Enhancing Air Quality Assessment through Data Assimilation in the Air4EU Project

1. Data assimilation in the air4EU project Martijn Schaap and Peter Builtjes Martijn Schaap

2. Air quality assessment Measurements Modelling To be combined using data assimilation Data assimilation

3. Data assimilation: Introduction • Data assimilation is the technique whereby observational data are combined with model output to produce an optimal estimate of the evolving state of the system. • In practise, data assimilation schemes boil down to variations of “least squares” and “weighted averages”. Data assimilation

4. Passive and active approaches • Passive approach: Modelled fields are synthesized with measurements without a feedback to the model state. • Synthesis is performed as a “finishing touch” using the model output Examples: Kriging, Optimum interpolation • Active approach: Modelled fields are synthesized with measurements with a feedback to the model state. • The system uses and updates the causal relations within the modelling system • Assimilation is performed on the fly Examples: Kalman filtering, 3D-VAR Data assimilation

5. The process: • Gather the measurement data • Perform quality control on the data • Uncertainty assessment, incl spatial representativity • If needed, determine model uncertainty • Set up the system and perform experiment • Analysis of the results including validation. Data assimilation

6. Passive data assimilation: Synthesizing in-situ observations, satellite data and modelled fields • Purpose: To generate regional PM10 map over Europe for 2003 • Methods: Geo-statistical methods • Input: AIRBASE PM10, LOTOS-EUROS PM2.5, MODIS AOD Jan van de kasteele et al. Data assimilation

7. PM10 maps based on different conbinatins of observations, modeled fields and satelite data Combination of these information sources gives an added value Validation: Measurements in combination with: Data assimilation

8. Examples of active data assimilation: EnKF • The LOTOS-EUROS air quality model • An Ensemble Kalman Filter Goal of these experiments: To generate regional background concentrations for European cities Ensembles created by putting noise to emissions and dry deposition speeds Examples: • Ozone: summer 2003 • PM: 2003 Data assimilation

9. AOD assimilation: May, 9th Retrieval Stdev (%) Data assimilation

10. AOD assimilation: May, 9th Model Assimilation Data assimilation

11. The Measurements • O3 • EMEP data assimilated • AIRBASE as validation • Hourly data • Stdev = 10% with max of 5 ug/m3 • PM10: • AIRBASE data assimilated • Validation with city data and some EMEP stations • Daily data • Stdev = 12.5% PM10 stations in AIRBASE Data assimilation

12. Results Ozone (ppb) Ozone measurements from the EMEP network assimilated Data assimilation

13. Validation Vredepeel Assimilation station Westmaas Validation station Data assimilation

14. Validation Data assimilation

15. Validation for PM10 Underestimation: 40-60% Bias correction: 50% Model Measurement Data assimilation

16. PM10 fields for March-April, 2003 Data assimilation

17. Validation plot • First results give a mixed picture • Small innovations point at too low model uncertainty in combination with high measurement uncertainty, in this case noise on rainout is needed • PM is complex, due to large measurement uncertainties Data assimilation

18. Applications active data assimilation Relatively unexplored area • Air quality assessment (the state) • Determine initial state for forecasting • Parameter estimation, e.g. emissions • Validation, yields directions for model improvement • Nudging of fields for process studies • Observation system experiments / network design • ……………… Creative minds will find more oppurtunities Data assimilation

19. Fast Sparse data No causal relations Application limited to assessment Knowledge on total model uncertainty Large computation Large number of data required Causal relations, influence unobserved quantities Wide range of applications Knowledge on uncertainties in parameterisations and input parameters Passive Active • Both have these benifits: • Forced attention to quality control • Errors in data and in model are accounted for • Filling in data poor regions based on process knowledge Data assimilation

20. Conclusions • Data assimilation yields an improved assessment of the air pollution regimes in Europe. • Passive assimilation techniques can be effectively used for (sparsely available) measurement data • Applying active assimilation techniques is feasible with present facilities and gives consistent fields through the model. Further, they give additional info on causal relations • Undertainty analysis very important and determines the end result. Model development and improved measurement techniques will improve the results of the system. • It is advisable that the EMEP programme makes use of data assimilation techniques and does not refrain itself to using passive techniques Data assimilation

21. Data assimilation