Methodology - OhioLINK ETD

2 downloads 0 Views 2MB Size Report
Travel Survey to represent the static and dynamic activity of Franklin County ..... “Less Than High School/High School” Subgroup Paired T-Test Daytime.
Assessment of Particulate Matter Exposure in Franklin County, Ohio: A Comparison of Static and Dynamic Approaches Thesis

Presented in Partial Fulfillment of the Requirements for the Degree Masters of Arts in the Graduate School of The Ohio State University By Jaclyn Sineri, B.S. Graduate Program in Geography

The Ohio State University 2010

Thesis Committee: Mei Po Kwan, Co-Advisor Desheng Liu, Co-Advisor Ningchuan Xiao

   

Copyright by Jaclyn Sineri 2010

   

Abstract The primary focus of this study is to determine whether a significant difference exists between static and dynamic particulate matter exposure at an intraurban level for Franklin County residents. Static exposure is analyzed at a single location (household) for a twenty four hour period while dynamic exposure takes into account the various locations that an individual travels to throughout the day (i.e. work, school). An air dispersion model will be utilized in conjunction with industrial point source emissions inventory data to generate a continuous particulate matter grid at an hourly resolution across the county. In addition to the dispersion model output, background concentrations from three local monitors will be incorporated into the surface grid as representation for all other pollution sources not included in the model run. A sample population was extracted from the 1999 Mid-Ohio Area Household Travel Survey to represent the static and dynamic activity of Franklin County residents. The 24 hour and daytime activity of the individuals was associated with particulate matter concentrations at hourly intervals throughout his or her recorded travel day. In turn, the total static and dynamic exposure to particulate matter was examined using statistical analysis methods such at paired t-tests and ANOVA to determine if a significant difference is present. Also, income, age, gender, and education of each individual will serve as predictor variables in a multiple linear regression to determine if a relationship exists with particulate matter exposure. ii

Results from the statistical analysis did not reveal that individuals are exposed to significantly different levels of particulate matter concentration when in a static versus dynamic state. The lack in variance was attributed to the distance decay affect from the modeled point sources and the strong influence of meteorological phenomena on the modeled output (i.e. wind direction). Additionally, a dense monitor network was unavailable for Franklin County thus limiting the ability to account for linear (road network) and minor sources which greatly contribute to particulate matter concentrations.

iii

Dedication Dedicated to my supportive and loving husband and family

iv

Vita June 2006……………………………………..B.S. Geography/Meteorology, Ohio University July 2008 to present…………………………..Wind Farm Optimization Analyst NextEra Energy, Juno Beach, FL

Fields of Study Major Field: Geography

v

Table of Contents ABSTRACT........................................................................................................................II VITA...................................................................................................................................V LIST OF TABLES..........................................................................................................VIII LIST OF FIGURES ............................................................................................................X CHAPTER 1: INTRODUCTION...................................................................................... 1 CHAPTER 2: LITERATURE REVIEW........................................................................... 6 CHAPTER 4: METHODOLOGY ................................................................................... 33 4.1

Gaussian Plume Model .......................................................................................... 33

4.2 ISCST 3 Dispersion Model.................................................................................... 35 4.2.1 Control Pathway................................................................................................... 36 4.2.2 Source Pathway .................................................................................................... 38 4.2.3 Receptor Pathway................................................................................................. 39 4.2.4 Meteorological Pathway....................................................................................... 41 4.2.5 Output Pathway .................................................................................................... 43 4.3

Model Output and Interpolation............................................................................. 43

4.4

Background Concentration .................................................................................... 48

4.5 Exposure Analysis of Dynamic vs. Static Population ........................................... 54 4.5.1 Static Exposure..................................................................................................... 54 4.5.2 Dynamic Exposure ............................................................................................... 57 4.6

Dynamic vs. Static Statistical Analysis ................................................................. 65

4.7

Dynamic Exposure by Subgroup ........................................................................... 65

4.8

Multiple Linear Regression of Dynamic Exposure ............................................... 66 vi

CHAPTER 5: RESULTS................................................................................................. 67 5.1

Dynamic vs. Static Paired T-Test All Individuals ................................................. 67

5.2

Gender Subgroup Dynamic vs. Static Paired T-Test ............................................. 69

5.3

Income Subgroup Dynamic vs. Static Paired T-Test............................................. 70

5.4

Education Subgroup Dynamic vs. Static Paired T-Test......................................... 74

5.5

Age Subgroup Dynamic vs. Static Paired T-Test .................................................. 76

5.6

Gender Subgroup Dynamic ANOVA .................................................................... 78

5.7

Income Subgroup Dynamic ANOVA.................................................................... 79

5.8

Education Subgroup Dynamic ANOVA................................................................ 80

5.9

Age Subgroup Dynamic ANOVA ......................................................................... 81

5.7

Daytime and 24 Hour Multiple Linear Regression of Dynamic Exposure............ 83

CHAPTER 6: CONCLUSION ........................................................................................ 86 APPENDIX A: ADDITIONAL TABLES....................................................................... 94

vii

List of Tables Table 1. Monitors 1, 2, & 3 Data Recovery...................................................................... 49 Table 2. Sample of Households and Individuals Assessed for Static Exposure............... 55 Table 3. Sample of Households and Individuals Assessed for Dynamic Exposure ......... 58 Table 4. All Individuals Paired T-Test for Daytime Static vs. Dynamic Exposure.......... 67 Table 5. Descriptive Statistics for Daytime Static & Dynamic Exposure Difference ...... 68 Table 6. Men Subgroup Paired T-Test for Daytime Static vs. Dynamic Exposure.......... 69 Table 7. Women Subgroup Paired T-Test for Daytime Static vs. Dynamic Exposure .... 70 Table 8. “Low” Income Subgroup Paired T-Test Daytime Travel vs. Daytime Home Concentration.................................................................................................................... 71 Table 9. “Middle” Income Subgroup Paired T-Test Daytime Travel vs. Daytime Home Concentration.................................................................................................................... 72 Table 10. “High” Income Subgroup Paired T-Test Daytime Travel vs. Daytime Home Concentration.................................................................................................................... 73 Table 11. “Less Than High School/High School” Subgroup Paired T-Test Daytime Travel vs. Daytime Home Concentration ......................................................................... 74 Table 12. “Vocational/Some College” Subgroup Paired T-Test Daytime Travel vs. Daytime Home Concentration .......................................................................................... 75 Table 13. “Undergraduate/Graduate” Subgroup Paired T-Test Daytime Travel vs. Daytime Home Concentration .......................................................................................... 75 Table 14. “Young” Age Subgroup Paired T-Test Daytime Travel vs. Daytime Home Concentration.................................................................................................................... 76 Table 15. “Middle” Age Subgroup Paired T-Test Daytime Travel vs. Daytime Home Concentration.................................................................................................................... 77 Table 16. “Old” Age Subgroup Paired T-Test Daytime Travel vs. Daytime Home Concentration.................................................................................................................... 77 Table 17. Gender Subgroup Daytime Dynamic Concentration ANOVA......................... 79 Table 18. Gender Subgroup 24 Hour Dynamic Concentration ANOVA ......................... 79 Table 19. Income Subgroup Daytime Dynamic Concentration ANOVA ........................ 80 Table 20. Income Subgroup 24 Hour Dynamic Concentration ANOVA......................... 80 Table 21. Education Subgroup Daytime Dynamic Concentration ANOVA .................... 81 Table 22. Education Subgroup 24 Hour Dynamic Concentration ANOVA..................... 81 Table 23. Age Subgroup Daytime Dynamic Concentration ANOVA.............................. 82 Table 24. Age Subgroup 24 Hour Dynamic Concentration ANOVA .............................. 82 Table 25. Multiple Linear Regression of 24 Hour Particulate Matter Concentrations and Income, Age, Education, and Gender Variables............................................................... 83 Table 26. Multiple Linear Regression of Daytime Particulate Matter Concentrations and Income, Age, Education, and Gender Variables............................................................... 85 Table 27. Industrial Point Source Dispersion Model Input Parameter ............................ 95 viii

Table 28. 1999 Hour 0 Monitor Readings for Monitors 1, 2, and 3 ............................... 101 Table 29. April 5th Derived Background Concentration (Δ) at Monitors 1, 2, and 3 ..... 103 Table 30. April 7th Derived Background Concentration (Δ) at Monitors 1, 2, and 3 ..... 104 Table 31. April 9th Derived Background Concentration (Δ) at Monitors 1, 2, and 3 ..... 105 Table 32. March 2nd Derived Background Concentration (Δ) at Monitors 1, 2, and 3... 106 Table 33. March 4th Derived Background Concentration (Δ) at Monitors 1, 2, and 3 ... 107 Table 34. March 5th Derived Background Concentration (Δ) at Monitors 1, 2, and 3 ... 108 Table 35. May 6th Derived Background Concentration (Δ) at Monitors 1, 2, and 3 ...... 109 Table 36. May 7th Derived Background Concentration (Δ) at Monitors 1, 2, and 3 ...... 110

ix

List of Figures Figure 1. Franklin County monitors that sample PM10 concentration levels (1999) ........ 26 Figure 2. Franklin County 119 major industrial point sources (1999) ............................ 28 Figure 3. Gaussian plume dispersion from point source................................................... 34 Figure 4. Dispersion Model Configuration Summary ...................................................... 38 Figure 5. Receptor Locations with 1 kilometer separation ............................................... 41 Figure 6. Highest 72 Hour Concentration Estimation for March 2nd, 4th, 5th ................... 44 Figure 7. Highest 72 Hour Concentration Estimation for April 5th, 7th, 9th ...................... 45 Figure 8. Highest 72 Hour Concentration Estimation for May 4th, 6th, 7th .................... 46 Figure 9. Potential Areas of High PM10 Concentrations (μg/m3) ..................................... 47 Figure 10. Monitor 1 and 2 Linear Regression ................................................................. 50 Figure 11. Monitor 1 Residual Plot from Monitor 1 & 2 Linear Regression ................... 50 Figure 12. Monitor 1 and 3 Linear Regression ................................................................. 51 Figure 13. Monitor 1 Residual from Monitor 1 & 3 Linear Regression Plot ................... 51 Figure 14. Franklin County Road Network ...................................................................... 60 Figure 15. Shortest route between start and destination locations.................................... 62 Figure 16. Point vertices along selected line segment ...................................................... 64 Figure 17. Industrial source in relation to receptor location and model output with interpolated raster for hour 11 on May 7th ........................................................................ 87 Figure 18. May 7th Hourly Model Output (2 – 9 A.M.)................................................... 90

x

 

