@prefix vivo: . @prefix edm: . @prefix ns0: . @prefix dcterms: . @prefix dc: . @prefix skos: . vivo:departmentOrSchool "Science, Faculty of"@en, "Earth, Ocean and Atmospheric Sciences, Department of"@en ; edm:dataProvider "DSpace"@en ; ns0:degreeCampus "UBCV"@en ; dcterms:creator "Suzuki, Natalie M."@en ; dcterms:issued "2009-03-24T23:03:13Z"@en, "1997"@en ; vivo:relatedDegree "Master of Science - MSc"@en ; ns0:degreeGrantor "University of British Columbia"@en ; dcterms:description """The Lower Fraser Valley (LFV) of outhwestern British Columbia is periodically subject to unacceptably high concentrations of ground-level ozone (O₃). To investigate the factors which influence elevated 0₃ concentrations in the LFV, a three-dimensional Eulerian model called the Urban Airshed Model was used to simulate an historical O₃episode which occurred between July 17-19, 1985. The results showed generally satisfactory model performance for predicted O₃ levels across the region, relative to model performance parameters established elsewhere. Peak unpaired accuracies ranged from -2% to -24% and normalized gross error ranged from 35.6-36.9% over the three-day simulation. However, there was a tendency for the model to underestimate O₃concentrations, as reflected in the normalized bias, which ranged from -29.1 to -31.3%. Spatial patterns of predicted O₃concentrations indicated that higher O₃concentrations were observed along the valley walls and tributary valleys than on the valley floor where most of the population resides. Overall model performance based on nitrogen dioxide levels was within established parameters, although concentrations were typically underestimated and performance at two stations was poor. As a further means of assessing model performance, 0three indicator species (NOy, 0₃/NOz and H₂O₂/HNO₃) were calculated to determine the ozone sensitivity of the modelled airshed. The findings suggest that with the exception of sites located far downwind of the major source region, conditions tended to be VOC-limited."""@en ; edm:aggregatedCHO "https://circle.library.ubc.ca/rest/handle/2429/6440?expand=metadata"@en ; dcterms:extent "7356614 bytes"@en ; dc:format "application/pdf"@en ; skos:note "A P P L I C A T I O N OF T H E U R B A N A I R S H E D M O D E L IN T H E L O W E R F R A S E R V A L L E Y , BRITISH C O L U M B I A By Natalie M . Suzuki B. A.Sc. (Chem. Eng.) University of British Columbia, 1983 Dipl. Meteorology University of British Columbia, 1991 A THESIS S U B M I T T E D IN P A R T I A L F U L F I L L M E N T O F T H E R E Q U I R E M E N T S F O R T H E D E G R E E O F M A S T E R O F S C I E N C E in T H E F A C U L T Y O F G R A D U A T E STUDIES A T M O S P H E R I C S C I E N C E We accept this thesis as conforming to the required standard T H E U N I V E R S I T Y O F BRITISH C O L U M B I A August .1997 © Natalie M . Suzuki, 1997 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Atmospheric Science The University of British Columbia 2075 Wesbrook Place Vancouver, Canada V6T 1W5 Date: 9? /v?/zc Abstract The Lower Fraser Valley (LFV) of southwestern British Columbia is periodically subject to unacceptably high concentrations of ground-level ozone (O3) . To investigate the factors which influence elevated 0 3 concentrations in the L F V , a three-dimensional Eulerian model called the Urban Airshed Model was used to simulate an historical O 3 episode which occurred between July 17-19, 1985. The results showed generally satis-factory model performance for predicted O 3 levels across the region, relative to model performance parameters established elsewhere. Peak unpaired accuracies ranged from -2% to -24% and normalized gross error ranged from 35.6-36.9% over the three-day simu-lation. However, there was a tendency for the model to underestimate O 3 concentrations, as reflected in the normalized bias, which ranged from -29.1 to -31.3%. Spatial patterns of predicted O 3 concentrations indicated that higher O 3 concentrations were observed along the valley walls and tributary valleys than on the valley floor where most of the popu-lation resides. Overall model performance based on nitrogen dioxide levels was within established parameters, although concentrations were typically underestimated and per-formance at two stations was poor. As a further means of assessing model performance, three indicator species (NOy, 0 3 / N O z and H 2 O 2 / H N O 3 ) were calculated to determine the ozone sensitivity of the modelled airshed. The findings suggest that with the excep-tion of sites located far downwind of the major source region, conditions tended to be VOC-limited. i i Table of Contents List of Tables vi List of Figures viii Acknowledgments xiii 1 Introduction 1 1.1 Research Objectives 11 1.2 Methodology 11 1.3 Thesis Outline 11 2 Technical Description of Models 13 2.1 Urban Airshed Model 13 2.1.1 Governing Equations 13 2.1.2 Modelled Processes 16 2.1.3 Advection 16 2.1.4 Turbulent Diffusion 17 2.1.5 Surface Removal Processes 19 2.1.6 Chemical Processes 19 2.1.7 Emissions 23 2.2 C S U - R A M S 24 3 Episode and Domain Selection 26 3.1 Episode Selection 26 ii i 3.1.1 Data Availability 27 3.1.2 Characterization of a Typical Episode 29 3.1.3 Characterization of the Selected Episode 30 3.2 Domain Selection 39 4 Model Inputs 41 4.1 Boundary and Initial Conditions 41 4.1.1 B O U N D A R Y 41 4.1.2 R E G I O N T O P 43 4.1.3 A I R Q U A L I T Y 44 4.1.4 T O P C O N C 44 4.2 Meteorological Data 45 4.2.1 WIND 45 4.2.2 D I F F B R E A K 47 4.2.3 T E M P E R A T U R 52 4.2.4 M E T S C A L A R S 52 4.3 Emissions Data 53 4.3.1 Mobile Sources 55 4.3.2 Point Sources 56 4.3.3 Gasoline Marketing Sources 58 4.3.4 Area Sources 61 4.3.5 US Sources 63 4.4 Terrain Data 63 4.5 Chemical Parameters 64 4.6 Simulation Control Parameters 64 iv 5 Model Results and Discussion 66 5.1 Model Performance Criteria 66 5.2 Ozone 67 5.3 Nitrogen Dioxide 84 5.4 Discussion 99 6 Summary 107 List of References 109 v List o f Tables 1.1 U A M Applications 10 2.1 State species in C B M - I V 22 3.1 Location of air quality monitoring sites in the L F V , 1985 27 3.2 Exceedances of ambient 0 3 objectives in the L F V , 17-21 July 1985. . . . 38 4.1 U A M input files 42 4.2 Clean boundary conditions, South Coast Air Quality Management District (SCAQMD, 1990) 43 4.3 Maximum predicted and observed values of Zj at sites in Delta, Surrey and QEP, Vancouver between 17-19 July 1985 52 4.4 Annual emissions estimates for the B C (1985) and Washington (1990) L F V 54 4.5 Episodic adjustments of motor vehicle emissions 56 4.6 Default landuse types and terrain factors (Sheih et al., 1986) 64 5.1 Observed and predicted maximum one-hour O 3 concentrations in ppb and corresponding hour (PST) at monitoring stations in the L F V , 17-19 July 1985 79 5.2 Model performance based on predicted and observed maximum one-hour O 3 concentrations at stations in the L F V , 17-19 July 1985 82 5.3 Model 0 3 performance statistics for the L F V , 17-19 July 1985 83 vi 5.4 Observed (Obs.)and predicted (Pred.) maximum one-hour NO2 concen-trations in ppb at monitoring stations in the L F V , 17-19 July 1985, along with the respective time of day (Hr) 96 5.5 Model performance based on observed and predicted maximum one-hour N 0 2 concentrations at stations in the L F V , 17-19 July 1985 97 5.6 Model N 0 2 performance statistics for the L F V , 17-19 July 1985 98 vii List of Figures 1.1 Map of the Lower Fraser Valley showing topography, coastline, major pop-ulation centres and geographical features 3 3.1 Map of L F V , showing location of air quality and meteorological monitoring sites present in July 1985. Coastline and 100 m contours are shown by solid and dashed lines, respectively. Units are in U T M coordinates (x 103 m) 28 3.2 (a) Sea level pressure (hPa); (b) 850 hPa temperature (°C); (c) 850 bPa geopotential height (m); and (d) 500 hPa geopotential height (m), all for 17 July 1985 at 1200Z (0400 PST). Small rectangle in southeast quarter of each plot identifies modelling domain 31 3.3 (a) Sea level pressure (hPa); (b) 850 hPa temperature (°C); (c) 850 hPa geopotential height (m); and (d) 500 hPa geopotential height (m), all for 18 July 1985 at 1200Z (0400 PST). Small rectangle in southeast quarter of each plot identifies modelling domain 32 3.4 (a) Sea level pressure (hPa); (b) 850 hPa temperature (°C); (c) 850 hPa geopotential height (m); and (d) 500 hPa geopotential height (m), all for 19 July 1985 at 1200Z (0400 PST). Small rectangle in southeast quarter of each plot identifies modelling domain 33 viii 3.5 (a) Sea level pressure (hPa); (b) 850 hPa temperature (°C); (c) 850 hPa geopotential height (m); and (d) 500 hPa geopotential height (m), all for 20 July 1985 at 0000Z (19 July at 1700 PST). Small rectangle in southeast quarter of each plot identifies modelling domain 34 3.6 (a) Sea level pressure (hPa); (b) 850 hPa temperature (°C); and (c) 500 hPa geopotential height (m), all for 21 July 1985 at 1200Z (0400 PST). Small rectangle in southeast quarter of each plot identifies modelling do-main 36 3.7 Hourly temperature (solid lines) and wind vectors (arrows) at (a) Y V R and (b) Y X X , 16-21 July 1985 37 3.8 Modelling domain, as delineated by thick, solid lines. Coastline and 100 m contrours depicted by thin, solid lines and dotted lines, respectively. . 40 4.1 Surface (level 1) wind fields prepared for U A M at (a) 0900 PST and (b) 1500 PST, 17 July 1985. Vector length is proportional to wind speed. Observations shown by thick arrows. Wind speeds in m/s 48 4.2 Surface (level 1) wind fields prepared for U A M at (a) 0900 PST and (b) 1500 PST, 18 July 1985. Vector length is proportional to wind speed. Observations shown by thick arrows. Wind speeds in m/s 49 4.3 Surface (level 1) wind fields prepared for U A M at (a) 0900 PST and (b) 1500 PST, 19 July 1985. Vector length is proportional to wind speed. Observations shown by thick arrows. Wind speeds in m/s 50 4.4 Mixed layer depth at sites in Delta (top panel), Surrey (middle panel) and Queen Elizabeth Park (QEP) (lower panel), Vancouver. Lines depict predicted values; circles show observed data 51 ix 4.5 (a) V O C and (b) NOx emissions (in t/d) from mobile sources in the L F V . Washington sources were not included 57 4.6 Annual (a) V O C and (b) NOx emissions (in t/y) from point sources in the L F V . Circles denote point source emissions. Circle diameter is pro-portional to emissions 59 4.7 V O C emissions (in t/d) from gasoline marketing sources in the L F V . Sources from Washington were not included 60 4.8 (a) V O C and (b) NOx emissions (in t/d) from area sources in the L F V . Washington sources were not included 62 5.1 Predicted and observed 0 3 concentrations at stations T02, T03, T04 and T05 in the L F V , 17-19 July 1985. Solid lines represent predicted concen-trations and hatched lines show the range in neighbouring grid cells. Open circles indicate observations 68 5.2 Predicted and observed 0 3 concentrations at stations T07, T09 and T14 in the L F V , 17-19 July 1985. Solid lines represent predicted concentrations and hatched lines show the range in neighbouring grid cells. Open circles indicate observations 69 5.3 Predicted and observed 0 3 concentrations at stations T15, T16, T i l and T12 in the L F V , 17-19 July 1985. Solid lines represent predicted concen-trations and hatched lines show the range in neighbouring grid cells. Open circles indicate observations 70 5.4 Contours of predicted 0 3 at 20 ppb intervals on 17 July 1985 at (a) 0900 PST, (b) 1200 PST, (c) 1500 PST and (d) 1800 PST. Large numbers indicate observed concentrations. Missing values are indicated by \"-99\". 