Review Articles

Predicting Ozone Levels

Review Articles: Special Issue "Ozone"

Predicting Ozone Levels A Statistical Model for Predicting Ozone Levels in the Shuaiba Industrial Area, Kuwait lSabah A b d u l - W a h a b , 1Walid B o u h a m r a , lHisham Ettouney, 2Bey Sowerby, aBarry D. C r i t t e n d e n i Chemical Engineering Dept., College of Engineering and Petroleum, Kuwait University, EO. Box 5969, Safat 13060, Kuwait 2 School of Chemical Engineering, University of Bath, Claverton Down, Bath BA2 7AY, England Corresponding author:

Dr. Bey Sowerby, tel.: +44q225-826-961; fax: -894; email: [email protected]

Abstract This paper presents a statistical model that is capable of predicting ozone levels from precursor concentrations and meteorological conditions during daylight hours in the Shuaiba Industrial Area (SIA) of Kuwait. The model has been developed from ambient air quality data that was recorded for one year starting from December 1994 using an air pollution mobile monitoring station. The functional re lationship between ozone level and the various independent variables has been determined by using a stepwise multiple regression modelling procedure. The model contains two terms that describe the dependence of ozone on nitrogen oxides (NOx) and nonmethane hydrocarbon precursor concentrations, and other terms that relate to wind direction, wind speed, sulphur dioxide (SOz) and solar energy. In the model, the levels of the precursors are inversely related to ozone concentration, whereas SO2 concentration, wind speed and solar radiation are positively correlated. Typically, 63 % of the variation in ozone levels can be explained by the levels of NOx. The model is shown to be statistically significant and model predictions and experimental observations are shown to be consistent. A detailed analysis of the ozone-temperature relationship is also presented; at temperatures less than 27 ~ there is a positive correlation between temperature and ozone c o n c e n t r a t i o n whereas at temperatures greater than 27 ~ a negative correlation is seen. This is the first time a non-monotonic relationship between ozone levels and temperature has been reported and discussed. Keywords: Ozone levels; statistical model; precursors; meteorological conditions; air quality data

1

Introduction

Interest in the concentration of ozone in the lower atmosphere has increased in recent years because of the effect ozone has on photochemical smog. Photochemical smog is now observed in many urban and industrial centres around the world and it has become a c o m m o n p h e n o m e n o n in many large cities. Intensive studies have been carried out into the nature, frequency of occurrence and duration of those pollutants which give rise to photochemical smog [1 ]. It has been reported that ozone is the most important index

ESPR - Environ. Sci. & Pollut. Res. 3 (4) 195-204 (1996) 9 ecomed publishers, D-86899 Landsberg, Germany

substance of photochemical smog and it has been recognised as one of the major pollutants degrading air quality [2]. H y d r o c a r b o n s (HC) and nitrogen oxides (NOx) have been identified as the two key chemical precursors of ozone and other photochemical oxidants [3]. The m a j o r sources of atmospheric HC's and N O x are the burning of oil, industrial processes and secondary photochemical production. These anthropogenic sources are responsible for more than 95 % of the ozone concentrations in the lower a t m o s p h e r e [4]. M a n y studies reported in the literature have examined the relationship between the levels of ozone and its precursors [1-3,5]. In a USA Environmental Protection Agency (EPA) study, an observational hypothesis a p p r o a c h was used [5]. The basic assumption in this a p p r o a c h is that early morning H C and N O x levels are indicators of the o x i d a n t levels that will occur later in the day. Specifically the 0 6 0 0 - 0 9 0 0 hour H C and N O x levels were c o m p a r e d with the daily 1-hr m a x i m u m o x i d a n t levels that n o r m a l l y occur between 1000 and 1400 hours. This w o r k led to the development of graphical Empirical Kinetic M o d e l l i n g A p p r o a c h (EKMA) diagrams. These diagrams provide a basis for relating maxi m u m h o u r l y average o z o n e c o n c e n t r a t i o n s to the 0 6 0 0 - 0 9 0 0 hour concentrations of n o n - m e t h a n e hydrocarbons ( N M H C ) and N O x . Originally E K M A diagrams were designed as a guidance tool for deciding whether N O x reductions w o u l d have a positive or negative effect on ozone concentrations and air quality. The study using E K M A diagrams concluded that within certain limits of ozone concentrations and N M H C to N O x ratios, ozone concentration or its production rate is a p p r o x i m a t e l y a linear function of its precursor concentrations [6]. Statistical-empirical m o d e l s r e p r e s e n t an a l t e r n a t i v e app r o a c h for expressing the O 3 - H C - N O x r e l a t i o n s h i p [7]. These models are constructed from a m b i e n t data and are statistical in nature. In one statistical model, ozone levels have been correlated as a function of T850 (temperature in ~ at 850 m b a r levels). It w a s found that the t e m p e r a t u r e

