University of Southern California Dissertations and Theses

Locating nearby sources of air pollution using air quality data and wind direction

(USC Thesis Other)

Locating nearby sources of air pollution using air quality data and wind direction

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content LOCATING NEARBY SOURCES OF AIR POLLUTION USING AIR QUALITY DATA AND WIND DIRECTION Copyright 2003 by Yu-Shuo Chang A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirement for the Degree DOCTOR OF PHILOSOPHY (ENGINEERING) May 2003 Yu-Shuo Chang Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UMI Number: 3103868 UMI UMI Microform 3103868 Copyright 2003 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. ProQuest Information and Learning Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, Ml 48106-1346 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UNIVERSITY OF SOUTHERN CALIFORNIA THE GRADUATE SCHOOL UNIVERSITY PARK LOS ANGELES, CALIFORNIA 90089-1695 This dissertation, written by under the direction o f h i 1 dissertation committee, and approved by all its members, has been presented to and accepted by the Director o f Graduate and Professional Programs, in partial fulfillment o f the requirements fo r the degree o f YU-SHUO C H A N G DOCTOR OF PHILOSOPHY Director D ate March 13, 2003 Dissertation Committee Chair Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ACKNOWLEDGEMENTS There are so many wonderful people who have assisted me and supported me through the time of my doctoral research. I am thankful to them and would like to acknowledge their contributions to my growth in my doctoral research career. First and most of all, I would like to acknowledge my advisor, Dr. Ronald C. Henry, who has provided me well-structured research objectives and an environment that any could ever possibly have. His enthusiasm, insight suggestions, generosity of sharing life and ideas conduct me and keep me excited through my doctoral research. Moreover, thanks to his encouragement on me have given me confidence through the program. I want to thank my doctoral research committee, Dr. Sioutas, Dr. Devinny, Dr. Yen, and Dr. James for their reviews of my dissertation and generous suggestions. I am honored to have them in my research committee. And I also want to thank Dr. Elisabeth Arnold for the writing helps to my dissertation. In addition to these, I would like to acknowledge the United States Environmental Protection Agency for the necessary funding to support my doctoral research work. I am sincerely and truly grateful to my uncle Dr. Biao-Hun Chang and his families who give me the warmest helps while I am in the United States. Their caring, ii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. generosity, and encouragement through the years had made my life easier. Also many thanks to the friendship of Peter Wan, Dr. Shih-Pin Tu, Dr. Shih-Jen Tu, and Dr. Ming-Chih Chang. Finally, I am grateful and thankful to my family for what they have gave to me in my life. My grandmother, brother Yu-Hao and Yu-Chi have been so supportive. Especially and mostly, I want to dedicate this work to my mother Chuan-Hui, my father Biao-Sheng, and my fiance Ya-Chun for their understanding, encouragement, and support through my research life. Not only the deepest supports, but also the unconditional love gave me the strength and confidence to stand up from where I fell. Without you, I have nothing. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. TABLE OF CONTENTS Pages ACKNOWLEDGEMENTS ii L IS T O F F IG U R E S v i LIST OF TABLES xi ABSTRACT x ii CHAPTER 1 INTRODUCTION 1 1.1 Background 2 1.1.1 Health Effects of Volatile Organic Compounds 3 1.1.2 Tropospheric Ozone Formation 5 1.1.3 Photochemical Assessment Monitoring Stations Program 9 1.2 Rationale for Locating Sources by Wind Direction 10 1.3 Nonparametric Regression 17 1.4 Application for Evaluation of Emission Inventories 19 1.5 Thesis Overview 21 CHAPTER 2 LITERATURE REVIEW 25 2.1 Application of Wind Direction Measurements 26 2.2 Application of Plot Smoothing 28 2.3 Evaluation of Emission Inventories 31 CHPATER 3 METHODOLOGY 35 3.1 Local Average 36 3.2 Kernel Estimator 39 3.3 Smoothing Parameter Selection 4 5 3 .4 Confidence Intervals 4 9 iv Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CHAPTER 4 STUDY AREA AND DATA DESCRIPTION 54 4.1 Study Area 55 4.2 Ambient Measurements 56 4.3 Emission Inventories 61 CHPATER 5 APPLICATION OF NONPARAMETRIC REGRESSION FOR LOCATING THE LARGEST SOURCE 65 5.1 Objectives 6 6 5.2 Data Screening 67 5.3 Results of Nonparametric Regression Smoothing 71 5.4 Comparison to Known Sources 77 5.5 Location of the Largest Source 82 5.6 Discussion 88 CHPATER 6 APPLICATION FOR VOC EMISSION INVENTORY EVALUATION 90 6.1 Introduction 91 6 .2 Description of Unmix multivariate receptor model 92 6.3 Data Description and Treatment 97 6.4 Results 103 6.4.1 Roadway Emission 110 6.4.2 Refinery Emissions 115 6.4.3 Petrochemical Industrial Emissions 129 6.4.4 Local Propane Source 134 6 .5 Comparison to Emission Inventory 13 8 6 .6 Discussion 148 CHAPTER 7 CONCLUSION AND SUGGESTIONS FOR FUTURE RESEARCH 150 7.1 Conclusion 151 7.1.1 Application for Locating the Largest Emission Sources 152 7.1.2 Application for Emission Inventory Evaluation 154 7.2 Suggestions for Future Studies 156 REFERENCES 162 v Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. LIST OF FIGURES Figure 1-1 Schematics of the reactions involved in NO-to-N02 conversion and 03 formation in (A) N 0-N 02-03 systems in the absence of VOCs, and (B) N 0-N 02-03 systems in the presence of VOCs (Atkinson, 2000). Figure 1-2 Hourly concentration of cyclohexane at the Deer Park PAMS site in Houston versus wind direction for 1997. Figure 1-3 Bar chart of average concentrations of cyclohexane in 10° bins of wind direction starting at zero. Figure 1-4 Bar chart of average concentrations of cyclohexane in 10° bins of wind direction with estimated uncertainty. Figure 3-1 The curve of average cyclohexane concentration calculated by a 1 0° wide sliding window. Figure 3-2 Full Width at Half Maximum Figure 3-3 The curve of average cyclohexane concentrations versus wind direction by using a Gaussian kernel with a 10° FWHM. Data with wind speed < 1 mile/hour are removed. Figure3-4 The polar coordinator plot of average cyclohexane concentrations versus wind direction by using a Gaussian kernel with a 1 0° FWHM. Data with wind speed <1 mile/hour are removed. Figure 3-5 Cross-validation error curve of cyclohexane concentrations at the Deer Park site. The plot shows the optimal FWHM is equal to 7. Figure 3-6 The curve of average cyclohexane concentrations versus wind direction by using a Gaussian kernel with a 7° FWHM. Data with wind speed <1 mile/hour are removed. The gray region shows the 95% confidence interval. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 4-1 Map of Houston area. The location of the Clinton Dr. site is at x and the Deer Park site is at +. Houston Ship Channel is located at the center of the map. The red solid lines are major interstate highways (I-10 and 1-45). VOC sources are shown as red circles with the area proportional to the emission rate. The data of emissions is from the AIRS emission inventory. 59 Figure 5-1 The scatterplot of hourly cyclohexane concentrations versus wind direction at the Clinton Dr. site in 1997. 69 Figure 5-2 The scatterplot of hourly cyclohexane concentrations versus wind direction at the Deer Park site in 1997. The marked points were identified as outliers. 70 Figure 5-3 Nonparametric regression curve of average cyclohexane concentrations at the Deer Park site by using the Gaussian kernel with a FWHM equal to 7°. Data with wind speed less than 1 mile per hour are removed. The gray region is the 95% confidence interval. 72 Figure 5-4 Nonparametric regression curve of average cyclohexane concentrations at the Clinton Dr. site by using the Gaussian Kernel with a FWHM equal to 10°. Data with wind speed less than 1 mile per hour were removed. The gray region is the 95% confidence interval. 73 Figure 5-5 The estimated location from nonparametric regression method is marked as black dot. The location of the Phillips Petroleum source in the inventory is shown as *. The Clinton Dr. site is located at x and the Deer Park site is at +. The red circles are the VOC emissions from the stacks of Phillips Petroleum Company. The dash lines show the directions estimated from two sites. 83 Figure 5-6 Nonparametric regression curve of average cyclohexane concentrations at the Deer Park site by using the Gaussian Kernel with a FWHM of 5°. Data are restricted to periods with wind speed greater than 6 miles per hour (about one hour travel time from the Phillips source to the site). The gray region is the 95% confidence interval. 85 vii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 5-7 Nonparametric regression curve of average cyclohexane concentrations at the Clinton Dr. site by using the Gaussian Kernel with a FWHM equal to 10°. Data are restricted th periods with wind speed greater than 5 mile/hour (about one hour travel time from the Phillips source to the site). The gray region is the 95% confidence interval. 86 Figure 6-1 Map of Houston area. It shows the spatial relationship of major highways (I-10 and 1-45) and two PAMS monitoring sites. Houston Ship Channel is parallel with route 225. 96 Figure 6-2 Sum of all identified species (as % of reported TNMOC) vs reported TNMOC at the Clinton Dr. site. 101 Figure 6-3 Estimated source contribution from Unmix based on the Clinton Dr. site data. 107 Figure 6-4 Estimated source contribution from Unmix based on the Deer park site data. 108 Figure 6-5 The plot of estimated contributions of roadway versus wind direction (a) at the Clinton Dr. site, FWHM = 25°, and (b) at the Deer Park site, FWHM = 28°. 114 Figure 6 - 6 The plot of estimated contributions of refinery 1 versus wind direction (a) at the Clinton Dr. site, FWHM = 10°, and (b) at the Deer Park site, FWHM = 20°. 119 Figure 6-7 Houston map shows the predicted direction of refinery 1 (isobutane) based on Figure 6 -6 . The “X” is the location of the Clinton Dr. site and the “+” is the Deer Park site. YOC sources are shown as red circles with the area proportional to the emission rates. 1 2 0 Figure 6 -8 The plot of estimated contributions versus wind direction for (a) refinery 2 at the Clinton Dr. site, FWHM = 12°, (b) refinery 3 at the Clinton Dr. site, FWHM =11°, and (c) refinery mixing sources at the Deer Park site, FWHM =16°. 124 viii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 6-9 Houston map shows the predicted direction of refinery 2 & 3 (refinery mixture) based on Figure 6 -8 . The “X” is the location of the Clinton Dr. site and the “+” is the Deer Park site. VOC sources are shown as red circles with the area proportional to the emission rates. 126 Figure 6-10 The plot of estimated contributions of refinery 4 versus wind direction (a) at the Clinton Dr. site, FWHM = 15°, and (b) at the Deer Park site, FWHM = 20° 127 Figure 6-11 Houston map shows the predicted direction of refinery 4 (n- hexane) based on Figure 6-10. The “X” is the location of the Clinton Dr. site and the “+” is the Deer Park site. VOC sources are shown as red circles with the area proportional to the emission rates. 128 Figure 6-12 The plot of estimated contributions of industry 1 versus the wind direction (a) at the Clinton Dr. site, FWHM = 10 °, and (b) at the Deer Park site, FWHM = 19°. 132 Figure 6-13 Houston map shows the predicted direction of industry 1 (cyclohexane) based on Figure 6-12. The “X” is the location of the Clinton Dr. site and the “+” is the Deer Park site. VOC sources are shown as red circles with the area proportional to the emission rates. 133 Figure 6-14 The plot of estimated contributions of industrial 2 at the Clinton Dr. site. The nonparametric regression parameter, FWFIM, is equal to 15°. 135 Figure 6-15 Houston map shows the predicted direction of industry 2 (xylene) based on Figure 6-14. The “X” is the location of the Clinton Dr. site and the “+” is the Deer Park site. VOC sources are shown as red circles with the area proportional to the emission rates. 136 Figure 6-16 The plot of estimated contributions of the propane source at the Deer park site. The nonparametric regression parametric, FWHM, is equal to 79°. 139 IX Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 6-17 The nonparametric regression plots for reported TNMOC and derived sources at the Clinton Dr. site. The blue line is the reported TNMOC measurements at the Clinton Dr. site. The red line is the sum of six refinery and industrial sources estimated by Unmix. The pink line is the estimated contribution of refinery 1 in the Clinton Dr. solution. 141 Figure 6-18 The plot of distance-weighted emissions from major reported VOC sources and refinery 1 (isobutane) derived from Unmix. The red line is the estimated source contributions of refinery 1. 145 Figure 6-19 The plot of distance-weighted emissions from major reported VOC sources and refinery 2 (n-butane) derived from Unmix. The red line is the estimated source contributions of refinery 2. 146 Figure 6-20 The plot of distance-weighted emissions from major reported VOC sources and refinery 3 and 4 (refinery mixture) derived from Unmix. The red line is the estimated source contributions of refinery 3 and the black line is the estimated source contributions of refinery 4. 147 Figure 7-1 The 3-D plot of wind direction and wind speed versus cyclohexane concentrations at the Clinton Dr. site. The FWFIM for wind direction and wind speed is 1 0° and 2 mile/hr. 158 Figure 7-2 The plot of back trajectory for cyclohexane concentrations at the Deer Park site. The size of grid cells is 450m*450m. The Deer Park site is marked as “+” and the Clinton Dr. site is marked as “x”. The Phillips Corporation is marked as 160 x Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. LIST OF TABLES Pages Table 4-1 List of the PAMS target VOCs. 60 Table 5-1 Largest peaks in the curve of average cyclohexane concentrations versus wind direction in Figure 5-3 and Figure 5-4. 78 Table 5-2 Emissions of cyclohexane of 1997 in Harris County, Texas. 80 Table 5-3 Largest peaks in the curve of average cyclohexane concentrations versus wind direction in Figure 5-5 and Figure 5-6. 87 Table 6-1 Basic information of species measured at the Clinton Dr. site. 98 Table 6-2 Basic information of species measured at the Deer Park site. 99 Table 6-3 Estimated source composition (mass percent)at the Clinton Dr. site. 105 Table 6-4 Estimated source composition (mass percent) at the Deer Park site. 106 Table 6-5 Summary of 1997 MTBE emissions in Harris County, Texas. 122 Table 6 - 6 Summary of 1997 n-butane emissions in Harris County, Texas. 130 Table 6-7 Summary of 1997 xylene emissions in Harris County, Texas. 137 Table 6 -8 Summary of VOC emission rates from the AIRS database (tons/year). 144 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ABSTRACT Regulatory agencies count on self-reported emission inventories of air pollutants to make regulatory decisions. Incorrect emissions used in air quality models may lead to poor control strategies. Thus, an independent method is required to evaluate these emission inventories. This research work brings in a statistical method to determine the locations of nearby sources based on wind direction and measured pollutant concentrations. This statistical method then can be applied to the source contributions estimated by receptor models to indicate the directions and strengths of nearby sources. These predicted locations and strengths of nearby sources provide the fundamental link between the emissions inventory and observed concentrations. Nonparametric regression is the statistical method introduced in this work to estimate the wind direction that gives a local maximum in the average concentration of an air pollutant. Nonparametric regression can distinguish real concentration peaks from random noise and determine the precise direction of a nearby source with much better accuracy than other traditional methods such as pollution roses. A test of the nonparametric regression method was carried out using measured cyclohexane concentrations at two monitoring sites near a heavy petrochemical region in Houston, Texas. The source location determined by triangulation demonstrates that the nonparametric regression method can estimate the direction of xii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the dominant cyclohexane source precisely and locate that dominant source to within 500m. The nonparametric regression method can be applied to the source contributions estimated by receptor models as well as individual species concentrations. Unmix multivariate receptor modeling and analysis software is used to determine the sources, compositions, and contributions from monitoring measurements. The most satisfactory Unmix models of 1997 Houston VOC measurements are presented. Nonparametric regression of source contributions on wind direction determines the direction and source impacts of nearby sources to the monitoring site. These observationally-based results are compared with the emission inventory to determine any inconsistencies within the self-reported emission inventories. These inconsistencies will guide the necessary correction to the emission inventories in the future. The nonparametric regression method is a powerful technique that is going to provide significant accomplishments in air quality studies and atmospheric science. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CHAPTER 1 INTRODUCTION Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1.1 Background The troposphere is the region of the Earth’s atmosphere in which we live. This region can be compared to like a reservoir to mix the substances that are released from human activities and natural resources. However, anthropogenic emissions of many substances are much greater than natural releases due to growing industrialization, fossil fuel consumption, and transportation networks. These man- made emissions increase concentrations of these substances to produce measurable effects on humans, animals, vegetation, or materials. Therefore, how to control and reduce the emissions from anthropogenic sources is the significant concept for air quality controlling strategies. This concept can be accomplished when the source locations and contributions have been verified. Volatile organic compound (VOC) is one kind of air substances released from anthropogenic sources. These organic compounds are defined as having a boiling point that ranges between 50°C and 260°C (Jones, 1999). The typical VOCs include alkanes, alkenes, and aromatic hydrocarbons. Most VOCs are from motor vehicle exhaust in urban areas. The others are emitted from refineries and chemical plants (defined as stationary point sources), gasoline vapor, coatings, solvent usage, and dry-cleanings (defined as area sources) (Sweet and Vermette, 1992; Smith, 1993; Mukund et al., 1996). These organic compounds give adverse effects on human health and play as the key substances in tropospheric chemistry. Although VOCs have not been classified as the criteria air pollutants, they have affected the public 2 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. health. Some of these organic compounds have been classified as toxic air pollutants and regulated by U.S. Environmental Protection Agency (EPA). In addition, VOCs have been demonstrated the relation to the formation of secondary pollutants in the photochemical smog, which is the current major air pollution problem in many urban areas. Therefore, emission control of VOCs is a significant topic not only because of their adverse health effects but also because of their critical position in the atmospheric chemistry. 1.1.1 Health Effects of Volatile Organic Compounds A wide array of health effects have been linked to VOCs. They are the primary public health concern because many adverse effects are associated with exposure to them. The evidence of epidemiological studies have indicated that VOCs are associated with several chronic respiratory symptoms and central nervous system effects; they also cause liver, renal, and hematological effects (Bolla, 1991; Ashley et al., 1992; Ware et al., 1993). In addition to the respiratory effects, some aromatic hydrocarbons such as benzene, toluene, xylene, and styrene have been demonstrated as carcinogens. Under a long-term exposure, these aromatic hydrocarbon compounds increase the cancer risk (Lynge et al., 1997; Gerin et al., 1998; Jones, 1999). Although VOCs are not the criteria air pollutants, 33 organic hydrocarbon substances have been defined as air toxics presenting a serious threat to human health. The emissions of these air toxics are regulated by the U.S. EPA, which 3 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. compiles the Toxics Release Inventory (TRI) data each year and makes it available for public knowledge about the emission of local air toxic sources (U.S. EPA, 2000). In addition to their own toxicity and carcinogenicity, VOCs also create adverse health effects by their secondary products called secondary organic aerosols. These secondary organic aerosols have even more adverse effects on human health because these aerosols can be transported through the lung airways to deposit into the respiratory system (Vyskocil et al., 1998). The secondary organic aerosols are formed by the complicated photochemical reactions of VOCs and nitrogen oxides. These semi-volatile aerosols are produced by the gas-phase reactions of organic compounds and nitrogen oxides (Seinfeld, 1986; Seinfeld, 1989; Pitts, JR, 1993). The formation of secondary organic aerosols has been confirmed by several smog chamber experiments. The amount of VOCs determines the yields of second organic aerosols in the troposphere (Wang et al., 1992; Forstner et al., 1997; Forstner et al., 1997; Odum et al., 1997; Kleindienst et al., 1999). Secondary organic aerosols are also the characteristic species of photochemical smog, which is a mixture of air pollutants formed by photochemical reactions in the lower troposphere. Photochemical smog has been recognized as a serious air pollution problem in many urban areas. Peroxyacetylnitrate (PAN) and peroxypropionyl nitrate (PPN) are two secondary organic aerosols that have been 4 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. recognized as the specific indicators to monitor the transportation of photochemical smog between the urban and downwind remote areas (Grosjean and Williams II, 1992; Pandis et al., 1992; Schmidt et al., 1998). Epidemiological investigations reported that photochemical smog causes many adverse effects on human health. Ozone is the major agent responsible for damage to public health and welfare caused by photochemical smog. And VOCs are the key substances associated with ozone formation in the photochemical smog. 1.1.2 Tropospheric Ozone Formation In 1952, Haagen-Smit first used the term “photochemical smog” to describe a mixture of air pollutants, which was a serious air pollution problem in the Los Angeles area (Haggen-Smit, 1952). This smog is characterized by decreasing visibility, crop damage, eye irritation, objectionable odor, and rubber deterioration. These effects are attributed to the ozone formed by photochemical reactions in the troposphere. Because the emissions of ozone precursors have been increased by human activities, the tropospheric ozone measurements are frequently over the ambient level in urban areas and in downwind remote areas. Ozone is the major substance responsible for the damages to public health and welfare caused by photochemical smog. Ozone has been shown to have its adverse effects on human and vegetation health in many epidemiological studies. Ozone irritates lung airways to cause symptoms 5 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. including wheezing, coughing, and painful breathing when taking a deep breath. With long-term exposure, ozone may cause permanent lung damage (Lippmann, 1991; Stern et al., 1994; Paige et al., 2000). Even at very low levels, ozone triggers a variety of health problems including aggravated asthma, reduced lung capacity, and increased susceptibility to respiratory illnesses like pneumonia and bronchitis (Brauer and Brook, 1997). Regarding the vegetation damage, ozone damages the leaves of trees and other plants, ruining the appearance of cities, national parks, and recreation areas (Westenbarger and Frisvold, 1994; McLaughlin and Downing, 1995). Consequently, ozone has been set as a criteria pollutant of National Ambient Air Quality Standards (NAAQS) established by the U.S. EPA. The standard of tropospheric ozone now is 80 parts per billion (ppb) for an eight-hour average in order to set the limits for the protection of public health and welfare. However, many people still live in urban areas with ozone concentrations exceeding the NAAQS standard. An effective strategy is needed to reduce the ozone concentration and improve the ambient air quality in these ozone nonattainment areas. Ozone is the principle product of photochemical reactions in the troposphere. It is formed in the presence o f sunlight, water vapor, and its precursors, which are VOCs and nitrogen oxides (NOx ). The current known route of tropospheric ozone formation is the photodissociation of nitrogen dioxide (N 02 ) by sunlight in the 6 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 1-1 Schematics of the reactions involved in NO-to-N02 conversion and 03 formation in (A) N 0-N 02-03 systems in the absence of VOCs, and (B) N 0-N 02- 03 systems in the presence of VOCs (Atkinson, 2000). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. lower troposphere (Chock and Heuss, 1987; Seinfeld, 1995; Atkinson, 2000). The photodissociation of NO2 produces the atomic oxygen to form ozone with oxygen gases. In the absence of other oxidizing species, ozone oxidizes nitrogen monoxide (NO) to form NO2, resulting in no net gain in ozone (Figure 1- 1A). However, the organic compounds change the way of ozone formation and destruction. The photodissociation of VOCs leads to the formation of alkyperoxy (RO2) and hydroperoxyl radicals (HO2). These free radicals react with NO and convert NO to NO2 (Figure 1-1B). As a result, the presence of VOCs provides another pathway leading to the oxidation of NO without destroying ozone. So VOCs contribute to the net formation of ozone in the lower troposphere. The current development of an ozone controlling strategy concentrates on the emission control of two ozone precursors: VOCs and NOx. The roles of these precursors in ozone formation have been firmly established. Thus, the reduction of anthropogenic emissions of these ozone precursors has been considered an effective strategy for ozone abatement. However, there is no firm scientific conclusion existing on whether reducing VOCs emissions of nitrogen oxides emissions or both is the most effective strategy for ozone abatement in the urban areas (Lindsay et al., 1989; W olff and Korsog 1992; Finlayson-Pitts and Pitts, Jr 1993; Cardelino and Chameides, 1995). In order to provide more advance studies for the ozone controlling strategy, an additional comprehensive database is required to collect ozone and its precursor measurements in these nonattainment areas. Therefore, a 8 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. monitoring network program has been established, called the Photochemical Assessment Monitoring Stations (PAMS) program. 1.1.3 Photochemical Assessment Monitoring Stations Program The 1990 Clean Air Act Amendments (CAAA) required U.S. EPA to promulgate the rules of enhancing the monitoring of ozone and its precursors for the ozone problem. Under this requirement, the PAMS program has been established as the part of the existing State Implementation Plan (SIP) monitoring network in the ozone nonattainment areas (U.S. EPA 1994; U.S. EPA 1997). This enhanced monitoring program provides comprehensive air quality data. This data consists of hourly ambient measured concentrations and meteorological measurements. The ambient measured concentrations provide ozone, VOCs, nitrogen oxides (NOx ), and additional criteria air pollutants such as carbon monoxide (CO) and sulfur dioxide (SO2). The meteorological measurements report wind direction, wind speed, atmospheric pressure, temperature, and sunlight radiation. This data can assist regulatory agencies to track the emissions of ozone precursors, evaluate the performance of photochemical models, and refine the control strategies for the ozone problem. Emission tracking is one of chief objectives of the PAMS monitoring network. The purpose of emission tracking is to corroborate the NOx and VOC inventories, 9 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. confirm the VOC species source profiles, and analyze the air toxics. The speciated VOC measurements provide the characteristics of emission source impacts. These characteristics can validate the VOC composition used as input to photochemical models. Furthermore, these speciated VOC measurements will assist to locate the toxic “hot spots” by wind direction analysis for corroboration of the toxic emission inventory. The toxic emission inventory can help to establish the annual mean concentrations and the size of affected population for preparing the toxic exposure assessment. In addition, the receptor modeling analysis can be applied to these measurements for VOC source apportionment. The results of source appointment will assist to recover unreported sources for verification of the VOC emission inventory. 