1

30 March, 2007

Statistical Prediction ModelStatistical Prediction Modelfor Daily Means of PM10for Daily Means of PM10

Ernst StadloberErnst StadloberSiegfried Siegfried HörmannHörmann, Brigitte Pfeiler, Brigitte Pfeiler

Institute of StatisticsInstitute of StatisticsGraz University of TechnologyGraz University of Technology

2

30 March, 2007

ContentsContents

•• Situation in Graz, Klagenfurt, Situation in Graz, Klagenfurt, BozenBozen•• Meteorological factorsMeteorological factors•• Anthropogenic factorsAnthropogenic factors

•• Statistical Prediction modelsStatistical Prediction models•• Quality of the modelsQuality of the models•• Practical experience in GrazPractical experience in Graz

3

30 March, 2007

The general SituationThe general Situation

•• Unfavorable meteorological conditionsUnfavorable meteorological conditions•• Low wind velocitiesLow wind velocities•• Low precipitationLow precipitation•• Days with temperature inversionDays with temperature inversion

•• Anthropogenic factorsAnthropogenic factors•• Traffic, domestic fuel, industryTraffic, domestic fuel, industry

•• ConsequenceConsequence•• Threshold value exceeded regularlyThreshold value exceeded regularly

4

30 March, 2007

MonthlyMonthly MeanMean valuesvalues of PM10of PM10in in threethree areasareas of Graz of Graz

20032003--20062006

residentialresidential areaarea| | inner inner citycity areaarea| | industrialindustrial areaarea

SimilarSimilar characteristicscharacteristics in Klagenfurt and Bozenin Klagenfurt and Bozen

5

30 March, 2007

Human Human impactimpact

•• Domestic fuelDomestic fuel•• IndustryIndustry

•• TrafficTraffic•• IndicatorIndicator

type of daytype of day

•• Sunday/HolidaySunday/Holiday

Traffic Traffic ≈ ≈ 6060--70%70%

PM10 PM10 ≈ ≈ 7070--80%80% Su/HoSaMo-Fr

80

70

60

50

40

30

20P

M10 (

µg

/m³)

PM10 BZ2 (µg/m³)PM10 KT1 (µg/m³)PM10 GT1 (µg/m³)

GrazGraz KlagenfurtKlagenfurt BozenBozen

6

30 March, 2007

Temperature InversionTemperature Inversion•• Most important factorMost important factor

•• Temperature Temperature on groundon groundlower thanlower than above groundabove ground•• Implies bad air convectionImplies bad air convection

•• Inversion ≈ Inversion ≈ 3030--550% of winter days0% of winter days

•• Graz/Klagenfurt: Graz/Klagenfurt: suitable measuring point suitable measuring point (360m/390m above ground)(360m/390m above ground)

•• BozenBozen: : not suitablenot suitable (1590m above ground)(1590m above ground)

7

30 March, 2007

GrazGraz--WestWestBalloon probe of ZAMG (6 am Balloon probe of ZAMG (6 am –– 8 pm)8 pm)

2

4

6

8

10

12

14

16

18

20

22

24

50

100

150

200

250

300H

oehe

ueb

er G

rund

[m]

0610

h

0700

h

0800

h

0900

h

1000

h

1100

h

1200

h

1300

h

1400

h

1510

h

1600

h

1700

h

1800

h

1900

h

2015

h

0

Zeit MEZ

Temperatur [ °C ] am 17.03.2004in Graz-Gries, Firma Roche

-100

0

10

20

30

40

50

60

70

80

90

100

110

120

50

100

150

200

250

300

350

400

Hoe

he u

eber

Gru

nd [m

]

0610

h

0700

h

0800

h

0900

h

1000

h

1100

h

1200

h

1300

h

1400

h

1510

h

1600

h

1700

h

1800

h

1900

h

2015

h

0

Zeit MEZ

PM10 Konzentration [ µg/m³ ] am 17.03.2004in Graz-Gries, Firma Roche

no data

Inversions- obergrenze [ m ]

8

30 March, 2007

Inversion and Human ImpactInversion and Human ImpactGrazGraz--Mitte:Mitte: 5 Winter seasons (1.10.inter seasons (1.10.––31.3.)31.3.)

No Inversion: No Inversion: 40% 40% lowerlower thanthan InversionInversionSun/Sun/HolidayHoliday:: 30% 30% lowerlower thanthan WorkingWorking DayDay

9

30 March, 2007

Precipitation and WindPrecipitation and Wind

•• Both reduce load of PM10Both reduce load of PM10

•• But in But in GrazGraz and and KlagenfurtKlagenfurt•• low wind velocities low wind velocities •• rare days with precipitationrare days with precipitation

10

30 March, 2007

Statistical Prediction ModelStatistical Prediction Model

• Prediction of PM10 load of next day based on information of current day

• Predictions for Graz, Klagenfurt, Bozen in the winter season

• Models simple and practicable• Reliable but not necessarily precise

11

30 March, 2007

•• Models with 6 parameters (r² ≈ 0.55 Models with 6 parameters (r² ≈ 0.55 –– 0.64)0.64)

•• Metric variableMetric variable• PM10 mean of last 24 hours from 12am-12am

•• Categorical variablesCategorical variables• Type of day (1=Mo-Fr/2=Sa/3=Su-Ho)• Month (1=October/.../6=March)