195

Predicting Ozone Levels

Review Articles

at 850 mbar correlates well with ozone levels in the south coast air basin of California and that the 850 mbar temperature can be used to explain the monthly variation of basin-wide ozone levels [7]. Air pollution is a serious problem in many areas in the world and the statistical assessment of air quality data and dispersion models are useful tools for air pollution studies. The mathematical models used in air pollution studies have been classified into two groups [8]: 1. Models which depend on the statistical analysis of previous air quality data, and 2. Models which depend on theories related to chemical processes and atmospheric movements. This paper presents a statistical model that has been developed using ambient air concentration data collected during

daylight hours (1000-1700 hours). It can be used to predict ozone levels from precursor concentrations and meteorological conditions. 2

Experimental Method

Collection of Ambient Air Quality Data Ambient air quality data has been recorded every five minutes for one year, starting in December 1994, using an air pollution mobile monitoring station located in the Shuaiba Industrial Area (SIA) in Kuwait. The site of the mobile laboratory at SIA, along with the major industrialised plants, is shown in Figure 1. The location of the SIA relative to Kuwait City is also shown in Figure 1. The SIA comprises several large industries which include three petroleum re-

Kuwait City

N

T

Wt

~ESE t

WSW

.NE

-r

'

~'~"~-

'

1

l

4E

Arabian Gulf

I S

Air

J

Power plant

Cement crushing

North (not

!

working) KNPC Mine AI-Ahmadi refinery

Power plant South

]

2 Om Sea

KNPC Mine AI- J I PIC plant A

ShuA|be Refinery

.,,,=-..-,~.~ _~

construction

KNPC Mine Abdullah 2000m

Refinery

,oo; :~ . Scale (km)o i i 0 2.25 4.5

Fig. 1"

196

Location of SIA relation to Kuwait City and the location of the mobile laboratory along with the major industrial plants in SIA

ESPR - Environ. Sci. & Pollut. Res. 3 (4) 1996

Review Articles

Predicting Ozone Levels

fineries, a complex of petrochemical plants, a cement plant, a chlorine and soda factory, two desalination and power plants, a commercial harbour and oil loading terminals. The location of the mobile laboratory was selected on the basis of several criteria: availability of power, security and topography of the area. Care was taken to ensure that no trees or buildings were in the immediate vicinity of the mobile laboratory. It is important to realise that, apart from pollution arising from general road transport and air conditioning in the Kuwait City area, pollution sources in Kuwait are more or less concentrated in the SIA. This is somewhat different from the situation commonly found in most European and American industrial areas where the pollution arising from the surrounding cities is significant. The wind rose for December 1994 to November 1995 is superimposed on Figure 1. This wind rose shows that the majority of the prevailing winds are from W and WNW and that the second prevailing wind direction is ESE. Thus if the level of pollution from Kuwait City were significant, it would not be blown towards SIA and it would not affect the analysis presented in this paper. The effect of wind direction on the distribution of pollutants in the SIA will be discussed in detail in an imminent publication [9] where it will be shown that the major sources of NOx, SO2, NMHC, CO and suspended dust are located to the NW of the monitoring station (this direction coincides with a petroleum refinery) and a less significant source of NOx and SO 2 is located due SSW and SW of the monitoring station (again the direction coincides with a petroleum refinery). Details of the mobile laboratory's chemical sensors and meteorological instruments are given in Tables 1 and 2 respectively. In terms of its operation, the mobile laboratory Table 1:

is characterised by the following: sampling inlets were located on top of the laboratory 10 m above the ground; all the monitors were controlled by an intelligent data logger; automatic zero and span calibrations were performed using a calibration gas once every 23 hours (thus the same hourly data was not lost each day); Envicom software was used to record the data and then Envaid software was used to edit and process the data. A quality check of the data was performed by examining all data in graphical form. The measurements of chemical concentrations and meteorological conditions were recorded every 5 minutes. These 5 minute data have been used to determine the variations of ozone with the other pollutants and with meteorological parameters. Correlations of the levels of these various parameters relative to each other were then obtained. For examining trends in ambient ozone, scatter plots of ozone as a function of some of the other variables were constructed. Multiple regression modelling, using the stepwise procedure, has been used to determine functional relationships between ozone and the other variables. 3

E x p e r i m e n t a l Results

Figures 2 and 3 (pp. 198, 199) show the mean distribution of the various pollutants and meteorological parameters, respectively, according to the hour of the day. The means have been computed from the 5 minute interval data collected using the mobile laboratory. Figure 2 shows large diurnal variations in the concentrations of many of the pollutants. Previous studies have also reported large diurnal variations [10-13]. The measurements recorded for the SIA show two types of variation. Firstly, diurnal variations with a single maximum occur for the secondary pollutants N O 2

Details of the chemical sensors located in the mobile laboratory

Species

Method

Instrument make & model

Measurement range

Accuracy

hydrocarbons

flame ionisation detector

MSA/Baseline model 1030A

0-20 ppm

+ 0.2 ppm

carbon monoxide

non-dispersive infrared

Thermo Environmental Instruments Inc model 48

0-20 ppm

+ 0.1 ppm

carbon dioxide

non-dispersive infrared

Thermo Environmental Instruments Inc model 41H

0-1000 ppm

+ 1%

NO and NOx

chemiluminance

Thermo Environmental Instruments Inc model 42

0-1000 ppb

_+0.5 ppb

sulphur dioxide

fluorescence

Thermo Environmental Instruments Inc model 43A

0-2000 ppb

+ 1 ppb

ozone

non-dispersive ultraviolet photometer

Measurement Controls Corp. model ML9812

0-1000 ppb

_+ 1 ppb

suspended dust (PM-10)

gravimetric method

Rupprecht & Patashnick co., Inc. model TEOM series 1400a

0-5000 #g/m 3

+ 5 #g/m 3

Table 2:

Details of the meteorological parameters recorded by the mobile laboratory

Parameter

Technique

Instrument & model

Range

Accuracy

wind speed

speedometer with a reed switch

Aanderaa Sensor model 2740

0.4-76.7 m/s

+ 0.2 m/s

wind direction

compass with a potentiometer

Aanderaa Sensor model 2750

0-360 degrees

_+5 degrees

solar energy

high sensitive thermistor bridge

Aanderaa Sensor model 2770

0-2000 W/m 2

__ 20 W/m 2

temperature

ohmic half-bridge principle

Aanderaa Sensor model 3455

0-60 ~

__.0.1% range

relative humidity

capacitive polymer

Aanderaa Sensor model 3445

0-100 %

_+3 % RH

air pressure

silicon chip as a sensing element

Aanderaa Sensor model 2810

918-1092 mbar

_+0.2 mbar

ESPR- Environ. Sci. & Pollut. Res. 3 (4) 1996

197

Predicting O z o n e Levels

R e v i e w Articles

Methane

Nitrogen

Dioxide

30

30 r-,

2.8

2s .o

26

8

~E

2 94

8 22 8

20

8 8

15

~3 [

20

l

2

I

4

I

6

~

8

I

i

I

lO 12 14 Time (hour)

l

16

~

18

20

t

I

I

t

2

4

6

8

10

1

0

22

Non-methane

0.30

I

0.25

~.

4s

._~

4 0

~.

35

I

1

10 12 14 Time (hour)

I

I

I

I

16

18

20

22

NOx

5O

E

P,

'~

w 30 8

o

~

o

0.20 0

I

I

I

I

I

2

4

6

8

I

10 12

I

I

I

t

I

14

16

18

20

22

25

I

t

I

I

2

4

6

8

20 0

I

Time (hour) Carbon

Sulphur

Monoxide

I

10 Time

t

12 14 (hour)

I

I

I

I

16

18

20

22

dioxide

80

0~7 E

8

0.6

=r

~o 8

0

6o

0.5

8 0.4

I

I

I

I

I

I

I

I

I

I

I

2

4

6

8

10

12

14

16

18

20

22

20 0

I

I

I

I

2

4

6

8

Time (hour) Carbon

I

I

I

I

I

10 12 14 Time (hour)

I

I

16

18

20

22

Ozone

Dioxide 3O

400 E

Q. t-, v

.~

390

~

g

20

o~

380

c

10

o 370 0

I

I

I

I

2

4

6

8

I

I

I

10 12 14 Time (hour)

Nitrogen

I

I

I

I

16

18

20

22

~

0

I

I

I

I

2

4

6

8

I

I

I

I

I

10 12 14 Time (hour)

I

I

16

18

20

22

Suspended

Oxide

dust

~.-,200 E

30 O.. v c o

0

---~150

8

2o

c

8 loo

8

c o

o

o

o

10

I

I

I

I

2

4

6

8

I

I

I

10 12 14 Time (hour)

I

I

I

I

16

18

20

22

50 0

I

I

I

I

2

4

6

8

I

I

I

10 12 14 Time (hour)

I

I

I

16

18

20

I 22

Fig. 2: Average annual diurnal variation of air pollutant concentrations in SIA for the period from December 1994 to N o v e m b e r 1995

1 98

E S P R - Environ. Sci. & Pollut. Res. 3 (4) 1996

Review Articles

Predicting Ozone Levels

Wind speed

5.0

1.0

Solar energy

~4.5

"~4.o

0.4 0.2

3.5

0.0 , - ~ ~

3.0 0 2 4 6 8 10121416182022 Time (hour)

0

2

6

8 10121416182022

Time (hour)

Air temperature

30 28 26 24 22 20

~:+-~

. . . . . . 4

Atmospheric pressure 994 992 v~E

990 988

0

2

4

6

8 10121416182022

0 2

Time (hour)

4

6 8 10121416182022 Time (hour)

Relative hum idity

50 45 40 35 0 2 4 6 8 10121416182022 Time (hour)

Fig. 3: Averageannual diurnal variation of meteorologicalvariables in S1Afor the period from December 1994 to November 1995

and 03, that is for the products of photochemical reactions, and the primary pollutant NOx. Secondly, diurnal variations with double maxima are associated mainly with primary pollutants such as suspended dust, SO2, NO, CO and hydrocarbons. Again this is in agreement with other studies [10-13]. Conditions that foster ozone episodes are of interest in this study. Therefore the analysis is concentrated on the observations recorded during the daylight hours (1000-1700 hours), selected because during this period of time higher mean concentrations of 0 3 were recorded. Statistical analysis of the 5 minute data collected using the mobile laboratory was performed using the data analysis software, SPSS (Statistical Package Social Science). Conventional statistical data manipulation has been used to calculate the coefficients of multiple regression. Descriptives for each of the parameters considered are listed in Table 3. Emphasis is placed on the variations of 0 3 with the other pollutants and with the meteorological conditions. Table 4 shows the correlation matrix that has been determined for the various parameters that have been measured and the O 3 level. It should be noted that no causal relationship has been presumed. The last column in Table 4 shows the tendency of the various parameters to change with the change in ozone level. Those parameters where a significant correlation exists have been highlighted in Table 4. The correlation coefficients between ozone and CH4, CO, CO2, dust, NMHC, NO, NO2, NOx, SO2, temperature, wind direction and wind speed are negative. However, with relative humidity and solar energy positive correlations are seen. The most significant correlations are ESPR- Environ. $ci. & Pollut. Res. 3 (4) 1996

Table 3:

Descriptives for the various variables measured in SIA during the daylight hours (1000 1700h) for the period from December 1994 to November 1995

Variable

Mean Value

Minimum

Maximum

value

value

Number of recorded, data pairs

non-methane hydrocarbon

0.16 ppm

0.0

5.81

12,891

solar energy

0.47 kW/m 2

0.0

1.19

20,369

aarbon dioxide

0.52 ppm

0.0

6.99

16,668

methane

2.22 ppm

1.21

20.47

13,063

total hydrocarbon

2.38 ppm

1.21

24.16

13,016

wind speed

4.31 m/s

0.32

13.56

20,345

NO

15.79 ppb

0.0

378.0

14,454

NO2

18.00 ppb

0.0

120.0

14,943

3zone

22.15 ppb

0.0

102.5

14,123

temperature

25.7 ~

7.48

47.34

20,367

relative ~urnidity

47.03 %

2.6

92.75

20,369

~ulphur dioxide

51.97 ppb

0.0

972.5

16,636

~uspended :lust

123.16 #g/m 3 0.0

3225.0

17,341

Nind direction

166.98 deg.

357.5

20,367

:arbon

393.15 ppm

674.0

12,863

!4.32 301.0

dioxide

199

Predicting Ozone Levels

Review Articles

with N O 2 and N O x and slightly less significant with NO. The negative coefficient of correlation with N M H C denotes that a rise in 03 concentration was associated with a drop in the level of NMHC. The correlation with solar energy was generally poor. The slightly negative overall correlation of 03 with temperature was surprising and this is rather interesting as previous studies have found positive correlations [14-18]. However, as described later, the experimental relationship between ozone and temperature at SIA does not monotonically increase nor monotonically decrease with temperature. The correlation between N O and NO2, also shown in Table 4, is rather interesting. It is positive and quite significant denoting that the increase in the level of NO is associated with an increase in the level of NO 2. The predominance of thermal inversion in the SIA and the high level of NO could cause this unusual positive correlation between NO and NO 2. The correlation coefficient between wind speed and all other variables (except relative humidity) was positive. Clearly this does not reflect the dissipating role of stronger winds. Rather, in the SIA, stronger winds carry the pollutants from the industrial stacks to the mobile laboratory's sampling point as already discussed. Figure 4 shows scatter plots of ozone as a function of some of the more important parameters; temperature, N O x concentration, N M H C concentration and wind speed. These figures are useful for more detailed comparisons and for examining trends in ambient ozone. The ozone-temperature scatter diagram shown in Figure 4(a) has been used to evaluate, quantify and gain an insight into the ozone-temperature relationship in order to help provide an explanation for the negative correlation already mentioned. Two separate groups of points appear in this plot. The first group (for which the temperature is less than 27 ~ indicates a positive correlation between ozone and temperature with a correlation coefficient of 0.424. In contrast, the second group (for which the temperature is equal to or greater Table 4:

3H 4 30

D e v e l o p m e n t of the Statistical M o d e l

Ozone is a secondary pollutant and its ambient air concentration is influenced by two independent factors: emission rates of the primary pollutants and meteorological conditions. It is also well known that the ozone production rate is not simply a linear function of its precursor concentrations. Therefore in this work, multiple linear regression modelling has been performed in order to find the predictive equation for ozone concentration as a function of various parameters that were being measured at the same time each day by the mobile monitoring laboratory. The independent parameters (predictors) that have been studied are NOx, N M H C , CH4, CO, CO2, SO2, wind speed, wind direction, temperature, relative humidity, solar energy and suspended dust. The dependent or response variable is ozone concentration. Owing to the cross correlation between the independent variables (as shown in Table 4), a stepwise multiple regression procedure has been used. This procedure is one of the many multiple regression analysis methods available in SPSS. It is a combination of forward selection and backward elimination multiple regression procedures. The procedure used automatically selects parameters that are of most importance and elimi-