1.2 Rationale for Locating Sources by Wind Direction The major concept of source locating is to determine the direction of the nearby source by high measured concentrations given through wind direction measurements. An atmospheric species is spread out in the atmosphere by the wind blowing from its emission source. When the emission source is not too far away, the measured species concentrations are apparently affected by the directional behaviors o f the wind. Based on this concept, the wind direction measurements with higher measured concentrations enable to indicate the possible direction of the nearby sources around the monitoring site. When the measured concentrations and 10 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. wind direction are plotted, the wind direction that gives the local maximum of measured concentrations can be taken as the direction of the nearby source. And the source location can be determined by triangulation using the directions estimated from multiple monitoring sites. The scatter plot is a common method to reveal the direction of the nearby source by using measured species concentrations and wind direction measurements. The possible direction of the nearby source can be estimated from the peak that gives the maximum measured concentrations in the scatter plot. However, it is very difficult to locate the maximum concentration peak in order to estimate the direction given by the peak from the scatter plot. For example, Figure 1-2 is a simple scatter plot of hourly measured concentrations of cyclohexane and wind direction measurements collected in 1997 at Deer Park, Texas. This monitoring site is near the Houston Ship Channel, an area dominated by large refineries and petrochemical facilities. The wind direction and speed were collected by hourly measurements. The wind direction is recorded as the azimuth angle (measured clockwise from north), which shows the direction the wind is blowing from. From this scatter plot, two peaks are shown by high concentration measurements when the wind comes from 0° to 50° and 300° to 350°. But it is hard to identify which one is the largest peak of measured concentration of cyclohexane. And the direction ranges of these two peaks from the scatter plot are too wide to estimate the possible reading of wind 11 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. direction. Therefore, no reliable results can be estimated to present the associated direction of cyclohexane sources from this scatter plot. An improved approach to analyze the data in Figure 1-2 is the bar chart. This approach groups the measured concentrations into bins of width A9 based on wind direction and calculates the average concentration in each bin. The results are shown in Figure 1-3, which displays the average concentration calculated by every 10° (A9 = 5°) of wind direction starting at 0°. The measured concentrations of cyclohexane used in Figure 1-3 were screened by wind speed. The measured concentrations with wind speed less than 1 mile per hour were excluded because the wind direction is not well detected in lower wind speed conditions. Based on Figure 1-3, the bar presenting the largest average concentration is located between 320° to 330° and the second largest average concentration is presented at 40° to 50°. Although the direction range of the peak can be narrowed by the bar chart, the more precise estimate of the peak location is still difficult to obtain because of the bar chart limitation. The major limitation of the bar chart is that the bar chart is highly dependent on the choice o f the bin size A9 and the location of the boundaries o f the bins. Peaks cannot be resolved when they are close together than 2A9 and the location of the peaks cannot be estimated smaller than ±A9. In order to obtain the precise direction 12 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 200 180 160 140 120 S 100 T O u . § 80 C o 60 40 o Q - Q . . 4- ....................... + ......" + t + + + ++4 : ' = * • ...... P . ........................ .. .. + 4. 100 150 200 wind direction 250 300 350 Figure 1-2 Hourly concentration of cyclohexane at the Deer Park PAMS site in Houston versus wind direction for 1997. 13 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. o 50 100 150 200 250 300 wind direction Figure 1-3 Bar chart of average concentrations of cyclohexane in 10' direction starting at zero. 350 bins of wind 14 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. estimate, a better peak resolution can be worked out by using a small A0 like one or two degrees. However, the wind direction frequency distribution limits the bin size selection. The winds are less frequent in some directions. A smaller A0 selection could result in some bins that have too few observations to calculate the average concentration in these bins. Usually, a bin size cannot be less than 10° even with the hourly data for an entire year. Another limitation of the bar chart is the uncertainty estimation of the average concentration. The estimated uncertainty (or confidence intervals) of the average concentration is required to identify the smaller peaks. For the peaks, estimated uncertainty would provide a measure of the error in peak location. However, the number of observations in a bin affects the uncertainty estimation for the average concentration. If the bin has too few observations, it makes a bigger uncertainty to the average concentration of the bin. And no answer can be given to the peak identification. Figure 1-4 shows the results of 95 percent confidence intervals to the average concentrations displayed every 10° bin. Depending on the estimated uncertainty, the small peaks at around 170°, 220°, 270°, and 290° are hard to identify as the real peaks or just the random fluctuations of data. These small peaks could not be identified as real peaks based on the data. Therefore, the limitation of the bar chart could affect the results to present the direction of emission sources. 15 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 12 10 o _Q Q_ € 6 0) O c o o " J / J N i \ A V ✓ 50 100 150 200 wind direction 250 300 350 Figure 1-4 Bar chart of average concentrations of cyclohexane in 10° bins of wind direction with estimated uncertainty. 16 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. An improved method is required to present the peak location more precisely and separate peaks more reliably even they are close to each other. This method should be able to calculate the average concentration in a small bin size of wind direction with limited observations. And the uncertainty estimation of the peak can be estimated for judging the peak reliability. Furthermore, any parameters needed for this method should be calculated from the reproducible and qualitative algorithm to avoid the subjectivity of the analysis. For the above requirements, a statistical method called nonparametric regression has been introduced to wind direction analysis. This method smooths the plot of wind direction measurements and measured species concentrations as a continuous curve, which can give the precise estimate of wind direction to reveal the direction of the nearby source. 1.3 Nonparametric Regression The major issue of wind direction analysis is to estimate the wind direction to give a local maximum in the observed average concentration of an atmospheric species. In order to estimate the wind direction precisely, regression analysis is a common approach to determine the curve of the wind direction measurements and measured species concentrations. A general aspect of regression analysis is to assume a particular family o f distribution that fits the curve to the data. However, to assume that observations come from any specified family of distribution may be inappropriate. A well-fitting curve could not be easy to find for the data, especially 17 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. in biomedical and engineering fields (Gasser et al., 1984). Therefore, an alternative nonparametric regression approach has been introduced to apply to the wind direction analysis. Nonparametric regression is a set of techniques for estimating a regression curve without making strong assumptions about the shape of the true regression function (Altman, 1992). This alternative fits a curve to the data, making no assumption about a particular distribution. The concept of nonparametric regression relies on the information supplied by the data to speak for itself concerning the actual form of the regression curve. The unknown regression function of the data is estimated by a small neighborhood of the data. This “local” idea gives the great flexible space to estimate that unknown regression function. Therefore, nonparametric regression is able to determine the best suitable curve to the data in situations where there is little or no prior information available about the regression curve (Eubank, 1999). This advantage makes nonparametric regression suitable for determining the regression curve of average species concentrations to give the precise direction of the peak. The feasibility o f nonparametric regression for wind direction analysis can be examined by the predicted position of the dominant source of the species. The largest peak of average species concentrations gives the expected direction of the dominant source to the site. When two site data are available, the predicted position of this dominant source is expected at the intersection of the two lines drawn from 18 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. each site in the direction of the largest peak. Therefore, the feasibility of nonparametric regression can be judged by how closely the predicted position of this dominant source is to the known location. The result has demonstrated that nonparametric regression predicts the location of the dominant source of cyclohexane precisely by using monitoring data collected in Houston, Texas (Henry et al, 2 0 0 2 ). 1.4 Application for Evaluation of Emission Inventories One o f the chief objectives of the PAMS program is to provide the speciated VOC measurements for tracking the individual components of emission sources and verifying the VOC emission inventories. The emission inventories offer the self- reported emission rates of organic gases from industrial sources. Regulatory agents and photochemical models count on these industrial emission rates. Incorrect self- reported emissions can severely impact the air quality models and regulatory decisions. Thus, an independent and objective method is needed to verify the emission inventory, which may often be outdated, incomplete, or inaccurate. Receptor modeling analysis is applied for this purpose. Receptor models provide an alternative method to estimate the composition and contribution o f VOC sources by using VOC measurements. Receptor models do not require the knowledge of emission rates, wind, or other meteorological data to 19 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. reconstruct the dispersion patterns on physical and chemical processes in the atmosphere. The fundamental equation of receptor models is the mass balance. Receptor models assume that species are conserved during transport between the source and the sampler; the species composition is known at the emission source as well as in the collected sample. And the source contribution is apportioned by the measured VOC concentrations and conserved species composition. Based on the estimated source composition and contribution, the VOC emission sources can be identified. Wind direction analysis can be applied to the estimated source contributions for extending the explanation o f solution from receptor models. The application of wind direction analysis gives the plots to reveal the directions o f multiple sources that release the same chemical species. The location o f reported sources in the inventory can be compared with the direction given by the peaks. Previous work on comparison o f the emission inventory for the Houston ship channel relied on simple bar charts to determine the direction of VOC sources (Henry et al., 1997). Nonparametric regression provides an improvement to produce a sharper plot for determining the precise direction of emission sources. This precise result gives the basis to evaluate the consistency between the emission inventories and observations. The extension of nonparametric regression provides a useful way to achieve the objective of PAMS program to verify the emission inventory. 20 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1.5 Thesis Overview This thesis is divided in seven chapters. Chapter 1 (Introduction) presents an overview of the significance of VOCs in health concerns and in the formation of tropospheric ozone. This chapter also establishes the rationale to locate VOC sources based on measured concentrations and wind direction. The major idea of this rationale is to estimate the wind direction that gives a local maximum in the observed average concentration of an atmospheric species. Nonparametric regression gives an improvement to treat the plot in order to have a precise estimate of the wind direction of the peak. This technique not only estimates the precise direction of the emission source but also predicts the position of the emission source accurately. In addition, this technique also can be applied to the source contributions estimated from receptor models. This application gives the sharper results to examine the consistency of the emission inventory and ambient observations. Chapter 2 and Chapter 3 are the literature review and the methodology respectively. In Chapter 2, previous studies related to wind direction analysis have been reviewed. These previous studies have demonstrated the feasibility of wind direction analysis in air quality studies. The statistical literatures related with curve smoothing have also been reviewed. The technique called nonparametric regression can smooth the curve and provide the benefits to enhance the performance of wind direction analysis for locating the nearby source of air pollution. And the detailed 21 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. methodology of nonparametric regression is described in Chapter 3. Hardle (1990) provided the foundations of nonparametric regression subjects applied in this dissertation. Chapter 4 is the description of the data. The study area is Houston, Texas, a large metropolitan area with heavy petrochemical industrial activities. Two types of data have been used for the study: ambient observations collected from PAMS sites, which provides species concentrations and wind direction measurements, and the emission inventory from the U.S.EPA, which offers the emission rates, locations, and simple description of the industrial emission sources in the Houston area. The ambient observations are used to estimate the location o f the emission source by nonparametric regression. And the emission inventory is used to examine the consistency of the results estimated from the ambient observations. Chapter 5 describes the application of nonparametric regression for locating the dominant industrial source of cyclohexane in the Houston area. The emission inventory shows that 70% of the industrial emissions of cyclohexane in the Houston area are known to be from a single source. When two site data are available, the position of this dominant cyclohexane source can be expected at the intersection of two lines drawn from each site in the direction of the largest peak. The result shows that the distance from the position of this dominant cyclohexane source determined by triangulation to the known location from the inventory is less than 500 meters. 22 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. This short distance demonstrates the source location predicted by the nonparametric regression is in agreement with the inventory. Henry et al. (2002) demonstrated this work. Chapter 6 presents the results of the source apportionment of VOCs from Unmix software by using the ambient monitoring data of 1997 in Houston. Unmix is a multivariate receptor modeling and statistical analysis software, which estimates the source number, composition, and contribution without any prior knowledge of sources in the Houston area. And nonparametric regression is applied to source contributions from Unmix to giving the sharper results for wind direction analysis. We identified seven VOC sources based on 1997 ambient monitoring data from Unmix model. Three major sources account for about 70 percent of total VOC contribution in Houston. These sources resemble roadway emissions, isobutane related to MTBE production, and a refinery source. The remaining sources are characterized by one or two significant species to relate with specific petrochemical industries. Nonparametric regression produces the plots of hourly source contributions with wind direction to give the direction of the sources by the peaks. We compare our results with the emission inventory and only match the smallest of these sources to a specific petrochemical plant. The other emissions in the inventory fail to agree with our observation-based results. We believe our empirical method shows the inventory of industrial VOC emission is inaccurate in consistency 23 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. in the location and emission rates of major sources. This result can be the indicator to improve the emission data in the inventory. And finally, conclusions and suggestions o f the future research are discussed in Chapter 7. 24 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CHAPTER 2 LITERATURE REVIEW Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2.1 Application of Wind Direction Measurements The wind directional behavior is important in predicting the concentrations of pollutants in the atmosphere. Therefore, the wind direction has been applied in air pollutant assessments to aid in investigating the relationship between source emissions and impacts measured at monitoring sites. For example, pollutant rose plots have been used to show the directional patterns of pollutants for revealing the possible directions of emission sources (Kretzschmar and Cosemans, 1979; Batterman et al., 1987; Kriews et al., 1988; Chu, 1995; Incecik, 1996). Then average pollutant concentrations have also been plotted by the wind direction to locate the emission sources (Cheng and Lam, 1998; Uchiyama and Hasegawa, 2000; Takahashi et al., 2001). And the field data have been generated to investigate the relationship between the wind direction frequency distribution and point sources (Carter et al., 1993; Eklund, 1999). These studies have demonstrated wind directions were associated with the samples having high concentrations, which have been influenced by emission sources. These results give the evidence to explore the correlation of ambient concentrations and wind direction in a qualitative manner. Several studies have explored the correlation of measured concentrations and wind direction by analyzing the samples collected from the known source. Statistical and graphical methods have been applied to the ambient observations for describing the measured concentrations by wind direction, which has been referred to as the wind direction analysis. Somerville et al. (1994) used the linear-angular rank correlations 26 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. to quantitatively determine the relationship between the wind direction and pollutant concentrations. Ziomas et al. (1995) developed a model based on the correlation analysis of wind direction to forecast pollution concentrations. Okabayashi et al. (1996) proposed an estimation method of calculating the maximum ground-level concentration for the large wind directional fluctuations by using the wind tunnel data over a terrain model. Then Somerville et al. (1996) presented a classic parametric modeling approach to estimate the wind direction corresponding to the maximum pollutant concentration; this corresponding wind direction fitted the direction of a known emission source referred as the monitoring site. In conclusion, these studies have given the evidences that the measured species concentrations can be described as a function of the wind direction. The wind direction that gives the maximum concentration of an air species is taken as the direction of the nearby emission source of the species. These previous studies demonstrated the feasibility of the wind direction analysis for quantitative determining the direction of the dominant emission source of an air species. The obvious extension of the wind direction analysis is to explore how to describe the species concentration as a function of the wind direction by using ambient observations. This function should be smooth as a continuous curve by the wind direction. Then the wind direction that gives a local maximum of the species concentration can be estimated by using the algorithm. Therefore, an advanced technique is required to smooth the plot of the species concentration versus the wind 27 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. direction. Based on the smoothed curve, the peaks give the evidence to indicate the number of the species sources. And the wind direction of the peaks is taken as the direction of dominant emissions of the species. According to this smooth curve, all emission sources that emit the same species can be recovered without any prior knowledge of the sources. 2.2 Application of Plot Smoothing The scatterplot is a general graphical technique to present the relationship between two variables. But the scatterplot cannot always give a sufficient explanation for the data. One improved technique to enhance the scatterplot is called moving statistics (Cleveland and Kleiner, 1975). This graphical technique enhances the information on scatterplots by plotting a smooth curve, which summarizes the change in the distribution of the dependent variable given by the independent variable. Sperber (1987) applied this method of moving statistics to display the behavior of the species distribution as a function of wind direction. The curve that was smoothed by the predetermined interval of the wind direction provided the trends of the species concentrations with the wind directional behaviors. And Cleveland and Devlin (1988) introduced a locally weighted regression to smooth ozone data as a function of meteorological variables by a locally linear fitting factor. This method can potentially penetrate a regression study most deeply when the dependent variable is a nonlinear function of the independent variable. And this method falls 28 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. into a class of regression procedures that some call nonparametric regression (Cleveland and Devlin, 1988). Nonparametric regression is a set of techniques for estimating a regression curve without making strong assumptions about the shape of the true regression function; it can often provide insights about data that go beyond those obtained from classical parametric procedures (Altman, 1992; Eubank, 1994). The basic concept of nonparametric regression is to use the estimators to obtain the estimates that are smooth functions summarizing the relationship of variables. These smooth functions are estimated by a small neighborhood of the variables. This “local” concern gives the great flexibility o f fitting the curve to the data. Therefore, nonparametric regression analysis relaxes the assumption of linearity and substitutes the much weaker assumption of a smooth population regression function with little or no prior information available about the regression function (Eubank, 1999; Fox, 2000). This advantage makes nonparametric regression analysis suitable to smooth the plot of the species concentrations and wind direction for giving the precise direction of the emission source. The practical aspects and methodology of nonparametric regression analysis can be found in several books (e.g., Eubank, 1999; Hardel, 1990; Hollander and Wolfe, 1999; Silverman, 1986; Sprent and Smeeton, 2000). 29 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Nonparametric regression analysis has been a powerful and well-developed technique to provide the estimated regression function for data analysis. The kernel method is one of the local location estimation used in nonparametric regression analysis. Because of its simplicity and popularity, the kernel method has been commonly selected for nonparametric regression analysis. Several studies have shown the applications of the kernel estimation method in air quality research. Lorimer (1986) was the first to introduce the kernel method for developing a new way to simulate the transportation and diffusion of the atmospheric particles. Schwartz (1994) used the kernel method to smooth the features of diagnostic plots for the analysis of environmental epidemiology. Haan (1999) applied the kernel density estimation method to particle dispersion models for minimizing the sum of the variance and the bias of the predicted concentration distribution. And Henry et al. (2 0 0 2 ) applied the kernel method to the ambient observations for locating the nearby source of air pollution. These studies showed that nonparametric regression provides the benefits to give sharper results for air quality analysis and atmospheric science. The further applications of nonparametric regression could be as the assistant tool for receptor models. Receptor modeling uses observed concentrations to derive the source compositions and contributions. The nonparametric regression can be applied to estimates of the source contributions from receptor models. The results will help reconcile the emission inventory to observed concentrations. 30 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2.3 Evaluation of Emission Inventories The emission inventory is a national or regional database of air emissions emitted from stationary, area, and mobile sources. Emission rates are the significant inputs for air quality models because the models calculations base on these data along with meteorological data and theoretical chemical transformation algorithms. And regulation agencies rely on the results from air quality models for making decision of emission controlling strategies. Although the poor characterization of chemical mechanisms or other potential causes could impair the performance of the models, the prediction error of the models is often attributed to deficiencies in emission inventories. Therefore, how to indicate the deficiencies and update the adjustments for emission inventories is the major issue for inventory improvement. Targets for the improvement of emission inventories can be based on comparison of model results to observation concentrations. This top-down evaluation was performed to compare model estimates to ambient quality data for identifying areas of the inventory that warrant improvement (Funk et al., 2001; Mannschreck, 2002). And the technique o f wind-direction-dependent has been used to find the difference between the model calculations and field measurements for indicating the inaccuracy o f emission inventories (Aardenne et al., 2002). In addition to top-down evaluation, observation-based or receptor-based model is another technique to evaluate the emission inventory from field observations. For example, inverse air pollution modeling has been used to field observations for deducing the location and 31 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. strength of sources to evaluate the emission inventories (Mulholland and Seinfeld, 1995). Receptor modeling is another method to start with ambient concentrations at a receptor for seeking the apportionment of observed concentrations to link the source and receptor. This method can independently verify the emission inventory to give a better improvement. Receptor models provide an alternative and sometimes complementary top-down modeling methodology (Henry, 2002b). Receptor models do not require the knowledge of emission rates, wind, or other meteorological data to reconstruct the dispersion patterns on physical and chemical processes in the atmosphere. The models assume the species are conserved during transport between the source and the receptor. As a result, the samples collected at the receptor have the same chemical compositions as the ones released at the source. The amount of contribution coming from each source can be apportioned from ambient concentrations based on mass balance and the knowledge of the chemical compositions of the sources. Chemical Mass Balance (CMB) model is one of receptor models to serve for the inventory evaluation. Previous works used the source estimates from CMB models to evaluate the emission inventory (e.g. Scheff and Wadden, 1993; Kenski et al., 1995; Fujita et al., 1995; Mukund et al., 1996). These works have demonstrated the performance of CMB models for describing the emission sources from ambient observations 32 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. More advanced analysis can be done by multivariate receptor modeling and wind direction analysis for evaluating the emission inventory. Multivariate receptor modeling is another receptor modeling technique that derives the source contributions from observed concentrations without any prior knowledge o f the number and chemical compositions of the sources. The basic idea is that species from the source will be correlated; these correlations can be used to estimate the composition of the sources (Henry, 1991). This character makes multivariate receptor models to overcome the possible deficiencies in prior knowledge of the sources from sampling. And wind direction analysis enables to be applied to the source composition estimated from multivariate receptor models to provide the direction of emission sources. The estimated source direction along with the source contributions could indicate the important sources are excluded from the emission inventory. This was found to be the case when using simple bar chart to determine the direction o f significant sources for verifying the VOC emission inventory in the Houston Ship Channel (Henry et al., 1997). The plot of wind direction analysis can be improved by nonparametric regression method to have a continuous curve instead of a simple scatterplot or bar chart (Henry et al., 2002). The improved plot of wind direction analysis helps to estimate more accurate direction of the emission sources. Therefore, the method joining multivariate receptor modeling and wind direction analysis can provide the sharper 33 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. results for evaluating the emission inventory. Chapter 6 will have detailed application of this method for inventory evaluation. 34 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CHPATER 3 METHODOLOGY Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3.1 Local Average The nonparametric regression method is self-explanatory in determining a regression function when we know little about the data. This method relies more heavily on the data for getting the regression function than a parametric regression (Eubank, 1999). If there is a general relationship between two variables, an independent variable (X) and a dependent variable (Y), a reasonable approximation for a regression function will be any representative point close to the center of the band of response variables. A natural choice of approximation is the mean of the response variables near a point x, which is a given value of the independent variable. This approximation is called the “local average”, defined only by observations in a small neighborhood around x. This local average is very different from mean values (Hardle, 1990). The concept of estimating local average can give the improvement to present the bivariate relationship. Let a data set be assumed as (X,,Y; ), where / = 1,...,« observations and there is an unknown function between X and Y;. The relationship of X, and Y, can be expressed as Y,=w(X,)+e, (3-1) with the unknown function m and the observation error e,. The m(X,) is the value of 36 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the unknown function m at the points X;,...,or X„. The unknown function m is called the regression function or regression curve of X, and Y,. For a given fixed point X = x, the local average estimate at the given point is defined as n m(x) = n-lJ ^ w ni(x)Yi , (3-2) where (wm (x)} " = 1 is a sequence of weights that is dependent on the whole vector {X,} ” =1. According to equation 3-2, the plot can be smoothed by the average over a sliding window of width Ax centered at x. For the application of ambient observations, let the observed average concentration for the time period starting at U be Cj, where i = 1 observations. Further, let the measurement of resultant wind direction at the zth time period be Wu Within a fixed window AO , the average concentration in the sliding window centered at the direction 0 is where C(9) is the average concentration at the direction 0 , n is the number of data points inside the sliding window, and wm(x) = 1 for x - A 0 / 2 < x < x + A0/2 and zero otherwise. Figure 3-1 shows the results of the sliding window method applied n C(e)=n-''Zwje-wl)cl, (3-3) ;=1 37 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 100 150 200 wind direction 250 300 350 Figure 3-1 The curve of average cyclohexane concentration calculated by a 10° wide sliding window. 