72 x 5.5 Contours of predicted O 3 at 20 ppb intervals on 18 July 1985 at (a) 0900 PST, (b) 1200 PST, (c) 1500 PST and (d) 1800 PST. Large numbers indicate observed concentrations. Missing values are indicated by \"-99\". . 74 5.6 Contours of predicted 0 3 at 20 ppb intervals on 19 July 1985 at (a) 0900 PST, (b) 1200 PST, (c) 1500 PST and (d) 1800 PST. Large numbers indicate observed concentrations. Missing values are indicated by \"-99\". . 76 5.7 Comparison of predicted vs. observed O 3 concentrations (above) and com-parison of residual (predicted-observed) vs. observed O 3 concentrations (below), all sites 17-19 July 1985 81 5.8 Predicted and observed N 0 2 concentrations at stations T02, T03, T04 and T05 in the L F V , 17-19 July 1985. Solid lines show predicted values. Open circles indicate observed data 85 5.9 Predicted and observed NO2 concentrations at stations T07, T09 and T14 in the L F V , 17-19 July 1985. Solid lines show predicted values. Open circles indicate observed data 86 5.10 Predicted and observed NO2 concentrations at stations T15, T16, T i l and T12 in the L F V , 17-19 July 1985. Solid lines show predicted values. Open circles indicate observed data 87 5.11 Contours of predicted NO2 at 20 ppb intervals on 17 July 1985 at (a) 0900 PST, (b) 1200 PST, (c) 1500 PST and (d) 1800 PST. Large numbers indicate observed concentrations 88 5.12 Contours of predicted NO2 at 20 ppb intervals on 18 July 1985 at (a) 0900 PST, (b) 1200 PST, (c) 1500 PST and (d) 1800 PST. Large numbers indicate observed concentrations 90 xi 5.13 Contours of predicted N 0 2 at 20 ppb intervals on 19 July 1985 at (a) 0900 PST, (b) 1200 PST. (c) 1500 PST and (d) 1800 PST. Large numbers indicate observed concentrations 92 5.14 (a) Comparison of predicted vs. observed O 3 concentrations, and (b) com-parison of residual (predicted-observed) vs. observed O 3 concentrations, all sites 17-19 July 1985 95 5.15 Sensitivity indicators based on predictions at stations T02, T03 and T05. Triangles show values for 18 July, diamonds for 19 July 104 5.16 Sensitivity indicators based on predictions at stations T07, T09 and T14. Triangles show values for 18 July, diamonds for 19 July 105 5.17 Sensitivity indicators based on predictions at stations T15, T i l and T12. Triangles show values for 18 July, diamonds for 19 July 106 xii Acknowledgments Throughout this project, I have benefited from the help of many people to whom I am deeply grateful. First and foremost, I would like to thank Dr. Douw Steyn, my supervisor, for being a constant source of ideas and encouragement, and for refusing to give up on me. I would also like to thank my other committee members, Dr. T im Oke and Dr. Ian McKendry, who provided useful comments on my draft thesis. Special thanks go to Dr. Xiaoming Cai who was responsible for the meteorological fields used in my simulations; Bruce Thomson of Environment Canada who provided support in carrying out this work and who continues to carry the torch in terms of airshed modelling in the Lower Fraser Valley; Dr. Don Singleton (NRC), Dr. Rob McLaren (York University) and Sue Boehme (NRC) for providing assistance in the development of the emissions input files; and A l Percival of the Greater Vancouver Regional District for providing the air quality data. Vincent Kujala and Jim Mintha provided valuable assistance in sorting out my com-puter woes. Paul Jance was responsible for the one great map in this document. Ad-ditional thanks go to the staff of the Air Resources Branch, Ministry of Environment, Lands and Parks, and particularly to Dr. Richard Bennett and Hu Wallis, for their encouragement and understanding. Finally, I would like to thank my family, and in particular, my husband Roy Hourston, for being a constant source of encouragement and the calm in the eye of the storm - I couldn't have done it without you! xii i Chapter 1 Introduction Ozone (O3) is a colourless and reactive gas that is found in trace amounts in the atmosphere. In the troposphere, it is formed during the reaction of nitrogen oxides (NO^) and volatile organic compounds (VOCs) in the presence of sunlight (Haagen-Smit, 1952). It is known to affect enzyme systems in the lungs and to cause a decrease in pulmonary function (Rowe et al., 1983; Bates and Vedal, 1994; Brauer and Brook, 1997), to cause foliar injury to vegetation (e.g. Runeckles and Chevrone, 1992; Lefohn et al., 1997; Treshow and Anderson, 1989; Wright, 1988)), and to cause the rapid degradation of materials such as rubber. The current Maximum Acceptable National Air Quality Objective (NAAQO) for O 3 is 82 ppb. Between 1984-1992, exceedances of this objective in the Lower Fraser Valley (LFV) of British Columbia (BC) occurred on 8.5% of all days during the months of July-September. Due to the periodically unacceptable levels of 0 3 in the L F V , the region was designated as one of three O 3 non-attainment areas under the national N O x / V O C Management Plan ( C C M E , 1990) and required to submit interim targets for emission caps to reduce O 3 concentrations. The South Atlantic Region and the Windsor-Quebec City Corridor were the other two designated non-attainment areas. Relative to the eastern non-attainment zones, the L F V is isolated from other source regions by virtue of local topography. As shown in Figure 1.1, the L F V is a funnel-shaped valley, beginning near the town of Hope and spreading westward to the Straight of Georgia, approximately 120 km away. The L F V is enclosed to the north by the Coast 1 Chapter 1. Introduction 2 Mountains, and to the southeast by the Cascade Mountains. The northern valley walls are punctuated by a series of north-south tributary valleys. The L F V has a current population of roughly 2 million persons which is growing rapidly at a rate of approximately 0.04 million per year (GVRD, 1994). The major population centre is the City of Vancouver, which lies on the western edge of the L F V . However, much of the current growth is occuring in areas to the east of Vancouver. Associated with the growth in population is an increase in the number of motor vehicles. Recent estimates indicate that motor vehicles in the L F V contribute roughly 80% of pollutants associated with 0 3 formation (GVRD, 1994). While the population and its associated activities influence pollutant emissions, the physical environment in which the L F V is situated has a profound influence on local circulation patterns, and, hence, on the transport of pollutants within the valley. Under certain synoptic conditions, temperature differences between the Straight of Georgia and the adjacent land surface give rise to daytime sea breezes and nighttime land breezes in the L F V (Emslie, 1968; Hay and Oke, 1985; CSC,1982; Steyn and Faulkner, 1986; Steyn et al., 1990; Miao, 1993, Cai and Steyn, 1995). The differential heating and cooling between valley walls and valley floor may also give rise to slope and drainage winds (Hay and Oke, 1985; Mass, 1982). The so-called \"chimney effect\", in which pollutants are vented along heated valley walls, has been shown to contribute to the development of polluted layers aloft (McKendry et al., 1997). The role of tributary valleys in providing a cleansing mechanism has also been observed, in which daytime up-valley flows carrying 03 - r i c h air are followed by nocturnal downslope flows characterized by low O 3 concentrations (Banta et al., 1997). Under stagnant conditions, these thermally induced winds may result in the recircu-lation and buildup of O 3 and its precursors to unsatisfactory levels. Steyn et al. (1990) characterized such an episode which took place between 1-3 September 1988. On 3 Figure 1.1: Map of the Lower Fraser Valley showing topography, coastline, major population centres and geographical features. Chapter 1. Introduction 4 September, a maximum one-hour O 3 concentration of 213 ppb was recorded. This was one of the highest O 3 concentrations ever recorded in Canada. Pacific'93 (Steyn et al., 1997) represented an intensive field study conducted during the summer of 1993 to better understand the chemical and physical factors, at the surface and aloft, which contribute to O 3 episodes in the L F V . Because O 3 is predominantly a secondary product in the troposphere, its control is achieved through the control of its precursors, NOx and VOCs. However, a very non-linear relationship exists between 0 3 and its precursors. The fundamental reactions involved in the photochemical formation of O 3 are as follows (from NRC, 1991): N02 + hv = NO + OCD) (1.1) 0 ( 3 P ) + 0 2 + M = 0 3 + M ( M = N2,02) (1.2) NO + 03 = N02 + 02 (1.3) Here, the production of NO2 requires the consumption of O 3 . Hence, in the absence of other species to enhance the NO-to-N02 conversion, a photostationary state is assumed, in which the concentration of O 3 can be estimated as [03]=ji[NO]/k2[NO] (1.4) where j i is the photodissociation constant for reaction 1.1 and k 2 is the rate constant for reaction 1.3. In the polluted troposphere, odd-hydrogen species play a key role in enhancing the conversion of NO to N02- These species include: the hydroxyl radical (OH), the hydro-gen peroxy radical (HO2) and the organic peroxy radical (RO2). The hydroxyl radical is formed through the photolysis of 0 3 and the subsequent reaction of 0( X D) with wa-ter vapour. The peroxy radicals are predominantly formed through the photolysis of Chapter 1. Introduction 5 aldehydes and other intermediate VOCs. The reactions are summarized as follows: Oz + hv = 02 + OCD),V < 320nm (1.5) 0(lD) + H20 = 20H (1.6) RH + OH = R + H20 (1.7) i? + 0 2 = # 0 2 (1.8) R02 + NO = RO + N02 (1.9) RO + 02 = RCHO + H02 (1.10) RCHO + hv = R + CHO (1.11) C H O + 0 2 = # 0 2 + C O (1.12) 0 # + 7V0 2 + M = HONO2 + M (1.13) Major sinks of the odd-hydrogen species include the following: H02 + H02 = H202 + 02 (1.14) R02 + H02 = ROOH + O2 (1.15) OH + N02 = HN03 (1.16) The formation of peroxyacetyl nitrate (PAN) is another potential sink, particularly in urban environments. The numerous competing reactions reflect the fact that the relationship between O3 and its precursors is highly nonlinear. Isopleths of constant O3 concentrations plotted for various mixtures of VOCs and NOx (e.g. NRC, 1991) show that reductions in NOx levels at constant V O C concentrations may result in higher or lower O3 levels, depending on whether the environment is V O C - or NOx-limited. To add to the complexity, these conditions will vary both temporally and spatially within a single airshed. Chapter 1. Introduction 6 Two fundamentally different approaches can be taken to determine whether NOx or V O C reductions are the most effective means of controlling O 3 levels: (i) an observation-based approach and (ii) an emission-based approach, utilizing photochemical air models. The observation-based approach utilizes atmospheric observations at a given site to determine the sensitivity of real air masses to changes in 0 3 precursors. Applications range in complexity from comparing these observations to expected values, to using the observations as inputs to a photochemical model. The simplest techniques utilize ambient concentrations of individual species, or the ratio of different species which consistently assume different values depending on whether the conditions are NOx- or VOC-sensitive. These species or ratios of species are referred to as sensitivity