A Methodology for Seasonal Photochemical Model Simulation Assessment

Transcription

1 A Methodology for Seasonal Photochemical Model Simulation Assessment Veronica Gabusi and Marialuisa Volta Dipartimento di Elettronica per l Automazione Università degli Studi di Brescia - Italy

2 Outline Seasonal photochemical model assessment Critical issues The proposed methodology Application cases Northern Italy domains GAMES Modelling System 1996 Summer Season The CityDelta Modelling Exercise Conclusions

3 Seasonal Simulation Assessment Critical issues: no standard performance procedure exists each run of the models results in a large amount of output data Comparison between punctual and average values to focus the analysis for a restricted number of typical concentration patterns to summarize the simulation results over long time period by means of proper performance indexes.

4 The Model Evaluation Methodology (1) Clustering process (2) Sites Identification Used for classifying patterns Ozone concentrations Daily shape, peak distribution Each cluster can be represented by a single station Representative station Virtual station (3) Performance indicators Statistical indicators (EPA, EC) Other Statistical indexes O3 exposure values Percentile values Model inter-comparison

5 Seasonal Application Cases Simulation performed with GAMES system over Northern Italy domains Preliminary Application: 1996 Summer Season - Contribution to EUROTRAC2/SATURN Project CityDelta Modelling Exercise

6 ECMWF model Wind and Temperature profile EMEP model concentrations IC_BC The GAMES system Gas Aerosol Modelling Evaluation System Topography Land Use Initial and boundary conditions CALMET CALGRID 3-dimensional pollutant Concentration patterns Local measurements -Synop - Regional network 3D wind and temperature fields turbulence parameters Emission fields Upper air sounding data temperature profiles Emission model CORINAIR Emission inventory Spatial disaggregating coefficients Emission time patterns VOC splitting profiles

7 UTM (km) UTM (km) 240km x 232km complex terrain Case I: the selected domain industrialised and populated area close road network critical anthropic emissions the simulated period: April-September 1996 (UTM -km) VA CO PV MI LC LO BG (UTM - km) SO CR BS MN VR

8 ALRENON MOTTARONE POST_121 POST_308 POST_903 POST_902 POST_674 POST_675 POST_717 POST_602 POST_606 POST_673 POST_616 POST_624 POST_676 POST_618 POST_715 POST_698 POST_716 POST_662 POST_905 POST_907 POST_906 POST_503 POST_209 POST_801 POST_201 POST_526 POST_304 POST_301 POST_305 SPIETROCA POST_654 POST_751 POST_711 mottar Clustering process and sites identification Distances renon Mottaron Varenna LegnLimbiate Vimercat Mi_Juva mean day [ppb] Gambara Renon Rural Urban Remote Case I spcapof GAMBARA LEGN_SMA LIMBIATE MI_JUVA MOTTARON RENON VARENNA VIMERCAT [hour]

9 US EPA guidelines: Statistical Indexes Mean Normalized Bias Error (MNBE), ±5 15% Mean Normalized Gross Error (MNGE), ±30 35% EC Directive: the 1 hour averages (daytime), ± 50% the 8 hours daily maximum, ± 50% Model intercomparison: CALGRID STEM 1 n n 1 = 1 n n 1 = 1 Legnano Limbiate Juvara Vimercate Ispra Parma Varenna 1 C C mod mod ( x, t) C C ( x, t) obs ( x, t) C C obs obs ( x, t) Case I obs ( x, t) ( x, t) O3-1-h daily max O3-8-h daily max NO2 - mean conc. mnbe [%] mnge [%] mnbe [%] mnge [%] mnbe [%] mnge [%]

10 Computed Computed LEDA - d_1hmax STEM CALGRID LEDA - d_8hmax STEM CALGRID Measured [ppb] Model inter-comparison Measured [ppb] Computed Computed HEDA - d_1hmax STEM CALGRID HEDA - d_8hmax Measured [ppb] STEM CALGRID Measured [ppb] Case I LEDA: low emission density area HEDA: high emission density area

11 Case II: the selected domain 300km x 300km the simulated period: April-September 1999

12 COSSATO VA_VIDOL CHIAVENN RE_MASSE PARMA PIACENZA CREMA AGRATE VIMERCAT LIMITO MEDA ARCONATE MOTTA_V MAGENTA CASTELLA Ozone pattern classification (I) Case II Cluster Tree UTM (km) Distances TORINO NOVARA VARESE SONDRIO TRENTO BERGAMO BRESCIA MILANO VERONA PIACENZA Cluster 1 ALESSANDRIA Cluster 2 PARMA Cluster 3 Cluster 4 Cluster 5 GENOVA Cluster UTM (km) MODENA

13 O3 [ppb] Ozone pattern classification (II) [hour] UTM (km) TORINO SONDRIO VARESE BERGAMO BRESCIA NOVARA MILANO VERONA PIACENZA Cluster 1 ALESSANDRIA Cluster 2 PARMA Cluster 3 Cluster 4 Cluster 5 GENOVA Cluster UTM (km) Case II TRENTO MODENA

14 O3 [ppb] Representative station definition -Representative station new clustering process -Virtual station averaging, hour by hour, the cluster 1 cluster 3 cluster 5 Case II Varese_obs Varese_comp average_obs average_comp [hour] measurements recorded in the stations belonging to the group Vimercate_obs average_obs [hour] Magenta_obs average_obs [hour] Vimercate_comp average_comp Magenta_comp average_comp Rural cluster Urban cluster Suburban cluster

15 Performance evaluation Model evaluation (I) Graphical analysis: time series, scatter plots, Statistical indicators: US EPA guidelines: Mean Normalized Bias Error (MNBE), ±5 15% Mean Normalized Gross Error (MNGE), 30 35% Max 1h Max 8h Cluster MNBE MNGE R RMSE MNBE MNGE R RMSE Case II O3 concentrations

16 Observed Model evaluation (II) Sim ulated 8h p25 8h p50 8h p75 8h p Urban cluster Simulated 1h p25 1h p50 1h p75 1h p95 1h p25 1h p50 1h p75 1h p95 (repres.) 1h p25 1h p50 1h p75 1h p95 (virtual) Case II

17 Conclusions Methodological approach to evaluate seasonal photochemical simulation Simulations performed in complex domains Main issues: 1996 Summer Season (SATURN) model inter-comparison 1999 Summer Season (CityDelta Project) Statistical and graphical methods Each station or representative station The evaluation of representative stations should be preferred with long-term simulation assessments