38 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. to cyclohexane concentrations measured at the Deer Park site in 1997. There is certainly an improvement showing the plot as a curve to describe the relationship of average concentrations and the wind direction measurements. However, the curve in Figure 3-1 is not smooth enough. The peak locations are still not easily determined by the curve. The problem is that the weight sequence { wni(x) } " = 1 gives equal weight to all the concentration measurements inside the sliding window. The equal weight is like to average the concentration measurements inside the sliding window. A more reasonable approach would be to give less weight to concentration measurements near the edges of the sliding window. Based on this idea, a shape function and density function that describe the weight sequence would be applied to the analysis. The kernel estimator is then introduced to our nonparametric regression analysis. This estimator will be discussed in detail in the following section. 3.2 Kernel Estimator Based on the equation 3-2, the weight sequence {wm (x)} " = 1 can be represented by the shape function and the density function with a scale parameter that adjusts the size and form of the weights near x. This shape function is often referred to as a kernel K (Hardle, 1990, p.24). The kernel K presented here is a continuous, bounded, and symmetric real function that integrates to one. 39 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The kernel K function can be described as / • o o j K(u)du = 1. (3-4) The weight sequence can then be defined by the shape function and the density function as wni= K h( x - X i) / f h(x), (3 -5) where f,(x )= n , ’ L K ^ x - X <\ (3-6) and Kh(u)=h'lK(u/h). (3 . 7 ) (u = x-X ,) 40 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The A'h(u) is defined as the kernel K with scale factor h. Depending on equation 3-6 and 3-7, the sequence of weights wn i can be rewritten as w =n- K ( H h tr a (3-8) Based on equation 3-2, the local average estimate is m h(x) n y — Y E * (—T-W _ H h n v — Y i= i (3-9) Equation 3-9 has been called the kernel estimate with scale factor h. The kernel K is determined by the parameter h, which is called the bandwidth or the smoothing parameter of K (Hardle, 1990, p.25). Now we can use the kernel method to express the estimates of species concentrations at a given wind direction. The wind direction is the azimuth angle, which is measured clockwise from north. The parameter A 0 is the selected smoothing parameter for kernel K. According to the equation 3-9, the estimated 41 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. average concentration at the direction 6 local average of measured concentrations estimated by the kernel method is defined as (3-10) where 9 is the given direction starting from 1 to 360, Wi and C, are the resultant wind direction and observed species concentrations at time period tu respectively. The C(9, A9) is the estimated average concentration at the given direction 9. There are many possible selections for kernel K, but the two most often used are the Gaussian kernel if(x) = (2^-)“1/2exp(-0.5x2), — oo < x < oo (3-11) and the Epanechnikov kernel. K(x) = 0.75(1 - x 2), -1<x<1. (3-12) 42 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Both of these kernels will give the maximum weight to the observations near 9 and less weight to observations further away. The major difference between these two kernels is that the Gaussian kernel is defined over an infinite domain and the Epanechnikov kernel is defined over a finite range. Thus, the Gaussian kernel is better for the wind direction and other circular data. For data limited to a finite range, the Epanechnikov kernel has less bias than the Gaussian kernel at the end points. Thus, the Gaussian kernel has been selected for our nonparametric regression analysis. Based on equation 3-10, the estimated average concentration at a given direction 9 is defined by using the Gaussian kernel as C(9,A9) ” 1 9 — W Z ~ 1 T = e x p (-°-5 (—^ — )2 Vi i= i V2 n A9 Z ~7z= exP(_ 0 -5 ( ~ r ^ ) ) (3-13) A 9 The smoothing parameter A9 has been defined in terms of the Full Width at Half Maximum (FWHM), an intuitive measure of the width of the kernel function (Figure 3-2). The A9 is simply defined as the full width of the peak measured at the point where the curve has fallen to half its value at the peak. 43 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 0.3 -4 ■ 3 ■ 2 1 0 1 2 3 4 FWHM —► ! Figure 3-2 Full Width at Half Maximum 44 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. For the Gaussian Kernel, the relationship of FWHM and smoothing parameter is defined as FWHM = 2.36xA0. (3-14) For example, 10-degrees is a common choice of bin size for wind direction analysis, i.e. the average concentration of species in every 10-degree bin. If the FWHM is 10, the smoothing parameter A0 is 10/2.36 = 4.24. Figure 3-3 shows the result of the Gaussian kernel estimator on the Deer Park cyclohexane data by using FWHM equal to 10. Figure 3-4 shows the polar plot of average cyclohexane concentrations versus wind direction measurements by using Gaussian kernel. The curve can be continued at the north. However, the smoothing parameter cannot be selected by subjective experience only. The smoothing parameter should be decided by a reproducible, quantitative algorithm to avoid any subjective analysis. A method of smoothing parameter selection will be discussed in detail in the following section. 3.3 Smoothing Parameter Selection The most important part of the nonparametric regression method is the selection of the smoothing parameter, the FWHM. If the FWHM is too large, the curve of 45 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 12 10 8 6 4 2 0 100 150 200 w in d direction 250 300 350 Figure 3-3 The curve of average cyclohexane concentrations versus wind direction by using a Gaussian kernel with a 10° FWHM. Data with wind speed <1 mile/hour are removed. 46 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 0 330 300 270 240 120 210 150 180 Figure3-4 The polar coordinator plot of average cyclohexane concentrations versus wind direction by using a Gaussian kernel with a 10° FWHM. Data with wind speed < 1 mile/hour are removed. 47 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. estimated average concentrations will be over-smoothed and peaks could be lost or not resolved. If the FWHM is too small, the curve will have many small, meaningless peaks dominated by noise or large peaks may be separated as false multiple peaks. There are several ways to select the best smoothing parameter for the data. One approach for smoothing parameter selection is called the cross-validation method (Hardle, 1990, p. 152). For each observed wind direction Wh j=l...n and its associated observed concentration Ch the cross-validation method uses equation 3- 10 to estimate the expected concentration, but leaving out the jth observation. The result can be given as C ^ , A0) w . - W A0 ^ Wt -W. (3-15) The optimal smoothing parameter is the one that minimizes V(A0) , the mean squared difference between the concentration estimated, leaving out one observation and the jth observed concentration. 48 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The equation can be described as n V(A0) = Z(C, -C /^A t)))2 (3-16) The minimum F(A 0)can be presented by plotting. Figure 3-5 shows the V(A0) plotted by the Deer Park cyclohexane data in 1997 with wind speed greater than 1 mph. The minimum value occurs at 7. This is close to the value of 10 used in Figure 3-3, but it indicates that the curve in Figure 3-3 might be over-smoothed. 3.4 Confidence Intervals The confidence intervals presented here are used to assist the peak identification. Peaks need to be judged if they are data noise or not before the further analysis. The confidence intervals of the curve can be calculated from a formula based on the asymptotic distribution of the kernel estimates, which is a normal distribution. The sample estimate of the variance o f the asymptotic distribution of C(&, A#) is given by (Hardle, 1990, pp. 98-101) (3-17) n A d m 49 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 46.2 46.1 45.9 45.8 ■ g 45.7 w 45.6 45.5 45.4 45.3 45.2 10 15 5 20 25 30 Full W idth a t Halh Maximum (FWHM) Figure 3-5 Cross-validation error curve of cyclohexane concentrations at the Deer Park site. The plot shows the optimal FWHM is equal to 7. 50 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. where “ 1 f / : 2(x)fi6c = ^ = , 2 yl7r and (3-18) (3-19) - 2 — " 0 - W - a (e)H »m r"L K (— i-)(c,-cw .m y . P-20) t=l ^ ( 7 Therefore, when we request 95% confidence intervals, a is equal to 0.025. And ca is the (lOO-a)-quantile of the normal distribution (1,96 for a = 0.025), the confidence limits on C(0, Ad) are given by C (0,A 0)±cas(0). (3-21) 51 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 14 12 10 6 o 4 2 0 0 50 100 150 200 250 300 350 wind direction Figure 3-6 The curve of average cyclohexane concentrations versus wind direction by using a Gaussian kernel with a 7° FWHM. Data with wind speed <1 mile/hour are removed. The gray region shows the 95% confidence interval. 52 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. So when a is equal to 0.025, the equation 3-21 gives a two-sided 95 % confidence interval. Figure 3-6 shows the 95% confidence intervals as the gray shaded area with the curve of the Deer Park cyclohexane data. The confidence intervals are used in a subjective way to judge if a peak is a merely noise or not. A small peak that has small confidence intervals is likely to be a real peak, while a peak with large confidence intervals is likely to be noise. Furthermore, if the estimated upper or lower confidence limits makes the peak is less or greater than the upper or lower limits of the adjacent peak, this peak cannot be identified as real. Therefore, in Figure 3-6 the small peak near 170° can be identified as real but the peaks near 270° and 290° are not. The peak near 220° is not an obvious call and requires the further analysis. 53 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CHAPTER 4 STUDY AREA AND DATA DESCRIPTION Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4.1 Study Area Houston, Texas, located in the seat of Harris County, is the fourth largest city in the United States and the largest one in the South and Southwest. It is located on the northern portion of the Gulf coastal plain, 50 miles from the Gulf of Mexico. Because of the flat terrain, the Houston area has excellent freeways and railroad systems that underlie business development and community transportation. In addition to the freeway and railroad systems, the system waterways in Houston are another factor in the city’s development. The surface water system in the Houston region consists of lakes, rivers, and an extensive system of bayous and manmade canals. The Houston Ship Channel is one of these manmade canals providing a link between interior Texas and the sea. This waterway can permit ocean-going vessels to traverse the shallow Galveston Bay all the way to Houston. Because of this character, Houston Ship Channel is one of the busiest waterways in the United States. Nearly half of the nation’s petrochemical production occurs along Houston Ship Channel and Galveston Bay’s shores. These industrial facilities not only constitute the largest petrochemical complex in the world, but also contribute huge emissions of VOCs and NOx that worsen the air quality in the Houston metropolitan area. Exceeded emissions of VOCs and NOx from anthropogenic sources accelerate the ozone formation in lower troposphere. For this reason, the Houston metropolitan area and its surrounding counties have been classified as one of nation’s severe 55 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ozone nonattainment areas due to extremely high ozone concentrations. In addition to worsen the ozone problem, some VOCs released from industrial sources also have been classified as air toxics. The potential community exposure to these toxic chemicals is another great concern of public health in the Houston area. Therefore, many studies have investigated the ozone problem and impacts of VOCs emissions in the Houston area and recommended the most effective strategy for making substantial progress for improving the air quality (e.g. Siddiqi and Worley, Jr., 1977; Walker, 1985; LaGrone, 1991; Evans et al., 1992; Miller and Sager, 1992; Kelly et al., 1993; Smith and Shively, 1995; Davis and Speckman, 1999; Hadjiiski and Hopke, 2000; Grover and Bradford, 2001; Kleinman et al., 2002). From these previous studies, reducing the emissions of VOCs is one suggestion to work out these air quality problems. In order to reduce the emissions of VOCs, locating, identifying, and quantifying the contributions of these VOC sources are the ways to achieve this objective. The industrial Houston Ship Channel region has distinctive stationary sources of VOCs, thereby providing a great area to investigate the approach for accomplishing VOC source verification. 4.2 Ambient Measurements Hourly ambient measurements used in this dissertation are obtained from the PAMS program. The PAMS program is part of the SIP monitoring network to establish a wealth of air monitoring information for ozone and its precursors monitoring in the 56 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ozone nonattainment areas. The air monitoring information provides long-term monitoring data on changes of atmospheric concentrations of ozone and its precursors for taking the development and evaluation of ozone control strategies and programs. In addition to ozone control strategies and programs, the PAMS program specifically helps to improve emission inventories, assist in evaluating the performance of photochemical models, and provide routine measurements of selected hazardous air pollutants. Data from PAMS also assist in deriving a more complete understanding of tropospheric ozone formation and transport. Furthermore, the data from PAMS can allow for the development of the complementary top-down modeling methodology such as receptor models. Through these efforts, we may move toward the best solution for this complex environmental problem. There are five PAMS monitoring sites in the Houston area. Hourly ambient measurements were obtained from two sites, Clinton Dr. and Deer Park, which are located near downtown and Houston Ship Channel respectively. These two sites are type 2 sites designed to be placed at the maximum ozone precursors impact area. In PAMS program, EPA has classified four types of monitoring sites based on their positions to collect the data for understanding and solving ozone nonattainment problem. The type 2 site is designed to be located immediately downwind of the area of maximum precursor emissions. This type of site is close to the precursors emissions mixture in order to obtain neighborhood scale measurements. Houston 57 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Ship Channel and the downtown area are the major regions with the maximum precursors emissions. The data collected from these two PAMS sites can monitor the magnitude and type of precursor emissions to estimate the maximum impact of precursor emissions and monitor the contribution of air toxic pollutants. The locations of these two PAMS monitoring sites and Houston Ship Channel are shown in Figure 4-1. The ambient data from these two sites consisted of the measurements of target VOCs, criteria pollutants, as well as surface and upper air meteorology. The period of data collection was from January 1, 1997 to December 31, 1997. These two sites measured 56 target VOCs including ten compounds classified as toxic air pollutants and three carbonyl compounds (formaldehyde, acetaldehyde, and acetone) on a hourly basis during the collection period. As data were obtained, quality control and assurance procedures were performed to ensure that the reported data were within a reasonable range or limit. Quality assurance procedures were guided by PAMS Performance Evaluation Program (U.S. EPA, 1997). These hourly concentration measurements of these 56 target VOCs plus total non-methane organic carbon (TNMOC) were reported by an in-site automated gas chromatograph. TNMOC is the sum of all identified and unidentified peaks in the chromatogram; the value provides the total amount of organic gases measured in an hour. Table 4-1 lists those 56 PAMS target VOCs. 58 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. V ‘ Figure 4-1 Map o f Houston area. The location of the Clinton Dr. site is at x and the Deer Park site is at +. Houston Ship Channel is located at the center of the map. The red solid lines are major interstate highways (I-10 and 1-45). VOC sources are shown as red circles with the area proportional to the emission rate. The data of emission is from AIRS emission inventory. v £ > Table 4-1 List of the PAMS target VOCs. species abbreviation ctiemical name AiKS Code 1 ethan ethane 43202 2 ethyl ethylene8 43203 3 propa propane 43204 4 prpyi propylene0 43205 5 acety acetylene 43206 6 nbuta n-butane 43212 7 isbtu i-butane 43214 8 t2bte trans-2-butene 43216 9 c2bte cis-2-butene 43217 10 13btu 1,3-butadiene 43218 11 npnta n-pentane 43220 12 ispna i-pentane 43221 13 lpnte 1-pentene 43224 14 t2pne trans-2-pentene 43226 15 c2pne cis-2-pentene 43227 16 2m2be 2-methyl-2-butene 43228 17 3mpna 3-methyl-pentane 43230 18 nhexa n-hexane 43231 19 nhept n-heptane 43232 20 noct n-octane 43233 21 nnon n-nonane 43235 22 ndec n-decane 43238 23 cypna cyclo-pentane 43242 24 ispre isopreene 43243 25 22dmb 2,2-dimethylbutane 43244 26 lhexe 1-hexene 43245 27 2m lpe 2-m ethyl-1 -p entene 43246 28 24dmp 2,4-dimethyl-pentane 43247 29 cyhxa cyclohexane0 43248 30 3mhxa 3-methylhexane 43249 31 224tmp 2,2,4-trimethylpentane 43250 32 234tmp 2,3,4-trimethylpentane 43252 33 3mhep 3-methylheptane 43253 34 mcyhx methyl-cyclohexane 43261 35 mcpna methyl-cyclopentane 43262 36 2mhxa 2-methylhexane 43263 37 lbute 1-butene 43280 38 3m lbe 3-metliyl-l -butene 43282 39 cypne cyclopentene 43283 40 23dmb 2,3 -dim ethy lbutane 43284 41 2mpna 2-methylpentane 43285 42 23dmp 2,3-dimethylpentane 43291 43 form formaldehyde0 43502 44 aceta acetaldehydeab 43503 45 acet acetone0 b 43551 46 nundc n-undecane 43954 47 2mhep 2-methylheptane 43960 48 m/pxy m/p-xylene 45109 49 benz benzene0 45201 50 tolu toluene0 45202 51 ebenz ethylbenzene0 45203 52 oxyl o-xylene0 45204 53 135tmb 1,3,5-trimethylbenzene 45207 54 124tmb 1,2,4-trimethylbenzene 45208 55 styr styrene 45220 56 123tmb 1,2,3-trimethylbenzene 45225 a Species are air toxics. b Species are carbonyl compounds. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Other data coming with organic species concentrations were criteria pollutants and air meteorological measurements. The measurements of criteria pollutants had hourly average concentrations of ozone, carbon monoxide, nitrogen oxides, and sulfur dioxide. And air meteorological measurements consisted of temperature, pressure, radiation, and wind data. Wind data included hourly average wind direction and speed, standard deviation of the wind direction, and hourly resultant wind direction and speed. Hourly resultant wind direction revealed the direction of wind measured clockwise from the north. However, over 50% of wind data at Clinton Dr. were missed. Deer Park had the more complete wind data in 1997. Initial wind data from Deer Park site consisted of 8207 hourly records of direction and speed measurements. We expect the wind data from Clinton Dr. site should be similar as the data collected at Deer Park site because the terrain of Houston is flat and the distance between these two sites is only 14.33 km. Based on these reasons, the wind data from Deer park site was used for data analysis at both sites. 4.3 Emission Inventories The emission inventory is a national or regional database of air emission information from federal, state, and local air agencies. This database contains the information on stationary, area, and mobile sources that emit criteria air pollutants, their precursors, and air toxic pollutants. The information includes source classification, source description, and source locations, source emission estimates. 61 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The emission estimates are usually gathered, reported on an annual basis for individual point and major sources, as well as county level estimates for area, mobile, and other sources. Emission inventories are used for a wide variety of purposes. Data from emission inventories are used to serve as the basis for modeling of predicted pollutant concentrations in ambient air, to identify sources and track the trends in emission over time, to evaluate the status of existing air quality models, to provide input for human health risk assessment studies, and to serve as a tool for supporting future trading programs. Emission inventories can provide the technical foundation for federal, state, and local programs designed to improve or maintain ambient air quality. Two types of emission inventories are used in this dissertation as the reference to examine the results obtained from receptor modeling and wind direction analysis. One is the self-reported emission inventory in the year of 1997, which was obtained from U.S. EPA’s Aerometric Information Retrieval System (AIRS) database. This emission inventory provided detailed descriptions, locations, and annual emissions of VOCs, NOx, SO2, and PMi0 of all stacks in each facility located within Harris County, Texas. Regulatory agencies and air quality models rely on these self- reported emission rates for making regulatory decision. The other detailed information includes the stack height, the stack diameter, the effluent gas velocity and temperature, and the stack status. Harris County is the part of Houston- Galveston ozone nonattainment area; this county covers the city of Houston and 62 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Houston Ship Channel. Emissions of hydrocarbons released from the petrochemical industrial region make Houston different from other urban areas. VOC emissions of all stacks are shown in Figure 4-1 along with the locations of the Clinton Dr. and the Deer Park sites. These VOC emissions were reported as total emissions of VOCs; there were no partial emissions of individual organic gases o f stacks in this emission inventory. Air toxic emission inventory is another type of emission inventory obtained from EPA’s Office of Air Quality Planning & Standards (OAQPS). This is a specific database to hold the air emission information of those sources releasing toxic compounds that are classified by the EPA. This database contains information including toxic chemical emissions and other waste management activities and reports annually by certain covered industry groups as well as federal facilities. The 1997 data of this emission inventory includes the facility name, facility description, facility location, and annual emissions of individual toxic organic compounds that are released in Harris County. The reported emissions of toxic compounds enable us to identify the dominant source as “hot spot” in the Houston area. Several air toxics have been measured their ambient concentrations at PAMS monitoring sites. Cyclohexane is an air toxic having a dominant source that contributes over 70% of total cyclohexane emissions in the Houston area. There will be a dominant peak shown in the plot of hourly measured cyclohexane 63 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. concentrations and wind direction measurements produced by nonparametric regression method. This cyclohexane dominant source reported in the inventory should lie on the direction given by the peak. The predicted location o f this dominant source can be determined by triangulation. Therefore, the distance between the predicted and reported location will give the empirical result to evaluate the performance of nonparametric regression method. The detailed processes and results are in the following section. 64 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CHPATER 5 APPLICATION OF NONPARAMETRIC REGRESSION FOR LOCATING THE LARGEST SOURCE Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 5.1 Objectives The success o f the nonparametric regression method can be judged by examining the location of the largest source from ambient observations. Nonparametric regression method is applied to smooth the plot of measured pollutant concentrations and wind direction measurements. The direction of the largest source of the pollutant is able to be estimated from the largest peak in the smoothed plot. Based on measured observations from two monitoring sites, the location of this largest source of the pollutant can be determined by the estimated direction from each site. Cyclohexane has been selected from the ambient observations because the dominant emission source, Phillips Petroleum Company, contributes over 70% of cyclohexane emissions in the Houston area. Therefore, the reported position of this dominant cyclohexane source is compared to the estimated location from ambient observations. The success of the location estimation will confirm the feasibility o f the nonparametric regression method to locate the largest source of cyclohexane emission. Cyclohexane is a colorless and volatile liquid with a slightly pungent odor resembling that of chloroform or benzene. This organic compound can be produced from the catalytic hydrogenation of benzene and fractional distillation of petroleum. Cyclohexane has many usages in industry including the manufacture of nylon fibers and nylon molding resin, as a solvent for paint, resins, varnish and oils, and the intermediate in the manufacture of other industrial chemicals. Because of its 66 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. toxicity and carcinogenicity, cyclohexane has been classified as one of the air toxics. The cyclohexane emission information is listed in the air toxic emission inventory. One company, Phillips Petroleum Company, is the dominant source of almost 70 percent of the total of cyclohexane emissions in Harris County. Therefore, the nonparametric regression method is applied to smooth the plot of cyclohexane concentrations and wind direction measurements measured at two sites. The smoothed curve is expected to show the largest peak that gives the direction associated with this dominant cyclohexane source. When two straight lines are drawn at the associated direction from each site, the intersection of these two lines should represent the location of the dominant cyclohexane source. Comparing the distance between the estimated location with the reported position of this dominant cyclohexane source will be the key to judge the success of the nonparametric regression method. 5.2 Data Screening The initial dataset had 6249 and 5717 hourly records of cyclohexane concentrations collected at the Clinton Dr. and Deer Park sites and 8207 hourly records of wind direction and speed measurements at both sites. However, not all records can be used for source location analysis. The initial data screening is used to remove the records with missing measurements and low wind speed. Because the local average concentrations are calculated by wind direction measurements, records are requested 67 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. to have both of cyclohexane measured concentrations and wind direction measurements for the nonparametric regression method. The records missing one of these measurements should be dropped from the dataset. In addition to missing measurements, the records also need to be screened by wind speed measurements. Because the wind direction is not well detected at lower wind speed, records with wind speed less than 1 mile/hour should not be factored into the analysis. It is desirable to identify and remove the records that missed cyclohexane concentrations or wind direction measurements and have wind speed less than 1 mile/hour from the initial dataset. After initial data screening steps, the measured concentrations of cyclohexane were then examined by using the scatterplot of cyclohexane concentrations versus wind direction measurements. The scatterplot was used to identify the possible outliers of measured cyclohexane concentrations with the simple visual inspection. During the monitoring period, some concentration measurements could be affected by very infrequently occurring conditions such as a large impact from an intermittent local source. These real but very infrequent events could confound the statistical description of the data. These data should be identified as outliers to be dropped from the dataset. Figure 5-1 and 5-2 show the scatterplots of measured cyclohexane concentrations and wind direction measurements at each site. In Figure 5-2, three outliers have been determined by the virtual inspection of the scatterplot because they had unusually high measurements compared to the other samples. 68 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 150 20 wind direction Figure 5-1 The scatterplot of hourly cyclohexane concentrations versus wind direction at the Clinton Dr. site in 1997. 69 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 800 700 600 -Q 500 £2. Q . 400 C O ( D c 300 200 100 M M M i M M M M r i J f a o 50 100 150 200 250 300 350 wind direction Figure 5-2 The scatterplot of hourly cyclohexane concentrations versus wind direction at the Deer Park site in 1997. The marked points were identified as outliers. 70 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. These three samples formed a small, self-consistent group (measuring period from 11/1 22:00 to 11/2 0:00). By checking the same measuring period at the Clinton Dr. site, no apparently high measurements were found. Therefore, these three records could be affected by the local source impact and should be dropped from the Deer Park dataset. On the other hand, there is no apparent outlier showing in Figure 5-1. No records have been eliminated from the Clinton Dr. dataset. After these data screening steps, there were 5722 and 5393 records of cyclohexane concentrations with wind direction measurements remaining in the final dataset of Clinton Dr. and Deer park site for source location analysis. 5.3 Results of Nonparametric Regression Smoothing The nonparametric regression method is applied to measured cyclohexane concentrations and wind direction measurements for source location analysis. Missing data and outliers have been screened from all data. The Gaussian kernel estimates were then carried out to calculate the average cyclohexane concentrations for each whole degree from 1 to 360. Based on the cross-validation method, the smoothing parameter, FWHM, was determined to be 7 for data at the Deer Park and 10 for data at the Clinton Dr. site. Figure 5-3 and Figure 5-4 show the plot of average cyclohexane concentrations versus wind direction measurements produced by the nonparametric regression method at two sites using optimal FWHM values. The final result was a smooth curve with 95% confidence intervals. 