Tuhin Subhra Konar S.N Chakrabarty

Indian Journal ofAir Pollution Control, VolX, No. 1, March 2011 pp 41-51

Development of Traffic Based Regression Models for Prediction of Roadside Criteria Air Pollutant Concentrations for the City of

Kol kata Tuhin Subhra Konar*,

Shibnath Chakrabarty** Environmental Engineering Section, Department of Civil Engineering, Jadavpur University,

188, Raja Subodh Mullick Road, Kolkata-700032, India Email: (tuhin.cv1 [email protected], [email protected])

Abstract A Correlation study between traffic and roadside air pollution has been made on a typical roadway in the city of Kolkata. A regression based approach has been taken, considering hourly volume of major motorised vehicles namely, taxi\private car, auto, bus\truck, two wheeler and light goods vehicles as independent variables and criteria air pollutants like NOx, PMlo and PM25, as dependent variables. Apart fiom these criteria air pollutants, NRPM has also been considered in the study. Separate regression models have been developed and statistically validated for all the pollutants. In all cases the Pearson's correlations between measured and model predicted values have been 0.75 or above. The developed regression models are further simulated in Monte Carlo's simulations for 100000 iterations, by including measured means and standard deviations of independent variables. The simulation out put means and percentage coefficient of variation values of concentrations of different pollutants are compared with the actually measured values and found quite resembling. The study also establishes the admissibility of simulated averages as annual averages.

Keywords: automobile pollution, traffic volume, regression, Monte Carlo's simulation

1. Introduction In urban area the contribution from automobile sources plays most significant role in air quality status. The increasing trend of demands for transport facilities is resulting acceleration in the rate of motorization. In Kolkata Metropolitan Area (KMA) also the number of motor vehicles registered in a single year has increased from 92,708 (of 2000) by approximately 1.42 times only in five years and the value becomes 1,31,583 in 2006 (Infrastructure Development Finance Company Ltd. et al, 2008), resulting in deterioration of air quality caused by different vehicular air pollutants. Out of different air pollutants, Respirable Particulate Matter or RPM (PMlo), Fine Particulate Matter or FPM (PM2.5), Nitrogen Oxides (NOx), Non Respirable Particulate Matter (NRPM) are some key constituents. Proper approach towards the management of the problem related with air pollution in city area is to some extent achieved by finding out the relationships between how the traffic characteristics influences air quality from different point of view. In the current study traflic and air quality have been tried to correlate from those various aspects.

2. Materials and Methods For establishing the interrelationships between air quality and traffic in Kolkata, a typical city roadway span of Raja Subodh Mullick Road between Jadavpur Police Station to Sulekha is selected for the study (figure 1). Initially, concentration data of major air pollutants like NOx, PMlo, PM2.s, and NRPM are collected in varying times of different weekdays and weekends (Saturdays) fiom November, 2009 to March, 2010, (Annexure A shows the sampling plan) for sampling time of 4hrs at a stretch at Jadavpur University Gate No.3 on the mentioned roadway. For the purpose of NOx, PMlo and NRPM measurement, APM 460 BL High volume sampler with gaseous sampling attachment, and for PM2.5 measurement APM 550 Fine Particle Sampler (both made by Envirotech Instruments Pvt. Ltd) are used with receptor height being Im fiom ground level and the methodology is followed

Figure I!: Map of the Study Area with Relevant Details (Google Maps, 2010)

with reference to the operation manuals provided by Envirotech Instruments Pvt. Ltd. A simultaneous traffic monitoring is also conducted to count hourly volume of major motorised vehicles i.e. taxi\private car, auto, bus\truck, two wheeler and light goods vehicles. Local meteorological data namely temperature and wind speed data are collected fiom Jadavpur University, measured by Envirotech WM 251 automated weather monitoring station (with reference to Envirotech WM 251 Operation Manual). A statistical analysis based on regression is followed in SPSS (SPSS Inc., 1999) by taking traffic constituents as independent and measured pollutant concentrations as dependent variables to drive separate regression models for all major air pollutants by selecting data of monitoring days. The derived models are further validated by evaluating correlation between measured and predicted pollutant concentrations (by the derived models) with the help of monitoring data from different days. The regression models are further

simulated in Monte Carlo's Simulation (MCS) by incorporating mean and standard deviation (SD) for independent variables separately for weekdays and weekends. In this case, Palisade @risk software (Palisade, 2009) has been used. The simulation output parameters like means and percentage Coefficient of Variation (%CV) are compared with actual statistics to verify its applicability in predicting mean annual pollutant levels.

