Modeling total reduced sulfur and ... - University of Montana

University of Montana University of Montana

ScholarWorks at University of Montana ScholarWorks at University of Montana

Graduate Student Theses, Dissertations, & Professional Papers Graduate School

2000

Modeling total reduced sulfur and sulfur dioxide emissions from a Modeling total reduced sulfur and sulfur dioxide emissions from a

kraft recovery boiler using an artificial neural network and kraft recovery boiler using an artificial neural network and

Investigating volatile organic compounds in an urban Investigating volatile organic compounds in an urban

intermountain valley using a TD/GC/MS methodology and intermountain valley using a TD/GC/MS methodology and

intrinsic tracer molecules intrinsic tracer molecules

Christopher L. Wrobel The University of Montana

Follow this and additional works at: https://scholarworks.umt.edu/etd

Let us know how access to this document benefits you.

Recommended Citation Recommended Citation Wrobel, Christopher L., "Modeling total reduced sulfur and sulfur dioxide emissions from a kraft recovery boiler using an artificial neural network and Investigating volatile organic compounds in an urban intermountain valley using a TD/GC/MS methodology and intrinsic tracer molecules" (2000). Graduate Student Theses, Dissertations, & Professional Papers. 10607. https://scholarworks.umt.edu/etd/10607

This Dissertation is brought to you for free and open access by the Graduate School at ScholarWorks at University of Montana. It has been accepted for inclusion in Graduate Student Theses, Dissertations, & Professional Papers by an authorized administrator of ScholarWorks at University of Montana. For more information, please contact [email protected].

https://scholarworks.umt.edu/

https://scholarworks.umt.edu/etd

https://scholarworks.umt.edu/etd

https://scholarworks.umt.edu/grad

https://scholarworks.umt.edu/etd?utm_source=scholarworks.umt.edu%2Fetd%2F10607&utm_medium=PDF&utm_campaign=PDFCoverPages

https://goo.gl/forms/s2rGfXOLzz71qgsB2

https://scholarworks.umt.edu/etd/10607?utm_source=scholarworks.umt.edu%2Fetd%2F10607&utm_medium=PDF&utm_campaign=PDFCoverPages

mailto:[email protected]

INFORMATION TO USERS

This manuscript has been reproduced from the microfilm master. UMI films the text directly from the original or copy submitted. Thus, some thesis and dissertation copies are in typewriter face, while others may be from any type of computer printer.

The quality of this reproduction is dependent upon the quality of the copy subm itted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleedthrough, substandard margins, and improper alignment can adversely affect reproduction.

In the unlikely event that the author did not send UMI a complete manuscript and there are missing pages, these will be noted. Also, if unauthorized copyright material had to be removed, a note will indicate the deletion.

Oversize materials (e.g., maps, drawings, charts) are reproduced by sectioning the original, beginning at the upper left-hand comer and continuing from left to right in equal sections with small overlaps.

Photographs included in the original manuscript have been reproduced xerographically in this copy. Higher quality 6* x 9” black and white photographic prints are available for any photographs or illustrations appearing in this copy for an additional charge. Contact UMI directly to order.

Bell & Howell Information and Learning 300 North Zeeb Road, Ann Arbor, Ml 48106-1346 USA

800-521-0600

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Maureen and MikeMANSFIELD LIBRARY

The University ofMONTANA

Permission is granted by the author to reproduce this material in its entirety, provided that this material is used for scholarly purposes and is properly cited in published works and reports.

** Please check "Yes" or "No" and provide signature **

Yes, I grant permission No, I do not grant permi

Author’s Signature

Any copying for commercial purposes or financial gain may be undertaken only with the author’s explicit consent.


M o d e l i n g TRS a n d S 02 E m i s s i o n s f r o m a K r a f t R e c o v e r y B o il e r U s in g a n A r t i f i c i a l N e u r a l N e t w o r k

and

In v e s t i g a t i n g V o l a t il e O r g a n i c C o m p o u n d s in a n U r b a n In t e r - M o u n t a i n V a l l e y U s in g a TD/GC/MS M e t h o d o l o g y a n d In t r in s ic

T r a c e r M o l e c u l e s

by

C hristopher L. Wrobel

B.S., Ursinus C ollege, Collegeville, Pennsylvania, 1991

Presented in partial fulfillm ent o f the requirem ents

for the degree o f

D octor o f Philosophy

The University o f Montana

2000

Approved by:

Chairperson

Dean, Graduate School

___________ 3 - 3 g - a o o oDate


UMI Number 9974233

_ ___ (5j

UMIUMI Microform9974233

Copyright 2000 by Bell & Howell Information and Learning Company. All rights reserved. This microform edition is protected against

unauthorized copying under Title 17, United States Code.

Bell & Howell Information and Learning Company 300 North Zeeb Road

P.O. Box 1346 Ann Arbor, Ml 48106-1346


W robel, Christopher L. Ph.D., March 2000 Chemistry

M odeling TRS and SO; Emissions from a Kraft Recovery Boiler Using an ArtificialNeural NetworkandInvestigating Volatile Organic Compounds in an Urban Inter-M ountain Valley Using a TD/GC/M S M ethodology and Intrinsic Tracer M olecules

Director: Garon C. Smith ( j

Back-propagation neural networks were trained to predict total reduced sulfur (TRS) and SO; em issions from kraft recovery boiler operational data. A 0.721 coefficient o f correlation was achieved between actual and predicted sulfur emissions on test data withheld from network training. The artificial neural netw ork (ANN) models found an inverse, linear relationship between TRS/SO; emissions and percent opacity. A number o f relationships among operating parameters and sulfur em issions were identified by the ANN models. These relationships were used to formulate strategies for reducing sulfur emissions. Disagreement between ANN model predictions on a subsequent data set revealed an additional scenario for sulfur release not present in the training data. ANN modeling was demonstrated to be an effective tool for analyzing process variables when balancing productivity and environmental concerns.

Five receptor sites distributed in the M issoula Valley, M ontana, were employed to investigate possible VOC (benzene, 2,3,4-trimethylpentane, toluene, ethylbenzene, m-ip- xylene, o-xylene, naphthalene, acetone, chloroform, a-p inene, P-pinene, p-cym ene and lim onene) sources. The most dom inant source o f VOCs was found to be vehicle emissions. Furthermore, anthropogenic sources o f terpenoids overwhelmed biogenic emissions, on a local scale. Difficulties correlating wind direction and pollutant levels could be explained by wind direction variability, low w ind speed and seasonally dependent meteorological factors. Significant evidence was com piled to support the use o f p-cym ene as a tracer molecule for pulp mill VOC em issions.

A pportionm ent techniques using o -xylene and p-cym ene as tracers for automobile and pulp mill emissions, respectively, were employed to estim ate each source’s VOC contribution. M otor vehicles were estimated to contribute between 56 and 100 percent o f the aromatic pollutants in the M issoula Valley airshed, depending upon the sampling location. Pulp mill emissions were estimated to account from 1 to 34 percent o f the aromatic chemicals in the airshed. Measured ambient chloroform levels were attributable to the pulp mill (12 - 70%) and non-point source urban em issions (7.5 - 30%).

ii


To my wife,

Amy

iii


ACKNOWLEDGEMENTS

Now that all o f the tasks have been com pleted and my goal attained, I realize how many other people have helped me along the way. I begin by thanking my research advisor Garon Smith. Garon allow ed me the freedom to pursue my ow n research ideas. His excellent reputation and the com m unity's respect for him enabled m e to obtain local research funding. G aron guided the developm ent o f both my research and writing skills.I will not forget the com passion, concern and support the Smith family show ed me after my accident.

Ted Christian, Bruce King, Tim Looper and M att and Karla B aehr have my sincerest thanks for all o f their help and support during my convalescence. I thank Bruce King for being a good friend and a valued colleague. I will always appreciate his computer expertise. I thank Tony Ward for allow ing m e to collect sam ples at his home, for collecting sam ples for me when I was away, and for his friendship. I thank Jon Speare, Dave Jones, G reg Muth, Craig French and D oug Kolwaite for all the hiking, fishing and skiing adventures.

1 am grateful to the Smurfit-Stone C ontainer Corporation for funding my research, especially the M issoula M ill's General M anager, Bob Boschee and then D irector o f Environmental A ffairs, Ed Scott. I thank Leif G riffin for understanding that my first priority had to be m y degree. He taught me a great deal about environm ental engineering and stood out as an excellent role model. I thank Dave Stengel and M ike W ills for their assistance with the recover}' boiler project. I thank my committee m em bers Charles Thompson, Richard Field, Robert Yokelson and Vicki Watson for their tim e and insightful suggestions.

I thank M ichael Doyle, the Coffeys and the Rileys for allowing me to collect samples on their properties. No chemistry graduate student could possibly succeed without the help o f Bonnie Gatewood, Barbara M cC ann and Gayle Zachariason. They have my gratitude and sincere appreciation.

Last but not least, I thank my wife, Amy. I will always appreciate the many sacrifices she made so I could earn my Ph.D. She has been a nurse, an editor, a sounding board for a myriad o f ideas, a hiking and fishing partner, and above all, my best friend. Without her support and constant encouragement, I could not have reached my goal.

iv


TABLE OF CONTENTS

A bstract.......................................................................................................................................iiAcknowledgem ents................................................................................................................ ivList o f T ab le s ........................................................................................................................ viiiList o f F igures..........................................................................................................................ix

P r e f a c e

1. Introductory Remarks .................................................................................................... xi2. Purpose o f Study

2.a. M odeling TRS and SO; Em issions from a KraftRecovery Boiler Using an Artificial Neural N etw ork ...................... xi

2.b. Investigating Volatile Organic Compounds in an UrbanInter-M ountain Valley Using a TD/GC/M S Methodologyand Intrinsic Tracer M olecules..............................................................xii

3. Overview o f Thesis O rganization................................................................................xii

P a r t A : M o d e l in g T R S a n d S 0 2 E m is s io n s f r o m a K r a f t R e c o v e r y Bo il e r Us in g a n A r t i f ic ia l N e u r a l N e t w o r k

C hapter 1 Introduction1.1 Kraft Recovery Boiler Physical and Chem ical Processes ...................................I

1.1.1 Black Liquor Processing...............................................................................21.1.2 Com bustion Air Delivery..............................................................................5

1.2 Neural Network Control o f Process Equipm ent ....................................................81.2.1 Neural Network Studies at the M issoula M ill......................................11

C hapter 2 Experimental M ethod2.1 Boiler Parameters Included in ANN M o d e l.......................................................... 132.2 ANN Model C rea tio n ................................................................................................. 17

C hapter 3 Results and Discussion3.1 Predicted Sulfur Em issions........................................................................................ 193.2 Analysis o f ANN M odel............................................................................................ 203.3 ANN Model A pplication ...........................................................................................25

C hapter 4 C onclusions...................................................................................................................... 28

v


P a r t B: I n v e s t i g a t i n g V o l a t i l e O r g a n i c C o m p o u n d s in a n U r b a n I n t e r - M o u n t a i n V a l l e y U s in g a TD/GC/MS M e t h o d o l o g y a n d

I n t r i n s i c T r a c e r M o l e c u l e s

Chapter 5 Introduction5.1 Significance o f Volatile Organic Compounds in the Troposphere................... 305.2 Sources o f V olatile Organic Compounds to the A tm osphere.............................32

5.2.1 Sources o f Alkane and Aromatic Volatile OrganicC om pounds.................................................................................................33

5.2.2 Sources o f C hloroform .............................................................................. 345.2.3 Sources o f Terpenes.................................................................................... 355.2.4 Sources o f A cetone..................................................................................... 37

5.3 Methods o f Source A pportionm ent..........................................................................385 .3 .1 Traditional Apportionm ent M ethods...................................................... 38

5.3.1.1 T race rs ......................................................................................... 405.3.2 Artificial Neural N etw orks....................................................................... 42

Chapter 6 Experimental M ethod6.1 Sampling D esign............................................................................................................46

6.1.1 Sam pling L ocations.................................................................................... 476.1.2 Sam pling Frequency...................................................................................52

6.2 Sample Collection M ethodology.............................................................................. 536.3 Thermal Desorption/Gas Chromatography/M ass Spectrometry

Methodology'.............................................................................................................556.3.1 Instrum ent C alibration ................................................................................586.3.2 Analytical Uncertainty'............................................................................... 62

6.4 Statistical A nalysis........................................................................................................636.5 Apportionment C alculations.......................................................................................65

6.5.1 A pportionm ent Using «-Xylene Tracer M ethod...................................656.5.2 A pportionm ent Using “Dispersion" Reference M ethod.................... 666.5.3 Apportionm ent Using Linear C om binations........................................ 67

Chapter 7 Results7.1 Qualitative S tu d y .......................................................................................................... 697.2 Quantitative R esults......................................................................................................747.3 Statistical R esults.......................................................................................................... 777.4 Apportionment R esu lts ................................................................................................ 77

vi


C hapter 8 Discussion8 .1................................Experimental M ethodology............................................................878.2 Overall Qualitative and Quantitative R esu lts ........................................... 90

8.2.1 Chloroform and the Terpenes - Group 1 Com pounds........................ 908.2.1.1 The Impact o f Chlorine Bleaching on the

V alley A irshed ....................................................................... 948.2.2 Aliphatic and Aromatics - Group 2 Com pounds..................................978.2.3 A cetone........................................................................................................... 99

8.2.3.1 Seasonal Fluctuations in Acetone L evels........................... 1008.3 M eteorological Param eters and Their Correlation with Chemical

C oncentrations.........................................................................................................1038.3.1 Seasonal W eather Patterns........................................................................1048.3.2 W ind V elocity and Chem ical C oncentrations.....................................1078.3.3 W ind Direction and Chemical D istributions.......................................108

8.4 Apportionm ent o f V olatile Organic Pollutants in theM issoula V alley.......................................................................................................1138.4.1 o-Xylene as a Tracer M olecule for Fuel/Vehicle

Related Em ission Sources...................................................................... 1148.4.2 /7-Cymene as a Tracer M olecule for Mill Em issions......................... 1188.4.3 A pportionm ent Using «-X ylene as a Tracer for

Fuel/Vehicle Related E m issions...........................................................1228.4.4 A pportionm ent Using /7-Cymene and o-Xylene as

''Dispersion'^ Reference C om pounds...................................................1258.4.5 A pportionm ent Using Linear C om binations.......................................130

C hapter 9 Conclusions and Considerations for Future Study9.1 Conclusions from the A irshed S tudy................................................. 1339.2 Improvements to the Valley Airshed Study and M odeling............................. 1349.3 Investigation o f Biogenic Em issions and Acetone Levels...............................136

Literature C ited ....................................................................................................................................137

vii


LIST OF TABLES

Table 2.1 Recovery boiler input-output values...................................................................... 14Table 2.2 The most influential input parameters according to the ANN m od e ls 21Table 5.1 Estimated tropospheric lifetimes o f various organic chem icals.................... 30Table 5.2 Health advisory levels for the VOCs quantified in this study, which are

considered hazardous air pollutants by the E P A ...................................... 31Table 6.1 Parameters for the thermal desorption unit operation.......................................55Table 6.2 M ethod parameters for the purge and trap u n it ................................................. 56Table 6.3 Specific chromatograph and mass spectrom eter param eters..........................56Table 6.4 Chem icals purchased from Aldrich in neat form...............................................57Table 6.5 Compounds purchased from Supelco as certified solutions in

m ethano l..............................................................................................................57Table 6.6 M inimum instrument response and resultant detection limit for

all analytes in the quantitation d a tab a se ..................................................... 60Table 6.7 The relative uncertainties associated with the analytical m ethod................. 62Table 7.1 Predom inant com pounds found in the city com pared to those

found at the m ill.................................................................................................72Table 7.2 Final list o f 15 com pounds studied........................................................................74Table 7.3 Summary o f aliphatic and aromatic com pound concentrations..................... 80Table 7.4 Summary o f acetone, chloroform and terpene concentrations...................... 81Table 7.5 Calculated Student's t values for site and group mean com parisons 82Table 7.6 A pportionm ent results using o-xylene as a tracer m olecule...........................83Table 7.7 A pportionm ent o f aromatic chemicals using

"dispersion" reference com pounds............................................................... 84Table 7.8 A pportionm ent o f chloroform and terpenoids using

"dispersion" reference com pounds...............................................................85Table 7.9 Estimated percentages o f total arom atic compounds from

various sources................................................................................................... 86Table 8.1 Comparison o f chloroform levels before and after

the bleach plant c lo su re ...................................................................................95

viii


LIST OF FIGURES

Figure 1.1 A schematic draw ing o f the recovery boiler modeled andthe major chemical reactions occurring in each zone................................ 3

Figure 3.1 Predicted and actual total sulfur emissions for withheld test data ................ 19Figure 3.2 Correlation plot for the actual and predicted T S U L ........................................ 20Figure 3.3 The relationship between TSUL and % 0 ; extracted from the

trained AN N .......................................................................................................21Figure 3.4 The relationship between liquor flow and TSUL extracted from the

trained AN N .......................................................................................................23Figure 3.5 The relationship between TSUL and %opacity extracted from the

trained AN N .......................................................................................................24Figure 3.6 Resulting TSUL predictions from the trained network on a new

set o f data ............................................................................................................25Figure 3.7 Liquor flow and TSUL emissions for the new' data s e t ................................... 27Figure 3.8 Percent 0 2 and TSUL emissions for the new data set......................................27Figure 6.1 Map o f M issoula Valley indicating the sampling locations...........................47Figure 6.2 Site 3; with PM -10 sam pler and weather station v isib le .................................48Figure 6.3 The Smurfit-Stone Container Pulp Mill viewed from site 3 ...........................48Figure 6.4 Site 4; near the intersection o f Kona Ranch and Big Flat R oads...................49Figure 6.5 Site 5; directly o ff Big Flat R oad.......................................................................... 50Figure 6.6 Site 5; note proxim ity o f sampling device to conifers......................................51Figure 6.7 Site 1; located on the western edge o f urban M issoula................................... 51Figure 6.8 Site 2; near the intersection o f Front and Madison Streets............................. 52Figure 6.9 Diagram o f CMS tube calibration dev ice ........................................................... 58Figure 7.1 Chromatogram o f an air sample collected at site 2 in urban M issoula 70Figure 7.2 Chromatogram o f an air sample collected near the evaporator hotwelIs...71Figure 7.3 Chromatogram o f an air sample collected ^300 yards southeast

o f the pulp milTs waste fuel bo iler.............................................................. 71Figure 8.1 Mean concentrations o f chloroform, p-cymene and a-pinene at

the five sam pling locations............................................................................ 91Figure 8.2 Mean concentrations o f P-pinene and limonene plotted with a-pinene ....92Figure 8.3 Chloroform and terpene trend, site 3 to urban M issoula................................ 93Figure 8.4 Mean concentrations o f toluene and 2,3,4-trimethylpentane......................... 97Figure 8.5 The trend in the group 2 com pounds...................................................................98Figure 8.6 Mean concentrations o f toluene and o-xylene

across the M issoula V alley ............................................................................98Figure 8.7 Mean relative response ratios for ace to n e ..........................................................99Figure 8.8 Seasonal trends in acetone level and daily sunlight........................................ 101Figure 8.9 p-Cym ene concentrations at site 3 ..................................................................... 105Figure 8.10 p-Cym ene concentrations at site 1..................................................................... 106Figure 8.11 Toluene concentrations at site 1 ......................................................................... 107Figure 8.12 Toluene concentrations at site 3 ......................................................................... 107Figure 8.13 Fall-off in toluene concentration with wind speed at site 5 ..........................108Figure 8.14 NWS wind direction and site 3 p-cym ene concentration............................. 109

ix


Figure 8 .15 Site 3 wind direction and site 3 p -cymene concentration ..............................111Figure 8.16 The linear relationships between fuel/vehicle em ission compounds

and o-xylene determined from the M alfunction Junction samples. .. 115 Figure 8.17 Actual and predicted toluene concentrations from site (1 +2)

using o-xylene as a tracer m olecule............................................................116Figure 8.18 Actual and predicted toluene concentrations at site (1 -2 ) using

2,3,4-trimethylpentane as a tracer m olecule.............................................118Figure 8.19 a-Pinene versus /?-cymene concentrations at site 3 ........................................ 121Figure 8.20 Percentages o f aromatic pollutants attributable to vehicle

emissions via the o-xylene tracer................................................................ 123Figure 8 .21 Predicted amounts o f aromatic compounds from vehicle em issions 124Figure 8.22 Percentages o f aromatic compounds estimated from the m ill..................... 126Figure 8.23 Predicted concentrations o f group 2 compounds attributable to

mill em issions.................................................................................................. 127Figure 8.24 Estimated amounts o f chloroform and terpenes from the m ill.................... 129Figure 8.25 Estimated percentages o f chloroform from the mill and

urban Missoula sources..................................................................................130

x


PREFACE

1. Introductory Remarks

This thesis consists o f two separate research projects, related by a common goal.

Each project was conceived in an effort to im prove the air quality o f the Missoula Valley.

The first project focused on reducing pollutant em issions at the source. The second

project exam ined volatile organic compounds in the valley to assist with the formulation

o f a strategy for airshed improvement. W hile these projects were confined to a specific

geographic setting, the investigative methods are applicable on a more general scale.

2. Purpose o f Study

2.a. M odeling TRS and SO ; Emissions from a Kraft Recovery Boiler I /sing an Artificial Neural Network

This part o f the thesis demonstrates how artificial neural network (ANN)

m odeling can reveal the relationships among process variables, boiler productivity' and

sulfur em issions (sulfur em issions = total reduced sulfur, TRS, plus SO ;) within a kraft

recovery boiler. A broader understanding o f the link between boiler variables and sulfur

em issions w ould be helpful to kraft paper mills worldwide. The release o f sulfur from

the recovery boiler stack is a two-fold economic problem. First, any sulfur losses must

be replaced with the purchase o f additional sodium sulfate and may result in costlier

annual em ission fee payments. Second, the em ission o f sulfur to the atmosphere is an

environm ental concern, which could com prom ise good community relations. This

portion o f thesis has been subm itted to TAPPI Journal as a peer-reviewed publication.

xi


2. b. Investigating Volatile Organic Compounds in an Urban Inter-Mountain ValleyUsing a TD O C MS M ethodology and Intrinsic Tracer Molecules

The M issoula Valley has a very limited industrial presence. Increasingly stringent

environmental regulations continually require industries to reduce em issions o f various

pollutants. W hile any reductions in the ambient levels o f pollution certainly are

environmentally positive, they may require expenditures o f several millions o f dollars.

W hat actual benefits these large capital expenses have bought, o r will buy, for the airshed

remain questionable. How can we best improve the health o f the valley airshed? This

research project sought to answ er two questions: 1) w hat are som e key volatile organic

compounds that have a significant presence in the airshed and, 2) from where do they

originate?

J. Overview o f Thesis Organization

This thesis is organized into two parts, A and B. Part A consists o f four chapters,

which detail the artificial neural network model o f the kraft recovery’ boiler. Part B

contains five chapters and provides a description o f the valley airshed study, the chemical

source apportionm ent and the analytical methodology used. No appendices have been

added, instead air sam ple data, QAJQC data and instrum ent calibrations are included on a

CD-ROM.

xii


Pa r t A

M o d e l i n g TRS a n d S 02 E m is s io n s f r o m a K r a f t R e c o v e r y B o il e r U s in g a n A r t if ic ia l N e u r a l N e t w o r k

C hapter 1. Introduction

1.1 Kraft Recovery Boiler Chemical and Physical Processes

This section provides information that will help clarify the key elements o f the

recovery boiler processes necessary for the artificial neural network study. Black liquor

combustion and kraft recovery boiler chemistry and engineering are discussed. The

physical and chemical stages undergone by black liquor as its inorganic constituents are

"recovered" are discussed. The operating parameters under direct and indirect control are

identified and described for their usefulness in the boiler model.

In the kraft process, white liquor, containing sodium hydroxide and sodium

sulfide is used to delignify wood chips. The Sm urfit-Stone Container Paper Mill uses the

pulp from this process to produce linerboard. The spent cooking liquid, known as black

liquor, undergoes a series o f conversion steps that regenerate the white liquor. The

central step in the regeneration process occurs in the recovery boiler. The boiler bums

the organic content o f the black liquor, which consists o f lignin, sugar acids, and

extractives, providing the pulp mill with heat required to generate steam [Adams and

Frederick, 1988], During the combustion process, the sulfate contained in the black

liquor is reduced to sulfide and other inorganic chem icals are recovered in molten form.

1


1.1.1 Black Liquor Processing

Black liquor starts out "w eak" as recovered from the brown stock washers. The

first step in the recovery process is to concentrate the black liquor solution. In this step,

the percent solids in the liquor is increased from an initial -1 5 % to a final -55% by-

cycling it through a series o f multiple-effect evaporators. The end result is "strong" black

liquor. A direct contact evaporator further concentrates the solution to -65% solids.

Purchased sodium sulfate, as well as sodium sulfate and sodium carbonate recovered

from the econom izer and precipitator hoppers are added to the liquor to make up for

sulfur and sodium loss. Before delivery into the boiler, the liquor is heated to -200° F to

reduce its viscosity and help give it reproducible spray characteristics. The heated black

liquor is then sprayed onto the inner boiler walls via two oscillating guns. The liquor

feed rate and nozzle pressure are two important factors influencing the size o f the

droplets [Adams and Frederick, 1988], Figure 1.1 provides a schem atic drawing o f a

recovery boiler including the m ajor chemical reactions by region.

Each liquor droplet undergoes four steps on its way to the char bed: drying,

devolatilization, swelling and char burning. While each step generally occurs in a

distinct region o f the furnace, it would not be surprising to find the first through third

steps occurring in the char bed as well [Brown et a l., 1990]. The physical and chemical

properties o f the black liquor will influence its droplet size and combustion

characteristics.

The first step o f black liquor combustion is droplet drying. The droplet loses

w ater molecules by both surface evaporation and swelling and subsequent bursting

[Adams and Frederick, 1988], Drying is primarily dependent upon the heating rate o f the


droplet [Wessel et ctl., 1997], Significant variations in the drying step have been

observed due to variations in droplet size, furnace tem perature and chemical com position

[Adams and Frederick, 1988].

-<r ■ uO O O TON P K C V O T r U N IT FOR HCEWCfl - W AU & RF COAPCNATlON

M ISSOULA M *J_ M ISSO U L A . MONTANA

e aw c o n t r a c t ho . p a - i47

Oxidizing ZoneCO + ' . 'S o 2 - co2 H, - '/i o 2 - h 2o H2S + 3/2 O, - S 0 2 + H ,0so2 - ■ i o2 - so,Na2S + 2 0 ; - Na2 S 0 4

Na,CO, — SO3 — Na2 S 0 4 r C 0 2

Na2S + 3/2 0 2 - C O ,'- N a,C 0 3 - S 0 2

Na2 C 0 3 - S 0 2 - CO, + Na,SO,Na2 SO, + ' 2 0 2 - Na2 S 0 4

Drying ZoneOrganics + A — Pyrolysis Products Organics * A + 0 2 — Combustion Products Na,S + CO, -r H ,0 - Na,CO, + H,S c h , + h 2o — CO - 3 H2

Na20 + C 0 2 - N a,C 0 3

Na20 + H2 6 - 2 NaOH

Reducing ZoneOrcanics + A —• Pvrolysis Products 2 C + 0 2 — 2 CO CO + O, - CO,CO, + C - 2 CO

•«SS N a,S0 4 + 2 C - Na2S + 2 CO,N a,S0 4 , 4 C - Na;S + 4 CO N a,S04 + C - N a ,0 + S 0 2 + CO H, + '/, O, - H ,0 C - H ,0 - CO + H,C + 2 H, — C tt,Na,S + H ,0 - N a ,0 + H,S

Figure 1.1 A schem atic drawing o f the recovery boiler m odeled and the m ajor chemicalreactions occurring in each zone.

The second step is referred to as devolatilization. This process can proceed in

both an oxidizing and reducing environment. Once the droplet dries, its tem perature rises

significantly. Som e o f the organic material within the droplet begins to thermally

degrade. Released are CO 2 , CO, H,, H2O, light hydrocarbons, tars, FTS, and


4

organosulfur compounds [Adams and Frederick, 1988], U nder reducing conditions, the

process o f pyrolysis leads to the release o f volatiles. Due to the lack o f oxygen in the

vicinity o f the droplet, however, com bustion o f the gases does not occur close to the

falling droplet. Under oxidizing conditions, these volatiles are immediately burned upon

liberation. This com bustion increases the tem perature o f the droplet, further increasing

the devolatilization rate.

The third step o f black liquor combustion is droplet swelling. In this step, the

droplet grows somewhere between 10 to 30 tim es its initial size from the expansion and

evolution o f gases. The degree o f swelling is influenced by the chemical com position o f

the droplet. A high ratio o f final to initial droplet size is desired, as a low ratio results in

a lack o f porosity and poor char burning characteristics [Hupa, 1990],

Char combustion is the final step in the com bustion o f black liquor. At this stage

all that remains o f the droplet is solid char. C har consists prim arily o f elem ental carbon

and inorganic chemicals. It is in this step where the porosity o f the droplet becom es

critical. The oxygen at the bottom o f the furnace reacts with the hot char to form CO and

CO:. The oxygen concentration within the char bed is low, and the combustion is

incomplete. Carbon in the char as well as evolved CO and FT reduce the sulfate to

sulfide [Adams and Frederick, 1988], The heat from the char combustion melts the

sodium salts and they percolate through the char bed to the furnace floor. The sm elt

flows downward along the sloping furnace floor toward several water-cooled spouts. The

m olten salts drip into the dissolving tank to produce a solution known as green liquor.

Subsequent steps convert green liquor back into white liquor.