71 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 14 12 10 O -Q Q - 3 8 c o ‘ 4 - * 2 4 - » g 6 o 1 = o o 4 2 0 0 50 100 150 200 250 300 350 wind direction Figure 5-3 Nonparametric regression curve of average cyclohexane concentrations at the Deer Park site by using the Gaussian kernel with a FWHM equal to 7°. Data with wind speed less than 1 mile per hour are removed. The gray region is the 95% confidence interval. 72 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. C L o 3 0 50 100 150 200 250 300 350 wind direction Figure 5-4 Nonparametric regression curve of average cyclohexane concentrations at the Clinton Dr. site by using the Gaussian Kernel with a FWHM equal to 10°. Data with wind speed less than 1 mile per hour were removed. The gray region is the 95% confidence interval. 73 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The smooth curve produced by nonparametric regression method is used to estimate the wind direction that gives a local maximum of average cyclohexane concentrations. The wind direction and the value of local maximum can be calculated by interpolation, which is used to predict an unknown value by taking an average of known values at neighbor points. If the data equally changed with an interval h, the interpolated point (x0 + ph) can be estimated by f(xo), f(xo + ph ), and / (xa - ph) from the following formula: ( /( * o - Ph)~f(x0+ph)) 2 ( /O o - Ph) ~ 2/ ( x 0) + / ( x 0 + ph)) The value at the interpolated point is then given by f i x o + p) * 0.5p (p -1 ) /( x 0 -1 ) + (1 - p 2)f(x 0) + 0.5p(p + 1 ) /( x 0 +1) . (5-2) Both these formulas are from Abramowitz and Stegun (1972). / Because the average concentrations are calculated based on each whole degree, the interpolation is able to find the direction that gives the local maximum of average concentrations at a specific degree with greater precision than 1°. The closest point to a local maximum is C(x0) at the degree x0; the interpolated direction that gives 74 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. this local maximum of average concentration is at (x0 + p ) . Based on equation 5-2, the h is equal to 1 andp can then be estimated by C(x0) , C(x0 -1 ), and C(x0 +1) as p = (C (X p-l)-C (x0+l)) 2(C(x0 - 1 ) - 2C(x0) + C(x0 +1)) Thus, the interpolated value of this local maximum average concentration C(x0 + p) is calculated by C(x0 + p) « 0.5p(p - l)C(x0 -1 ) + (1 - p 2)C(x0) + 0.5 p(p + l)C(x0 +1). (5-4) In order to estimate the direction of the concentration peak, the edges of the peak were decided by the first and second derivative of the curve. The estimated uncertainty of the curve was then used to identify whether the peak was real or not. When the peak was identified, the direction and its value of average concentrations were calculated by interpolation. Table 5-1 shows the estimated directions and average concentrations of the four largest identified peaks shown in Figure 5-3 and Figure 5-4. The uncertainty of peak locations was estimated by 1000 bootstrap calculations. The peak locations were estimated by normal distribution with the standard deviation given from these 1000 calculations. The azimuth range was estimated by standard deviation with 95% confidence intervals. 75 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Depending on Table 5-1, the largest peak of average cyclohexane concentrations was given at around 328° at the Deer Park site and around 87° at the Clinton Dr. site. These estimated directions indicate that the largest emission source of cyclohexane should be located at 328 azimuth of the Deer Park site (or at 87 azimuth of the Clinton Dr. site.). The estimated average concentration of the largest peak is 11.89 ppbC at Deer Park and 7.53 ppbC at the Clinton Dr. site. The smaller value of the largest peak at the Clinton Dr. site implies that this largest emission source could be closer to the Deer Park site. Based on the estimated direction of each site, the predicted location of this largest cyclohexane source can be predicted at the intersection of these two estimated directions given by the largest peak. The result of this largest source identification will be discussed in the later section. The second and third largest peaks shown in Figure 5-3 are located at 20° and 44° respectively. These peaks indicate the other possible directions associated with smaller cyclohexane sources. Considering the azimuth, these smaller cyclohexane sources are located in the eastern sector of the Clinton Dr. site. Consequently, the emissions from these cyclohexane sources might be part of the largest peak shown in Figure 5-4. The fourth largest peak in Figure 5-3 is at around 209°. This peak could indicate another location of the smaller cyclohexane source. And the emission from this cyclohexane source should be much smaller than the other three larger sources. 76 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The second largest peak shown in Figure 5-4 is located at 161 degrees, which points to an area with many nearby sources to the Clinton Dr. site. The distance between the Clinton Dr. site and these sources are within 5 km. Several major refineries and petrochemical industries are located in this area. Therefore, this peak could associate with these nearby sources. However, the cyclohexane emissions of these sources cannot be identified by Deer Park data. By the locations of these facilities, the azimuth of these sources to the Deer Park site is around 295°. As a result, there is a peak showing at this direction in Figure 5-3. But this peak cannot be identified as a real peak because of its estimated uncertainty. One possible reason could be due to the fact that winds from this direction are infrequent. Another reason is these sources are too far away from the Deer park site. The average of distance is 14 km (8.75 miles) from these sources to the Deer Park site. This distance is greater than the average of the species traveling distance within one hour. Most of wind speed measurements are less than 9 miles/hr. Thus, the concentrations of species cannot be measured at the Deer Park site in many sampling period, even the wind blowing from the sources to the site. Insufficient data make greater uncertainty. Under this circumstance, we cannot conclude that there is a peak at this direction. 5.4 Comparison to Known Sources The estimated peaks in Table 5-1 must compare with the sources reported from the emission inventory. Table 5-2 lists the information of reported cyclohexane sources 77 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Table 5-1 Largest peaks in the curve of average cyclohexane concentrations versus wind direction in Figure 5-3 and Figure 5-4. Maximum Deer Park Maximum Clinton Dr. Azimuth estimated uncertainty Azimuth Range Azimuth estimated uncertainty Azimuth Range peak #1 11.89 328.68 1.09 326.54-330.82 7.53 87.19 1.17 84.90-89.48 peak #2 5.12 20.66 1.19 18.33- 22.91 4.52 161.00 1.98 157.12- 164.88 peak #3 5.00 44.47 4.45 35.75- 53.19 2.15 312.25 1.71 308.90-315.60 peak #4 2.28 209.61 2.63 204.46 - 214.76 1.95 249.01 1.99 245.11 -252.91 o 00 in Harris County, Texas from Air Toxic Emission Inventory. This table provides a summary of major cyclohexane sources about their annual emissions, positions, the azimuth and distance to both monitoring sites. In Table 5-2, one company, Phillips Petroleum, contributes almost 70 % of the total emissions of cyclohexane in Harris County. The azimuths of this dominant cyclohexane source are 83° and 330° to the Deer Park and Clinton Dr. site respectively. Therefore, the largest peak shown in Figure 5-3 and Figure 5-4 should be associated with this dominant cyclohexane source and shown at the azimuth given by Table 5-2. Not surprisingly, at both sites the largest peak is consistent with the location of this dominant emission source reported from the inventory. Although the estimated azimuth range of the largest peak at the Clinton Dr. site does not exactly cover the azimuth in Table 5-1, the estimate is still very close. This situation could be due to the low wind speed effects. A lot of sources locate at this azimuth to the Clinton Dr. site. The cyclohexane measurements could be affected by the emissions from the other sources during the low wind speed period. Under this circumstance, the data with higher wind speed should avoid the low wind speed effects and give more consistent results of estimated directions. Wind speed data can be a parameter to screen the concentration measurements before taking nonparametric regression analysis. The result of data screened by wind speed will be discussed in section 5-5. In Figure 5-3, the second and third largest peak is close with the same estimated average concentrations. But the second largest peak at 20° does not correspond to 79 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Table 5-2 Emissions of cyclohexane of 1997 in Harris County, Texas. Facility Name Release Type Total Release Percent of (lbs/year) Total --? - w* Latitude Longitude Accuracy (m) Deer Park Distance Azimuth (km) Ginton Dr. Distance Azimuth (km) PHILLIPS PETROLEUM CO. STACK 167000 56.52% 29.74 95.18 50 330.27 9.25 83.25 7.91 PHILLIPS PETROLEUM CO. FUG 33000 11.17% 29.74 95.18 50 330.27 9.25 83.25 7.91 ENICHEM AMERICAS INC. STACK 16402 5.55% 29.77 95.02 11000 43.24 15.65 79.44 23.56 ENICHEM AMERICAS INC. FUG 1193 0.40% 29.77 95.02 11000 43.24 15.65 79.44 23.56 EXXONMOBIL BAYTOWN REFINERY. STACK 15282 5.17% 29.74 95.01 80 56.33 14.05 88.32 24.15 EXXONMOBIL BAYTOWN REFINERY. FUG 3777 1.28% 29.74 95.01 80 56.33 14.05 88.32 24.15 LY ONDELL-CITGO REFINING L.P. FUG 8472 2.87% 29.72 95.23 50 298.79 11.23 123.13 3.11 LYONDELL-CITGO REFINING L.P. SHELL CHEMICAL L.P. SHELL CHEMICAL L.P. STACK STACK FUG 7509 7000 820 2.54% 2.37% 0.28% 29.72 95.23 50 298.79 11.23 123.13 3.11 VALERO REFINING CO. STACK 6628 2.24% 29.72 95.25 20 296.43 13.48 161.34 1.17 VALERO REFINING CO. FUG 2323 0.79% 29.72 95.25 20 296.43 13.48 161.34 1.17 WESTHOLLOW TECH. CENTER STACK 6365 2.15% 29.73 95.63 11000 277.34 49.19 268.63 36.36 WESTHOLLOW TECH. CENTER FUG 1814 0.61% 29.73 95.63 11000 277.34 49.19 268.63 36.36 MILLENNIUM PETROCHEMICAL INC. FUG 4360 1.48% 29.71 95.07 80 49.40 7.60 96.72 18.34 MILLENNIUM PETROCHEMICAL INC. CROWN CENTRAL PETROLEUM STACK 1626 0.55% 29.71 95.07 80 49.40 7.60 96.72 18.34 CORP. CROWN CENTRAL PETROLEUM FUG 3594 1.22% 29.72 95.21 50 308.00 9.84 102.59 4.81 CORP. STACK 655 0.22% 29.72 95.21 50 308.00 9.84 102.59 4.81 FMC CORP. BAYPORT PLANT FUG 2576 0.87% 29.63 95.04 80 116.11 9.33 118.25 23.65 AKZO NOBEL CHEMICALS INC. STACK 1244 0.42% 29.70 95.09 80 45.64 5.48 101.27 16.69 EXXONMOBIL CHEMICAL CO. STACK 1195 0.40% 29.75 95.02 500 49.91 13.92 85.35 23.17 SPECIFIED FUELS & CHEMICALS Total Em issions STACK 578 295492 0.20% 29.80 95.12 600 1.64 14.42 60.33 14.79 00 o any source given in Table 5-2. The third largest peak is shown at 40°, which corresponds closely with the location of Enichem Americas Inc. Enichem Americas Inc. is the second largest source accounting for about 5.77 % of the total emissions of cyclohexane in Harris County. However, the distance from this source to Deer Park is 15.65 km away with an accuracy of 11 km. This peak cannot be identified to associate with this source because of the accuracy of this distance. The estimated average concentration of the remaining small peak is much smaller than the other three peaks and this small peak does not correspond to any reported source given in Table 5-2. In Figure 5-4, the second largest peak is at around 160°, which corresponds well to Valero Refining CO. in the inventory. Although Valero Refining CO. accounts for 3.15 % of the total emissions of cyclohexane in this area, this source is located only 1.17 km away from the Clinton Dr. site. As a result, it seems reasonable to associate this second largest peak with this nearby cyclohexane source. The remaining two smaller peaks in Figure 5-4 do not correspond to any reported source in Table 5-2. These smaller peaks might indicate the emission from a non-industrial source such as roadway emission. But there is no information of facilities regarding these smaller peaks available. 81 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 5.5 Location of the Largest Source The largest peak at both sites has been identified to associate with the dominant source in Table 5-2. The estimated location of this dominant source can then be determined from the azimuth of the largest peak. All calculations involving azimuth, longitude, latitude, and distances were performed using functions from the Matlab Mapping Toolbox. The location of the Deer Park and the Clinton Dr. sites from the AIRS database are 29.6694 N, 95.1281 W, and 29.7333 N, 95.2569 W, respectively. Based on these positions and the azimuth of the largest peak in Table 5-1, the estimated locations of this dominant source is 29.7368 N, 95.1752 W. The distance between this estimated location and the reported position given from Table 5-2 is about 0.546 km. Figure 5-5 shows the estimated location of this dominant cyclohexane source. The red circles reveal the stack positions of Phillips source with the area proportional to the emission rates. This map shows the estimated location of this dominant source is in agreement with the inventory. As the result in Table 5-1, the estimated direction corresponds to an error of 1.09° from the Deer Park site, or 1.17° from the Clinton Dr. site. It is better than the result that could be obtained from the bar chart, which would be to estimate the location with ±10°. Consequently, these confirm that the nonparametric regression method is a major improvement to estimates the location of this dominant cyclohexane source by using wind direction and concentration measurements. 82 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Ii V ' Figure 5-5 The estimated location from nonparametric regression method is marked as black dot. The location of the Phillips Petroleum source in the inventory is shown as *. The Clinton Dr. site is located at x and the Deer Park site is at +. The red circles are the VOC emissions from the stacks of Phillips Petroleum Company. The dash lines show the directions estimated from two sites. 00 U ) A more accurate estimate for locating this dominant source can be achieved by screening the data with wind speed. Wind speed measurements are considered as the pollutant traveling distance within one hour. Thus, a clear impact will be seen if the travel time from the source to the sampler at the site is less than one hour. Based on this knowledge, the concentration measurements at the Deer Park site are restricted to periods with wind speed greater than 6 miles/hour (9.6 km/hour) because the site is 9.25 km away from the Phillips source. And the concentration measurements at the Clinton Dr. site are also restricted to periods with wind speed greater than 5 miles/hour (8 km/hour) because the distance is 7.91 km from the site to the Phillips source. The distance between the site and the Phillips source is used to screen the data for the advanced analysis. The results of the data screened by higher wind speed are shown in Figure 5-6 and Figure 5-7. Figure 5-6 and Figure 5-7 show the plot of average cyclohexane concentrations versus wind direction measurements at higher wind speed. The smoothing parameter FWHM was determined as 5° at the Deer Park and 10° at the Clinton Dr. site. We can find the plots are very similar to the ones shown in Figure 5-3 and Figure 5-4, but the peaks are much sharper. In Figure 5-6, the largest peak is much sharper than the same peak in Figure 5-3 because the higher wind speed limits available concentrations for nonparametric calculation. But the higher wind speed screening does not change the largest peak in Figure 5-7 significantly. Table 5-3 84 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 20 18 16 14 10 8 6 _ 4 2 0 0 50 100 150 200 250 300 350 wind direction Figure 5-6 Nonparametric regression curve of average cyclohexane concentrations at the Deer Park site by using the Gaussian Kernel with a FWHM of 5°. Data are restricted to periods with wind speed greater than 6 miles per hour (about one hour travel time from the Phillips source to the site). The gray region is the 95% confidence interval. 85 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. C L C L 0 50 100 150 200 250 300 350 wind direction Figure 5-7 Nonparametric regression curve of average cyclohexane concentrations at the Clinton Dr. site by using the Gaussian Kernel with a FWHM equal to 10°. Data are restricted th periods with wind speed greater than 5 mile/hour (about one hour travel time from the Phillips source to the site). The gray region is the 95% confidence interval. 86 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Table 5-3 Largest peaks in the curve o f average cyclohexane concentrations versus wind direction in Figure 5-5 and Figure 5-6. Maximum Deer Park Maximum Clinton Dr. Azimuth estimated uncertainty Azimuth Range Azimuth estimated uncertainty Azimuth Range peak#l 14.65 329.26 1.55 326.22 -- 332.30 7.16 86.55 1.72 83.18- 89.92 peak #2 4.66 43.26 1.43 40.46 -- 46.06 4.24 160.06 1.86 156.25 - 163.83 peak #3 2.18 20.75 2.80 15.26 -- 26.24 1.15 332.89 2.16 328.81 -336.97 peak #4 1.74 169.11 0.60 167.93 ~ - 170.29 1.00 240.42 1.79 236.89 - 243.88 O O '- O lists the estimated directions and average concentrations of the four largest identified peaks shown in Figure 5-6 and Figure 5-7. Using the positions of sites and the azimuth o f the largest peak in Table 5-3, the new estimated locations of this dominant source is 29.7376 N, 95.1747 W. The distance between this new estimate and the reported position given from Table 5-2 is about 0.464 km. This distance shows the estimated location is closer to the reported location of this regression dominant source after using higher wind speed to screen the measurements. Since the nonparametric method has been validated to estimate the location o f the source by using the wind direction and concentration measurements, wind speed would be another parameter we should consider applying to the source location analysis next. 5.6 Discussion Nonparametric regression method has been applied successfully to the wind direction analysis for estimating the largest emission source by using measured concentrations and wind direction measurements. It is logical to extend this method to reveal the directional pattern of the chemical species that is released from many sources. In addition to applying for single chemical species, this method can also be applied to the source contributions estimated from receptor models. Receptor models derive the source contributions along with the source compositions to verify the emission inventory. Previous work relied on simple bar charts to determine the direction of sources by using estimates from receptor models for verifying the 88 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. emission inventory in the Houston area (Henry et al., 1997). The application of this method should present the sharper results to assist in this work. The other applications of nonparametric regression techniques could be also used in other air quality applications, such as trend analysis o f time series or simply providing data smoothing for exploratory analysis of air quality data. This work has demonstrated the usefulness of the nonparametric regression method for air quality data analyzed by wind direction. Nonparametric regression techniques have a complete package of estimating processes and statistical analysis to provide the accurate determination of the wind direction of maximum concentration. However, ground level concentrations are also functioned by factors other than wind direction. For elevated sources, ground level concentrations can be a complex function of emission rates, wind speed, and atmospheric stability. In section 5-5, the data screened by higher wind speed gives the better direction estimates. This indicates wind speed is another factor that affects the measured concentration at the monitoring site. An advanced method that includes wind speed would be desirable. Simultaneous nonparametric regression of concentrations on wind direction and wind speed is possible and could help throw light on, among other things, the distance to the sources and whether the sources are ground level or elevated. This will be the subject of a sequel to the application of nonparametric regression techniques. 89 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CHPATER 6 APPLICATION FOR VOC EMISSION INVENTORY EVALUATION Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 6.1 Introduction The emission inventory is a national or regional database of air emission information from Federal, State, and local air agencies. This database includes the emission rates of criteria air pollutants, their precursors, and air toxic pollutants that are emitted from stationary, area, and mobile sources. The VOC emission inventory is the one providing the emission rates of organic gases from those anthropogenic sources. Photochemical models and regulatory agencies rely on the emissions rates of these organic gases from the VOC emission inventory. Any deficiency of emission rates of these organic gases could severely impact the results of photochemical models and regulation decision. Thus, the inventory evaluation is one of significant issues in the air quality studies. Receptor modeling technique is a method applied to explicate the individual component and the strength of sources based on the ambient observations. The results from receptor models can compare with the emission inventory for the source evaluation. Several studies have used the estimates from receptor models to evaluate the emission inventory (e.g. Scheff and Wadden, 1993; Kenski et al., 1995; Fujita et al., 1995; Mukund et al., 1996). This application is applying the nonparametric regression method to source contributions estimated from multivariate receptor modeling for showing the relationship of wind direction to the estimated source contributions. The ultimate goal of this work is then to compare the emission inventory in Houston area based on the results of multivariate receptor modeling and nonparametric regression 91 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. method. Multivariate receptor modeling is one kind of receptor models using statistical methods to derive the source contributions as well as the source number and compositions from ambient observations themselves. Henry et al. (1997) used a less advanced form of multivariate receptor modeling to analyze the PAMS data from Houston Ship Channel area. In this earlier work, the directions of emission sources were determined by source contributions from the model along with the simple wind direction bar chart. They found the significant difference with the VOC emission inventory based on the results of the model and wind direction analysis. Here an advanced multivariate receptor modeling and analysis software, Unmix, has been developed by Henry (2001). This analysis software can derive the number, composition, and contribution of sources by selected species based on ambient observations. Nonparametric regression of source contributions on wind direction is used to locate the directions of nearby sources (Henry et al, 2002). Therefore, Uumix and nonparametric regression method is applied to analyze the 1997 ambient observations collected from two PAMS sites in Houston area. The results determined by Unmix and nonparametric regression method will give the great assist to evaluate the emission inventory. 6.2 Description of Unmix multivariate receptor model Multivariate receptor modeling is the term applied in the field o f receptor models to the solution o f the general mixture problem (Henry, 1991). For a set of air quality 92 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. data, let Cv , i =1, ... m, j - 1, ... n, be the measured species concentrations collected at a monitoring site for m species and n sampling periods. If the atmospheric species are conservative, the principle of mass balance can be applied. Let there be N sources, aik be the mass fraction of species i in source k (source composition). Then the mass balance of species / can be written as N = Z a>A > /=1> J =1> (6-!) *:=i Equation 6.1 is the basis of all receptor models. In this equation, Cfj is the ambient measurement of the species / in sample j ; Sk j is the total amount of emissions contributed from source k in sample j (source contribution). Thus, the source contribution Sk ] can be estimated from the ambient measurements Ctj collected at the site and externally supplied source composition aik (i.e. CMB model). Or the source contribution Sk j can be estimated as well as the source composition aik only relying on the ambient measurements C,y (i.e. multivariate receptor model). Obviously, the challenge of the CMB model application is to obtain the source profiles from other external supplies. Another common problem in the CMB model 93 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. is the near-collinearity when estimating S k ] by using Cy and given aik (Henry, 2002b). Hence, multivariate receptor models serves as an alternative to overcome these problems. Multiple observations taking from field sampling can provide the information about the number and composition of the sources. This notion helps multivariate receptor models not need any prior knowledge about the number and compositions of the sources. As a result, multivariate receptor models can estimate the source contribution as well as the source number and composition. A history of this approach can be found in Henry, 1997. Unmix is a multivariate receptor modeling and analysis software and originally written in the MATLAB scientific programming language application. A standalone version that does not required MATLAB is available as EPA Unmix version 2.3 for Windows, which has been released by the U.S. E.P.A. and is available on request. This work uses Unmix version 2.4. Unmix provides the solution for the air data that are assumed to be a linear combination of an unknown number of sources of unknown composition, which contribute an unknown amount to each sample (Henry, 2001). In this case, the data points can be presented in the space spanned by m arbitrary dimensions when the number of species m is greater than the number of sources N. Unmix model uses Numfact algorithm (Henry et al., 1999) and edge-finding algorithm (Henry 2002a) to estimate the number of sources A and to find the data “edges” (or the hyperplanes) in a (N-1) dimensional space for estimating the contributions and the compositions of these N sources. The matrix 94 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. operation of singular value decomposition (SVD) is used to provide the link between these two algorithms. Based on these fundamentals, Unmix model gives the solutions to any problem in which the data are a convex combination of underlying factors (Henry 2002a). The application o f Unmix model to analyze the data of Phoenix PM2.5 aerosols can be found in Lewis et al., 2002. Unmix is a data self-explanation software to estimate the source number, source compositions, and source contributions based on ambient observations. For using this software, the most important decision made by users is the selection of species to be used in the model. The comprehensive Unmix user’s manual provides the discussion of species selection and suggestion of alternative approaches. Several air quality studies also provided the useful information for species selection (e.g. Shreffler, 1993; Henry et al., 1994; Lewis et al., 1998). The results from Unmix are given based on selected species and the samples having complete records of these selected species. Before running the model, Unmix provides several useful advanced operations for diagnosing the data to assist in developing the best possible solution of the data; the function of diagnostic plots is one of them. This function produces the diagnostic scatterplots of all selected species against species set as a tracer. The diagnostic plots can facilitate to identify the possible outliers and other questionable data points. Usually, TNMOC is set as a tracer to produce the diagnostic plots against the other selected species for identifying the outliers and 95 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. r Houston S {Clinton Dr. site Galena: Park ^VVest: U niversity PljBce j:::. . •: : S o u th si d e P la c e ! ? ; Deer Park site : Porte i.South Houston f \ < & ? £ & # ¥ $ , ^Bfbo kS'tte-Village SReariarvl ^ Q e a f L a k e kNass«u'Bay Friendswood ^Highlands ■ - ...................... ,M en a r sS}S S i1 - - - ! ' tviw » & * ■ - * ? — ............. % * ? < > .co«, ■ ■ ■ ■ ’ • ' * - T : L„/. ; \ * v v ( j a r ^ / f ^Baytown / .jp 1996 P«Lazmt Str«titol£s USA: : y '^ T e u * f? , - .: - - : X ; : 0 * * ^ :: 3 | V ¥ ? - ; :-rr f'i.lf ia tw e C ity;; :: l^ a y l e r £ a k e Y iflS g * ;: < a > * S a B b w * ! [ - I 4 6 J : ; $ .K e n r t a l’ f - x Figure 6-1 Map of Houston area. It shows the spatial relationship of major highways (I-10 and 1-45) and two PAMS monitoring sites. Houston Ship Channel is parallel with route 225. n o O N questionable data points in the dataset. The detailed processes will be described in the following section. 6.3 Data Description and Treatment Ambient observations in Houston area were obtained from two PAMS sites, Clinton Dr. and Deer Park site. These two sites are located at the downwind area of the Houston Ship Channel, the region of the petrochemical complex. The distance between these two sites are about 16 km (or 10 miles). Figure 6-1 is the map showing the major highways around the Houston metropolitan and the locations of Houston Ship Channel and two monitoring sites. Ambient measurements of volatile organic compounds were measured by on-site automated gas chromatography. The data were measured for the period from January 1, 1997 to December 31, 1997 at both sites. The initial data from Clinton Dr. consisted of hourly average concentrations of 63 organic species plus total non-methane organic carbon (TNMOC). Another data obtained from Deer Park had hourly average concentrations of 64 VOCs without TNMOC. The TNMOC measurements of Deer park were rebuilt by the sum of all identified species in samples. In addition to VOC concentrations, the data from Deer Park also reported wind data as hourly resultant measurements of wind direction and speed. These wind data from Deer Park were used for wind direction analysis to both data sets. Reported VOCs data from the Clinton Dr. and Deer park site are listed in Table 6-1 and Table 6-2. 