indicators. Examples are the VOC-to-NOx ratio, total nitrogen (NOy) concentration, formaldehyde (HCHO) to NOy ratio, 03-to-NOz ratio (where NOz=NOy-NOx) and hydrogen peroxide ( H 2 O 2 ) to nitric acid (HNO3) ratio. Each is described in the following. The VOC-to-NOx ratio is an obvious indicator to describe how reductions in V O C and NOx emissions affect 0 3 concentrations. The transition between VOC-sensitive and NOx-sensitive conditions typically occurs near a VOC-to-NOx ratio of about 8:1 (NRC, 1991). NOx-limited conditions occur at lower values while VOC-limited conditions occur at higher values. In practice, ratios obtained from measurements taken in the early morning (e.g. 0600-0900 LAT) at the urban centre are typically used. Milford et al. (1990) expressed the following concerns with this approach: • The air mass sampled in the urban centre during the morning may not be the one that produces peak O 3 concentrations later in the day and further downwind. • Contributions from precursor sources after the sampling period are ignored. • The reactivity of the V O C mixture is not considered. Chapter 1. Introduction 7 • Urban plumes tend to shift from VOC-sensitive to NOx-sensitive as they move downwind because NOx becomes depleted more quickly than VOCs. NOy includes all nitrogen species which can be converted into NOx during photo-chemical activity: NOx, P A N , H N 0 3 , nitrous acid (HONO), all organic nitrates and nitrate radicals. It can therefore be viewed as a NOx reservoir. Unlike NOx, which is relatively short-lived, NOy has a similar lifetime as O 3 . Studies have shown that where afternoon NOy is greater than 20 ppb, conditions are VOC-sensitive (Milford et al., 1989; Rao et al., 1993; Sillman et al., 1993; Milford et al., 1994). A limitation with respect to this indicator is the difficulty in measuring NOy. A reactivity-weighted alternative to the VOC-to-NOx ratio is the ratio of H C H O to NOy. Formaldehyde is readily photolysed, yielding by-products which react to form odd-hydrogen species. As such, the HCHO-to-NOy ratio is a much better indicator of 0 3 sensitivity than the VOC-to-NOx ratio. Studies indicate that a transition from V O C - to NOx-sensitive conditions occurs near ratios of 0.28 (Sillman, 1995). The photolysis of 0 3 is another major source of odd-hydrogen species. As the concen-tration of aldehydes generally increases with O 3 concentration, it may be assumed that the source of odd-hydrogen species is proportional to 0 3 . NOz represents the reaction products of NOx. Hence, the ratio of 0 3 - t o - N O z may be used to describe the transition from NOx- to VOC-sensitive chemistry. Sillman (1995) reported this transition near a ratio of 7, where values less than 7 represent VOC-sensitive conditions. The ratio of H 2 0 2 - t o - H N 0 3 represents the competition between reactions 1.17 and 1.18 for odd hydrogen. H02 + H02 = H202 + 02 (1.17) OH + N02 = HNOz (1.18) Both H 2 O 2 and H N 0 3 are major sinks of odd hydrogen. Where H 2 0 2 dominates, O H Chapter 1. Introduction 8 (and therefore 0 3 ) increases with increasing NOx: HO2 + NO = OH + NO2 (1.19) and decreases with increasing VOCs: RH + OH (+02) = R02 + H20 (1.20) Where H N O 3 dominates, OH (and therefore O3) decreases with increasing NOx. The transition from NOx to VOC-sensitive chemistry occurs near ratios of 0.3-0.5 (Sillman, While sensitivity indicators can provide guidance on which precursor should be re-duced to attain the most efficient reduction in O3 levels, they cannot provide a quan-titative estimate of what that change will be, nor can they predict how the changes will vary spatially over an entire airshed. For that, it is necessary to employ a photo-chemical model which can represent the pertinent physical and chemical processes in the atmosphere which affect ozone concentrations. Photochemical models were first developed in the early 1970's, in response to the passage of the Clear Air Act Amendments in the United States. These amendments established provisions for uniform national air quality standards and ordered states to develop abatement programs to meet these standards. Modelling became a necessary tool to demonstrate attainment. There are three basic types of photochemical model: (i) box models, (ii) trajectory (Lagrangian) models, and (iii) grid-based (Eulerian) models. Seinfeld (1988) provides an overview of each model type. Box models are the most simplistic, as they treat the domain as a single cell. The ground and the inversion height represent the lower and upper boundaries. Emissions are assumed to mix uniformly and instantaneously throughout the domain. A characteristic 1995). Chapter 1. Introduction 9 wind is specified, and the inversion height is allowed to vary with time, thereby providing a ventilation mechanism for the domain. Lagrangian models follow a single column of air that is advected by the mean hor-izontal wind. The column is defined by the ground and the inversion height, and may be divided into vertical layers. Emissions are injected into the column as it passes over emission sources. Chemical transformations, removal processes and vertical diffusion can be simulated. In Eulerian models, the domain is divided into a number of grid cells. The governing equations representing atmospheric processes are then solved for each grid cell. Eulerian models typically require more detailed input data and greater computer resources than the other two model types. However, they are the only models that can predict the tem-poral and spatial variations in pollutant patterns in a three-dimensional domain. Hence, Eulerian models are best-suited for urban air quality studies such as in the L F V where a prognostic capability is required. Major Eulerian photochemical models in use in North America include the following: Urban Airshed Model Version IV (UAM-IV) (Morris and Myers, 1990); U A M - V (Haney et a l , 1990); CIT (McRae et al., 1982; McRae and Se-infeld, 1983); California Air Resources Board airshed model (CALGRID) (Yamartino et al., 1992); the Regional Acid Deposition Model (RADM) (Chang et al., 1987); and the Regional Oxidant Model (ROM) (Lamb, 1983). Within the L F V , the use of two different photochemical models has been recently re-ported. The C A L G R I D model was used in conjunction with the Mesoscale Compressible Community Model (MC2) to evaluate model performance for an 0 3 episode occurring between 17-21 July 1985 (Hedley and Singleton, 1997; Hedley et al., 1997) and to eval-uate the effectiveness of various alternative fuel scenarios (Hedley et al.,1996). Systems Applications International applied the U A M - V to evaluate model performance for the same July 1985 episode (Lolk et al., 1995) and to the evaluate emission control scenarios Chapter 1. Introduction 10 Table 1.1: U A M Applications Location References Atlanta Morris et al. (1990a) Baton Rouge Haney et al. (1990) Denver Dennis and Downton (1984) Los Angeles Basin S C A Q M D (1990) Netherlands Builtjes and Reynolds (1982) New York Rao et al. (1987) Santa Barbara/Ventura Co. Tesche and McNally (1991) St. Louis/ Morris et al. (1990b) Philadelphia Tokyo, Japan Wakamatsu et al., (1990) for B C Hydro (Haney, 1997). In this research, the U A M - I V will be used. It is the preferred model for O 3 regulatory analysis in the United States, as specified by the U.S. Environmental Protection Agency (U.S. EPA) . It has been used extensively throughout the United States and in other countries, as shown in Table 1.1. This research was begun in parallel with the C A L G R I D application in the L F V . While it represents a less advanced treatment than the U A M - V or C A L G R I D models, the U A M - I V is an established model which has been well-documented. Although not the purpose of this research, this study provides further reinforcement of the findings obtained in the other modelling applications in the L F V . Chapter 1. Introduction 11 1.1 Research Objectives The goal of this research is to apply the U A M - I V to the L F V . Model performance will be evaluated in terms of the ability of the model to recreate an historical O 3 episode. Established statistical parameters will be applied to assess the results. Sensitivity indi-cators will also be calculated from model output as a further means of assessing model performance. 1.2 Methodology The first step will be the selection of an historical multi-day ozone episode to be modelled. This will be followed by model domain selection, which is dependent on the circulation patterns observed or expected during the episode as well as the important sources that are to be included in the study. Episode and domain selection will then define the meteorological, emissions and air quality data required by the model, including the spatial and temporal scales to be used. The model will be run to simulate the historical episode, and predicted O 3 and NO2 concentrations compared against observed values. Model performance will be characterized in terms of the acceptable range of statistical performance indicators established from a review of other modelling studies. Finally, the model results will be compared with established indicators to assess the sensitivity of the modelled airshed to V O C and NOx. 1.3 Thesis Outline In Chapter 2, the theory behind the U A M - I V is presented. Individual modules comprising the model are described. In Chapter 3, the characteristics of the selected O 3 episode and a description of the modelling domain are presented. The preparation of model input files to describe the Chapter 1. Introduction 12 meteorology, emissions and air quality during the episode is described in Chapter 4. Model results are compared against observational data in Chapter 5. Model perfor-mance is characterized based on the ability of the model to predict spatial and temporal variations in O 3 and NO2 concentrations. Sensitivity indicators calculated based on model results are also presented. Major findings are summarized in Chapter 6. Chapter 2 Technical Description of Models Although the focus of this study is the application of U A M - I V , additional models have been used by others to produce meteorological fields and eission estimates. In the following, a detailed description of the U A M and its modelled processes will be followed by brief descriptions of the mesoscale model, CSU-RAMS, and the models used to estimate source emissions. 2.1 Urban Airshed Model The development of the U A M dates back to 1969 when the National Air Pollution Control Administration (NAPCA) , a predecessor of the U.S. EPA, contracted with Sys-tems Applications Inc. (SAI) to develop an air quality model which could be used to assess urban O 3 and to evaluate control strategies for abatement programs. A n overview of model development is provided by Scheffe and Morris (1993). 