v represents the uncertainties of the model. The observations are related

Transcription

1 HARMO June 010, Paris, France - 13t Conference on Harmonisation witin Atmosperic Dispersion Modelling for Regulatory Purposes H DATA ASSIMILATION IN AIR QUALITY MODELLING OVER PO VALLEY REGION Gabriele Candiani 1, Claudio Carneale 1, Gioanna Finzi 1, Enrico Pisoni 1 and Marialuisa Volta 1 1 Department of Information Engineering (DII), Uniersity of Brescia, Brescia, Italy Abstract: An accurate description of te air quality state is a ery callenging task due to te complexity and non-linearity of te mecanisms taking place in atmospere. Tis is especially true wen dealing wit secondary pollutants suc as ozone (O 3). Deterministic cemical transport models are aluable tools to understand te spatial distribution of an air pollutant oer a certain domain. Tese models can proide spatially consistent air quality data as tey consider te main pysical and cemical processes goerning te atmospere. Howeer, uncertainties in formalization and input data (emission fields, initial and boundary conditions, meteorological patterns) can eaily affect simulation results. Tis issue can be addressed troug te use of Data Assimilation (DA), wic combines concentration fields of secondary atmosperic pollutants simulated by a cemical transport model (background) wit measured data (obserations). In tis work, O 3 concentration fields oer te Po Valley region (Nortern Italy) ae been simulated by te cemical transport model TCAM. For te period from 15 t May to 15 t July 007, te Optimal Interpolation (OI) algoritm as been used to assimilate O 3 and NO data measured by bot ground stations and satellite data, into te TCAM simulations. Te study area ( km ) is caracterized by complex topograpy, ig antropogenic emissions and frequent stagnant meteorological conditions leading to ig concentration leels of ozone in summer. Preliminary results sow tat te estimate of O 3 concentrations improes significantly assimilating O 3 data, wile no impact or een a negatie impact can be seen wen performing te assimilation of NO for bot ground-based and satellite data. Te researc as been deeloped in te framework of te Pilot Project QUITSAT (QUalità dell aria mediante l Integrazione di misure da Terra, da SAtellite e di modellistica cimica multifase e di Trasporto - contract I/035/06/0 ttp:// sponsored and funded by te Italian Space Agency (ASI). Key words: Air Quality, Data Assimilation, Optimal Interpolation INTRODUCTION An accurate description of te air quality state is a callenging task due to te complexity of te atmospere mecanisms. In fact, te dynamic of a pollutant concentration is usually non-linear and partially unknown. Tis is especially true wen dealing wit secondary pollutants suc as ozone (O 3 ), wic constitutes one of te major concerns in Europe, due to its armful effects on uman ealt and on natural ecosystems. Ozone formation and accumulation are non-linear processes, wic depends on a large number of reactions taking place in te atmospere. Tese penomena are faoured by meteorological stagnating conditions and by primary emissions of precursors (namely nitrogen oxides and olatile organic compounds). Due to tese reasons, modeling tools are required to proide a good estimate of te air quality state. In fact, Cemical transport model (CTM) are able to reproduce in a 3D-gridded domain te complex cemical and pysical processes goerning te atmospere. Howeer, uncertainties in formalization and input data (emission fields, initial and boundary conditions, meteorological patterns) can eaily affect simulation results. Tus, a Data Assimilation (DA) tecnique suc as Optimal Interpolation (OI; Kalnay, E., 003), can be used to assimilate measurements in te simulations of a deterministic cemical transport model, in order to obtain a better estimate of te atmosperic state. Te paper is organized as follows: Section describes te formalization of te implemented sceme; Section 3 illustrates a case study in wic O 3 and NO data, measured by ground stations and a satellite sensor, are assimilated into a cemical transport model to