97 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 6-1 Basic information o f species measured at the Clinton Dr. site. species N on-m issing # trnnoc 5197 ethane 7043 ethylene 7038 propane 7043 propylene 6950 acetylene 6962 n-butane 7018 i-butane 6402 trans-2-butene 7022 cis-2-butene 6885 1,3-butadiene 4878 n-pentane 6452 i-pentane 6717 1-pentene 4975 trans-2-pentene 6377 cis-2-pentene 4668 3 -methyl-pentane 7005 n-hexane 6950 n-heptane 6862 n-octane 6742 n-nonane 6474 oyclo-pentane 5888 isopreene 5234 2,2-dimethylbutane 6237 2-methyl-1 -pentene 3617 cyclohexane 6249 3 -methy lhexane 6789 2,2,4-trimethylpentane 6802 2,3,4-trimethylpentane 6531 3-methylheptane 6044 methyl-cyclohexane 6845 methyl-cyolopentane 6720 2-methylhexane 6553 1-butene 6218 3-methyl-1 -butene 6843 cyclopentene 3954 2,3-dimethylbutane 6993 2-methylpentane 7043 2,3-dimethylpentane 4172 2-methylheptane 6020 m/p-xylene 6957 benzene 6974 toluene 6977 ethylbenzene 6747 o-xylene 6872 1,3,5-trimethylbenzene 6388 1,2,4-trimethylbenzene 6868 n-propylbenzene 5327 i-propylbenzene 3827 o-ethyltoluene 4848 m-ethyltoluene 5416 p-ethyltoluene 3515 m-diethylbenzene 4767 p-diethylbenzene 4988 styrene 3735 1,2,3 -trimethylbenzene 5355 mean (ppbC) Standard deviation (ppbC) 421.28 768.10 24.69 25.45 12.03 19.38 24.51 37.72 9.48 28.46 2.50 3.60 90.96 418.18 101.11 519.50 2.49 4.65 1.76 3.63 3.04 7.71 11.06 16.03 23.23 30.83 1.20 2.39 1.95 4.05 1.09 2.19 4.75 7.15 7.73 16.74 2.83 4.89 1.85 3.17 0.90 0.92 4.59 35.78 1.16 1.45 0.96 2.14 0.86 2.66 2.67 4.95 2.64 3.60 7.66 26.85 3.10 11.34 1.09 1.61 2.45 3.30 3.28 5.81 2.21 2.88 2.78 6.52 2.49 5.42 0.79 1.27 2.67 6.98 6.67 9.78 1.89 2.82 0.94 1.24 8.36 12.75 5.08 10.22 13.23 21.93 2.21 3.42 3.13 4.58 0.97 1.17 2.94 3.37 0.67 0.66 0.67 2.39 0.77 0.78 2.12 2.23 1.09 1.07 1.25 9.16 0.90 0.97 1.61 3.25 1.51 1.99 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 6-2 Basic information o f species measured at the Deer Park site. species Non-missing # mean (ppbC) standard deviation (ppbC) tnmoc 26 121.50 103.74 ethane 6235 28.46 34.20 ethylene 6204 10.88 23.05 propane 6235 27.68 80.92 propylene 5907 15.07 39.93 acetylene 6219 2.42 4.18 n-butane 6235 16.10 29.28 i-butane 6235 13.41 22.37 trans-2-butene 4198 1.34 1.79 cis-2-butene 6084 0.85 1.48 1,3-butadiene 4878 1.66 6.35 n-pentane 6235 6.32 10.38 i-pentane 6235 11.63 21.59 1-pentene 4869 0.73 1.57 trans-2-pentene 5303 0.79 1.73 cis-2-pentene 2791 0.58 1.04 3-methyl-pentane 6229 2.58 3.74 n-hexane 6205 5.19 8.31 n-heptane 5929 1.29 1.76 n-octane 5882 0.80 1.04 n-nonane 5386 0.42 0.51 n-decane 4520 0.45 0.55 cyclo-pentane 5645 0.82 1.06 isopreene 4671 0.93 1.01 2,2-dimethylbutane 5150 0.56 0.74 cyclohexane 5717 2.71 15.09 3-methylhexane 5994 1.14 1.57 2,2,4-trimethylpentane 6020 2.04 4.10 2,3,4-trimethylpentane 5628 0.77 1.66 3 -methy lheptane 5027 0.50 0.65 methyl-cyclohexane 6203 1.85 2.78 methyl-cyolopentane 6002 2.08 3.76 2-methylhexane 5578 0.85 1.19 1-butene 5824 1.69 5.12 3 -methyl-1 -butene 6192 0.91 2.08 2,3 -dimethylbutane 6053 0.89 1.29 2-methylpentane 6223 3.37 4.85 2,3-dimethylpentane 2405 1.05 1.11 n-undecane 4494 0.39 0.38 2-methylheptane 5067 0.51 0.84 m/p-xylene 6202 2.74 4.10 benzene 6209 4.08 6.64 toluene 6224 7.78 14.16 ethylbenzene 6158 1.00 1.38 o-xylene 6111 1.11 1.75 1,3,5-trimethylbenzene 4671 0.51 0.80 1,2,4-trimethylbenzene 6030 1.36 2.13 n-propylbenzene 5062 0.33 0.43 i-propylbenzene 3441 0.52 1.14 o-ethyltoluene 3649 0.44 0.53 m-ethyltoluene 4943 0.85 1.20 p-ethyltoluene 2475 0.77 0.80 m-diethylbenzene 3312 0.49 0.86 p-diethylbenzene 3840 0.51 0.60 styrene 5819 0.58 0.72 1,2,3-trimethylbenzene 4393 0.64 0.80 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 6-1 and 6-2 reveal the initial data collected at PAMS sites were not complete. Although most of species have over 6000 concentration measurements, many hourly samples could have one or more species measurements missing. The initial screening showed the missing measurements of individual species were concentrated in some special periods. These could be due to the instrument errors or the measurements below the detection limits at those sampling periods. Because Unmix relies on the principle component analysis (PCA), only the samples having complete records of species measurements can be used for its analysis. Therefore, the samples having any missing measurements should be dropped from the data. The number of remaining samples having the complete species measurements were determined automatically by Unmix based on the selected species. The ambient data were then screened by TNMOC measurements. TNMOC is the sum of all identified and unidentified peaks showing in the chromatogram; it is always greater than the sum of all identified species. For this reason, TNMOC is a good choice than any other individual compound to examine the data (Lewis et al., 1998). In Clinton Dr. data, the sum of the concentrations of the identified species was computed as the percentage of the reported TNMOC. The result is presented as the ratio of the sum of identified species to the reported TNMOC shown in Figure 6- 2. From Figure 6-2, there are two types of questionable samples having been identified: the samples have the ratio greater than 1, and the samples have the low 100 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4- Q ________ I ________ 1 ________ I ________ I ________ I ________ I ________ I ________ I ________ L 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 reported TNM OC (ppbC) Figure 6-2 Sum of all identified species (as % of reported TNMOC) vs reported TNMOC at the Clinton Dr. site. 101 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ratio. The source contributions derived from Unmix are the results of the apportionment of TNMOC measurements. Thus, the samples having the ratio >1 must need to be removed because these samples will confound the analysis. In addition, the samples with high portion of unidentified species could not have enough information to derive the correlation between identified species. Lewis et al. (1998) found that the PAMS compounds measured in ambient air constitute about 80% of TNMOC after they reviewed five site data. Based on this conclusion, the samples having the ratio less than 0.8 were also removed from the data. These defective samples could result from some of start-up situation such as inappropriate instrument operations or instrument errors. However, these defective samples are suspected to cause the analytical problems. These suspected samples could confound the results from Unmix. Therefore, it is desirable to remove those samples that have the ratio >1 and ratio <0.8 shown in Figure 6.2. i The ambient data were next screened by the scatterplots to identify the outliers. The scatterplot has also been suggested to diagnose the hydrocarbon data from the PAMS program (Lewis et al., 1998). The outliers were identified by the scatterplots, which use TNMOC and acetylene versus every other species. The screening process determined suspected samples that contained outliers with respect to the main body of data by visual inspection of scatterplots. The measured concentrations of outliers were not entirely random; sometimes they formed small and self-consistent groups. These outliers could be real but represent very unusual 102 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. conditions. However, these real but very infrequent events could confound the data statistical description that the model depends. Therefore, these outlier samples have been thought appropriate to remove from the data. After removing missing data and outliers, the final data sets had 2814 hourly samples for Clinton Dr. data and 2022 hourly samples for Deer Park data. The average TNMOC of Clinton Dr. and Deer park data were 597 ppbC and 305 ppbC respectively. 6.4 Results The most satisfactory results of applying Unmix model to Clinton Dr. and Deer Park data were a 7-source solution and a 6-source solution respectively. Acetylene was set as the tracer and the estimated solutions were normalized by TNMOC. These possible source solutions were determined by the diagnostic indicators (r-squared and signal-to noise ratio), which were calculated by Unmix based on the input data. Henry (2001) recommends the values of these diagnostic indicators to obtain the satisfied solution. In this ambient data analysis, 20 species plus TNMOC were selected from Clinton Dr. data for running Unmix. And there were 21 species plus TNMOC selected from Deer Park data. These species selections included the light paraffins, olefins, and the aromatic compounds based on the previous study in Houston area (Henry, 1997). In these species, acetylene was selected as the tracer due to its specialty. Lonneman et al. (1986) reported that acetylene is able to be as the vehicular tracer for automobile exhaust because it produces from the vehicular 103 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. combustion processes. Therefore, acetylene is a selected species to distinguish the roadway emission from the other industrial emissions. The source characterized by acetylene is identified as the roadway emission. At final, TNMOC is selected to normalize the source composition results in order to present the mass percentage of constituent species in each source. Table 6-3 and 6-4 list the estimated source compositions of the 7-source and 6- source solution given by Unmix. TNMOC was the average of source apportionments and presented as the concentration (ppbC) contributed from the source. The constituent species of each source presented as the percentage of average TNMOC of each source to reveal the mass concentration contribution to the source. In Deer Park data, most of TNMOC measurements were missed. Therefore, TNMOC measurements of Deer Park data were rebuilt by the sum of all identified species measurements. For this reason, the estimated average TNMOC contributions could be underestimated in Deer Park solution because TNMOC should be the sum of all peaks in the chromatogram including identified and unidentified species. According to the solutions derived from Unmix, the roadway emission contributes about 25% of total average TNMOC contributions in Houston area. The industrial sources including refinery and other petrochemical industries contribute the other 75% of total average TNMOC. Table 6-3 and 6-4 are the solutions from Unmix to present the estimated average TNMOC contributions and 104 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 6-3 Estimated source composition (mass percent)at the Clinton Dr. site. roadway refinery 1 refinery 2 industry 1 industry 2 refinery 3 refinery 4 propane 9.11 1.53 0.78111 I I I ! 2.10 1.37 5.42 acetylene l l l l l l l 0.00 0.00 0.00 0.00 0.00 0.00 n-butane 3.78 i3- 38l i i l l l l l l l 2.78 -1.18 -4.47 -0.77 i-butane 0.40 l l l l l l l 0.12 -0.51 -1.88 -0.03 0.39 n-pentane 2.64 -0.24 0.78 5.07 0.79 7.66 ! ! | I lllllll i-pentane 5.01 -0.50 4.60 7.38 2.38i!§ I ! ! ! ! ! 10.96 trans-2-pentene 0.32 0.03 0.06 0.10 0.11 ||! l l l l l l l -0.03 3 -methylpentane 1.58 0.06 0.20 0.29 0.24 1.49 5.82 n-hexane 2.18 0.04 0.12 2.08 0.60 1 30 l l l l f i f 2,2-dimethylbutane 0.23 0.00 0.04 0.16 0.00 0.56|1| I lllllll cyclohexane 0.16 0.13 0.01.... i i s l l l 0.16 0.18 0.44 3-methlyhexane 1.09 0.06 0.11 0.26 0.55 0.34|!! illlllll methyl-cyclohexane 0.94 0.07 0.08 1.12 0.76 0.39!!! l l l l l l l 2-methylhexane 0.90 0.05 0.08 0.21 0.49 0-42HI l l l l l l 3 -methyl-1 -butene 0.50 0.03 0.08 0.08 0.08l|| illlllll.. -0.03 2-methylpentane 2.09 0.05 0.39 0.72 0.20 2.71 III I lllllll meta/para-xylene 2.89 -0.02 0.18 0.38 i l l ! ! 1.03 0.85 ethylbenzene 0.11 -0.01 0.01 0.16 i l l ! ! 0.23 0.07 1,2,4-trimethylbenzene m i ! 0.02 0.11 -0.23 2.04 0.14 0.30 1,2,3 -trimethylbenzene i i i i i i i -0.02 0.04 0.13 0.94 -0.09 0.06 % of total TNMOC 25.27 29.45 15.38 6.95 4.95 8.65 9.35 105 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 6-4 Estimated source composition (mass percent) at the Deer Park site. roadway industry 1 propa source refinery 1 refinery 2&3 refinery 4 propane 6.71 4.45 i i l l l i l t 2.72 3.32 5.33 acetylene : ||||! 1 1 0.00 0.00 0.00 0.00 0.00 n-butane 6.49 8.04 1.44 11.20|| IIIIIIIII 0.05 i-butane 1.21 1.53 -0 1 6 l l l l l l l . . . 6.40 -2.56 n-pentane 1.28 4.85 1.27 1.4311 i i i i i i i i i ! 5.44 i-pentane 4.13 8.98 1.82 0.411 | .........m i 7.60 1-pentane 0.47 0.36 0.06 -o.ioll l l l l l l l 0.19 trans-2-pentene 0.73 0.46 0.09 -0.421!i i i i i i i i i 0.43 3-methylpentane 1.78 1.79 0.60 0.05 2 1 3 I I I ! ! ! n-hexane 1.12 3.25 1.07 4.65 i .0 6 |||!1 IIIIIIIII 2,2-dimethylbutane 0.21 0.41 0.10 -0-0211i i i i i l 0.39 cyclohexane 0.26 1136 0.06 0.15 0.31 0.30 3-methlyhexane 1.13 0.80 0.27 -0.21 0.5 2 l l l l 1111111: 2-methylhexane 0.79 0.75 0.20 -0.18 0-43 i | l | I l l l l l l l 3 -m ethyl-1 -butene 1.00 0.63 0.10 -0.44:11l l l l l l l l l l 0.40 2 -methylpentane 2.32 2.39 0.82 -0.18 3.15 l l l l l l l l l l i i 2,3 -dimethylpentane 0.54 0.45 0.15 -0.15 o.3ol|lil i i i i i i i m/p-xylene l l l l l l i i 1.52 0.59 -0.91 0.45 3.17 ethylbenzene i i i i i i i 0.55 0.20 -0.30 0.13 1.13 1,3,5 -trimethylbenzene I l l l l l l l 0.24 0.12 -0.36 0.00 0.60 1,2,4 -trimethylbenzene 1 1 1 1 1 1 1 1 1 1 1 0.88 0.37 -1.04 0.30 1.77 % of total TNMOC 22.39 9.40 16.01 19.62 13.44 19.14 106 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Estimated Source Contribution industrial 2 (xylene) 5%- industrial 1 (cyclohexane) refinery 1 (isobutane) 30% refinery 4 (n-hexane) 9% refinery 3 (isopentane) 9%- refinery 2 (n-butane) 15% -roadway emissions 25% Figure 6-3 Estimated source contribution from Unmix based on the Clinton Dr. site data. 107 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Estimated Source Contribution industrial 1 (cyclohexane) •roadway emissions 23% mixture of refinery 2 & 3 13% local propane source 16% •refinery 1 (siobutane) 20% refinery 4 (n-hexane) 19% Figure 6-4 Estimated source contribution from Unmix based on the Deer park site data. 1 0 8 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the source compositions derived from the ambient observations collected at Clinton Dr. and Deer park site in 1997. Figure 6-3 and 6-4 present the pie figures to illustrate the percentage of each source contribution shown in Table 6-3 and 6-4, respectively. The nonparametric regression method was used to make the plots of estimated source contributions derived from Unmix and the wind direction measurements. These plots present the directional pattern of each source for wind direction analysis. In Chapter 5, we are able to locate the dominant cyclohexane source by species measured concentrations and wind direction measurements. In this case, Unmix apportioned the source contributions from TNMOC measurements in samples; the results of source apportionments were presented as the species concentrations in the same sampling period. Based on this, the nonparametric regression method was applied to determine the plots of estimated source contributions versus wind direction measurements. The peaks in the plots were able to indicate the possible directions of emission sources. Additionally, the peaks also provide the information regarding the area or stationary sources by their shapes. The plots given by the nonparametric regression method can support the advanced analysis for source identification. The following sections will discuss the results of nonparametric regression plots of sources and source compositions from Unmix to identify the specific VOC emission sources in Houston area. 109 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 6.4.1 Roadway Emission The roadway emission is the one of major contributor to VOC in urban/suburban areas. The exhaust from on-road gasoline power vehicles and diesel trucks are classified in this kind of VOC emission source. Sometimes contributions of the roadway emission are more than 50% of total VOC emissions due to the areas with heavy commutation (i.e. Lin and Milford, 1994; Fujita et al., 1995; Harley and Sawyer, 1997). In the cities having industrial activities, the contributions of the roadway emission to VOC present between 14% and 34% (Scheff and Wadden, 1993; Kenski et al., 1995; Scheff et al., 1996). In this Houston data study, acetylene is used as the vehicular tracer for the source apportionment to identify the roadway emission. Based on the results derived by Unmix model, the source characterized by acetylene has been identified as the roadway emission. The average estimated contributions of VOC from the roadway emission are 25 % and 23% in the Clinton Dr. and Deer Park result, respectively. Table 6-3 shows the Clinton Dr. solution. The source characterized by acetylene represents the emission profile of the roadway emission, which could reveal the exhaust of vehicles fueling the reformulated gasoline. Previous studies reported the mass fraction of acetylene to the roadway emission is various from 1.7% to 6.5% based on the dynamometer studies (Slgsby, Jr. et al., 1987; Stump et al., 1992; Doskey et al., 1992) and the roadside measurements at Lincoln Tunnel, NY (Lonneman et al., 1986), Raleigh, NC (Zweidinger et al., 1988), Atlanta, GA 110 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (Conner et al., 1995), and Caldecott Tunnel, CA (Kirchstetter et al., 1996). Based on Table 6-3, the mass fraction of acetylene to the roadway emission is about 2.19%, which resembles the result collected at the Caldecott Tunnel. The mass fraction of isobutene, isopentane, n-pentane, 2-methylpentane, and 3-methylpentane are also close to the Caldecott Tunnel result. It is reasonable to find this consistency because the measurements of Caldecott Tunnel were collected in 1994 while the oxygenated gasoline has been sold in the San Francisco Bay area. In additional, the location of Clinton Dr. site is beside the junction of two interstate highways (I-10 and 1-45); this site is close to monitor the emissions from the high traffic volume. Therefore, the road emission found in Table 6-3 should be reasonable to present the composition of highway vehicle exhausts in Houston area. In Table 6-4, the Deer Park solution shows propane, acetylene, n-butane, isopentane, 2-methylpentane, and m/p-xylene are major species of the roadway emission composition. The mass fraction of acetylene of the roadway emission is about 5.47%, which presents a higher fraction to the Clinton Dr. solution. The reason of this discrepancy is probably because of the locations of these two monitoring sites. The position of the Deer park site is located at the residential area. We presume the roadway emission monitored at the Deer park site could be the mixture of highway emissions and the local vehicle exhaust. The local vehicle exhaust is contributed by incomplete combustion with unburned gasoline from the frequently stopping. Doskey et al. (1992) reported the profile of the cold start 111 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. emissions based on the observations collected from starting vehicles at the evening rush hours in Chicago. They showed propane, acetylene, n-butane, isopentane, 2- methylpentane, and m/p-xylene were the major species in the profile of cold start emissions. Comparing our results with this previous study, the ratio o f major species (excluding propane) to acetylene at the Deer Park is essentially consistent with the species-to-acetylene ratio in the profile of cold-start emissions. However, the fraction of propane in Deer Park roadway emissions is much lower than the profile of cold start emissions. This factor could be related to the propane source found by Unmix model. This propane source will be discussed in the following section. Figure 6-5 shows the plots of the estimated source contributions of the roadway emission versus wind direction measurements smoothed by nonparametric regression method for (a) at Clinton Dr. and (b) at Deer park site. The smoothing parameters of FWHM estimated by the cross validation method are 25° and 28°, respectively. Both of plots show continuous curves with some broad, un-sharp peaks. These broad and smooth peaks are expected to see because the roadway emission should be as the area emission from a wide range, not as the stationary emission from a specific direction. In Figure 6-5 (a), a broad peak is shown approximate from 200° to 300° (winds from SW to NW). This broad peak should present the roadway emissions at the 112 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. southwest and northwest sectors to the Clinton Dr. site. Another small peak is shown around 50°, which indicates a smaller impact of the roadway emission from NE. Comparing these with Figure 6-1, interstate highway 1-610 lies on the west and has the junction with highway 1-45 at the southwest of the Clinton Dr. site. And interstate highway I-10 passes through the northeast sector to the Clinton Dr. site. In additionally, interstate highway 1-610 is much closer to the Clinton Dr. site than I-10. It is reasonable to find the impact from the west sector is greater than the impact from the northeast sector. Thus, these findings are consistent with the spatial relationship between the Clinton Dr. site and these major highways in Houston area. We can confirm that the composition of roadway emissions in Table 6-3 associates with the vehicle exhaust from these interstate highways. Figure 6-5 (b) shows two peaks at around 50°, 100° and a broad peak from 225° to 300°. These peaks are consistent with the position of the highways around to the Deer Park site. As the presumption, the roadway emission in the Deer Park solution could be from the local traffic. The lower impact of Figure 6-5 (b) can approve this presumption due to the low traffic volume in residential area. From Figure 6-1, the route # 8 is located on the west sector and the route # 225 is located on the north sector to the Deer park site. The vehicle exhaust from the route # 8 should associate with that broad peak showing from 225° to 300° (SW to NW). The other two peaks around at 50° (NE) and 100° (E) should be associated with the emissions from the 113 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 220 200 1 6 0 O 160 140 1 2 0 £ 100 50 100 0 150 200 300 250 350 w ind direction (a) 140 1 2 0 1 0 0 o £ 2 C L C l O 80 * £ £ O O 60 40 20 0 0 50 1 0 0 150 200 250 300 350 wind direction (b) Figure 6-5 The plot of estimated contributions of roadway versus wind direction (a) at the Clinton Dr. site, FWHM = 25°, and (b) at the Deer Park site, FWHM = 28°. 114 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. route # 225. From southeast to southwest there is no major roads passing through this area. This is consistent with the curve presenting a low contribution between 130° and 230°. This sector presents the background emissions or we say as the local traffic emissions. Therefore, the plots in Figure 6-5 demonstrate the presumption of the roadway emission in the Clinton Dr. and Deer Park solutions by wind direction analysis. The nonparametric regression method provides the advanced analysis to extend the feasibility of Unmix model. 6.4.2 Refinery Emissions Excluding the roadway emission, the other sources derived from Unmix model are related with the stationary sources that are refining and petrochemical industries in Houston area. Usually, the refining emissions are characterized by paraffins and olefins coming from crude oil distilling and cracking processes. And the emissions from petrochemical industries are associated with one or two specific organic compounds. Thus, we use this knowledge to simply classify these stationary sources by checking the source compositions derived from Unmix. In Table 6-3, the Clinton Dr. solution shows the greatest YOC contribution is from a source characterized by isobutane, which contributes about 30% of total average TNMOC. This source consists of 84%, 13%, and 1.5% for isobutane, n-butane, and propane. This source is marked as refinery 1. The third largest source is characterized by n- butane, which contributes about 16% of total average TNMOC. This source 115 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. consists of 74% n-butane and 5% isopentane. This source is marked as refinery 2. The fourth and fifth largest sources account for 9% and 8% of the total average TNMOC, respectively. These two sources are characterized by relatively high amount of C-4 and C-5 paraffins and olefins. Therefore, we mark these two sources as refinery 3 and 4, respectively. A similar solution of source compositions is also found from the Deer Park data. Table 6-4 shows a source also characterized by isotunate, n-butane, and propane for mass proportion of 32%, 11%, and 3% in its source composition. This source contributes about 20% of total average TNMOC, which is the second greatest source contribution behind the roadway emission. Based on this, we link this source to the refinery 1 shown in Table 6-3. The fourth largest source in Table 6-4 is characterized by 3-methylpenatne, n-hexane, 3-methylhexane, 2-methylhexane, and 2-methylpentane. Because this source profile is very similar to the refinery 4 shown in Table 6-3, it is marked as refinery 4. The fifth largest source is characterized by n-butane, isopentane, trans-2-pentane, and 3-methyl-1-butene. These species are the characterized compounds in refinery 2 and 3 shown in Table 6-3. Thus, this source is reasonable to be marked as the mixture of refinery 2 and 3. The refinery 1 and 2 seem to relate to the specific refining processes because these two sources are specifically characterized by isobutane and n-butane, which are two essential ingredients in reformulated gasoline production. One fact is n-butane has 116 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. been used to produce isobutane through the isomerization process that converts straight chain paraffins into their high octane number isomers. Another is isobutane formed by isomerization of n-butane goes through the alkylation process to produce the alkylate by reacting with light olefins, which is the essential blending component of reformulated gasoline production (Meyers 1997). In addition to associate with gasoline production, isobutane is also used for synthesis of methyl tert-butyl ether (MTBE) to be the oxygenate agent in reformulated gasoline (Schmidt et al., 1993). Based on above knowledge, the refinery 1 may present the refinery emission associated with the alkylation process or the production of MTBE. The refinery 2 may associate with the isomerization process of n-butane. Furthermore, the refining industries involve complex processes to crack the heavy crude oil into light hydrocarbons for the gasoline production. Fugitive emissions of these processes could be characterized by relatively high amounts of C-4 and C-5 paraffins and olefins. Therefore, refinery 3 and 4 may be related to the fugitive emissions of these intermediaries released from the gasoline production. The plots of industrial source contributions versus wind direction show a strong tendency for high source contributions occurring at certain wind direction. This can be seen in Figure 6-6, 6-8, and 6-10, which plot the hourly estimated source contributions of those refinery-related sources versus the hourly wind direction at the Clinton Dr. and Deer Park site. Figure 6-6 presents the plots of refinery 1 source in (a) Clinton Dr. and (b) Deer park solution. At Clinton Dr., there is a very distinct 117 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. peak shown at around 166° and only this peak can be identified based on the uncertainty estimation. Depending on the scale of estimated contributions, this peak may show a nearby source associated with isobutane at the southeast to the Clinton Dr. site. At Deer Park, the largest peak is at around 50° and there is a small peak at around 330°. Figure 6-7 shows the predicted directions of refinery 1 estimated from the plots in Figure 6-6. These predicted directions indicate the locations of possible refinery sources characterized by isobutane. Based on Figure 6-7, two peaks shown in the Deer Park plot could be associated with a small peak shown at 100° in the Clinton Dr. plot. But their impacts are much smaller than the nearby sources located at 166°. In addition to this, the winds were not frequent from northwest based on the 1997 data. This situation results in no distinct peak at Deer Park associated with the largest peak shown in Clinton Dr. plot. The emissions of MTBE have been reported in EPA Air Toxic Emission Inventory. Table 6-5 presents the summery of MTBE emissions in Harris County, Texas. Although there is no detailed speciated inventory available, Air Toxic Inventory provides the emissions of the single species, which may correspond to the peaks shown in the source contribution plot. Based on the source composition from Unmix, we speculate the refinery 1 is related with MTBE production. Therefore, the sources in Table 6-5 can be used to verify the peaks shown in Figure 6-6. At Clinton Dr., the largest peak shown at 160° corresponds well to Valero Refining in 118 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1200 1000 800 600 200 0 0 50 100 150 200 250 300 350 w ind direction (a) 160 140 1 2 0 u J 2 § 1 0 0 c 0 1 ac o o T 5 d > 0 3 60 g T > iu 40 20 0 0 50 1 0 0 150 200 250 300 350 wind direction (b) Figure 6-6 The plot of estimated contributions of refinery 1 versus wind direction (a) at the Clinton Dr. site, FWHM = 10°, and (b) at the Deer Park site, FWHM = 20° 119 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. z Figure 6-7 Houston map shows the predicted direction of refinery 1 (isobutane) based on Figure 6-6. The “X” is the location of the Clinton Dr. site and the “+” is the Deer Park site. VOC sources are shown as red circles with the area proportional to the emission rates. to o the inventory. Though Valero Refining accounts only for 12% of the emissions, this source is located only 1.23 km away from the Clinton Dr. site. It seems reasonable to associate this largest peak with this source. The largest emission shown in Table 6-5 is Exxonmobil Baytown Refinery contributing about 29% o f MTBE emissions in Harris County. This refinery seems to associate with the small peak shown at 100° in the plot (a) and the largest peak shown at 50° in the plot (b). Because this refinery is located 24.22 km away from the Clinton Dr. site, the impact from this refinery cannot be distinct shown in the plot. The location of second largest peak at Deer Park seems to correspond closely to Crown Central petroleum in Table 6-5. In addition, there are two sources located at the NE sector to the Clinton Dr. site (Lyondell Chemical and Eqaistar Chemicals). They seem to relate with the small peak shown at around 60° in the plot (a). However, the distance could affect source impacts not be identified as a real peak. Figure 6-8 shows the plots of refinery 2 and 3 in Clinton Dr. solution and the source of refinery mixture in Deer Park solution. In plot (a) it shows a board but high contribution peaks covered a sector from 125° to 190° (from SE to S), and a lower and clear peak at around 94° (E). A similar plot of refinery 3 shows two significant peaks at around 95° (E) and 158° (SSE). Then it is obvious to see two major peaks occurring almost at the same direction but showing the different source contributions. Previous knowledge demonstrated there were a number of nearby 121 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Table 6-5 Summary o f 1997 MTBE emissions in Harris County, Texas.