2.1.1 Governing Equations The U A M simulates the physical and chemical fate of inert or chemically reactive species in the atmosphere through solution of the species continuity equation: S C J _ _ 6t _ 6(u-Cj) 5(v • Cj) S(w • Cj) 5x 5y Sz + fa<*»S> + + S<*TJ> 13 Chapter 2. Technical Description of Models 14 + Ri + Si + Lt (2.21) where Q is the concentration of species i which has been averaged spatially (x,y,z ) over the volume of a grid cell and temporally (t) over the interval of an integration time step; u , v and w are wind velocities in the x , y and z directions, respectively; KH and Ky are horizontal and vertical eddy diffusivities; Ri is net rate of production of species i by chemical reactions; Si is emission rate of species i ; and Li is net rate of removal of species i by dry deposition. By assuming that the chemical species do not alter the meteorology to any significant degree, Equation 2.21 can be solved independently of the coupled Navier-Stokes and energy equations (Reynolds et al., 1974). Due to the presence of nonlinear terms in the equation, a numerical solution is required. To facilitate the use of finite difference methods, the vertical dimension is normalized to an independent variable p: p = *-Hb{x,y,t) Ht(x,y,t)-Hb(x,y,t) V ' ; where Hb and Ht are the elevations of the surface and top of the domain, respectively. Upon transformation of vertical coordinate z to p, and after neglecting cross-derivative diffusion terms which are assumed to be negligibly small, Equation 2.21 becomes A A A A — (AH • Q ) + — (u • AH • Cj) + —(vAH-Ci) + —(W-Ci) = ot ox dy dp + RiAH + SiAH (2.23) Chapter 2. Technical Description of Models 15 where W = w — u( 6Hb + P 5AH Sx )-v( + P 5AH 5y AH (2.24) 5x 5y and AH = Ht(x,y,t) - Hb(x,y,t) (2.25) As noted by Reynolds et al. (1974), Equation 2.23 is an approximate equation as opposed to the fundamental equation governing the dynamic behaviour of pollutants in the atmosphere. Its application is subject to the following limitations: • the time resolution A i is large compared with the Lagrangian time scale of turbu-• A i is small compared with the characteristic temporal scales for — gradients in the mean velocity field; — gradients in the mean turbulence velocity correlations; — gradients in the source emission rates; — changes in the rate of production or depletion of a species by chemical reac-• the average distance that a fluid particle travels in A i is small compared with the characteristic spatial scales for — gradients in the mean velocity field; — gradients in the mean turbulence velocity correlations; — gradients in the source emission rates. lence; tions. Chapter 2. Technical Description of Models 16 Based on data obtained in the Los Angeles Basin, Reynolds et al. (1974) found that Equation 2.23 was applicable for perturbations in the concentration field with horizontal scales greater than 2 km, vertical scales greater than 20 m, and temporal scales greater than 103 s. Equation 2.23 is solved using the method of time-splitting, or fractional steps, as proposed by Yanenko (1971). During each time step, the equation is solved for 1. advection/diffusion in the x-direction; 2. advection/diffusion in the y-direction; 3. injection of pollutant emissions and advection/diffusion in the z -direction; 4. chemical transformations. The master or advection time step is determined based on grid size and maximum wind speed. To maintain numerical stability, the chemistry and vertical diffusion time steps must be an integral portion of the advection time step. 2.1.2 Modelled Processes The mechanisms used to describe the transport, removal and chemical transforma-tion of pollutants in the atmosphere are presented in the following. 2.1.3 Advection The horizontal dispersion of pollutants is primarily through advection. The advective part of Equation 2.23: A A - (AH • Ci) + — (u • AH • a) = 0 (2.26) Chapter 2. Technical Description of Models 17 and -(AH-ci) + —(vAH-cl)=0 (2.27) is solved using the Smolarkiewicz scheme (1983), which is a positive definite upstream scheme reported to be less diffusive and more computationally efficient than the SHASTA scheme (Boris and Book, 1973) applied in previous versions of the U A M . A detailed description of this scheme is provided in Morris and Myers (1990). 2.1.4 Turbulent Diffusion The diffusive part of Equation 2.23: is solved using an explicit finite difference method. Turbulent fluxes are parameterized using a closure approximation frequently referred to as either gradient transport theory or K-theory, where it is assumed that turbulent fluxes are proportional to local gradients. The proportionality constant (KH in the horizontal; Ky in the vertical) is known by a variety of names, including eddy viscosity and eddy diffusivity. KH and Ky are difficult to measure. Given that in the horizontal, advection generally dominates over diffusion, a nominal constant value of 50 m 2 - s _ 1 is given to KH. Greater care is given to the treatment of Ky, as diffusion often dominates over advection in the vertical. Ky is calculated as a function of stability class, ground-level wind speed, reference height, surface roughness and height of the grid cell. For stable conditions, (2.28) Ky = (fc\".*)exp(-fetg) 1 + 4.7* (2.29) for neutral conditions where 0 < z < 0.45%, Kv = -f(a0 + an A 2 + a3X3 + a 4 A 4 ) (2.30) Chapter 2. Technical Description of Models 18 for neutral conditions where z > 0.45y, Kv = 0.01m2/s (2.31) and for unstable conditions: Kv = w*Zl(30 + A C + /32(2C2 - 1) + &(4C3 - 3C) + /?4(8C4 - 8c?2 + 1) (2.32) where C = 2 £ - l , « . = « . - ( i t ) 1 / 3 . /=Coriolis parameter, A;=von Karman constant, u*=friction velocity, Zj^inversion height, i>9=geostrophic wind component, L=Monin-Obukhov length, and the coefficients a* and $ are c*o=7.396xlO-4 #,=0.152 ai=6.082 x 10- 2 A =0.080 a2=2.532 #,=-0.039 a3=-1.272 x 10 /?3=0.032 a 4=l-517 x 10 /?4=0.020 For lapse rates less than -0.011°C/m above the mixing layer, Equation 2.30 or Equa-tion 2.31 for neutral conditions is used. Otherwise, if neutral or unstable conditions are present in the lower levels, then Ky aloft is set to 0.01 m 2 . If stable conditions are present in the lower levels, then Ky aloft is calculated using Equation 2.29. Chapter 2. Technical Description of Models 19 2.1.5 Surface Removal Processes Dry deposition is the only surface removal process treated in the U A M . It is treated as a three-stage process: 1. transport of pollutants through the atmosphere to just above the surface; 2. transport of pollutants to the actual surface; 3. uptake of the pollutants by vegetation or surface material through absorption, adsorption or chemical reaction. The U A M utilizes a deposition velocity concept, in which the deposition velocity Vd% is inversely proportional to the sum of the resistance to transport Rt and the resistance to surface removal Rsf Vdi = (2-33) Rt is a function of wind velocity at 10m elevation and friction wind velocity, while Rsi is a function of pollutant and surface types. The deposition velocity is then related to the uptake flux F^ by the following equation: Fdi = Vdi • Cgi (2.34) where Cgi is the ground level concentration of species i . A more detailed description of the parameterization of surface removal processes is given in Killus (1984). 2.1.6 Chemical Processes Chemical transformations taking place in the atmosphere are described by a chemical kinetic mechanism. This mechanism is comprised of a set of reactions and associated rate constants in the form of coupled ordinary differential equations. Chapter 2. Technical Description of Models 20 To explicitly model all the pertinent species and reactions taking place in the at-mosphere would require a prohibitive number of computations and C P U time. In fact, Peters et al. (1995) report that the integration of chemical rate equations can con-sume as much as 98% of the total C P U time in Eulerian modelling simulations. Two major approaches have been developed to reduce the number of hydrocarbon species which are included in the chemical mechanism: the lumped molecule approach and the lumped structure approach. The former method organizes hydrocarbons using molecular surrogates. Examples are the S A P R C mechanism (Carter, 1990) which is used in the C A L G R I D photochemical model, and the Stockwell mechanism (Stockwell, 1990) which is used in the Regional Acid Deposition Model (Chang et al., 1987). The latter method disaggregates hydrocarbons on the basis of bond type. As this approach requires fewer chemical reactions, it is computationally more efficient. The Carbon Bond IV Mechanism (CBM-IV) , which is used in the U A M , represents a lumped structure approach (Whitten et al., 1980; Gery et al., 1988, 1989). Four types of species are considered in the CBM-IV: • inorganic species which are treated explicitly: ozone, NOx and HOx chemistry; • organic species which are treated explicitly: — formaldehyde, because it is formed in all oxidation reactions involving hydro-carbons and is very reactive; — ethene, because it constitutes a large fraction of hydrocarbon emissions, is un-usually unreactive for an alkene, and yields a high percentage of formaldehyde under most conditions; — isoprene, because it constitutes a large fraction of biogenic hydrocarbon emis-sions and is very reactive. Chapter 2. Technical Description of Models 21 • organic species which are represented by carbon (C) bond surrogates: — paraffins: single C bonds; — olefins: C-C double bonds; — aldehydes: C-CHO bond and also 2-alkenes which tend to react very quickly to produce aldehyde products. • organic species which are represented by molecular surrogates: — toluene: seven-C species used as surrogates for monoalkylbenzene structures; — xylene: eight-C species used as surrogates for dialkyl- and trialkylbenzenes. The C B M - I V represents a condensed version of the C B M - E X which is described in Gery et al. (1988). C B M - E X is a more detailed mechanism that contains over 120 reactions. A number of steps were carried out to condense this mechanism: • eliminate unimportant reactions and products, • create a universal peroxy radical, thereby eliminating many organic peroxy radicals, • apply mathematical and algebraic manipulations to limit the number of reactions, and • lump secondary reaction products, especially relating to isoprene chemistry. The condensed version contains 33 state species, as shown in Table 2.1. More than 80 reactions are included in the mechanism. Due to the wide range of rate constants involved, the differential equations describing the mechanism are referred to as a \"stiff\" system. To facilitate computations, quasi-steady state assumptions are used for the low-mass, fast-reacting (i.e. stiff) species. The Crank-Nicholson algorithm is applied to solution of the other state species. Chapter 2. Technical Description of Models Table 2.1: State species in C B M - I V Species Name Representation nitric oxide NO nitrogen dioxide N02 nitrogen trioxide N 0 3 dinitrogen pentoxide N205 nitrous acid HONO nitric acid H N 0 3 peroxynitric acid P N A oxygen atom (singlet) O l D oxygen atom (triplet) 0 hydroxyl radical O H water H20 ozone 03 hydroperoxy radical H02 hydrogen peroxide H202 carbon monoxide CO formaldehyde F O R M high-molecular-weight aldehydes A L D 2 peroxyacyl radical C203 peroxyacyl nitrate P A N paraffin carbon bond P A R secondary organic oxy radical R O R olefinic carbon bond O L E ethene E T H toluene T O L cresol and higher-molecular-weight phenols CRES toluene-hydroxyl radical adduct T 0 2 methylphenoxy radical CRO high-molecular-weight aromatic oxidation ring fragment O P E N xylene X Y L methylglyoxal M G L Y isoprene ISOP NO- to-N0 2 operator X 0 2 NO-to-nitrate operator X 0 2 N Chapter 2. Technical Description of Models 23 Dodge (1989, 1990) reviewed three chemical mechanisms used in photochemical mod-els: C B M - I V , the Carter-Atkinson-Lurmann-Lloyd (CAL) mechanism (Carter et al., 1986; Lurmann et al., 1987) which is part of the S A P R C family of models, and a mech-anism developed by Stockwell (1988) for use in R A D M (Chang et a l , 1987). Both the S A P R C and Stockwell mechanisms use a lumped