Forecasting Ozone in Treasure Valley using CART

Idaho DEQJune 3, 2011

1

Ozone in Treasure Valley

Ozone in Treasure Valley

0

20

40

60

80

100

2001

2002

2003

2004

2005

2006

2007

2008

Year

Ozo

ne

Co

nce

ntr

atio

n

(pp

b)

8hr max

4th Highest

3year Ave 4thHighest

Standard

2

Treasure Valley Ozone Monthly Statistics2001-2008

0

20

40

60

80

100

MAY JUN JUL AUG SEP

Month

Ozo

ne

(pp

b)

Month Average

Ave Month Max

Month Max

3

Forecasting for AQI and CRB

• Daily AQI forecast for public• Daily AQI forecast for residential burning bans (AQI<60

outdoor, AQI <74 all burns)• Forecast for Crop Residual Burning • Regional offices utilize AIRPACT, WRF data, apply various

methods for the forecasting • Need more reliable, easy to use tools

4

CART Model

• Classification And Regression Tree (CART) is a statistical procedure designed to classify data into dissimilar groups.

• CART helps to develop a decision tree to predict pollutant concentrations based predictor variables that are well correlated with pollutant concentrations.

5

Forecasting Methods

Methods Development Effort Operational Effort Accuracy

Persistency Low Low Low

Climatology Low-Moderate Low Low

Criteria Low-Moderate Low Low-Moderate

CART Moderate Low Moderate-high

Regression Moderate Moderate Moderate-high

Neural Networks Moderate-high Moderate Moderate-high

3-D Air Quality

Models

Very High Moderate-high Moderate-high

Phenomenological

/Intuition

High Moderate High

From Guidelines for Developing an Air Quality (Ozone and PM2.5) Forecasting Program. EPA-456/R-03-002 June 20036

Data

• Eight year ozone data (2001-2008)• Eight year meteorological data including

surface data and upper air data: temperature, wind, humidity, pressure, etc.

• WRF forecasting data

7

Correlation TablePositive correlation

No correlation

Negative correlation

8

G=Green 0-59ppbY=Yellow 60-75 ppbO=Orange 76-95ppbR=Red 96-115ppbP-Purple 116-374ppb

CART tree for Treasure Valley Ozone Forecasting (1)Observation Data: Year 2001-2008, May-Aug

9

G=Green 0-59ppbY=Yellow 60-75 ppbO=Orange 76-95ppbR=Red 96-115ppbP-Purple 116-374ppb

CART tree for Treasure Valley Ozone Forecasting (2)Observation Data: Year 2001-2008, May-Aug

10

Performance EvaluationTS Performance  

Total days 79

Observed Green Days 59

Observed Yellow Days 20

forecasted Green 52

Forecasted G for G 48

Forecasted Yellow 21

Forecasted Y for G 4

Forecasted G for Y 4

Forecasted Y for Y 17

Accuracy of Green 81%

Accuracy for Yellow 85%

Over forecasting 7%

Under forecasting 20%11

Performance

Ozone Forecast for Treasure ValleyOzone season 2010

0.0300.0350.0400.0450.0500.0550.0600.0650.0700.0750.080

6/6/

2010

6/8/

2010

6/10

/201

0

6/12

/201

0

6/14

/201

0

6/16

/201

0

6/18

/201

0

6/20

/201

0

6/22

/201

0

6/24

/201

0

6/26

/201

0

6/28

/201

0

6/30

/201

0

7/2/

2010

7/4/

2010

7/6/

2010

7/8/

2010

7/10

/201

0

7/12

/201

0

7/14

/201

0

7/16

/201

0

7/18

/201

0

7/20

/201

0

7/22

/201

0

7/24

/201

0

7/26

/201

0

7/28

/201

0

7/30

/201

0

8/1/

2010

8/3/

2010

8/5/

2010

8/7/

2010

8/9/

2010

8/11

/201

0

8/13

/201

0

8/15

/201

0

8/17

/201

0

8/19

/201

0

8/21

/201

0

8/23

/201

0

8/25

/201

0

8/27

/201

0

8/29

/201

0

8/31

/201

0

9/2/

2010

Date

Ozo

ne

(p

pm

)

Obs Max 8hr Max TS Forecast

12

Source of Errors

• Small changes near the split point may end larger errors.

• Bias in the meteorological forecast.• Emission changes. e.g. Holidays, economy driven

sources.• Boundary conditions. e.g. Stratosphere intrusion (ST)

due to stratosphere-troposphere exchange (STE); long range transport.

13

An unusual Ozone Episode in May 2011Graphs From AIRPACT

Conditions on May 15, 2011 in Treasure Valley:A cold front was reaching the areaMax temperature ~ 53°F, breezy, rainMax 8hour average O3 reached 63ppb in early afternoon, the highest in the month.

14

Monthly average relative humidity was 26.7%Average 6 year Relative humidity in May is 38.4%. (average from 10:00am-6:00pm)

Warm (~80’s °F) and Dry

15

Weekend Effect

16

CART- Limitations

• Requires large set of data, a modest amount of expertise and effort to develop.

• Small changes in predictor variables may produce large changes in the predictions.

• Does not predict unusual events.• Requires periodic updates due to emission

and land use changes.

17

Future Work

• Improve the model for Boise • Experiment for Coeur d’Alene• Explore more parameters• Study for unusual events.

18