______________________________________ Release Total release Percentage Accuracy Facility name type (lbs/year) of total Latitude Longitude (m ) Deer Park_______ Clinton Dr. Distance Distance __________________________________________________________________________ Azimuth (km)_____ Azimuth (km) EXXONMOBIL BAYTOWN REFINERY STACK 107442 28.83% 29.74 95.01 80 57.68 14.00 89.60 24.22 SHELL CHEMICAL L.P. FUG 36000 9.66% 0.00 0.00 0.00 0.00 0.00 0.00 SHELL CHEMICAL L.P. STACK 17000 4.56% 0.00 0.00 0.00 0.00 0.00 0.00 VALERO REFINING CO. STACK 26961 7.23% 29.72 95.25 20 297.03 13.51 163.35 1.23 VALERO REFINING CO. FUG 18086 4.85% 29.72 95.25 20 297.03 13.51 163.35 1.23 EGP FUELS CO. STACK 31984 8.58% 29.68 95.01 11000 84.68 11.12 105.00 24.29 CROWN CENTRAL PETROLEUM FUG 17597 4.72% 29.72 95.21 50 308.68 9.83 103.98 4.85 CROWN CENTRAL PETROLEUM STACK 10738 2.88% 29.72 95.21 50 308.68 9.83 103.98 4.85 LYONDELL CHEMICAL CO. STACK 22333 5.99% 29.81 95.10 80 10.22 16.22 60.48 17.54 LYONDELL CHEMICAL CO. FUG 4915 1.32% 29.81 95.10 80 10.22 16.22 60.48 17.54 EQUISTAR CHEMICALS L.P. FUG 14000 3.76% 29.83 95.11 50 6.41 17.90 53.93 17.79 EQUISTAR CHEMICALS L.P. STACK 5000 1.34% 29.83 95.11 50 6.41 17.90 53.93 17.79 TEXAS PETROCHEMICALS CORP. FUG 15246 4.09% 29.70 95.26 20 284.71 12.66 178.12 4.10 EXXONMOBIL CHEMICAL PLANT STACK 13076 3.51% 29.75 95.02 500 51.08 13.86 86.57 23.21 EXXONMOBIL CHEMICAL PLANT FUG 3149 0.84% 29.75 95.02 500 51.08 13.86 86.57 23.21 TEXAS PETROCHEMICALS CORP. STACK 12831 3.44% 29.70 95.26 20 284.71 12.66 178.12 4.10 LYONDELL-CITGO REFINING L.P. STACK 8462 2.27% 29.72 95.23 50 299.45 11.27 124.64 3.13 LYONDELL-CITGO REFINING L.P. FUG 648 0.17% 29.72 95.23 50 299.45 11.27 124.64 3.13 EGP FUELS CO. FUG 3960 1.06% 29.68 95.01 11000 84.68 11.12 105.00 24.29 SOUTHWEST SHIPYARD L. P. STACK 2000 0.54% 29.79 95.06 11000 25.54 14.85 72.06 19.75 JOHANN HALTERMANN LTD. STACK 1290 0.35% 29.76 95.10 80 18.50 9.87 82.49 15.65 Total emissions 372718 to to refineries at the southeastern sector and large but remote refineries at the eastern sector to the Clinton Dr. site. Hence, this may be the evidence to proof that these two refinery-related sources may present the emissions associated with the different refining processes. At Deer Park, the plot of refinery mixture gives the largest peak at around 312° and a smaller peak at around 260°. Figure 6-9 shows the predicted directions of refinery mixture estimated from the plots in Figure 6-8. These predicted directions indicate the locations of possible refinery sources characterized by n-butane, n-pentane, isopentane, and 3-methyl-1-butene. Based on triangulation, one possible location of this refinery source can be located at 94° to the Clinton Dr. site and 312° to the Deer Park site. But there is no reported source located at 260° to the Deer Park site. The source identification can be accomplished by comparing with the detailed emission inventories in this area. Figure 6-10 presents the plots of refinery 4 in (a) Clinton Dr. solution and (b) Deer Park solution. As the other refinery-related sources, the plot of refinery 4 at Clinton Dr. also shows two major peaks at 94° (E) and 159° (SSE). The shapes of peaks are much similar to the peaks in Figure 6-8 (b), which is the plot of refinery 3 in Clinton Dr. solution. This can approve the assumption that refinery 3 and 4 are the fugitive emissions from refineries, could be related with the gasoline production. At Deer Park, the plot of refinery 4 shows a board, high contribution peak covered a sector from 230° to 310°, which presents the maximum at around 258° and a shoulder at around 295°, and the other two peaks shown at 51° and 112°. Figure 6-11 shows the 123 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. "8 ioo 100 50 200 wind direction (a) 100 90 80 8 - < to 1 8 1 30 20 10 0 0 50 150 1 0 0 200 250 300 350 (b) 160 140 1 2 0 i Is 60 E T 3 40 20 0 0 50 1 0 0 150 200 250 300 350 (c) Figure 6-8 The plot of estimated contributions versus wind direction for (a) refinery 2 at the Clinton Dr. site, FWHM = 12°, (b) refinery 3 at the Clinton Dr. site, FWHM = 11°, and (c) refinery mixing sources at the Deer Park site, FWHM =16°. 124 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. predicted directions of refinery 4 from the Clinton Dr. and Deer Park site. Based on the triangulation, a possible location of this refinery source could be located at the intersection of 94° to the Clinton Dr. site and 51° to the Deer Park site. However, there is no reported source associated with 258° to the Deer Park site. Additionally, the peak shoulder at 295° could be associated the nearby refineries at the southeast of Clinton Dr. site. But this cannot be sure because of the uncertainty. The refinery sources associated with the peaks at 51° and 112° can be identified by detailed emission inventory. Because n-hexane is one of characteristic species in refinery 4, we can use n-hexane emissions from EPA Air Toxic Emission Inventory to compare with the peaks shown in Figure 6-10. Table 6-6 presents the summary of n-hexane emissions in Harris County, Texas. In Figure 6-8 (a), the largest peak shown at 94° seems to correspond closely to the first two largest emissions in Table 6-6 (Exxonmobil Chemical Plant and Solvay Polymers). These two emissions account for 24% of n- hexane emissions in Houston area. The second largest peak shown in 158° corresponds well to Valero Refining in the inventory. Although Valero only accounts for 5% of n-hexane emissions, it is located only 1.7 km the Clinton Dr. site. Therefore, it seems reasonable to associate this second largest peak with this nearby sources. Another near source Lyondell-Citgo Refining could be also associated with the shoulder of this second largest peak. But the plot does not show a clear impact from this source. 125 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. 1 Figure 6-9 H ouston map shows the predicted direction o f refinery 2 & 3 (refinery mixture) based on Figure 6-8. The “X” is the location o f the Clinton Dr. site and the “+” is the Deer Park site. YOC sources are shown as red circles with the area proportional to the emission rates. ro O n 1 0 0 G Cl Cl £ Q O 0 50 1 0 0 150 200 250 300 350 wind direction (a) 120 100 £ 80 CL CL c o '■ 5 S O 40 20 0 0 50 1 0 0 150 200 250 300 350 wind direction (b) Figure 6-10 The plot of estimated contributions of refinery 4 versus wind direction (a) at the Clinton Dr. site, FWHM = 15 °, and (b) at the Deer Park site, FWHM = 20°. 127 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. M m \ 3 ? \ Figure 6-11 Houston map shows the predicted direction o f refinery 4 (n-hexane) based on Figure 6-10. The “X” is the location o f the Clinton Dr. site and the “+” is the D eer Park site. VOC sources are shown as red circles with the area proportional to the emission rates. to 00 At Deer Park, the second largest peak shown at 112° could be associated with the third largest emission shown in Table 6-6 (Fina Oil & Chemical). However, the location of this source is given an accuracy of 11 km, it cannot definitely associate this source with this peak. The peak shown at 51° could be associated with two sources (Aristech Chemical and Akzo Nobel Chemicals), which are located at 45° to the Deer Park site. But the largest peak shown at 258° does not correspond to any source in Table 6-6. 6.4.3 Petrochemical Industrial Emissions There are two small sources derived from Unmix associated with one or two specific species, which are expected to relate with specific petrochemical industries. In Table 6-3, the smallest two sources are contributed about 6% and 5% of the total average TNMOC and characterized by cyclohexane and xylene, respectively. In Table 6-4, the smallest source contributed 10% of total average TNMOC is also characterized by cyclohexane. Previous work has confirmed a dominant source responded for the cyclohexane emission (Henry et al, 2002). The source characterized by xylene has not been found in Deer Park solution because xylene was not selected. However, the study of 1993 data at the Clinton Dr. site has confirmed a nearby emission from a rubber production by using the results from receptor modeling and detailed emission inventory (Henry et al., 1997). Thus, the xylene source derived from Unmix could be the same petrochemical source 129 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Table 6-6 Summary o f 1997 n-butane emissions in Harris County, Texas. Facility name Release type Total release (lb/year) Percentage of total ----------— ~ j , ----------------- Latitude Longitude Accuracy (m) Deer P ark Clinton Dr. Azimuth Distance (km) Azimuth Distance (km) EXXONMOBIL CHEMICAL PLANT FUG 51350 6.31% 29.75 95.02 500 49.91 13.92 85.35 23.17 EXXONMOBIL CHEMICAL PLANT STACK 49335 6.06% 29.75 95.02 500 49.91 13.92 85.35 23.17 SOLVAY POLYMERS INC. FUG 92000 11.30% 29.72 95.09 20 35.83 7.01 94.86 16.61 FINA OIL & CHEMICAL CO. FUG 80000 9.82% 29.60 95.02 11000 125.65 13.19 122.51 27.48 FINA OIL & CHEMICAL CO. STACK 10546 1.30% 29.60 95.02 11000 125.65 13.19 122.51 27.48 CHEVRON CHEMICAL CO. FUG 71484 8.78% 29.82 94.92 10 49.04 26.37 72.48 33.92 CHEVRON CHEMICAL CO. STACK 13658 1.68% 29.82 94.92 10 49.04 26.37 72.48 33.92 LY ONDELL-CITGO REFINING L.P. STACK 21453 2.63% 29.72 95.23 50 298.79 11.23 123.13 3.11 LYONDELL-CITGO REFINING L.P. FUG 37974 4.66% 29.72 95.23 50 298.79 11.23 123.13 3.11 ARISTECH CHEMICAL CORP. FUG 58400 7.17% 29.71 95.08 10 44.96 6.11 99.37 16.99 AKZO NOBEL CHEMICALS INC. STACK 36528 4.49% 29.70 95.09 80 45.64 5.48 101.27 16.69 EXXONMOBIL BAYTOWN REFINERY. STACK 24415 3.00% 29.74 95.01 80 56.33 14.05 88.32 24.15 EXXONMOBIL BAYTOWN REFINERY. FUG 11476 1.41% 29.74 95.01 80 56.33 14.05 88.32 24.15 VALERO REFINING CO. STACK 21943 2.69% 29.72 95.25 20 296.43 13.48 161.34 1.17 VALERO REFINING CO. FUG 13328 1.64% 29.72 95.25 20 296.43 13.48 161.34 1.17 REED TOOL CO. STACK 31378 3.85% 29.75 95.31 10 296.57 19.43 287.71 5.18 PHILLIPS PETROLEUM CO. STACK 30000 3.68% 29.74 95.18 50 330.27 9.25 83.25 7.91 ARISTECH CHEMICAL CORP. STACK 17040 2.09% 29.71 95.08 10 44.96 6.11 99.37 16.99 EQUISTAR CHEMICALS L.P. STACK 8900 1.09% 29.63 95.09 80 133.60 5.60 123.56 19.81 EQUISTAR CHEMICALS L.P. FUG 11000 1.35% 29.83 95.11 50 5.47 18.00 52.59 17.81 CELANESELTD. CLEAR LAKE PLANT STACK 13655 1.68% 29.62 95.06 80 130.74 8.19 123.67 22.42 MILLENNIUM PETROCHEMICAL INC. STACK 10203 1.25% 29.71 95.07 80 49.40 7.60 96.72 18.34 CNA HOLDINGS INC. FUG 9500 1.17% 29.60 95.01 20 124.85 13.61 122.16 27.92 Total emissions 814321 O demonstrated by 1993 data. The plot of source contributions of this xylene source will provide the advanced analysis to confirm this petrochemical industrial source. Figure 6-12 shows the plots o f industrial 1 at (a) Clinton Dr. and (b) Deer Park site. The plots are very similar to the results plotting by cyclohexane measured concentrations and wind direction measurements in Chapter 5. The largest peak shown in both plots is consistent with the position of dominant cyclohexane source in the inventory. Hence the industrial 1 can be confirmed as that dominant cyclohexane source in Houston area. Figure 6-13 presents the predicted directions of industry 1 from the Clinton Dr. and Deer Park site. Figure 6-14 shows the plot of industry 2 characterized by xylene in Clinton Dr. solution. This plot shows several peaks at around 50°, 105°, 190°, 283°, and 330°. The largest peak is shown at 194°, which corresponds closely to the direction of the xylene source shown in 1993 Houston study. Therefore, the industrial 2 may be related with that xylene source near the Clinton Dr. site. Figure 6-15 shows the predicted directions of industry 2 from the Clinton Dr. site. The directions of 105° and 194° are definite associated with the nearby sources to the Clinton Dr. site. The industries located at these two directions will be corresponded to xylene emissions in the Houston area. 131 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1 0 0 o £ 3 a . Cl 0 50 100 150 200 250 300 350 wind direction (a) 120 100 £ 3 a . Cl C o = 5 £ 3 - £ O O T 5 • s E T 5 ■ D 1 0 0 150 200 wind direction 250 300 350 (b) Figure 6-12 The plot of estimated contributions of industry 1 versus the wind direction (a) at the Clinton Dr. site, FWHM = 10 °, and (b) at the Deer Park site FWHM =19°. 132 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Figure 6-13 Houston map shows the predicted direction o f industry 1 (cyclohexane) based on Figure 6-12. The “X ” is the location o f the Clinton Dr. site and the “+” is the Deer Park site. VOC sources are shown as red circles w ith the area proportional to the emission rates. U > U ) Table 6-7 shows the summery of xylene emissions from EPA Air Toxic Emission Inventory. The xylene emissions will be compared with the peaks shown in Figure 6-14 because the industry 2 is characterized by xylene. In Figure 6-14, this xylene source peaks the maximum contribution at 194°. Based on the location, it seems Bayer Corporation is associated with this largest peak shown in the plot. Bayer source is the third largest emission in the inventory and the distance to the Clinton Dr. site is 4 km. Hence, it is reasonable to speculate that this source corresponds to the largest peak at 194°. On the other hand, Table 6-7 shows that the largest xylene source is Lyondell-Citgo Refining that accounts for 40% of xylene. This source is a nearby source and locates at 125° to the Clinton Dr. site. But there is no any peak corresponding to this largest source in Figure 6-14. The closest peak to this facility is at 105° and is not the largest peak in the plot. Because the xylene source is the smallest estimated source in the Houston area, we speculate this discrepancy could indicate the inaccurate emission rates reported by the facility. In addition, no any other source from the inventory corresponds to the remaining peaks in Figure 6-14. 6.4.4 Local Propane Source This derived propane source is only presented in Deer Park solution. In Table 6-4 the propane source is the fourth largest source contributed about 16% of total average TNMOC in the Deer Park 6-sources solution. This propane source consists of 70% propane and n-butane, n-pentane, and isopentane with small fraction. As we 134 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. S 25 8 20 0 50 1 0 0 150 200 250 300 350 w in d direction Figure 6-14 The plot of estimated contributions of industry 2 at the Clinton Dr. site. The nonparametric regression parameter, FWHM, is equal to 15 °. 135 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Figure 6-15 Houston map shows the predicted direction o f industry 2 (xylene) based on Figure 6-14. The “X” is the location o f the Clinton Dr. site and the “+” is the Deer Park site. VOC sources are shown as red circles with the area proportional to the emission rates. U J O s Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Table 6-7 Summary o f 1997 xylene emissions in Harris County, Texas. Facility name Release type Total Release Percentage (lb/year) of total ----5 - - Latitude Accuracy Longitude (m) Deer P ark Distance Azimuth (km) Clinton Dr. Distance Azimuth (km) LYONDELL-CITGO REFINING L.P. FUG 470752 39.81% 29.72 95.23 50 299.45 11.27 124.64 3.13 LYONDELL-CITGO REFINING L.P. STACK 2401 0.20% 29.72 95.23 50 299.45 11.27 124.64 3.13 TRINITY INDS. INC. PLANT 19 STACK 115787 9.79% 32.50 94.64 11000 9.36 317.57 11.82 312.65 TRINITY INDS. INC. PLANT 19 FUG 6074 0.51% 32.50 94.64 11000 9.36 317.57 11.82 312.65 BAYER CORP. HOUSTON FUG 57800 4.89% 29.70 95.26 80 285.93 13.21 184.86 3.71 BAYER CORP. HOUSTON STACK 1200 0.10% 29.70 95.26 80 285.93 13.21 184.86 3.71 VALERO REFINING CO. - TEXAS FUG 44150 3.73% 29.72 95.25 20 297.03 13.51 163.35 1.23 VALERO REFINING CO. - TEXAS STACK 4761 0.40% 29.72 95.25 20 297.03 13.51 163.35 1.23 EXXONMOBIL BAYTOWN PLANT FUG 36365 3.08% 29.75 95.02 500 51.08 13.86 86.57 23.21 EXXONMOBIL BAYTOWN PLANT STACK 2564 0.22% 29.75 95.02 500 51.08 13.86 86.57 23.21 EXXONMOBIL BAYTOWN REFINERY. FUG 26923 2.28% 29.74 95.01 80 57.68 14.00 89.60 24.22 EXXONMOBIL BAYTOWN REFINERY. STACK 7264 0.61% 29.74 95.01 80 57.68 14.00 89.60 24.22 CHEVRON CHEMICAL CO. FUG 26800 2.27% 29.82 94.92 10 50.27 26.28 73.78 33.95 CHEVRON CHEMICAL CO. STACK 5395 0.46% 29.82 94.92 10 50.27 26.28 73.78 33.95 GREIF BROS. CORP. STACK 28217 2.39% 29.70 95.06 11000 63.83 6.93 102.90 19.08 GREIF BROS. CORP. Total em issions FUG 1642 1182406 0.14% 29.70 95.06 11000 63.83 6.93 102.90 19.08 know, propane is the essential ingredient of liquefied petroleum gas (LPG). LPG is used as a fuel for domestic, industrial, horticultural, agricultural, cooking, heating and drying processes. Therefore, this propane source could be an area source associated with LPG usage around the Deer Park site. Figure 6-16 is the source contribution plot of this propane source in the Deer park solution. The plot shows a board and smooth peak covered a wide range from 150° to 300° (SSE to NW) and the FWHM is 79°. Therefore, this board peak presents an evidence to infer this propane source as an area source associated with local LPG usage. On the other hand, the impact of this propane source to the Clinton Dr. site could be reduced as the background emission due to its source strength is much lower than the other industrial sources around the Clinton Dr. site. This could result in this propane source hasn’t been found from the Clinton Dr. data. 6.5 Comparison to Emission Inventory Only the data collected at the Clinton Dr. site have complete TNMOC measurements, the rebuilt TNMOC concentrations at Deer Park could result in the source contributions underestimated. According to this, the Unmix solution based on Clinton Dr. data is selected to evaluate the emission inventory. The nonparametric regression plots of derived source contributions compare with the total VOC emission rates and locations reported by AIRS emission inventory. Any 138 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1 0 0 90 80 t _ ) in Q. a. c o n • 6 o u T 3 < u IS e T 5 V 0 50 1 0 0 150 200 250 300 350 w in d direction Figure 6-16 The plot of estimated contributions of the propane source at the Deer park site. The nonparametric regression parametric, FWHM, is equal to 79°. 139 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. discrepancy between the plots and reported emission rates will give the basis for the evaluation of emission inventory. At the Clinton Dr. site, the derived sources show the amount of source contributions increased strongly with winds from south-southeast sector to the Clinton Dr. site. This indicates the major YOC sources should be located at this direction. Figure 6- 17 presents the nonparametric regression plots of reported TNMOC, sum of all six derived sources, and refinery 1 at the Clinton Dr. site. All of three curves present the similar shape and have the maximum peak at around 166°. This not only shows the consistency between the TNMOC measurements and Unmix results, but also indicate refinery 1 is the most significant source at this direction to the Clinton Dr. site. The largest peak of refinery 1 locates at 166° and accounts for over 60% of TNMOC contribution. Based on the source composition, this largest peak has been identified to correspond with Valero Refining by MTBE emissions from Air Toxic Inventory. Therefore, Valero Refining should be the largest VOC emissions at around 166° to the Clinton Dr. site. Table 6-8 gives the self-reported VOC emissions from AIRS in tones per year for a 10° sector from 70° to 200° relative to the Clinton Dr. site (all directions are azimuth angels measured clockwise from north). In the Table, Valero Refining and Lyondell-Citgo Refining are two significant VOC sources located from 120° to 180° 140 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1400 1200 <-> 1000 _ £ 2 £2. Q. 800 600 400 200 0 50 100 150 200 250 300 350 wind direction Figure 6-17 The nonparametric regression plots for reported TNMOC and derived sources at the Clinton Dr. site. The blue line is the reported TNMOC measurements at the Clinton Dr. site. The red line is the sum of six refinery and industrial sources estimated by Unmix. The pink line is the estimated contribution of refinery 1 in the Clinton Dr. solution. 141 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. in the vicinity of Clinton Dr. site and there are other nearby sources located at the east to the Clinton Dr. site. Based on emission rates and distance to the Clinton Dr. site, the distance-weighted YOC emissions for each facility can be calculated. The distance-weighted YOC emissions are presented as the bar chart for every 10° sector to compare with the plots of source contributions of refinery sources derived from Unmix, as seen in Figure 6-18, 6-19, and 6-20. Emissions from Valero Refining lay on the 150-180° sector and accounts for over 50% of the list VOC emissions in this sector. However, the highest amount of VOC emissions from Valero Refining is in 170-180° sector, which is not consistent with the direction given by the largest peak. Figure 6-18 shows this inconsistency between Valero Refining and refinery 1. Although Valero Refining matches the conjecture from the source composition determined by Unmix, the discrepancy is found between the reported emission rates and peak location. The comparison shows the inconsistency and the emission rates of this facility should be went over to rid of inaccuracy. Figure 6-19 and 6-20 show the same inconsistency for reported VOC emissions and derived refinery sources from Unmix. The derived source contributions of refinery 2 along with refinery 3 and 4 also peak at 158° that matches the location of emissions from Valero Refining and Lyondell-Citgo Refining. Based on Table 6-8, Lyondell-Citgo Refining is the largest VOC emissions lying on the 120-160° sector. 142 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. However, the maximum emission from this facility is at 130-140°, which is not consistent with the direction given by the peak. Furthermore, the amount of emissions at 120-140° is over 60% of the list emissions from this source. Neither three refmery-related sources nor observed TNMOC shows an apparent peak at this sector. If these three refinery-related sources are inferred as the emissions from different processes but in the same facility (Valero Refining), Lyondell-Citgo Refining would not fulfill our expectations for the direction and the size of the impact. Therefore, this largest emission reported from the emission inventory fails to agree the results from our observation-based method. In Clinton Dr. solution, we were only able to match the industrial 2 with a specific chemical plant in the emission inventory. The amount of TNMOC from industrial 2 shows the largest peak at 190°, which corresponds to Bayer Corporation based on Table 6-7. According to Table 6-8, Bayer Corporation is the largest VOC emission in 180-190° sector excluding some partial emissions from Valero Refining. In addition, the 1993 Houston data study also demonstrated the emissions from Bayer Corporation were characterized by xylene. An industrial source determined by receptor modeling was associated with Bayer Corporation at 180-190°. For these reasons, we can positively identify industrial 2 as the emissions from Bayer Corporation. 143 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Table 6-8 Summary of VOC emission rates from the AIRS database (tons/year). Azimuth to Clinton Dr. Company (km) 70-80 80-90 90- 100 100- 110 110- 120 120- 130 130- MO 140- 150 150- 160 160- 170 170- 180 180- 190 190- 200 Grand total VALERO REFINING COMPANY 1.25 3 1 251 371 570 111 27 1334 LYONDELL-CITGO REFINING LP 3.23 648 1482 985 225 13 15 3368 GOODYEAR TIRE AND RUBBER 3.31 283 17 299 MOBIL CHEMICAL COMPANY 3.34 9 237 246 BAYER CORPORATION 3.73 59 59 GULF COAST WASTE DISPOSAL TEXAS PETROCHEMICALS CORPORATION 3.79 3.85 23 146 18 377 20 169 414 PASADENA PAPER COMPANY 4.28 16 360 376 WARREN PETROLEUM COMPANY 5.05 101 74 175 CROWN CENTRAL PETROLEUM 5.19 68 1006 61 1135 SOLVAY POLYMERS, INC. 5.19 993 993 GATX TERMINALS CORPORATION 5.38 607 173 1 1 782 AMERADA HESS CORPORATION 5.67 143 408 551 PHILLIPS CHEMICAL COMPANY 6.95 43 436 102 581 SHELL OIL COMPANY 12.90 1069 173 1242 ROHM & HAAS INCORPORATED 14.82 234 1186 1420 EQUISTAR CHEMICALS, L.P. 18.50 1038 46 1084 HOECHST CELANESE CHEMICAL 22.23 642 642 EXXON CHEMICAL AMERICAS 23.01 1095 35 1130 EXXON COMPANY, U.S.A. 24.25 4376 1848 6224 Grand Total 1108 6230 8151 1925 567 1291 1486 1004 476 393 1482 206 27 24345 - t x . -px V alero, 1 ,3km Lyondell-Citgo, 3.3km B ayer, 3.7km C row n C entral, 5.2km Solvay Polym ers, 5.2km G atx Term inals, 5.4km A m erada H e s s , 5.7km Phillips. 6.9km 3UU v . 150 200 250 azimuth to Clinton Dr. site 300 350 Figure 6-18 The plot of distance-weighted emissions from major reported VOC sources and refinery 1 (isobutane) derived from Unmix. The red line is the estimated source contributions of refinery 1. 145 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ;timated contribution (ppb) E 03 Q J != -, CO O CO c o CO to 'E 0 3 T 3 0 3 C33 03 to T 3 V alero, 1 .3km Lyondell-Crtgo, 3.3km B ayer, 3.7km C row n Central, 5.2km Solvay Polym ers, 5.2km G atx Term inals, 5.4km A m erada H e s s, 5.7km Phillips. 6.9km 150 .5 100 “ - U a 150 200 250 azimuth to Clinton Dr. site Figure 6-19 The plot of distance-weighted emissions from major reported VOC sources and refinery 2 (n-butane) derived from Unmix. The red line is the estimated source contributions of refinery 2. 146 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. V alero, 1 ,3km Lyondell-Citgo, 3.3km H H B ayer, 3.7km C row n C entral, 5.2km Solvay Polym ers, 5.2km G atx Term inals, 5.4km m i A m erada H e s s, 5.7km Phillips, 6.9km 0 50 100 150 200 250 300 350 azimuth to Clinton Dr. site Figure 6-20 The plot of distance-weighted emissions from major reported VOC sources and refinery 3 and 4 (refinery mixture) derived from Unmix. The red line is the estimated source contributions of refinery 3 and the black line is the estimated source contributions of refinery 4. 147 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The comparison to the current emission inventory indicates the discrepancy between our observation-based results and reported emissions. Only the smallest source has been identified as the emissions from a specific chemical plant at south-southeast sector. The previous study of 1993 Clinton Dr. data reported the same results by using a less advanced multivariate receptor modeling and simple wind direction plot (Henry et al., 1997). The maximum emission from Valero Refining is not consistent with the direction determined by observations. And the largest emission source in the inventory did not show up in the observations. Based on above facts, our results present that the inventory of industrial VOC emissions is inaccurate in the location and its emission rates of major sources. 6.6 Discussion The multivariate receptor model Unmix is a statistical software to perform an independent source appointment analysis without any prior knowledge regarding to the source composition. Based on the 1997 observations analysis, Unmix estimated source compositions and average source contributions for seven sources. Three largest sources account for about 70% of the average TNMOC in the Houston area. The remaining 30% of the average TNMOC are contributed from the other four smaller sources. The application of nonparametric regression produces the smooth plots of wind direction analysis to determine the possible direction of sources. The plots show the major TNMOC contributions come from several nearby sources in 148 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the vicinity of the monitoring site. However, we are only able to match the smallest source to a petrochemical plant from the emission inventory. The apparent discrepancy has been shown from the comparison of our results and the emission inventory. These findings are agreement with the results from a previous study (Henry et al., 1997). This previous study of 1993 field data in Houston reported an inconsistency by comparing the observations with the emission inventory. A less advanced receptor modeling and the simple wind direction plot determined three major sources in the vicinity of the Clinton Dr. site. But only the smallest source characterized by xylene has matched to a specific chemical plant reported in the inventory. The other emissions in the inventory failed to agree with remaining sources determined by observations. The same inconsistency reported from 1993 filed data has been found in this 1997 data study. For this reason, we believe that the inventory of industrial VOC emissions is inaccurate in the location and emission rates of major sources. And the inaccuracy still remains in the emission inventory. The future work should go over these emission data and get rid of inaccuracy from the emission inventory. 149 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CHAPTER 7 CONCLUSION AND SUGGESTIONS FOR FUTURE RESEARCH Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 7.1 Conclusion The work of this dissertation focuses on development and application of the wind direction analysis to air quality data by nonparametric regression. Wind direction analysis has been used as a supplementary method to predict the source location by the plot of pollutant concentrations and wind direction data. The simple scatterplot or bar chart is the common method to produce the plot for showing the possible direction of pollutant sources. However, these graphic methods could not give the precise prediction of the source direction due to their limitations. In order to predict the direction precisely, nonparametric regression has been introduced to improve the plot of pollutant concentrations and wind direction data for wind direction analysis. Nonparametric regression is a set of technology to seek the smooth curve for showing the relationship of pollutant concentrations and wind direction data. This smooth curve is then used to estimate the wind direction given by pollutant concentration peaks. The estimates of peak directions from nonparametric regression plots enable to predict the more precise location of pollutant sources. In addition, nonparametric regression can be also applied to the source contributions estimated from receptor models. This application of nonparametric regression gives the sharper analysis o f source directions in order to evaluate of emission inventories along with the source compositions from receptor models. Results of nonparametric regression application for wind direction analysis as well as suggestions in the future studies are discussed in the following sections of this chapter. 