Emission optimization

Among the uncertain input data, emissions of atmospheric pollutants pose one major contribution to forecast errors. In this respect, anthropogenic and biogenic emissions must be treated differently. Biogenic emissions are calculated online using meteorological model data. Anthropogenic emissions are taken from emission inventories and distributed temporally using predefined functions. These predefined functions do not take actual weather or societal conditions such as heat waves or local traffic restrictions into account leading to possibly large over- or underestimations of the concentration of atmospheric pollutants. Thus, optimizing the emissions by assimilating given observational data is essential for reliable air quality analyses. Further, such analyses provide information to control emission regulations and guide actions regarding improved air quality.

Research areas

1. Immission controlled assessment of pollutant emissions: assessment of emission data using in-situ and remote-sensing observations of air pollutants.
2. Evaluation of German emission inventory: analysis of the spatial and temporal distribution of anthropogenic emissions of air pollutants using 4D-var.
3. Emission optimization of exceptional aerosol events (volcanic eruptions, biomass burning): development of advanced assimilation strategies to quantify highly uncertain emission events.
4. Polluter specified optimization of emissions: assessment of the potential and limitations to quantify emissions for different causer groups separately.

Predictability

Deterministic data assimilation methods provide the analysis of the most probable state that may lead to large forecast errors if model input data, such as initial values and emissions are erroneous. Given uncertain observations and model input data, the analysis of forecast uncertainty (“predictability”) aims to find a representation of the full probability density function (PDF) of the model state. As the complexity of atmospheric models allows only Monte Carlo simulations, an approximation of the full PDF can be simulated by ensemble techniques. Here, one major task is the optimal setup of the ensemble to keep the number of ensemble members manageable. This optimal ensemble setup can be realized by singular vector decomposition or Karhunen-Loeve expansion to optimally represent the largest uncertainties in the model. Further, ensemble data assimilation approaches are developed to make use of observations of the atmospheric state for analysing the uncertainty of the predicted model state leading to a particle smoother that utilizes the adjoint model code. It aims to estimate the uncertainty of the model input data, which leads to deviations from the observations and is realized in the Ensemble for Stochastic Integration of Atmospheric Simulations (ESIAS).

Research areas

1. Estimation of the forecast uncertainty: How probable is the modelled atmospheric state given observations of air pollutants?
2. Identification of areas characterized by low predictability: Can different observations lead to more reliable model predictions?
3. Evaluation of parameter uncertainties (emissions, deposition, meteorological model parameters – e. g. surface parameter and parametrization setups) → identification of sources of uncertainty: What governs the predictability of the modelled atmospheric state?

Observability

Largest uncertainties for successful atmospheric model predictions lie in the initial and boundary values chosen. Thus, observations provide essential information to data assimilation systems allowing for the optimisation in model predictions and analyses. The research field of observability investigates the impact of utilized observations on the value of forecasts that are important to society. If initial values and e. g. emissions are ineligibly chosen for model simulations, the forecast system can react very sensitively leading to growing forecast errors. The improvements regarding the analysis quality as a product of observation configurations and data assimilation can be evaluated in qualitative measures as well as in a probabilistic manner. Consequently, to minimize forecast uncertainties, targeting observations must be placed into areas, where the analysis errors reinforce large forecast errors.

Research areas

1. Assessment of observation contribution to reliable forecasts and analyses
2. Identification of areas and chemical compounds that are well constrained by the information content provided by observations
3. Optimization of observation networks and optimal sensor placement
4. Evaluation of quality gain in the analysis due to individual or types of observations

Chemical Weather Forecast

We provide daily forecasts and analyses of European-wide transport and transformation of atmospheric trace gases and aerosols. Therefore, a grid nesting technique is used allowing for simulations with different horizontal resolution from 45 km x 45 km to the particularly high resolution of 1 km x 1 km. In these simulations, up to 120 variables per grid point are calculated, even if only a small fraction of these, as e. g. Nitrogen Oxides, Ozone, and particulate matter, are provided every day for the use by environmental protection agencies or the public. Daily forecasts and analyses and yearly re-analyses of the European air quality on the 9 km scale are published via the Copernicus Atmosphere Monitoring Service (CAMS) multi model ensemble. Forecasts and analyses for the higher resolved model grids up to 1 km x 1 km are here available.

Research areas

1. Regional chemistry transport modelling with EURAD-IM
2. Air quality prediction and analysis in the framework of the European Copernicus Atmosphere Monitoring Service (CAMS2_40)
3. Parameterization of atmospheric processes