Chapter 1: Introduction Increased concern involving human health effects associated with exposure to intraurban air pollution has amplified recent attempts to determine the spatial distribution of harmful pollutants and the daily impacts on surrounding communities. In efforts to understand characteristic dispersal patterns of primary pollutants, past studies have integrated various forms of air pollution exposure models to analyze and estimate spatial and temporal variation within the realms of populated city areas. A critical component of this study is to obtain an understanding of the essential elements that comprise these models and select an appropriate method to analyze the distribution of particulate matter across Franklin County, Ohio. Earlier research utilized a basic proximity assessment approach which estimated exposure based on individuals’ proximity to primary sources of pollution, particularly road networks and industrial sites. More advanced methods relied on deterministic and stochastic geospatial techniques involving interpolation of concentration measurements obtained from local government monitoring sites throughout the study area. Additionally, researchers developed area specific methods through the application of land use regression models which predicted air pollution concentration surfaces based on surrounding land use and traffic characteristics. An alternative to these methods is the application of air dispersion modeling software that was also a common procedure used for creating continuous pollutant 1

 

surfaces in past research. This option offers a high spatial and temporal substitute if a dense government monitoring network is not accessible. In terms of technicality, air dispersion software relies on deterministic assumptions (Jerrett et al., 2005) by making use of meteorological, topographic, and emissions data in order to establish exposure estimations for specific regions. As expected with any research, several advantages and disadvantages can be accredited to all forms of air pollution exposure models that have been practiced in past and present studies. In the case of proximity assessment, the simplistic and straight forward approach can offer a relatively low cost alternative when monitors or modeling software is unattainable. However, meteorological, topographic and distance decay effects on traffic related pollution is often times disregarded as an influential variable on dispersal patterns. Additionally, interpolation techniques and land use regression models make use of observed air pollution concentrations but the accessibility to dense monitoring networks is often times unfeasible. This is especially relevant when a high temporal and spatial representation of particular air pollutants is sought after. On the other hand, air dispersion modeling can accommodate for the lack in spatial and temporal resolution. However costly input and unrealistic assumptions pertaining to dispersion patterns may result in a biased outcome. Another problematic issue that is often encountered in air pollution research is the methodology associated with assigning exposure values to the individuals involved in the study. Often times, individuals are analyzed in a static state meaning exposure estimations are solely allocated to household, work, or school locations. Generally, this 2

 

is prevalent when proximity assessment is the primary form of exposure evaluation. To account for daily activity outside of these principal locations, it is recommended to incorporate time activity data for those who are analyzed in the study. Utilizing comparable methodologies from past studies, an air dispersion modeling software package will be integrated into this research in order to obtain a high spatial and temporal degree of concentration levels across the study area. Given the limitations of an extremely sparse government monitoring network in Franklin County, the air dispersion software will accommodate for the drawbacks associated with the lack in spatial and temporal coverage of particulate matter exposure. Furthermore, the software will incorporate meteorological and topographic influences on dispersal patterns based on accurate emissions inventory for pollution sources positioned throughout the study area. In terms of pollution sources, particulate matter can be described by the source category, aerodynamic diameter, and the relation of the particle size to the primary supplier. Principal sources of particulate matter are: industrial processes (39%), transportation (24%), fuel combustion (22%), miscellaneous (indoor sources, environmental tobacco smoke, wood burning, and cooking) (12%), and solid waste (3%). The associated categories can be divided into burning of fuel (power plant emissions), unpaved roads, industry, wood burning stoves, pollen, dust, salt spray, erosion, and mold. Furthermore, particulate matter emissions can be attributed to certain portions of the country (Bell et al., 2005). Typically, the eastern United States experiences particulate matter concentrations comprised primarily of sulfate which is characteristic of emissions from power plants. On the other hand, the western portion of the country is generally exposed to particulate matter that is comprised of nitrate due to heavy 3

 

transportation emissions. Finally, the northwest region experiences wood burning related particulate matter while the southwest is exposed to windblown dust from desert climates. Lastly, the size of the particle in relation to the source is another characteristic aspect of the various forms of particulate matter. In practice, the aerodynamic diameter of the particle is used to classify particulate matter. The four types include: total suspended particles (45 microns or less), PM10 (10 microns or less), PM2.5 (2.5 microns or less), and ultrafine (0.2 microns or less). In general, PM10 is attributed to fuel combustion, industrial processes, transportation, and road dust while PM2.5 is characteristic of combustion processes, particularly diesel exhaust (Bell et al., 2005). Given the input requirements for air dispersion modeling software, attainable emissions data, and knowledge of particulate matter, the focal point of this study will revolve around particulate matter of ten microns or less in diameter that is emitted from stationary industrial point sources located throughout Franklin County. Another advantage to this research is the use of time activity data for a select group of Franklin County residents. As previously mentioned, daily travel patterns associated with every day activity are often times unaccounted for when determining exposure estimations. In reality, the average individual travels to several locations throughout a twenty four hour period thus neglecting the dynamic nature of human activity could force substantial error on exposure assessment. This study will strive to determine an accurate exposure for the sample population by incorporating travel activity diaries recorded by residents over a twenty four hour period.

4

 

As an overview, the primary focus of this study is to determine whether a significant difference exists between static and dynamic particulate matter exposure at an intraurban level for Franklin County residents. An air dispersion model will be utilized in conjunction with industrial point source emissions inventory data to generate a continuous particulate matter surface grid at an hourly resolution across the county. In addition to the dispersion model output, background concentrations from three local monitors will be incorporated into the surface grid as representation for all other pollution sources not included in the model run. Additionally, a sample population will be extracted from the 1999 Mid-Ohio Area Household Travel Survey to represent the static and dynamic activity of Franklin County residents. The 24 hour and daytime activity of the individuals will be associated with particulate matter concentrations at hourly intervals throughout his or her recorded travel day. In turn, the total static and dynamic exposure to particulate matter will be examined using statistical analysis methods such at paired t-tests and ANOVA to determine if a significant difference is present. Also, income, age, gender, and education of each individual will serve as predictor variables in a multiple linear regression to determine if a relationship exists with particulate matter exposure.

5

 

Chapter 2: Literature Review Understanding the dynamic nature of air pollution dispersion and associated human health effects at an intraurban scale has been a major focus for recent epidemiological studies. An attempt to determine which portion of the population lies at greatest risk must begin with analyzing the major air pollution sources of concern. First, what are the primary contributors of a specific pollutant? What external factors influence the diffusion of air borne particles? Furthermore, how can one accurately portray relevant pollution concentrations across a particular study area? To examine these factors, several studies have developed and utilized air pollution exposure models to determine pollution concentrations and associated health effects at an intraurban level. The initial objective of this research is to adopt an air pollution exposure model that will create a continuous surface for particulate matter concentrations across Franklin County, Ohio. The creation of this surface will provide the spatial variations in particulate matter concentrations throughout the county. Additionally, a series of time activity patterns for Franklin county residents will be analyzed with respect to the varying levels of particulate matter concentrations over time; therefore individuals at greatest risk for high exposure levels can be determined. An analysis of existing air pollution exposure models has been categorized into six main classes based on common methodology. The six methods include: proximity based assessments, statistical interpolation, land use regression models, line dispersion 6

 

models, integrated emission-meteorological models, and hybrid models (Jerrett et al., 2005). Each method entails varying levels of complexity, data requirement, time, and sophistication. Several researchers have conducted health related studies using various combinations of these six air exposure models. However, this review will focus on the comparison between proximity based assessments, statistical interpolation, land use regression models, and dispersion models. It will also examine parameters pertaining to pollutant data, specifically particulate matter, such as: the number of monitors used to collect data, monitor locality with respect to the study area, how often data was collected, and what time period was implemented. Additionally, spatial characteristics of the study area will be explored, particularly scale. For example, did the analysis take place over a state, city, neighborhood, or isolated locations? Furthermore, did the study incorporate external factors such as meteorological conditions or topography? Proximity based assessments establish relationships between air pollution and exposure by assuming that humans face greater risk if they are located near major emission sources (Jerrett et al., 2005). A common approach to this method is to relate distance to heavily traveled roadways, traffic density, and surrounding populations. Venn et al. (2001) adopted this approach while studying asthma related symptoms in school children living close to major highways. The study area, located in Nottingham, United Kingdom, was divided into a one meter resolution grid with the center point corresponding to an individual postcode (approximately 15 adjacent households). The study population consisted of a case control sample of 6,147 primary school children (4

7

 

to 11 years) and a random cross sectional sample of 3,709 secondary school children (11 to 16 years). The proximity based approach was utilized by applying a buffer around major roadways that intersected the study area. Venn et al. (2001) determined from previous studies that effects of road vehicle traffic pollution were most prominent 150 meters from major highways. Several studies have agreed that pollutants are higher than ambient background levels at this particular distance. After establishing this relationship, a correlation was made between proximity of the school children’s home to the nearest main road and the odds of experiencing asthma related symptoms. A logistic regression was implemented to determine that the majority of the increased risk occurred at approximately 90 meters from the major roadways. Similarly, Janssen et al. (2001) expanded on Venn’s study by gathering actual measurements of traffic related air pollutants outside and inside of schools located 400 meters from a major highway in the Netherlands. Common traffic related pollutants analyzed in this study were PM2.5, NO2, and benzene. Also, traffic density of each major roadway, distance to the roadway, and percentage of time downwind during measurements were observed. Compared to the previous study, Janssen et al. (2001) applied a more complex approach by recognizing external meteorological factors and implementing personal monitored data of specific pollutants of concern. Temporal and spatial variations existed between the three monitored pollutants. In particular, PM2.5 measurements were taken both inside and outside of the schools. Indoor sampling occurred during school hours only, while outdoor sampling took place every day of the week for 15 minute increments. Finally, a weekly average was assigned 8