3. Results and Discussion

3.1 Regression Analysis To develop a regression models to predict roadside PMIo, NRPM, PM2 I( and NOx concentration the measured independent variables (hourly volumes of different motorised vehicles) and dependent variables (measured concentrations of the respective variables in pg/m3) from Annexure B are used. For construction of model the data of certain days have been chosen for each pollutant (Table 1). The developed regression model for predicting pollutant concentrations are validated or cross validated by taking the dependent and independent variable values of same or other days from the same datasets (Annexure B). The days selected for validation study of respective pollutants are also shown in Table 1. The correlation analysis between measured and regression model predicted pollutant concentrations are conducted in SPSS (SPSS Inc., 1999). The Pearson's Correlations 'r', (significant at 5% level) are evaluated. Figure 2 displays the best fit straight lines between measured and regression model predicted pollutant concentrations for all the pollutants. Annexure C gives the details of developed regression models.

Table 1: The sets of days for different pollutants for development and validation of regression models

Pollutants Days Used in Model Days Used in Development Model -

Validation PMm 1 ,2 ,3 ,4 ,5 ,6 ,7 ,8 ,9 , 10,21, 11, 12,13, 14,15,

NRPM 1,2,3,4,5,6,7,8,9,10,11, 26,27,28,29,30, 12, 13, 14, 15, 16, 17, 18, 19, 31,32,33,34,35 20,21,22,23,24,25

NOx 7,9, 10,13,14,15,16,18,19,20, 7, 9, 10, 14, 16, 23,25,27,28,29,32, 33 18, 19, 20, 25, 27,

29,32,33

From Annexure C it is evident that in all cases the 'r' values are 0.75 or above, which signifies that there are fare correlations between measured and model predicted values.

3.2 Simulations of Regression Models The regression models for different pollutants as represented in Annexure C are simulated in MCS for 100000 times with Palisade @risk software (Palisade, 2009) by incorporating means and SDs of independent variables separately for weekdays and weekends from 7AM to 21PM, calculated for total period of study (November, 2009 to March, 2010). The mean and SD values for all independent variables of regression equations (hourly average traffic volumes) are represented in Table 2 for weekdays and weekends.

J 200 - s % 150

200.00 250.00 300.00 350.00 400.

PM10 Concentration Measured NRPM Concentration Measured

- 0 100 MO 300 400 100 120 140 160 180

PMZS Concentration Measured NOx Concentration Measured

Table 2: Mean and SD Values for Hourly Average Traffic Volumes for Weekdays and weekends,

TaxiWri Two Light vate Wheelerslh Bus\TrucWh Goods Carlh Vehicleslh

Weekday 1008.9 536.4 372.5 222.4 5 1 mean over 7AM to 2lPM WeekdaySD 204.59 92.2 60.7 26.6 20 over 7AM to 21PM Weekend 1051 55 1 346 204 5 1 mean over 7AM to 21PM Weekend SD 124 100 76 32 19 over 7AM to 21PM

Simulated means and %CV values are compared with measured statistics for same period of time to verify its applicability in predicting long term average. Table 3 shows the measured and simulated mean and %CVs of different air pollutants for weekdays and weekends, where as, the figure 3 and figure 4 shows the simulation distributions of outcomes respectively for weekdays and weekends. From the Table 3, it is evident that simulation means in all cases are almost equal to measured means and simulation %CVs are similar or less, which indicate the acceptability of simulated results.

172.4 345.1

0.005

0.004

0.002

0.w1

0 0 0 0 0 0 0 d . Valws in Thousands

Figure 3: Distributions Observed in Simulation runs for Weekdays.

Table 3: Measured and Simulated Mean and %CV of different Air Pollutants for Weekdavs and Non Weekdays over the Time Span of 7AM to 21PM.

Weekday mean over 7AM to 2lPM

Weekday %CV over 7AM to 21PM

Weekend mean , over 7AM to 21PM

Weekend %CV over 7AM to 21PM

PMlo con measured

269

44.23

276

28.98

PM1o con

simulate d

259.02

20.22

206.86

19.83

NRPM con

measure d

624

26.44

572

13.81

NRPM con

simulat ed

602.25

12.30

596

12.75

PMz.5 measur

ed

198

37.37

217

32.25

pM2.5 simulat

ed

200.33

15.52

199.49

11.53

NOH measured

125

32

126

32.54

NOx simulated

129.57

1 1.4996

127.42

9.78

Figure 4: Distributions Observed in Simulation runs for Weekends.

It is mentioned that the simulation is done by taking diurnal means and SDs of different independent variables (hourly traffic volumes) calculated through a prolong period of 5 months (November 2009 to March 2010) and compared with means and CVs of measured pollutant concentrations over the same time span. The important points are that this covers more than 50% of a year (excluding rainy season) from worm November, 2009 to worm March, 2010, totally including cold December-January and avoiding rainy season. The traffic characteristics in weekdays and weekends are of same nature through out the year and road geometry is invariant. Only the meteorology varies. So, whether these simulations out puts may be taken as annual average or not is decidable on the basis of the fact that whether the study period average (November, 2009 to March, 2010) meteorological data, namely wind speed and temperature acceptably represent the annual values or not. From Annexure B mean of wind speed and temperature of 35 days sample in varying time are calculated as 1.27 mls and 25.56'~ respectively, whereas SD of wind speed is 0.56mls and that of temperature is 4.74'C. The average annual mean wind speed for Kolkata is 0.84 and temperature is 26.25 (Indian Society of Heating Refrigerating and Air-conditioning Engineers, 2005). The absolute relative deviation (Z) of mean of monitoring time wind speed with respect to the value as availed for annual mean (provided by Indian Society of Heating Refrigerating and Air-conditioning Engineers, 2005) is numerically 2.0 and that for temperature is 0.81. These imply that study time wind speed and temperature represents annual values well at 5% level of significance. Thus, under this set of arguments the simulation estimated average values can be accepted as annual average concentration levels of respective pollutants.