5

Efficient sulfate reduction is dependent upon several key variables. W hile sulfate

reduction has been shown to be independent o f the sulfate concentration for reduction

efficiencies up to 95% [Grace et al., 1990], it does depend strongly on the shape and

tem perature o f the char bed. Char bed temperatures o f at least 900° C are required to

efficiently reduce the sulfate to sulfide. Laboratory experim ents have dem onstrated that

no sulfide is formed when the temperature is lowered to 700°C [Sricharoenchchaikal et

al., 1997], A significant reduction in volatile sulfur release was noted as char particle

tem perature increased [Wag el al., 1997], A high average char bed temperature, while

resulting in good sulfate reduction, also results in high sodium particulate generation or

fuming [Wag et al.. 1997; Tavares et al., 1998; Verril and Wessel, 1998], In addition, it

has been observed that a highly uniform temperature distribution within the char bed

results in lower fume generation, as compared to a bed with the same average

tem perature, but with a widely varied temperature configuration [Tavares and Tran,

1998],

I. 1.2 ( 'ombustion Air Delivery

A pivotal factor in recovery boiler chemistry is the amount o f available m olecular

oxygen. Outside air, containing 21% molecular oxygen is supplied to the boiler by

forced draft fans via primary, secondary, and tertiary ducts. These three air systems

establish three distinct zones in the furnace, a drying zone where the liquor is sprayed, a

reduction zone surrounding the char bed and an oxidation zone in the turbulent upper

section o f the boiler [Smook, 1997]. Figure 1.1 lists the critical chemical reactions

thought to occur in each o f the three zones. A large induction fan, located post-


6

precipitator, maintains the internal negative pressure required for proper overall drafting

o f the boiler.

The primary' air system serves two functions: 1) a source o f more than half the

oxygen needed for com bustion and, 2) a physical means to shape the char bed which

forms on the furnace floor. The primary air system is com posed o f the forced draft fan,

the primary windbox, a gas fired heater, a series o f dampers and thirty two small ports on

each o f the four boiler walls. The air is heated to 300° F before entering the furnace in

order to enhance boiler efficiency by preventing cool air from entering the hottest zones

o f the boiler. The velocity' o f the primary' air is the lowest o f the three air injection

systems. The portion o f the furnace below the primary air system comprises the reducing

zone.

A secondary air system , located just above the primary a ir je ts, establishes an

oxygen-rich region through which black liquor droplets will fall and combust. It is

sim ilar in engineering design to the primary system, but the a ir is introduced at a higher

velocity through a sm aller num ber o f larger ports to assure efficient, turbulent mixing

[Smook, 1997], The liquor guns are situated above the secondary air ports. The black

liquor is sprayed onto the walls o f the boiler by two oscillating guns. It is desirable to

m aintain the lowest velocity o f a ir from the secondary air system , which will result in

adequate mixing. Too high o f a velocity and an increase in droplet entrainment will

occur [Adams and Frederick, 1988]. This situation will result in less char reaching the

furnace floor and an excess o f particulate in the boiler exhaust.

A ir entrainment o f liquor droplets and char particles is a serious concern. The

sm allest particles, with a diam eter less than 1 mm will be most easily entrained. Droplets


7

with a diam eter greater than 1 mm but less than 1.5 mm may be entrained, but will likely

fall back to the bed once they coalesce as smelt. The ideal diameter for a droplet is about

3 mm. A larger size will not permit sufficient drying before reaching the bed [Adams

and Frederick, 1988], H igher nitrogen oxide em issions are another consequence o f

entrainment. Longer particle residence tim e in the lower boiler has been correlated with

low er NOn em issions [Janka et a l., 1988],

The region o f the boiler above the prim ary air ports, including the liquor guns and

the secondary' a ir ports is the drying zone. Above the liquor guns to the bullnose, the

oxygen concentrations are highest and com prise the oxidation zone. A bimodal size

distribution o f em itted particles has been shown to occur in recovery' boilers. M ost

particles are 0.5-1.0 um in size, and are form ed from inorganic fume em itted from the

char bed. W hen char particles becom e entrained the resulting particulate form ed are 5

um in size [M ikkanen et al., 1994],

The tertiary' a ir system is located in the oxidation zone and provides the final

infusion o f oxygen into the boiler. The ports are located on two opposing walls o f the

furnace. Tertiary air is injected into the boiler at a high velocity to ensure total

penetration into the furnace and facilitate com plete m ixing o f the stack gases. The

tertiary air is introduced at am bient tem perature in an effort to cool the boiler exhaust.

Lowering the tem perature o f the entrained particles helps to solidify them so they are less

likely to stick to the superheaters and other boiler heat exchange components.


1.2 Neural Network Control o f Process Equipment

An artificial neural network is a com puter program, which is capable o f learning

the relationships between a set o f input parameters and an observed outcome. Input

signals are processed through interconnected neurons that predict an outcom e based on

"experience." In a back propagation network the predicted outcome is com pared to the

actual value and the root m ean squared error is used to adjust the weighting factors o f

each neuron. In this m anner highly influential input parameters are weighted more

heavily, while less important parameters are de-emphasized in the resulting model.

Neural networks and neurocontrol strategies are rapidly gaining the attention o f

industrial managers and environm ental scientists around the world. The attraction

towards neural networks and neurocontrol lies in the promise that, for the price o f some

software and a few sensors, routine data collection can be transform ed into a

sophisticated modeling endeavor. Neural network-derived models have an uncanny

ability to predict future outcom es that are site- or equipment-specific. At the sam e time,

they sift through complex, noisy or im precise data to reveal the underlying quantitative

and phase relationships present in a collection o f input variables.

The strengths o f the artificial neural network model match up well with the

mercurial kraft recovery boiler. Even under careful control, recovery boilers may

misbehave unpredictably. In one particular example, a boiler maintaining an acceptable

2.5% oxygen in the flue gas, had a total reduced sulfur com pound (TRS) concentration

over 15 ppm and a CO concentration greater than 500 ppm. Unstable com bustion in the

lower furnace was also a constant concern [Walsh et al., 1998], The degree o f

com plexity associated with recovery boiler combustion processes has hindered the


9

creation o f an all-encompassing model. A com putational fluid dynamics m odel has been

reasonably successful at predicting some o f the relationships between droplet size, gas

flow patterns, oxygen concentrations, and in-flight particle burning [Grace et a l., 1989],

To our knowledge, no com prehensive model o f em ission behavior has been published.

An emission-focused model extracted from a neural network may provide initial insights

into many o f the interrelated boiler processes.

Neural process control offers industry a co st effective means to achieve higher

product quality, increased productivity and low er costs. The paper industry has already

em ployed neural network modeling to improve pulp yield and strength [Tessier et a l.,

1995], as well as to diagnose web breaks on new sprint paper machines [M iyanishi and

Shimada. 1998]. But the benefits provided by neural network models do not stop with

higher profit margins. Neural networks can also help factories, mills, pow er plants, and

other industrial facilities lower their impact on th e environment. Finding the best

"recipe” for a process can also include provisions to cut energy consumption, reduce the

tonnage o f toxic ingredients or process chem icals, and lower stack emissions. Thus, both

the global environment and the corporate bottom line are winners.

One o f the simpler industrial com bustion sources to model is a natural gas

powered utility boiler. The success o f neural networks in the modeling o f these boilers is

reflected in the acceptance o f neural models as predictive emission monitors (PEM s)

[Baines et a l., 1997], Correlation between m easured and predicted NOx em issions has

been demonstrated to exceed 95%. United States environmental regulations already

allow PEMs to be used to calculate total em issions. Some states are in the process o f

passing legislation, which would allow industries to employ PEMs in place o f traditional


10

continuous em ission monitors (CEM s). These predictive models can be easier to

maintain as they do not have any direct contact with the harsh sm okestack environm ents

[Rumph and Rudd, 1998]. They may also aid the environm ent by forecasting, and thus

avoiding excess em issions rather than merely recording them.

Neural netw ork technology has been applied to the em issions o f nitrogen oxides

(NOn) from coal fired power plants. Since NOx is a principal ingredient in the

photochemical sm og reactions that lead to ozone in urban airsheds, its emissions are

strictly regulated in the U.S. and elsewhere. One way for a pow er com pany to lower its

NOx em issions is through neural network modeling. Reinschm idt and Ling [1994]

successfully trained a neural network using a back-propagation algorithm. Their results

indicate that it can advise a plant operator in real time about optimal settings for lower

NOx em issions from coal-fired boilers.

A more advanced control strategy for utility boilers involves the use o f electronic

cameras to discern em ission patterns in the boiler's flame. A neural network can process

digital images and has been able to successfully respond to sim ulated load excursions in a

closed loop control system [Allen et a l., 1993],

M unicipal incinerators have also been targeted as places where neural networks

may help to control the em issions o f hazardous air pollutants such as dioxins and furans.

Since the com position o f the waste material burned can change greatly over time,

automated control has been difficult to achieve. A neural network-based fuzzy logic

control algorithm , however, has had some success in enhancing incinerator perform ance

[Chen and Chang, 1996].


Artificial neural networks have been used to model the heat recovery efficiency o f

the gas/w ater scrubbers used in paperm aking. The network was designed to predict

process w ater outlet temperatures. The resulting model identified the parameters needed

to design a scrubber with superior heat exchange properties [M ilosavljevic and Heikkila,

1999],

1.2 .1 Neural Network Studies al the M issoula M ill

The goal o f the recovery boiler research was the creation o f an emission-focused

neural network model for the largest kraft recovery boiler at the M issoula paper mill.

Initially, the construction o f a boiler model that would predict the em issions o f odor-

causing TRS com pounds (e.g., dimethyl sulfide, methyl mercaptan), particulate, and SO 2

was attem pted. A superior protocol for boiler operation was sought, that would lower

em issions w ithout impacting productivity. Unfortunately, most o f the electrostatic

precipitator's operating parameters were not m easured or recorded. W ithout the

precipitator parameters, the feasibility o f a m odel for particulate em issions seemed

rem ote and was not pursued any further.

Work on an ANN model for the recovery boiler commenced in July 1996 based

on the premise that if TRS emissions could be reduced, the odor problem , and possible

health risks [Haahtela et al., 1992] in the M issoula valley could be eased. During various

stages o f the modeling, it became evident that the boiler TRS em issions were inversely

related to SO 2 emissions. By creating an operating formula that lowered TRS, SO 2

em issions were ultimately increased. Trading one pollutant for another was not the

outcom e desired for the project.


Continuing refinements to early ANN models led the project toward its ultimate

goal, the reduction o f incompletely oxidized sulfur em issions (TRS and SO:). Sulfur

exits the boiler in three general forms. The first is as volatile TRS compounds. These

com pounds have strong, objectionable odors and may cause acute health effects

[Haahtela et a l 1992], The second is as gaseous SO:. Sulfur dioxide is an EPA criteria

pollutant and while it possesses a less offensive odor, it is a lung irritant and contributes

to acid deposition. The third is as solid sodium sulfate, the primary chemical in the

particulate emitted by the recovery boiler. O f the three forms o f sulfur that can leave the

boiler, the easiest to control (via electrostatic precipitator) is the sodium sulfate. The

final stages o f the neural network modeling were focused on the creation o f a boiler

model that could be used to reduce TRS and SO: em issions from the stack, while not

com prom ising productivity. This could be accom plished by reducing particle

entrainm ent, m inimizing the formation o f reduced sulfur com pounds in the upper boiler

regions, and by fully oxidizing a larger portion o f those sulfur species to SO.T'; the form

least detrimental to the environment. To accom plish this goal TRS and SO: emission

concentrations were summed into a single value. This com posite value represented

undesirable sulfur losses and was termed total sulfur (TSUL).


Chapter 2 Experim ental Method

Neural networks require large am ounts o f data for training. The effort required to

enter process variables by hand is impractical. The recovery boiler studied was the only

unit interfaced to a comprehensive data acquisition system (DAS). In addition it was the

largest recovery boiler and the object o f local environm ental scrutiny. N ew source

performance standards (NSPS) required that boiler operating parameters and stack

emissions are continuously recorded as 6-m inute averages. These data were downloaded

from the DAS and used in the creation o f the ANN models.

2.1 Boiler Parameters Included in ANN ModeI

Five categories o f boiler parameters under operator control, e ither directly or

indirectly, were used as inputs to the neural network: liquor, air flow/pressure, furnace

draft, particulate emissions and boiler productivity. Table 2.1 lists each param eter used

in network construction, as well as their ranges and units o f measure.

13


14

Table 2.1 Recovery boiler input-output values.

A b b r e v i a t i o n P a r a m e t e r D e s c r i p t i o n U n i t o f M e a s u r e L o w V a l u e H i g h V a l u e

TSUL Total Sulfur (TRS S 0 2) ppm 5 7 209.3LIQF Black Liquor Flow gpm (Ls) 249 9 ( 1 5 . 7 6 ) 282.5 ( 1 7 . 8 2 )

NOZP Liquor Gun Nozzle Pressure psi (kPa) 30.9 ( 2 1 3 ) 36 I ( 2 4 9 )

LIQT Black Liquor Temperature degrees F ( K ) 211.7 ( 3 7 3 . 0 ) 21 9 6 ( 3 7 7 . 4 )

DENS Black Liquor Solids percent 61.94 67.27SRPM Saltcake Addition Rate rpm 9 14PAFL Primary Air Flow 103*Ibs/hour (kgs) 176.1 ( 2 2 . 1 9 ) 207.8 ( 2 6 . 1 8 )

SAFL Secondary Air Flow 103* lbs/hour (kgs) 2 2 1 A ( 2 8 . 6 5 ) 318 2 ( 4 0 . 0 9 )

TAFL Tertiary Air Flow 103*Ibs/hour (kgs) 78.1 ( 9 . 8 4 ) 95.0 ( 1 2 . 0 )

PAIR Primary Air Damper Position % open 49 9 58.1SAIR Secondary Air Damper Position % open 49.0 69.0TAIR Tertiary Air Damper Position % open 43.9 53.2PWB Primary Windbox Pressure in. o f H ;0 ( m ) 1.47 ( 0 . 0 3 7 3 ) 2 79 ( 0 . 0 7 0 9 )

SVVB Secondary Windbox Pressure in. o f FLO ( m ) 5.17 ( 0 . 1 3 1 ) 13.65 ( 0 . 3 4 6 7 )

TWBa Tertiary Windbox A Pressure in o f FLO ( m ) 10.59 ( 0 . 2 6 9 0 ) 16.88 ( 0 . 4 2 8 8 )

TWBb Tertiary Windbox B Pressure in. o f FLO ( m ) 11 ( 0 . 2 8 ) 17 ( 0 . 4 3 )

PAT Primary Air Temperature degrees F ( K ) 274 ( 4 0 8 ) 320 ( 4 3 3 )

SAT Secondary Air Temperature degrees F ( K ) 298 ( 4 2 1 ) 303 ( 4 2 4 )

DRFT Furnace Draft in o f H ;0 ( m ) -0.32 ( - 8 . 2 s ) -0.19 ( - 4 . 8 ' )

FNSP Induced Draft Fan Speed rpm 250 3340 2 Oxygen in Flue Gas percent 1.1 3.4OP AC Opacity percent 5.5 30STFL Steam flow 103*lbs/hour (k g s) 227 ( 2 8 . 6 ) 509 ( 6 4 . 1 )

SPHO Superheater Temperature degrees F ( K ) 663 ( 6 2 4 ) 753 ( 6 7 4 )

PPTT Electrostatic Precipitator Temperature degrees F ( K ) 405 ( 4 8 0 ) 465 ( 5 1 4 )

* SI units and values are shown in parenthesis where applicable.

Liquor variables influence the size o f the droplets form ed as the black liquor is

sprayed into the boiler. Black liquor flow, spray gun nozzle pressure, liquor tem perature,

liquor percent solids and saltcake (NaiSCL) addition rate were obvious choices for

network inputs. These variables influence the size o f the droplet formed when the black

liquor is sprayed into the boiler. Furtherm ore, droplet size and percent solids influence

particle entrainm ent, as well as the drying, devolatilization, swelling and char burning

characteristics o f the liquor. Black liquor flow indicates boiler loading, and will affect


15

char bed temperature. Increasing the black liquor flow will provide m ore fuel for the

char bed resulting in a higher bed temperature.

Air flow/pressure param eters dictate the amount o f oxygen entering the boiler and

control combustion efficiency. Prim ary, secondary and tertiary boiler airflow parameters

were included in the ANN. A irflow refers to the amount o f air entering the boiler. The

windbox pressure relates to the force to which the a ir enters the boiler. Primary and

secondary air are both preheated in order to enhance boiler efficiency by preventing cool

air from entering the hottest zones o f the boiler. Damper position will influence airflow

and pressure. The am ount o f oxygen entering the boiler will dictate combustion

efficiency, and therefore can be indicative o f boiler temperatures.

Primary air param eters are critical in boiler operation. The quantity o f air

entering the boiler at this location supplies -5 0 % o f the oxygen required for combustion.

The windbox pressure is an indication o f the depth to which the air penetrates into the

boiler. This helps create hom ogeneity within the boiler environm ent in the horizontal

plane. Primary air also shapes the char bed forming on the furnace floor. The shape o f

the bed is indicative o f sulfate reduction efficiency and fuming. Primary air is also

responsible for supplying oxygen to the char bed. Too little oxygen at the bed surface

will result in a low' bed tem perature and heterogeneous temperature distribution, which

will also lead to insufficient sulfate reduction and excess fuming. Secondary airflow

provides more oxygen and its increased introduction velocity aids in the m ixing o f boiler

gases. Tertiary airflow adds the last infusion o f oxygen into the flue gas. Possessing the

highest entrance velocity in order to increase turbulent mixing, it ensures high

com bustion efficiency.


16

Furnace draft parameters characterize internal gas velocity and are also indicative

o f smooth boiler operation. Fumace draft, induced draft fan speed and percent oxygen

com prised this input group. A sudden increase in fum ace draft or induced draft fan speed

might suggest plugging due to hot entrained particles adhering to heat exchanger tubing.

The percent oxygen in the flue exhaust was also used to register the boiler's operating

conditions. Too little oxygen in the flue gas may indicate incomplete combustion, while

an excessively high percentage could be indicative o f poor mixing conditions. Fum ace

draft and induction fan speed were used to give an indication o f internal gas velocity.

These parameters are also indicative o f smooth boiler operation. A sudden increase in

fum ace draft o r induction fan speed might indicate plugging due to hot entrained particles

adhering to heat exchanger tubing.

Percent opacity was included as an input variable since it tracks escaping sulfate

particulates. Com plete oxidation o f reduced sulfur com pounds present in the upper

fum ace would lead to an increase in sodium sulfate particulate in the stack gas. This

increase would lead to more particulate entering the precipitator and could cause higher

opacities.

Boiler productivity was described by steam production, superheater temperature

and electrostatic precipitator temperature. Superheater temperature is correlated with

boiler operating tem perature and particle entrainment. To maintain good heat transfer,

the superheater must be free from significant fume deposits and be exposed to adequate

flue gas temperatures. The temperature within the precipitator was also used to show

boiler efficiency by providing an indication o f how much heat was lost to the flue gas.

The list o f variables presented in Table 2.1 is a refinement o f the original set


17

employed. O ther param eters dem onstrated no forecast value in earlier trials; these

param eters were system atically pruned from the network as it underwent development.

Some potentially useful param eters, such as droplet size and internal boiler temperature

distribution, remain unknown. Their role was indirectly accounted for by the inclusion o f

parameters known to influence them. The sulfidity o f the liquor was a parameter that was

explored and subsequently dropped. This was done for two reasons: 1) the test for

sulfidity was only perform ed tw ice per day, while all other data used were recorded every'

six minutes, and 2) the available sulfidity data showed no changes beyond the inherent

error o f the analytical testing m ethod utilized.

2.2 T Y Y Model Creation

All neural network activities were performed with Version 3.11 o f the

BrainM aker Professional softw are (California Scientific Software, Nevada City, CA,

USA) and Version 3.1 o f its accom panying neural network spreadsheet NetMaker.

Substantial m anipulations were performed on the data sets before they were

im ported into the neural netw ork softw are using M icrosoft Excel. The data sets

dow nloaded from the boiler DAS w ere parsed, cleaned and reorganized. Any missing

dates and times were either corrected o r flagged. Missing or invalid data points were also

flagged. Since the chronological order o f the data was important, missing tim e periods or

those with invalid data had to be retained at this stage o f network construction. Input

parameters were normalized to generally fall within a range betw een 10 and 99. This was

done to prevent the network from over-em phasizing variables w ith high absolute values

and ignoring those with fractional absolute values.


18

Using Netmaker, a historical tim e series was created for each o f the input

param eters so that a training fact contained the current value plus a 30-m inute history (the

five previous 6-m inute interval values). This stage o f network construction required the

deletion o f the spurious/missing values during periods o f sensor calibration or sensor

m alfunction, which had been previously flagged. Since a spurious/missing value would

appear in the 30-minute history for the next five facts, a total o f six facts had to be

expunged on each occurrence. Finally, the rows (facts) were randomly shuffled.

The architecture o f the ANN was refined to produce the best overall testing

results. The final generation o f networks had 144 input neurons (the current input value

plus a 30-m inute history for each o f the 24 parameters listed in Table 2.1) and two hidden

layers. At the onset o f training there were 66 neurons in the first hidden layer and 33 in

the second. The network size was m odified while training; one neuron was added to each

hidden layer, in turn, if the root mean square error did not decrease by 0.01 after 50

iterations. The sole output neuron was total sulfur (TSUL) in parts per million. Each o f

the final networks was trained using different sets o f data, containing betw een 500 to

2500 facts. Training was performed using 90% o f the facts; 10% were w ithheld from

training and used for testing. A sigmoid neuron transfer function was used with a

minimum value o f zero and a maximum value o f one. A linear learning rate o f 0.75 was

selected to reduce the size o f the changes m ade to the weighting factors. Tolerance

tuning initially set at 0.4 o f the TSUL range was also employed. Tolerances were

successively reduced until network learning stalled o r a minimum tolerance o f 0.1 was

achieved. A noise value o f 0.05 was added to the training data to help the network

generalize rather than memorize the relationships am ong variables.


C hapter 3 Results and Discussion

3 .1 Predicted Sulfur Emissions

Predicted and actual TSUL values are plotted in Figure 3.1 for the test facts from

the network described above. The network successfully trained and tested to a 0.16

tolerance level after 2000 iterations. The chosen variables for the ANN model are good

predictors o f the boiler's total sulfur em issions as dem onstrated by the broad dynamic

range displayed by the output neuron. The ANN was trained over a large range o f TSUL

values (5.7 - 209.3). This helped to ensure that the network had "seen" the output for

many types o f anticipated boiler conditions. This was necessary because the trained

network can only choose output values that fall within the training range.

160 —

140120

| 100 Q.' 80

6040200

1 116 231 346 461 565 669 773 888 1003 1118 1222 1326 1430Fact

Predicted Actual

Figure 3.1 Predicted and actual total sulfur em issions for withheld test data.

Figure 3.2 correlates the actual versus predicted TSUL for the test data. The plot

indicates that the network can predict TSUL over the entire range o f training values with

equal success. Random, hom oscedastic scatter is responsible for the R2 value o f 0.721.

19


20

Equal distributions o f points both above and below the line support the use o f a linear

relationship. The resulting line through the points has a slope slightly less than one,

which indicates that the network predictions a re generally lower than the actual TSUL

values.

160

140

£120

a.& 100_i3(O 80t-"53 60O< 40

20

0

y = 0.9555x -7.1555= 0.721

20 40 60 80 100Predicted TSUL (ppm)

120 140 160

Figure 3.2 Correlation plot for th e actual and predicted TSUL.

3.2 Analysis o f ANN Model

The trained networks were analyzed to determ ine which input neurons were most

important in predicting TSUL. Initially, each input neuron was varied by ± 10% o f its

range, while all other inputs were held constant. This was done for every neuron and fact

in the set o f training data. The resulting values w ere paired with the average response o f

the output neuron (change in TSUL with +10% input neuron variation plus change in

TSUL with -1 0 % variation divided by 2). This produced a ranking o f input neuron

influence from those with the largest positive effec t on the TSUL output, to those with

the largest negative effect. The 25 most positively and the 25 most negatively influential


21

input param eters were then com pared among several networks trained using different sets

o f data. Those input parameters that were consistently im portant in all ANN models

w ere re-exam ined in greater detail (Table 2.2). The mathematical relationships between

these highly influential inputs and the output neuron were extracted from the networks.

Table 2.2 The most influential input parameters according to the ANN models.

Positively NegativelyBlack liquor flow Tertiary windbox pressure

Super heater output temp. Percent oxygen in flue gasPrimary air flow Black liquor percent solids

Primary wind box pressure Black liquor temperatureSecondary wind box pressure Secondary air temperature

Tertiarv air flow OpacityFum ace draft Induction fan speed

This was accom plished by varying the param eter o f interest, while holding the other

param eters constant at their m idrange value, and recording its effect on TSUL. Figures

3 .3 and 3.4 are the A N N 's representation o f the relationships found between TSUL and

two input variables, % CL and liquor flow.

6059

TSUL = 0.0051(%Oa)2 - 0.5503(%02) + 64.4458

E 57 £ 56_i13V)I— 54

5352

10.0 15.0 25.0 30.0 35.020.0

% o 2 *10

Figure 3.3 The relationship betw een TSUL and % CL extracted from the trained ANN.


The % O; rem aining in the flue gas has a direct bearing on the TSUL output

(Figure 3.3). C om bustion engineers are well aw are o f this relationship; the fact that the

network model found it w ithout prom pting gives the modeling technique a level o f

validation. Clearly, when the % CF in the exhaust gas is low, the level o f excess oxygen

is insufficient to com plete the oxidation o f su lfur species to sulfate (S +6). This produces

low opacity and a high concentration o f su lfur dioxide (S +4) and TRS (S - 2 ) in the stack

emissions. The % CF trace is non-linear, so TSUL emissions increase exponentially as

oxygen levels decrease. Figure 3.3 dem onstrates that an increase in the excess CF from

1.1 to 3.4% produces a 12% reduction in TSU L emissions.

As black liquor loading to the boiler increases, so do the overall stack em issions

(Figure 3.4). This relationship makes intuitive sense, since higher liquor loads tax the

combustion process for adequate gas m ixing, low er the amount o f oxygen available for

com plete oxidation and increase the quantity o f sulfur within the boiler. W hen the liquor

flow is increased from 250 to 285 gpm, a 13% increase, the amount o f TSU L em itted is

increased by only 3 ppm or 5.5%. This indicates that there is room to im prove TSUL

em issions without adversely affecting boiler productivity.


57TSUL = 0.0585(LIQF/10)2 - 2.1922(LIQF/10) + 72.00156.5

_ 56Eg; 55.5

_ i 3 cn i-55

54.5

54

53.524.5 25.0 25.5 28.026.0 27.5 28.527.026.5

LIQF (gpm/10)

Figure 3.4 The relationship between liquor flow and TSUL extracted from the trainedANN.

The partitioning o f sulfur between the completely oxidized sulfate form and

incompletely oxidized TSUL forms is shown in Figure 3.5. Since opacity is dominated

by sulfate particulate, it is a good indicator o f sulfate levels. W hen sulfate levels are

high, there is a corresponding reduction in TSUL. When TSUL is high, the % opacity is

low. The relationship is highly linear (R : = 0.9996), suggesting a simple, inverse

proportionality. M anipulating the boiler parameters to reduce either TSUL or opacity

may result in higher em issions o f the other. A balance must be struck between the two

pollutants. Retaining more sulfur in the char bed would increase the amount recovered as

molten sulfide, curtailing both TSUL and particulate emissions, thus, lowering the entire

trace in Figure 3.5.


24

57TSUL = -0.2819(%OPAC) + 58.25356

R = 0.999655

I 54 &Q .

53- I

w 52

50

499.0 24.04.0 14.0 19.0 29.0

% OPAC

Figure 3.5 The relationship between TSUL and % opacity extracted from the trainedANN.

Several strategies for lowering TSUL em issions can be suggested based on the

relationships uncovered by the ANN models. Decreasing both the primary and secondary

windbox pressures results in 8.5 and 7.7 percent reductions in TSUL emissions,

respectively. If the tertiary windbox pressure is increased simultaneously, TSUL

emissions are lowered another 12.3%. Altering the characteristics o f the black liquor

prior to its introduction into the boiler also lowers TSUL values. An increase in the

liquor temperature o f 10° F is predicted to reduce TSUL by 14%. The models also

predict that a 4% increase in the solids content o f the liquor will decrease the TSUL

emissions by another 5%.

Figures 3.3 and 3.4 indicate that many relationships in the ANN are nonlinear.

These complex relationships are often not adequately represented using multiple

regression techniques. Correct use o f regression procedures requires advanced

knowledge o f the form o f the underlying relationship (e.g., linear, exponential,


25

logarithm ic, etc.). Even with an AN N to help uncover the appropriate model for each

input variable, regression techniques still proved inadequate. W hile it is not difficult to

elucidate the general effect each input neuron has on the output o f the network, it is quite

difficult to use these resulting relationships to create a sim plified model. The ANN

model responded far better to the com plexity o f the interrelated variables than any

regression model extracted from the trained network.

3.3 l . V . V Model Application

The value o f a model can be dem onstrated in what it reveals when applied to new,

unknown data. Model shortfalls can often prove more instructive and interesting than the

successes. Figure 3.6 shows the ne tw ork 's predictions o f total su lfur loss on one

particular set o f boiler data that w as not used in its training. The ANN reasonably

forecasted the sulfur loss for m uch o f the period, but clearly failed to predict the large

excursion betw een facts 150 and 200.

250

200

a. 150 a .

_i3COt-100

50

1 18 35 52 69 86 103 120 137 154 171 188 205 222 239 256 273 290Fact

Actual Predicted

Figure 3.6 Resulting TSUL predictions from the trained network on a new set o f data.


Failure o f the model was prom pted by boiler conditions that were not

encom passed in the training set. The ANN model interpreted the input parameters in the

unseen data set and determ ined that during this period o f tim e the boiler should be

operating sm oothly with little change in the TSUL emissions. For most o f this data set,

this is true. But between facts 150 and 200, the TSUL em issions show two dramatic

spikes. In order to understand what caused the model to break down, the raw data were

re-examined.

The large TSUL em issions correspond to a period o f high boiler loading (black

liquor flow) and very low levels o f excess oxygen in the flue gas (Figures 3.7 and 3.8).