 

to each school over a 16 week period. Additionally, traffic counts (both cars and trucks) for a one year period were provided for all major roadways along with hourly wind direction data that helped to determine if schools were located downwind. Once the variables were established, the relationship between air pollution concentrations, traffic density, distance, and percentage of time downwind was analyzed using multiple regression analysis (Janssen et al., 2001). As a result, an association was found between PM2.5 concentrations for both indoor and outdoor air with an increase in truck traffic density and also a decrease with increasing distance from the roadways. Applying the proximity approach to intraurban air pollution exposure studies can be beneficial yet restricting in several aspects. In a practical sense, Jerrett et al. (2005) provide that proximity assessment offers a simple approach for long term exposure analysis. Also, assessment of pollution dose in relation to distance decay for individuals living along major roadways may be accurately determined from proximity buffers. Furthermore, cost in terms of additional software and time as well as updated data are not essential factors when conducting this type of study. On the contrary, a number of limitations arise as the proximity method is implemented. To begin, the dynamic nature of human activity is disregarded as a variable. English et al. (1999) claim potential exposure to individuals narrowly focuses on residence, work, and school locations while other daily activity is overlooked. Clearly, all human activity is not limited to three isolated localities throughout a typical 24 hour day thus a static analysis may lead to biased results. Additionally, proximity based assessment relies heavily on traffic emissions which is a combination of various types of vehicles. For the majority, studies fail to distinguish vehicle type with respect to 9

 

emission estimates. Typically, emissions will vary depending on the number of cars and trucks that occupy the roadways (Kanaroglou et al., 2000; Gertler, 2003) hence estimation of pollution exposure may be misclassified. Finally, dispersion characteristics of air pollution may be influenced by topographical and meteorological factors. From a proximity based assessment view point, these external features may infringe upon isotropic dispersion assumptions (Jerrett et al., 2005). Consequently, the idea that air borne particles experience the same dispersion pattern in all directions surrounding the study area would be in question in terms of validity. In a realistic sense, it is difficult to assume that wind speed, wind direction, weather patterns, atmospheric stability, and land patterns have no effect on air pollution dispersal. Thus, proximity based assessment tends to oversimplify the analysis of intraurban air pollution exposure which can be applicable to static human activity and near road distance decay pollution dose over an extended period of time. However, reality encompasses a more complex nature where human activity changes over time and external factors influence the dispersal of intraurban air pollution, thus a more sophisticated method may be utilized for the analysis of air pollution exposure. Another approach to establishing a continuous surface of pollution concentrations that is often used in exposure analysis is interpolation. The application of interpolation methods require deterministic and stochastic geostatistical techniques (Jerrett et al., 2005) which in turn provides estimated values at unsampled locations across a specified study area. The critical component with interpolation models in terms of air pollution analysis is a dense network of monitoring stations with accessible data that is distributed throughout the target study area. 10

 

A number of interpolation techniques are offered as a spatial analysis tool. Common procedures include inverse distance weighting, splines, Theissen triangulation, and kriging. The latter of these methods is the most commonly used technique in air pollution studies (Jerrett et al. 2001a). An advantage of kriging interpolation over the other methods is the establishment of both predicted values at unsampled locations along with their standard errors (Jerrett et al. 2005). Having access to error estimates indicate where predicted values may be more dependable if error is small versus unreliable predictions where error is relatively large. An example of the interpolation approach was presented in an environmental justice study in Hamilton, Canada. In this research, the authors sought to determine if populations of lower socioeconomic status were more likely to be exposed to higher levels of particulate air pollution versus populations with a higher socioeconomic standing. They also compared the sensitivity of the association between levels of particulate air pollution and socioeconomic status to exposure estimates or statistical models (Jerrett et al., 2001a). Similar to previous air pollution exposure studies, the scale of the study area was concentrated at an intraurban level where the city of Hamilton, Canada was analyzed. The air pollutant of concern was total suspended particles (TSP) and the time frame involved a ten year period. The selection of Hamilton, Canada was determined by three important factors: data, locality, and ambient air pollution issues (Jerrett et al., 2001a). To begin, Hamilton is credited for its dense network of ambient air pollution monitors that are located throughout the city. Additionally, Hamilton is situated near major industrial zones hence elevated levels of air pollution are found across significant 11

 

portions of the city. Lastly, time series studies have associated ambient air pollution problems and premature deaths within the city limits therefore raising the question of air quality and exposure to hazardous pollutants. The process of interpolation started with the collection of one year TSP averages for a ten year time frame from 23 monitoring sites throughout Hamilton. It is important to note that TSP consists of air borne particles that are less than 50 micrometers in aerodynamic diameter (Jerrett et al., 2001a) therefore PM10 and PM2.5 are both identified as TSP. Monitors gathered TSP samples on a six day cycle hence a one year average is determined by approximately 61 samples taken throughout a year. Afterwards the researchers utilized two methods of exposure assessment: chronic average exposure and probability of extreme events (Jerrett et al., 2001a). Average exposure represented a ten year average for TSP concentrations while the probability of an extreme event was determined by the ratio of total monitoring days and concentrations that read above the Ontario Ministry of the Environment (MOE) 24 hour daily objective. Universal kriging was the method of interpolation that was used for this study. This technique was preferred over ordinary kriging due to earlier evidence indicating a spatial trend in TSP distribution within this intraurban setting (Farhang, 1983; Pengelly et al, 1984). After two pollution surfaces were generated and converted into isobands, one for each method of exposure assessment, an overlay analysis was completed coupling a census tract layer with the TSP concentrations. Here, each isoband was converted into a polygon to perform a polygon-on-point overlay operation (Jerrett et al., 2001a) which allowed for a pollution exposure and socioeconomic interpretation. Finally, multiple ordinary least squares and simultaneous autoregressive models were integrated to 12

 

determine whether a relationship existed between high pollution exposure to TSP and low socioeconomic status. Acquiring a correlation between income and pollution exposure was also done at a neighborhood level using similar techniques as the previous study. For this analysis, Finkelstein et al., (2003) selected 5,228 individuals from the Hamilton-Burlington area of Ontario to target mortality rates throughout the study area in relation to higher levels of air borne particles. Once again, researchers hypothesized that mortality variability in an intraurban setting would be associated with socioeconomic status and pollution exposure (Finkelstein et al., 2003). Some techniques in this research paralleled the previous Hamilton, Canada study conducted by Jerrett et al. (2001) while other aspects differed, especially pertaining to the study population. Similarities included the use of universal kriging to interpolate monitored data along with the hypothesis that lower income neighborhoods would be subjected to higher levels of air pollution concentrations. Additionally, 23 monitoring sites were utilized to obtain TSP and sulfur dioxide concentrations. On the contrary, Finkelstein et al. (2003) went beyond a census tract analysis for the study group by implementing Hamilton and Burlington residents who have undergone pulmonary function testing during a 14 year period. From this group, the researchers sought to establish a relationship between mortality risk, income ranking, and pollutant levels. First, this was done by establishing income levels for each individual by relating his or her postal codes to a census tract which contained mean household income estimations. Next, universal kriging was applied to three years of 24 hour average TSP samples which in turn created a continuous surface of concentration levels. Finally, 13

 

proportional hazards regression models were used to compute mortality risk in relation to income and air pollution levels (Finkelstein et al., 2003). Similar to Jerrett et al. (2001a), the outcome once again confirmed that individuals living in low income neighborhoods experienced higher levels of pollutant concentrations. From this conclusion, Finkelstein et al. (2003) were also able to determine that individuals living in lower income neighborhoods were subjected to higher mortality rates. Implementing interpolation techniques to air pollution studies portrays a more sophisticated approach compared to proximity based assessments but also provides some limitations. On the upside, interpolation methods utilize real pollution measurements when computing exposure estimates (Jerrett et al., 2005). Due to distance weighting effects, interpolation methods such as inverse distance weighting, spline, and global/local polynomials (Jerrett et al., 2005) are typically avoided when creating an air pollution concentration surface. Instead, most researchers depend on kriging techniques such as ordinary and universal to better represent pollution variability across a study area and to obtain error estimates at unsampled locations. On the other hand, data availability may pose an issue when considering this method. Jerrett et al. (2005) estimate that a dense network of sampling sites typically ranging from 10 to 100 stations are required for meaningful interpolation results depending on the size of the study area. Generally, it is difficult to obtain an adequate amount of air pollution concentration readings since monitored data serves as the main source of reference. Furthermore, government monitoring stations are sparsely scattered throughout urban areas thus introducing large error to rural locations or regions where minimal observations are available (Jerrett et al., 2005). 14

 

A study conducted in the northern Georgia region provides evidence for the limitations presented by sparse networks. This research utilizes a high ozone episode to address some of the draw backs of weighted and interpolated data from monitoring networks. Rather than depending solely on monitoring networks, Bell (2006) argues that air quality modeling can accommodate for several restrictions that are identified throughout the report. To support her argument, she estimated individual and population based ozone exposures using approaches such as area weighted average of modeled estimates, nearest monitor, and inverse distance weighting and kriging spatial interpolation techniques (Bell 2006). The ozone data obtained from eight monitoring sites scattered throughout the study area was recorded as hourly estimates for 2,268 four by four kilometer squared grid cells which covered the northern portion of Georgia. Bell’s first critique addresses the misclassification of concentration levels from ambient monitoring sites at an individual level of exposure. The researcher points out that monitored data assumes a uniform exposure for a particular period of time over a given area which in turn fails to account for personal activity patterns (i.e. time spent at home, work, and school) and time spent indoors vs. outdoors (Bell 2006). As a result, predicted concentration levels generally tend to underestimate the true association between individuals and exposure. The next shortcoming pertains to the location of ambient monitors. Typically, regulated air monitoring networks are situated in heavily populated areas that generally experience higher levels of exposure due to proximity to main roadways or industrialized areas. As a result, most measurements do not accurately represent individual, community level, or rural exposures (Bell 2006) therefore misclassified results may become an issue. 15

 

Lastly, the temporal coverage of monitored data is usually measured on six day intervals thus a lack in observations may lead to crude estimations for days when concentrations were not sampled. Clearly interpolation methods go a step beyond the simplistic structure of proximity based models but both techniques still lack in several areas when analyzing air pollution effects on human health. As mentioned before, topographic and meteorological parameters should not be overlooked as a contributing factor to the spatial and temporal characteristics of air pollution variability. Additionally, the dynamic nature of human activity fails to be accounted for in most health assessment studies involving these two methods. A third technique that is utilized in health assessment analysis pertains to land use regression models. This air pollution exposure model adopts similar features from proximity assessment and interpolation methods but also introduces a different approach to estimating pollution concentration levels. The two main factors that set the framework for land use regression is the utilization of surrounding land use attributes and traffic characteristics. This model predicts pollutant concentrations for a particular area by implementing three variables: a measured pollution concentration value (y), a central location (s), and the surrounding land use types (x) within a specified buffer zone (Jerrett et al., 2005). After implementing least-squares regression modeling, a pollution surface can be created based on the monitored data at location s. Similar to the previous two methods, land use regression analysis requires measured data for the target pollutant and a proximity measurement to a given location. 16

 

Once again, the availability of a dense monitoring network plays a crucial role in the validity of land use regression pollution estimations hence an increase in monitors leads to a decrease in prediction error. In addition to these variables, traffic count data is incorporated into the analysis to establish a link between traffic volume and elevated levels of a particular pollutant. A Small Area Variations in Air Quality and Health study (SAVIAH) conducted in four European cities established a land use regression model that described spatial variations in nitrogen dioxide concentrations associated with traffic related emissions. From the combination of monitored pollution data and exogenous information (Briggs et al., 1997), predictive models of pollution surfaces were created using least squares regression techniques. An analysis of geographical variations in nitrogen dioxide pollution was performed over relatively short distances to obtain a better understanding of intraurban exposure characteristics. The spatial and temporal attributes associated with the monitoring network in this study illustrated significant variation compared to previous studies. Rather than obtaining averaged yearly data, samples were collected during four two-week periods due to the idea that spatial patterns of urban pollution remain relatively stable from year to year (Briggs et al., 1997). Furthermore, access to an extremely dense monitoring network allowed the researchers to collect data from 80 sites in each of the four cities. Also, eight to ten reference sites at each location collected continuous data on a monthly cycle throughout the entire study. These samples were used to establish a mean annual pollution concentration and to serve as validation for the predicted values.

17

 

The development of the land use regression model called for four sets of data. First, road traffic characteristics describing the road network, road type, and traffic volume were collected for each location. Next, land cover and land use data was used to characterize commercial and industrial locations. Furthermore, the altitude of each monitor was recorded to determine the topographic layout of the area and lastly the monitored nitrogen dioxide concentrations were applied. Due to local variations in the four parameters previously mentioned, a land use regression model was created for each city based on two restrictions (Briggs et al., 1997). First, each model must consider traffic volume, land cover, and topography. Also, the buffering distance must remain constant at 300 meters for each monitor in all four locations. Finally, the application of ARC/INFO could be utilized to make computations for each parameter. The computations included in the land use regression contained weighted traffic volume and land cover factors within 300 meters of each monitor site. To begin, daytime traffic volumes were estimated and then compared against nitrogen dioxide concentrations using multiple regression analysis. Next, land cover areas for high density housing and industrial regions were compared to the traffic volume residuals using multiple regression analysis. Subsequently, another multiple regression analysis was used which combined the previously calculated traffic and land use factors along with the altitude of the sampler height. Finally, the combination of these factors was calculated against the measured average nitrogen dioxide concentrations; the final regression equation was used to create a complete air pollution coverage for the corresponding study

18

 

area (Briggs et al., 1997). Predicted results were validated using the measured concentrations from the reference sites. Conclusions made from this study acknowledge the limitations and advantages to using land use regression models. On the upside, Briggs et al. (1997) have successfully demonstrated that traffic related air pollution concentrations vary over small distances. These findings help to understand the true spatial variation of air pollutants. Contrarily, these results reveal that data derived from a single monitoring location can only represent a small area surrounding a particular site. Therefore reliance on a sparse monitoring network will result in misclassification of exposure levels with a high degree of uncertainty and error. The final approach to modeling spatial variations of intraurban air pollution concentrations is the utilization of dispersion models. Dispersion modeling makes assumptions about deterministic processes through the application of Gaussian plume equations (Bellander et al., 2001). The core components that formulate these processes include data on emissions, meteorological conditions, and topography (Jerrett et al., 2005). In turn, the combination of these factors will estimate spatial variations of air pollution concentrations across a specified region. To begin, modeling criteria relies on four input parameters: pollution concentrations, meteorological and topographical data, and emissions inventory (Jerrett et al., 2005). First, pollution concentrations, also referred to as background concentrations, are usually obtained from nearby government monitoring stations and serve as a model calibrator (Clench-Aas et al., 1999b). Additionally, Gualtieri and Tartaglia (1998) specify that wind speed, wind direction, outside surface temperature, solar radiation, and 19

 

atmospheric stability comprise the meteorological component. Furthermore, topographical information relating to the elevation of the study area is essential along with emission data that can be subdivided into two main categories: stationary and mobile. The process of classifying emission sources as stationary or mobile will determine the type of input parameters that are required for the dispersion model. First, stationary sources are categorized by several release variables and are identified as residential homes and localized industries. Typically, release variables include annual mass emissions, stack height, stack diameter, temperature, and vertical emission velocities (Jerrett et al., 2005). On the other hand, mobile sources are composed of various transportation units such as cars and trucks whose parameters are characterized by traffic counts. Compared to stationary emissions data, mobile source variables are determined by vehicle type, vehicle speed, and road network gradients (Jerrett et al., 2005). After establishing the parameter criteria previously mentioned, the dispersion model calculates estimated air pollution concentrations for receptors that are located throughout a particular study area. The placement of receptors is determined using a grid with specified cell sizes which covers the entire study region. Each receptor is situated in the center of each cell with a corresponding estimated concentration value. In turn, the cell in which the receptor is located will also adopt the same concentration value. Clench-Aas et al. (1999a) utilizes a dispersion modeling approach by introducing an integrated air quality monitoring method which combines monitored pollution concentrations with data extracted from a dispersion model. Using a Geographical Information System, the authors suggest that a combination of dispersion model results 20

 

and spatial distributions of population, households, and workplaces can be used to analyze the spatial and temporal variations of air pollution. In general, the principle purpose of this study was to introduce the different ways in which dispersion modeling can be coupled with spatial data to understand how individuals, communities, cities, and regions are exposed to several pollutants. First, three different levels concerning the extent of the study area are described in terms of scale. Studies that focus on street segments, areas surrounding chimneys, or street canyons would be classified as a local scale (Clench-Aas et al., 1999a). Here, dispersion models can be used to examine air pollution concentrations within a residential area, neighborhood, or community. Typically, emission parameters would be determined by small road segments, individual households, nearby local restaurants, and laundromats. Next, urban scale is characteristic of city centers and surrounding rural areas. Cleary, this scale is characteristic of intraurban studies which is the focus of this research. In terms of grid resolution for receptor placement, a one kilometer by one kilometer cell size is commonly used for urban scale studies while two kilometers squared is characteristic of rural areas (Clench-Aas et al., 1999a). Lastly, a regional scale study can be described as a combination of two or more urban areas. Following scale classification, another criterion that is crucial to dispersion modeling is the type and time frame of emissions inventory. Clench-Aas et al. (1999a) divide the inventory into three categories: traffic, home heating, and industry. Traffic emission factors are described in terms of vehicle type, various speeds, and road gradients. Additionally, home heating is characteristic of the quantity and quality of 21

 

heating units while industrial parameters are determined by stack emissions which can be obtained through industry surveys (Clench-Aas et al., 1999a). Once all parameters are fulfilled, dispersion calculations are performed to represent hourly variations in pollution concentrations. Generally, short term analysis is a common approach due to the nature of the emissions inventory. Often times, emissions data is collected on an hourly or daily (24 hour average) basis thus models are used to reflect shorter time intervals (Clench-Aas et al., 1999a). This in turn allows for individual short term exposure analysis. However, long term exposure can also be estimated when using yearly averages for emissions data. The last procedure for this study is to determine which portion of the population is exposed to elevated levels of air pollution. Here, pollution exposure can be analyzed in terms of population as a whole or at an individual level. The authors suggest that population exposure can be determined using two methods. First, the total number of people living in each square kilometer can be examined in relation to varying hourly concentrations for that cell (Slørdal 1998). On the other hand, point estimations can be made for all homes and buildings within a given area (Clench-Aas et al., 1999a). For individual assessment, long and short term exposure can be determined using different approaches. When concerned with long term exposure, dispersion models can be used to estimate concentrations surrounding the home, work place, or school for a specific individual based on yearly, monthly, or seasonal averages (Bartonova et al., 1999). Conversely, short term exposure can be analyzed using a diary method (Duan 1982). Diary entries provide information regarding location and time of targeted individuals which in turn can be associated to the spatial and temporal variations of air 22

 

pollutants that are calculated by the dispersion models (Bartonova et al., 1999 and Clench-Aas et al., 1999b). A traffic exposure study in Oslo, Norway provides an example of individual assessment related to the diary method. In this research, dispersal patterns of particulate matter and nitrogen oxides related to traffic and spatial heating sources were examined using the dispersion model method. Here, estimations were provided for a one kilometer resolution grid on an hourly basis which reflected air pollution patterns for an urban region (Bartonova et al., 1999). In addition to activity diaries, a cross sectional study was performed to provide exposure characteristics in residential areas where the individuals spend the majority of their time (Bartonova et al., 1999). The extent of this area was determined by the individuals that participated in this research. First, 1,100 people were selected to represent the residential six kilometer squared region located in central Oslo. Additionally, a sub group of these individuals provided activity diaries that contained a detailed description of their daily movement. Participant activity was recorded by the hour for a two to three week period (Bartonova et al., 1999). Furthermore, receptor locations were determined by the participant’s households along with the various places that were visited throughout the study period. Following the establishment of receptor locations, calculations were made using EPISODE dispersion model for line, point, and area sources. Estimated traffic counts and density for all road links situated within the two by three kilometer study area served as the input line sources for air pollution. Additionally, main road segments outside of the study area were treated as line sources while smaller roads were treated as area 23