151 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 7.1.1 Application for Locating the Largest Emission Sources The success of the nonparametric regression method has been judged by examining the location of the largest emission source from ambient observations. PAMS program provides the measured cyclohexane concentrations of 1997 from two sites in the Houston area. A dominant cyclohexane source, Phillips petroleum Company, demonstrated by Air Toxic Emission Inventory contributes over 70% of total cyclohexane emissions in the Houston area. Therefore, a large peak should be shown in the plot of wind direction and measured cyclohexane concentrations from each site. The estimated direction of the largest peak will give the direction of this dominant cyclohexane source to the site. And the success of the nonparametric regression method enables to be judged by the comparison of the predicted location of this source determined by triangulation of these directions and given position from Air Toxic emission inventory. The direction given by the largest peak is 87.19° and 328.68° from the Clinton Dr. and Deer Park site respectively. Based on the site positions and the direction given by the largest peak, the estimated location of the dominant cyclohexane source is 29.7368N, 95.1752W. The distance between this estimated location and reported position from the inventory is about 0.546 km. This short distance shows the estimated location from nonparametric regression is in agreement with the inventory. The estimated direction corresponds to an error of 1.17° from the Clinton Dr. site and 1.09° from the Deer Park site. It is better than the result that 152 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. could be obtained from the bar chart, which would estimate the location with ± 1 0°. Consequently, these confirm that the nonparametric regression method provides a significant improvement to estimate the direction of the dominant cyclohexane source by using wind direction and concentration measurements. The location of this dominant source can be predicted precisely by using the directions estimated from two monitoring datasets. In addition, the wind speed data would be another factor to affect the accuracy of source location estimation from wind direction and concentration measurements. According to the knowledge of reported location from the inventory, the distance from Phillips to the Clinton Dr. and Deer Park site is about 5 miles and 6 miles respectively. We use these distances as one-hour traveling distance to screen wind direction and concentration measurements by high wind speed. The directions from new plots estimate a new location of the dominant cyclohexane source is 29.7376N, 95.1747W; the distance is 0.464 km between this new estimated location and the reported location from the inventory. We find the estimated location is closer to the reported location after using high wind speed to screen the measurements. This finding indicates wind speed is another factor we should consider applying to the source location analysis. The combination of nonparametric regression along with receptor modeling will give an advanced analysis for the identification of multiple emission sources. 153 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 7.1.2 Application for Emission Inventory Evaluation The VOC emission rates are the significant data for regulatory agencies and photochemical models. Any deficiency of emission rates of organic gases could severely impact the results of air quality models and the regulatory decision. Receptor modeling technique is a method providing an independent way to evaluate the emission inventory based on ambient observations. Unmix is a statistical receptor modeling and analysis software to derive the source compositions as well as source contributions. Nonparametric regression can be applied to the source contributions from Unmix for showing the source strength by wind direction. The combination of nonparametric regression along with Unmix will give the sharper results to evaluate the emission inventory. The 1997 monitoring VOC data from PAMS sites are analyzed by Unmix receptor modeling. Acetylene is set as the tracer to distinguish the roadway emission from the other industrial emissions. The most satisfactory results of applying Unmix to Clinton Dr. and Deer Park data are a 7-source and a 6 -source solution respectively. These two solutions are similar to each other. Three major sources account for about 70 % of the average TNMOC. These sources resemble the roadway emission, isobutane emission related to MTBE production, and nbutane emission related to the refinery source. The remaining 30% of the average TNMOC are contributed from the other smaller sources that are characterized by one or two significant species to relate with the special petrochemical industries. 154 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Nonparametric regression is applied to show the relationship of wind direction to the source contributions derived from Unmix. The plots indicate the amount of TNMOC contributed from industrial sources increases strongly with wind from the SSE sector to the Clinton Dr. site. This implies that these sources should be located close to the site in the direction given by the peak. Based on the plots, we speculate the largest industrial source should be located at around 161° to the Clinton Dr. site. Because the other industrial sources also show the major peak at 161°, the industrial sources located at this direction are considered as the same facility releasing the emissions from different refining processes. The discrepancy of the emission inventory has been found by comparing our results to the nearby reported emissions in the vicinity of the Clinton Dr. site. The comparison finds the direction given by the peaks is not consistent with the major emissions reported from the inventory. We are only able to match the smallest source to a nearby petrochemical plant in the vicinity of the site. The other major emissions reported from the inventory failed to agree with the direction shown by the source contribution plots. The same inconsistency reported from 1993 field data has been found in this 1997 data study. We believe that the inventory of industrial YOC emission is inaccurate in the location and emission rates of major sources. And the inaccuracy still remains in the emission inventory. The future work should go over these emission data and get rid of inaccuracy from the emission inventory. 155 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 7.2 Suggestions for Future Studies This research has conducted to development and application of wind direction analysis to air quality data by nonparametric regression. Nonparametric regression improves the estimation of source direction from the plot of wind direction analysis. Based on the estimated direction given by the peaks, we are able to locate the largest emission sources precisely. In addition, the nonparametric regression can be applied to the source contributions derived from receptor modeling for showing the plot of source strength by the wind direction. The plot enables to give an advanced analysis to evaluate the emission inventory along with the source composition and find the inaccuracy. As the previous discussion in Chapter 5, a precise estimate of the source location can be determined when data are screened by higher wind speed. This finding suggests wind speed is a potential factor to determine the source location by ambient observations. Hourly wind speed data can be considered as the pollutant traveling distance within one hour. Based on this concept, the pollutant concentrations will be high when the value of wind speed is equal to the distance from the source to the site. Therefore, the distance from the source to the site can be determined from the peaks given by the plot of wind speed and pollutant concentration measurements. Combination of wind speed into wind direction analysis should provide a plot to illustrate not only the source direction but also the 156 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. distance to the source. This will make it possible to identify the locations of multiple sources from single monitoring dataset. In order to involve wind speed for the calculation, the equation 3-10 is modified to calculate the estimated average concentration at the direction 0 and speed c o as the following: C (6, A0,co, A c o ) = ±K(^-yK(^.yci Ad______ A c o A 0 A c o (7-1) The co is the given wind speed and Ut is the wind speed measurement from the observations. The Acu is the smoothing parameter for kernel K. Based on this equation, a three-dimensions plot can be produced to illustrate the concentration peaks by wind direction and speed, as seen in Figure 7-1. Figure 7-1 presents a three-dimensions plot to show the cyclohexane concentration peaks at the Clinton Dr. site. The estimated average concentrations are calculated by Gaussian kernel based on wind direction and wind speed. The reading of wind speed is considered as the distance from the site to the source. Therefore, the peaks shown at 90° but different wind speed could imply the distance from the site to those different remote sources lying on this direction. The source location could be determined by the direction and distance given by the concentration peaks. 157 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. w in d direction Figure 7-1 The 3-D plot of wind direction and wind speed versus cyclohexane concentrations at the Clinton Dr. site. The FWFCVl for wind direction and wind speed is 1 0° and 2 mile/hr. 158 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Another method to involve wind speed and wind direction for showing the source location is the back trajectory model. This method uses wind direction and speed to describe the paths taken by air parcels. The end-point of paths will fall in one of grid cells that are presumed in the monitoring area. In the grid cell, the air parcel is assumed to collect the pollutant in that cell. Therefore, the measured concentration can be considered as the emissions from one of grid cells determined by the wind direction and speed measurements. Nonparametric regression is used to calculate the weighted concentration of each path in the same cell. The sum of weighted concentrations is presented as the average emission in that cell. The results can show a concentration map to indicate the possible location of the emission sources, as seen in Figure 7-2. The combination of wind speed and wind direction analysis is much complicated than wind direction analysis. The possible reason is the concentration effects made by wind speed cannot be simple to isolate as the individual measurements, especially in low wind speed. For example, the pollutants from remote sources can be detected by the low wind speed when wind blows from the same direction for several hours. However, the measurements of wind speed sometimes are extremely changeable. This will increase the difficulty for source analysis. However, nonparametric regression doesn’t have appropriate formula to determine the parameters for bi-variables analysis. The current results for wind speed analysis is 159 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. * 115 - k 10 • -5 I q Figure 7-2 The plot of back trajectory for cyclohexane concentrations at the Deer Park site. The size of grid cells is 450m*450m. The Deer Park site is marked as “+” and the Clinton Dr. site is marked as “x”. The Phillips Corporation is marked as 160 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. not satisfied. But we believe the analysis of wind speed should give the specific viewpoint to speculate the emission source based on ambient observations. This study will provide the valuable insight to the source location identification. 161 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. REFERENCES Aardenne, J.A. van, Builtjes, P.J.H., Hordijk, L., Kroeze, C., and Pulles, M.P.J. (2002). Using wind-direction-dependent difference between model calculations and field measurements as indicator for the inaccuracy of emission inventories. Atmospheric Environment, 36, 1195-1204. Abramowitz, Milton and Stegun, Irene A. (1972). Handbook o f Mathematical Functions with Formulas, Graphs, and Mathematical Tables. New York: John Wiley & Sons. Altman, N. S. (1992). An introduction to kernel and nearest-neighbor nonparametric regression. The American Statistician, 46, 175-185. Atkinson, Roger (2000). Atmospheric chemistry of VOCs and NOx. Atmospheric Environment, 34, 2063-2101. Ashley, David L., Bonin, Michael A., Cardinali, Frederick L., McCraw, Hoan M., Holler, James S., Needham, Larry L., and Patterson, Donald G, Jr. (1992). Determining volatile organic compounds in human blood from a large sample population by using purge and trap gas chromatography/mass spectrometry. Analytical Chemistry, 64, 1021-1029. Batterman, S. A., Fay, J.A., and Golomb, D. (1987). Significance of regional source contributions to urban PM-10 concentrations. Journal o f Air Pollution Control Association, 37, 1286-1291. Bolla, Karen I. (1991). Neuropsychological assessment for detecting adverse effects of volatile organic compounds on the central nervous system. Environmental Health Perspectives, 95, 93-98. Brauer, Michael and Brook, Jeffrey R. (1997). Ozone personal exposures and health effects for selected groups residing in the Fraser valley. Atmospheric Environment, 31, 2113-2121. Cardelino C. A. and Chameides, W. L. (1995). An observation-based model for analyzing ozone precursor relationship in the urban atmosphere. Journal o f the Air and Waste Management Association, 45, 161-180. Carter, Ray E., Lane, Dennis D., Marotz, Glen A., Chaffin, Charles T., Marshall, Tim L., Tucker Melissa, Witkowski, Mark R , Hammaker, Robert M., Fateley, William G., Tomas, Mark J., and Hudson, Jody L. (1993). A method of predicting point and path-average ambient air VOC concentrations, using meteorological data. Journal o f the Air and Waste Management Association, 43, 480-488. 162 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Cheng, Shouquan and Lam, Kin-Che (1998). An analysis of winds affecting air pollution concentrations in Hong Long. Atmospheric Environment, 32, 2559-2567. Chu, Shao-Hang (1995). Meteorological considerations in siting photochemical pollutant monitors. Atmospheric Environment, 29, 2905-2913. Cleveland, William S. and Kleiner, Beat (1975). A graphical technique for enhancing scatterplots with moving statistics. Technometrics, 17, 447-454. Cleveland, William S. and Devlin, Susan J. (1988). Locally weighted regression: an approach to regression analysis by local fitting. Journal o f the American Statistical Assoication, 83, 596-610. Chock, David P. and Heuss, Jon M. (1987). Urban ozone and its precursors. Environmental Science and Technology, 21,1146-1153. Conner, Teri L., Lonneman, William A., and Seila, Robert L. (1995). Transportation-related volatile hydrocarbon source profiles measured in Atlanta. Journal o f the Air and Waste Management Association, 45, 383-394. Davis, J. M. and Speckman, P. (1999). A model for predicting maximum and 8 h average ozone in Houston. Atmospheric Environment, 33, 2487-2500. Doskey, Paul V., Porter, Joseph A., and Scheff, Peter A. (1992). Source fingerprints for volatile non-methane hydrocarbons. Journal of the Air and waste Management Association, 42, 1437-1445. Eklund, Bart (1999). Comparison of line and point source releases of tracer gases. Atmospheric Environment, 33, 1065-1071. Eubank, Randall L. (1994). A simple smoothing spline. The American Statistician, 48, 103-106. Eubank, Randall L. (1999). Nonparametric Regression and Spline Smoothing (second edition). New York: Marcel Dekker, Inc Evans, Gary F., Lumpkin, Thomas A., Smith, Deborah L., and Somerville, Matthew C. (1992). Journal o f the Air and Waste Management Association, 42, 1319-1323. Finlayson-Pitts, B. J. and Pitts, J. N. Jr (1993). Atmospheric chemistry of tropospheric ozone formation: scientific and regulatory implications. Journal o f the Air and Waste Management Association, 43, 1091-1100. 163 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Forstner, Hali J. L., Flagan, Richard C., and Seinfeld, John H. (1997). Secondary organic aerosols from the photooxidation of aromatic hydrocarbons: molecular composition. Environmental Science and Technology, 31, 1345-1358. Forstner, Hali J. L., Flagan, Richard C., and Seinfeld, John H. (1997). Molecular speciation of secondary organic aerosol from photooxidation of the higher alkenes: 1-octane and 1-decene. Atmospheric Environment, 31, 1953-1964. Fox, John (2000). Nonparametric Simple Regression: Smoothing Scatterplots. In lewis-Beck (Michael S. Ed.). Sage University Papers Series: Qauantitative Applications in the Social Sciences. Thousand Oaks: Sage Publications, Series/Number 07-130. Fujita, Eric M., Watson, John G., Chow, Judith C, and Magliano, Karen L. (1995). Receptor model and emissions inventory source apportionments of nonmethane organic gases in California’s San Joaquin valley and San Francisco Bay area. Atmospheric Environment, 29, 3019-3035. Funk, T. H., Chinkin, L. R., Roberts, P. T., Saeger, M., Mulligan, S., Paramo Figueroa, V. H., and Yarbrough, J. (2001). Compilation and evaluation of a Paso del Norte emission inventory. The Science o f the Total Environment, 276, 135-151. Gasser, Theo and Muller, Hans-Georg (1984). Estimating regression functions and their derivatives by the kernel method. Scand J. Statist, 11, 171-185. Grover, Rajiv and Bradford, Mike L. (2001). Texas NOx state implementation plan for Houston-Galveston area. Environmental Progress, 20, 197-205. Grosjean, Daniel and William II, Edwin L. (1992). Photochemical pollution at two sourthern California smog receptor sites. Journal o f the Air and Waste Management Association, 42, 805-809. Gerin, Michel, Siemiatycki, J., Desy, M., and Krewski, D. (1998). Associations between several sites of cancer and occupational exposure to benzene, toluene, xylene, and styrene: results of a case-control study in Montreal. American Journal o f Industrial Medicine, 34, 144-156. Haagen-Smit, A. J. (1952). Chemistry and physiology of Los Angeles smog. Industrial and Engineering Chemistry, 44, 1342-1346. Haan, Peter De. (1999). On the use of density kernels for concentration estimations within particle and puff dispersion models. Atmospheric Environment, 33, 2007- 2021 . 164 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Hadjiiski, Lubomir and Hopke, Phillip (2000). Application of articial neural networks to modeling and prediction of ambient ozone concentrations. Journal o f the Air and Waste Management Association, 50, 894-901. Harley, Robert A. and Sawyer, Robert F. (1997). Undated photochemical modeling for California’s South Coast Air Basin: Comparison of chemical mechanisms and motor vehicle emission inventories. Environmental Science and Technology, 31, 2829-2839. Henry, Ronald C. Multivariate receptor Models in Receptor Models fo r Air Quality Management (Hopke, P. K. Ed.). Amsterdam: Elsevier, (1991), pp. 117-148. Henry, Ronald C. (1997). History and fundamentals of multivariate air quality receptor models. Chemometrics and Intelligent Laboratory Systems, 37, 525-530. Henry, Ronald C. Unmix version 2.3 Manual (2001), available with Unmix software (rhenry@usc.edu). Henry, Ronald C. (2002a). Multivariate receptor modeling by N-dimensional edge detection. Chemometrics and Intelligent Laboratory Systems, submitted for publication. Henry, Ronald C. Receptor Modeling in The Encyclopedia o f Environmetrics (El- Shaarawi, Abdel H. and Piegorsch, Walter W. Ed.). Chichester: John Wiley & Sons Ltd, (2002b), Vol. 3, pp. 1706-1721. Henry, Ronald C., Chang, Yu-Shuo, and Spiegelman, Clifford H. (2002). Locating nearby sources of air pollution by nonparametric regression of atmospheric concentrations on wind direction. Atmospheric Environment, 36, 2237-2244. Henry, Ronald C., Lewis, Charles W., and Collins, John F. (1998). Vehicle-related hydrocarbon surce compositions from ambient data: The GRACE/SAFER method. Environmental Science and Technology, 28, 823-832. Henry, Ronald C., Park, E. S., and Spiegelman, C. H. (1999). Comparing a new algorithm with the classic methods for estimating the number of factors. Chemometrics and Intelligent Laboratory Systems, 48, 91-97. Henry, Ronald C., Spiegelman, Clifford H., Collins, John F., and Park, EunSug (1997). Reported emissions of volatile organic compounds are not consistent with observations. Proceeding o f the National Academy o f Sciences, 94, 6596-6599. 165 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Hollander, Myles and Wolfe, Douglas A. (1999). Nonparametric Statistical Methods (second edition). New York: John Wiley & Sons Inc. Incecik, S. (1996). Investigation atmospheric conditions in Istanbul leading to air pollutaion episodes. Atmospheric Environment, 30, 2739-2749. Jones, A. P. (1999). Indoor air quality and health. Atmospheric Environment, 33, 4535-4564. Kenski, Donna M., Wadden, Richard A., Scheff, Peter A., and Lonneman, William A. (1995). Receptor modeling approach to VOC emission inventory validation. Journal of Environmental Engineering, 121, 483-491. Kirchstetter, Thomas W., Singer, Brett C., and Harley, Robert A. (1996). Impact of oxygenated gasoline use of California light-duty vehicle emissions. Environmental Science and Technology, 30, 661-670. Kleindienst, T. E., Smith, D. F., Li, W ., Edney, E. O., Driscoll, D. J., Speer, R. E., and Weathers, W. S. (1999). Secondary organic aerosols formation from the oxidation of aromatic hydrocarbons in the presence of dry submicron ammonium sulfate aerosol. Atmospheric Environment, 33, 3669-3681. Kleinman, L.I., Daum, P.H., Imre, D., Lee, Y.-N., Nunnermacker, L. J., Springston, S. R., Weinstein-Lioyd, J., and Rudolph, J. (2002). Ozone production rate and hydrocarbon reactivity in 5 urban areas: a case of high ozone concentration in Houston. Geographical Research letters, 29, 10.1029/2001GL014569. Kretzschmar, J. G. and Cosemans, G. (1979). A five year survey of some heavy metal levels in air at the Belgian north sea coast. Atmospheric Environment, 13, 267-277. Kriews, M., Naumann, K., and Dannecker, W. (1988). Aerosol specification in the sourthern north sea region by wind dependent sampling and multielement analysis. Journal o f Aerosol Science, 19, 1051 -1054. LaGrone, F. Scott (1991). Potential community exposure to toxic chemicals. Environmental Science and Technology, 25, 366-368. Lewis, Charles W., Henry, Ronald C., and Shreffler, Jack H. (1998). An exploratory look at hydrocarbon data from the Photochemical Assessment Monitoring Stations network. Journal o f the Air and Waste Management Assoication, 48, 71 -76. Lewis Charles W., Norris, Gary A., Henry, Roland C., and Conner, Teri L. (2002) Source apportionment of Phoenix PM2.5 aerosol with the Unmix receptor model. Journal o f the Air and Waste Management Association, submitted in published. 166 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Lin, Chitsan and Milford, Jana B. (1994). Decay-adjusted chemical mass balance receptor modeling for volatile organic compounds. Atmospheric Environment, 28, 3261-3276. Lindsay, Ronald W., Richardson, Jennifer L., and Chameides, William L. (1989). Ozone trends in Atlanta, George: Have emission controls been effective?. Journal of Air Pollution Control Association, 39, 40-43. Lippmann, Morton. (1991). Health effects of tropospheric ozone. Environmental Science and Technology, 25, 1955- 1962. Lonneman, William A., Sella, Robert L., and Meeks, Sarah A. (1986). Non-methane organic composition in the Lincoln Tunnel. Environmental Science and Technology, 20, 790-796. Lorimer, Graeme S. (1986). The kernel method for air quality modeling-!: mathematical foundation. Atmospheric Environment, 2 0 , 1447-1452. Lynge, Elsebeth, Anttila, A., and Hemminki, K. (1997). Organic solvents and cancer. Cancer Causes and Control, 8 , 406-419. Mannschreck, K., Klemp, D., Kley, D., Friedrich, R , Kuhlwein, J., Wickert, B., Matuska, P., Habram, M., and Slemr, F. (2002). Evaluation of an emission inventory by comparisons of modeled and measured emission ratios of individual HCs, CO andNOx. Atmospheric Environment, 36 Supplement, S81-S94. Mayers Robert A. (1997). Handbook o f Petroleum Refining Processes (second edition). New York: McGraw-Hill McLaughlin, S. B. and Downing, D. J. (1995). Interactive effects of ambient ozone and climate measured on growth of mature forest trees. Nature, 374, 252-254. Miller, Alison A. and Sager, Thomas W. (1994). Site redundancy in urban ozone monitoring. Journal o f the Air and Waste Management Association, 44, 1097-1102. Mukund, R., Kelly, Thomas J., and Spicer, Chester W. (1996). Source attribution of ambient air toxic and other VOCs in Columbus, Ohio. Atmospheric Environment, 30, 3457-3470. Mulholland, Mechael, Seinfeld, John H. (1995). Inverse air pollution modeling of urban-scale carbon monoxide emissions. Atmospheric Environment, 29, 497-516. 167 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Okabayashi, K., Ide, Y., Kitabayashi K., Okamoto, S., and Kobayashi, K. (1996). Effects of wind directional fluctuations on gas diffusion over a model terrain. Atmospheric Environment, 30, 2871-2880. Odum, J. R., Jungkamp, T. P. W., Griffin, R. J., Flagan, R. C., and Shinfled, J. H. (1997). The atmospheric aerosol-forming potential of whole gasoline vapor. Scinece, 276, 96-99. Paige, Renee C., Wong, Viviana, and Plopper, Charles G. (2 0 0 0 ). Long-term exposure to ozone increases acute pulmonary centriacinar injury by 1- nitronaphthalene: II. Quantitative histopathology. The Journal o f Pharmacology and Experimental Therapeutics, 295, 942-950. Pandis, Spyros N., Harley, Robert A., Cass, Glen R., and Seinfeld, John H. (1992). Secondary organic aerosol formation and transport. Atmospheric Environment, 26A, 2269-2282. Pitts, JR., James N. (1993). Atmospheric formation and fates of toxic ambient air pollutants. Occupational Medicine, 8 , 621-662. Scheff, Peter A. and Wadden, Richard A. (1993). Receptor modeling of volatile organic compounds. 1. Emission inventory and validation. Environmental Science and Technology, 27, 617-625. Scheff, Peter A., Wadden, Richard A., Kenski, Donna M., Chung, Joseph, and Wolff, George (1996). Receptor model evaluation of the Southeast Michigan Ozone Study ambient NMOC measurements. Journal o f the Air and Waste Management Assoication, 46, 1048-1057. Schmidt, Robert J., Bogdan, Paula L., and Gilsdort, Normal L. (1993). Meeting the challenge of reformulated gasoline. Chemtech, 23, 41-46. Schmidt, R. W. H., Slemr, F., and Schurath, U. (1998). Airborne peroxyacetyl nitrate (PAN) and peroxypropionyl nitrate (PPN) measurements during tract 1992. Atmospheric Environment, 32, 1203-1227. Schwartz, Joel (1994). Nonparametric smoothing in the analysis o f air pollution and respiratory illness. The Canadian Journal o f Statistics, 2 2 , 471-487. Seinfeld, John H. (1986). Atmospheric Chemistry and Physics o f Air Pollution. New York: Wiley. Seinfeld, John H. (1989), Urban air pollution: State of the science. Science, 243, 745-752. 168 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Sheinfeld, John H. (1995). Ozone formation and photochemical smog in urban and regional atmosphere. Special Publication, 170, 65-72. Shreffler, Jack H. (1993). Comparison of nonmethane organic compound concentration data collected by two methods in Atlanta. Journal o f the Air and Waste Management Association, 43, 1576-1584. Siddiqi, Aziz A. and Worley, Jr. Frank L. (1977). Urban and industrial air pollution in Houston, Texas - 1 . Hydrocarbons. Atmospheric Environment, 11, 131-143. Silverman, B. W. (1986). Density Estimation fo r Statistics and Data Analysis. Londan: New York: Chapman & Hall. Slgsby, Jr., John E., Tejada, Silvestre B., Ray, William D., Lang, John M., and Duncan, John W. (1987). Volatile organic compound emissions from 46 in-use passenger cars. Environmental Science and Technology, 21, 466-475. Smith, Richard L. and Shively, Thomas S. (1995). Point process approach to modeling trends in tropospheric ozone based on exceedances of a high threshold. Atmospheric Environment, 29, 3489-3499, Smith, Simon (1993). Volatile organic compounds. Environmental Elealth, 101, 156-159. Somerville, M. C., Mukerjee, S., Fox, D. L., and Stevens, R. K. (1994). Statistical approaches in wind sector analysis for assessing local source impacts. Atmospheric Environment, 28, 3483-3493. Somerville, M. C., Mukerjee, S., and Fox, D. L. (1996). Estimating the wind direction of maximum air pollutant concentration. Environmetrics, 7, 231 -243. Sperber, Kenneth R. (1987). The concentration and deposition of nitrate, sulfate, and ammonium as a Junction of wind direction from precipitation samples. Atmospheric Environment, 21, 2629-2641. Sprent P. and Smeeton, N. C. (2000). Applied Nonparametric Statistical Methods (third edition). Boca Raton: Chapman & Hall/CRC. Stern, Bonnie R., Raizenne, Mark E., Burnett, Richard T., Jones, Linda, Keareny, Jill, and Franklin, Claire A. (1994). Air pollution and childhood respiratory health: Exposure to sulfate and ozone in 10 Canadian rural communities. Environmental Research, 66, 125-142. 169 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Sweet, Clyde W. and Vermette, Stephen J. (1992). Toxic volatile organic compounds in urban air in Illinois. Environmental Science and Technology, 26,165- 173. Stump, Fred D., Knapp, Kenneth T., Ray, William D., Snow, Richard, and Burton, Charles (1992). The composition of motor vehicle organic emissions under elevated temperature summer driving conditions (75 to 105°F). Journal o f the Air and Waste Management Association, 42, 152-158. U.S. EPA (1994). Recommdenations fo r analysis o f PAMS data. Office of Air Quality Planning and Standards, Research triangle Park, North Carolina, EPA 6 8 - D3-0019. U.S. EPA (1997). Photochemical Assessment Monitoring Stations (PAMS) performance evaluation program. Office of Air Quality Planning and Standards, Research triangle Park, North Carolina, EPA 68-D3-0095. U.S. EPA (2000). Taking toxics out o f the air. Office of Air Quality Planning and Standards, Research triangle Park, North Carolina, EPA-452/K-00-002. Vyskocil, Adolf, Viau, C., and Lamy, S. (1998). Peroxyacetyl nitrate: review of toxicity. Human & Experimental Toxicology, 17, 2 1 2 -2 2 0 . Walker, Harry M. (1985). The-year ozone trends in California and Texas. Journal o f the Air Pollution Control Association, 35, 903-912. Wang, Shih-Chen, Paulson, Suzanne E., Grosjean, Daniel, Flagan, Richard C., and Seinfeld, John H. (1992). Aerosol formation and growth in atmospheric organic/NOx systems-I. Outdoors smog chamber studies of C7 and C8 hydrocarbons. Atmospheric Environment, 26A, 403-420. Ware, James H., Spengler, John D., Meas, Lucas M.,Samet, Jonathan M., Wanger, Gregory R., Coultas, D., Ozkaynak, H., and Schwab, M. (1993). Respiratory and irritant health effects of ambient volatile organic compounds. American Journal o f Epidemiology, 137, 1287-1301. Westenbarger, David A. and Frisvold, George B. (1994). Agricultural exposure to ozone and acid precipitation. Atmospheric Environment, 28, 2895-2907. Wolff, George T. and Korsog, Patricia E. (1992). Ozone control strategies based on the ratio of volatile organic compounds to nitrogen oxides. Journal o f the Air and Waste Management Association, 42, 1173-1177. 170 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Wiyschi, Hanspeter (1988). Ozone, nitrogen dioxide, and lung cancer: a review of some recent issues and problems. Toxicology, 48, 1-20. Ziomas, Ioannis C., Melas, Dimitrios, Zerefos, Chaistos S., and Paliatsos, Athanasios G. (1995). Forecasting peak pollutant levels from meteorological variables. Atmospheric Environment, 29, 3703-3711. Zweidinger, Roy B., Slgsby, Jr., John E., Tejada, Silvestre B., Stump, Fred D., Dropkin, Davis L., and Ray, William D. (1988). Detailed hydrocarbon and aldehyde mobile source emissions from roadway studies. Environmental Science and Technology, 22, 956-962. Scheff, Peter A., Wadden, Richard A., Bates, Barbara A., and Aronian, Paul F. (1989). Source fingerprints for receptor modeling of volatile organics. Journal o f the Air Pollution Control Association, 39, 469-478. 171 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Test generation for crosstalk noise in VLSI circuits