•• Meteorological forecastsMeteorological forecasts• Precipitation (0=no/1=yes)• Wind velocity (mean of the next day in m/s)• Inversion (temp(ground) - temp(above ground))

• → No traffic frequencies needed

Multiple Linear Regression ModelsMultiple Linear Regression Models√√PM10 mean of the next dayPM10 mean of the next day

12

30 March, 2007

QualityQuality functionfunction

20 40 60 80 100 120 140

20

40

60

80

100

120

140

PredictedPredicted valuesvalues vs. vs. MeasuredMeasured valuesvalues•• QualityQuality of of predictionprediction: 5 : 5 levelslevels

•• greengreen (1=excellent)(1=excellent) to to red (5=very bad)red (5=very bad)

13

30 March, 2007

Graz: Graz: QualityQuality of of PredictionPrediction ((theoreticaltheoretical))

1: 50.6%1: 50.6%, , 2: 29.2%2: 29.2%, , 3: 12.5%3: 12.5% 92.3%92.3%4: 6.0%, 5: 1.8%4: 6.0%, 5: 1.8%

14

30 March, 2007

Klagenfurt: Klagenfurt: QualityQuality of of PredictionPrediction ((theortheor.).)2005/20062005/2006

1: 55.1%1: 55.1%, , 2: 23.3%2: 23.3%, , 3: 10.8%3: 10.8% 89.2%89.2%4: 8.4%, 5: 2.4%4: 8.4%, 5: 2.4%

15

30 March, 2007

Bozen: Bozen: QualityQuality of of PredictionPrediction ((theortheor.).)

1: 62.1%1: 62.1%, , 2: 20.3%2: 20.3%, , 3: 6.0%3: 6.0% 88.4%88.4%4: 8.2%, 5: 3.3%4: 8.2%, 5: 3.3%

2005/20062005/2006

16

30 March, 2007

GrazGraz--MitteMitte: Practical Experience: Practical Experience

• October 16, 2006 – March 15, 2007

• Meteorological forecasts by ZAMG Steiermark in the morning

• Prediction available at 1 pmwebsite as traffic light system

17

30 March, 2007

QualityQuality of of PredictionPrediction withwith forecastforecast

1: 39.0%1: 39.0%, , 2: 19.9%2: 19.9%, , 3: 21.3%3: 21.3% 80.1%80.1%4: 12.5%, 5: 7.4%4: 12.5%, 5: 7.4%

18

30 March, 2007

Model Model withwith meteorologicalmeteorological forecastsforecasts

19

30 March, 2007

PredictionPrediction errorerror

Small Small deviationsdeviations exceptexcept Silvester Silvester effecteffect

20

30 March, 2007

ForecastForecast of Wind in 4 of Wind in 4 categoriescategories

PeriodsPeriods withwith underunder--estimationestimation of wind of wind velocitiesvelocities

21

30 March, 2007

ForecastForecast of of DeltaDelta--TemperatureTemperature

AcceptableAcceptable forecastsforecasts exceptexcept forfor extreme extreme situationssituations

22

30 March, 2007

FC of wind to highFC of wind to highFC of FC of deltadelta--temptemp to highto high

FC of wind to FC of wind to lowlowFC of FC of deltadelta--temptemp to highto high

FC of wind to highFC of wind to highFC of FC of deltadelta--temptemp to to lowlow

FC of wind to FC of wind to lowlowFC of FC of deltadelta--temptemp to to lowlow

PredictionPrediction to highto high

PredictionPrediction to to lowlow

ForecastForecast errorserrors: Wind vs. : Wind vs. DeltaDelta--TemperatureTemperature

23

30 March, 2007

9 9 daysdays withwith predictionprediction PM10PM10 ≥ ≥ 75 75 µgµg/m³/m³

Good Good predictionprediction, , ifif observedobserved valuevalue PM10 > 50 µg/m3PM10 > 50 µg/m3→ 2 → 2 PredictionsPredictions twotwo highhigh

24

30 March, 2007

15 15 daysdays withwith observationobservation PM10PM10 ≥≥ 75 75 µµg/mg/m³³

4 4 daysdays: : PredictionPrediction PM10PM10 >> 75 75 µg/m³µg/m³ (red)(red)9 9 daysdays: : PredictionPrediction PM10 > PM10 > 50 50 µg/m³µg/m³ (orange)(orange)2 2 daysdays: : PredictionPrediction PM10 < PM10 < 50 50 µg/m³ (µg/m³ (greengreen))→ → 2 2 PredictionsPredictions to to lowlow (30.11., 30.1.)(30.11., 30.1.)

25

30 March, 2007

•• Graz Mitte,Graz Mitte, Graz SüdGraz Süd•• Models & Test Models & Test runsruns: 3 : 3 winterwinter seasonsseasons

2004/05 2004/05 –– 2006/072006/07

•• withwith weatherweather forecastforecast ≈ 80% ≈ 80% reliablereliable predictionspredictions

•• Klagenfurt VölkermarkterstraßeKlagenfurt Völkermarkterstraße

•• Model Model withwith knownknown wheatherwheather ≈ 90% ≈ 90% reliablereliable predictionspredictions

•• Bozen BZ5, Meran1Bozen BZ5, Meran1

•• Models Models withwith knownknown wheatherwheather ≈ 90% ≈ 90% reliablereliable predictionspredictions

SummarySummary PredictionPrediction Models Models forfor PM10PM10