 

sources. Lastly, ship and air traffic along with spatial heating units were classified as area and point sources (Bartonova et al., 1999). All emission inventory data for NOx, PM2.5, and PM10 and meteorological data were obtained for hourly increments for a three year study period. Finally, exposure levels could be determined for the cross sectional area and the individuals who provided activity dairies. Within the study area, concentration estimations for each receptor location were provided for each hour during a three month period. Furthermore, these calculations were used to represent annual average concentrations for the three year time frame (Bartonova et al., 1999). On the other hand, individual exposure levels were provided for each location that was visited on an hourly basis. Here, receptor values at each location were assigned to the individual when they were situated within the study area while grid values were assigned outside of the study area (Bartonova et al., 1999). If multiple locations were visited during a one hour time frame, an average was taken of the outdoor concentrations at each stop. Air pollution exposure studies that utilize dispersion models illustrate the advantages of integrating both spatial and temporal disparities of pollutant levels across a given region (Bartonova et al., 1999; Clench-Aas et al., 1999a). To begin, this method overcomes the need for extensive monitoring networks that are common with previously mentioned techniques such as proximity assessment, interpolation, and land use regression. Furthermore, external parameters which diffuse air pollutants such as wind, atmospheric stability, topography, and traffic flows are accounted for in dispersion frameworks (Jerrett et al., 2005). 24

 

Moreover, dispersion models can be applied to various scales regarding the extent of the study area. Examples include urban areas that are characteristic of short term air pollution episodes along with regional areas which are used to address pollution transfer over space and time (Jerrett et al., 2005). Therefore, high resolution air pollution variability can be illustrated and analyzed using dispersion models. However, various drawbacks can be associated with this approach as well. In general, data input can be timely depending on the quantity of point, area, and line sources that are entered into the model. As previously mentioned, multiple parameters must be entered for each individual source hence a substantial amount of time must be devoted to data preparation. Additionally, Gaussian assumptions about dispersion patterns are often times unrealistic thus outcome values may be misrepresented (Jerrett et al., 2005). Lastly, estimate error may be presented when emissions and meteorological data contain temporal mismatches (Jerrett et al. 2005). In other terms, the sample intervals for both input values may be expressed in different scales. This is presented when meteorological data is collected by the hour while emissions data for point and area sources is represented as an annual rate (Jerrett et al., 2005). As a result, exposure estimations have the potential to contain substantial error.

25

 

Chapter 3: Data The two main sources of data that support this research are twenty four hour particulate matter concentrations and travel activity data for Franklin County residents. Initially, particulate matter concentration levels were to be obtained from local monitoring sites that are located throughout Franklin County. Figure 1 (below) displays the three monitoring sites in Franklin County for the year 1999.

Figure 1. Franklin County monitors that sample PM10 concentration levels (1999) 26

 

Monitor one is a continuous monitor that is located in an urban setting. This monitor collects one sample of particulate matter at the beginning of every hour for every day of the year starting at hour zero (twelve o’clock A.M.) and ending at hour twenty three (eleven o’clock P.M). Conversely, monitors two and three are located in urban (monitor three) and suburban (monitor two) settings. They collect one sample of particulate matter every six days at hour zero which clearly does not provide the appropriate temporal resolution for this study. Furthermore, due to the lack in spatial coverage of the monitoring network, an alternative method of estimating particulate matter concentrations had to be acquired. As an alternative approach, the process of utilizing an air dispersion modeling system served as the primary method for estimating concentration levels. In simple terms, an air dispersion model is a computer program that requires the input of various parameters in order to calculate both spatial and temporal measures of a specific air pollutant. A more in depth description of this software and its functionality will be discussed in the methodology chapter. In terms of data input, air dispersion models require three main sources of information: emissions data, meteorological data, and topographic data. First, emissions data must be obtained for the type of source (point, line, and/or area) the user wishes to analyze. Due to data availability, this study solely focuses on stationary, continuous point sources (i.e. industrial processes). The emissions data for Franklin County facilities (point sources) was obtained from the Ohio Environmental Protection Agency’s 1999 emissions inventory. The input parameters for point source emissions include: latitude, longitude, and base elevation of 27

 

each source (UTM coordinates), stack height (meters), stack temperature (degrees Kelvin), exit diameter of the stack (meters), exit velocity, exit flow rate (actual feet per minute), and the total amount of pollution emitted (tons per year). Figure 2 displays the 119 major point sources in Franklin County that were included in this study. Several industrial point sources are located within close proximity of one another therefore many points are overlapping at this scale.

Figure 2. Franklin County 119 major industrial point sources (1999)

28

 

Next, meteorological data is utilized to determine the dispersal patterns of the pollution. As a requirement, both surface and upper air data must be obtained for the region of interest. For this research, surface data was collected from the Columbus International Airport on an hourly basis for the year 1990 (1999 data was unavailable in the required format). Additionally, upper air data was obtained from the Dayton Wright Patterson Air Force Base (1990). From both sets of data, wind direction, wind speed, temperature and temperature differences, humidity, precipitation, pressure, radiation, (etc.) are used to determine atmospheric turbulence characteristics, mixing heights, friction velocity, surface heat flux, (etc.) which in turn influences the overall dispersal patterns of the emitted pollution. The third type of data input refers to the topographic features of the study area. Here, USGS 7.5 minute digital elevation model (DEM) quads are entered into the air dispersion model and are used to calculate the base heights of all point sources and receptor locations, which will be discussed in the following section. It is important to include all digital elevation models that will cover the entire extent of the study area and receptor locations since all elevations must be calculated before running the model. Typically, the elevation of the study area will have a stronger influence on the dispersal patterns if mountainous or hilly regions are present. However, Franklin County is relatively flat therefore topographic influences may not be as prevalent in this study. Lastly, a critical component that must be taken into consideration when using air dispersion techniques is background concentrations. This data is obtained from local monitoring sites that are situated within the study area. More importantly, it accounts for all other pollution sources that are not entered into the dispersion model. For this study, 29

 

other sources would include: major industrial sources located outside of Franklin County and/or Ohio, minor sources that do not have emission inventories (restaurants, auto body shops, gas stations, and dry cleaners), mobile sources, fugitive emissions (land fills, storage piles), dust from unpaved roadways, and environmental tobacco smoke. These concentration values are added onto the results of the estimated concentrations from the air dispersion model and will be obtained from the Franklin County monitors that were previously introduced. The second set of data that is essential to this study is travel activity data that reflects the daily travel patterns of Franklin County residents. In 1999, the Mid-Ohio Regional Planning Commission (MORPC) collaborated with the Licking County Area Transportation Study (LCATS) to conduct the 1999 Mid-Ohio Area Household Travel Survey which was funded by the US Department of Transportation and the Ohio Department of Transportation (1999 Mid-Ohio Area Household Travel Survey Final Report). The study area for this research included Franklin, Licking, and Delaware counties along with portions of Fairfield, Union, Madison, and Pickaway counties. Additionally, data was collected over a five month period from February to June with a total of 5,555 participating households located throughout the selected counties. Overall, 13,524 persons, 10,488 vehicles, and 52,031 trips were represented by these households. The process of data collection involved the completion of a travel diary which reflected the individuals’ daily travel activities. During the recruitment procedure, each household was assigned a travel day in which all individuals residing in that specific household were responsible for tracking their travel activities over a twenty four hour period (1999 Mid-Ohio Area Household Travel Survey Final Report). Following 30

 

completion and retrieval of all travel diaries, a geocoding process was applied to each travel entry including household, work, school, and non-work/school locations in order to accurately display a visual representation of all travel activity. In total, 100% of household locations, 95% of work and school locations, and 92% of non-work/school locations were successfully geocoded. Within the realms of this air pollution research, the data selected from the 1999 Mid-Ohio Area Travel Survey solely focused on households that resided in Franklin County. Furthermore, any travel outside of the Franklin County borders, regardless of household location, was extracted from the dataset as well. Hence any individual who lived in Franklin County but traveled to neighboring counties during their twenty four hour recording period was excluded from the study. This process ensured that all activity would reside within the proposed study area. In terms of descriptive travel and household information, four attribute tables were established and coded with quantitative variables which in turn illustrated the daily activities of each individual involved. In general, the four tables disclosed household, trip, personal, and vehicle information. With regards to this research, the household, trip, and personal attribute tables maintained the sole focus while the vehicle table was disregarded. A combination of household, trip, and personal datasets allows for a more in depth description of the sample population’s daily activity. Amongst several attributes contained within the household dataset, the most relevant to this study include: household ID, travel day, travel month, travel day of week, county ID in which the household resides, household latitude and longitude, and income. Additionally, the trip 31

 

dataset comprises: household ID, person ID, descriptive location of each trip, county ID in which the trip took place, latitude and longitude of trip location, arrival hour of trip, arrival minute of trip, departure hour of trip, departure minute of trip, and trip duration in total minutes. Finally, within the personal dataset the key attributes include: household ID, person ID, household county ID, relation to individual who recorded the travel diary, gender, age, and ethnicity of each individual, education level, and total number of trips recorded for each individual within the twenty four hour travel period.

32

 

Chapter 4: Methodology As an overview, the methodological procedures that support this research involve the utilization of an air dispersion modeling software package and ArcGIS Desktop version 9.2. Through the combination of these applications, an investigation of particulate matter exposure to Franklin County residents can be carried out at a high spatial and temporal resolution. In terms of procedural components, the supporting methodology can be divided into three processes: air pollution dispersion modeling for concentration estimations, extrapolation of monitored concentration readings, and the utilization of geographic information system technologies for spatial and temporal evaluation of the model outputs accompanied with travel activity data of Franklin County residents. 4.1

Gaussian Plume Model As previously discussed, air dispersion models make the assumption that steady

state conditions exist with regard to air pollutant emissions and meteorological changes through the application of Gaussian plume equations. From a Gaussian model, air pollution is represented as a plume coming from the top of a stack (point source) which is characterized by a height and diameter. The vertical displacement of the plume is determined from the stack dimensions along with stack gas exit velocity and temperature. After the plume has reached the stack height, the pollution is dispersed in three dimensions. Figure 3 depicts this process and also illustrates the assumed Gaussian 33

 

distribution in the two dimensional cross wind and vertical direction (Image source: http://www.colorado.edu/geography/babs/empact%20draft%20web%2018apr%2004/abo ut%20aermod.htm).