4. Conclusions The study presents a regression approach for modelling roadside concentration variations of some of the criteria pollutants for a typical roadway of Kolkata. Separate regression models developed for different pollutants have been found to show good correlation with measured values. Based on Monte Carlo's simulation approach, the models have shown perfect concurrence in prediction of pollutant concentration arising from long term variation in dependent variables over the study period. The study also justifies the acceptability of these long term average concentration levels to represent annual average levels of respective pollutants in study area. The modelling approaches as

shown in the study thus would be helpful for management of roadside air quality for the city of Kolkata.

Acknowledgement The authors are grateful to Prof. A. Debsarkar, Professor, Department of Civil Engineering, Jadavpur University, Anirban Kundu Chowdhury, Research Scholar, Department of Civil Engineering, Jadavpur University and Abhishek Chakraborty, Post Graduate Student, Department of Civil Engineering, Jadavpur University for their constant helps in this study.

References 1. Envirotech Instruments Pvt. Ltd., Envirotech APM 460 BL Respirable Dust Sampler

Operation Manual. 2. Envirotech Instruments Pvt. Ltd., Envirotech APM 550 Fine Particle Sampler, Operation

Manual. 3. Envirotech Instruments Pvt. Ltd., Wind Monitor Envirotech WM 25 1, Operation Manual. 4. Google Maps 2010. 5. Indian Society of Heating Refrigerating and Air-conditioning Engineers, 2005, Weather

Data Set for Kolkata, West Bengal, www.ishrae.org.in 6. Infrastructure Development Finance Company Ltd., Superior Global Infrastructure

Consulting Pvt. Ltd., 2008. 'Comprehensive Mobility Plan- Back to Basics: Kolkata Metropolitan Area', pp. 27.

7. Palisade Corporation, Ithaka, New York, USA, 2009, (DTS) Decision Tools Suite. 8. SPSS Inc, 1999, SPSS 9.0.

Annexure-A

The sampling plan (WD Stands for Weekday and WND Stands for Weekend)

Day Wise Traffic Volume (W Stands for Weekday and N Stands for Weekend), Measured Concentrations of NOH, PMZj, PMlo and NRPM and Meteorological Data (Wind Speed and Temperature)

- - -

NRPM NOx PMts PMlo Concent- Concent- Concent- Concent-

Light ration ration ration ration Wind Tax~Wriva te Two Bus\ Goods Measured Measured Measured Measured Speed Temperature

Day Date Type Carlh Autolh Wheelerdh TrucWh Vehicledh (pg/m3 (pglm (pglm (pglm (mh) (degree C)

I Details of Developed Regression Models

PMlo NRPM PMzs NOX PMlO concentration (pgh3) = NRPM concentration PM2.5 concentration NOx concentration (pglm3) Regression -0.5 174 +0.20004 (pg/m3) = 27 +0.113 (pg/m3) = 43.96 +0.1311 =74.53 +0.0576

Equations (Taxi\Private Carh) (TaxWrivate Cadh) (Taxi\Private Carh) (Taxi\Private Carih) +0.0767 (Autoh)-0.0495 1 + 0.58 (Autoh) +0.27 (Two -0.0343 (Autoh) +O. 1062 -0.084 (Autoh) +0.00 14 (Two Wheelersh) + 0.49 Wheelersh)+ 0.67 (Two Wheelers/h)+ 0.1655 (Two Wheelersh)+ 0.1 7982 (Bus\TrucMh) -1.45 (Light (Bus\TrucWh) -1.95 (Light (Bus\Truck/h) -0.664 (Light (Bus\Truck/h)+ 0.0288 Goods Vehiclesh) Goods Vehiclesh)) Goods Vehiclesh) (Light Goods Vehicleslh)

Equation Relating

y = 0.8335~ + 32.143 Value with Measured Value

Actual PM 10 concentration Actual NRPM concentration Actual PM2.5 concentration Actual NOx concentration Calibration (pg/m3) = 0.7573 (Model (p@) = 0.56 (Model (pglm3) = 1.2319 (Model (pglm3) = 1.4374 (Model Equation Predicted Concentration) + Predicted Concentration) + Predicted Concentration) - Predicted Concentration) -

--

earso on's Correlation 0.795 (r)