PDF

Nucleotide-based transport in ultrasonic fields

PDF

Development and evaluation of a personal particle sampler and a mobile high-capacity particle size classifier

PDF

Ultrasonic wave techniques and characterization of filled elastomers and biodegradable polymers

PDF

Understanding air quality data using nonparametric regression analysis

PDF

Space-time adaptive acquisition and demodulation in mobile cellular

PDF

The feasibility study of desulfurization of diesel fuels by ultrasound -assisted oxidative desulfurization (UAOD)

PDF

Investigation of sacrificial layer and building block for layered nanofabrication (LNF)

PDF

Technologies for assessment of personal (population) exposure to toxic components of airborne particulate matter

PDF

Nanofiltration of natural and synthetic chemicals with hydrogen peroxide/UV pretreatment process

PDF

Study of biopolymer-modified concrete system: Geopolymerization of biopolymers

PDF

Instrumentation and methods for PM(2.5) and ultrafine particulate matter exposure assessment

PDF

Biofiltration of petroleum hydrocarbon vapors and co-treatment of volatile organic compounds in low-pH sulfide biofilters

PDF

Techniques for audiovisual data confidentiality, integrity and copyright protection

PDF

SINR analysis and power control for downlink W-CDMA and multiuser OFDM systems

PDF

Technologies for assessment of physical, chemical and toxicological characteristics of ambient ultrafine particles

PDF

Mechanisms of acquisition of molecular genetic changes during tumor development and progression

PDF

Investigation of the physical and chemical characteristics of ambient coarse particulate matter in indoor and outdoor environments

PDF

Membrane bioreactor process for removing biodegradable organic matter and disinfection by -product precursors from water: Modeling and process efficiency

PDF

Probing the chemical control of mineral scale and metal corrosion at the microscopic level

Asset Metadata

Creator Chang, Yu-Shuo (author)

Core Title Locating nearby sources of air pollution using air quality data and wind direction

Degree Doctor of Philosophy

Degree Program Engineering

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag engineering, environmental,environmental sciences,OAI-PMH Harvest

Language English

Contributor Digitized by ProQuest (provenance)

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c16-358516

Unique identifier UC11334841

Identifier 3103868.pdf (filename),usctheses-c16-358516 (legacy record id)

Legacy Identifier 3103868.pdf

Dmrecord 358516

Document Type Dissertation

Rights Chang, Yu-Shuo

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA