Inverse modelling of emissions using in situ and satellite (OMI, GOME2, IASI) data J.Resler, K.Eben, P.Juruš, P.Krč Institute of Computer Science Academy of Sciences of the Czech Rep., Prague
Setup of the experiments - models and inputs Pair WRF - CMAQ 4DVar data assimilation: 2 domains (resolution 27km and 3km) 8 days (June 28 - July 5, 2008) Emission inputs for experiments Based on slightly outdated EMEP inventories (will the corrections reflect some known trends in some regions?)
4DVar with the CMAQ model CMAQ is one of the few regional models having an adjoint code under development: The adjoint model was implemented in CalTech and VirginiaTech ) It includes experimental adjoint code for gas phase (physical processes and chemical mechanism CB4) 4DVar based on this CMAQ adjoint code: Finalized parallelization of the code Implemented observation operators for observations from in-situ stations tropospheric columns or profiles from satellite instruments ) A. Hakami et al.: The Adjoint of CMAQ, Environ.Sci.Technol., 2007
Setup of the experiments - in-situ observations NO 2 stations from CZ (AIM), DE (UBA), B, GB (EEA) Coarse domain - cca 180 stations Fine domain - cca 30 stations O 3 stations from Europe (EEA) About 325 stations Background rural and suburban stations, altitude < 900 m In situ stations at coarse domain (left) and fine domain (right)
Setup of the experiments - satellite observations Satellite instruments OMI, GOME2 and IASI Tropospheric columns of NO 2 from the TEMIS service The lowest layer of the profiles of O 3 from IASI ) Tropospheric NO 2 columns and O 3 profile layer at July 1 2008. GOME2 10:24 (left), OMI 13:24 (center), IASI 12:00 (right) The data contain other necessary quantities for the observation operator (averaging kernels, standard deviations,...) Overpass of the satellites occurs for our domains usually twice a day. ) In collaboration with Jean-Luc Attié, Laboratoire d Aérologie, Univ. de Toulouse III
Enhanced assimilation design Enhancements in assimilation method configuration (compared to previous experiments with single multiplicative emmision factors): Tuning of the diurnal emission profiles. Implementation of adjoint for SAPRC99 mechanism. Observation of O 3 - stations and satellite columns Multivariate assimilation in IC and emission factors Species with significant sensitivity. Excluded radicals and non-reactive species Preconditioning of the minimisation by transformation of variables.
Emission profiles optimization Optimizing each emission coefficient independently in every time step would make the problem ill-conditioned, parameterization is necessary. Emission profile corrections parametrized using five B-splines. Gradient of emission coefficients calculated separately Basis functions for parametrization of the correction of diurnal emission profile.
Multivariate assimilation of NO 2 and O 3 Experiment with all proposed enhancements. Assimilation shifts emissions from evening peak to morning in some regions. Maps of the optimized emission coefficients for 30.6.2008 for differents parts of the day Upper from left: night, morning, lower from right: midday, evening, night.
Results - correction factor and emission Graphs of the daily profile of the correction factor and emission of NO 2 in Prague for 2.7.2008. Black: emission modelled by emission model, red: corrected emission.
Results - correction factor and emission Graphs of the daily profile of the correction factor and emission of NO 2 in Brno for 2.7.2008. Black: emission modelled by emission model, red: corrected emission.
Results - correction factor and emission Graphs of the daily profile of the correction factor and emission of NO 2 in Usti nad Labem for 2.7.2008. Black: emission modelled by emission model, red: corrected emission.
Results - correction factor and emission Graphs of the daily profile of the correction factor and emission of NO 2 in Koln for 2.7.2008. Black: emission modelled by emission model, red: corrected emission.
Results - performance of the method Improvement of the analysis and forecast of the NO 2 and O 3 NO 2 28.6.2008 29.6.2008 30.6.2008 1.7.2008 2.7.2008 3.7.2008 4.7.2008 5.7.2008 Freerun 5,38 5,90 7,26 9,39 9,99 9,36 6,80 6,57 Analysis 3,88 3,57 5,34 7,32 7,33 7,59 6,05 - Forecast - 4,27 6,20 8,28 8,26 6,93 5,06 5,20 Analysis(%) 27,88 39,41 26,54 22,02 26,59 26,04 25,61 - Forecast(%) - 27,59 14,66 17,26 17,26 18,95 11,00 20,84 O 3 28.6.2008 29.6.2008 30.6.2008 1.7.2008 2.7.2008 3.7.2008 4.7.2008 5.7.2008 Freerun 18,60 20,79 24,43 29,94 29,40 26,37 20,69 24,36 Analysis 16,38 17,39 18,36 23,54 24,93 21,58 15,97 - Forecast 18,60 20,74 24,04 27,59 27,58 24,77 18,19 21,99 Analysis(%) 11,91 16,37 24,87 21,39 15,21 18,16 20,80 - Forecast(%) - 0,26 1,60 7,83 6,21 6,07 12,08 9,70 MAE for freerun, analysis and forecast and relative percentage improvement of analysis and forecast. Upper: NO 2, lower: O 3
Thank you for your attention E-mail: jurus@cs.cas.cz, resler@cs.cas.cz References: K.Eben et al.: An ensemble Kalman filter for short term forecasting of tropospheric ozone concentrations A.Sandu et al.: Adjoint Sensitivity Analysis of Regional Air Quality Models A. Hakami et al.: The Adjoint of CMAQ H.Elbern et al.: Emission rate and chemical state estimation by 4-dimensional variational inversion A.Sandu,et al.: Inverse Modeling of Aerosol Dynamics Using Adjoints: Theoretical and Numerical Considerations D.K.Henze,et al.: Inverse modeling and mapping US air quality influences of inorganic PM2.5 precursor emissions using the adjoint of GEOS-Chem
Conclusions The proposed method gives optimized emissions, which improve the agreement of the model with the observations. The ability to improve the forecast of NO 2 is good, the improvement in the forecast of ozone is still only moderate. The method can be used for designing an adaptive forecasting system. The method can be used for validation and optimization of the emission model. We suppose to use it for tuning an emission model based on detailed national inventories.
Outlook Next supposed steps for improving of the method: More sophisticated statistical model for explaining daily and weekly fluctuations in the emission corrections would be needed. Better background error covariance structure, e.g. anisotropic nonhomogenous covariance function diffusion operator used as covariance function Optimization of other model parameters (deposition rates) Better acounting for variances of observation errors. We intend to participate on efforts in development of the CMAQ adjoint model for aerosols.