estimate te ozone concentration; in Section 4 conclusions are gien. METHODOLOGY Aim of te DA sceme is to compute te best possible estimate (analysis state x a ( t) concentration (true state x ( t) ) starting from a model simulation (background x b ( t) (obserations y o ( t) ). Te background is modeled as were x n b ( t ) R, f represents te CTM and ( t ) to te true state x ( t) by ( t) f ( x ( t),( t) ) ) of te unknown pollutant ) and from a series of measured data x & (1) b = b represents te uncertainties of te model. Te obserations are related ( t) H ( x( t),t) w( t) y o = + () were y p o ( t ) R, te matrix ( ) p n H t R, called obseration operator, maps te model true state ( t) obserations true state y ( t) and w ( t) is te obseration error. Te model error ( t) and te obseration error ( t) assumed to be wite, gaussian and mutually uncorrelated: ( t) N(,B) x into te w are 0 (3) ( t) N(,R) w 0 (4) Session Enironmental impact assessment 75

2 HARMO June 010, Paris, France - 13t Conference on Harmonisation witin Atmosperic Dispersion Modelling for Regulatory Purposes n n p p [ ( ) ] = Ε[ w( ) ] = Ε[ ( ) w( ) '] = 0 Ε (5) B R and R R are te error coariance matrices of te background and te obserations respectiely. Te specification of te error coariance matrices B and R is usually a difficult task. Te matrix B is of utmost importance as it describes te model error spread oer te domain. In tis work, B is estimated using te Gaussian exponential function d d B = s exp exp = s L L B ~ (6) were: d and L and d are te distances between two grid cells of te domain in te orizontal and ertical direction; L are two parameters defining te decay of coariance in te orizontal and ertical direction; s is te model error ariance computed on te basis of preious simulations. Tis approximation states tat te model error ariance is constant for eac cell of te domain, wile te error coariance between two grid points is a function of te orizontal and ertical distance between tem. Te R matrix can be considered diagonal wen te measurements performed by te monitoring stations are independent. Moreoer, if te same type of instruments are used for te measurements, it could be possible to assume tat all te monitoring stations ae te same error ariance r, tus R could be rewritten as R = r I (7) Te assimilation sceme follows a two steps approac: 1) at time t, te model is soled to obtain te background field t x a t is computed troug te OI algoritm merging te information of te background wit te x b ( ) ; ) te analysis ( ) information coming from te obserations ( t) Step 1 Te computation of ( t) y o. Te analysis represents te initial state for te step 1 at time t + 1. x b is performed using a CTM, able to reproduce te complex and non-linear penomena inoling pollutants in atmospere. Suc model describes te concentration of eac pollutant oer a certain domain, troug a mass conseration equation, considering transport, emission, deposition and cemical transformation penomena. To sole te resulting partial differential equation system it is necessary to proide te initial and boundary conditions for eac inoled species: x a t 1 ). Initial conditions, te alues for te concentrations of transported species oer all te domain ( ( ) Boundary conditions, te alues for te concentrations of transported species on te boundaries of te application domain during te simulation time. Oter fundamental inputs are te emission and meteorological fields, tat must be supplied to CTMs wit detailed spatial and temporal resolution. In general, te CTM equation system is soled on a regular 3D grid troug a numerical approac. Te simulations performed using te equation (1) oer te simulation period, proide te background x b ( t) needed for te assimilation step. Step At eac time step t, te analysis state x a ( t) is computed assimilating obserations y o ( t) into te background state ( t) estimated at step 1, applying te OI algoritm x a ( H ~ ) ( t) x ( t) + K y ( t) x ( t) = b o b, (8) were H ~ is a linear operator wic performs te interpolation (based on te Inerse Distance Weigted algoritm) of te x b t at te obseration points. background state field ( ) x b Te matrix n p K R is computed by: ( H ~ BH ~ ' + ) 1 K = BH ~ ' R. (9) 76