molecule approach, although the latter utilizes a generalized organic species to represent similar organics. In general, Dodge found good agreement between the three mechanisms, and points out that this should be expected, since the developers relied on the same kinetic data evaluation sources during mechanism development. The major differences involve predictions under high levels of N O x and aromatics, and at low temperatures. Of particular note, the C B M - I V consistently predicted lower concentrations of ozone at low temperatures. Jefferies and Tonnesen (1994) compared results obtained using the SAPRC90 (Carter, 1990) and C B M - I V chemical mechanisms in a Lagrangian box model. Both mechanisms gave similar predictions of total reactivity and O 3 maxima. However, at low NOx levels, the SAPRC90 mechanism was more reactive and produced higher O 3 maxima, while the C B M - I V mechanism produced more nitrate and became NOx-limited earlier in the day. 2.1.7 Emissions Emissions within an airshed originate from a wide range of sources, including point, area, mobile and biogenic sources. Methodologies used by others to develop emission estimates in the L F V are discussed in Chapter 4. For the purposes of simulating these emissions within an Eulerian model such as the U A M , emissions data must be processed so that (i) spatial and temporal distributions reflect the simulation period, (ii) emissions are injected into the appropriate vertical grid cell, and (iii) organic species are attributed to the appropriate surrogate group. The Emissions Preprocessor System Version 1.0 (EPS1.0) was developed to prepare Chapter 2. Technical Description of Models 24 emission inventories that are compatible with the format required by the U A M (Causley, 1990). The spatial and temporal allocation of emissions and the chemical speciation are achieved by associating spatial, temporal and source codes with the appropriate distribution factors. Emissions injected into a grid cell are mixed instantaneously throughout the cell. Ground-level emissions are input to the lowest level of grid cells. Elevated emissions are injected to the appropriate vertical cell corresponding to the total effective plume rise A h p , which is calculated by an emissions preprocessor using the algorithms recommended by Briggs (1971) and summarized in Morris and Myers (1990). 2.2 C S U - R A M S The lack of an extensive wind monitoring network, the complexity of the terrain in the L F V , and the importance of mesoscale flows such as land/sea breezes and moun-tain/valley winds necessitated the use of a prognostic non-hydrostatic model. For these reasons, the Colorado State University Regional Atmospheric Modelling System (CSU-R A M S ) , hereafter referred to as R A M S , was chosen to generate the three-dimensional wind fields which are required by the U A M . R A M S is a second-generation model to the CSU Mesoscale Model (CSUMM) de-veloped by Pielke (1974). It is very flexible, representing a merging of three different atmospheric models: a non-hydrostatic cloud model and two hydrostatic mesoscale mod-els (McNider and Pielke, 1981; Tremback et al., 1986; Tripoli and Cotton, 1982). It is based upon the full set of primitive dynamical equations, and includes optional pa-rameterizations for such processes as turbulent diffusion, terrestrial radiation, and moist processes (Walko and Tremback, 1991). Solution of these equations is through finite dif-ference methods. To improve the efficiency of numerical solutions, the Arakawa-C grid Chapter 2. Technical Description of Models 25 stagger is employed. R A M S has been applied in a number of studies in the L F V . Steyn and McKendry (1988) used a primitive version of R A M S to simulate meteorological fields in the L F V for 23 August 1985. Miao and Steyn (1994) simulated conditions on the same day using RAMS2a. Steyn and Cai (1994) applied RAMS2a to an 0 3 occurring between 17-20 July 1985. Cai and Steyn (1995) repeated this simulation using the next-generation mode, RAMS3a. Chapter 3 Episode and Domain Selection Model performance is evaluated on the basis of its ability to replicate observed 0 3 concentrations during a particular historical episode. Domain selection is dependent on defining the area of interest and then adjusting for the meteorological conditions during the episode which may affect airflow into and out of the region. Hence, both episode and domain selection must precede the preparation of model inputs, as they help define the air quality, emissions and meteorological databases to be used. In the following, the criteria and rationale for episode and domain selection are presented. 3.1 Episode Selection A n 0 3 episode which occurred between 17-20 July 1985 was selected for further study. Episode selection was based on the following criteria (after Seinfeld, 1988): • The episode exhibited elevated O3 concentrations. • The episode was a multi-day event. • The episode was characteristic of typical O3 episodes in the region. • Satisfactory databases of aerometric, emissions and meteorological data were avail-able to characterize the modelling domain during the episode. A more detailed description of the episode and of the available data is provided in the following. 26 Chapter 3. Episode and Domain Selection 27 Table 3.1: Location of air quality monitoring sites in the L F V , 1985. Station Location U T M E U T M N ID (x 103m) (x 103m) T01 Downtown Vancouver 491.3 5458.7 T02 Kitsilano, Vancouver 488.2 5456.6 T03 Marpole 492.1 5450.4 T04 Kensington Park, Burnaby 502.1 5458.4 T05 Confederation Park, Burnaby 499.9 5459.0 T06 Second Narrows 598.5 5460.8 T07 Anmore 510.2 5461.9 T08 Lions Gate STP 490.3 5462.9 T09 Rocky Point Park, Pt. Moody 511.1 5458.4 T10 Eagle Ridge 512.9 5459.0 T i l Abbotsford Airport 545.5 5430.4 T12 Chilliwack Airport 576.8 5444.6 T13 North Delta 507.3 5445.0 T14 Burnaby Mountain 506.8 5458.3 T15 Surrey East 522.4 5442.2 T16 Pitt Meadows 521.7 5450.9 3.1.1 Data Availability The G V R D and the B.C. Ministry of Environment, Lands and Parks (MELP) oper-ate a relatively dense network of air quality monitors in the L F V . In July 1985, O 3 data were monitored at 16 sites as shown in Figure 3.1 and Table 3.1. The highest density of stations was found along Burrard Inlet, in the vicinity of major oil refineries. The spatial resolution of O 3 data in the central valley was and still is comparatively poor, although sites in Abbotsford, Chilliwack and, more recently, Hope provide valuable information on the eastern extent of O 3 episodes. Chapter 3. Episode and Domain Selection 28 CO 6 o o O 5 I -z > 1 1 1 1 1 1 1 1 1 O air quality stat ions + airports • e c h o s o u n d e r A te thersonde \\ v 7 / i f > r • A • + • • • • \\ \\ \\ A s . ? \\ \\ \\ > ' / — > / } • / n \\ s \\ > > / fs s > M \\ 460 480 500 520 540 560 580 600 UTM Coordinates (x 10*3) Figure 3.1: Map of L F V , showing location of air quality and meteorological monitoring sites present in July 1985. Coastline and 100 m contours are shown by solid and dashed lines, respectively. Units are in U T M coordinates (x 103 m) Temperature, humidity, cloud cover and wind data are regularly collected at the Vancouver International Airport (YVR) , which is located along the coast, and Abbotsford International Airport ( Y X X ) , which is located in the central valley. Additional wind and temperature data are collected at some of the G V R D and M E L P monitoring stations, as listed in Cai and Steyn (1995). Augmenting the regularly available air quality and meteorological data are the fol-lowing datasets obtained during or for 1985: • an emissions inventory prepared for base year 1985 ( G V R D , 1988); • tethersonde data from Queen Elizabeth Park (QEP) in Vancouver for 17 and 19 July 1985, obtained as part of a study into the development of the mixed layer in Chapter 3. Episode and Domain Selection 29 the Lower Mainland (Steyn and Wallis, 1986); • back-trajectory analyses from three sites in the region for 17, 19 and 20 July 1985 (Coligado, 1988). Such information is important for both model input preparation and model performance evaluation. 3.1.2 Characterization of a Typical Episode Between May and September, the most prevalent synoptic regimes for surface O 3 events in southwestern British Columbia are persistent or slow-moving anticyclones which are eastward extensions of the North Pacific High (CSC, 1985). Anticyclones are gen-erally associated with light winds and large-scale subsidence resulting in the adiabatic warming of the subsiding air, clear skies, and the formation of temperature inversions aloft. The lack of synoptic forcing allows for the development of local thermally induced flow patterns such as sea breeze/land breeze circulations and mountain/valley circula-tions, which may result in the recirculation and accumulation of pollutants in the airshed (Steyn and Faulkner, 1986). In a study of the climatology of O 3 episodes in the L F V , McKendry (1994) charac-terized the various synoptic patterns related to ozone episodes at station T09 in Port Moody. He found that of the days in which O 3 concentration exceeded 82 ppb at T09, 47% were associated with the presence of a low-level thermal trough and a 500 hPa upper level-ridge of high pressure, in agreement with the general findings of Taylor (1991). The remaining episode days were characterized by either an upper-level low off the British Columbia coast combined with a weak surface pressure gradient (44% of episode days) or a persistent southerly flow at 500 hPa and weak flow at the surface (6% of episode days). McKendry also found that episode days are associated with a reduction in the Chapter 3. Episode and Domain Selection 30 strength of the sea breeze, suggesting that the thermal trough acts to suppress sea breeze development. 