Figure 3. Gaussian plume dispersion from point source

From Figure 3, the dispersal process in the downwind direction is a function of the average wind speed while dispersion in the cross wind and vertical direction are determined by the Gaussian plume equations. Here it is shown how the two dimensional dispersion takes the form of a normal Gaussian curve with the maximum concentration around the center of the plume. Furthermore, the lateral dispersion about the cross wind

34

 

and vertical axis is governed by the stability of the atmosphere which is provided by the meteorological input data. Several air dispersion models incorporate Gaussian equations including: ADMS (Advanced Dispersion Modeling System (UK)), AERMOD (AMS/EPA), CALPUFF (non steady state, Gaussian puff model), BLP (Buoyant Line and Point), and CALINE3 (linear model). Due to software availability, the type of air dispersion modeling software package that is used in this study is BEEST Suite for Windows version 9.00. This software package contains a user interface with various types of air dispersion models including ISCST, AERMOD, ISC-PRIME, and SCREEN3. As a result of data limitations, the Industrial Source Complex Short Term 3 (ISCST 3) model was used to estimate particulate matter concentrations from point sources. 4.2

ISCST 3 Dispersion Model The User’s Guide for the Industrial Source Complex (ISC3) Dispersion Models

Volume 1 was referenced for a thorough investigation of the functionalities of dispersion models. To begin, the guide describes the basic input data requirements: an input runstream file and meteorological data files. The runstream setup file is divided into six functional pathways which are described by a two character pathway ID. These pathways include the overall job control options (CO), source information (SO), receptor information (RE), meteorological information (ME), terrain grid information (TG), and output options (OU). The terrain grid information is an optional pathway that is only used when implementing the dry depletion algorithm which was not practiced in this study.

35

 

4.2.1 Control Pathway The control pathway (CO) allows the user to manage the modeling options which consist of dispersion options, averaging time options, and terrain height options. The dispersion option that was selected for this study was the regulatory default options which are specified in the Guideline of Air Quality Models and also serve as the default options for ISC models. A more thorough explanation of these options is identified in Appendix A of the Guideline on Air Quality Models (Revised) (EPA, 1987b). To provide a brief description of these default settings, the model incorporates the use of stack-tip downwash, the use of buoyancy induced dispersion, the exclusion of gradual plume rise, the use of calms processing routines, the use of upper bound concentration estimates for sources influenced by building downwash from super squat buildings, the use of default wind speed profile exponents, and the use of default vertical potential temperature gradients. Additional dispersion options that are specified within the control pathway include rural versus urban dispersion parameters and the output type which consists of concentration or deposition. The user selects rural or urban dispersion parameters based on the characteristics of the source location. For this study, the rural option was selected given the locations of the majority of the point sources that were analyzed. Furthermore, concentration (μg/m3) was chosen as the model output. In addition to the dispersion model option, the averaging time is also determined within the control pathway. The averaging time refers to the time frame in which the model performs calculations. Examples include short term periods such as one or twenty four hour averages and long term periods such as monthly or annual averages. One hour 36

 

averaging was performed due to the high temporal resolution of this study. Also defined is the pollutant type. The user specifies a pollutant ID however this identification does not have impact on the dispersion results unless SO2 is the defined pollutant. The final option that is defined under the control pathway is the elevated terrain information. If specific height information is desired for receptor locations, the user must supply terrain data to avoid a model assumption of flat land. This holds significant accountability when an assessment is being performed in complex terrain where topographical influences are relevant. A default terrain height of zero meters is applied to receptor locations if no values are provided. The following image (figure 3) provides an overview of the control option settings displayed on the output file.

37

 

Figure 4. Dispersion Model Configuration Summary

4.2.2 Source Pathway The next pathway that requires user specifications are the source options (SO). Here, the user defines the source information for a particular model run. Sources options for ISC models include point, volume, area, or open pit. For all sources, the user provides a source ID (not relevant for calculations), source type, and source coordinates in the form of x, y, and z. For point and volume sources, the x, y coordinate coincides 38

 

with the center of the source and for area and open pit sources the coordinates correspond to the southwest corner of the source. Additionally, the z coordinate signifies the source elevation. Along with source identification, the user must designate the source release parameters which vary according to source. This study solely focuses on point sources therefore a more in depth analysis of the point release parameters will remain the primary concern. Generally speaking, modeling of point source emissions refers to industrial stacks and isolated vents. In comparison, volume sources include building roof monitors and multiple vents, area sources consist of low level or ground level releases with no plume rise (i.e. storage piles, slag dumps) and open pit sources are associated with surface coal mines and rock quarries. In terms of point release parameters, the following data requirements include: point emission rate in grams per second (g/s), release height above ground in meters, stack gas exit temperature in degrees Kelvin (K), stack gas exit velocity in meters per second (m/s), and the stack inside diameter in meters. More importantly, all parameters must be present to perform model calculations. All point sources and associated attributes that were included in this study are located in Appendix A. 4.2.3 Receptor Pathway The next essential step in the dispersion model set up is defining the receptor pathway (RE). To briefly describe the functionality of receptors, once all required input parameters are entered into the model (source release parameters, meteorological, and terrain data), the model will estimate concentration values from the multiple sources at the predetermined receptor locations. The user is able to define a Cartesian grid receptor 39

 

network and/or polar grid receptor network with either uniform or non-uniform grid spacing. The grid density may vary across the area of interest depending on the user’s primary focus for the analysis. For example, a dense grid may be generated for areas experiencing maximum impact from the input sources while a coarse grid may be more appropriate for exterior regions that undergo less impact. For this study, a 38 by 40 grid with one kilometer of separation between each receptor was created thus 1,520 receptor locations were used to cover the extent of Franklin County. A one by one kilometer resolution was implemented based on previous intraurban studies which suggest that this is the appropriate scale for an analysis of urban air pollution. Figure 4 (below) displays the receptor locations.

40

 

 

Figure 5. Receptor Locations with 1 kilometer separation

4.2.4 Meteorological Pathway ISCST models can utilize various file formats to satisfy the input meteorological data requirements for the ME pathway. Essential inputs include surface meteorological data and upper air data. Both levels are required to calculate mixing heights which is described as the height of the atmosphere above ground level that is well mixed due to mechanical or convective turbulence. Atmospheric mixing characteristics greatly influence vertical and horizontal dispersal patterns thus playing a significant role in the model outputs. 41

 

Standard formats include unformatted sequential files of meteorological information generated by PCRAMMET and MPRM preprocessors. In general, MPRM (Meteorological Processor for Regulatory Models) processes National Weather Service and on site data where it first lists missing, suspect, and invalid data, merges quality assured and corrected meteorological data, and then creates meteorological data files for input to air quality dispersion models. Similarly, PCRAMMET is a processor which combines hourly National Weather Service surface and twice daily mixing heights into a single file, computes a mixing height for each hour and then incorporates surface characteristic parameters (i.e. Monin-Obukhov length, surface roughness length, and Bowen ratio) for deposition modeling. Other valid inputs include formatted ASCII files that contain sequential hourly records of meteorological variables. Typically, the first record of the input file contains the station number and year for both the surface station and the upper air station. Other essential variables include: year, month, day, flow vector (direction), wind speed (m/s), ambient temperature, stability class, rural mixing height, urban mixing height, wind profile exponent, vertical potential temperature gradient, friction velocity (m/s) (dry/wet deposition only), Monin-Obukhov length (m) (dry/wet deposition only), surface roughness length (m) (dry/wet deposition only), precipitation code (wet deposition only), precipitation rate (mm/hr) (wet deposition only). Additional, required information is the height of the anemometer which represents the height above ground at which the wind speed data was collected. The model will use this information to adjust the input wind speeds from the anemometer height to the

42

 

release height of the source. Understanding wind characteristics at the release height plays an essential role in obtaining valuable model results. 4.2.5 Output Pathway The final parameters that the user must specify are the output options. Critical information that is acquired from the output pathway includes summaries of high values by receptor for each averaging period, summaries of overall maximum values for each averaging period, and tables of concurrent values summarized by receptor for each averaging period. Lastly, the output file contains all coordinate and concentration values for the receptor locations which allows for importation into graphic plotting packages to generate concentration plots. 4.3

Model Output and Interpolation For this study, concentration estimations were made every hour at every receptor

location for the following dates: March 2nd, 4th, 5th; April 5th, 7th, 9th; May 4th, 6th, 7th (1999). In total, eight days were analyzed on a twenty-four hour basis therefore the final dataset contained concentration estimations for 1,520 receptor locations for a total of 192 hours (May 4th was removed from the analysis due to no recorded travel activity). The extent of this time period will coincide with the travel activity dataset provided by the 1999 Mid-Ohio Area Household Travel Survey Final Report which in turn will allow for an analysis of twenty-four hour particulate matter exposure for all individuals involved in the study. In total, ISCST 3 generated three contour maps displaying concentration values (μg/m3) for each month that was analyzed. The following figures (5 – 7) display the highest 72 hour particulate matter concentration estimation at each receptor location. 43

 

Figure 6. Highest 72 Hour Concentration Estimation for March 2nd, 4th, 5th

44

 

Figure 7. Highest 72 Hour Concentration Estimation for April 5th, 7th, 9th

45

 

Figure 8. Highest 72 Hour Concentration Estimation for May 4th, 6th, 7th

Here, it is evident that certain hot spots with elevated levels of particulate matter exist across Franklin County thus ensuring that certain individuals may be subjected to higher concentrations depending on his or her location throughout the day. Also, by monthly comparison, it is clear that time of year will influence the dispersal patterns of the industrial emissions. This may be contributed to typical weather patterns that are characteristic of the changing seasons. After obtaining results from the air dispersion model, the next step is to import the hourly concentration values for all receptor locations into ArcGIS version 9.2 to perform 46

 

interpolation on the receptor values. From here, a continuous surface of particulate matter hourly concentrations is created for the entire study area at a high resolution. For this study, the kriging method was selected as the primary form of interpolation. As an end result, 192 particulate matter concentration continuous surface raster datasets are created which in turn will be utilized for individual exposure assessment for Franklin County residents. The eight day dispersion model output is provided in Figure 8.

