The following text was presented as a Poster at the AFO 2000 Status Meeting (Oktober 2002) and summarises the result of the ongoing activities of the SATEC4D group at EURAD.

Objectives

Figure 1: CONTRACE (14.11.2001 - Flight 1); left panel: wind fields at 12 UTC as a 36 hour forecast showing strong up-winds in the east Mediterranean region; right panel: forecasted NO concentrations at 500hPa (including the flight track), interpreted as a feature of a warm conveyor belt.

A skillful combination of models with data from measurement campaigns is feasible only with spatio-temporal data assimilation. The objective of the project is to analyse mesoscale measurement campaigns. With the 4 dimensional variational data assimilation (4D-var) algorithm as an advanced inversion technique the option is given to obtain a consistent combination of both model and observations, or, alternatively, strong indications of incompliances.
The underlying model is the mesoscale EURAD chemistry transport model. In addition to assimilation runs, the project work includes development issues: variational data assimilation with grid nesting, which is in progress and, likewise, the extension of the assimilation system to non-hydrostatic simulations.
SATEC4D is scheduled to collaborate with four measurement campaigns within the AFO2000 funding program, approximately one per year, preceded by an initial assimilation analysis of the BERLIOZ experiment. SATEC4D meanwhile provided forecast simulations for these four AFO2000 campaigns and assimilated the BERLIOZ campaign.

Model description and assimilation technique

Figure 2: SPURT (19.01.2002); Richardson number (left hand side) at 400hPa and the ozone concentrations at sigma level 20 (~400hPa) (right hand side).

The EURAD CTM2 is a comprehensive tropospheric Eulerian model operating on the mesoscale-alpha. The chemistry transport model calculates the transport, diffusion, and gas phase transformation of about 60 chemical species with 158 reactions. The present grid configuration is 100 hPa model top with 77 x 67 x 15 or 33 x 27 x 15 grid points with 54 km grid resolution. The associated adjoint operators include the gas phase mechanism, the transport schemes and an implicit vertical diffusion scheme. As the specification of error covariances is crucial in data assimilation techniques, special efforts were devoted to identify proper weights in the case of joint emission rate and initial value optimisation. The evolving forecast skill is taken as criterion for identification of the pertinent weights. Given the limited set of observations compared to the much

Figure 3: AFOHAL (16.05.2002); left panel: chemogram at location Dagebüll; right panel: NO2 mixing ratio as a surface plot for the Nest1-area.

larger number of model concentrations, the inversion is an ill-posed problem in a mathematical sense. In these cases the solution is constrained to the least deviation from given background information (= first guess estimates) of emission rates and initial state values. Data assimilation algorithms, which comply with requirements mentioned above must provide a "Best Linear Unbiased Estimate'' (BLUE) in the space-time domain. The four dimensional variational data assimilation (4D-var) method provides for a consistent state analysis in the BLUE sense, while smoothing of observational noise.

Campaign support

To foster a closer collaboration by direct campaign support, it was decided to provide forecast simulations for the measurement episodes in addition to the project commitments, as service to the data providers. The supported AFO2000 campaigns are CONTRACE (02.-30.11.2001), VERTIKO (17.05.-10.7.2002), SPURT (17.-19.01.2002) and AFOHAL (17.04.-24.05.2002). In extension to the scheduled work package, for all these campaigns chemical and meteorological forecast simulations were provided as campaign support and made available on the institute's web server (examples are given in Fig. 1-4).

Figure 4: VERTIKO (08.07.2002); daily maximum in the ozone concentration for the additional saxonian nest (left panel) and the temperature field overlying the horizontal wind field at 12 UTC (right panel).

Graphical output of all campaign simulations is held available on a SATEC4D archive. The various objectives of these campaigns may be found in the respective abstracts/posters.
CONTRACE flight schedule planning was actively supported by 48 hours real time chemical forecast with concentration fields of a variety of constituents (Fig. 1).
Chemograms for the location Dagebüll were provided in support of the AFOHAL experiment (Fig. 3). The VERTIKO campaign forecast was furnished with a second nest level by 5 km x 5 km horizontal resolution over Saxony and the provision of a emission inventory of corresponding resolution.

Assimilation of the BERLIOZ campaign

Figure 5: Assimilation and simulation results for the Berlin area, starting on July 20, 1998, 06:00 UTC and ending after 42 hours (blue solid line). The assimilation process (initial value & emission rate optimisation) ingested observations within the assimilation interval, spanning the first 14 hours (grey shaded). Later observations are given for quality control. A legacy and control simulation without assimilation is shown for comparison (black dash-dotted). Performance results are in general superior for the assimilation run within the assimilation window, especially for ozone. Observations of other species are more affected by the representativity error and hence more difficult to simulate. The striking missprediction for SO2 on the second day around hour 30, given by the control simulation, is remarkably corrected.

The BERLIOZ campaign is re-simulated as the project's first assimilation study and observations of SO2, O3, NO, NO2, and CO are assimilated with the 4D variational algorithm. In addition to campaign data also routine observations from various central European observation networks are assimilated. Assimilation experiments with separate and combined optimisation of initial values and emission rates are performed.

Figure 6: All available observations for grid box (27,16,1) compared with first guess and analysis run; see text for further information.

The potential and limits of data assimilation with too a coarse model grid of 54 km is demonstrated in Fig. 6 for a densely observed Berlin grid box. Clearly, the variation exceeds the assumed observational accuracy, and making it possible to estimate the error of representativity for this grid box and each multiply observed constituent. In general, the error appears to have a non-Gaussian density distribution. With a significant number of NO measurement sites having concentration levels exceeding 100 ppbv, it can be concluded that many of them are too close to roads to have some reasonable information for the assimilation procedure. An analog picture is exhibited for NO2, and, with a qualitatively inverse display visible for ozone. SO2 concentration levels are only modestly scattered, with only one outlier. A critical case is illustrated by CO, where concentration levels given in multiples of 43 ppbv only are available.

Figure 7: Emission optimisation factors for four emitted species for July 20 as inferred for a joint initial value - emission rate optimisation. The BERLIOZ area indicates a moderate increase for Butene, Xylene and SO2, and a moderate decrease for NO2 emission rates.

Outlook

The assimilation for the supported campaigns will be continued according to the project description approximately one per year, with presently assimilation of CONTRACE I measurements being in progress.

Authors: A. Strunk, H. Elbern