3.1.3 Characterization of the Selected Episode Synoptic Meteorological Profile Daily surface pressure fields, temperature fields at 850 hPa, and geopotential height fields at 850 hPa and 500 hPa between 17-21 July 1985 are shown in Figures 3.2-3.6. A persistent feature throughout the episode was the broad surface high extending north-ward over the Pacific Ocean, off the west coasts of British Columbia and Washington. Superimposed on this were two additional features which McKendry (1994) identified as characteristic of ozone episodes in the region: a low-level thermal trough and an upper-level ridge. The thermal trough was aligned over the coast of southwestern B C on 18 and 19 July. Associated with this trough was warm air advection from the southeast, as evident in the 850 hPa temperature plots which showed temperatures over the region increase from about 16°C on 18 July to 20°C the next day. During this period, the upper-level ridge at 500 hPa migrated slowly eastward. Geopotential heights over southwestern B C increased from 5800 m to greater than 5820 m. Surface O 3 concentrations measured in the L F V indicated elevated 0 3 levels (i.e. >82 ppb) at several sites between 18-20 July. By 21 July, the O 3 episode was over. The surface trough had migrated inland and 850 hPa temperatures had dropped to approximately 18°C. Upper-level geopotential heights remained above 5820 m, but flow patterns changed from meridional to nearly zonal. Chapter 3. Episode and Domain Selection 31 (a) (b) Figure 3.2: (a) Sea level pressure (hPa); (b) 850 hPa temperature (°C); (c) 850 hPa geopotential height (m); and (d) 500 hPa geopotential height (m), all for 17 July 1985 at 1200Z (0400 PST). Small rectangle in southeast quarter of each plot identifies modelling domain. Chapter 3. Episode and Domain Selection 32 (a) (b) Figure 3.3: (a) Sea level pressure (hPa); (b) 850 hPa temperature (°C); (c) 850 hPa geopotential height (m); and (d) 500 hPa geopotential height (m), all for 18 July 1985 at 1200Z, (0400 PST). Small rectangle in southeast quarter of each plot identifies modelling domain. Chapter 3. Episode and Domain Selection 33 (a) (b) Figure 3.4: (a) Sea level pressure (hPa); (b) 850 hPa temperature (°C); (c) 850 hPa geopotential height (m); and (d) 500 hPa geopotential height (m), all for 19 July 1985 at 1200Z (0400 PST). Small rectangle in southeast quarter of each plot identifies modelling domain. Chapter 3. Episode and Domain Selection 34 (a) (b) (d) / x ' ^ | / 0 Figure 3.5: (a) Sea level pressure (hPa); (b) 850 hPa temperature (°C); (c) 850 hPa geopotential height (m); and (d) 500 hPa geopotential height (m), all for 20 July 1985 at 0000Z (19 July at 1700 PST). Small rectangle in southeast quarter of each plot identifies modelling domain. Chapter 3. Episode and Domain Selection 35 Mesoscale Meteorological Profile The episode was characterized by high temperatures, clear skies, light onshore breezes during the daytime and extended periods of calm or offshore winds at night, as indicated by measurements at Y V R and Y X X which are shown in Figures 3.7(a) and (b), respec-tively. Maximum surface temperatures during the episode peaked at 27.4°C at Y V R and 33.2°C on 19 July, and remained high on the 20th. Daytime winds at Y V R and Y X X on 18 and 19 July were predominantly westerly. On 20 July, the day in which the highest 0 3 concentrations were observed, a southerly component was evident at both sites. Of note, these patterns were also observed during an extreme 0 3 episode occurring in September 1988 (Steyn et al., 1990). While several hours of calm were observed at Y X X on the morning of 21 July, strong easterlies were observed at Y V R . While still warm, daytime temperatures on 21 July were significantly lower than observed on the preceding days. Acoustic sounder measurements at sites in Delta and Surrey indicated that with the exception of 17 July, elevated daytime thermal structures were strong and consistent throughout the episode (Steyn and Wallis, 1986). Maximum mixed layer depths (MLDs) at the Delta site decreased from about 620 m on 17 July to 590 m on 19 July. At the Surrey site, maximum MLDs decreased from approximately 840 m on 17 July to 650 m on 18 and 19 July. For comparison purposes, the maximum M L D observed at the Surrey site during the September 1988 episode was 470 m (Steyn et al., 1990). Tethersonde measurements from Queen Elizabeth Park, Vancouver indicated that maximum MLDs decreased from approximately 530 m on 17 July to 320 m on 19 July. Air Quality Profile A number of stations in the L F V reported exceedances of the one-hour Maximum Acceptable Level for 0 3 (82 ppb) between 18-20 July 1985, and four of the stations Chapter 3. Episode and Domain Selection 36 Figure 3.6: (a) Sea level pressure (hPa); (b) 850 hPa temperature (°C); and (c) 500 hPa geopotential height (m), all for 21 July 1985 at 1200Z (0400 PST). Small rectangle in southeast quarter of each plot identifies modelling domain. July 1985 Figure 3.7: Hourly temperature (solid lines) and wind vectors (arrows) at (a) Y V R and (b) Y X X , 16-21 July 1985. CO Chapter 3. Episode and Domain Selection 38 Table 3.2: Exceedances of ambient O 3 objectives in the L F V , 17-21 July 1985. Date Number of Stations Exceeding: Maximum Maximum Maximum Ozone Desirable Acceptable Tolerable Maxima (51 ppb) (82 ppb) (153 ppb) (ppb) July 17 6/11 0/11 0/11 69 July 18 9/12 3/12 0/12 104 July 19 7/12 4/12 0/12 107 July 20 12/12 6/12 0/12 110 July 21 7/12 0/12 0/12 80 reported exceedances on consecutive days. A summary of the number of exceedances is presented in Table 3.2. The highest recorded concentrations during the episode were 107 ppb at station T12 in Chilliwack on July 19 and 110 ppb at station T15 in Surrey on July 20. The lowest O 3 concentrations and the highest NO2 concentrations were observed at sites located slightly upwind or within the urban core (stations T02, T03, T04 and T05). Conversely, the highest O 3 concentrations and the lowest NO2 concentrations were found at sites downwind of the urban plume (T15, T16, T i l and T12). Elevated O 3 levels were also found at T07 and T09, which are located at the eastern end of Burrard Inlet and in close proximity to oil refineries which were operating at the time. High nighttime O 3 concentrations at T14, located on Burnaby Mountain, reflect the buildup of 0 3 aloft and the lack of available NOx sources near this elevated site. No exceedances of the one-hour Maximum Acceptable Level for N 0 2 (210 ppb) were observed during the episode. However, elevated concentrations were observed at stations Chapter 3. Episode and Domain Selection 39 T04 and T05 in Burnaby. The highest concentrations recorded during the episode were 100 ppb at T05 on 18 July and 86 ppb at T05 on 19 July. 3.2 Domain Selection The selected modelling domain is as shown in Figure 3.8. The dimensions of the domain were 140 km by 110 km, with the origin located at U T M coordinates 460000 E and 5390000 N . For a grid cell size of 5 km by 5 km, this translated to a domain of 28 by 22 cells. Domain selection was based on the following criteria (after Seinfeld, 1988): 1. domain boundaries should contain all important current and future emission sources; 2. necessary air quality, emissions and meteorological data should be available for the domain; and 3. local circulation cells should be well-contained within the domain boundaries. As described in the previous section, air quality, emissions and meteorological data were available for the British Columbia side of the L F V . However, similarly detailed data were not available for the Washington State side. Clearly, both the British Columbia and Washington sides of the L F V had to be included in the modelling simulation in order to capture the major sources and recirculating wind patterns transporting pollutants in the valley. A back-trajectory study by Coligado (1988) for 17, 19 and 20 July 1985 indi-cated that air parcels originating from the Washington State side of the L F V produced maximum one-hour O 3 concentrations that were as high or higher than air parcels origi-nating in the G V R D . Preliminary trajectory studies performed by Miao (1993) for a sea breeze day (23 August 1985) indicated that sea breeze circulation and upslope/downslope winds were important pollutant transport mechanisms in the L F V . A larger domain is also preferable from the point of view of minimizing boundary effects on the main area Chapter 3. Episode and Domain Selection 40 Figure 3.8: Modelling domain, as delineated by thick, solid lines. Coastline and 100 m contrours depicted by thin, solid lines and dotted lines, respectively. of interest. However, this is tempered by the added computational requirements and the availability of observational data to help characterize the expanded domain. The selected domain size encompasses most of the L F V , with the exception of the town of Hope, B C to the east, and areas to the south of Bellingham, Washington. It is very similar to the typical domain size of 140 km x 120 km (4 km by 4 km grid cells) that was found by Tesche (1991) in reviewing 67 papers and reports on U A M studies. Chapter 4 Model Inputs U A M simulations require a total of 13 input files, as listed in Table 4.1. These files are used to define boundary and initial conditions, meteorological fields, emissions, terrain, chemical parameters and simulation control parameters. Further details regarding the preparation of these files are provided in the following. 4.1 Boundary and Initial Conditions The preparation of files describing model boundaries, associated boundary conditions and initial conditions are described in this section. 4.1.1 B O U N D A R Y The B O U N D A R Y files specifies (i) time-invariant lateral boundaries, and (ii) average hourly pollutant concentrations in each boundary cell. Dimensions of the modelling domain were 140 by 110 km, beginning at 460000 U T M E and 5390000 U T M N , as discussed in Section 3.2. Due to a lack of monitoring data near the domain boundaries, clean boundary conditions were based on those used by the South Coast Air Management District for the Los Angeles Basin (SCAQMD, 1990). These values are listed in Table 4.2. Reactive organic gas (ROG) and carbon monoxide (CO) concentrations of 10.3 ppbv and 200 ppbv, respectively, compare well with total non-methane hydrocarbon and CO concentrations of 9.4 ppbv and 161 ppbv measured during low-level (0.1 km) flights off the west coast of Vancouver Island in Apri l 1985 (Greenberg et al., 1990). This suggests 41 Chapter 4. Model Inputs 42 Table 4.1: U A M input files Input File Description BOUNDARY specifies the lateral domain boundaries and the concentrations of each species along each boundary. REGIONTOP specifies the height of the top of the domain; this height may vary in time and space. AIRQUALITY specifies the concentrations of each species in each grid at the start of the simulation. TOPCONC specifies the hourly concentrations of each species for each grid cell along the top of the domain. TERRAIN specifies the roughness and deposition factors for each grid cell along the bottom of the domain. DIFFBREAK specifies the hourly mixed layer depth (or diffusion break) during the daytime and the hourly inversion height at night. WIND specifies the hourly u and v wind components for each grid cell, the hourly maximum wind speed, and the hourly average wind speed along each boundary. TEMPERATUR specifies the hourly gridded temperature field. METSCALARS specifies the hourly values of meteorological parameters that do not vary spatially. This includes the N O 2 photolysis rate constant, the concentration of H 2 0 vapour, the temperature gradients above and below the diffusion break, atmospheric pressure and exposure class. EMISSIONS specifies hourly gridded emissions from low-level species. PTSOURCE specifies hourly emissions from elevated sources which are injected into the appropriate grid cells. This file includes stack parameters. CHEMPARAM specifies the species to be simulated, reaction rate constants, upper and lower bounds, activation energies, reference temperature, and resistances to surface sinks. SIMCONTROL specifies the period of simulation, model options, and information on the integration time steps. Chapter 4. Model Inputs 43 Table 4.2: Clean boundary conditions, South Coast Air Quality Management District (SCAQMD, 1990). Species Concentrations (ppm) NO 0.00025 N02 0.0005 03 0.04 O L E 0.00055 P A R 0.00622 T O L 0.0000914 X Y L 0.00004 F O R M 0.0005 ALD2 0.0005 E T H 0.00011 M E O H 0.0001 E T O H 0.0001 ISOP 0.0001 total R O G 0.0103 CO 0.2 that the S C A Q M D boundary conditions at least provide the correct order of magnitude for the L F V application. 4.1.2 REGIONTOP The R E G I O N T O P file specifies the top of the modelling domain. General practice is to set it to a constant level approximately 50 to 200 m above the maximum mixed layer depth (Morris and Myers, 1990). In the present application, it was set to to coincide with a height of 1372 m in the R A M S sigma coordinate system. The following conversion was Chapter 4. Model Inputs 44 used to obtain terrain-parallel coordinates: (H -zs) . . ztop = V „ 1 • 1372m (4.35) ti where z t o p is the height of the top of the modelling domain, H is the top of the R A M S domain (19 km) and z s is the surface elevation The height of 1372 m was obtained through consideration of the number and heights of the vertical layers used in R A M S which would provide adequate resolution of the mesoscale flows important in pollutant transport in the L F V . 