3 HARMO June 010, Paris, France - 13t Conference on Harmonisation witin Atmosperic Dispersion Modelling for Regulatory Purposes It sould be noted tat, under tese assumptions about B and R matrices, te Kalman gain can be written as: 1 r B ~ H ~ ' s H ~ B ~ H ~ ' + I = B ~ H ~ ' ( H ~ B ~ H ~ ' + ) 1 K = s σ I. (10) s r were te only degree of freedom σ = is te ratio between te obserations and model error ariances. s CASE STUDY In tis study, OI algoritm is used to assimilate obsered and simulated ourly ozone concentration oer a km domain placed in Nortern Italy (Figure 1). Tis region is caracterized by densely inabited and industrialized area, and by ig antropogenic emissions, frequent stagnating meteorological conditions and Mediterranean solar radiation regularly causing ig ozone leels in particular during summer monts. Te period selected for te analysis ranges from 15 t May to 15 t July 007. Figure 1. Study area wit te ground stations used for te assimilation (black crosses) and for te alidation (red triangles). Background fields Te background field is computed by means of te Transport Cemical Aerosol Model (TCAM; Carneale, C. et al., 008). TCAM is a part of te Gas Aerosol Modeling Ealuation System (GAMES; Volta, M. and G. Finzi, 006) sown in Figure, wic also includes: te meteorological pre-processor PROMETEO, tat proides TCAM all te meteorological input fields in te correct spatial-temporal resolution, starting from te output of continental scale models; te emission processor POEM-PM (Carneale, C. et al., 006); a boundary condition pre-processor, tat computes te boundary conditions for TCAM model in te application domain starting from te simulation of continental scale models. Figure. Te GAMES modelling system. Te emission fields are estimated by POEM-PM pre-processor starting from te 004 CTN emission inentory (Deserti, M. et al, 009). Meteorological fields are computed by means of MM5 model (Grell, G. et al., 1994). Te boundary conditions are computed starting from continental scale simulation of CHIMERE model (Scmidt, H. et al., 001). Session Enironmental impact assessment 77

4 HARMO June 010, Paris, France - 13t Conference on Harmonisation witin Atmosperic Dispersion Modelling for Regulatory Purposes Obserations data Te obserations consist of data measured by ground stations and OMI sensor during te same period of te TCAM simulations. Data from 6 and 3 stations were used for te assimilation of O 3 and NO respectiely. Tese measurements were collected from te regional monitoring networks for Italy, and from AirBase database for France and Switzerland. Te 75% of te stations (black crosses in Figure 1) are used in te assimilation sceme, wile te remaining stations (red triangles in Figure 1) are used for te alidation. Measurements of te troposperic NO column proided by te OMI sensor are aailable for 0 days during te simulation period. DA parameters Te parameters used in te OI algoritm ae been cosen on te base of sensitiity analysis and literature alues. In te case of ground stations, te orizontal and ertical influence lengts L and L are equal to 80 km and 80 m respectiely. In te case of OMI data, L is set to 10 km. Te σ is set to 0.1 as suggested in Kalnay, E., 003. Results Te alidation is carried out comparing te results of te model simulations, before a new assimilation step is performed, wit te alues measured by te stationss cosen for te alidation and terefore not used in te assimilation process. Figure 3 and Figure 4 sow te results of te ozone statistics in terms of box plots, for te assimilation of O 3 and NO respectiely. Te box plots present te comparison between te case in wic te assimilation is performed (OI) or not (TCAM). Figure 3 sows te results for te ozone obtained assimilating O 3 measurements. It can be noticed tat all te box plots confirm an improement due to te assimilation of O 3 data. In particular, all te error box plots sow a reduction of te error alue wen te assimilation is performed. Te Correlation coefficient instead