CH 4

CO

CO 2

Dust

NMHC

NO

NO 2

NO 3

RH

1

0.115 1 -

0.359 0.360

0.292 -0.060

0.275 0.260

0.521 0.500

0.473 0.428

0.553 0.512

-0.368 -0.158

0.510 0.419

0.012 -0.026

1 -

0,190 1 -

0.259 0.291 1

0.497 0.272 0.249 1

0.428 0.37.9 0.291

0.512 0.337 0.288

-0.158 -0.447 I-0.352

0.419 0.399 0.310

0.666 1

0.959 0.850

-0.359 ! --0.584

0.472 0.608 0.565

-0.026 0.125 -0.129 0.009 0.079 0.037 -0.325 0.023 1

-

NO 2 NO x RH SO 2 Solar Temp WD ~VS

4

The correlation matrix of the various variables in the SIA for 1000 to 1700h for the period from December 1994 to November 1995

302 Dust NMHC NO

than 27 ~ shows a negative correlation, with a correlation coefficient of -0.523. Table 5 gives a summary of the relationships between ozone and season/temperature that have been published in the literature for various regions around the world and, when given, the range of conditions studied. This shows that other workers have found that the level of ozone increases with rising temperature and that no other workers have differentiated between low and high temperatures; this could have been because the studies to date have been carried out in more temperate regions than Kuwait. To date, no publication has reported the change reported in this study of Kuwait in the correlation between ozone and temperature as the temperature rises above 27 ~

-

. . . . . . .

. . . . . . .

.

. .

. .

. .

. . . .

. .

. .

. . .

1 .

.

. .

. .

.

-0.479 1

. . . .

~)3

. .

. .

.

.

SO 2

-0.491 1

Solar

Temp

WD

0.400

0.350 0,270 0.270 0.397

03

0.253 0.102 0.102

-0.330 -0.267 -0.267

0.308 0.367 0.558 0.473 -0.659 0.535

0.566 0.198 0.250 0.304 0.289 -0.408 0.349

--0.284 -0.455 -0.495 -0.582 -0.590 0.376 -0.366

0,059 0.290

0.120 0.297

0.071 -0.172

0.466

-0.483 -0.166 1

-0.094 -0.094

0.468 -0.101 0.250 0.353 0.314 :-0.632 0.265 0.397

WS

1 1 -

1 -

where

NMHC is non-methane hydrocarbon, ppm; Solar is solar energy, W/m 2 ; THC is total hydrocarbon, ppm; WS is wind speed, m/s; WD is wind direction, degrees; RH is relative humidity, %; Dust is suspended particles ~g/m 3

highlighted numbers denote a significantcorrelation

200

ESPR - Environ. Sci. & Pollut. Res. 3 (4) 1996

Review Articles

~& ~, ~ c |{3

Predicting Ozone Levels

120 100

80

(a) C3-

80 6O 40 tO 2O o~ 0' -2(" 0

&

60

0

40

g g 8

d 10

20

30

40

(b)

20 0' -2(" -100

50

0

100

Temperature (%)

.-~

300

400

(c)

(d) {D_

8O

&

60'

g

4e

Em

8

2O

d

O' "2C .5

500

NOx concentration (ppb)

100 r~

200

o to

o

o

0.0

.5

1.0

8 G

o

1.5

2.0

0

2

4

NMHC concentration (ppm)

6 8 Wind speed (m/s)

10

12

14

Fig. 4: Scatter plots of ozone concentration as a function of temperature, N O x concentration, N M H C concentration and wind speed Table 5:

S u m m a r y of ozone and temperature/season correlations reported in the literature

Author

Country Region

Finding/correlation ] Reference with ozone source

SALOP et al

USA

SE Virginia

ozone level positively correlated with temperature

14

MASSAMBANI

; Brazil

Sao Paulo

Highest ozone levels in summer (Jan - March).

15

Pisa

Ozone level 16 positively correlated with temperature & solar radiaton. Ozone levels: maximum in summer, minimum in winter.

and ANDRADE LORENZINI et al

VARSHNEY

Italy

India

Delhi

and AGGARWAL

BOWER et al

UK

Stevenage, London, Lullington etc.

Ozone levels high in summer (max T 46 ~ and low in winter (min T 1 ~

where the concentration of SO 2 is in ppb and solar is the solar energy in kW/m 2.

Ozone levels higher in summer

Figure 5 shows a plot of R 2 for the model according to the independent variables which are included. The first parameter, NOx concentration, describes 63 % of the variation in ozone. The R 2 increases markedly when the number of parameters is increased from 1 to 2, that is when N M H C concentration is added. Adding more variables leads to only small further increases m R . It should be noted that in this work no attempt was made to separate the effects of solar radiation (UV) and solar energy (temperature) on ozone formation. 9

nates those that are of least importance. The expression that best correlates ozone concentration with the other parameters was determined to be as follows : (1)

where [i] is the concentration of parameter i. For 0 3 and NOx the concentration is in ppb, for N M H C and CO the

ESPR - Environ. Sci. & Pollut. Res. 3 (4) 1996

Statistical evaluation of this model using the 6 variables, [NOx], WD, [NMHC], WS, log[CO] and T yields a coefficient of determination (R2) of only 0.56. This means that only 56 % of the variations in ozone data can be explained by the variations of the independent variables used in equation 1. Transformation of the dependent variable data has been carried out and many different formulae have been used in an attempt to increase R 2. The same approach was used with the independent variables. It was found that using multiple linear regression of log[O3] instead of [03] increases R 2 from 0.56 to 0.753. The result of the last step of the stepwise multiple regression analysis with log[O3] as the dependent variable is as follows: log [03] = 1.543-0.00585 [NOx]-0.2599 [NMHC]-O.O005392 WD + 0.0162 WS + 0.0003132 [SO2] - 0.038 log[solar] (2)

17

[03] = 31.0597-0.1177 [NOx]-O.0316 WD-13.086 [NMHC] +0.9587 WS-6.577 Log [COl-0.138 T

concentration is in ppm. Wind direction (WD) is in degrees, wind speed (WS) is in rrds and temperature (T) is in ~

5

.