4.1.3 A I R Q U A L I T Y Initial conditions are provided in the A I R Q U A L I T Y file. The initial concentra-tions are based on observations made at the simulation start time. To obtain a three-dimensional concentration field, station values were first interpolated horizontally using an inverse distance weighting factor. Chemical species were assumed to be well-mixed below the inversion height, and were scaled back to the concentrations at the top of the modelling domain which were specified by the R E G I O N T O P file. Where initial con-centrations were not available for certain species, clean air conditions as used in the B O U N D A R Y file (Table 4.2) were applied. 4.1.4 T O P C O N C Boundary conditions aloft are defined in the T O P C O N C file. In the absence of monitoring data at this elevation, clean boundary conditions as shown in Table 4.2 were used. Chapter 4. Model Inputs 45 4.2 Meteorological Data Meteorological output to describe wind fields, mixing heights and surface tempera-tures were generated in a mesoscale modelling study using the RAMS2a model (Steyn and Cai, 1994). Treatment of the data within the current research was limited to conversion of the data to a format compatible with the U A M . 4.2.1 WIND The WIND file contains gridded, hourly-averaged values of horizontal wind compo-nents u and v, hourly maximum wind speeds, and hourly averaged wind speeds for each boundary. The values of u and v were obtained by adjusting R A M S output to the format and grid coordinates required for the wind preprocessor U A M W N D . A feature of U A M - I V is that cell heights vary with the height of the mixed layer depth. In contrast, the vertical structure of R A M S is fixed. Vertical interpolation of R A M S output to a structure compatible with U A M input requirements can mask such important local-scale features as sea breezes and land breezes. For the purpose of this application, cell heights in the WIND preprocessor were fixed to levels coinciding with cell heights in the lower eight layers of the R A M S model. Cell interfaces were set at the following heights above ground level: 100 m, 215 m, 347 m, 499 m, 674 m, 875 m, 1106 m and 1372 m. Cell heights were adjusted from sigma coordinates to terrain-following coordinates as described in Section 4.1.2. Horizontal wind fields also required adjustments. The wind vectors u and v are calculated at the centre of each grid cell in the U A M , while they are calculated along perpendicular grid faces in R A M S , as summarized in Steyn and Cai (1994). The R A M S wind fields were for a grid size of 2.5 km by 2.5 km and a domain origin of 453750 Chapter 4. Model Inputs 46 U T M E and 5385750 U T M N . To convert to the U A M domain, a six-point inverse-distance-weighted interpolation scheme was applied to RAMS-generated values of u and v. Interpolations near domain boundaries were based on four- or three-point schemes, depending on the availability of neighbouring grid cells. The wind fields were subject to further adjustments in U A M W N D . Vertical velocity w is calculated based on three-dimensional divergence minimization. A stability adjustment is then applied to ensure zero vertical velocities at the top of the domain. Horizontal velocities u and v are then adjusted until the calculated divergence is within acceptable parameters. U A M W N D produces hourly, gridded horizontal wind speeds (in km/h) for each layer, in addition to hourly maximum wind speeds and average wind speeds along the boundaries. Wind fields produced by U A M W N D at 0900-1000 PST and 1500-1600 PST for each day of the episode are shown in Figure 4.1-4.3. Vector length is proportional to wind speed. Observed winds at Vancouver International Airport (YVR) , the University of British Columbia (UBC - located on the northwest 'tip of the City of Vancouver) and Abbotsford International Airport (YXX) are indicated by thicker arrows. Aside from the morning of 17 July, when initialization effects are still evident, morning wind pat-terns were characterized by strong northwesterly winds channelling down the Straight of Georgia. The simulated wind fields captured the start of the onshore sea breeze, which typically began about 1000 PST. However, it is evident that the modelled wind fields had a much stronger southerly component than the winds observed at Y V R and U B C . By late afternoon, the onshore wind pattern had been firmly established, while wind speeds over the Straight of Georgia had slackened considerably. Another characteristic of the afternoon wind fields is the strong southerly winds over such north-south channels as Indian Arm, Pitt River/Pitt Lake, Stave River/Stave Lake and Harrison River/Harrison Lake. These channels lie to the north of the major emissions Chapter 4. Model Inputs 47 area, and as such these flow patterns have large implications on pollutant transport to these wilderness/recreational areas. 4.2.2 D I F F B R E A K The D I F F B R E A K file specifies hourly, gridded heights of the daytime mixed layer and the nocturnal inversion base (hereafter referred to as z^). The D I F F B R E A K prepro-cessor calculates Zj using an algorithm developed by Kelly (1981) in which Zj is defined as the level at which the potential temperature exceeds that based on screen-level tem-perature measurements. To maintain dynamical consistency with other meteorological parameters, potential temperature estimates from the R A M S model were used. Modelled results and observed data for sites in Delta, Surrey, and Queen Elizabeth Park (QEP), Vancouver are shown in Figure 4.4. Predicted and observed daily maxima are presented in Table 4.3. Predicted values represent hourly averages. Data obtained from Q E P are based on an evaluation of tethersonde soundings. Data from the Delta and Surrey sites are from acoustic sounder measurements. No observations from Q E P and Delta were available for 18 July. Figure 4.4 indicates that aside from the first day of the simulation, when model start-up effects may have dominated, predicted values generally agreed with observed data in terms of the magnitude and timing of the daily maxima. However, there was a tendency to underestimate z;. At the Surrey site, the simulated development of the mixed layer was somewhat delayed compared to observed data, although the magnitude of the daily maximum z; agreed well. At QEP on 19 July, the morning development of z, was well-characterized, but the second maxima during the mid-afternoon was not captured. Chapter 4. Model Inputs 48 (a) 1 1 1 1 r - i — . 1 t « < * \\ / <• . « . x , t \\ \\ t * i » t \\ * <: j.\"V: /. .; . . . fl\\ .'•'>-•.•':':• • • |<::':: < \\ \\ -i.x, - - • X.X^X.x^xix. x, x. x x - -* • X X. X. X. ^ ~X X X, >'<\"\".: ' ^ x x x x x x . ^ x ^ - - • k X N X X x xix x x x_ V -- . y . X X X X \\ x x x x.-x -\\ * \\ \"x X \\ X X x x ^ /x -' V 1 v \\ \\ \\ X X X x x x\\-I \\ * * \\ X X X X X x *-'•• I'' 1 1 J 1 1_ . - \\ \\ , . . \\ \\ . \" . . \\ 1 . , / I 1 , • • I I I ,r-i t /.•-'•' S i t i i 460 500 520 5<0 UTM E 110**3 M l 0.M1IT01 UAXSIDH *KCTO* (b) \" T \" \"T\" > . . . / t / y\\ 1 I s I \\ I I. - . * — / / / y \\ \\ / y / 1 / , \\ \\ \\ I s ^ ^ ^ r I s ^ t \\ , ^ , , / , z.UJr*';. '': i - - '('~.K ' — ' t': ' ' ' ' S\\\\., ± ~ s y'\\A\\k - — / .7: ' I s ^ i f- .-1. \" >• S. y:Ji j/ y — y f / ^ ^ ^ - ^ ^ ' W . — - i -./-~ J s .—. — / ' ~ x.-x.-^ a^-xOx.\" x. C x-x. - — - *r - — . — ' * -x.-x.x.x_ x; x x x x - - xix.?- — S.X.X.XV..X x x x x x x x - - - - -X X. X. N X X. x -k X x x x x x, A • \\— - * - - - <••' > > ^ ^ ^ — • . t t t t t 1 I • I 1 I r \\ i-I I .!.•• r:r /: : » i » \\ i 500 520 540 UTM E (10**3 Ml Figure 4.1: Surface (level 1) wind fields prepared for U A M at (a) 0900 PST and (b) 1500 PST, 17 July 1985. Vector length is proportional to wind speed. Observations shown by thick arrows. Wind speeds in m/s. Chapter 4. Model Inputs (a) (b) • 1 M \\ ' 1 1 u r J M % \" - \\ \\ t • t ' / / . / / * 4 1 % 4 \\ \\ \" ' / \\ * / \\ - \\ . ; - \\ i / / , v N t ' * s -- 1 M \\ \\ t 1 / i \\ \\ - (•• < r ' < i ;C> t t \\ \\'\\ * - * x ' • / / » \\ • y>i 4 .» 1 1 /If;)/ , / \\ jlj ' * * y s I i \\ \" r ' / / ' i Jr } i t Ah/./ /Ji;/ v ' \\ \\ '. i r'/ v. t * , y v ' . / / ? . ^ / \\ \" \\ '. - i t i-.t t .•• •» ; r i . t r t'~-y , - -. • • ' / : / . • • \\ ' ! - v < • - ... < - - - . • . \\ - • • -V \\ i >-?. . . . . - - ; • -—- S / l 1 • \\ v -- \\ \\ \" - - -1. •/ ' ' ' / / / ' 1 V *, . . . 't : \\ \\ . -. I \\ > • /••'/ • < • • - . • t \\ 1 . v \\ V \\ - — - ** ^ ~ • • V 1 V •* - -* 1 i ,1 * 528 54B UTM E 110**3 Ml \\ . . I I , . / / _ ^ - ^ / ^ _ / \\ * „ r , / I \\ 1 -;i'V ^ _/% - / / ^ - , ^ / ? r _ ^ _-: j^; ^ — / / / ^ •^ _. / s ^ _ ^ - ^ I i ^ ^ ^ >.' . . ^. .^/ / _ _ ^ ^ . — , T / — i . 1 ' /.-''/ 520 • 540 UTM E (10**3 MI 49 0 6*6BtOI lanwuw VECTOR Figure 4.2: Surface (level 1) wind fields prepared for U A M at (a) 0900 PST and (b) 1500 PST, 18 July 1985. Vector length is proportional to wind speed. Observations shown by thick arrows. Wind speeds in m/s. Chapter 4. Model Inputs 50 (a) 520 540 UTM E (10**3 HI (b) - \\ u \" T \" .'.'•'A i -~ — ^ _ / / \\ I / / \\ t t , _ ^ , _ _ _ / / s \\ \\ i / s * t / . / / ^ , \\ \\ s ^ , . . . X. — ~- V ^ s t ^ — / 1 / 4, _ s\\ j \\ j ^ t ;'/: s ^ s f / . ^* A x ^- —../f;/ s / / / ;^z^*^'-*>- _4 - r / - ^ > * ^ ^ ^ ^ / / - ^ r — ' • ^ -. .. . . ^ ^ v . ^ x x N % \\ % ^ ••> N Y \\ \\ ^ \\ V \\ ' ' .. . . \\ . -. . , \\ \\ < -. . . \\ \\ \\ -. . ; \\ \\ \\: . _ , 1 1 /\" - - t I f ./,•' . / . / ; / / / • ^ V ; \" - - • <-'^ ^ ^ ^. ^. — —< * — ' - r — — — ^ % . 520 540 560 UTM E <10»*3 M) 0.TP5EHI VECTOR Figure 4.3: Surface (level 1) wind fields prepared for U A M at (a) 0900 PST and (b) 1500 PST, 19 July 1985. Vector length is proportional to wind speed. Observations shown by thick arrows. Wind speeds in m/s. Chapter 4. Model Inputs 51 1000 —. 1 1 . 1 . ' . 1 r-Q E P 800 - --p-600 _ o CD 0 0 ^ 4 0 0 - V / o o o n 0022-/ ^ ° o ° / ^ e A p 0 12 24 36 48 60 7i 17 July 18 July 19 July 140 17 July 18 July 19 July 140 120 100 80 60 40 20 °(hnnnr,rP T11 O n n n n n n r v i 0 12 17 July 36 18 July 60 7; 19 July 140 120 100 80 60 40 20 T12 12 17 July 36 18 July 60 7! 