presents an improement, wit an increment in te statistic from 0.4 to 0.6. Figure 4 and 5 sow te impact on te ozone due to te assimilation of NO measured by ground stations and OMI sensor. From te box plots it can be seen tat ozone estimate as no benefit from te NO assimilation, neiter using sparse ground stations obserations frequent in time (ourly) nor wit dense OMI obserations wit a poor temporal resolution (daily). In particular, all te box plots present a ery similar statistics for bot te case wit (OI) and witout (TCAM) te assimilation. CONCLUSION In tis work, te impact on te estimatee of ozone fields is ealuated assimilating two different species (O3 and NO) from bot ground stations and a satellite sensor. Te data assimilation is performed using te OI algoritm, a sequential process wic performs te assimilation in two steps. In te first step, te model is integrated forward in time to obtain te background field. In te second step, te obserations coming from measurement stations are assimilated into te background field to produce te analysis; tis new field is te input for te next model integration. Te comparison of te assimilation of te two species sows different results. Te statistics sow tat te assimilation of O 3 significantly improes te estimation of ozone concentrations. On te contrary, te NO assimilation does not seem to gie any benefit to te estimate of ozone fields, independently on te use of ground-based data or te OMI sensor. AKNOWLEDGEMENTS Tis work as been deeloped in te frame of Pilot Project QUITSAT (QUalità dell aria mediante l Integrazione di misure da Terra, da SAtellite e di modellistica cimica multifase e di Trasporto - contract I/035/06/0 - ttp:// sponsored by te Italian Space Agency (ASI). We also acknowledge te Italian Ministry of Uniersity and Researc (MIUR), te COST78 action (Enancing mesoscalee meteorological modeling capabilities for air pollution and dispersion applications) and te EU Network of Excellence ACCENT (T&TP). Figure 3. Validation of te impact of O3 data assimilation oer te O3 concentrations. Box plots sow: NME; NMAE; RMSE; Correlation. 78

5 HARMO June 010, Paris, France - 13t Conference on Harmonisation witin Atmosperic Dispersion Modelling for Regulatory Purposes Figure 4. Validation of te impact on te O 3 concentrations due to te assimilation of NO dataa measured by ground stations. Box plots sow: NME; NMAE; RMSE; Correlation. Figure 5. Validation of te impact on te O 3 concentrations due to te assimilation of NO dataa measured by OMI sensor. Box plots sow: NME; NMAE; RMSE; Correlation. REFERENCES Carneale, C., E. Decanini, M. Volta, 008: Design and alidation of a multipase 3D model to simulate troposperic pollution. Science of te Total Enironment, ol. 390, p , ISSN: , doi: /j.scitoten Carneale, C., V. Gabusi, and M. Volta, 006: POEMPM: an emission model for secondary pollution control scenarios, Enironmental Modelling and Software, ol. 1, pp Denby, B., Horlek, J., S. E. Walker, K.E., Fiala, J.: Interpolation and assimilation metods for Europe scale air quality assessment and mapping. Part I: reiew and recommendations. Tecnical report, ETC/ACC Tecnical Paper 005/7 (005) Deserti, M., E. Minguzzi, M. Stortini, S. Bande, E. Angelino, M. Costa, G. Fossati, E. Peroni, G. Pession, F. Dalan, S. Pillon, C. Carneale, G. Finzi, E. Pisoni, G. Piroano, and M. Bedogni, 009: A performance ealuation of cemical transport models in te Po Valley, Italy, in Proceedings of 7 t International Conference in Air Quality. Grell, G., J. Dudia, and D. Stauffer, 1994: A description of te Fift generation Penn State/NCAR Mesoscale Model (MM5), tec. rep., NCAR Tec Note TN STR. 1 pp. Kalnay, E., 003: Atmosperic modelling, data assimilation and predictability, Cambridge Uniersity press. Scmidt, H., C. Derognat, R. Vautard, and M. Beekmann, 001: A comparison of simulated and obsered ozone mixing ratios for te summer of 1998 in Western Europe, Atmosperic Enironment. Volta, M. and G. Finzi, GAMES, 006: a compreensie Gas Aerosol Modeling Ealuation System, Enironmental Modelling and Software, ol. 1, pp Session Enironmental impact assessment 79