneously the different kinds of data. We have used these new facilities to assimilate the ozone measurements from SMR and GOMOS instruments in addition to MIPAS (Massart et al., 2005 ; Massart and Cariolle, 2006 ; Massart et al., 2007). Our development work on Mocage-Palm has also served to extend the chemical analysis to other species than ozone, the system being then able to assimilate each of the chemical species included in the chemical scheme of the CTM.
Moreover, we have developed linear chemical scheme for ozone (Cariolle and Teyssèdre, 2007) that prove to be very computationally efficient for data assimilation and that has been included in the IFS operational forecasting suite at ECMWF. The latter gives routine forecasts of the ozone evolution, and was able to predict in autumn 2007 the deepening of the ozone hole ten days in advanced. This work has been performed in cooperation with the ECMWF research department (A. Dethof, J.J. Morcrette, A. Untch). Based on the same approach a linear scheme has been developed for CO and is actually under evaluation in cooperation with modelling groups of the LA and the CNRM. The linear schemes reduce the computational cost of the assimilation process; this was a request of the ADOMOCA’s community that wants to produce analyses during long period. To save computational time, we also optimized Mocage-Palm depending on the computational platform. And finally, we added a complete set of post-treatment scripts to Mocage-Palm that generate graphical outputs. This aimed at helping users in the interpretation of the outputs of the system for each assimilation experiment.
The development work occurred by steps, each step producing a new release of the Mocage-Palm system. Each new release is provided with its documentation and is available at the website of Ether for the scientists involved in the ADOMOCA project. The Mocage-Palm system was indeed used by several ADOMOCA’s scientists to carry on their own research. For example, Mocage-Palm has contributed to the study of particular atmospheric events (Semane et al., 2007, Bencherif et al., 2007), to study the transport pathways of tropospheric CO (Barret et al., 2008), to explore the impact of the ozone analysis on the radiative budget (Cariolle and Massart, 2007) and to evaluate the ozone loss in the 2002-2003 Arctic vortex (El Amraoui et al., 2008).
In parallel of the development work, we have improved the Mocage-Palm system in order to increase its quality. The most noticeable improvement came from the work on the characterization of the forecast error (Massart et al., 2007, Pannekoucke et Massart, 2008). This work was conducted in synergy with CERFACS assimilation studies in the area of oceanography. To improve the analyses, several other ways were explored. In particular, we studied the best mean to vertically spread the information brought by vertically integrated information, like the information provide by a total column. We have also studied the effect of a re-linearization iterative process around the analysis step. This study will be extended to include in our suite a four dimensional assimilation method. This new implementation is still under validation, but it already provides interesting results and will be used to analyze date from new instruments like IASI (Massart et al., 2009).
The birth of Valentina
Recently, the assimilation system was made independent to the Mocage model in order to apply the assimilation suite to any other chemistry-transport model. The assimilation part, still developed under the Palm framework, was renamed Valentina. Nowadays, the Valentina system could work with various geometries of the CTM (regular or Gaussian global grid, various vertical grids and resolution), at different resolutions. Moreover, Valentina can now work with regional CTMs. It was tested with a limited area version of Mocage and provided promising perspectives for the assimilation of high-resolution datasets over specific regions.