This exam ple dem onstrates one weakness o f the neural network model. It can only

predict the outcom e for situations for which it has seen adequate repetitions. While the

network learned the correct relationship between these inputs and the output neuron, only

a few exam ples o f these conditions were in its training base. Since the network had not

learned this relationship to the degree presented in the new data set, it responded too

weakly to the causes for the spikes. The network predicted only the very small increases

in TSUL visible in the predicted trace immediately underneath the two spikes. Inclusion

o f this unseen data to the training basis for the network should yield improved

perform ance under sim ilar circum stances in future data sets.


TSUL

(p

pm)

TSUL

(p

pm)

27

250

200

150

50

01 19 37 55 73 91 109 127 145 163 181 199 217 235 253 271 289

27.4

27.2

27

26.8

26.6

26 4

26.2

Fact

-TSUL — ■ LIQF

Figure 3.7 Liquor flow and TSUL em issions for the new data set.

24250

22200

20150 18

1610014

50

1001 19 37 55 73 91 109 127 145 163 181 199 217 235 253 271 289

Fact

-TSUL - •02

Figure 3.8 Percent O ; and TSUL em issions for the new data set.


% 02

*10

LIQ

F (g

pm/1

0)

C hapter 4 Conclusions

ANN techniques offer a viable method for modeling the com plex nature o f kraft

recovery boilers. The ANN m odels were validated by their confirm ation o f accepted

boiler physical and chemical processes. A num ber o f relationships between boiler

operating param eters and sulfur em issions have been recognized. These relationships

were used to form ulate several strategies for the reduction o f sulfur emissions.

Run in a real-tim e application, ANN models could help assure that mill operations

are proceeding in a predictable fashion. Discrepancies between the predictions o f a well-

established ANN model and the current boiler perform ance indicate unusual unit

behavior has occurred. An exam ination o f the incongruous data with the trained network

can be a useful tool for assessing underlying causes and providing guidance for avoiding

excess emissions. Furtherm ore, the network can be periodically updated by the inclusion

o f new data sets into its training basis. This allows for parameter evolution such as

equipment aging or changes in black liquor composition. A revised network can help

formulate a new operational procedure for the minim ization o f TRS and SO : emissions,

while maintaining desired green liquor and steam production.

Several im portant additions to the set o f input neurons could lead to the creation

o f an improved A N N model. These data were not available for the boiler studied. The

inclusion o f actual o r calculated black liquor droplet sizes should lead to better TSUL

predictions. Tem perature distributions and internal gas flow patterns would also be

useful in forecasting TSUL emissions. W hile these parameters are difficult to m easure in

the hostile environm ent o f an operating recovery boiler, theoretical constructs can be

28


29

em ployed to calculate their values for use as input neurons. The combination o f many

related predictive neurons into several ensembles has been shown to enhance the

predictive capability o f ANNs with respect to the sulfur emissions from the recovery

boiler [Opitz et al., 1999], Future efforts should be directed toward blending these ANN

and traditional modeling techniques. The resulting models will improve our ability to

accurately predict the outcom es o f highly com plex processes.


Pa r t B

In v e s t ig a t in g V o l a t il e O r g a n ic C o m p o u n d s in a n U r b a n In t e r - M o u n t a in V a l l e y U s in g a TD/GC/MS M e t h o d o l o g y a n d In t r in s i c

T r a c e r M o l e c u l e s

C h ap ters Introduction

5. 1 Significance o f Volatile Organic Compounds in the Troposphere

Atmospheric chem istry involves the study o f remarkably complex and interrelated

processes. Volatile organic compounds (VOC) play a significant role in these processes.

VOC as used in this thesis refers to any gas phase organic com pound found at low levels

in the atmosphere. VOCs have both biogenic and anthropogenic origins. These

chem icals possess atm ospheric lifetimes that range from seconds to years. Table 5.1 lists

the average tropospheric lifetimes o f a selection o f VOCs quantified during this study.

Table 5.1 Estim ated tropospheric lifetimes o f various organic chemicals.

i i Estimated Lifetime in the Presence of:I Chemical OH' (0.06 ppt) O ’, (30 ppb) NO 3 (l ppt) hv

Benzene 1 2 days 1

i Toluene 2.4 days 1 1.9 yearsm-Xylene 7.4 hrs 2 0 0 davs

I Acetone 6 6 days 38 days| Chloroform 0.55 yearsi a-P inene 3.4 hrs 4.6 hrs 2 . 0 hrsi (3-Pinene 2.3 hrs 1 . 1 days 4.9 hrsi p-Cym ene 1 . 0 days > 330 days 1.3 years! Limonene 1 . 1 hrs 1.9 hrs 53 minI M yrcene 52 min 49 min 1 . 1 hrs

3-Carene 2 . 1 hrs 1 0 hrs 1 . 1 hrs[Seinfeld and Pandis, 1998; Corchnoy and Atkinson, 1990; Finlayson and Pitts, 1986]

30


31

Researchers have demonstrated that chem icals with atmospheric concentrations,

often below one part per million, produce significant environmental consequences. The

em issions o f VOCs increase the atm ospheric concentrations o f CO and C O : through their

subsequent oxidation. Elevated levels o f CO: niay lead to global warming. Some VOCs

may cause acute or chronic health problems. Benzene, for example, has been identified

as a human carcinogen. Table 5.2 lists the health advisory levels for some o f the VOCs

tracked during this study. The oxidation products o f various VOCs in the troposphere

cause acute respiratory- problems. For exam ple, dark, gas phase experim ents have

dem onstrated that com m on monoterpenes readily react with NO? producing various

carbonyl compounds, which are lung irritants [Hallquist et a l ., 1999; Atkinson, 1990],

Table 5.2 Health advisory' levels for the VOCs quantified in this study, which are considered hazardous air pollutants by the EPA.

Chemical M SHA A C G IH ( TLV)

OSHA(PEL)

NIO SH(REL)

EPA(CRL)

Benzene 24 ppm 9.8 ppm 1.0 ppm 98 ppb 31 pptToluene 20 ppm 20 ppm 20 ppm 20 ppm

Ethylbenzene - 100 ppm 100 ppm - -

Xvlenes - 150 ppm 150 ppm 100 ppm -

Naphthalene 9.5 ppm 9.5 ppm 9.5 ppm - -

Chloroform - 10 ppm 2.0 ppm 2.0 ppm 8.2 ppt[UATW (EPA website). 1998]

Y1SHA - Mine Safety and Health AdministrationACGIH (TLV) - American Conference o f Governmental and Industrial Hygienists' threshold limit value,

a time w eighted average; the concentration o f a substance to which most workers can be exposed without adverse health effects

OSHA (PEL) - Occupational Safety and Health Administration s permissible exposure limit, a time weighted average; the concentration o f a substance to which most workers can be exposed without adverse effect averaged over an 8-h workday.

NIOSH (REL) - National Institute o f Occupational Safety and Health's recommended exposure limit.EPA (CRL) - Environmental Protection Agency's cancer risk level, a substance's concentration in air,

which an individual could breath over an entire lifetime and theoretically have no more than a one in a million increased chance o f developing cancer.


Terpenes will also react with hydroxyl radicals and ozone, resulting in a variety o f

carbonyl compounds [Grosjean et a l ., 1993], In a reaction described by Kotzias et al.

[1990], sulfur dioxide, ozone and common m onoterpenes reacted in the absence o f

sunlight to form sulfuric acid.

M ost VOCs are photochem ically reactive and play a critical role in the

tropospheric production o f ozone. Increases in tropospheric ozone threaten human

health, result in lower agricultural crop yields and may lead to global warming. On the

other hand, halogenated organic compounds are involved in the destruction o f

stratospheric ozone. Oxidation products formed from many o f these compounds

contribute to acid deposition. Terpene and arom atic compounds can form organic

aerosols [Calogirou. et a l., 1999], Increases in aerosol levels im pair visibility, cause

acute health effects and may cool the earth by reflecting solar radiation back into space

[Seinfeld and Pandis, 1998],

5.2 Sources o f V olatile Organic Compounds to the Atmosphere

The VOCs exam ined during this project can be divided into four categories:

aliphatic and aromatic, halogen containing, terpenoid and oxygen containing. The

aliphatic/arom atic category included the compounds: benzene, 2,3,4-trimethylpentane,

toluene, ethylbenzene, m-!p-xylene, o-xylene and naphthalene. The only halogenated

com pound followed was chloroform. a-Pinene, P-pinene,p-cym ene, limonene, 3-carene

and P-m vrcene com prised the terpenoid category. A cetone was the only chem ical in the

oxygen-containing category'. Anthropogenic and biogenic sources o f these compounds

are discussed in the following sections.


5.2 .1 Sources o j Alkane and Arom atic Volatile Organic Compounds

The largest contributors o f alkane and arom atic chemicals to the atmosphere are

automobiles [Seinfeld and Pandis, 1998]. Volatile organic compounds (VOCs) are

emitted from vehicles via exhaust and evaporative emissions. Automobile exhaust

contains both unbum ed fuel and volatile chemicals formed during com bustion

[Kirchstetter et al., 1999], In addition, the process o f refueling liberates the vaporized

fraction contained in the headspace over the liquid gasoline in an autom obile 's fuel tank.

Diesel fueled vehicles em it hydrocarbons mainly in the form o f fine particulate

polvcvclic arom atic hydrocarbons (PAHs) [W esterholm and Egeback, 1994],

Naphthalene, l-/2-m ethylnaphthalene and dimethyl naphthalene are the m ost abundant

PAHs in both diesel fuel and exhaust [Lee et al., 1992; Lowenthal et al.. 1994],

Naphthalenes are also com ponents o f gasoline and have been detected in the exhaust o f

gasoline powered autom obiles [K irchstetter et al., 1996; W esterholm and Egeback,

1994],

The Sm urfit-Stone Container pulp mill is a local industrial source o f these types

o f VOCs. Benzene, toluene, ethylbenzene, naphthalene and xylenes have all been

detected in the stack gases from different plant process origins, at varying levels

[Som eshwar et al., 1995].

The sm oldering com bustion o f wood and wood products results in the release o f

volatile hydrocarbons. One o f the predom inant volatile aromatic com pound associated

with com bustion gases is benzene [Zhang and Peterson, 1996], The em issions o f volatile

organic com pounds from biom ass com bustion are highly variable. They depend upon the

type o f fuel, fuel moisture content, am ount o f oxygen, tem perature and com bustion


34

phase. Hauk et al. [1994] found only traces o f benzene in the combustion gas evolved

from the burning o f beech wood.

5.2.2 Sources o f I 'hloroform

Chloroform is added to the atmosphere by direct emissions from a variety o f

sources. Estimates o f global chloroform emissions from the pulp and paper industry are

30 r 8 G g y r -1. while other industries account for 10 = 2 Gg vr '. The paper industry is the

major source o f chloroform to the global atmosphere through the bleaching process;

however, it is not the only significant source o f chloroform in the M issoula Valley. The

addition o f chlorine to potable w ater and wastewater results in the formation o f

chloroform and accounts for the release o f — 11 Gg yr to the global atmosphere.

Swim m ing pools treated with calcium hypochlorite will also emit chloroform to the

airshed [Aucott el a /., 1999], C hloroform 's low solubility and high vapor pressure favor

its partition from the aqueous into the vapor phase. The use o f sodium hypochlorite in

laundry bleaching also contributes to the formation and emission o f chloroform.

Natural processes may release significant quantities o f chloroform to the

atmosphere. Oceanic sources o f chloroform, while not important for Missoula, are

significant in the global budget. Chloroform concentrations measured by Frank et al.

[1991] were highest at m arine locations, relative to forested and urban sites in Europe.

Chloroform has even been m easured in anaerobically decom posing landfill wastes.

Animals, such as cows and other ruminants, which are large sources o f methane, emit

small amounts o f chloroform.


35

Trace quantities o f chloroform have been detected in autom obile exhaust and in

other combustion gases [Aucott et al., 1999], Furtherm ore, chloroform is used as a

solvent and an extractant in many laboratory procedures and is em itted to valley airshed

via fume hoods and fugitive emissions. Typical household products used in the Missoula

valley may contain chloroform. Sack et al. [1992] analyzed 1,159 com m on household

products from eight different categories: automotive products, household cleaners, paints,

fabric leather treatments, electronic cleaners, oils/greases/lubricants, adhesives and

miscellaneous products. Household cleaners, fabric/leather treatm ents and the

miscellaneous categories were found to contain the m ost chloroform.

5.2.3 So urces o j Ter penes

Most plant species emit volatile chemicals as part o f their natural lifecycle. The

principal hydrocarbon released by deciduous trees, agricultural crops and grasses is

isoprene. Coniferous trees em it a class o f com pounds known as terpenes. In the

M issoula Valley and western M ontana, coniferous vegetation and agricultural crops are

far more abundant than deciduous trees. Emissions from crops are estim ated to account

for approxim ately three percent o f the total U.S. biogenic em issions [Lamb et al., 1987]

and therefore, terpenes are the most important non-anthropogenic VOC in the valley.

Terpene emission rates for some common western M ontana conifer species range from

0.1 to 5.0 g tre e -1 day -l [Benjamin and Winer, 1998].

Plants em it a large variety o f terpene com pounds, however, each plant species’

em issions are dominated by relatively few specific chem icals [Juuti et al., 1990], The

two most dom inant terpenes emitted by coniferous vegetation are a - and P-pinene


36

[Isidorov et al., 1985; Helmig and A rey, 1992; Juuti et al., 1990; A ltshuller, 1983; Bertin

et a!., 1997; Konig et al., 1995].

The biogenic emission rates o f m onoterpenes from vegetation depend on several

factors such as, overall health, nutrients and water, physical disturbances and

temperature. The quantities o f terpenes stored in a p lan t's tissues are correlated with

em ission rates [Yatagi et a!.. 1995]. The am ount o f terpenes stored are a function o f the

genetics o f the particular plant species as well as long term nutrient and w ater availability

[Drewitt el al., 1998], Physical dam age to plant tissue provides open pathways for

terpene release. It is therefore not surprising that physical damage results in increased

terpene em issions [Drewitt et al.. 1998]. Although relative humidity has not been shown

to affect terpene emission rates. Lam b et al., [1985] did find that the em ission rates from

wet branches were approximately one order o f magnitude greater than dry branches. This

observance is in agreement with the generally accepted presumption that terpene

em ission occurs primarily via diffusion through the stom ata [Lindskog and Potter, 1995],

Ambient tem perature changes are the most overwhelm ing environmental factor

influencing terpene emission rates [D rew itt et al., 1998], Terpene em issions increase

exponentially with temperature [Lamb et al., 1985; Juuti et al., 1990; Fehsenfeld et al.,

1992], This dependence on tem perature explains the seasonal variations observed in

terpene em ission rates. Emissions peak in the warm sum m er months and are at their

lowest during the w inter [Yatagai et al., 1995; Staudt et al., 1997].

Sunlight does not seem to influence terpene em issions [Fehsenfeld et al., 1992].

Two factors are generally cited to explain the lack o f light dependence. The first is that

most m onoterpene production, although dependent on sunlight (photosynthesis), occurs


37

during initial needle development. Thus, conifers have large stable terpene reservoirs.

Since terpenes are always present, exposure to sunlight and the associated increase in

photosynthesis do not add appreciable quantities to the already large supply o f terpenes

stored in the plant's tissues. Staudt et al. [1997] presented the results o f a study that

contradict the non-dependency on light theory. Their results may be questionable since

am bient tem peratures were not constant during the experiments.

Processes that use wood as a raw material or fuel have the potential to release

large quantities o f terpenes. Terpenes have been detected in both the fugitive and stack

em issions o f pulp mills [Stromvall and Petersson, 1993] and a-pinene emission rates o f

over 200 lbs/hr have been m easured in digester re lief vapors and tall oil reactor vents

[Som eshwar, 1995], Biomass com bustion, such as forest fires or residential wood

burning, is an additional source o f terpenes to the troposphere.

5.2.4 So urces o f A celone

Acetone is emitted into the atm osphere from a number o f sources. The

breakdown o f lignin during the chemical pulping o f wood is known to release acetone

[Someshwar, 1995; Smook, 1992], Acetone has also been detected in the smoke given

o ff during biomass combustion [Hauk et al., 1994]. Direct biogenic em issions also

account for a portion o f the acetone found in the atmosphere [Guenther et al., 1994;

Isidorov et al., 1985; Lamb et a!., 1987],

The oxidation o f various hydrocarbons, such as isobutane, isobutene and propane

are believed to be a major secondary source o f acetone to the atmosphere [Arnold et al.,

1997]. Acetone has also been identified as an atmospheric oxidation product o f several


38

monoterpenes [Vinckier et al., 1998; Grosjean et al., 1993; Atkinson, 1990; Hallquist et

a!.. 1999],

5.3 Methods o f Source Apportionment

The apportionment o f atmospheric pollutants among different sources typically

involves either o f two modeling approaches. Source oriented models take point-source

emissions data, then using fluid mechanics and meteorological parameters, predict

pollutant concentrations at the receptor locations [Glen et al., 1996], Conversely,

receptor oriented models use chemical measurements at various sites to construct the

"best fit" linear combination o f source emissions necessary' to reconstruct the am bient

pollutant compositions found at the sites [Schaur et al., 1996], The chemical mass

balance (CM B) and related techniques are discussed in the following section. Various

tracer molecules are also described. The final section presents less conventional artificial

neural network applications to source apportionment and atmospheric modeling.

5.3.1 Traditional Apportionment Methods

There have been many modifications to the CMB model approach. For

example, multivariate receptor models have been designed that can predict the num ber o f

sources and their profiles without explicit source information. These models may include

chemical transformations occurring during transport. Principal com ponent analysis

models are multivariate receptor models capable o f source apportionm ent [Swietlicki et

al., 1996], Positive matrix factorization utilizes the known or estim ated errors o f the data


39

matrix to create boundary constraints, thus excluding negative factors in the result

[Paterson et a l 1999],

The basic chem ical mass balance has been applied to an immense num ber o f

source-apportionm ent situations. G ertler et al., [1996] used a CMB technique to

determ ine the relative contributions o f evaporative and exhaust emissions from

automobiles. Source compositions were determ ined by analyzing the headspace over

gasoline samples and literature values were used for tailpipe emissions. Their data

indicated that the tw o EPA automotive emission rate m odels under-predict evaporative

emissions. Fujita et al. [1995] applied a CMB m odel to determ ine m ajor contributors o f

VOCs to the San Joaquin Valley, California. Vehicle em issions and industrial oil

production were found to be the largest VOC sources. They also noted that previous

chemical em ission inventories overestim ated biogenic inputs, while underestim ating

motor vehicle contributions. Mukund et al., [1996] constructed a CMB m odel to

elucidate the spatial and temporal dependence o f VOC sources in Colum bus, Ohio.

Vehicle exhaust and evaporative emissions were determ ined to be the main VOC source,

especially during m orning and evening rush hours.

In one A ustralian study the relationship betw een light extinction coefficients and

aerosol com position was derived using linear regression techniques. Published source

profiles were then used to ascertain the fractions o f each species to the total aerosol mass.

A CMB was em ployed to apportion the particulate am ong relevant sources [Chan et al.,

1999],

Im provem ents to the basic source receptor model often involve m erging two or

more mathematical techniques. Henry et al. [1994] used a combination o f graphical


40

analysis and multivariate receptor modeling. The graphical ratio analysis for com position

estim ates (GRACE) method uses the slopes o f several linear relationships elucidated

from each scatter plot o f pollutant versus tracer molecule. In this way a series o f

constraints are created for use in the source apportionm ent by factors with explicit

restrictions (SAFER) model. This multivariate receptor model predicts the number o f

sources and their compositions. Due to the physical constraints placed on the matrix by

GRACE, negative values in the source contributions are avoided and a semi-intuitive

upper and lower limit for each chemical is obtained.

5.3.1.1 Tracers

The com parison o f a unique tracer m olecule to pollutant concentrations is one o f

the m ost straightforward and robust methods o f determ ining the chemical contributions

from a variety' o f sources. Lead em itted from gasoline combustion had long been an

excellent m arker for motor vehicle pollutants. The elimination o f lead from gasoline has

resulted in the need for new autom obile em ission tracers. The investigation o f tracer

molecules for a variety o f atm ospheric pollutants can help to target the most serious

sources o f pollution.

Grosjean et al., [1998] m easured the concentrations o f over 60 hydrocarbons in

the am bient air at Porto Alegre, Brazil. Scatter plots o f the concentration o f one

com pound versus that o f another dem onstrated that many o f these chem icals exhibited a

high degree o f correlation. This suggested a com m on source. Carbon monoxide and

acetylene, two chem icals not in gasoline but em itted in vehicle exhaust, were correlated


41

with several hydrocarbons. The resulting linear relationships from the scatter plots were

used to estim ate hydrocarbon contributions from vehicle exhaust.

The use o f elemental markers for m otor vehicle emissions was investigated by

Huang et al., [1994], Aluminum was found to be one o f the most common elem ents in

both fine (< 2.5 um) and coarse (> 2.5, < 10 um) particles. Antimony, zinc and bromide

were determined to be the most prom ising elemental markers, however, the authors

cautioned that their importance as tracers is likely to be minimal. This is because they

are emitted as large particulates and tend to quickly deposit on the earth 's surface.

Another problem with their use as tracers is that only very small quantities are em itted by

vehicles, but relatively large quantities are emitted from coal and oil-fired power plants.

The low polarity organic fraction was extracted from aerosol samples collected in

Albuquerque, New Mexico by Sheffield et al., [1994], Radiocarbon analysis linked

retene with softwood combustion emissions. Pyrene concentrations correlated well with

fossil fuel derived carbon. This would support its use as a motor vehicle marker. A

potential problem with its use, however, is its propensity to split its concentration

between the gas and particulate phases. Therefore, both particulate and gaseous samples

would need to be collected and analyzed.

Klouda and Connolly [1995] used radiocarbon measurements o f "clean air" CO,

wood burning CO and fossil fuel burning CO to determ ine the contributions these sources

made to CO concentrations in Albuquerque, New Mexico. In all cases, fossil fuel

derived CO dominated the urban samples.

Sim oneit et al., [1993] proposed using fine particulate aerosols as markers for

different types o f biomass combustion. They determ ined pine wood smoke emitted


42

mainly dehydroabietic acid, not found in hardwoods due to their lack o f resin acids. It

was also suggested that sterols could be used to partition aerosol contributions between

the burning o f plant and animal (meat cooking) materials. p-Sitosterol was found to be

the major phytosterol in smoke from pine and oak wood combustion, while cholesterol

was linked with algal detritus and faunal sources.

The purposeful release o f unique chemicals is another technique that has been

used when no intrinsic tracer molecule can be associated with a source. Lamb et al.

[1995] dem onstrated the utility o f sulfur hexafluoride as a tracer for isolated point

sources. Since the tracer release rate is predeterm ined, only the target analyte and SFr,

downwind concentrations are needed to calculate the emission rate. The method used

sulfur hexafluoride release coupled with real-time analyzers to measure methane

emission rates from a variety o f sources, w hile correcting for contributions from

combustion by simultaneously m easuring C O ; levels.

5.3.2 Artificial Neural Networks

Artificial neural networks are a recent addition to the list o f tools used for solving

source apportionm ent problems. One o f the most complex atmospheric contam inants to

apportion is particulate matter because it is often defined by size as well as chemical

composition. As a result, battles over the source o f particulates exact a considerable

effort and expense. W ienke and Hopke [ 1994] have designed a visual neural network

mapping technique for locating airborne particle sources. It involves projecting a P rim 's

minimal spanning tree into a trained K ohonen neural network. The result is a two-

dimensional neuron map, which is visually interpretable to the geography and industry


43

around a sample location. Airborne particulate samples near G ranite City, Illinois were

collected and elementally fingerprinted using X-ray fluorescence spectroscopy and

neutron activation analysis. Each sam ple fingerprint was represented by 48 elem ental

concentrations. The resultant map was in close agreement with studies based on principal

com ponent analysis and chemical mass balance techniques. In a subsequent paper,

W ienke et al. [1994] successfully applied the same technique for particulate source

apportionm ent to a multiple sampling site network in Southern California.

Song and Hopke [1996] apportioned particulates among nine sim ulated sources

using a neural network to solve the classical chemical mass balance. The input pattern

consisted o f ambient aerosol concentrations and the output pattern was the relative

fractions o f speciated particulates at each source. The resulting netw ork gave reasonable

predicted source contributions when applied to simulated aerosol com position data

obtained from the National Bureau o f Standards.

Reich et al.. [1999] employed a standard back-propagation artificial neural

network to apportion daily SO: concentrations between three main sources. They used

daily meteorological data and SO; levels a t one receptor location to estim ate the em ission

rates at the sources. The results were excellent for the specific types o f w eather patterns

used in training, but were poor w'hen the meteorology differed from the training set.

Source oriented models have a lso been created using artificial neural networks.

Boznar et al. [1993] built a short-term prediction model for am bient levels o f SO; in the

com plex terrain that surrounds the Sostanj coal-fired power plant in Slovenia. This

model solved shortcomings observed in earlier attempts, which were based on traditional

physical and statistical approaches to atm ospheric dispersion. Rege and Tock [1996]


44

developed a neural netw ork model to estimate hydrogen sulfide and am m onia release

rates. Emission rate predictions were based on wind speed, atm ospheric stability,

ambient tem perature, relative humidity, downwind distance, crossw ind distance, and

downwind concentration. The resulting model predictions were within 20% o f the

measured emission rates.

Neural netw ork models that include chemical reactions and source information

are proving useful in m odeling tropospheric ozone production. The formation o f low-

level ozone is a recurrent problem in industrialized urban areas. Adequate modeling o f

ozone has evaded atm ospheric researchers due to the uniqueness o f each location and the

complexity o f the m eteorological and photochemical variables that lead to ozone

formation. Yi and Prybutok [1996] developed a neural network to predict the daily

maximum ozone concentration in the Dallas-Ft. Worth area that has dem onstrated

excellent results when com pared to actual measured ozone levels. The neural network-

based model was com pared to two traditional models: linear regression and Box-Jenkins

autoregression integrated moving average (ARIMA). The network model exceeded the

accuracy o f the regression approach by one and h a lf orders o f magnitude and the Box-

Jenkins method by a factor o f four.

The ozone forecasting capabilities o f neural network based models have been

extended by Roadknight et al. [1997] to predict the com plex interactions between ozone

pollution and subsequent plant damage. Their work provides a direct way to forecast a

pollutant's effect on the econom ics o f crop production. Principal com ponent analysis

was used to improve the network model. The neural netw ork/principal com ponent

analysis com bination was able to achieve correlation coefficient o f 0.770 on test data


45

versus 0.544 for simple linear regression, 0.591 for m ultiple linear regression and 0.459

for non-linear regression.


Chapter 6 Experimental M ethod

6.1 Sampling Design

Two groupings o f air sam pling sites were established to investigate the

relationship am ong VOCs near the mill and two likely recipient vectors. All o f the

sample sites w ere chosen for their specific geographic locations, security and uniqueness.

The number o f sites was limited to five in order to reduce the expense and tim e required

for sampling. The sampling sites form a general line from the mill southeastward toward

the Sapphire M ountains. W ithin this general layout, sites 3, 4 and 5 describe a line that

begins at the mill and follows the Clark Fork/Bitterroot R iver corridor, where a visible

fog band from the mill often travels. Sites 1 and 2 are offset to the east in the Missoula

urban airshed, with site 2 situated sufficiently close to H ellgate Canyon to experience

periodic influence from the Clark Fork River valley. The physical relationships among

the five sam pling sites can be seen in Figure 6.1.

46


47

w rV '- F' T'SST H

V ’ \ - L

SSC Puln Mill

JU>’ T IM « >

■ ' • t l v - .

unanu.

v ‘‘v .

r i - v

rani -

■naouu^-- / MONTANA•• • ' — k f > s N ^ 1- ' '

•— ^ X p S ^ - S ----------------:

• , Y~

sat»

CENTRAL ^ MISSOULACOUNTY

- / > -

Figure 6.1 M ap o f M issoula Valley indicating the sam pling locations.

6 .1 .1 Sampling Locations

The following site descriptions are arranged in order beginning at the mill and

proceeding eastward. The site num bering schem e was dictated by the usual order o f

setup and collection. The placem ents o f the sampling apparatus were chosen with the

intention o f m aintaining uniform ity am ong the five sites. Therefore, each sam ple was

collected at a range o f four to six feet above the ground surface.

Site 3 (Sm urfit-Stone C ontainer Corporation, A m bient Station 1 A) (Figures 6.2

and 6.3 ) is w ithin the mill property boundary. This site is part o f the EPA aerom etric

information and retrieval system (AIRS). A group o f autom ated instrum ents track

several param eters important to valley air quality. Hydrogen sulfide and N O x are

measured and reported as hourly averages. Ten-m icron particulate m atter (PM -10) is

also m easured as a daily average every three to six days, dependent upon the season. The


48

site has its own w eather station, recording tem perature, w ind speed and wind direction as

hourly averages.

Figure 6.2 Site 3; with PM-10 sam pler and w eather station visible.

Figure 6.3 The Smurfit-Stone Container Pulp M ill viewed from site 3.


49

Site 4 (Coffey residence) (Figure 6.4), is located at the intersection o f Kona

Ranch and Big Flat roads, about five miles south-southeast from the mill. The mill had

received several complaints from the C offey's who claimed their respiratory problems

stem m ed from mill emissions. Smurfit-Stone Container requested the inclusion o f this

site in our study in order to ascertain which chemical or chemicals could be responsible

for the reported respiratory irritation. This sampling location com prises the middle

station on the sampling line that reaches from the mill in a southerly direction toward the

Bitterroot Valley.

Figure 6.4 Site 4; near intersection o f Kona Ranch and Big Flat Roads.

Site 5 (Riley residence) (Figures 6.5 and 6.6) is on Big Flat Road, near the

intersection with River Pines Road. Nearly four miles from site 4, ju s t w est o f Kelly

Island, this site is above the C lark Fork River, surrounded by pine trees, on the edge o f


50

the B itterroot M ountains. The forested location provides a com parison betw een

vegetative and mill emissions.

