320 likes | 463 Views
COST-733 WG 4 -Meeting. INSTITUTE OF METEOROLOGY AND WATER MANAGEMENT. Department of Monitoring and Model l ing Air Pollution, Krakow, Poland. TITLE : Comparison of selected weather types classifications , for air pollution data from different areas. Jolanta Godłowska
E N D
COST-733 WG4-Meeting INSTITUTE OF METEOROLOGY AND WATER MANAGEMENT Department of Monitoring and Modelling Air Pollution, Krakow, Poland TITLE :Comparison of selected weather types classifications, for air pollution data from different areas Jolanta Godłowska Anna Monika Tomaszewska Ioannina 9-10.05.2008
My questions are: • Are there similar results comparing different classifications: • by different methods (EV, WSD, WSD_U) • for different air pollutants (PM10, CO, NO2, SO2, ozone) • for different sites (Poland, Slovakia, Germany, Belgium) • What is the nature of EV, WSD, WSD_U parameters ? • modification of WSD_U and WSD • How results depend on domain ? • comparing results for 7, 8 and 5 domains • What kind of classifications is the best for forecasting situations with high concentrations? • WSD_U - Ustrnul weighted standard deviation index
Comparison of selected weather types classifications for forecasting the days with high air pollution • Data: • SO2 PM10 NO2 CO • NDJF • daily mean concentrations -SO2, PM10, NO2 • maximal daily 8-hour concentrations – CO • from: • Poland • Cracow 1994 -1999 • Upper Silesia 1999-2002 • Belgium • Uccle (1996-2002) – only PM10
Methods of classification evaluation : the best: EV=1-(SSi/SSt) between 0 and 1the highest k WSD = (1/k)*∑ sdi depending on standard deviation the lowest ki=1 k WSD_U = (∑ sdi*ni)/(∑ni) depending on standard deviation the lowest i=1 i=1 Relation between EV (left), WSD (center), WSD_U(right) and number of classes N for different air pollutants Conclusions: 1.WSD and WSD_U methods arenot good for comparing results for different air pollutants 2. Normalisation of WSD and WSD_U parameters are necessary.
New methods of classification evaluation after normalisation: • WSD and WSD_U normalised:nWSD = WSD/sd • nWSD_U = WSD_U/sd • sd - total standard deviation Relation between EV (left), nWSD (middle), nWSD_U (right) and number of classes N for different air pollutants (Upper Silesia) • Conclusion: • For all methods and species better quality is observed for classifications • with number of classeslarger than 15 • Probably classifications with number of classes larger then 15 • are better for air pollution forecasting • For NO2 it is observed the worst evaluation
Comparison of different methods of classification evaluation Conclusion: EV and nWSD_U are correlated the most
Comparison of differentmethods of classification evaluation (EV, nWSD, nWSD_U) for SO2, PM10, NO2, CO Upper Silesia
Comparison of classification ESLPC30 with LWT2 for SO2 Upper Silesia
Comparison of differentmethods of classification evaluation (EV, nWSD, nWSD_U) for SO2, PM10, NO2, CO Upper Silesia
Comparison of EV evaluation for different pollutants at different places SO2, PM10, NO2, CO Upper Silesia, Cracow, Brussels
Comparison of different methods of classification evaluation EV vs Index of Performance R2
Comparison of different methods of classification evaluation EV vs Index of Performance R2
Comparison of different methods of classification evaluation EV vs Index of Performance R2
Comparison of different methods of classification evaluation EV vs Index of Performance R2
Comparison of selected weather types classifications for forecasting the days with high air pollution • Data: • OZONE • AMJJA • 8-hour concentration of ozone (for 17 UTC) • from: • Poland 1997-2002 • central and east monitoring stations: Warszawa IOŚ (urban) and Diabla Góra, Jarczew, Belsk, Zbereże (rural) • south monitoring stations: Zabrze and Katowice (urban), Kuźnia Nieborowska (rural), Kędzierzyn (suburban, industrial), • German 1997-2002 • central and east monitoring stations: • Hoyeswerda (urban), Goerlitz (urban, traffic), Mittelndorf (rural) • Slovakia 1997-1998, 2000 • east monitoring station: • Humenne (urban) • Belgium 1990-2002 • monitoring stations: • Moerkerke and Vezin (rural)
Comparison (EV) of different classifications for ozone domain 7 LWT2 LWT2
Mean ozone concentrations for different types of LWT2 domain 7 Germany Belgium Poland
LWT2 ERA40 Composites Type 4 High ozone concentrations in Germany Type 3 The highest ozone concentrations in Belgium Type 5 High ozone concentrations in Poland
Mean ozone concentrationsfor different types of LWT2 domain 7 Germany Belgium Poland
LWT2 ERA40 Composites Type 22 High ozone concentrations in Poland and Germany, Low ozone concentrations in Belgium Type 13 The highest ozone concentrations in Germany High ozone concentrations in Poland Mean ozon concentrations in Belgium
Comparison (EV)of different classifications for ozone domain 7 LITtc LITtc
Germany Mean ozone concentrations for different types of LITtc domain 7 Belgium Poland
type 12 type 13 The highest values of ozone in Germany, Poland and Belgium The high values of ozone in Poland, middle in Germany, and low in Belgium
Comparison of (EV) different classifications for ozone domain 5,7,8 Diabla Góra, Poland
Comparison (EV) of different classifications for ozone domain 5,7,8 Diabla Góra, Poland
Comparison of(EV) different classifications for ozone domain 5,7,8 Jarczew, Poland
Comparison (EV) of different classifications for ozone domain 5,7,8 Jarczew, Poland
Comparison of (EV) different classifications for ozone domain 7,8 Humenne - Slovakia
Conclusions: • Evaluation of classifications: • WSD and WSD_U parameters are not good for comparing results for different air pollutants. • Normalisation of WSD and WSD_Uparameters is necessary. • By comparing EV, nWSD and nWSD_U with variability of PM10 and SO2 for classification ESLPC30 it is found that nWSD is not good parameter for evaluation classification. • By comparing EV and R2 for DJF (Poland, Belgium - PM10, Romania - TSP) is observed the similar behavior for both parameters. It seems that parameter EV is sometimes better. • The best classifications for winter urban air pollution are: • Classifications with number of classes greater than 15 • Objective classifications: LWT2, LITTC, Sandra, Sandras, • Manual classifications: HBGWL, OGWL and Polish Tc21 classification prepared by Niedźwiedź • The best classifications for summer ozone concentrationsare : • objective classifications: CEC, GWT, LITTc, LWT2, Petisco for all areas (Poland, Slovakia, Germany, Belgium) • objective WLKC733 classifications for Polish stations • objective Sandras classification for Belgian stations • manual Polish Tc21 and Tc11 classifications prepared by Niedźwiedź for Slovak, German and south or central Polish stations • manual ZAMG classification for German and south or central Polish stations • manual HBGWL, OGWL and Perret for Belgian stations • There are not considerable differences between classifications evaluations prepared on the basis of calculations made for different domains • (when sites of stations are at the border areas in different domains).
Thank you for your attention KONTAKT: Tel: 12 6398119 E-mail: zigodlow@cyf-kr.edu.pl zitomasz@cyf-kr.edu.pl IMGW 01-673 Warszawa, ul.: Podleśna 61 tel.: (022) 56 94 000 fax: (022) 00 00 000 kom.: 0 503 000 000 nazwa@imgw.pl www.imgw.pl