2

Discussion

The empirical ozone model shown in equation 2 contains three concentration terms (NOx, N M H C and SO2) and

201

Predicting Ozone Levels

Review Articles

00t

i

0.75

7= 0.70 0.65 0.60

0

I 1

I 2

I 3

I 4

I 5

t 6

t 7

I 8

I 9

10

tween the measured and the calculated values. However this in itself does not provide a quantitative assessment of the agreement between the observation and theory since the calculated value is itself evaluated using the experimentally observed data. A strong correlation does suggest that most of the data being examined have similar photochemical characteristics. A weak correlation, on the other hand, could have meant that large random errors exist [24].

No. of independent variables

Fig. 5:

Changem the coefficient of determination, R2, as independent variables are added to the regression

1.8 j

1.6"

three other meteorological condition terms (wind direction, wind speed, and solar energy). The relationship between ozone concentration and these 6 independent variables can be explained on theoretical grounds as follows: 9 N O x and N M H C have been identified by many investigators as the two key chemical precursors that produce ozone in the presence of solar energy [3-5,7]. In addition, many controlled studies that have used environmental smog chambers have shown that ozone formation is dependent on the concentrations of its precursors N O x and N M H C as well as light intensity [19-21]. 9 Ozone concentration is dependent on wind speed and direction. These two parameters characterise mechanical turbulence and dilution/concentration of atmospheric concentrations. 9 The significance of SO 2 on ozone concentration is interesting. In Table 2 a negative correlation exists between ozone and SO 2. However in equation 2 a small positive correlation is shown. This indicates that on a one-to-one basis SO 2 is positively correlated with ozone but when many parameters are considered a small negative correlation is found. In a study of the relationship between ambient air SO 2 and 03 concentrations using air monitoring data from the Delmarva Peninsular on the eastern shore of Maryland, USA, a linear relationship with a positive correlation between 0 3 and SO 2 existed [22]. It was suggested that theoretical atmospheric chemistry supports this relationship; when SO 2 or 03 concentrations are high, SO 2 absorbs light and forms an excited electronic state of molecular oxygen which readily combines with O-atoms to form 03 . It is important to note that this theory is only applicable to the daytime levels of SO 2 and 03, and so in the absence of solar radiation, the explanation given above might be invalid. 6

C a l i b r a t i o n o f the M o d e l

The observed and predicted values of log [03] are shown in Figure 6 which shows that the model prediction and experimental observations are consistent. It is important to note that the derived model is only applicable during daylight hours (1000-1700 hours). Linear regression analyses were performed on the predicted and observed values of log [03]. The results of this analysis are shown in Figure 7. This figure shows that the calculated and measured values cluster around the diagonal which indicates the validity of the derived model. Thus there is a strong correlation be-

202

1.4; 1.2:

0

_o 1.~ 2~

•@•

'~'Log ozone

.8"

.6

1;4 1'8

22o 3o

9 Predicted value

Time (day)

Fig. 6:

Observed and predicted log(O3) concentrations during daylight hours

-o 0..

O Q

-.5

0.0

.5

1.0

1.5

2.0

Log ozone (ppb)

Fig. 7:

7

Predicted values of log (03) vs the observed values during daylight hours

Test of the A s s u m p t i o n s o f the M o d e l

The assumptions that are made when a statistical model is developed may be summarised as follows [24, 25]: 9 the distribution of the error should be normal 9 the model should be able to predict the observed concentration maxima 9 the residuals should be normally distributed 9 there should be a constant distribution of variance. These assumptions are discussed in the following subsections. 7.1

Distribution of residuals

In many respects, statistical analysis of the deviation between predictions and observations is the heart of the model performance evaluation [26]. It has been reported that although raw statistical comparison of observed and predicted values may not reveal the cause of discrepancies, it can tell much about the nature of the mismatch [26]. In the present study of Kuwait considerable attention has been given to the statistical measures for comparing predicted and observed air pollution concentrations. Figure 8 shows the frequency distributions of the residuals, (residual ESPR - Environ. Sci. & Pollut. Res. 3 (4) 1996

Review Articles

Predicting Ozone Levels

800'

1.0

.8

~, 600. g

g- 400" LL 200'

.6

r

J

-3.3

-1".8

.4'

g Std. Dev = 1.00 Mean = .0 N = 6089.00 -.3

1.3

/

.2

Lid

o Normal Line

0.C 0.0

2.8

gill " * zresid

.2

.74

.6

.8

1.0

Observered cum (%)