Site 1 (W robel residence) (Figure 6.7) lies between site 5 and the bulk o f

Missoula. Located a few blocks east o f Reserve Street, this site represents the western

edge o f urban M issoula.

Figure 6.5 Site 5; directly o ff Big Flat Road.


51

Figure 6.6 Site 5; note proximity o f sampling device to conifers.

Figure 6.7 Site 1, located on the western edge o f urban Missoula.


52

Site 2 (W ard residence) (Figure 6.8) is located at the eastern edge o f the urban

airshed. This sam pling location is the farthest from the mill. Its proxim ity to Hellgate

Canyon periodically exposes the location to strong easterly winds.

Figure 6.8 Site 2; near the intersection o f Front and M adison Streets.

6.1.2 Sampling Frequency

A sampling frequency o f one run per week was attempted. A sam pling run

consisted o f pumping all five sites for an ~-8 hour period during the same day. Sampling

was begun in the m orning at approximately 7:30 a.m. and curtailed around 4:00 p.m.

Interruptions in the sam pling frequency were due to weather and scheduling conflicts.

Sampling was suspended on very wet days, as extreme humidity interferes with analyte

sorption.


6.2 Sample Collection M ethodology

Samples were collected using SKC Model 222-3 air sam pling pum ps (Eighty

Four. PA) with flow rates o f -9 5 m L/m in ± 10% [Bromenshenk et al., 1998]. Target

analytes were sorbed onto Supelco Carbotrap™ 300 air sampling tubes (CMS tubes).

Each sampling pump flow was calibrated at the onset o f the sam pling period using a

BIOS DryCal primary flow m eter (Pom pton Plains, NJ). The flow rate was re-verified at

the end o f the sampling period a n d the average o f the initial and final flow rates was used

in subsequent calculations. Sam ple collection at site 1 was conducted in duplicate. This

provided a consistent measure o f th e random error associated with the sampling

methodology.

Each CMS tube was com posed o f three different carbon sorbents separated by-

small plugs o f silanized glass w ool. The first bed contacted by incom ing sample air was

300 mg o f 20/40 mesh C arbotrap™ C, a graphitized carbon black adsorbent with a 10

n r/g ram surface area. This layer w as efficient at trapping and releasing the largest

molecules, those with nine or m ore carbon atoms. The second contact layer was

composed o f 200 mg o f 20/40 m esh Carbotrap™ B, graphitized carbon black adsorbent

with a 100 m7gram surface area. T his layer trapped and released m olecules containing

as few as four or five carbons and a s many as 12 carbons. The final sorbant layer was

125 mg o f 60/80 mesh C arbosieve™ S-III, a spherical carbon m olecular sieve possessing

an 820 m‘/gram surface area as w ell as 15 to 40 angstrom pores. The Carbosieve™ layer

trapped the smallest organic m olecules. This combination o f adsorbents/absorbents has

proved superior for the concentration o f low level, low polarity, volatile contaminants

[Helmig and Greenberg, 1994].


54

Each CM S tube was conditioned before each use by flushing with -6 0 m L/m in o f

purified nitrogen gas and heating for 20 m inutes at 350° C. Cleaned and conditioned

sam ple tubes were sealed in individual storage vials and then placed in glass ja rs with

Teflon-lined lids. Storage vials holding field sam ples were placed in the sam e type ja rs

separate from unused tubes, in the dark at tem peratures below 4°C. The m ethod is a

variant o f EPA M ethod TO-2, which has been approved for the analysis o f volatile

com pounds in am bient air for more than 15 years [Keith, 1995],

The analytes o f interest have sorption/desorption efficiencies ranging from 99.3 to

100.5 % with the sorbents utilized in this study [Hazard, 1994], A com parison o f our

methodology with those o f several other studies indicated our procedure would also

produce acceptable results for the analysis o f terpenes [Larsen et al., 1997; H elm ig and

Arey, 1992; Juuti et al., 1990; Isidorov et al., 1985], O ther researchers have em ployed

sim ilar sorption tube techniques in order to characterize headspace contam ination over

industrial waste tanks, with acceptable results [Cheng et a l., 1997], Thermal

Desorption;Gas Chrom atography/M ass Spectrom etry (TD/GC/M S) has been used to

exam ine various pollutants, including some terpenes, in indoor air [Heavner et al., 1992].

One study used a TD/GC/M S method to identify nicotine as a tracer for office air

pollution resulting from cigarette smoke [Bayer and Black, 1987], Jenkins et al. [1996]

used sim ilar pumps with flow rates greater than 500 mL/min to determine environm ental

tobacco smoke exposure among nonsm oking individuals in an eight-hour workday. A

Belgian air pollution study used a related m ethod to quantify vehicle em issions in a

heavily used traffic tunnel [Fre et al., 1994],


55

6.3 Thermal Desorption/Gas Chromatography/Mass Spectrometry Methodology

The analytes were rem oved from the CMS tubes by thermal desorption. For this

step in the procedure a Dynatherm M TDU model 910 therm al desorption unit (Kelton,

PA) was employed. A T ekm ar LSC 2000/ALS 2016 liquid sam ple concentrator

(Cincinnati, OH) was used to focus the sam ple stream prior to its introduction onto the

column head. Analyte separations were perform ed on a Hewlett Packard 5890 series II

gas chromatograph equipped with a Restek RTX 502.2 capillary colum n (Bellefonte,

PA). The instrument configuration allow ed the GC/MS to handle three methods o f

sam ple injection: direct syringe injection, thermal desorption, and purge and trap without

any significant modifications. Helium flows and the type o f injection port liner were the

only alterations needed to sw itch am ong the three sample injection modes. The following

three tables outline tem perature settings, gas flows, colum n information and other

specifics used in this m ethodology. Tables 6.1 and 6.2 list the various parameters for the

thermal desorption and the purge and trap units, respectively. Table 6.3 describes

specific chromatograph and m ass spectrom eter parameters.

Table 6.1 Param eters for thermal desorption unit operation.

Dynatherm MTDU M odel 910 ParametersGas: HeliumPurge Flow: 25.0 m L/m inutePurge Time: 4.0 m inutesD esorb Time: 10.0 minutesD esorb Tem perature: 250° CC ooling Time: 6.0 m inutesInterface Tem perature: 120°CT ransfer Line Tem perature 120° C


56

Table 6.2 M ethod param eters for the purge and trap unit.

Tekmar LSC 2000 - ALS 2016 ParametersMethod: 2Standby Temperature: 30° CPurge Tim e: 16.00 minutesDry Purge Time: 4.00 minutesGas. HeliumPurge Flow: 15.0 mL/minuteDesorb Preheat Tem perature: 260° CDesorb Mode: 6.0 minutes (a), 260° CLSC 2000 Transfer Line/Valve Temperatures

Line: 125° CValve: 125°C

ALS 2016 Transfer Line/Valve TemperaturesLine: 120°CValve: 120°C

Table 6.3 Specific chrom atograph and mass spectrometer parameters.

Chromatograph and Mass Spectrometer Parameters !

M ethod: AIR.M j

j Inlet Temperature: 220° C j

1_______ D etector Temperature:________ 270°____ C___________i_______ Injection Mode:______________ Split_________________

Split Time:___________1.00 minuteSplit Flow: 20.0 mL/minute

j Septum Purge Flow: 3. mL/minute| Oven Programi Initial Temperature: 40° C

Initial Hold: 5.0 minutesRam p Rate: 5° C/minuteFinal Temperature: 220° CTotal Run Time: 50.0 minutes

Colum n ParametersColum n Type: Restek RTX 502.2Diameter: 0.32 mmLength: 60 metersFilm Thickness: 1.8 umCarrier gas: HeliumFlow: 1.0 mL/minute

D etector ParametersSolvent Delay: 5.0 minutesM ass Range: 35 - 260 m/z


57

Reagent grade standards for chem ical quantitation o f six com pounds were

purchased from Aldrich and Supelco in both solution and neat forms. Neat com pounds

were dispensed by volume, and their densities were utilized to create a working solution

with a concentration o f -2 ,0 0 0 ug/mL. Table 6.4 lists the chemical analytes that were

purchased in neat form.

Table 6.4 Chem icals purchased from A ldrich in neat form.

/V eat Chemicals-------------2,3,4-T rim ethy 1 pentane p - M yrcene

3-Carene ar-PineneLim onene | p- Pinene

The compounds listed in Table 6.5 were purchased as certified solution mixtures

for environmental analyses. In addition to those chemical analytes listed in Tables 6.4

and 6.5, acetone was purchased from Restek in a certified solution at a 5,000. ug/mL

concentration in methanol. The calibration standards were created by com bination and

dilution o f the working standard solutions.

Table 6.5 Compounds purchased from Supelco as certified solutions in methanol.

Supelco VOC M ix I Supelco VOC M ix 2 Supelco VOC M ix 4Chlorobenzene Benzene 1,1 -D ichloroethane

/7-Xylene Toluene 2,2-Dichloropropaneo-Xylene Ethylbenzene Chloroform

Isopropylbenzene m-Xylene Brom ochlorom ethane/7-Propylbenzene Styrene 1,1,1 -T richloroethane2-Chlorotoluene Bromobenzene Carbon tetrachloride4-Chlorotoluene 1,3,5-TrimethyIbenzene Dibrom om ethane

/e/-/-Buty (benzene 1,2,4-Trim ethylbenzene T etrachloroethenevec-Buty 1 benzene /7-Isopropyltoluene (77-Cym ene) Brom oform

1,3-Dichlorobenzene tf-Butylbenzene1,4-Dichlorobenzene 1,2,4-T richlorobenzene1,2-Dichlorobenzene Naphthalene

1,2,3-Trichlorobenzene----------- ----------- ------------


58

6.3.1 Instrument Calibration

An apparatus was designed and fabricated that simplified the process o f creating a

calibration curve for use in quantifying TD/GC data (Figure 6.9). In addition, it

presented a means through which an internal standard could be added to each field

sam ple prior to thermal desorption. The introduced compounds were flash volatilized at

125°C within a deactivated quartz "tee" and swept onto a sample collection tube at room

tem perature by purified nitrogen gas.

N 2 SupplyElectronicTherm om eter

-9 volts supplied

Sam pleinjection

Flow m eter r "Flow control C3

L w J

V ariable Power Supply

H eated D eactivated Q uartz Tee

CM S T ube

Figure 6.9 Diagram o f CM S tube calibration device.

The outside o f the quartz tee was w rapped with chromel wire. The spacing

between the coils o f wire was varied so that near the sample introduction area coils were

close together (-1 /1 6 inch apart), but tow ard the exit o f the tee, coils becam e widely

spaced (-1 /2 inch). This provided a hot spot within the tee at the analyte introduction

point for rapid volatilization, but allow ed for the exit end o f the tee to rem ain cool. The


59

tem perature o f the quartz tee was controlled by a variable voltage power supply. The

internal surface tem perature o f the volatilization area o f the tee was maintained by

determ ining the relationship to the temperature immediately adjacent to its outer surface.

Both the tem peratures o f the tee and the exiting gas were m onitored by thermocouples

connected to an electronic thermometer. The tee was flushed with nitrogen gas at a flow

rate o f 55.0 to 60.0 mL/min, monitored with a Restek Veri-Flow 500 electronic flow

meter. This flow rate allowed a consistent flash volatilization temperature within the tee,

and a gas exit tem perature cool enough (< 30° C) to ensure efficient analyte sorption

within the sample tube.

The standard solutions were injected through a septum (Supelco Thermogreen™

LB-2, 10mm) fitted into a union at the top o f the tee. The CMS tube was attached to the

terminal end o f the tee by another union. The unions were m anufactured from Teflon, as

its low conductivity coupled with its high degree o f inertness were properties desired for

this application.

Calibration standard tubes were created by injecting 1.0 uL o f an internal standard

solution containing o f 100. ng/uL o f fluorobenzene (internal standard) plus 4-

brom ofluorobenzene and 1,2-di-dichlorobenzene (system monitoring compounds) into

the equilibrated calibration device. This was followed by the injection o f 1.0 uL o f the

standard solution containing each analyte. The tube rem ained on the calibration device

for 5 minutes to allow a sufficient volume o f nitrogen (275-300 mL) to completely flush

all the analyte from the tee through the sample tube. This sequence was repeated for the

remaining standard concentrations, always beginning at the least concentrated and

finishing with the most concentrated solution. M ethod blanks were prepared by injecting


60

1.0 uL o f the internal standard solution and 1.0 uL o f pure m ethanol onto the sample

tube. Standard and method blank tubes were run on the TD/GC/M S using the identical

method as was used to analyze field samples.

Fifteen o f the 40 analytes present in the standard mixes were examined in detail.

Table 6.6 lists those 15 compounds regularly tracked by this study and the minimum

instrum ent response (MIR) necessary for their reliable identification and quantitation.

The associated minimum detection limit (MDL) was calculated using the average volume

o f air collected while sampling and the average ambient air tem perature for the sampie

days.

Table 6.6 M inimum instrument response and resultant detection limit for all analytes inthe quantitation database.

Analyte M IR (ng) M DL (ppt)Acetone 1.00 10.2

Chloroform 2.00 10.0Benzene 1.00 7.60

2,3,4-T rimethylpentane 2.00 10.4Toluene 1.00 6.50

Ethvlbenzene 1.00 5.60m- & ^-X ylene 1.00 5.60

o-Xylene 1.00 5.60a-Pinene 1.50 6.60

0-M yrcene 5.00 21.80-Pinene 1.50 15.4

/>-Cymene 1.00 4.4Limonene 1.50 6.603-Carene 5.00 21.8

Naphthalene 1.00 4.6

A six-point calibration curve was created for each com pound in the quantitation

database using concentrations o f 5.00, 25.0, 50.0, 100., 200. and 400. ng/uL. The system

m onitoring compounds in the internal standard solution provided a second quality check


61

on the quantitation. T he mass spectrom eter perform ance was verified each day by

evaluating the spectrum produced from the direct injection o f 25 ng o f 4-

brom ofluorobenzene. The instrument perform ance and all calibration curves were

verified daily by the analysis o f a CMS tube spiked with 100 ng o f each analyte.

Instrum ent calibration was performed once per m onth o r when the trauma

inflicted by dirty honey bee samples upon the G C/M S rendered it necessary. System

blanks, m ethod blanks, trip blanks, and calibration checks were utilized to ensure a

maximum achievable level o f QA/QC. Each sam ple tube was spiked with 1.0 uL o f the

internal standard solution prior to analysis. Q uantitation o f the target analytes was based

upon the internal standard response as well as the target response relative to the

calibration curve, as described by the following equation.

Quantity o f target analyte (ng) = ( A t C i s ) / ( A i s R F ) (6.1)

where:A r = Area o f characteristic m/z for the target analyte Cis = M ass o f the internal standard (ng)Ais = Area o f characteristic m/z for the internal standard RF = R esponse factor for target analyte

and

RF = ( A t C , s ) / ( A , s C t ) (6.2)

where:C r = M ass o f target analyte (ng)


62

6.3.2 Analytical Uncertainty•

T he relative uncertainty for each analyte was determ ined by statistical analysis o f

replicate, mid-range (200. ng) standards. These uncertainties are listed in Table 6.7. The

m inimum relative uncertainty was 3.50% fo r 2,3,4-trim ethylpentane, while the maximum

uncertainty was 8.95% for (3-pinene. The average uncertainty for all analytes studied was

5.81%. O nly one value for the percent relative uncertainty was used in subsequent

calculations, the maximum o f 8.95%. This was done in order to m ake calculations

sim pler and to represent the maximum error expected from the analytical methodology.

T able 6.7 The relative uncertainties associated with the analytical method.

| A n a l y t e

\

A c t u a l

M a s s

A v e r a g e

M e a s u r e d M a s s

S t a n d a r d

D e v i a t i o n

C o n f i d e n c e

I n t e r v a l (95%)% R e l a t i v e

U n c e r t a i n t y

! Acetone 200 198.21 16.97 1 1.26 5.68! Chloroform 200 203.21 15.69 10.41 5.12i Benzene 200 205.14 16.51 10.96 5.34

2.3,4-Trimethylpentane 176 183.62 9.67 6.42 3.50Toluene 200 207.05 12.65 8.39 4.05

! Ethylbenzene 200 208.71 14.60 9.69 4.64| m - & p - X ylene 400 432.48 36.39 24.15 5.58

o-Xylene 200 206.76 14.70 9.76 4.72oc-Pinene 214 218.52 14 38 9.54 4.37

i p-Pinene 180 193.42 22.94 15.23 7.87P-Myrcene 214 220.17 29.69 19.71 8.95

1 p-Cymene 200 198.37 20.10 13 34 6.73 !Limonene 204 198 63 15.57 10 33 5.20

| 3-Carene 193 205 81 22.64 15.02 7.30Naphthalene 200 201.39 24.77 16.44 8.16

Both the percent relative uncertainty in the analytical m ethod and the error

associated with the volume o f air collected during the sam pling period (± 10%) were used

to calculate the overall percent error associated with the study. As a result, all

concentrations reported have an uncertainty betw een 10.6 and 13.4 percent.


63

6.4 Statistical Analysis

Several statistical methods were applied to the refined data set in order to

determine the significance o f observed trends above the variation introduced by random

error. First, frequency histograms were plotted to identify which type o f population

distribution would best represent each specific data set. As expected, the sample

populations had a degree o f nonnormal ity. The nonnorm ality o f the sample distributions

was characterized as being skewed to the right. W henever feasible, the nonsymmetrical

F distribution was selected to be a more accurate representation o f the population

distribution. The mean, median, minimum, maximum, variance, standard deviation and

range were then calculated for each parameter at each sam ple location. The degree o f

skewness in the concentrations for any individual com pound can be assessed by the

difference between the mean and median.

The means for each o f the sampling locations were com pared using the Student's

t method. The W ilcoxon rank sum test was not used to com pare site means because the

validity criteria were met for the t method. The most im portant assumption required for

the t method was that each set o f samples was drawn from a different population. This

also assum ed the elem ents o f one sample site were unrelated to the other sample sites.

The final assum ption was that the population distributions were normal. Initially, it

appeared this assum ption would be invalid, forcing the use o f the W ilcoxon rank sum

test. However, since in all cases the combined sample sizes had n > 30, the Central Limit

Theorem was applicable and the sampling distributions could be considered normal [Ott,

1993],


64

The form ula used to com pare the means (Xa) o f two sets o f data is dependent

upon each set's variance. When the variances for each set o f data were similar, m eaning

they did not differ by more than a factor o f three, the pooled standard deviation m ethod

was used [Ott, 1993J. The pooled standard deviation (S) for both data sets was calculated

using the following equation.

S = [ i : ( X 1- X aI): - i : ( X J - X ^ r / C n , - n ; - 2 ) ] ' ' - (6.3)

The resultant value for the pooled standard deviation was then used in calculating a value

for Student's t.

t = [(Xai - Xai) / S ][(n ,n2)/(ni + n2)]1/2 (6.4)

The combined degrees o f freedom (df) were calculated as:

d f = ni - n2 - 2 (6.5)

The majority o f the data sets that w ere com pared had large differences between their

sample variances. Therefore, a second method was needed to compare the means

[Johnson and Leone, 1964], In th is case the value for S tudent's t was calculated using.

t = (Xai - Xa2) / [ ( S f / n,) + (S22 / n2)]1/2 (6.6)

The degrees o f freedom were calculated as:

d f = [ ( n , - 1 )(n2 - l ) ] / [ ( n 2 - l ) c 2 + (l - c ) 2 ( n , - I)] (6.7)

where:c = (S,2/ n,) / [(S,2 / n,) + ( S22 / n2)j (6.8)


65

The calculated t values were com pared to tabulated values, in order to determine the

probability that random error was, or was not, responsible for the observed difference

between the com pared m eans [Beyer, 1984].

6.5 Apportionment Calculations

A pportionm ent o f the volatile organic pollutants tracked during this study was

attempted using three different methods. The first method used o-xylene as a tracer for

other arom atic hydrocarbons. The second method employed o-xylene and p-cym ene as

reference com pounds to estim ate contributions to the valley airshed from two sources:

urban M issoula (vehicles) and the pulp mill, respectively. The third method used linear

com binations o f source profiles o f o-xylene and /?-cymene to determ ine the relative

contributions o f total arom atic chem icals from automobiles and the mill at each sam pling

site.

6.5.1 Apportionment Using o-Xylene Tracer M ethod

o-Xylene, an effective tracer for fuel/vehicle emission related compounds

[Zweidinger et a/.. 1990], was used to quantitate the mean contributions from the vehicle

fleet to the valley airshed. A ir sam ples were collected at the intersection o f South Ave.,

Russell and Brooks Streets (M alfunction Junction) to be representative o f M issoula's

vehicle fleet. Traffic volum e at M alfunction Junction assured the predominance o f

vehicle em issions to the surrounding air. Sampling at M alfunction Junction was

performed in the early sum m er and restricted to days when the w eather patterns assured

good air m ovem ent. In addition, sam ple days where chosen so as to avoid collection


66

when forest fires, fireplace woodsm oke, prescribed bum s and stagnant air episodes would

have significantly contributed or altered the levels o f volatile organic pollutants within

the airshed. Concentrations o f benzene, toluene, ethylbenzene, m/p-xylene and

naphthalene were plotted against the concentration o f o-xylene. A least squares fit was

used to extract a linear correlation factor between each analyte and o-xylene. The mean

concentrations o f the five pollutants at each site were subsequently com puted by

m ultiplying the mean o-xylene concentration by the correlation factor.

y = mx (6.9)

where:y = Predicted concentration o f analyte, pptm = Slope o f line (correlation factor)x = M ean concentration o f o-xylene m easured at site, ppt

6.5.2 Apportionment ik in g "Dispersion" Reference M ethod

The second set o f apportionm ent calculations was perform ed using o-xylene and

/7-cymene as "dispersion" references. W ith o-xylene, for exam ple, its concentration at a

sam pling site was divided by its concentration at M alfunction Junction to produce a

"dispersion factor." The predicted mean value for all o ther contam inants, assum ed to

com e from urban M issoula air o f a sim ilar com position, were com puted by multiplying

the M alfunction Junction level by the dispersion factor. Analogously, samples

representative o f a ir parcels in the vicinity o f the Sm urfit-Stone C ontainer Pulp Mill were

collected -3 0 0 meters southeast o f the waste fuel boiler. p-Cym ene, the most unique

terpene from the mill, was used to com pute m ill-based dispersion factors.


67

Estimated quantity o f analyte from source, ppt = ( T .s /T r )A r (6.10)

where:T s = Mean concentration o f the tracer com pound at site o f interest T r = Concentration o f the tracer at source reference A r = Concentration o f the analyte at source reference

6.5.3 Apportionment Using Linear Combinations

The third method o f apportionm ent used linear combinations o f source profiles

obtained from Malfunction Junction and the Smurfit-Stone Container Paper Mill samples

(described in the two previous sections) to estim ate the amounts o f total arom atic

chem icals attributable to vehicles and the mill at each site. For each sam ple site, the

mean quantities o f benzene, toluene, ethylbenzene, m-ip-xylene, naphthalene, o-xylene

and /7-cymene were summed and term ed total aromatic compounds (TAC). It was

assum ed that the TACs originated from vehicles, the mill and other sources.

[TACJtotal = [TAC]vehicles + [ T A C ] m,II -r [T A C ]other (6 . 1 1 )

o-Xylene concentrations were dom inated by vehicle emissions, however, the mill and

other sources contributed some quantities to the airshed. The mill was the m ost

important source o f /7-cymene to the valley airshed, but natural em issions and the

presence o f other industries utilizing wood as a raw material may have contributed

additional small quantities. The total concentrations o f each tracer were expressed by the

following two equations.

[/7-cymene],oUli = [/7-cymene]maUjunc + j/7-cymene]min (6.12)

[o-xylene],oiai = to-xylene]^^,™ + [o-xylene]m,n (6.13)


68

The portion o f the TACs due to vehicle emissions, the mill and others were represented

by V, M and O, respectively.

[TAC] = V + M + O (6.14)

The ratios o f the TACs to o-xylene and /7-cymene at both sources were substituted into

equations 6.12 and 6.13.

j/?-cymene]tomi = { ( [T A C ^ p -c y m e n e ] )^ - ,^ V} +

{([TAC]/[/?-cymene])miiiM} (6.15)

[o-xylene]totai = ‘( [ T A C l/ to - x y le n e ] ) ^ ^ V] -r

!([TAC]/[o-xylene])miu M{ (6.16)

This algebraic system o f two equations and two unknowns was then solved for each

sampling site using the mean concentrations o f p-cym ene and o-xylene. The values o f V

and M were then used along with the mean TAC concentration from each site to solve for

O in equation 6.14, other sources.


C hapter 7 Results

The first part o f the results section characterizes the early qualitative phase o f the

research. It details the developm ents that lead to the ideas, which ultim ately shaped the

final project. The remaining m aterials present the quantitative tables, the statistical tests

and the source apportionment predictions.

7. / Q ualitative Study

The list o f compounds follow ed in this study evolved through several stages. The

first phase o f this project was begun in the w inter o f 1995. During that tim e, periodic air

sam pling was conducted at many locations around the valley and in a variety o f settings

at the pulp mill. The resulting chrom atographic peaks were identified by m atching their

spectra with those found in the m ass spectral library database. Initially, as m any peaks as

possible were identified in an effort to determ ine the breadth o f chemicals present in the

airshed.

A large num ber o f com pounds were followed in order to determ ine w hich were

consistently present. Fifty-three w ere contained in the list o f 58 com pounds associated

with EPA m ethod 524 for analysis o f volatile organic compounds in drinking w ater by

purge and trap. A quantitation database was constructed for all 53, but the database

proved too extensive to feasibly digest. A careful study o f many sample sites w ithin the

valley provided a basis for trim m ing the quantitation database. Often, closely related

com pounds showed redundant trends from sample to sample. The quantitation database

was continually refined, as a greater fam iliarity with the airshed was achieved.

69


70

A dramatic difference was visible between samples collected from several valley

locations. A prom inent gasoline fingerprint was consistently displayed in samples

originating within the City o f M issoula (Figure 7.1). On the other hand, air samples

taken at or near the mill were dom inated by terpenes and carried low levels o f

organosulfur com pounds, which are associated with kraft pulping (Figures 7.2 and 7.3).

Table 7.1 compares the compounds detected with the highest frequencies and responses

at both the urban and mill sample locations.

! 1. ^ t ' u i u l d r i c w '

600000

s o o o o o

300000

200000

1 o o o o o

Figure 7.1 Chrom atogram o f an air sample collected at site 2 in urban Missoula.


71

t 3 00000

t 2 0 0 0 0 0

lSOCOOO

I oooooo

9 0 0 0 0 0

aooaoo

7 0 0 0 0 0

aocooo

5 0 0 0 0 0

4 0 0 0 0 0

3 OCOOO

200000 1OCOOO

iu L l_ L i— a l i l l. OO oo OO OS.OO 30 .00

J ill,3 * QO - + Q CO

Figure 7.2 Chromatogram o f an air sample collected near the evaporator hotwells.

Figure 7.3 Chromatogram o f an air sam ple collected -300 yards southeast o f the pulpm ill's waste fuel boiler.


72

Table 7.1 Predom inant compounds found in the city com pared to those found a t the mill.

Urban M issoula Sm urfit-S tone Container M il1Butane Ethanol

2-M ethylbutane PropanolEthanol AcetonePentane Dimethyl sulfideAcetone Acetic acid

2-Methyl pentane Dimethyl disulfideHexane 2-Butanone

2,4-Di methyl pentane a-PineneAcetic acid p-M yrcene

2,2,4-trimethyIpentane p-PineneBenzene a-Phellandrene

2,3,4-trimethyIpentane 3-CareneToluene a-Terpipene

Ethylbenzene Limoneneo-, m- & p-Xylene /7-Cymene

Propylbenzene P-Phellandrene1,3,5-T rimethylbenzene 1 -M ethyl-4-( 1 -methylethenyl )-

benzene1 -Ethyl-4-m ethylbenzene 2-M ethoxyphenol

Naphthalene Junipene

In an effort to narrow the quantitation list, likely sources for some o f the

com pounds were considered: gasoline and diesel engines, the mill, natural vegetative

em issions, etc. At this stage in the study, the determ ined objective was to apportion the

VOCs found in the valley airshed am ong tw o o r three main sources.

Based on initial observations, organosulfur com pounds seemed like reasonable

candidates as tracers for mill VOC emissions. The only major source o f these

com pounds to the M issoula airshed was believed to be the Smurfit-Stone C ontainer Mill.

W hile these types o f compounds are not abundantly studied, a similar analytical

technique reported a minimum detection lim it o f 2.4 ppt for dimethyl sulfide [Yokouchi


73

et al., 1993]. Unfortunately, these com pounds were at too low a concentration or were

too transient to be detected anywhere within the valley except on the pulp mill premises.

Several terpenes, lignin pyrolysis products and resin acids have been proposed as

tracers for specific plant classes in biomass com bustion [Simoneit et a l ., 1993]. Terpenes

were seen in both urban and mill a ir samples and seemed likely secondary candidates as

VOC tracers from the mill. A lthough they are em itted naturally from trees, their

concentrations and types are lim ited [Arey et al., 1995], During the pulping process,

large amounts o f terpenes are released. W hile investigating the use o f terpenes as tracers

for the m ill's VOC emissions, it was noted that /?-isopropyltoluene (/7-cymene) was found

in large concentrations at all o f the mill locations sampled. /7-Cymene was also found

throughout the valley in lower concentrations at sites farther from the mill.

Analytical measurements m ade a strong case for the overall im portance o f vehicle

emissions to the airshed from the very first round o f samples. Many o f the samples

analyzed were dom inated by benzene, toluene, ethylbenzene and xylenes. Zw eidinger et

al. [1990] had already com piled evidence for the use o f 2,3,4-/2,2,4-trim ethylpentane, n-,

rn-ip-x ylene, m ethylcyclohexane, 2- and 3-methylhexane, and 2-methyl pentane as tracers

for fuel/vehicle related emissions.