Figure 9. Potential Areas of High PM10 Concentrations (μg/m3)

47

 

4.4

Background Concentration A critical and influential exposure factor that is not taken into consideration

within the dispersion model is the background concentration of the study area. The background concentration is a combination of all pollution sources and is obtained from on site monitors. Sources include industrial processes, transportation, fuel combustion, indoor sources, environmental tobacco smoke, wood burning, cooking, and solid waste. The contribution from each source varies by region and density (i.e. road networks, industrial plants). For this study, the process of incorporating background concentrations consisted of three main components. First, obtain hourly concentration readings from the three available monitors. Next determine the dispersion model output at each monitor for every hour then subtract that value from the monitored value. The result will represent the background concentration from all contributing sources with the exception of industrial sources. Finally, perform a nearest neighbor function from the receptors to the monitor locations. From here, a background concentration can be assigned to the individual receptor locations and eventually interpolated to create a smooth grid covering the study area. As previously discussed, there are three monitors located within Franklin County that collect particulate matter concentration readings. From these three monitors, only one collects data at an hourly resolution (monitor one). The other two monitors take samples every six days at hour zero (monitors two and three). To accommodate for the lack in temporal resolution for monitors two and three, a regression analysis was performed between monitors one and two and monitors one and three. 48

 

To begin, all data at hour zero for the entire year was extracted from the dataset for monitors one, two, and three. Hour zero was selected based on the sampling process for monitors two and three. Next, a comparison of time stamps was performed so that all hour zero data from monitor one matched the sampling date for the zero hour readings at monitors two and three. Therefore the final data set was based on the sampling times of the six day monitors. The following table summarizes the data recovery for each monitor.

Data Recovery January 1 – December 31, 1999

Monitor 1 (Hourly)

Monitor 2 (6 day, 0 hour sampling)

Monitor 3 (6 day, 0 hour sampling)

59/60

52/60

50/60

Table 1. Monitors 1, 2, & 3 Data Recovery

Following the data collection process, a linear regression was completed first between monitors one and two, then between monitors one and three. The hourly concentration values from monitor one served as the independent variable while the samples obtained from monitors two and three served as dependent variables for each respective regression analysis. Raw data values from all three monitors that were used in the regression can be referenced in Appendix A. The following figures (9 – 10) illustrate the linear regression results and residuals.

49

 

Monitor 2 (6 Day) (ug/m3)

PM Concentration Hour 0, January - December 1999 80 60 40 20 0 0

10

20

y = 0.5145x + 11.328 R2 = 0.3581

30

40

50

60

70

Monitor 1 (Hourly) (ug/m3)

 

Figure 10. Monitor 1 and 2 Linear Regression

Monitor 1 (Hourly) Residual Plot 40

Residuals

30 20 10 0 -10 0

10

20

30

40

50

60

70

-20 -30

Monitor 1 (Hourly)

Figure 11. Monitor 1 Residual Plot from Monitor 1 & 2 Linear Regression

The linear regression results and residual plot offer some insight to the statistical model that was used to analyze the strength of the relationship between the zero hour samples from monitors one and two. An R2 value of 0.358 indicates a relatively weak, positive relationship between the two variables. However, the residual plot displays a random distribution therefore a linear model is the appropriate approach for this analysis. 50

 

The necessary measure of outlier removal was carried out to improve the R2 value to 0.629. The same process was repeated between monitor one and monitor three. The following figures (11 – 12) provide linear regression and residual plot results.

Monitor 3 (6 Day) (ug/m3)

PM Concentration Hour 0, January - December 1999 120 100 80 60 40 20 0 0

10

20

y = 0.7732x + 14.184 R2 = 0.3303

30

40

50

60

70

Monitor 1 (Hourly) (ug/m3)

             

Figure 12. Monitor 1 and 3 Linear Regression

Monitor 1 (Hourly) Residual Plot 60

Residuals

40 20 0 -20

0

10

20

30

40

50

-40

Hourly

Figure 13. Monitor 1 Residual from Monitor 1 & 3 Linear Regression Plot

51

60

70

 

The results are comparable to the regression outcome between monitor one and two. An R2 value of 0.330 provides that a relatively weak yet positive relationship exists between the zero hour samples however the process of outlier removal is necessary to improve the linear relationship. The resulting R2 value following the removal of outliers improved to 0.620. After establishing a linear regression for both models with the removal of outliers, the resulting trendline equations were used to predict hourly concentration values at monitors two and three for the study days. Now, all three monitors have hourly particulate matter concentration values over a twenty four hour period for eight days. The predicted concentration values for monitors two and three can be viewed in Appendix A. Performing a measure, correlate and predict method between the three monitors introduces caveats that are worth noting. First, the relationship between zero hour concentration values at each monitor may not be consistent for various hours throughout the day. However this assumption is implied for this study due to an extreme lack of data. This is especially relevant during high traffic activity hours when urban monitors may be subjected to higher exposure compared to the suburban monitor. Hence this method fails to portray diurnal patterns that are influenced by typical daytime activity. Also, a linear relationship was established for an entire year’s worth of data which may overlook seasonal trends. For example, on average monitor one may experience lower concentration exposure than monitor three during spring months when mid latitude cyclonic activity enhances southerly to westerly wind flows. On the other hand, winter weather patterns are typically associated with northwesterly to northerly wind flows 52

 

which may increase pollution exposure at monitor one compared to monitor two or three. Once again, this method falls short of incorporating these patterns. Lastly, a positive linear relationship has been established between the three monitors yet, the R2 values suggest that other variables account for the variation at monitors two and three. This is especially true since several outliers were removed from the dataset to improve the model. Given these limitations to the model, the derived background concentrations were still incorporated into the study in order to provide more spatial variability across the study area. This leads to the next step that was performed in association with the background concentration data. Since the monitored data incorporates all pollution sources, the portion that was contributed from industrial sources must be removed from the monitored concentration values. Using the dispersion model interpolated grids, the concentration value derived from the kriging process of receptor data was determined at each monitor location and then subtracted from the monitored concentration value for every hour. In turn, this procedure provides a concentration value at each monitor that represents the pollution from all polluting sources with the exception of industrial sources. The derived concentration values for each monitor are provided in Appendix A and are categorized by twenty four hour concentration values for each travel day at each monitor. Lastly, in order to integrate the newly derived background concentration values, a nearest neighbor function was applied using the receptor locations as reference points. A simple nearest neighbor calculation was performed from each receptor location to the nearest monitor. Once the nearest monitor was determined, the hourly concentration value was applied to each receptor location. Finally, kriging was utilized to interpolate 53

 

the values across all receptor locations to in turn produce a smooth grid that represented the background concentrations across the study area. 4.5

Exposure Analysis of Dynamic vs. Static Population The next goal of this study is to utilize the hourly concentration raster datasets

(raw dispersion outputs plus background concentrations) in conjunction with the travel activity data to obtain a better understanding of which Franklin County individuals are exposed to higher levels of particulate matter concentrations. This portion of analysis can be divided into two separate entities. First, an investigation will be performed on a sample portion of the population while maintaining a static state. This implies that all individual exposure will be assessed at the household location for all members of the family over a twenty four hour period hence all travel is disregarded at this time. On the other hand, the second analysis will involve a comparison of particulate matter exposure between traveling individuals within the each household. 4.5.1 Static Exposure Taking a closer look at the first scenario, the initial step in the analysis is to obtain the household dataset that was previously described in the data chapter. Next, all household locations within the dataset must be projected in ArcGIS in order to determine the sample population. From here, the sampled individuals are obtained using a select by attribute query which will identify all household locations that are situated in Franklin County through the use of the county household ID attribute field. Furthermore, an additional attribute query must be performed which relates to the activity travel day. As previously mentioned, each household involved in the Mid-Ohio Area Household Travel Survey was assigned a travel day in which all individuals within 54

 

the household had to record his or her daily activity. Therefore, an attribute query must be completed for the fifteen days that coincide with the particulate matter concentration raster datasets. Upon gathering this information, it will be possible to associate hourly concentration values from a particular day to all individuals who partook in the travel survey. Table 2 (below) summarizes the sample population located in Franklin County for each day of analysis.

ISCST 3 model run day

# of households sampled

# of individuals sampled

Tuesday March 2, 1999

25 households

51 individuals

Thursday March 4, 1999

20 households

49 individuals

Friday March 5, 1999

41 households

103 individuals

Monday April 5, 1999

49 households

143 individuals

Wednesday April 7, 1999

55 households

146 individuals

Friday April 9, 1999

59 households

140 individuals

Thursday May 6, 1999

35 households

76 individuals

Friday May 7, 1999

30 households

61 individuals

Table 2. Sample of Households and Individuals Assessed for Static Exposure

In summary, a total of 314 households and 769 individuals will serve as the sampled population for the analysis of particulate matter exposure at household locations. Next, the process of associating twenty four hour particulate matter concentration sums to all of the individuals in the above chart must be performed. This analysis will be 55

 

completed according to the model run days thus the following methodology must be repeated eight times. To begin, the projected household locations are now displayed in ArcGIS according to the assigned travel day. Next, the corresponding twenty four continuous raster datasets with the interpolated particulate matter concentrations will be utilized to assign a value to each household location. This is simply done by extracting the cell values from the interpolated rasters (raw concentration values plus the background concentration) to all household point locations. This process must be repeated for every hour hence all households will have twenty four appended values which will represent the hourly concentrations that he or she is exposed to while at the household location. After repeating this procedure for all model run days, it is now possible to determine twenty four hour total particulate matter concentration estimation for each person in the dataset. A simple summation of all twenty four readings will provide the total exposure for a twenty four hour period. Also important to note is the absence of a model estimation. This will occur when calm winds are recorded for that particular hour. In the event of this situation, all receptor estimations from the model will be presented as a zero therefore the only recorded value for that hour will be the background concentration which is provided by the nearest monitor. The final step involved in examining household location exposures is to relate the final particulate matter sum to the attributes that describe each individual. As previously mentioned, a personal dataset containing information regarding ethnicity, age, education, and income will now be linked to the household dataset according to the person identification numbers. Now, a relationship can be established between the twenty four 56

 

hour particulate matter concentration summations and the socioeconomic attributes of each individual. 4.5.2 Dynamic Exposure The second part to this analysis involves the dynamic nature of household individuals. Rather than examining twenty four hour exposure at one location, the travel locations will be taken into account thus providing different exposure levels based on the locality of all household individuals over a twenty four hour period. Prior to examination of particulate matter exposure, a spatial join must be performed involving the household and travel datasets. As mentioned before, eight household datasets have been imported according to the assigned travel day. Now, the travel dataset attribute table that was previously described must be joined to each one of these tables. The common field that is shared by both attribute tables is the household ID number. The newly updated travel table now provides information regarding how many trips each individual recorded along with trip location and duration. All trips that do not take place on the specified travel day will be disregarded. Finally, all trip locations can now be projected onto the study area in ArcGIS according to the geocoded latitude and longitude coordinates. Similar to the previous procedure, the concentration values from the interpolated model output and background concentration rasters are extracted to the joined household and travel dataset in order to evaluate the concentration values at each point location. All travel locations that are recorded in the travel diary will be taken into consideration for analysis. Additionally, households will be eliminated from the analysis if travel takes 57

 

place outside of Franklin County or if any travel locations were unsuccessfully geocoded according to the Mid-Ohio Area Household Travel Survey. The final household and travel attribute table will once again have twenty four concentration readings appended to all locations that are traveled to during his or her assigned travel day. Table 3 reveals the number of households and individuals that were analyzed for this portion of the study.

ISCST 3 model run day

# of households sampled

# of individuals sampled

Tuesday March 2, 1999

11 households

22 individuals

Thursday March 4, 1999

10 households

20 individuals

Friday March 5, 1999

26 households

52 individuals

Monday April 5, 1999

26 households

52 individuals

Wednesday April 7, 1999

15 households

30 individuals

Friday April 9, 1999

31 households

62 individuals

Thursday May 6, 1999

12 households

24 individuals

Friday May 7, 1999

15 households

30 individuals

Table 3. Sample of Households and Individuals Assessed for Dynamic Exposure

To summarize, a total of 146 households and 292 individuals serve as the sample population for this portion of the study. This sample is significantly smaller compared to first section of analysis due to various restrictions. First, many individuals traveled outside of the study area during his or her twenty four hour travel day which is beyond the extent of the particulate matter raster datasets. Also, several locations were not 58

 

successfully geocoded thus a concentration value could not be assigned for the corresponding hours which in turn prevents an accurate summation for the twenty four hour exposure value. Additionally, some households consisted of single individual. Upon extracting the individuals from each household, the following procedure involves an in depth analysis of his and her hourly travel patterns. The fields of interest from the travel attribute table include the locations of each trip, the arrival and departure times and the duration of each trip. With this information, it is now possible to examine where the individual is located at the beginning of each hour for the entire twenty four hour period. If the individual is at a set location (i.e. work) at the start of the hour then the concentration value associated with that particular location will be applied to his or her twenty four hour summation of particulate matter exposure. On the other hand, an individual may be traveling at the start of a given hour therefore additional analysis and calculations must be performed. First, it must be determined whether the individual is stationary or traveling at the beginning of each hour. This is accomplished by analyzing the arrival and departure times of each trip. Next, the travel hour is recorded in order to obtain the corresponding hourly particulate matter raster dataset. Also, the start and destination locations in which the trip takes place must also be noted. Finally, a process must be implemented to determine approximately where the individual is located along the route which connects the start and destination locations. In a general sense, the total travel distance and the total travel time between the start and destination locations will be utilized to establish an approximate position along the traveled route. In order to accomplish this task, the network analyst extension in 59

 

ArcGIS must be utilized. As an overview, the network analyst extension which is comprised of several components permits the user to build a network dataset and then perform an analysis involving the dataset. In terms of applicable components, the processes performed in this study involve the construction of a network dataset, the creation of new routes, importation of location stops, and calculation of travel routes. To begin, the creation of a network dataset requires an accurate road network that is representative of the study area. Figure 13 illustrates the Franklin County road network that is utilized for all network analysis.

 

Figure 14. Franklin County Road Network

60

 

Upon obtaining the road network, a network dataset is created which allows the user to specify connectivity settings, model turns, cost attributes for network analysis, and driving directions. For this research, all settings will remain defaulted with the exception of the cost attribute field. Here, the length of the road network will be selected as the cost attribute. Thus an assumption is made that all traveled routes between start and destination locations will be dependent on the shortest path of the road network. Clearly this may introduce an inaccurate assumption for several trips within the travel dataset since the shortest path established by the network dataset may not be representative of the routes chosen by the traveling individuals. Examples include newly created road networks that are absent from the dataset, short cuts via back roads that are absent from the dataset, or longer routes in terms of distance that may be traveled at faster speeds due to increased speed limits (i.e. highways). Regardless, the only attribute that is taken into consideration is distance therefore all alternative routes are disregarded. Upon creating and importing the network dataset into ArcGIS, the following steps involve the establishment of stop locations and the approximate location of the traveling individual along the shortest path. First, a new route must be created from the Network Analyst dropdown menu. Next, the stop locations must be loaded into the newly created route. Prior to this import, a start and destination location must be selected from the travel dataset. Finally, the shortest route between the start and destination locations is determined by solving the network analyst. The final output from the network analyst is a polyline with attributes describing the total length of the line segment (units in feet). Figure 14 provides an example of two selected locations connected by the shortest route. 61

 

 

Figure 15. Shortest route between start and destination locations

Next, additional calculations must be implemented in order to determine an approximate location of the traveling individual with respect to the start of the hour. As previously mentioned, the total travel time between two locations is determined from the trip duration attribute field. This information accompanied with the total travel distance will allow for an estimation of the individuals’ location along the route that he or she is traveling. The proceeding steps describe how this process is carried out. First, the total travel distance is divided by the total travel time in order to determine how far the individual travels for a given unit of time. For this study, the units 62

 

are in feet per minute. Next, the total number of minutes from the turn of the hour is recorded with respect to the departure time. For example, an individual departs from a location at 7:55 A.M. and arrives at his or her destination at 8:10 A.M thus the total number of minutes from the start of the hour is five minutes. This indicates that within five minutes of departure the individual is at a given location along the traveled route at 8:00 A.M. Here, another assumption is introduced from this process which is the individual is traveling at a constant speed throughout the entire trip hence traffic stops and speed restrictions are neglected. After obtaining this information, the previously established shortest route must be divided into segments according to the distance traveled per minute. Next, activation of the attribute table will allow for the selection of the line segment that corresponds to the turn of the hour (i.e. the first line segment corresponds with the first traveled minute; the second line segment corresponds with the second traveled minute, etc.). In general, the selected line segment now provides an approximate location of the traveling individual at the start of the hour. Once the selected road segment is isolated, the following step is to establish point features at each vertex. Figure 15 (below) illustrates this process.  

63

 

 

Figure 16. Point vertices along selected line segment

Now, the particulate matter raster dataset with the corresponding hourly concentration values can be extracted to the point features along the line segment. The final step is to determine an average particulate matter concentration value for the line segment using the extrapolated values from the point features. This is done by examining the attribute table which now contains a particulate matter concentration value for each point. Finally, the particulate matter concentration value is determined by taking the average of the end point values thus an hourly exposure value is now established for the traveling individual. 64

 

4.6

Dynamic vs. Static Statistical Analysis The first statistical analysis that was performed involved a paired t-test for

dynamic vs. static exposure for all 292 individuals. The main purpose of this analysis is to determine if a significant difference is evident in average particulate matter exposure when an individual is static (located on home all day) versus dynamic (moving from one location to the next). Additionally, this analysis will only involve exposure during the daytime hours of nine o’clock a.m. and five o’clock p.m. which is representative of a standard work day. A twenty four hour analysis of dynamic and static exposure would not generate desirable results since early morning and late night hours are generally spent at home. Next, paired t-tests were performed for all individuals by subgroup. The data was divided into gender, income, age, and education subgroup. Age was categorized by young (19 – 40), middle (41 – 61), and old (62 – 83) age groups. The age categories were determined by identifying the range from the youngest to oldest individuals within the sampled dataset. Income was defined by nine categorical values within the time activity dataset therefore three groups were characterized by low, middle, and high income with each category representing three categorical values. Lastly, education was defined by six categorical values within the time activity dataset therefore the three groups were composed of two values: “Less than High School/High School,” “Vocational/Some College,” and “Undergraduate/Graduate.” 4.7

Dynamic Exposure by Subgroup The second statistical analysis was to perform ANOVA to test for significant

differences within dynamic exposure using the same subgroups that were outlined above. 65

 

Compared to the first analysis, both twenty four hour concentration values and daytime concentration values were examined. 4.8

Multiple Linear Regression of Dynamic Exposure The final statistical analysis involved the implementation of multiple linear

regressions of dynamic exposure in relation to age, income, gender, and education. Due to the categorical nature of the independent variables, age, income, and education were once again arranged into the same groups that were used for the paired t-tests (i.e. young, middle, and old age groups). Also, prior to building the multiple linear regression models, exploratory data analysis was performed using R Statistical Software Package version 2.7.1. Daytime and twenty four hour travel concentration sums were analyzed using graphical techniques such as histogram plots and box plots for all travel days.

66

 

Chapter 5: Results 5.1

Dynamic vs. Static Paired T-Test All Individuals Table 4 summarizes the results for the static versus dynamic paired t-test for all

292 individuals. The paired t test null hypothesis states that the difference between the two observations (individual static and dynamic exposure) is equal to zero. A significance level of 0.05 was implemented for all paired t-tests.

t-Test: Paired Two Sample for Means All Individuals Mean Variance Observations Pearson Correlation Hypothesized Mean Difference df t Stat P(T