Regression standardized residual

Fig. 8: Histogram of concentration residuals for ozone during daylight hours

Fig. 10: Normal plot of regression standardised residual of ozone .6

is the observed concentration minus the predicted concentration) for ozone over the whole period of study during the daylight hours (1000-1700 hours). It can be seen that the frequency distribution of residuals appears to be normal. 7.2

1.4' g o 1.3

o

9 LOGO3 1 1.1

10.00

,

12'.00

I

14'.00

'

9 PRE 1 1

16'.00

Time (hour)

7.3

Mean and predicted logarithmic concentrations of ozone according to daylight hours of the day

Normally distributed residuals

The normal probability plot for the residuals of the multiple linear regression fit of the empirical 03 model should be a straight line. Figure 10 confirms this to be the case. 7.4

Constant variance distribution

A plot of the residuals against the predicted values of log[O3] yields roughly a horizontal band of points. This is consistent with the assumption that the residuals have comparable variances. When Figure 11 is examined, this test seen to be valid. ESPR - Environ. Sci. & Pollut. Res. 3 (4) 1996

o.

.2

o

~

~o o o _ eL

o oooo

"O

rr

q'~

0.0 o

-.2

Predicted and observed concentration maxima

An important criterion in evaluating an air pollution model is its ability to predict the observed concentration maxima [26]. Figure 9 shows a comparison of the magnitude of the mean predicted and observed ozone concentrations for the hours under study. It is clear from this Figure that the log of the observed maximum ozone concentration was 1.32 whereas the predicted value was 1.34. Both occurred at about 1400 hours and so the predicted hour of occurrence of the ozone maximum agrees with that observed.

Fig. 9:

.4

o

.5

~

o

%

oo

0.0

o o"

,

.5

o~_oo. % o ~_~oo

~

1.0

~

1.5

2.0

Predicted value of log ozone (ppb)

Fig. 11: Scatter plot between residual and predicted value of log(O3) according to daylight hours of the day

8

Conclusions

Air quality data for SIA during daylight hours (1000-1700 hours) for one year starting from December 1994 has been used to develop a statistical model to describe the dependency of ozone on its precursors as well as on the meteorological conditions. This model was obtained by multiple linear regression analysis. The model contains two terms to fit the dependence of O 3 on its precursor concentrations NOx and NMHC, one term to describe the dependence of O 3 on another independent pollutant SO2, and three other meteorological terms (wind direction, wind speed, and solar energy). According to this model, the levels of the precursors are inversely related to ozone concentration, whereas SO 2 concentration, wind speed and solar radiation are positively correlated. 63 % of the variation in ozone levels can be explained by NOx. Inclusion of the N M H C concentration term improves the model by increasing the R 2 value from 0.629 to 0.704. However inclusion of wind direction, wind speed, SO 2 and solar energy terms only improves the model by a further 5 %; R 2 increases from 0.704 to 0.753. The findings are attributable only to the period that was selected for the study. During the period 1000 and 1700 hours, the range of variability in wind speed, wind direction and solar energy was limited, leaving a greater role for the precursors. The derived model is statistically significant and the model predictions and experimental observations are consistent. Therefore, the model can be used to predict the levels of ozone expected in SIA during the daylight hours (1000-I 700 hours) when the values of the precursors, SO2, solar energy, wind speed and direction are known. The model will be useful in other locations that

203

Predicting Ozone Levels have similar pollution sources and weather conditions. It may also be possible to extend the usefulness of the model presented in this paper by applying it to, and validating it with, data collected in other industrial areas. The ozone temperature relationship for SIA in Kuwait has been found to be particularly interesting; at temperatures less than 27 ~ there is a positive correlation between temperature and ozone concentration whereas at temperatures greater than 27 ~ a negative correlation is seen. In the ozone model presented in this paper, temperature was not a particularly important parameter. Any future study should be directed towards determining a theoretical insight into this interesting ozone-temperature relationship. For example, a multiple regression analysis method could be used twice; once for temperatures below 27 ~ and again for temperatures above 27 ~ The results of the developed models for the two cases could then be examined. The effect of solar energy (UV) should also be elaborated in future studies. The developed model could be used as a guidance tool for the purpose of designing pollution control strategies. It can be used for estimating the impacts of different N M H C and N O x emission rates, together with local meteorological conditions and concentrations of other pollutants (namely SO2) , for reducing the tropospheric ozone concentrations in SIA. Different scenarios could be envisaged and simulated with the model with controls on just NOx, just NMHC, or on a combination of N O x and NMHC, and under different meteorological conditions. Such studies would help to identify the most effective and beneficial methods of reducing ozone concentrations in this industrialised area.

9

References

[1] BOUCHER,K. (1991). The monitoring of air pollutants in Athens with particular reference to nitrogen dioxide, Energy and Buildings, 15-16, 637-645 [2] Xu, J., and ZHU, Y. (1994). Some charateristics of ozone concentrations and their relations with meteorological factors in Shamghai, Atmospheric Environment, 20, 3387-3392 [3] HAAGEN-SMITH,A.J. (1952). Chemistry and Physiology of Los Angeles Smog, Ind.Eng.Chem, 44, 1342-1346 [4] SEINFELD,J. (1986). Atmospheric Chemistry and Physics of Air Pollution, Wiley, New York. [5] Environmental Protection Agency, EPA, (1971). Air Quality Criteria for Nitrogen Oxides, Washington, D.C. [6] K1NOSIAN,J.R. (1982). Ozone-precursor relationships from EKMA diagrams, Environ.Sci.Technol, 16(12), 880-883 [7] KUNTASAL,G., and CHANG, T.Y. (1987). Trends and relationships of 03, NOx and HC in the south coast air basin of California, J. Air Poll. Contr. Associ., 37,1158-1163

Review Articles [8] ToPcu, N., KESKINLER, B., BAYRAMOGLU,M., and AKCAY, M. (1993). Air pollution modelling in Erzurum city, Environmental Pollution, 79, 9-13 [9] ABDUL-WAHAB,S., BOUHAMRA,W., Eq'TOUNEY,H.M., 8OWERBY,B., CRITTENDEN,B.D.. Trends and analysis of air pollution at Shuiaba industrial area in Kuwait, paper in preparation [10] VALEROSO,I.I., MONTEVERDE,C.A., and ESTOQUE,M.A. (1992). Diurnal variations of air pollution over metropolitan Manila, Atmosfera, 5, 241-257 [11] BIZJAK,M., BENNER,W.H., HANSEN,D.A., HRCEK,D., HUDNIK,V., and NOVAKOV, T. (1988). Spatial and temporal variations of aerosol sulphate and trace elements in a source-dominated urban environment, Atmospheric Environment, 22(12), 2851-2862 [12] GARCIA,B.A., FERNANDEZDIAE, J.M., RUIPEREZ,L.G., and LOPEZ, A. (1988). Concentrations, sources and particle size distribution of the atmospheric aerosol of Oviedo urban nucleous (Spain), Atmospheric Environment, 22(12), 2963-2969 [13] STEVENS,C.S. (1987). Ozone formation in the greater Johannesburg region, Atmospheric Environment, 21(3), 523-530 [14] 8ALOP,J., WAKELYN, N.T., LEVY, G.E, MIDDLETON, E.M., and GERVIN,J.C. (1983). The application of forest classification from landsat data as a basis for natural hydrocarbon emission estimation and photochemical oxidant model simulations in south eastern Virginia, J. Air. Poll. Contr. Associ., 33, 17-22 [15] MASSAMBANI,O., and ANDRADE,E (1994). Seasonal behavior of tropospheric ozone in the SaG Paulo (Brazil) metropolitan area, Atmospheric Environment, 28(19), 3165-3169 [161 LORENZINI,G., NALI, C., and PANICUCCI,A. (1994). Surface ozone in Pisa (Italy) : A six-year study, Atmospheric Environment, 28(19), 3155-3164 [17] VARSHNEY,C.K., and AGGARWAL,M. (1992). Ozone pollution in the urban atmosphere of Delhi, Atmospheric Environment, 26B(3), 291-294 [18] BOWER,J.S., BROUGHTON,G.E, DANDO, M.T., STEVENSON,K.J., LAMPERT,J.E., SWEENE'r B.P., PARKER,V.J., DRrVER, G.S., CLARK, A.G., WADDON, C.J., WOOD, A.J., and WILLIAMS,M.L. (1989). Surface ozone concentrations in the U.K. in 1987-1988, Atmospheric Environment, 23(9), 2003-2016 [19] GLASSON,W.A., and TUESDAY,C.S. (1970). Hydrocarbon reactivities in the atmosphere photooxidation of nitric oxide, Envir. Sci. Technol, 4, 916-924 [20] WILSON, K.W., and DOYLE, G.J. (1970). Investigation of photochemical reactivities of organic solvents: final report, SRI Project, PSU-8029, Stanford Research Institute, Irvine, California [21] LAITY,J.L., BURSTEIN,EG. and APPLE,B.R. (1973). Photochemical smog and the atmospheric reactions of solvents: In Solvents Theory and Practice, Adv. Chem. Series 124, 95-112 [22] GUPTA,G., 8ABARATNAM,S., and DADSON, R. (1986). Linear regression analyses of ozone and sulphur dioxide in ambient air, The Science of the Total Environment, 50, 209-215 [23] JACOBSON,J.S., and MCMANUS,J.M. (1985). Pattern of atmospheric SO2 occurrence, Atmos. Environ., 19(3), 501-506 [24] BOUHAMRA,S. (1996). Personal communication, Statistical Department, Kuwait University [25] ZIZILA,Y. (1996). Personal communication, Kuwait University [26] MCRAE, G.J., and SEINFELD,J.H. (1983). Development of a second-generation mathematical model for urban air pollution-P. Evaluation of Model Performance, Atmospheric Environment, 17(3), 501-522

Call for Papers Endocrine Disruptors ESPR would like to contribute to and bundle the scientific discussion on the subject "Environmental Chemicals with Endocrine Activity". We invite reviews, full and short papers as well as other information and news from our readers.

204

Please contact us via phone, fax or email: +49-8191-125-469 (phone) +49-8191-125-492 (fax) [email protected]

ESPR - Environ. Sci. & Pollut. Res. 3 (4) 1996