Finally, the compounds in the quantitation database were broken down into three

categories based on chemical sim ilarity and the theorized major source. Table 7.2

identifies the source categories and the com pounds associated with each. The list o f 53

had been pruned to include eight o f the com pounds, both o f interest and found with

regularity in the airshed. Added to this were five terpenes o f potential value, acetone and


74

2,3,4-trimethyIpentane. Quantitation o f the 15 final compounds formed the basis for all

samples analyzed after May 1998.

Table 7.2 Final list o f 15 compounds studied.

Terpenes F uel-Related M iscellaneousa-Pinene Benzene AcetoneP-Pinene Toluene Chloroform

p-M yrcene EthylbenzeneLimonene o-Xylene3-Carene m- & p-X ylene

/7-Cymene 2,3,4-T ri methyl pentaneN aphthalene

7.2 Quantitative Results

The results from the quantitation o f the am bient air samples, the subsequent

statistical analysis and source apportionment are listed in Tables 7.3 through 7.8. The

format o f each table is described in the following section.

Results from the quantitative am bient a ir study are summarized in Tables 7.3 and

7.4. Table 7.3 is com prised o f those com pounds that have been linked to automotive

fuels and vehicle emissions; Table 7.4 lists those com pounds that have possible sources

other than direct vehicle/fuel emissions. Data in both tables are organized in the sam e

manner. Each m ajor column heading represents a separate analyte. The column beneath

each analyte is sometimes subdivided into tw o entries. One column holds concentrations

expressed as parts per trillion by volume (ppt). The other column holds relative response

ratios (RRR). Relative response ratios were initially used for acetone, 2,3,4-

trimethylpentane, a-pinene, P-pinene and lim onene because they were not part o f the

original 53 compounds in the quantitation database. M ajor row headings represent either


75

individual sample sites, o r com binations o f sites that were grouped to characterize

broader regions within the M issoula Valley. Rows are further subdivided to delineate the

minimum, mean, median and maximum concentrations o f the analytes detected at each

site or site grouping.

Tw o types o f numerical representations were used to depict the concentrations o f

the analytes measured during this study, parts per trillion (ppt) and relative response ratio

(RRR). From January through July 1999, all o f the analyte concentrations were reported

as ppt, calculated as follows:

A nalyte (ppt) = [A / ((1 x 109 ng) X)J / [PV / RT] (1 x 1012) (7.1)

where:A = Mass o f analyte from quantitation database (ng)X = M olecular weight o f analyte (g)P = Average atm ospheric pressure (0.889 atm)V = Volume o f air sam pled (L)R = Gas constant (0.08206 L atm m o l'1 K '1)T = Average tem perature for sample day (K)

Since the initial phase o f this study was done qualitatively, simply to determ ine

what types o f chem icals w ere present in the valley airshed, standards for seven

compounds that were eventually included in the study were absent from the early

quantitation databases. These seven com pounds were acetone, 2,3,4-trimethylpentane, a -

pinene, P-pinene, P-m yrcene, limonene and 3-carene. The relative response ratio was

computed to provide a site-to-site concentration comparison among identical species in

samples collected between M ay 1998 and January 1999.

RRR = [(Aa / A IS) / V ] ( l x 103) (7.2)


76

where:A,\ = Area o f characteristic m/z for the analyteAis = Area o f characteristic m/z for the internal standardV = V olum e sampled (L)

The presence o f those chemical species in early samples was retroactively affirm ed by

com parison with standards purchased in January 1999. The calculation o f the RRR

values was m aintained throughout the study in order to preserve consistency. The

validity'' o f the RRR values were further substantiated by overlaying daily RRR plots with

ppt plots for the same analyte and com paring the trends observed in both. In all cases,

the difference between the RRRs and the concentrations expressed in ppt was simply a

scalar factor.

Three different site groupings were assembled in order to establish sub-regions

within the valley airshed. Each group, thereafter, was treated statistically the same as a

sam pling site. For example, the group labeled (1+2) in Tables 7.3 and 7.4 was com piled

from the data collected at both urban sites. Therefore, the m ean and the m edian were

calculated using a larger sample population. The use o f larger populations improved the

reliability o f the statistical analysis as the histograms o f the com bined data more closely

approached a normal distribution.

A com prehensive listing o f individual daily chemical concentrations for each

com pound at each site is found on the CD-ROM included with this text. The file is

entitled "M slaA irC om pounds.xlsT The results from the analysis o f system, m ethod and

trip blanks, as well as daily calibration curve verifications are found in the file

"A irQ A Q C .xlsT The entire ensem ble o f calibration curves constructed and used during

this study are located in the file “GCM S-M ethods.”


77

7.3 Statistical Results

Table 7.5 lists calculated Student's t values for com parisons among the six site

and/or group means. The columns are organized as previously described for Tables 7.3

and 7.4. Each row represents a comparison between tw o different sites or site groupings.

Statistically significant differences are coded in their typeface, as described at the bottom

o f the table, for the 90, 95 and 99 percent confidence levels.

7.4 Apportionment Results

The apportionm ent o f the volatile organic pollutants in the valley airshed was

accom plished using the three different methods that w ere described in the chapter 6.5.

Results from the o-xylene tracer technique are listed in Table 7.6. Results from

apportioning contam inants with o-xylene and £>-cymene "dispersion" factors appear in

Table 7.7 (fuel/vehicle-related compounds) and Table 7.8 (terpenes and chloroform).

The resulting apportionm ent o f total aromatic com pounds (TAC) using linear

com binations can be found in Table 7.9. Explanations o f the values reported in the tables

and the table formats are explained in the following section.

For Tables 7.6 through 7.8 each chemical quantified is divided into five rows,

which represent separate sampling sites or site groupings. The first column in each o f the

three tables is the actual measured mean concentration (ppt) for each chemical at each

site.

The second colum n in Table 7.6 holds the 'p redicted mean contribution from

vehicle em issions.’ These concentrations w ere calculated as described in the chapter

6.5.1. The following colum n labeled ‘% vehicle em issions’ was calculated by dividing


78

the 'predicted mean contribution from vehicles' by the 'm easured m ean,' (column

2/column 1) x 100%. The subtraction o f the 'predicted mean contribution from vehicles'

from the measured m ean’ resulted in the column entitled 'o ther contributions,' column 1

- column 2. The '% o ther' values were calculated by dividing 'o ther contributions' by

the measured m ean,' (colum n 4/column 1) x 100%.

The second colum n in Tables 7.7 and 7.8 list the predicted mean contribution

from urban M issoula,' calculated as described in the chapter 6.5.2. The '% from

M issoula' column was calculated by dividing the predicted mean contribution from

urban M issoula' by the 'm easured mean,' (column 2/column 1) x 100%. The 'predicted

mean contribution from SSC m ill' concentrations were calculated as described in the

chapter 6.5.2. The '% from m ill' values were determined by dividing the predicted

mean contribution from SSC m ill' by the measured m ean,' (column 4/column 2) x

100%. The total contribution from urban Msla & m ill' colum n is the summation o f the

% from M issoula' and '% from m ill' (column 3 + column 5).

Table 7.9 lists the estim ated percentages o f TACs attributable to vehicles, the mill

and other sources, obtained from the three methods o f apportionm ent described in chapter

6.5. Two additional variations on the apportionment m ethod described in chapter 6.5.1

were included as further validation. Based on published em ission factors, it was

calculated that 78.6 percent o f the o-xylene in an air sample, comprised o f equal volumes

o f vehicle exhaust and woodsm oke, would be due to the vehicle exhaust. Thus, this

fourth apportionm ent m ethod is a worse case scenario that could only occur under

extreme forest fire situations.


The fifth apportionm ent m ethod is a second variation on the method described

6.5.1. The analyte to o-xylene correlation factors were replaced w ith analyte to 2,3,4-

trimethyl pentane correlation factors. This variant was explored in an effort to draw a

comparison between the use o f an arom atic versus an aliphatic tracer molecule.



Site

Ben

zene

ppt

2,3,

4-T

rim

ethy

lpen

tane

R

RR

pp

t *

Tol

uene

ppt

Eth

ylbe

nzei

iepp

t

hi

- &

/>-X

ylen

e

ppt

o -X

ylen

e

ppt

Nap

htha

lene

ppt

1M

inim

um98

.80.

0771

< 10

.416

14.

0622

.3<5

.61

<4.6

4M

ean

464

3 24

57 9

946

146

475

165

38.2

Med

ian

301

2.19

48.8

711

87.2

295

103

33.7

Max

imum

1,33

015

.514

63,

990

712

1.88

068

313

62

Min

imum

249

0.00

557

< 10

.454

255

.133

.774

.47.

58M

ean

540

3.30

87.5

1.16

016

055

019

238

.9M

edia

n48

52.

272.

271,

020

140

480

171

36.6

Max

imum

1.25

013

.313

.32,

790

446.

51.

590

555

ii3

3M

inim

um33

.80.

00<

10.4

24.2

<5.6

1<5

.61

<5.6

1<4

.64

Mea

n17

00.

690

10.6

247

37.9

121

429

14 3

Med

ian

153

0.41

3<

10.4

212

28.2

82.7

30.5

10.4

Mvd

mum

387

’3.

2354

.51,

150

205

: •

693

248

85.0

4M

inim

um52

.60.

00<

10.4

10.6

<5.6

1<5

.61

<5.6

1<4

.64

Mea

n21

10.

846

14.3

307

388

128

45 3

12.1

Med

ian

173

0.42

913

.122

622

.470

.224

88

19M

aixi

mum

569

.5.

6147

.01,

220

255

935!

'

334

49.3

5M

inim

um25

.70.

00<

10.4

28.8

<5.6

17.

43<5

.61

< 4.

64M

ean

138

0.37

24.

8817

221

.866

024

012

.2M

edia

n12

40.

159

< 10

.414

217

.348

,618

67.

15M

axim

um35

22.

4920

.263

696

.428

4;10

472

.9(1

+2)

Min

imum

98.8

0.00

557

< 10

.416

14.

0622

.3<5

.61

<4.6

4M

ean

510

3.22

72.7

1.07

015

853

118

539

.8M

edia

n44

72.

2079

.086

910

840

713

434

5M

axim

um1,

330

15.5

146

3,99

071

21.

880

683

136

(3+4

+5)

Min

imum

25.7

0.00

< 10

.410

.6<5

.61

<5.6

1<5

.61

< 4.

64M

ean

174

0.64

110

.024

333

.010

537

612

9M

edia

n14

90.

322

< 10

.419

021

.366

.324

189

2M

axim

um56

95.

61:

54.5

1,22

025

593

533

485

.0(1

+2+4

+5)

Min

imum

25.7

0.00

< 10

.410

.6<5

.61

< 5

61<5

.61

<4.6

4M

ean

331

1.89

41 7

626

89 1

295

103

24.8

Med

ian

251

1.02

25.5

456

48.9

165

594

17.3

Max

imum

1,33

015

.514

63,

990

712

1,88

068

313

6C

once

ntra

tions

hav

e a

rela

tive

unce

rtain

ty o

f 13

.4%

V

alue

s ba

sed

on d

ata

colle

cted

fro

m 5

/98

- 7/9

9 o

* V

alue

s ba

sed

on d

ata

colle

cted

fro

nt 1

/99

- 7/9

9

Tab

le 7

.3

Sum

mar

y o

f alip

hatic

and

aro

mat

ic c

ompo

und

conc

entr

atio

ns.


Ace

tone

Chl

orof

orm

«//>

//«-P

inen

ebe

ta -P

inen

ep

-Cyn

tene

l.iin

oiie

neSi

teR

RR

ppt

*PP

‘R

RR

ppt*

RR

Rpp

t*PP

‘R

RR

ppt

*1

Min

imum

5.58

599

<9.9

70.

00<6

.55

0.00

< 15

.4<4

.43

0.00

< 6.

65M

ean

21.7

1.40

019

10.

855

8.23

0 37

73.

9814

40.

170

4 94

Med

ian

15.9

1,18

012

.40.

530

4.30

0 16

9<

15.4

112

0.06

20<

6.65

Max

imum

65.7

3,43

012

73.

8028

.01.

8816

,752

.91.

6143

.82

Min

imum

4.09

394

<9.9

70.

00<6

.55

0.00

< 15

.4<4

.43

0.00

<6.6

5M

ean

25.5

1,52

029

.03

1622

.80.

425

3.36

18.1

0.12

9<

6.65

Med

ian

17.1

1,16

014

.40.

455

< 6.

550.

146

< 15

.417

.00.

00<

6.65

Max

imum

105

3,92

032

518

.013

62.

5823

.376

.91.

159.

063

Min

imum

0.01

1518

2<9

.97

0.00

<6.5

50.

00<

15.4

<4.4

30,

00<6

.65

Mea

n22

.01,

290

57.7

8.27

78.7

269

45.9

53.1

0 53

397

6M

edia

n15

.299

711

80.

855

< 6.

550

199

< 15

.416

.20.

00<

6.65

Max

imum

104

3,22

047

783

.31,

340

31,0

815

•52i

5.33

176

4M

inim

um6.

2333

8<9

.97

0,00

<6,5

50.

00<

15.4

<4.4

30.

00<6

.65

Mea

n21

.71,

300

326

1 48

7.85

0.50

54.

7416

00

104

2.70

Med

ian

14.9

991

10.2

0 16

4<6

.55

0.02

98<

15.4

6.73

0.00

< 6.

65M

axim

um82

.93,

770

519

14.3

80.5

5.49

46;9

99.8

1.19

48.7

5M

inim

um2.

4820

5<9

.97

0.00

<6.5

50.

00<

15.4

<4.4

30.

00<6

.65

Mea

n18

.41,

000

14.7

0.60

12.

880.

294

2.57

9.34

0,01

45<

6.65

Med

ian

11.5

695

< 10

.00.

0461

<6.5

50.

00<

15.4

5.75

0.00

<6.6

5M

axim

um81

.13.

950

128

8.71

40.1

4.00

28.2

36.7

0.25

0<6

.65

(1+2

)M

inim

um4.

0939

4<9

.97

0,00

<6.5

50.

00<

15.4

<4.4

30.

00<6

.65

Mea

n23

11,

460

23.3

1.88

15.5

0.39

33

6715

90

149

2.89

Med

ian

16.1

1,18

014

.30.

527

< 6.

550.

161

< 15

413

60.

0353

< 6.

65M

axim

um10

53,

920

325

18.0

136

2.58

23.3

76.9

1.61

43.8

(3+4

+5)

Min

imum

2.47

9918

2<9

.97

0.00

< 6.

550.

00<

15.4

<4.4

30.

00<6

.65

Mea

n20

.71,

200

35.4

3 50

30.3

118

18.0

264

0.22

04

23M

edia

n13

.293

211

.00.

132

<6

550.

0219

< 15

.490

20.

00<

6 65

Max

imum

104

3,95

051

983

.31,

340

31.0

815

521

' ■

5.33

176

(1+2

+4+5

)M

inim

um2.

4820

5<9

.97

0.00

<6.5

50.

00<

15.4

<4.4

30.

00<6

.65

Mea

n21

.71,

310

23.8

1.47

10.6

0.40

13

6814

30.

104

2.15

Med

ian

13 5

1,03

011

.90

192

<6.5

50.

0584

< 15

48

660.

00<

6.65

Max

imum

105

3,95

051

918

.013

65.

4946

.999

.81.

6148

.7C

once

ntra

tions

hav

e a

rela

tive

unce

rtain

ty o

f 13

4%

Val

ues b

ased

on

data

col

lect

ed fr

om 5

/08

- 7/9

0 *

Val

ues

base

d on

dat

a co

llect

ed f

rom

1/0

9 - 7

/99

Tab

le 7

.4 S

umm

ary

of ac

eton

e, c

hlor

ofor

m a

nd te

rpen

e co

ncen

tratio

ns.

00


Ace

tone

, Chl

orof

orm

and

Ter

pene

sLo

catio

nsA

ceto

neC

hlor

ofor

mal

pha-

Pinc

ncbe

ta-P

inen

ep-

Cym

ene

Lim

onen

eco

mpa

red

RRR

ppt

ppt

RRR

. PP

tRR

Rpp

tpp

tRR

Rpp

t(1

+2) 4

30.

270.

471.

872.

200.

852.

220.

932.

351.

860.

70(1

+2) &

40.

380.

600.

630.

640.

920.

660.

350.

020.

840.

07(1

+2) 4

51.

221.

721.1

22.

742.

150.

790.

572.

283.

852.

273

& 5

0.84

0.92

2.37

2.66

1.02

2.30

0.96

2.77

2.54

1.00

(1+2

)&(3

+4+5

)0.

831.

341.3

11 5

30.

582.

180.

951 8

40.

920.

373

4(1+

2+4+

5)0.

110.

111.

862.

360.

912.

210.

932.

462.

090.

78A

lipha

tic a

nd A

rom

atic

Che

mic

als

Loca

tions

Benz

ene

2,3,

4-T

nmet

hylp

enta

neT

olue

neEt

hylb

enze

ncm

- 4p-

Xyl

ene

o-X

ylen

eIN

apht

hale

neco

mpa

red

ppt

RRR

ppt

ppt

ppt

ppt

ppt

ppt

(1+

2)4

38.

916.

078.

54S.

S26.

437.2

S7.0

S5.

SS(1

+2)

4 4

7.15

5.42

S. 66

7. S3

6.25

6.S9

6.72

6.80

(1+

2)4

59.

946.

9510

. S3

10. O

S7.

75S.

9IS.

666.

153

45

1.73

2.09

1.24

1.97

2.22

2.33

2.20

0.60

(1+2

) 4 (3

+4+5

)9.

166.

3410

.05

9.28

7.06

S.05

7.S4

7.09

3 4(

1+2+

4+5)

7.32

5.15

4.72

7.22

4.94

5.43

5.23

2.87

Varia

nces

with

in a

fact

or o

f 3__

____

____

_]

Test

stat

istic

sign

ifica

nt a

t the

99%

leve

lTe

st s

tatis

tic si

gnifi

cant

at t

he 9

5% le

vel

Varia

nces

diff

er si

gnifi

cant

ly

\ Te

sl st

atist

ic sig

nific

ant a

t tire

90%

leve

l

Tab

ic 7

.5 C

alcu

late

d St

uden

t’s t

valu

es f

or s

ite a

nd g

roup

mea

n co

mpa

riso

ns.

to

83

Site

MeasuredMeanppt

Predicted Mean Contribution

from Vehicles, ppt% Vehicle Emissions

OtherContributions

ppt% Other

(1+2) 510 604 118 N/A N/A3 £ 170 140 82.4 29.9 17 64 Ns 211 147 69.8 63.9 30.25 138 78 56.5 60 I 43.5

(3+4+5) 174 122 70.5 5 1 2 29 5

(1+2)

Tolu

ene

1,070 936 87.4 134 12.63 247 216 87.5 30.8 12.54 307 228 74.4 78.5 25.65 172 121 70.3 51.1 29.7

(3+4+5) 243 190 78.0 53.5 22.0

(1+2)

Ethy

lben

zene 158 147 92.6 11.7 7.42

3 37.9 33.9 89.6 3.95 10.44 38 8 35 8 92.2 3.03 7 805 21.8 18.9 87.0 2.82 13.0

(3+4+5) 33.0 29.7 90.1 3.27 9.92

(1+2)

m- &

p-X

ylen

e 531 527 99.3 3 93 0.743 121 122 101 N/A N/A4 128 129 101 N/A N/A5 66 0 68.1 103 N/A N/A

(3+4+5) 105 107 101 N/A N/A

d+2)

laph

thal

ene 39.8 32.1 80.8 7.64 19.2

3 14.3 7.44 52.0 6.86 48.04 12.1 7.85 64.9 4.24 35.15 12.2 4.15 34.0 8.06 66.0

(3+4+5) 12.9 6.52 50.6 6 36 49.4

Table 7.6 Apportionm ent results using o-xylene as a tracer molecule.



Site

0+2

)

(3+4

+S)

d+2)

(3+4

+S)

d+2)

(3+4

+S)

H+

E

(3+4

+S)

d+2)

(3+4

+5)

4* e v 9 ■© H s s tt JS ss u X L * i S * jt £ a « z

Mea

sure

d

npt

5)0

170

211

138

174

1,07

024

730

717

224

3

158

37.9

38.8

21.8

33.0

531

121

128

66.0

105

398

14.3

12.1

12.2

12.9

Pred

icte

d M

ean

Con

trib

utio

n fr

ont

Urb

an M

isso

ula,

ppt

615

142

150

79 125

960

222

234

124

195

138

31.9

33.7

178

28.0

530

123

129

68 107

35.0

8.

10

8 55

4.

52

7.10

% f

rom

U

rban

Mis

soul

a12

1 83

.9

71.1

57

.5

71 8

89,7

89

8 76

4 72

.1

80 I

87.2

84.4

86,8

82.0

848

100

102

101

104

102

880

56.6

70.7

37

.0

55 1

Pred

icte

d M

ean

Con

trib

utio

n fr

om

SSC

Mill

, ppt

169

56.5

17.0

9.

9428

.1

27.6

92.1

27.8

16.2

45.8

1.67

5.

58

1 68

0,98

2.

78

4 51

15

1

4 54

2.

65

7.49 1.75

5.

84 1.76

1.

03

2.90

% fr

om

Mill

3 32

33

.3

8.06

7

20

16.2

2.58

37

.3

9,04

9,

43

18 8

1.06

14

.7

4.33

4

51

841

0 84

9 12

5

3.55

4.01

7.

10

4.40

40

.9

146

842

22 6

Tot

al

Con

trib

utio

n fr

ont

Urb

an M

sla &

Mill

124

117

79.1

64.7

88.0

92.3

12

785

.4

81.6

98

.9

88.2

99.1

91.2

86.5

93.2

101

114

105

108

109

92 4

97

.5

85.3

45

4

77.7

oo

Tab

le 7

.7 A

ppor

tionm

ent

of a

rom

atic

che

mic

als

usin

g "d

ispe

rsio

n" r

efer

ence

com

poun

ds.

72


Site

Mea

sure

dM

ean

ppt

Pred

icte

d M

ean

Con

trib

utio

n fr

ont

Urb

an M

isso

ula,

ppt

% f

rom

l)

r ban

M

isso

ula

Pred

icte

d M

ean

Con

trib

utio

n fr

om

SSC

Mill

, ppt

% f

rom

M

ill

Tot

al

Con

trib

utio

n fr

om

Urb

an M

sla

& M

ill(1

+2)

ft12

23.

6229

.71.

4812

141

.83

! o7.

020.

838

11.9

4.94

70 3

82.3

4£ e

7.32

0884

12.1

1.49

20.3

32 4

5o 2

6.20

0.46

87.

540.

869

14.0

21.6

(3+4

+5)

u6.

850.

734

10.7

2.45

35 8

46.6

RR

RR

RR

d+2)

4* e1.

880.

131

6.94

11.6

617

624

344 S

8.27

0.03

020

365

38.7

469

469

4A

, i1 4

80.

0319

2 15

11.7

789

791

5X

t &06

00.

0169

2.80

6.82

1,1.10

1,13

0(3

+4+5

)*

3.50

0.02

650.

756

19.3

550

551

RR

RR

RR

d+2)

440.

393

0.04

210

84.

051,

030

1,04

03

6 44 e2.

690.

0097

80.

364

13 5

503

503

4Qm

0.50

50.

0103

2.05

4.08

807

809

55 44

0.29

40.

0054

61.

862.

3880

981

1(3

+4+5

)2

1.18

0.00

858

0.72

76.

7257

057

1RR

RR

RR

(1+2

)0.

149

00

15.1

10,1

0010

,100

3e 44

0.53

30

050

.39,

440

9,44

04

e ©0.

104

00

15.2

14,6

0014

,600

5E

0.01

450

08.

8561

,000

61,0

00(3

+4+5

)0.

220

00

25.0

11,4

0011

,400

^Chl

orof

orm

app

ortio

nmen

t was

don

e fo

r th

e sa

mpl

ing

peri

od a

fter

the

blea

ch p

lant

was

clo

sed

Tab

le 7

.8

App

ortio

nmen

t o

f chl

orof

orm

and

ter

peno

ids

usin

g "d

ispe

rsio

n" r

efer

ence

com

poun

ds.

86

Total Aromatic Compounds 3 4 5 (1+2)Linear Vehicles 73 78 69 97

Combinations Mill 34 9 9 2Other 0 13 22 1

Dispersion Vehicles 89 80 72 99Reference Mill 30 8 8 2Molecules Other 0 12 20 0o-Xylene Vehicles 88 79 71 97

100% from vehicles Other 12 21 29 3o-Xylene Vehicles 69 62 56 76

78.6% from vehicles Other 31 38 44 24W,4-T rimethylpentane Vehicles 78 87 65 100

100% from vehicles Other 22 13 35 0

Table 7.9 Estimated percentages o f total arom atic compounds from various sources.


Chapter 8 Discussion

This chapter is divided into four main sections. The first section provides a

discussion o f the experimental methodology used in this study. The quantitative results

and their im plications are discussed in the second section. The second section is further

subdivided into three parts; benzene, 2,3,4-trim ethylpentane, toluene, ethvlbenzene. m-lp-

xylene. r;-xylene and naphthalene levels and trends are addressed in the first part. The

second part discusses the concentrations and trends o f the terpenes and chloroform , while

the third part deals with acetone. The third section in this chapter explains the

relationships between weather and the concentrations o f the volatile chemicals quantified.

The final section discusses the apportionm ent results and their implications.

8.1 Experimental Methodology

The experim ental methodology em ployed in this study offers several advantages

over ev acuated canister procedures. First, the nature o f the sorbent tubes reduces the

possibility o f reactive transformations o f the trapped analytes and the retention o f higher

m olecular weight components. Second, dram atic sample concentration can be achieved

which improves detection limits. Third, longer collection times guarantee that the sam ple

will be more representative o f the valley airshed. W hile the advantages o f this m ethod

are numerous, it also has several shortcomings. Both advantages and disadvantages o f

the methods used are discussed below.

Analytes trapped in multi-bed sample tubes are less apt to undergo chem ical

degradation before analysis. Once the m olecules are sorbed by the various carbonaceous

87


88

materials, they are protected from collisions with other m olecules that could result in

chemical transform ations. In contrast, chemical analytes captured within an evacuated

canister are free to collide w ith any reactive species present. The sorbents used in the

CMS tubes are chem ically sim ilar to the organic analytes. Thus, analyte sorption is akin

to the condensation o f m olecules into/onto an inert surface. In contrast, compounds

adhering to the walls o f an evacuated canister are more likely to experience

heterogeneous catalysis. Even if the chem icals sorbed to the inner walls o f a canister are

not chemically altered, they are removed from the sample analysis, resulting in erroneous

chemical data [Fehsenfeld et a /., 1992; Altshuller, 1983], Because o f these problems,

canister samples m ust be analyzed immediately. Sorbed chem icals are removed from

CMS tubes by flash heating w hile flushing with helium gas. This assures that a high

percentage o f the sorbed analytes are transferred into the gas chrom atograph.

Experiments perform ed as QA/QC verification demonstrated that air samples taken using

CMS tubes are stable for several weeks, provided they are kept tightly sealed, stored

below 4°C, and not exposed to direct sunlight.

The analyte detection limits associated with the CMS tube methodology are at

least two orders o f m agnitude lower than those obtained using canister methods. Over an

eight-hour sampling interval, the contam inants from up to 50 liters o f air were trapped by

the sorbent materials. Typical evacuated canisters have volum es o f only 5 to 10 liters.

Second, the thermal desorption step exhaustively releases all o f the trapped molecules.

Most canister methods use a syringe or cryotrap to remove an aliquot o f the trapped air

for analysis, further eroding the detection limit.


89

The eight-hour sam ple collection period used with the CMS tube technique better

represents the true overall exposure levels at a sample site. Evacuated canisters remove a

brief, grab sam ple from the a ir over a period o f minutes. This is analogous to a snapshot,

revealing only what was present at that instant. The CMS tubes collect a composite,

tim e-weighted sample. It is a more accurate description o f what an individual would be

exposed to while breathing the valley air as the day progresses and weather patterns vary.

The com posite sample is also less vulnerable to anomalous, short duration episodes. For

example, if a canister sample were taken at the same time a vehicle with exceptionally

heavy hydrocarbon emissions passed by, the resulting analysis would produce misleading

results. A com posite sample would not be appreciably affected by a sim ilar episode, as it

would have been taken using a constant flow over a long period o f time. The composite

sample also eliminates the bias that could occur if one had to choose when to actually

take a grab sample. One might wait until a vehicle with heavy em issions had cleared an

area before filling the canister. In so doing, a form o f bias would enter into the study.

Although the CMS tube technique has many advantages over the evacuated

canister method, it does possess several drawbacks as well. The most serious

disadvantage is that the CMS tubes com prise a destructive sample method; the technician

has only one chance to perform the analysis. If the instrument malfunctions, the CMS

tube is damaged, or an anomalous result is obtained, there is no second aliquot to analyze.

Generally, the CMS tubes can be conditioned and reused numerous times. However, on

occasion, they are damaged either by repeated use or irreversible sorption. This makes it

necessary to keep track o f each tube in order to flag any spurious behavior. Improperly

conditioned tubes, or those past their useful lifetimes, are vulnerable to analyte


90

breakthrough during sam ple collection. In addition, high hum idity conditions also cause

analyte breakthrough as the excess w ater molecules interfere with the sorption o f target

molecules. Finally, by the nature o f the CMS tube methodology, any inform ation on

acute pollution episodes is sm oothed out.

8.2 Overall Qualitative and Quantitative Results

The data collected during this study indicated that VOC concentrations in the

Missoula Valley airshed follow three distinct patterns. A m bient concentrations o f

chloroform, a-pinene, P-pinene, p-cvm ene and limonene (group 1) w ere found to be the

highest near the Smurfit-Stone C ontainer Pulp Mill at site 3. Their concentrations

diminished as the distance from the pulp mill increased. The concentrations o f benzene,

2.3.4-trimethvlpentane, toluene, ethvlbenzene. /w-/p-xylene, o-xvlene and naphthalene

(group 2) are clearly linked with the heavy vehicle traffic associated w ith urban Missoula

(sites 1 and 2). A dram atic reduction in the levels o f group 2 com pounds was observed as

the sample sites moved westward away from the city'. A cetone (the only mem ber o f

group 3) was found to be ubiquitous throughout all o f the sam ple locations. Acetone

levels are likely the result o f photochemical initiated degradation o f m any different

primary chemicals.

-S'. 2.1 ( hloroform and the Terpenes - Group 1 Compounds

The highest concentrations for the group 1 com pounds were all found at site 3

(Figure 8.1). The site 3 m ean concentrations o f chloroform, p-cym ene and a-pinene were

about twice as high as those found at any o f the other four sam pling locations. Mean site


91

3 concentrations o f chloroform and /?-cymene were 57.7 and 53.1 ppt, respectively. The

mean RRR for a-pinene was 8.27 over then entire period o f study. Between January and

July 1999, when absolute concentration values were available, a-p inene had a mean

value o f 78.7 ppt.

} 8.00

{ 7.00

• 6.00

i 5.00

| 4.00

60.0

50.0 <

40.0CLa.

30.0j 3.00

2.0020.0

- 0.00 51 32 4

Site

—&— Chloroform # p-Cymene —& - alpha-Pinene

Figure 8.1 Mean concentrations o f chloroform , /?-cymene and a -p inene at the fivesam pling locations.

As shown in Figure 8.2, the m ean RRRs for the other two group 1 compounds

(lim onene and (3-pinene) exhibited the sam e site 3 maximum. The a-p inene trace is

reproduced in the figure to em phasize the similarity in trend. Limonene show ed a slight

disparity between sites 1 and 2, how ever the observed difference betw een the two means

was not statistically significant.


92

9.00 —[ 0.608.00

7.00

6.00 0.40at 5.00

I 0.30 £4.00 r

3.00 f

2.00 \ 1 00 k { 0.10

0.00 0.0021 3 54

Site~® ~alpha-Pinene —* — beta-Pinene • Limonene

Figure 8.2 M ean concentrations o f P-pinene and limonene plotted with a-pinene.

The concentrations o f the group l compounds dim inished with distance from the

mill along the southeasterly line described by sites 3, 4 and 5 (Figure 8.3). This

distribution pattern supports the contention that the m ill's emissions are responsible for a

portion o f the VOCs found in this region o f the valley. The linearity o f the chloroform

line is indicative o f its stability' in the atmosphere. The terpenes' drop in concentration

was more dramatic. This was due to their high reactivities and dramatically shorter

atmospheric lifetimes (Table 5.1).

W hile the concentrations o f the group 1 com pounds generally decreased with

increasing distance from site 3, site 2 was an exception to the rule. Located on the

eastern side o f the city, site 2 had slightly higher mean concentrations o f the group 1

compounds than site 1. The elevated levels o f these com pounds measured at site 2

suggested that they have sources in addition to the pulp mill. Although this conclusion

makes intuitive sense, probabilities that were calculated for differences between means

from sites 1 and 2 were about 50 percent.


93

7 0 . 0 ----------------------------------------------------- 3.00

60.0 2.50

50.02.00

40.0a.& 1.50

30.01.0020.0

0.5010.0

0.0 0.003 4 5 (1+2)

Site♦ Chloroform —* — p-Cymene alpha-Pinene

Figure 8.3 Chloroform and terpene trend. site3 to urban M issoula.

Since sites 1 and 2 appeared statistically similar; they were com bined into a single

data set representing urban Missoula. Concentration means for each o f the group 1

chemicals are plotted from west to east across the valley in Figure 8.3. The mean

chemical concentrations decreased from site 3 through site 4 to a m inim um at site 5. Site

(1 -2 ) had group 1 mean concentrations that were greater than site 5. bu t generally less

than site 4.

The mill is a significant source o f the group 1 compounds to the valley airshed.

The concentration means from site 3 were statistically compared to the urban sites ( K 2).

These results were summarized in Table 7.5. The mean terpene and chloroform

concentrations found at site 3 were statistically higher than those m easured at site ( H 2).

a-Pinene, P-pinene and /7-cymene were higher at the 95% confidence level; limonene and

chloroform were higher at the 90% confidence level. The lower level o f confidence

associated with the limonene and chloroform levels indicates that there are unidentified

sources o f these compounds affecting the urban Missoula samples.


94

The dom inance o f the mill as a group 1 em itter still holds when the two rural sites

(4 and 5) are added into the picture. The mean terpene concentrations at site 3 were still

significantly higher (95% confidence level) than those at all non-mill sites ( H 2+4+5).

The mean chloroform concentration was also higher at site 3. but only a 90% confidence

level could be associated with that value.

Added support for unidentified inputs o f group 1 chem icals is found in a

comparison o f site 4 to site (1 -2 ). Statistically, there was no difference between

concentration means o f group 1 com pounds. Dispersion and dilution would dictate lower

means for site (1-^-2) if the mill were the sole source o f these compounds.

The mean concentrations o f group 1 compounds at site 5 were surprising. First,

the concentrations o f a-pinene, /?-cvmene and limonene were lower at site 5 than at any

other site, despite the fact that it sits in the middle o f a conifer stand. Anthropogenic

terpene em issions exceed the quantities released in the vicinity o f this forested site. Both

the pulping process and the com bustion o f wood liberate terpenes; with chemical pulping

responsible for em itting the greater quantities. Site 3 mean terpene concentrations w'ere

much greater than those found at site 5, significant at the 95-99% level. Even site (1 -2 )

had significantly higher terpene concentrations, except for p-pinene. While these

anthropogenic terpene em issions are small on a regional scale, this study suggests they

are significant to the M issoula Valley.

H.2.1.1 The Impact o j Chlorine Bleaching on the Valley Airshed

D uring the course o f this study, it was possible to assess the impact o f Smurfit-

Stone C ontainer's chlorine bleaching plant on the valley. In response to changes in


95

env ironmental regulations and econom ics, the mill discontinued the bleaching o f pulp

stock in February 1999. Thus, sam ples collected prior to February 1999 included

contributions from the bleach plant, while those collected after did not. Table 8.1

com pares the chloroform concentrations at the sampling sites before and after the bleach

plant closure.

Table 8.1 Comparison o f chloroform levels before and after the bleach plant closure.

Chloroform (ppt)Site 1 Site 2 Site 3 Site 4 Site 5

Mean while BP operating 27.2 43.6 109 62.6 24.5Mean after BP shutdown 12.0 12.3 7.02 7.32 6.20

Variance while BP operating 1167 25 5362.13 20065.47 14956 95 1209 44Variance after BP shutdown 184.55 57.89 42.96 33.95 49.85Number samples (up/down) 25/14 20/14 26/14 26/14 25/13

t value 1.96 1 80 3.66 2.30 2.54c 0.78 0.99 1.00 1.00 0.93

Degrees o f Freedom 34.40 17 47 25 20 25.21 28.77

As expected, the greatest reduction in chloroform levels derived from the bleach

plant closure was observed at the mill. The mean chloroform concentration at site 3

plummeted 94%, from 109 ppt w hen the bleach plant was in operation to 7.0 ppt

afterwards. The test statistic for the tw o means was 3.66, an indication that the two

m eans were different at the 99% confidence level. Dramatic reductions in chloroform

concentrations were also observed at the other two rural sites after the bleach plant was

closed. At sites 4 and 5 there was an 88.7% and 74.7% fall o ff in m ean chloroform

levels, respectively. In both cases, calculated t values supported the observed differences

at the 95% confidence level.

The behavior o f the chloroform levels in urban Missoula was m ore interesting as

the bleach plant ceased operation; they provide significant evidence that the mill is no


96

longer the largest overall source o f chloroform to the M issoula Valley. W hile the bleach

plant was up, mean chloroform concentrations were 27.2 and 43.6 ppt at sites 1 and 2.

After the bleach plant shutdown, the mean concentrations dropped to 12.0 and 12.3 ppt,

respectively. These reductions, 55.9% at site 1 and 71.8% at site 2 are not as m arked as

those found in the western ha lf o f the airshed. Pre- and post-bleachplant closure levels at

sites 1 and 2 differed at the 90% confidence level. Even more intriguing is that after the

bleach plant closed, the M issoula urban center became a larger source o f chloroform

releases into the valley airshed than the pulp mill. Post-bleachplant mean chloroform

concentrations at both urban sites were higher than the post-bleachplant mean at the mill

(site 3). Mean chloroform levels o f 12.0 and 12.3 ppt were measured at sites 1 and 2

compared to only 7.0 ppt at site 3.

A supplemental piece o f evidence indicating that the mill was a much larger

source o f chloroform to the airshed while bleaching is seen in the chloroform

concentration variances. A dam ping o f the fluctuations is manifest in the post-

bleachplant chloroform concentration variances at each site. The sample variances were

between 6 and 467 times greater while the bleachplant was operational than after its

closure. This was due to the sporadic operation o f the bleachplant, dictated by the

m arket's periodic dem and for white linerboard. The on again/off again nature o f the

bleachplant caused the extremely large chloroform variances calculated at the sample

sites while the bleachplant was online.


97

H.2.2 Aliphatic and Aromatics Group 2 Compounds

Group 2 compounds are strongly associated with vehicle em issions, although they

may also em anate from residential w ood burning. Not unexpectedly, then, the highest

mean concentrations o f the group 2 com pounds were measured at sites 1 and 2. Overall,

toluene was present at the highest concentrations o f any com pound from groups 1 or 2.

The highest daily toluene concentration o f 3,990 ppt was m easured at site 1. The mean

toluene concentrations at sites 1 and 2 w ere 946 and 1,160 ppt. Figures 8.4 and 8.5

illustrate the trend exhibited by the group 2 compounds. The correlation o f the group 2

chemicals with urban Missoula is conspicuous in the six-fold decrease in their

concentrations as the sampling distances moved westward away from the city.

1400 4.00

3.501200

3.0010002.50

800Q.Q. 2.00 ac

6001.50

400 1.00

200 0.50

0.003 5 24 1

Site

♦ Toluene —■— 2.3.4-Trimettiylpentane

Figure 8.4 Mean concentrations o f toluene and 2,3,4-trim ethylpentane.


98

700 250

600200

500

150400

300 100200

100

03 4 5 1 2

- Benzene

Site

I—o-Xylene —A— Ethylbenzenem-&p-Xy1ene —• —o-Xylene —A— Ethylbenzene — Naphthalene

Figure 8.5 The trend in the group 2 compounds.

Data from multiple sites were grouped to evaluate regional differences. M ean

concentrations o f group 2 compounds in Urban M issoula (site 1+2) were significantly

higher than sites 3. 4 and 5. These are sum m arized num erically in Table 7.5 and visually

in Figure 8.6. The differences were significant at the 99% confidence level. The low est

mean concentrations o f the group 2 compounds w'ere m easured at site 5.

1200

1000

800

g- 600

400

200

3 54

Site

-Toluene —• — o-Xylene

200180160140120100806040200

Figure 8.6 M ean concentrations o f toluene and o-xylene across the M issoula Valley.


99

8.2.3 Acetone

The occurrence o f acetone was noticeably d ifferent from any group 1 or group 2

com pound; it had a uniform distribution across the M issoula Valley (Figure 8.7). Over

the entire course o f the study, the m ean RRRs o f acetone only varied from 18.4 to 25.5.

A statistical com parison o f mean acetone concentrations from sites 3, 4 and 5 to site

(1 -2 ) revealed no significant differences (Table 7.5). The highest measured acetone

concentration after January 1999, when absolute quantitation standards were in use. was

3.950 ppt. Estim ates based on RRR calculations indicate the highest acetone

concentration during the study period exceeded 5.000 ppt.

30.0 ;

25.0 4-

20.0

15.0

5.0 j

0.0 *---------------------------------------------------------------------------------------------------------1 2 3 4 5

Site

Figure 8.7 Mean relative response ratios for acetone.

O f all the com pounds followed, acetone has one o f the longest potential

atm ospheric lifetimes. During the sum m er, its lifetim e is between 16 and 66 days, but it

can be as long as 170 days in the w inter [Singh et a l., 1995; Seinfeld and Pandis, 1998],

This is com parable with the time scale necessary for upper tropospheric turnover,


100

approximately 10 days [Arnold et al., 1997], Photolysis in the presence o f oxygen is the

principal mechanism for acetone 's atmospheric degradation [Cox et al., 1981 ].

C H 5C O C H 3 + hv - 20; -* C H ,C O O ;- + C H jO ;’ (8 .1 )

Based on acetone 's longevity, it is reasonable to assume it will be well mixed within the

Missoula Valley airshed. O ther possible explanations for the ubiquitous nature o f

acetone are that there are many sources affecting the airshed concentration or that it is not

a direct emission, but rather formed from other chemicals as they degrade.

S.2 .3 .1 Seasonal l-'luct nations in Acetone Levels

Air samples collected in the Missoula Valley indicated that the acetone

concentrations followed a seasonal pattern (Figure 8.8). The lowest acetone levels were

observed during the winter. As the summer progressed, the levels o f acetone measured in

the valley airshed increased, finally reaching a maximum concentration late in the season.

Acetone has been detected in the smoke given o ff during biom ass combustion [Hauk et

al., 1994], however 1998 did not have a major forest fire season.

The kraft pulp mill in the Missoula Valley is a continual source o f acetone to the

airshed. Acetone is produced as a byproduct o f lignin digestion [Someslnvar, 1995;

Smook. 1992]. The mean acetone levels measured near the mill were no greater than

were found at the other sites. Furthermore, the observed concentration trend is opposite

from what a constant release would produce. Under constant release, summer levels

would be low because photolytic rates are accelerated. In winter, a constant release

should lead to higher levels. The observed pattern places the highest concentrations in


late summer and the minimum in mid-winter. These facts elim inate the mill as the source

o f the seasonal acetone concentration fluctuations.

70 t 1000

IE 40 fDC !

20

60

10 i

50

05/1/98

w

7/1/98 9/1/98 11/1/98 1/1/99 3/1/99 5 /1/99 7/1/99

Date

Acetone Sunlight

Fitiure 8.8 Seasonal trends in acetone level and dailv sunliuht.

The bulk o f the acetone measured in the airshed is probably due to secondary

formation. A cetone is a natural vegetative emission [Guenther et al., 1994; Isidorov et

al.. 1985], however, em issions from coniferous trees are dom inated by monoterpenes.

Acetone along with other non-m ethane hydrocarbons constitute only a minor portion o f

these biogenic em issions [Lamb et al., 1987], Consequently, the observed increase in

acetone appears due to the photochem ical initiated degradation o f other organic

compounds. Also, the production o f acetone must exceed either its rate o f destruction or

its dispersion out o f the valley airshed.

Many chem icals have photochemical degradation pathways that lead to the

formation o f acetone. The oxidation o f various hydrocarbons, such as isobutane,

isobutene and propane are believed to be a major secondary source o f acetone to the

atmosphere [Arnold et al., 1997; Seinfeld and Pandis, 1998],


102

CH 5CH :C H 5 + OH- + NO + 2 0 2 - NO : + C H 3CO CH 3 + H20 - HOy (8.2)

Most o f these small alkanes and alkenes are com ponents o f gasoline fuel. As a group,

these molecules com prise between 54.8 and 70.9 percent o f the exhaust and between 6 6 . 8

and 93 .9 percent o f the evaporative emissions from autom obiles, as measured under

typical urban driving conditions [Sigsby et al., 1987],

Terpenes may be one o f the most significant secondary sources o f acetone.

Terpenes are em itted into the atmosphere by both biogenic and anthropogenic

mechanisms. S ignificant amounts o f terpenes are produced and em itted to the

atm osphere by trees [Helm ig and Arey, 1992; Arey et a!., 1995]. These biogenic

em issions are dependent upon a seasonal cycle as well as tem perature [Altshuller, 1983].

Terpenes are oxidized by reaction with H ( \ and N O 3 radicals as well as ozone

[Corchney and A tkinson, 1990; Atkinson, 1990; A ltshuller, 1983], During their

atmospheric degradation, they produce significant am ounts o f secondary chemicals, such

as acetone. V inckier et al., [1998] m easured the acetone yields from a-pinene/hydroxyl

radical reactions and estim ated 2.8 Tg yr~' are produced in the troposphere globally.

Grosjean et al. [1993] and Atkinson [1990] reported that the reactions o f a - and P-pinene

with ozone yielded significant amounts o f acetone. H allquist et al. [1999] studied the

reactions o f a-pinene, p-pinene, i^-carene and limonene with the NO 3 radical. All o f the

terpenes studied produced significant am ounts o f carbonyls (12 - 78% yield), with a -

pinene producing the greatest quantities. Yatagai et a l . ,[ 1995] found that the emissions

o f these biogenic chem icals peak late in the summer. This natural peak coincides with

the elevated levels o f acetone measured in the M issoula Valley at the end o f the summer

season.


103

Processes utilized in forest products industries, such as pulping and lumber

production, increase the quantities o f terpenes emitted into the atmosphere. The kraft

pulping process adds other unmeasured chemicals to the airshed in addition to terpenes;

some o f which, may also be secondary sources o f acetone. During the chemical pulping

process, lignin is chem ically and thermally broken down. Since lignin is a large,

com plex, highly polymerized structure, the number o f organic breakdown products is

vast. The propane linkages between phenolic moieties are prime candidates as acetone

precursors. Qualitative analysis o f valley air has indicated the presence o f a large number

o f organic compounds, many with the potential to increase acetone concentrations within

the airshed.

A cetone's large num ber o f direct emission sources and chemical precursors,

coupled with its long atm ospheric lifetime, appear to be responsible for its uniform

distribution and seasonal fluctuation. The seasonal trend in acetone levels was observed

in the M issoula airshed because o f the geography and topography unique to the valley.

This afforded the time necessary to measure the concentration before it dispersed.

8.3 Meteorological Parameters and Their Correlation with Chemical Concentrations

The reconciliation o f daily weather data with the measured concentrations o f

contam inants was accom plished through several intuitive approaches. Daily

tem peratures, am ount o f precipitation, snow cover, wind speed, wind direction and

am ount o f sunlight from the National W eather Service (M issoula International Airport)

and Sm urfit-Stone Container's Am bient 1A w eather station (site 3) were analyzed in an

effort to determ ine their influences over the volatile organic chem icals found in the


104

airshed. Seasonal variations in the daily temperatures account for observed concentration

increases o f group 1 and 2 compounds during the winter. High w ind speed correlates

with a predom inant northwest wind, indicating the passing o f a w eather front. This

situation results in low ambient contam inant levels. Valley wind direction was found to

be highly variable and can be a misleading piece o f information w hen used to predict the

sources o f volatile compounds to the valley airshed.

-S’. 3 .1 Seasonal Weal her Patterns

Seasonal weather patterns and daily temperatures during the late fall and w inter

can lead to poor dispersion and tend to cause inversion conditions in the valley.

Overnight, as cold air sinks lower into the valley, boundary layer compression

concentrates contam inants in the air column. Low levels o f sunlight decrease the

removal o f volatile compounds by convection. This helps to trap the atmospheric

contaminants near ground level and results in stagnant air episodes, which are evidenced

by an increase in the concentration o f the pollutants. As expected, an increase in the

ambient levels o f the terpenes was observed in late October at site 3, closest to the

principal release point in the airshed. This is demonstrated by the /?-cymene

concentration shown in Figure 8.9.


105

600 - 550 - 500 - 450 -

g 400 - S t 350 --1 300 - | 250 - £ 200

150 - 100 - -

50 ---it0 -i

CO O)

Date

Figure 8.9 /7-Cymene concentrations at site 3.

The lack o f data points in Decem ber 1998 and January 1999 was due to instrument

downtime. This data gap must be taken into consideration when discussing the seasonal

trends and certainly weakens the body o f evidence.

Sites 1. 2, 4, and 5 did not show a strong w inter time increase for the group 1

compounds. Their distance from the m ill, coupled with the lack o f turbulent m ixing in

stagnant inversion conditions, kept them isolated from the terpene rich air mass. This is

evident in Figure 8.10, which shows a rather uniform low level o f /?-cymene at site 1

across several seasons. Diffusion, by itself, is insufficient to move volatile chem icals

over the large distances between the mill and the other four sampling sites.

The increase in the terpene concentration during the winter is unlikely to be the

result o f natural emissions. Since the increase occurred during the coldest part o f the

year, it does not fit with the widely dem onstrated dependence o f vegetative em issions on

am bient tem perature [Staudt et al., 1997; Isidorov et al., 1985; Lamb et al., 1985;


106

Fehsenfeld et al., 1992], Furthermore, only site 3 dem onstrated an increase in terpene

levels. If the increase was due to biogenic em issions, site 5 should also show an increase

and this was not the case.

The chloroform concentration seem ed to follow the same seasonal increase as the

t e r p e n e s . However, the lack o f data points for Decem ber and January, coupled with the

variability in the bleachplant operation and closure, dem and a more cautious assessment

o f t h e available evidence.

600 - 550 —500 - 450 —

g 400 - — 350 -0g 300 —1 250 •£ 200 -

150 •100 ■

50 - 0

CO

in

Date

Figure 8.10 /7-Cymene concentrations at site 1.

The concentrations o f the group 2 com pounds increased during the winter season

in a trend sim ilar to that o f the terpenes. The only difference was that the occurrence o f

the trend was at the opposite side o f the valley, near the source o f the group 2 com pounds

(urban M issoula). The basic premise for this trend is the same as described above for the

terpenes. Figure 8 .11 illustrates the cold-w eather increase in toluene concentrations at

site 1, while Figure 8.12 records a much less pronounced increase at site 3. This

o>o CM

0>CO


107

reinforces the evidence for urban M issoula and m otor vehicle use as the m ajor source o f

group 2 compounds to the valley airshed.

o.Q.

O3

4500

4000

3500

3000

2500

2000

1500

1000

500

0a t a t a t

5 — <vi —

Date

Fiszure 8.11 Toluene concentrations at site I .

4500

4000 -

3500 •-

~ 3000 -

f 2500 - | 2000 -

1500 •

1000

500 •

0 .1 n , i i 11

CO

.n i i io>

fM CM CO ^

Date

Figure 8.12 Toluene concentrations at site 3.

S. 3.2 Wind I 'elocity and C 'hemical C 'oncent rat ions

The measured chemical concentrations were coupled to the peak wind speed for

each day to expose any underlying dependency. There was reasonable correlation


108

between high pollutant levels and low wind velocity. D uring the entire period o f study,

there were 22 days that had a peak wind speed greater than 15 mph. On 18 o f those 22

days, the predom inant wind direction was from the northwest. This type o f weather

pattern is indicative o f a high-pressure system displacing a low-pressure air mass from

the area. The result is an increase in turbulent mixing and the dilution o f volatile

chem icals as a cleaner parcel o f air moves into the region. Figure 8.13 plots the daily

toluene concentration at site 5 against the average wind speed for each day. The scatter

around the line is expected, as a number o f variables actually influence the concentration

o f toluene on any given day. This trend was identified in every compound studied and at

each sample site.

700 |-

y = -23.643X + 283.75 R2 = 0.2243

500

400

300ot-

80 1 2 3 5 7 9 1064Average Wind Speed (mph)

Figure 8.13 Fall-off in toluene concentration with wind speed at site 5.

H.3.3 Wind Direction and Chemical Distributions

Com parisons between the chemical concentrations from the eight-hour sam pling

periods and the average wind direction data are easily m isinterpreted. To investigate the

sources o f the various quantified compounds, daily wind direction data from the National


109

W eather Service (NWS) and hourly data from Smurfit-Stone C ontainer's Ambient 1A

w eather station (site 3) were used to create polar coordinate plots o f wind direction versus

com pound concentration, sim ilar to wind rose diagrams. In Figure 8.14 each point

represents the head o f a vector whose angle on the graph indicates the direction from

which the wand originated. Each vector begins at the origin and its length or magnitude

corresponds to the concentration o f /?-cymene. Thus, the farther a point is from the

origin, the greater the daily concentration, while the placem ent o f the point on the graph

corresponds to a compass reading o f the wind direction. W ind direction as plotted, is the

day 's fastest gust o f wind lasting at least two minutes (m easured by the NWS).

North• 520-------—

260

West r - ■ - ■ East-520 -260 # 0 • 260 520

* 620------

South

Figure 8.14 NWS wind direction and site 3 /?-cvmene concentration.

According to evidence previously presented, most i f not all o f the points should

com plete vectors pointing northw est tow ard the mill, the m ost probable source o f the p -

cvmene. A cursory look at the plot indicates that there are 10 data points that do not fall

in the northwest quadrant. Even m ore alarming, the point associated with the largest


110

concentration (red) com pletes a vector, which points to the south-southwest, suggesting

som e counter-intuitive source of/7-cymene.

A detailed exam ination o f this confounding data point proves the directional

inform ation spurious. The point represents a sample collected on November 11. 1998

when the predom inant wind direction was 185° (south-southwest) according to the NWS.

As recorded at the site 3 weather station, it was 125° (southeast) for five o f the eight

hours when sam pling took place. Thus, reasonable agreem ent is demonstrated between

the two w eather stations. The wind direction seemed to be fairly stable, originating from

a southerly direction. But the wind direction and /7-cymene concentration plot do not

reveal the entire story. Biogenic em issions were out o f the question as a source o f p-

cvm ene; the maxim um temperature on November 11, 1998 was only 32° F (0° C), the

average was 26° F (-3° C) and the m inim um was 19° F (-7 ’ C). Furthermore, the average

wind velocity was 0.9 mph. These conditions indicate a stable high-pressure system

dom inated the valley weather, causing stagnation in the airshed. The misleading nature

o f this data point is also substantiated by the /?-cvmene concentration gradient around the

valley. Sites 5, 4 and 3 are arranged in a south to north line. The/7-cymene

concentrations in the direction o f w ind movement were found to be 6.32, 99.8, and 521

ppt, respectively. W ere a source o f the jD-cymene in the southern end o f the valley, its

concentration gradient would have to have been in the opposite direction.

The w ind direction in the M issoula Valley was extrem ely variable. Variability

was present both within and between the measuring stations. First, the wind direction

recorded at either w eather station was often so irregular as to m ake the determination o f

its predom inant direction impossible. Second, it was more com m on for the two weather


stations to disagree on the wind direction than to agree. The variability o f the wind

direction often led to erroneous conclusions. This can be appreciated by comparing

Figure 8.14 with Figure 8.15. Figure 8.15 was prepared in the same m anner as Figure

8.14, except that the wind direction measured at site 3 was substituted for the NWS

direction. Since site 3 directional data were recorded on an hourly basis, an average wind

direction was determ ined by analyzing the hourly wind directions between 8 a.m. and 5

p.m. The resulting predominant wind direction was used to create the plot.

North

225 •

West - • - East-S60 -225 3 ^ ^ 225 *50

«0------South

Figure 8.15 Site 3 wind direction and site 3 /7-cymene concentration.

The differences between the two plots are so striking that one might think they are

two, com pletely different sets o f data. The red point in the southeast quadrant is from

November 11, 1998, which was discussed previously. The green point in the southeast

quadrant represents data from October 8, 1998. On that day sample collection occurred

between the hours o f 8:00 a.m. and 5:00 p.m. D uring that time period, interpretations o f

the site 3 wind direction data are problematic. For three hours the wind cam e from a

southeasterly (128°) direction. For another three hours a northwest (321 °) wind direction


112

was predominant. The remaining three hours recorded wind from the east-northeast

(71°), northeast (59°) and west-southwest (232°). The single datum from the NWS wind

direction was west-southwest (250°). How does one decide the best wind direction to use

for this day? As with the November 1 I, 1998 data, the concentration gradient for p-

cym ene is inconsistent with a southeastern source: site 3 (187 ppt), site 4 (45.7 ppt), site 5

(36.7 ppt). Site ( H 2 ) had a 53.5 ppt concentration of/?-cymene, which is indicative o f

another source's influence.

There are two reasons for the disparity in m easured wind direction between the

two w eather stations. The first is geography. Each location is affected by the topography

o f the valley. The NWS station is located in the center o f the valley, at the M issoula

International Airport. The site 3 station sits in a trough, close to the Bitterroot

M ountains, at the intersection o f M ullan Road and M occasin Lane. It is separated

topographically from the airport by a ridge and valley system. The second reason for

direction disparity is anem om eter height. The NWS measures wind speed and direction

10 m eters (32.8 ft) above the ground. Site 3 m easures these parameters at about h a lf that

distance above the ground. Wind direction is more variable close to the ground. Site 3

records a more localized picture, while the NWS obtains a more generalized picture o f

the valley weather.

W hen the wind direction was fairly constant throughout an entire day,

contam inant levels w ere easily reconciled. February 12, 1999 w as an exam ple o f such a

sam pling day. The NWS wind direction was north-northwest (330°) and the site 3

predom inant wind direction was northwest (322°) (Figures 8.14 and 8.15, pink data

point). The m easured concentration of/?-cym ene a t site 3 was 204 ppt. Sites 4, 5 and


(1 -2 ) had /7-cymene concentrations o f 44.8, 22.0 and 25.4 ppt, respectively. The

concentration gradient falls o ff dow nw ind from the presumed source.

The three days detailed in this section o f the discussion are indicative o f the

w ind 's fickle nature and its influence on the volatile chem icals found in the airshed.

M ost o f the remaining sample days can be categorized with one o f the three exam ples

that have already been described: low wind speed, variable wind direction with higher

speed, consistent wind direction with higher speed. The choice to discount the

im portance o f wind and emphasize average concentrations in pursuing apportionm ent

interpretations is a reflection o f the unreliable correlation between wind data and

concentrations.

8.4 Apportionment o f Volatile Organic Pollutants in the Missoula Valley

Two studies completed in the past decade have examined various sources o f

particulate (PM -10) within the valley [Schm idt, 1996: and Cooper, 1998], however, this

thesis represents the first attem pt to apportion the volatile organic com pounds found in

the M issoula Valley airshed between autom obiles and the Smurfit-Stone C ontainer Pulp

Mill. The volatile organic com pounds apportioned in this study were chosen because o f

their constant presence and relative abundance in the airshed. The evaluation o f various

volatile com pounds as tracers for autom obile em issions has been an ongoing area o f

study since federal regulations called for the phase out o f leaded gasoline. This is the

first application o f /7-cymene as a m olecular tracer for pulp mill emissions. The

justifications supporting the use o f o-xylene and /7-cymene as tracers for the appropriate


114

sources o f pollution are presented in the following two sections. A discussion o f the

three methods o f apportionm ent and results follow.

K4.1 o-Xylene as a Tracer Molecule fo r Fuel Vehicle Emission Sources

Zweidinger et al. [1990] conducted a study in Boise, Idaho, which indicated that

the xylenes, methylcvclohexane, 2-/3-methyIhexane, 2,3,4-/2,2,4-trimethylpentane and 2-

m ethylpentane held prom ise as tracer compounds for fuel/vehicle related emissions.

They examined each prospective tracer's correlation with lead, and fine particulate

potassium. Lead had been proven an excellent tracer for gasoline related emissions;

however, its elimination from fuel was m andated by federal law. Wood com bustion was

traced using soil corrected fine particulate potassium concentrations. Those chem icals

that exhibited the best relationship with lead, while having the poorest with potassium

where tagged as potential tracers. o-Xylene correlated well with lead ( r = 0.852) and

poorly with potassium (r2 = 0.147). Estimates o f the contributions o f volatile organic

chem icals were shown to be the same when using both lead and potassium or o-xvlene

and potassium.

For this study, correlation between o-xvlene and each group 2 compound

(benzene, 2,3,4-trimethylpentane, toluene, ethylbenzene, m-/p-\y lene and naphthalene)

was accom plished by collecting air samples where autom obile emissions would heavily

outweigh other sources. The intersection o f South Ave., Russell and Brooks Streets

(M alfunction Junction) was an obvious choice. The large, steady volume o f traffic at

M alfunction Junction assured that vehicle emissions would dom inate the surrounding air.

The EPA verifies M issoula’s CO compliance at M alfunction Junction because the site has


115

been dem onstrated to consistently have the highest levels. Carbon m onoxide and motor

vehicle emissions are well correlated [Grosjean et al. , 1998; Klouda and Connolly, 1995],

M alfunction Junction also provided an excellent opportunity to ob ta in a composite

sample o f a variety o f autom obiles both idling and moving at speeds encompassing the

EPA 's standard urban driving (19.56 mph) and normal high speed driv ing conditions

(34.79 mph) [Sigsby et al.., 1987]. The samples were collected in th e early summer and

restricted to days when the w eather patterns precluded airshed stagnation. Sample days

w here chosen to avoid forest fires, fireplace woodsmoke and prescribed bum s that might

hav e significantly contributed or altered the levels o f volatile organic pollutants within

the airshed.

Figure 8.16 shows the concentrations o f benzene, toluene, ethylbenzene, m-ip-

xylene, 2,3,4-trimethvIpentane and naphthalene plotted against the concentration o f o-

xylene from the Malfunction Junction samples. A least squares fit, forced through the

ormin. was used to extract a linear correlation factor between each analvte and o-xvlene.

5000 : 4500 j

4000 • 35003000 ^

a.a

ooinCM

2000 ' ' -- ■: =-0 4 ?

1500 ! y = 2 84!Sx

1000 | R- = 0 9765

500 i 0

y * 0.1733* R* = 0 9596

800

700

600

500

400 g-Q»300

| 200

100

100 200 300 400 500 600o-Xytene (ppt)

700 800 900 1000

Toluene • Benzene • m- & p-Xylene - 2,3.4-Trimethytpentane • Ethylbenzene • Naphthalene

Figure 8.16 The linear relationships between fuei/vehicle em ission com pounds and o- xylene determined from the Malfunction Junction sam ples.


16

The validity o f the relationship between each group 2 com pound and o-xylene

was evaluated by com paring the actual and predicted concentrations for site (1+2).

Figure 8.17 plots the actual and predicted toluene concentrations at site (1+2).

4500 - - ------------ --------- —- .............- - -

4000 n = 73

3500

= ; 3000

f 2500 ; \o 2000

® 1500

1000 •

500

0 100 200 300 400 500 600 700 800

o-Xylene (ppt)

• Toluene (Actual) • Toluene (Predicted)

Figure 8 .17 Actual and predicted toluene concentrations from site ( K 2 ) using o-.xyleneas a tracer m olecule.

There is precedence in the literature to use am bient concentrations o f VOCs as a

reasonable average o f a variety o f engine-operating conditions. The emission averages o f

three different autom obile operating conditions com pared very well with the am bient

levels o f volatile chem icals measured along an open roadway. M ugica ct al. [1998]

measured the em issions from automobiles at various stages o f engine operation. Normal

driving conditions were represented by samples taken from a traffic tunnel. Cold start

(high fuel/oxygen) conditions and hot soak (evaporative em issions from a non-running

vehicle at operating tem perature) were also measured. Calculated ratios o f benzene,

toluene, ethylbenzene and m-/p-xylene to o-xylene during all three conditions were

consistent. Benzene to o-xylene ratios displayed the most variability (19%), while the m-

/p-xylene ratios were the least variable (m axim um change o f 0.04%). For benzene, the


117

com pound with the m ost variability with respect to o-xylene ratio, only a three-percent

difference was found betw een the average emissions during the three tested conditions

and those measured in am bient air.

The addition o f ethanol to the gasoline sold in M issoula from November through

February caused a reevaluation in the choice o f tracer molecule originally chosen for the

fuel/vehicle emissions. 2,3,4-Trim ethylpentane was initially slated to track the

fuel/vehicle emissions, but i f used, the final apportionm ent o f these compounds could not

include air samples when oxygenated fuel was burned. Reform ulated gasoline contains

'1 0 percent ethanol by weight. Along with the addition o f oxygenates, winter blended

fuel has an increased am ount o f lighter alkenes and branched alkanes. The aromatic

content is also reduced by 8 to 15 percent [Kirchstetter et al., 1996; Gaffney et al., 1997;

Stump et al.. 1996; K irchstetter et al., 1999], This overall reduction in the total quantity

o f arom atic species is not reflected in their respective ratios. The proportions o f benzene,

toluene, ethyibenzene and m-/p-\y len e to o-xylene in vehicle em issions only vary from

0.17 to 0.63 percent between base and reform ulated fuels [K irchstetter et al., 1996]. The

correlation plots in Figures 8.16 and 8.17 are therefore valid at all times o f the year. In

contrast, the increased am ounts o f branched alkanes in reform ulated gasoline cause an

overestim ation o f the group 2 com pounds during the w inter (Figure 8.18).


118

6000

5000

4000 • •CL

30000)3O 2000

1000

0

&

0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0 16.0 18.0

2,3,4-T rim ethylpefitane (RRR)

• Toluene (Actual) » Toluene (Predicted)

Figure 8.18 Actual and predicted toluene concentrations at site (1-^2) using 2.3.4-trim ethylpentane as a tracer molecule.

The typical operational behavior o f automobiles during the sum m er can differ

slightly from the winter. In the warm sum m er months evaporative em issions increase,

but combustion efficiency also increases. More efficient combustion leads to a decrease

in tailpipe emissions. During the w inter when the ambient temperature is lower,

evaporative emissions decrease. But these colder temperatures result in a reduction o f

combustion efficiency. These variations tend to offset, resulting in rather uniform overall

em issions [McLaren et a l., 1996; Stum p et a /., 1996; Sigsby et al., 1987].

'S‘.4.2 p -( ym ene as a Tracer M olecule fo r M ill Emissions

/7-Cymene was chosen to be the tracer molecule for mill emissions. Data

collected during this study, coupled with the that reported in the literature, suggest that

the surrounding forests are not the m ajor source of/7-cymene to the valley airshed, but

rather that the mill is the largest contributor. This chemical may be useful as a tracer for

forest products industries as a whole.


/?-Cymene is a terpenoid commonly found in the tissues o f coniferous plants,

however, it typically com prises only one percent or less o f the total quantity o f

monoterpenes [Fengel and W egener, 1984], Biogenic em issions from coniferous plants

are dominated by terpenes. A large number o f terpene species have been detected in

these vegetative em issions, but only three to five o f these com pounds are present in

significant percentages. Helmig and Arey [ 1992] collected air samples, near the top o f

the forest canopy in the Sierra N evada Mountains. A large num ber o f compounds were

identified including aldehydes, ketones and 17 terpenoids. They found the most abundant

terpene compounds were a- and P-pinene. Several studies have investigated the

emissions from specific tree species. Eighty percent o f the volatile emissions from a

monterev pine were found to be a - and P-pinene [Juuti et al.. 1990], The ponderosa pine

was found to emit mostly a - and P-pinene as well [Altshuller, 1983]. Isidorov et al.

[1985] quantified terpene em issions from coniferous forests in Russia. They determ ined

that 40-65 percent o f the vegetative emissions were a - and p-pinene. Staudt et al. [1997]

measured the quantities o f terpenes emitted from Pinus pinea. /?-Cymene emissions were

characterized as low, while a-p inene was one o f the five terpenes with the highest overall

emission rates. Konig et al. [1995] measured the em ission rates o f more than 50 volatile

compounds from eleven plant species and reported the m ost abundant terpenes were a -

pinene, P-pinene and sabinene.

The types and proportions o f terpenes detected in a ir samples collected from

western Montana forests were in agreem ent with those found in the literature. Nine

different terpenes as well as several aldehydes and ketones were identified. a-/p*Pinene,

limonene and /7-cymene accounted fo ra total o f 333 ppt, terpenes. The concentration o f


120

p-cym ene was 8.8 ppt and it com prised 2.6 percent o f the four terpenes quantified. The

concentration ratio o f a-p inene to p-cym ene was 26.1.

The mean p-cym ene concentrations in the valley were all higher than those

m easured in the forest and the m axim um concentrations were between 4 and 59 times

greater than the forest samples. The highest concentrations o fp-cym ene in the valley

were m easured at site 3. where a maximum o f 521 ppt and a mean o f 53.1 ppt were

quantified. The ratio o f a-p inene to p-cym ene ranged from 1.48 to 4.45. This location is

near the mill, centered between the Bitterroot and Rattlesnake M ountains. There are no

trees in the vicinity o f the site. The lowest levels o f p-cymene were found at site 5. The

mean concentration o f 9.3 ppt is in line with those measured in the forests around

M issoula. The maximum value o f 36.7 ppt was probably influenced by anthropogenic

contributions. As previously discussed in the section on group 1 chem icals (chapter

8.2.1) the concentration ofp -cym ene decreased with distance from the mill.

p-C ym ene is a m ajor chemical com ponent in the turpentine that is collected from

the digester relief vapors and stored at the mill. Its increase in concentration is at the

expense o f other terpenes. m ainly a-pinene. p-C vm ene is believed to be the result o f a -

pinene decom position during the pulping process [Sjostrom, 1981], Isidorov et al. [1985]

found that at 400° C, a-/p-p inene can undergo isomerization and dehydrogenation to

form p-cym ene and l-m ethyl-4-isopropenylbenzene. During delignification, the thermal

degradation o f a-pinene proceeds under very basic conditions and probably involves the

sulfur species present in the cooking liquor. Dehydrogenation reactions have been

carried out using sulfur or selenium with hydrogen sulfide or selenide produced as a

byproduct [House and O rchin, 1960; M arch, 1977], Silverwood and O rchin [1962]


121

carried out dehydrogenation reactions o f guaiene using aryl diselenides. Copious

amounts o f hydrogen sulfide are produced during the pulping process: some o f which

may be byproducts o f a-p inene dehydrogenation by organosulfur compounds. Phenolic

functional groups from the breakdown o f lignin may act in a manner analogous to

benzoquinone, first rem oving a hydride then a proton to dehydrogenate the a-pinene to

form/7-cymene [M arch, 1977].

The correlation betw een the terpenes and /7-cymene at site 3 were good, as

dem onstrated by a-p inene (Figure 8.19). This linearity' broke down as the distance from

the mill increased. The correlation coefficients for the a-p inene versus /7-cvmene plots at

sites 3. 4, 5. 1 and 2 were 0.8469, 0.7768, 0.0567. 0.5541 and 0.7403, respectively.

100

90y = 0 .1685x-0.6676

Rr = 0.84698070605040302010

0100 200 5000 400 600300

p-Cymene (ppt)

Figure 8.19 a-P inene versus /7-cymene concentrations at site 3.

Linearity drop-off was due to the disparity in the atm ospheric lifetimes o f the terpenes.

/7-Cymene possesses an atm ospheric lifetime o f at least a day. The remaining terpenes:

a-/p-pinene, lim onene, 3-carene and myrcene are much m ore reactive. They are quickly

destroyed by hydroxyl radicals and ozone and typically survive no more than a few


122

hours. The worst correlation between a-pinene and /7-cymene was found at site 5. This

implies that biogenic emissions o f /7-cymene by the surrounding forest were much less

significant than those from the mill and other anthropogenic sources. Furthermore, the

improved correlation at site 2 over site 1 (0.7403 vs. 0.5541) is indicative o f another

source o f terpenes to urban Missoula. The process o f particleboard manufacturing at

Louisiana Pacific is the most likely second source o f these terpene compounds. Its

location and the observed weather patterns could explain the slight increase o f terpene

concentrations found at site (1 -2).

Air samples representative o f the emissions from the Smurfit-Stone Container

Pulp Mill were collected -300 meters southeast o f the waste fuel boiler (WFB). The

same criteria used in choosing the sample days at malfunction junction, applied to the

WFB air samples. M alfunction junction and WFB sampling occurred simultaneously.

Localized w eather patterns, sampling problem s and instrument malfunction affected the

sample analysis. In an effort to maintain accurate representations relative to each source,

only analytical results from days with the highest assurances o f validity at both locations

were used in the apportionment calculations.

-S'. 4.3 Apportionment Using o-Xylene as a Tracer fo r Euel Vehicle Emissions

A utom obile emissions accounted for the largest fraction o f the aromatic

com pounds quantified during this study (Figure 8.20, Table 7.6). N early 100 percent o f

the /W-//7-xylene detected at each site could be linked to vehicle em issions via the 17-

xylene tracer. The percentages o f ethylbenzene attributable to vehicles were also fairly

consistent at all sites, ranging from 87 to 93 percent. Benzene and toluene percentages


correlated with the traffic volumes at the sampling sites. Experim ental error may account

for a portion o f the over-estim ated percentage o f benzene at site (1+2), although the value

seems anom alous since it only occurred in this one instance. About eighty percent o f the

toluene and naphthalene in urban Missoula was estim ated to originate from vehicle

emissions. The rural sites were found to have the greatest percentage o f benzene, toluene

and naphthalene attributable to other sources. The tw o m ost likely sources o f these

chemicals to the airshed are the pulp mill and w oodsm oke/forest fires.

3 4 5 (1+2)

Site■ Benzene □Toluene ■ Ettiylbenzene □ m- & p-Xylene ■ Naphthalene

Figure 8.20 Percentages o f aromatic pollutants attributable to vehicle emissions viao-xylene tracer.

W hile the percentages o f aromatics derived from vehicle emissions are relatively

high among all sites, there is a substantial difference in the absolute concentrations.

Figure 8.21 com pares the concentrations o f arom atic pollutants from automotive sources

at each sampling location. The levels o f aromatic com pounds estim ated in urban

Missoula dw arf the quantities predicted for the rem aining valley sites. The levels at site 3

and 4 are very sim ilar, although the percentages o f benzene and toluene attributed to

vehicles decrease progressively from site 3 to 4 to 5.


124

1000

900800 j700 {■600 f

400 | 300 | 2 0 * !

10 ^ ~ L n3 4 5 (1- 2 )

Site

■ Benzene □Toluene ■ Ethylbenzene O m- & p-Xytene ■ Naphthalene

Figure 8.21 Predicted amounts o f arom atic compounds from vehicle emissions.

One possible explanation for vehicle percentage losses o f benzene and toluene is that

increasing contributions due to wood com bustion from residences near sites 4 and 5

begin to overtake the vehicle sources as the traffic volumes are reduced in the rural areas

o f the v alley. A second possibility is that mill emissions impact these sites to a greater

degree. Both benzene and toluene are stack em issions and the height o f the point o f

em ission may cause site 3 to occasionally be passed over by the plume. Looking

westward from M ount Sentinel on a clear cold day, the plume from the mill can be

observed hanging above the valley. W hen these conditions exist, the higher elevation o f

site 5 m ay allow a m ore direct route o f exposure to mill emissions. The decrease in the

benzene and toluene percentages due to m otor vehicles, coupled with the similarity in

their lev els m easured at sites 3 and 4. suggest other sources are significant at these sites.


125

R.4.4 Apportionment Using p-C 'ymene and o-Xylene as "Dispersion ” ReferenceC 'ompounds

The second m ethod o f apportionment was based on two assumptions: the first is

that all o f the /7-cvmene in the valley originates from the mill and all o f o-xylene in the

valley is em itted from vehicles in urban Missoula. O ther source contributions o f these

chemicals to the airshed must be considered minor. The bases for this assumption have

been discussed throughout this thesis and the first m ethod o f apportionment substantiated

the assumption that urban Missoula is the predominant source o f group 2 compounds.

The dispersion-based contributions o f group 2 chem icals from urban Missoula

were summarized in Table 7.7. Not surprisingly, the values agree with those calculated

using the o-xylene tracer method. Figure 8.22 shows the estim ated percentages o f the

group 2 chem icals that are attributable to the mill. It is important to note that the p-

cvmene "dispersion” reference assumes all o f the group 2 compounds measured at the

waste-fuel boiler reference point originated as mill em issions. This is not the case as any

v ehicle emissions and residential woodsmoke in the mill vicinity would be included as

mill emissions. Therefore, the percentages in Figure 8.22 represent the maximum

possible contributions from the mill.


126

45 ,--------------

3 4 5 (1+2)Site

■ Benzene □ Toluene ■ Ethytbenzene Q m - & p-Xytene ■ Naphthalene

Figure 8.22 Percentages o f arom atic com pounds estimated from the mill.

The trend in percent contribution from the m ill at each sample site makes intuitive

sense. As the distance from the mill increases, its im pact on the valley diminishes. The

percentages o f benzene, toluene, ethylbenzene and /w-//?-xylene from the mill at sites 4

and 5 are very' similar. This can be explained by the proxim ity o f the two sites, the higher

elevation at site 5 and the stack height effect described above. Naphthalene does not

follow the same trend as the other group 2 com pounds. This suggests stack emissions are

not the main source o f naphthalene. The predom inant source o f naphthalene at the mill

may be the diesel engines used to m ove boxcars around the railyard and the nearby

M ontana Rail Link activity.

The mill is not a serious source o f aromatic com pounds to the urban Missoula

area. The aromatic compound contributions to urban M issoula from the mill were all

estim ated to be < 5 percent. While the percentages o f the group 2 com pounds associated

with the mill are important, the actual predicted quantities o f these chem icals help

underscore this point (Figure 8.23). For instance, the am ount o f toluene in urban


127

M issoula from vehicular sources (1,070 ppt) is 38 tim es larger than the predicted

contribution from the mill (27.6 ppt). Even at site 3, only 92 ppt o f toluene could be

linked to mill emissions.

100.0 ; -

90.0 f-----80.0 f—70.0 -60.0 f -

g- 50.040.030.020.0 10.0

0.03 54

Site

■ Benzene □Toluene ■ Ettiylbenzene D m - & p-Xyfene ■ Naphthalene

Figure 8.23 Predicted concentrations o f group 2 compounds attributable to millemissions.

The m aximum possible contribution o f VOCs from the mill decreases with

distance. This makes intuitive sense as dilution and dispersion o f the mill plum e would

result in lower chem ical concentrations with greater distance from the source. The slight

elevation of/7-cym ene m easured in urban M issoula constitutes a disparity in the expected

trend. Since additional source/sources must contribute to the /7-cymene concentration,

the contribution from the mill is over-predicted. Toluene levels associated with the mill

decrease by a factor o f 3.3 between site 3 and site 4. Assuming this rate o f

dispersion/m ixing rem ains constant, the am ount o f toluene at site 5 is predicted to be 17

ppt. This estim ation is in agreem ent with the concentration predicted from the /7-cymene

"dispersion" factor at site 5 (16.2 ppt).


128

The presence o f an additional source o f /7-cymene near urban M issoula can

noticeably perturb the predicted levels o f volatiles assigned to the mill. Applying the p-

cymene concentration fa ll-off rate seen between sites 3 and 4. the toluene concentration

at site (1 -2 ) associated with the mill should be ~ 9 ppt. But the toluene level that was

predicted from the /7-cymene concentration m easured at site (1 -2 ) was 27.6 ppt.

Therefore, the mill contributions o f group 2 chem icals have been over-predicted by - 67

percent because o f the additional /7-cymene source.

A com parison betw een the predicted and m easured m ean values for the group 1

chem icals, shown in Table 7.8, emphasizes the differences betw een the tropospheric

lifetime o f /7-cymene and those o f the remaining terpenes and chloroform . /7-Cvmene's

lifetime is as much as 21 tim es that o f the other terpenes. By the tim e the mill plume

reached each sampling site, the terpene concentrations were reduced not only by

dispersion^dilution, but also by reaction with hydroxyl radicals and ozone. Therefore,

estim ates o f the terpene contributions from the mill are significantly higher than were

measured. Conversely, chloroform has an average atm ospheric lifetim e o f 0.55 years, so

estim ates o f its contribution from the mill and urban M issoula are lower than expected.

The levels o f the group 1 compounds, predicted from the m ill's /7-cvmene

emissions, are shown in Figure 8.24. The predicted values at each site were considerably

higher than those measured. The ratio o f predicted to m easured a-p inene at site 3 was

~5. The a-p inene ratios at sites 4 and 5 were -8 and -1 1 , respectively. This

dem onstrates that the lifetim e-based error between the estim ated and m easured

concentrations for the terpenes increases with distance from the em ission source as

expected.


129

_ 60K

3 4 5 (1*2)

Site

□ Chloroform* Oalpha-Pinene Dbeta-Pmene ■Limonene

Figure 8.24 Estimated amounts o f chloroform and terpenes from the mill.

Although the absolute quantities o f chloroform are underestimated from both

sources, the relative proportions at each site should be fairly accurate (Figure 8.25). This

assumption is reasonable because the tropospheric lifetimes o f each tracer (/7-cvmene and

o-xylene) are similar. The increase in the error with distance from the source

demonstrated above in the predicted to actual a-pinene concentration ratios, must be

taken into consideration when discussing the chloroform apportionment. Because o f this

progressive error, the percentage o f chloroform in urban Missoula due to mill emissions

must be larger than estimated. In accordance, the percentage o f chloroform at site 3

attributed to urban M issoula is underestimated.


130

100908070

Q.

0 u3 4 5 ( 1+2 )

Site

□ Chloroform (urban Missoula) ■ Chloroform (SSC Pulp Mill)

Figure 8.25 Estim ated percentages o f chloroform from the mill and urban Missoulasources.

K-/.5 Apportionment ( 'sing, Linear C 'ombinations

Since the group 2 aromatic VOCs exhibited the highest levels within the airshed,

their origins are the m ost critical for the valley. Linear com binations o f source profiles

representing the mill and automobiles were used to apportion the total aromatic

compounds (TAC) a t each sample site. The results from this method, along with a

summation o f the results from the other two methods are provided for comparison in

The results from the linear combinations method agreed with both the o-xylene

correlation factor and the dispersion reference methods. According to the linear

com binations m ethod, vehicle emissions accounted for 97 percent o f the aromatic VOCs

in urban Missoula. The o-xylene correlation and dispersion reference methods estimated

97 and 99 percent o f the TACs were due to vehicles, respectively. Only two percent o f

the aromatic VOCs in urban M issoula could be attributed to the mill.

Table 7.9


131

At site 3, the sampling site closest to the mill, vehicles were estimated to

contribute 73 to 89 percent o f the arom atic com pounds. M ill emissions were estim ated to

account for 34 percent. This percentage is a m axim um value, as the source profile from

the mill inevitably contained vehicle emissions. The o-xylene correlation factor m ethod

estim ated that only 12 percent o f the arom atics were attributable to any source other than

autom obiles. Thus, the m ill's actual contributions to the arom atic VOCs at site 3

probably falls within the range o f 12 to 34 percent.

The linear combination and the dispersion reference methods estimation o f the

m ill's impact on sites 4 and 5 were within one percent o f each other. The mill was

determ ined to be responsible for eight to nine percent o f the aromatic chemicals at those

sites.

o-Xvlene was highly correlated with vehicle em issions in the Missoula Valley,

however, in order to ascertain the m aximum erro r that could be associated with the

apportionm ent, a slight variation on the o-xylene correlation method was employed.

Seventy eight percent o f the o-xylene in an air sam ple consisting o f equal volumes o f

vehicle exhaust and woodsmoke, would be due to the vehicle exhaust. Thus, even in a

worse case scenario, vehicle emissions would account for the majority o f the TACs in the

valley airshed.

2.3.4-Trimethylpentane was used as another check on the apportionment results.

2,3,4-Trim ethylpentane is not associated with the m ill's VOC emissions. It was used in

place o f o-xylene in the correlation factor method. The results were in line with the other

apportionm ent methods and, as expected, were slightly higher than estimates obtained

using o-xylene. The reason for this was discussed in section 8.4.1. The 2,3,4-


tri methyl pentane correlation factor method estim ated that 78 percent o f the TACs at site

3 were associated with vehicles. The 22 percent o f the aromatic chemicals estim ated to

com e from another source falls w ithin the range o f mill contributions established by the

other m ethods o f apportionment.


Chapter 9 Conclusions and Considerations for Future Study

9.1 Conclusions from the Airshed Study

The results from this quantitative investigation o f the valley airshed clearly

associated the highest concentrations o f VOCs with urban air. The concentrations o f

arom atic compounds in urban M issoula were four and six times as great as those found at

sites 3 (m ill) and 5 (cleanest), respectively. When all quantified chemicals, including

terpenes. are compared, urban levels o f VOCs were still nearly two and three times those

measured at sites 3 and 5, respectively. The chemical apportionment estimates indicated

m otor vehicles were the dominant VOC source in the airshed. Across the valley, vehicle

em issions were estimated to contribute on average, between 56 and 100 percent o f VOCs.

Another M issoula Valley study estim ated vehicles w'ere responsible for 60 percent o f the

CO [Schmidt, 1992]; w'hile in other urban areas, as much as 98 percent o f the CO has

been linked to vehicles [Grosjean et a/., 1998], The VOC results presented in this thesis

are within the range o f estimated CO contributions from motor vehicles determined from

these other studies.

The population o f the M issoula Valley may be incurring a small health impact

from these levels o f pollution. The mean concentrations o f benzene found throughout the

valley airshed, especially in urban Missoula, pose a one in one hundred thousand

increased risk for developing cancer. The cancer risk due to the concentration o f

chloroform in the airshed falls between one in one million, and one in one hundred

thousand.


134

These findings lead to the conclusion that the m ost effective strategy for air

quality improvement would target m otor vehicle emissions. Based on 1994 traffic

volumes, there are over 300 m illion vehicle m iles traveled in urban M issoula each year.

The city' can be thought o f as a m otor vehicle point source with a sm okestack nearly four

feet in diameter, emitting exhaust and evaporative emissions 24 hours a day, 365 days a

year. The concentrations o f these pollutants will grow along with the population o f

Missoula. Therefore, any strategy to improve air quality should focus on decreasing the

vehicle miles driven, reducing autom otive VOC em issions and the use o f alternate fuels.

9.2 Improvements to the V alley Airshed Study and Modeling

This project could be im proved in several ways. Ethene, propene and acetylene

have all been identified as tracers for the exhaust portion o f vehicle emissions.

Analyzing for these compounds w ould allow the separation o f vehicle evaporative and

exhaust emissions. In order to analyze for these chem icals along with the other VOCs,

the gas chromatograph would need a 105-meter m egabore capillary colum n and/or a

cryogenically cooled oven. Although these options are presently available, none o f them

were when this project began. A nother piece o f information that would be useful for

chemical source apportionm ent is the ratio o f ,4C to l-C present in the hydrocarbons

collected from the valley air. This could determ ine the fractions due to fossil fuels and to

wood burning. An accelerator m ass spectrom eter and extensive sample preparation are

required for this analysis.

The evaluation of/?-cym ene as a mill em ission tracer, beyond that described in

this thesis, is less straightforward. M ill VOC em issions would not be distinguishable


135

from wood com bustion sources by UC. The highly reactive nature o f terpenes would

invalidate the source profiles, thus a chemical mass balance technique would not be

helpful in apportioning terpenes between biogenic and pulp mill sources. A source

oriented m odel that included the rates o f destruction o f /?-cvmene and the other terpenes

would be the best approach. A favorable com parison between predicted and actual

concentrations o f /7-cvmene at various receptor sites would provide further evidence for

its use as a tracer molecule.

Sev eral artificial neural network models o f the airshed using the quantitative and

m eteorological data were created. Not surprisingly, all produced unsatisfactory results.

The neural network approach could lead to an excellent airshed model; however, this

would require considerably more time, work and money. The current meteorological

data are inadequate and my quantitative data set is too small. To create a model inclusive

o f the valley weather, several weather stations placed in specific valley locations must be

constructed. Furtherm ore, each station should be equipped with two anemom eters, one

placed high (above 30 feet) and one low (-1 5 feet), to encom pass both upper and lower

wind patterns.

The ANN models created had too many input param eters for the small num ber o f

facts presented. To satisfy this problem valley sam ple sites could be reduced to three.

The waste fuel boiler site, M alfunction Junction and a third centrally located site.

Ideally, sam ple collection could occur at least every other day and a duplicate would be

collected at each site to insure against instrument m alfunction and accidental sample loss.


itI! 136

9.3 investigation o f Biogenic Emissions and Acetone Levels

The seasonal change in ambient acetone concentration observed during this study,

was an intriguing observation. Calculations o f tropospheric terpene concentrations based

on the published biogenic emission rates exceed the actual measured concentrations.

Acetone concentrations in the surrounding forests (1,870 ppt) were slightly elevated

above those m easured in the valley (1,200 ppt) around the same date. Acetone is

produced as a relatively stable product from the oxidation o f terpenoids. M any studies

have exam ined the products o f terpenes reacted with ozone, hydroxyl or nitrate radicals.

The reaction conditions are designed to look for primary' products. In the atm osphere

these reactions do not stop after the primary products are formed. Furthermore, as in the

case o f pinon aldehyde from a-pinene, the oxidative products can be even m ore reactive

than the parent molecule. Calogirou et a/. [1997] looked for terpene oxidation products

in the forest air, and found only nopinone. Nopinone has a longer lifetime than its

precursor terpenes. They suggested deposition or heterogeneous reactions on aerosols

might be m ore important for these chemicals than currently thought. This is an

interesting area o f research, which will lead to a better understanding o f tropospheric

chemistry.


137

L it e r a t u r e C it e d

Adams. T.N., Frederick, J., Kraft Recovery Boiler Physical and Chem ical Processes, The American Paper Institute, Inc., 1988.

Allen, M.G., Butler, C.T., Johnson, S.A., Lo, E.Y., Russo, F., Im aging Neural Network Combustion Control System for Utility Boiler Application, Combust. Flame, Vol. 94, 1993, pp'. 205-214.

Altshuller, A.P., Review: Natural Volatile Organic Substances and Their Effect on Air Qualit\r in the United States, Atmospheric Environment, Vol. 17, No. 11, 1983. pp. 2131-2165.

Atkinson. R., Gas phase tropospheric chemistry' o f organic com pounds: A Review.Atmos. Environ., Vol. 24A, No. I, 1990, pp. 1-41.

Atkinson. R.. Kinetics and mechanisms o f the gas-phase reactions o f NO; radical withorganic compounds, J. Phvs. Chem. Ref. Data, Vol. 20, No. 3. 1991, pp. 459-507.

Arey. J.A., Crowley, D.E., Crowley, M., Resketo, M., Lester, J., Hydrocarbon Emissions from Natural Vegetation in California's South Coast Air Basin, Atmospheric Environ., Vol. 29. No. 21, 1995. pp. 2977-2988.

Arnold. F., Schneider. J., Gollinger, K., Schlager, H., Schulte, P., Hagen, D., W hitefield, P., van Velthoven. P., Observation o f upper tropospheric sulfur dioxide and acetone pollution: Potential implications for hydroxyl radical and aerosol formation. Geophysical Research Letters, Vol. 24. No. 1, 1997, pp. 57-60.

Arnold, F., Burger, V., Droste-Fanke, B., Grimm. F., Krieger. A., Schneider, J., Stilp, T., Acetone in the upper troposphere: Impact on trace gases and aerosols.Geophysical Research Letters, Vol. 24, No. 23, 1997, pp. 3017-3020.

Aucott. M L., McCullough, A., Graedel, T.E., Kleiman, G„ M idgley, P., Yi-Fan Li, Anthropogenic Emissions o f Trichloromethane and Chlorodofluoromethane: Reactive Chlorine Emissions Inventory, Journal o f Geophysical Research, Vol. 104, No. D7, 1999, pp. 8405-8415.

Baines, G.H., Hayes, R.L., Stabell, J.L., Predicting Boiler Em issions With Neural Networks, Tappi Journal, Vol. 78, No. 6, 1995, pp. 167-174.

Bayer, C.W., Black, M.S., Thermal Desorption/Gas Chromatographic/M assSpectrometric Analysis o f Volatile Organic Compounds in the Offices o f Smokers and Nonsmokers, Biomedical and Environmental Mass Spectrometry, Vol. 14, 1987, pp. 363-367.


138

Benjamin, M .T., W iner, A.M., Estimating the Ozone-Form ing Potential o f Urban Trees and Shrubs, Atmospheric Environment, Vol. 32, No. 1, 1998, pp 53-68.

Bertin, N., Staudt, M., Hansen, U., ei al., D iurnal and Seasonal Course o f M onoterpene Em issions from Quercus Ilex Under Natural Conditions - Application o f Light and Tem perature Algorithms, A tm ospheric Environment, Vol. 30, No. SI, 1997, pp. 135-144.

Beyer. W.H., Standard M athematical Tables, CRC Press, 27th Ed., 1984, p. 546.

Boznar, M., Lesjak, M., M lakar, P., A Neural Network-Based Method for Short Term Predictions o f Ambient S 02 Concentrations in Highly Polluted Industrial Areas o f Com plex Terrain, Atmos. Environ., Vol. 27B, 1993, pp. 221-230.

Bromenshenk, J.J., Smith G.C., King B E., Seccom b, R.A., Alnasser, G.H., Loeser, M.R., Henderson, C.B., W robel, C.L., New and Improved Methods for M onitoring Air Quality and the Terrestrial Environment: Applications at Aberdeen Proving Ground— Edgewood Area. Annual Report 1998, U.S. Army Center for Environm ental Health Research, Ft. Detrick, MD. U.S.A. Contract DAMA 17-9 5-C-5072, 1998.

Brown. C.A.. Grace, T.M .. Lien, S.J., Clav. D.T., Char Bed Burning Rates -Experim ental Results, Kraft Recovery O perations Short Course, Tappi Press,1990. pp. 145-151.

Calogirou. A., Larsen, B.R., Kotzias, D., G as-Phase Terpene Oxidation Products: A Review, Atmospheric Environment, Vol. 33, 1999, pp. 1423-1439.

Calogirou, A., Duane, M., Kotzias, D., Lahaniati. M., Larsen, B.R.,Polyphenylenesulfide, Noxon®, An O zone Scavenger for the Analysis o f O xygenated Terpenes in Air. A tm ospheric Environment. Vol. 31, No. 17, 1997, pp.*2741-2751.

Chan, Y.C., Simpson, R.W., M ctainsh, G.H., et a !., Source Apportionment o f Visibility Degradation Problems in Brisbane (A ustralia) Using the Multiple Linear Regression Techniques, Atmospheric Environm ent, Vol. 33, 1999, pp. 3237- 3250.

Chen, W.C., Chang, N., Com bined Fuzzy Logic and Neural Network for Combustion Control o f M unicipal Incinerators, Proc. Int. Conf. Solid Waste Tech. Mgmt, (Univ. o f Pennsylvania, Philadelphia, 1996).

Cheng-Yu Ma, Skeen, T., Dindal, A., Bayne, C „ Jenkins, R., Performance Evaluation o f a Thermal Desorption/Gas Chrom atographic M ethod for the Characterization o f W aste Tank Headspace Samples, Environ. Sci. Technol., Vol. 31, No. 3, 1997, pp. 853-859.


139

Cooper. J.A ., PM Issues in M issoula, Montana. A Review and Evaluation with Recom m endations Relevant to Kraft Recovery Boiler O pacity Rules,Cooper Environmental Services, 1998.

Corchnov, S B., Atkinson, R., Kinetics o f the Gas Phase Reactions o f OH and NOSRadicals with 2-Carene, 1,8-Cineole, p-Cymene, and Terpinolene. Environ. Sci. Technol., Vol. 24, No. 10. 1990, pp. 1497-1502.

Cox. R., Patrick, K., Chant, S., Mechanism o f atm ospheric photooxidation o f organic com pounds. Reactions o f alkoxy radicals in oxidation o f n-butane and simple ketones. Environ. Sci. Tech., Vol. 15, No. 5. 1981, pp. 587-592.

Drewitt, G .B., Curren, K., Steyn, D.G., Gillespie, T.J., Niki, H., Measurement o fBiogenic Hydrocarbon Emissions From Vegetation in the Lower Fraser Valley. British Colum bia, Atm ospheric Environment, Vol. 32, No. 20, 1998. pp. 3457- 3466.

Fengel. D., W egener, G., Wood: Chemistry. U ltrastructure, and reactions, W alter de Gruyter. 1984, pp. 182-222.

Fehsenfeld. F., ct a /., Emissions o f Volatile Organic Com pounds from Vegetation and the Im plications for Atmospheric Chemistry, G lobal Biogeochemical Cycles. Vol. 6. No. 4, 1992, pp. 389-430.

Finlayson, B.J.. Pitts, J.N ., Atm ospheric Chemistry': Fundam entals and Experimental Techniques. W iley-Interscience, 1986.

Frank, H., Frank, W., Neves, H.J.C.. A irborne Ci and C :-H alocarbons at FourRepresentative Sites in Europe, Atmospheric Environment, Vol. 25A, No. 2.1991, pp. 257-261.

Fre, R.D., Bruynseraede, P., Kretzschmar, J.G., A ir Pollution Measurements in Traffic Tunnels, Environmental Health Perspectives, Vol. 102, Supplement 4, No. 10, 1994, pp. 31-37.

Fujita, E.M., W atson, J.G., Chow, J.C., M agliano, K.L., Receptor Model and Emissions Inventory Source Apportionm nets o f N on-m ethane Organic Gases in C alifornia 's San Joaquin Valley and San Fransisco Bay Area, A tm ospheric Environment, Vol. 29, No. 21, 1995, pp. 3019-3035.

Gaffney. J.S ., Marley, N.A., Martin, R.S., Dixon, R.W., Potential A ir Quality Effects o f Using Ethanol-Gasoline Fuel Blends: A Field Study in Albuquerque, New M exico, Environm ental Science and Technology, Vol. 31, No. 11,1997, pp. 3053-3061.


140

Gertler, A.W., Fujita, E.M ., Pierson, W.R.. W ittorff, D.N., Apportionm ent o f NMHC Tailpipe vs N on-Tailpipe Emissions in the Fort M cHenry and Tuscarora Mountain Tunnels, Atmospheric Environment. Vol. 30. No. 12. 1996, pp. 2297- 2305.

Glen. W.G., Zelenka, M .P., Graham , R.C., Relating M eteorological Variables and Trends in Motor Vehicle Emissions to Monthly Urban Carbon M onoxide Concentrations, Atmospheric Environm ent, Vol. 30, No. 24, 1996, pp. 4225-4232.

Grace, T.M ., Cameron, J.H., Clay, D.T., Role o f the Sulfate/Sulfide Cycle in Char Burning-Experimental Results and Implications, Kraft Recovery Operations Short Course, Tappi Press, 1990, pp. 153-161.

Grace. T.M ., Walsh, A., Jones, A., Sumnicht. D., Farrington. T., A Three-Dimensional Mathematical M odel o f the Kraft Recovery Furnace. Kraft Recovery Operations Short Course. Tappi Press, 1989, pp. 173-180.

Grosjean. E.. Grosjean, D., Rasmussen, R.A., Ambient Concentrations, Sources,Emission Rates and Photochemical Reactivity o f C ; - C |(. Hydrocarbons in Porto Alegre, Brazil, Environ. Sci. Technol.. Vol. 32, No. 14, 1998, pp. 2061-2069.

Grosjean, D., W illiams, E.L., Grosjean, E., Andino. J.M ., Seinfeld, J.H., Atmospheric Oxidation o f Biogenic Hydrocarbons: Reaction o f Ozone with b-Pinene, D- Limonene and trans-Caryophyllene. Environ. Sci. Technol.. Vol. 27, No. 13.1993, pp. 2754-2758.

Guenther. A., Zim m erm an, P., W ildermuth, M., Natural Volatile Com pound Emission Rate Estimates for the U.S. Woodland Landscapes, Atmospheric Environment, Vol. 28. No. 6. 1994, pp. I 197-1210.

Haahtela, T., Marttila, O., Vilkka, V., Jappinen, P., Jaakkola, J.K , The South Karelia Air Pollution Study: Acute Health Effects o f M alodorous Sulfur Air Pollutants Released by a Pulp M ill, American Journal o f Public Health, Vol. 82, No. 4,1992, pp. 603-605.

Hallquist, M., Wangberg, I., Ljungstrom, E., Barnes, I., Becker, K.H., Aerosol andProduct Yields from N 0 3 Radical-Initiated Oxidation o f Selected Monoterpenes, Environ. Sci. Technol., Vol. 33, No. 4, 1999, pp. 553-559.

Hauk, A., Sklorz, M., Bergman G., Hutzinger, O., Analysis and Toxicity Testing o f Combustion Gases I. A New Sampling Unit for Collection o f Combustion Products, Journal o f Analytical and Applied Pyrolysis, Vol. 28, 1994, pp. 1-12.

Hazard, S., Desorption Efficiencies o f Air Samples Collected on Thermal Desorption Tubes, Supelco Publication, Pittsburgh Conference, 1994.


141

Heavner, D.L., Ogden, M.W., Nelson, P.R., M ultisorbent Thermal Desorption/Gas Chromatography/M ass Selective Detection M ethod for the Determination o f Target Volatile Organic Com pounds in Indoor Air, Environ. Sci. Technol., Vol. 26, No. 9, 1992, pp. 1737-1746.

Helmig, D., Arey, J., Organic Chem icals in the Air at W hitaker's Forest/Sierra Nevada Mountains, California, Science o f the Total Environ., Vol. 112, No. 2. 1992, pp. 233-250.

Helmig, D., Greenburg, J.P., Autom ated in situ gas chromatographic-mass spectrom etric Analysis o f ppt level volatile organic trace gases using multistage solid-adsorbent Trapping, Journal o f Chrom atography A, 677, 1994, pp. 123-132.

Henry, R.C., Lewis, C.W., Collins, J.F., Vehicle-Related Hydrocarbon SourceCompositions from Ambiet Data: The GRACER/SAFER Method, Environ. Sci. Technol., Vol. 28, No. 5, 1994, pp. 823-832.

House, W.T., Orchin, M., A Study o f the Selenium Dehydrogenation o f Guaicol and Related Compounds. Selenium as a Hydrogen Transfer Agent, Journal o f the American Chemical Society, Vol. 82, No. 2, 1960, pp. 639-642.

Hupa. M., Combustion o f Black Liquor Droplets, Kraft Recovery Operations Short Course, Tappi Press, 1990, pp. 135-144.

Huang, X., Olmez, L, Aras, N.K., Gordon, G., Emission o f Trace Elements from M otor Vehicles: Potential M arker Elem ents and Source Composition Profile, A tmospheric Environment, Vol. 28, No. 8, 1994, pp. 1385-1391.

Isidorov, V.A., Zenkevich, LG., Ioffe, V.. Volatile Organic Compounds in theAtmosphere o f Forests, A tm ospheric Environment, Vol. 19, No. 1, 1985. pp. 1-8.

Janka K., Ruohola, P., Siiskonen, P., Tam m inen, A., A Comparison o f Recovery' Boiler Field Experiments Using Various NOx Reduction Methods, Tappi Journal,Vol. 81, No. 12, 1998, pp. 137-141.

Jenkins. R.A., Palausky, A., Counts, R .W ., et al., Exposure to Environmental TobaccoSmoke in Sixteen Cities in the United States as Determined by Personal Breathing Zone Air Sampling, Journal o f Exposure Analysis and Environmental Epidemiology', Vol. 6, No. 4, 1996, pp. 473-502.

Johnson, N.L., Leone, F.C., Statistics and Experimental Design in Engineering and the Physical Sciences, vol. 1, 2nd Ed., John Wiley and Sons, 1964, pp. 258-264.

Juuti, S., Arey, J., Atkinson, R., M onoterpene Emission Rate Measurements from a Monterey Pine, Journal o f Geophysical Research, Vol. 95, No. D6, pp., 1990, 7515-7519.


142

Keith, L., EPA Method TO -2 - VOCs in Ambient A ir by Carbon M olecular Sieve and GC/MS, Handbook o f A ir Toxics, Lewis, 1995, pp. 53-54.

Kirchstetter. T.W., Singer, B.C., Harley, R.A., et al., Impact o f California Reformulated Gasoline on M otor V ehicle Emissions. 2. Volatile Organic Com pound Speciation and Reactivity, Environ. Sci. Technol., Vol. 33, No. 2, 1999, pp. 329-336.

Kirchstetter, T.W., Singer, B.C., Harley, R.A.. et al., Impact o f Oxygenated Gasoline Use on California Light-Duty Vehicle Emissions, Environ. Sci. Technol.. Vol. 30. No. 2. 1996, pp. 661-670.

Klouda, G.A., Connelly, M .V ., Radiocarbon M easurem ents to Q uantity Sources o fAtmospheric Carbon M onoxide in Urban Air. A tm ospheric Environment, Vol. 29. No. 22. 1995, pp. 3309-3318.

Konig. G.. Brunda, M., Puxbaum , C H., Hewitt, C., Duckham. S.C.. RelativeContribution o f O xygenated Hydrocarbons to the Total Biogenic VOC Emissions o f Selected M id-European Agricultural and Natural Plant Species. Atmospheric Environment. Vol. 29, No. 8, 1995, pp. 861-874.

Kotzias, D., Fvtianos. K.. Geiss, F., Reaction o f M onterpenes with Ozone, Sulfur Dioxide And Nitrogen D ioxide-G as Phase Oxidation o f S 0 2 and Formation o f Sulfuric Acid, Atmos. Environ., Vol. 24A. No. 8, 1990, pp. 2127-2132.

Lamb, B.. Guenther, A., Gay, D., Westburg, H., A National inventory o f BiogenicHydrocarbon Em issions. Atmospheric Environment, Vol. 21. No. 8, 1987, pp. 1695-1705.

Lamb, B.. Westberg, H., A llw ine, G., Quarles, T., Biogenic Hydrocarbon Em issions from Deciduous and C oniferous Trees in the United States, Journal o f Geophysical Research. Vol. 90, No. D l, 1985, pp. 2380-2390.

Lamb, B.K.. et al.. Developm ent o f Atmospheric Tracer M ethods to Measure Methane Emissions from Natural Gas Facilities and Urban Areas, Environ. Sci. Technol., Vol. 29, No. 6, 1995. pp. 1468-1479.

Larsen et al.. Sampling and Analysis o f Terpenes in Air: An Interlaboratory Comparison, Atmos. Environ., Vol. 31, 1997, pp. 35-49.

Lee, L.S., Hagwall. M., Delflno, J.J., Rao, P.S., Partitioning o f Polycyclic Arom atic Hydrocarbons from Diesel Fuel into W ater, Environ. Sci. Technol., Vol. 26,No. 11, 1992, pp. 2104-2110.

Lindskog, A., Potter, A., T erpene Emission and Ozone Stress, Chem osphere, Vol. 30,No. 6. 1995, pp. 1171-1181.


143

Lovventhal, D.H., Zielinski, B., Chow, J.C ., et al., Characterization o f Heavy-Duty Diesel Vehicle Emissions, Atmospheric Environment, Vol. 28, No. 4, 1994, pp. 731-743.

M arch. J.. Advanced Organic Chemistry: Reactions, M echanisms and Structure, 2nd Ed., M cgraw Hill, New York, 1977, pp. 1077-1078.

M cLaren, R., Singleton, D.L., Lai, J.Y .K ., et al., Analysis o f M otor Vehicle Sources and their Contribution to Ambient Hydrocarbon Distributions at Urban Sites in Toronto During the Southern O ntario Oxidants Study, Atm ospheric Environment, Vol. 30, No. 12, 1996. pp. 2219-2232.

M ikkanen. M.P.. Kauppinen, E.I., Jokiniem i, J.K., Sinquefield, S.A., Frederick. W.J., Bimodal Fume Particle Size D istributions From recovery Boiler and Laboratory Scale Black Liquor Combustion, Tappi Journal, Vol. 77, No. 12, 1994, pp. 81-84.

M ilosavljevic, N., Heikkila, P., M odeling a Scrubber Using Feed-Forward Neural Networks, Tappi Journal, Vol. 82, No. 3. 1999. pp. 197-201.

M iyanishi. T., Shimada. H., Using Neural Networks to Diagnose Web Breaks on a Newsprint Paper Machine, Tappi Journal, Vol. 81, No. 9, 1998, pp. 163-170.

M ugica. V., Vega, E., Arriagw, J.L., Ruiz. M.E., Determination o f M otor VehicleProfiles for Non-M ethane Organic Compounds in the M exico City Metropolitan Area, Journal o f the Air and W aste M anagem ent Assoc., Vol. 48, No. 11, 1998.pp. 1060-1068.

M ukund, R., Kelly, T.J.. Spicer, C.W., Source Attribution o f Am bient Air Toxic and other VOCs in Columbus, Ohio, Atmospheric Environm ent, Vol. 30. No. 20,1996, pp. 3457-3470.

Opitz. D.. Prabu, G., Smith, G., W robel, C.L., Neural Network Ensembles for Control, IASTED International Conference on Control and Applications, Banff, Canada, 1999, pp. 464-467.

Ott. R.L., An Introduction to Statistical M ethods and Data Analysis, Duxburv Press,1993, pp. 234-242, 260-273, 354-361.

Paterson, K.G, Sagady, J.L., Hooper, D.L., et al., Analysis o f Air Quality Data Using Positive M atrix Factorization, Environ. Sci. Technol., Vol. 33. No. 4, 1999, pp. 635-641.

Rege, M .A., Tock, R.W., A Simple Neural Network for Estimating Emission Rates o f Hydrogen Sulfide and Ammonia from Single Point Sources, J. Air W aste M anage. Assoc., Vol. 46, 1996, pp. 953-962.


144

Reich, S.L., G om ez, D.R., Dawidowski, L.E., Artificial Neural Network for theIdentification o f Unknown Air Pollution Sources, Atmospheric Environment, Vol. 33, 1999, pp. 3045-3052.

Reinschmidt, K.F., Ling, B., Neural Networks With Multiple-State Neurons for Nitrogen Oxide Em ission M odeling and Advisory Control. IEEE Int. Conf. Neural Networks Conf. Proc. V ol. 6. 1994, pp. 3834-3839.

Roadnight. C .M ., Balls, G.R., Mills, G.E., Palmer-Brown, D., Modeling Com plex Environmental Data, IEEE Trans. On Neural Networks, Vol. 8, 1997, pp. 852-861.

Rumph, H P., Rudd, J.B., Design Guidelines for Continuous Emissions monitors andPredictive Emissions M onitors, Tappi Journal, Vol. 81, No. 3, 1998, pp. 133-138.

Sack, T.M., Steele, D.H., Hammerstrom, K., Remmers, J., A Survey o f HouseholdProducts for Volatile Organic Compounds, Atmospheric Environment, Vol. 26A, No. 6. 1992, pp. 1063-1070.

Schaur. J.J., Rogge, W.F., Hildemann, L.M., Source Apportionment o f Airborne Particulate M atter Using Organic Compounds as Tracers, Atmospheric Environment, Vol. 30, No. 22, 1996, pp. 3837-3855.

Schmidt, B.. Chem ical Mass Balance Source Apportionment o f Missoula, M ontana 1995/1996 W inter Suspended Particulate Matter, Missoula City - County Health Department, 1996.

Schmidt, B., M issoula Carbon M onoxide Saturation Study, M.S. Thesis, The University o f M ontana, 1992.

Seinfeld. J.H., Pandis, S.N., Atmospheric Chemistry and Physics: From Air Pollution to Clim ate Change, John Wiley and Sons, Inc., 1998, pp.234-330.

Sheffield, A.E., Gordon, G.E., Curries, L.A., et al., Organic, Elemental and IsotopicTracers o f A ir Pollution Sources in Albuquerque, NM, Atmospheric Environment, Vol. 28, No. 8, 1994, pp. 1371-1384.

Sigsbv, J.E., Tejada, S., Ray, W., Volatile Organic Emissions from 46 IN-Use Passenger Cars, Environmental Science and Technology, Vol. 21, No. 5, 1987, pp. 466-475.

Silverwood, H.A., Orchin, M.. Aromatization Reactions with Selenium and ArylDiselenides, Journal o f the American Chemical Society, Vol. 27, No. 10, 1962, pp. 3401-3404.


145

Sim oneit, B.R.T., Rogge, W.F., M azurek, M .A., Standley, L.J., Hildemann, L.M., Cass, G.R.. Lignin Pyrolysis Products, Lignans, and Resin Acids as Specific Tracers o f Plant Classes in Emissions from B iom ass Com bustion, Environ. Sci. Technol., Vol. 27, No. 12, 1993, pp. 2533-2541.

Singh, H.B., Kanakidou, M., Crutzen, P.J., Jacob, D.J., High Concentrations andPhotochemical Fate o f Oxygenated H ydrocarbons in the Global Troposphere, Nature, Vol. 378, No. 6552, 1995, pp. 50-54.

Sjostrom, E., W ood Chemistry: Fundam entals and Applications, Academic Press, 1981, pp. 83-97, 104-144.

Smook, G.A., Handbook for Pulp and Paper Technologists, Angus W ild Publications, Second Edition, 1992.

Someshwar, A., Dillard, D.S., Hinton, S.. C om pilation o f Air Toxic and Total Hydrocarbon Emissions Data for Sources a t Chemical W ood Pulp Mills,Volumes I and II, Technical Bulletin No. 701, O ctober 1995.

Song. X.H.. Hopke, P.K., Solving the Chem ical M ass Balance Problem Using anArtificial Neural Network. Environ. Sci. Technol., Vol. 30, No. 2, 1996, pp. 5 3 1 - 535.

Sricharoenchaikul, V., Frederick, J R., G race, T .M ., Sulphur Species Transform ations During Pyrolysis o f Kraft Black Liquor, Journal o f Pulp and Paper Science,Vol. 23. No. 8, pp. 394-400.

Staudt. M., Bertin, N., Hansen, U., Seufert. G ., et a l., Seasonal and Diurnal Patterns o f M onoterpene Emissions from Pinus Pinea Under Field Conditions, Atmospheric Environment, Vol. 31, No. SI, 1997, pp. 145-156.

Stump. F.,D., Knapp. K.T., Ray, W.D., Influence o f Ethanol-Blended Fuels on theEm issions from Three Pre-1985 Light-D uty Passenger Vehicles, Journal o f the Air and W aste M anagement A ssociation, Vol. 46, 1996, pp. 1149-1161.

Strom well, A.M., Petersson, G., Photooxidant-Form ing M onoterpenes in Air Plumes from Kraft Pulp Industries. Environ. Pollution, Vol. 79, 1993, pp. 219-223.

Swietlicki, E., Puri, S., Hansson, H., Urban A ir Pollution Source A pportionm ent Using a Com bination o f Aerosol and Gas M onitoring Techniques, Atmospheric Environment, Vol. 30, No. 15, pp. 2795-2809.

Tavares, A.J., Tran, H., Reid, T.P., Effect o f C har Bed Tem perature and Tem perature Distribution on Fume Generation in a K raft Recovery Boiler, Tappi Journal,Vol. 81, No. 9, 1998, pp. 134-138.


146

Tavares, A., Tran, H„ Field Studies on Fume Chem istry and Deposition in Kraft Recovery Boilers, Tappi Journal, Vol. 80, No. 12, 1997. pp. 117-122.

Tessier. P.. Yu Qian, Dunont, G.A.. M odeling a W ood Chip Refiner Using Artificial Neural Networks, Tappi Journal, Vol. 78, No. 6, 1995, pp. 167-174.

UATW , Unified Air Toxics W ebsite, U.S. EPA,http://ww w .epa.gov/ttn/uatw /hapindex.htm l, 1998

Verril, C.L., W essel, R.A., Detailed Black Liquor Drop Com bustion Model forPredicting Fume in Kraft Recovery Boilers. Tappi Journal, Vol. 81. No. 9. 1998. pp. 139-148.

Vinckier. C.. Com pem olle, F.. Saleh, A .M ., Van Hoof, N., Van Hees, 1.. Product Yields o f the a-P inene Reaction with Hydroxyl Radicals and the Implication on the Global Emission o f Trace Com pounds in the Atmosphere, Fresenius Environmental Bulletin, Vol. 7, 1998, pp. 361-368.

Wag. K.J.. Reis. V.J.. Frederick, W .J.. Grace. T.M .. M athematical Model for the Release o f Inorganic Emissions During Black Liquor C har Combustion. Tappi Journal. Vol. 80~ No. 5, 1997, pp. 135-145.

W alsh, A.R., La Fond, J.F., Fom etti, M .A., Legault, T., Optim ization o f Combustion Air Delivery, Tappi Journal, Vol. 81, No. 9. 1998. pp. 125-133.

W essel, R.A.. Parker, K.L.. Verril, C.L., Three-Dim ensional Kraft recovery Furnace Model: Implementation and Results o f Improved Black Liquor Combustion M odels, Tappi Journal, Vol. 80. No. 10, 1997, pp. 207-211.

W esterholm , R., Egeback, K.E., Exhaust Em issions from Light and Heavy DutyVehicles: Chemical Com position, Impact o f Exhaust after Treatment, and Fuel Param eters, Environmental Health Perspectives, Vo. 102. Supplement 4, No. 10, 1994, pp. 13-23.

W ienke, D., Hopke, P.K., Visual Neural M apping Technique for Locating Fine Airborne Particulate Sources, Environ. Sci. Technol., Vol. 28, 1994, pp. 1015-1022.

W ienke. D., Gao, N., Hopke, P.K., M ultiple Recptor M odeling by a Minimal Spanning Tree Projected into a Kohonen Neural Network, Environ. Sci. Technol., Vol. 28, 1994, pp. 1023-1030.

Yatagai, M., Ohira, M., Ohira, T., N agai, S., Seasonal Variations o f Terpene Emission from Trees and Influence o f Tem perature, Light and Contact Stimulation on Terpene Emission, Chem osphere, Vol. 30, No. 6, 1995, pp. 1137-1149


http://www.epa.gov/ttn/uatw/hapindex.html

147

Yi, J., Prybutok, V.R., A Neural Network Model Forecasting for Prediction o f Daily M aximum Ozone Concentration in an Industrialized Urban Area, Environ.Pollut., Vol. 92, 1996, pp. 349-357.

Yokouchi, Y., Bandow, H., Akimoto, H., Development o f Autom ated GasChromatographic-M ass Spectrometric Analysis for Natural Volatile Organic Compounds in the Atmosphere, Journal o f Chromatography, 642, 1993, pp. 401-407.

Zhang, X.J., Peterson, F., Emissions from CO-Combustion o f W ood and Household Refuse, Department o f Energy, Royal Institute o f Technology, 1996, pp. 997- 1000 .

Zweidinger. R.B.. Stevens, R.K.., Lewis, C.W., Identification o f Volatile Hydrocarbons as Mobile Source Tracers for Fine-Particulate Organics, Environ. Sci. Technol.,Vol.24, No. 4, 1990, pp. 538-542.


Modeling total reduced sulfur and ... - University of Montana

Documents

Transcript of Modeling total reduced sulfur and ... - University of Montana