19 July Figure 5.3: Predicted and observed O 3 concentrations at stations T15, T16, T i l and T12 in the L F V , 17-19 July 1985. Solid lines represent predicted concentrations and hatched lines show the range in neighbouring grid cells. Open circles indicate observations. Chapter 5. Model Results and Discussion 71 • The model tends to underpredict O 3 concentrations during the day, and overpredict at night when observed values approach zero. The latter feature is particularly noticeable at stations T i l and T12, which are located downwind of the main source region. • With the exception of the first day of the simulation, the morning buildup of O 3 is represented quite well at most stations. However, the predicted morning buildup of O 3 levels at stations T07 and T09 is seen to lag the observations, even though there is good agreement in the timing of the daily maximum. The buildup of O 3 at rural sites T i l and T12 is also very poorly reproduced on 19 July. Spatial patterns of predicted O 3 concentrations (surface layer) are depicted in Fig-ures 5.4-5.6. For each day of the simulation, 0 3 isopleths at 0900, 1200, 1500 and 1800 PST are presented. Contours are plotted at 20 ppb intervals, beginning at 40 ppb. As shown in Figure 5.4, the morning of 17 July is characterized by relatively low O 3 concentrations throughout the entire modelling domain. By 1500 PST, concentrations in excess of 60 ppb are found over a wide band extending eastward along the Coast Mountains from Burrard Inlet, and a smaller area extending up-valley from Abbotsford to Chilliwack. As expected, 0 3 levels are depressed over the major source region, most likely the result of NO scavenging. By 1800 PST, ozone patterns are displaced further up the valley. The predicted patterns for 18 July shown in Figure 5.5 are characterized by higher concentrations over a much broader area than the previous day. At 0900 PST, spatial patterns are still very disorganized, but it is clear that levels are depressed over an area including Burrard Inlet and downtown Vancouver. At 1200 PST, O 3 levels in excess of 60 ppb are located over the Coast Mountains, north of Indian Arm and Pitt Lake, and over a wide band extending eastward from Boundary Bay in South Surrey towards Figure 5.4: Contours of predicted 0 3 at 20 ppb intervals on 17 July 1985 at (a) 0900 PST, (b) 1200 PST, (c) 1500 PST and (d) 1800 PST. Large numbers indicate observed concentrations. Missing values are indicated by \"-99\". Figure 5.4 (Continued). Figure 5.5: Contours of predicted O 3 at 20 ppb intervals on 18 July 1985 at (a) 0900 PST, (b) 1200 PST, (c) 1500 PST and (d) 1800 PST. Large numbers indicate observed concentrations. Missing values are indicated by \"-99\". Chapter 5. Model Results and Discussion 75 (c) AVERAGE , 85199/15.00 - 85199/16.00, level 1. species 03 \"I 5450 V-(d) AVERAGE , 85199/18.00 - 85199/19.00, level 1, species 03 520 540 X ( m e t e r s ) 560 580 60 Figure 5.5 (Continued). Figure 5.6: Contours of predicted O3 at 20 ppb intervals on 19 July 1985 at (a) 0900 PST, (b) 1200 PST, (c) 1500 PST and (d) 1800 PST. Large numbers indicate observed concentrations. Missing values are indicated by \"-99\". Chapter 5. Model Results and Discussion Figure 5.6 (Continued). Chapter 5. Model Results and Discussion 78 Chilliwack. An additional region of elevated 0 3 is found over the Straight of Georgia, to the south of Saturna Island. In contrast, depressed concentrations less than 40 ppb are found over an area encompassing Burrard Inlet. At 1500 PST, a large band of elevated 0 3 concentrations extends in an almost north-south fashion over the central valley. Maximum concentrations are found at the northern and southern ends of this band, with O 3 concentrations in excess of 80 ppb at the north end of Pitt Lake and more than 100 ppb to the southeast of Abbotsford. The very high concentrations observed at downwind sites T15 (97 ppb) and T12 (103 ppb) and the moderately high concentrations at T07 and T09 were not predicted by the model. The small region of elevated 0 3 concentrations located off Saturna Island at 1200 PST has now moved in a southeasterly direction down the Straight of Georgia. The area of depressed concentrations observed over Burrard Inlet at 1200 PST has now grown to encompass most of Vancouver, North Burnaby and Port Moody. By 1800 PST, the large area of elevated O 3 concentrations stretching across the valley has been displaced to the southeast, suggesting transport over the Cascade Mountains. The area of depleted O 3 has grown further to include most of the G V R D and much of the Washington State side of the L F V . Similar patterns are observed for 19 July, although maximum concentrations are somewhat higher. At 1200 PST, 0 3 concentrations in excess of 80 ppb are found over the Coast Mountains. The model fails to capture the elevated concentrations observed at T07 and T09. In contrast to the previous day, elevated 0 3 levels are not found off Saturna Island. At 1500 PST, the O 3 patterns are displaced further northward than found on the previous day. Peak concentrations in excess of 100 ppb are found to the north of Pitt Lake. High concentrations observed at stations T15 and T16 are not captured. By 1800 PST, the entire pattern is shifted eastward. Maximum predicted and observed ground-level 0 3 concentrations are presented in Table 5.1 on a daily basis for each station. The times of the respective maxima are also Chapter 5. Model Results and Discussion 79 Table 5.1: Observed and predicted maximum one-hour 0 3 concentrations in ppb and corresponding hour (PST) at monitoring stations in the L F V , 17-19 July 1985. 17 July 1985 18 July 1985 19 July 1985 Station Obs. Hr Pred. Hr Obs. Hr Pred. Hr Obs. Hr Pred. Hr 1 29.6 15 33.2 13 42.2 13 39.4 13 33.0 14 40.8 12 2 39.5 15 39.3 12 39.9 14 44.3 12 31.7 12 45.7 12 3 38.4 14 36.6 12 51.9 14 46.4 13 45.6 13 43.3 12 4 19.1 10 26.6 13 53.6 15 34.7 13 50.3 14 40.5 13 5 41.7 14 26.9 13 66.6 14 34.6 13 63.4 12 39.7 13 6 - - - - - - - - 42.0 14 43.4 13 7 63.7 14 41.7 14 74.2 15 49.5 14 79.9 13 53.2 14 9 61.0 14 32.0 14 84.0 14 41.9 14 95.0 15 45.3 13 11 69.2 15 60.5 16 95.4 14 89.7 16 106.7 14 63.1 14 12 73.9 17 72.6 17 102.5 16 71.9 18 107.2 16 81.7 16 14 42.1 14 29.8 13 58.7 14 37.1 14 39.0 2 43.4 13 15 71.3 16 47.6 14 104.4 15 70.1 13 87.8 15 69.9 13 16 66.0 15 40.6 13 79.0 15 61.3 15 95.0 12 59.9 14 presented. Maximum observed 0 3 concentrations increased over the three-day period, registering 73.9 ppb (T12), 104.4 ppb (T15) and 107.2 ppb (T12) on the 17th, 18th, and 19th, respectively. Maximum predicted concentrations over the same period were 72.6 ppb (T12), 89.7 ppb ( T i l ) and 81.7 ppb (T12), respectively. The time of the predicted maxima coincided with observations on the 17th and 19th, and lagged by one hour on the 18th. As indicated previously by the time series in Figures 5.1-5.3, there is an overwhelming tendency for the model to underpredict O 3 levels. A l l predicted concentrations and all residuals calculated between the predicted and observed concentrations are compared .against hourly observations in Figure 5.7. These plots indicate that there is a tendency for the model to overpredict low concentrations Chapter 5. Model Results and Discussion 80 (i.e. <20 ppb) and to underpredict concentrations in excess of 40 ppb. Mean bias and error based on predicted and observed O3 concentrations are presented for each station in Table 5.2. Paired observations and model predictions were only in-cluded where observations exceeded 20 ppb, and the number of pairs is also provided in Table 5.2. Although a slight tendency to overpredict 0 3 concentrations is noted for the two stations located on the west side of Vancouver (T01 and T02) on each day of the simulation, data from the other stations indicate an overwhelming tendency to under-predict peak 0 3 concentrations. This is evident even when relaxing spatial criteria to include the predicted concentrations in the nine nearest grid cells. Mean normalized bias ranged from 15.2% at T01 to -49.9% at T09 on the 17th, 14.8% at T01 to -57.7% at T09 on the 18th, and 38.8% at T01 to -61.8% at T09 on the 19th. Mean normalized gross error ranged from 14.4% at T03 to 49.9% at T09 on the 17th, 17.2% at T03 to 57.7% at T09 on the 18th, and 14.2% at T03 to 61.8% at T09 on the 19th. From the time series shown in Figures 5.1-5.3 and these performance statistics, it is evident that model performance is worst for those sites located immediately on the downwind side of the urban core, which is the main source region. Model performance was particularly poor for stations T04, T05 and T09, which are each located toward the eastern end of Burrard Inlet and in close proximity to petroleum refineries, and station T14, which is an elevated site. Averaged over all paired data from all stations, normalized bias ranged from -28.5 to -31.3% and normalized gross error ranged from 35.6% to 36.9% over the three days of the simulation. As a further measure of model performance, the accuracy of peak 0 3 concentrations paired and unpaired in space and time for the three-day simulation are summarized in Table 5.3, together with mean bias and gross error. Peak unpaired accuracy increased over the three-day simulation from -1.8% to -23.8%. However, results generally agreed with the performance goal of ±15-20% recommended by Tesche et al. (1990). The Chapter 5. Model Results and Discussion 81 Observed Ozone (ppb) 60 | 40 - • . 0 20 40 60 80 100 120 Observed Ozone (ppb) Figure 5.7: Comparison of predicted vs. observed O3 concentrations (above) and com-parison of residual (predicted-observed) vs. observed 0 3 concentrations (below), all sites 17-19 July 1985. Chapter 5. Model Results and Discussion 82 Table 5.2: Model performance based on predicted and observed maximum one-hour 0 3 concentrations at stations in the L F V , 17-19 July 1985. Station 17 July 1985 18 July 1985 19 July 1985 Bias Error No. Bias Error No. Bias Error No. 1 0.152 0.153 4 0.148 0.174 5 0.388 0.388 4 2 0.122 0.263 6 0.096 0.402 8 0.263 0.263 6 3 -0.012 0.144 6 -0.122 0.172 7 -0.079 0.142 9 4 - - - -0.468 0.467 8 -0.470 0.470 10 5 -0.353 0.405 9 -0.493 0.493 14 -0.551 0.551 12 6 - - - - - - -0.168 0.168 2 7 -0.336 0.336 10 -0.277 0.348 14 -0.231 0.387 20 9 -0.499 0.499 10 -0.577 0.577 12 -0.618 0.618 13 11 -0.225 0.250 12 -0.191 0.219 12 -0.250 0.288 12 12 -0.355 0.389 11 -0.223 0.285 10 -0.188 0.262 13 14 -0.388 0.484 13 -0.490 0.490 22 -0.384 0.385 11 15 -0.344 0.348 13 -0.297 0.312 14 -0.325 0.325 19 16 -0.383 0.383 10 -0.203 0.223 7 -0.256 0.311 12 Total -0.291 0.356 104 -0.313 0.375 133 -0.285 0.369 143 timing of peak predictions agreed well with observations on 17 and 19 July, and lagged the observed maxima by one hour on 18 July. The mean normalized gross error, which averaged 37% over the three-day period, was just outside of the performance goal of 30-35%. The poorest performance was for mean normalized bias, which averaged 30% and compared poorly to the goal of 5-15%. This statistic reflects the fact that there was an overwhelming tendency for the model to underestimate O 3 concentrations. Chapter 5. Model Results and Discussion 83 Table 5.3: Model 0 3 performance statistics for the L F V , 17-19 July 1985. Performance measure 17 July 1985 18 July 1985 19 July 1985 Maximum observed 0 3 concentration (ppb) 73.9 (T12) 104.4 (T15) 107.2 (T12) Maximum predicted O 3 concentration (ppb) 72.6 (T12) 89.7 ( T i l ) 81.7 (T12) Peak accuracy paired in space and time unpaired in space unpaired in space and time -1.8% -1.8% -1.8% -32.6% -19.4% -14.1% -23.8% -23.8% -23.8% Time of observed O 3 maximum 1600-1700 PST 1400-1500 PST 1500-1600 PST Time of predicted O 3 maximum 1600-1700 PST 1500-1600 PST 1500-1600 PST Mean normalized bias -29.1% -31.3% -28.5% Mean normalized gross error 35.6% 37.5% 36.9% Chapter 5. Model Results and Discussion 84 5.3 Nitrogen Dioxide Model performance is also evaluated in terms of the model's ability to reproduce N 0 2 concentrations. Time series of modelled and observed N 0 2 concentrations are compared in Figures 5.8-5.10. The following observations can be made from these time series: • The model can reproduce diurnal patterns of N 0 2 at those stations located near Burrard Inlet (T04,T05,T07 and T09) and in South Vancouver (T03). • Reproduction of observed patterns is very poor at stations T02 and T14, where the timing of peak levels is out of phase. • The model tends to underpredict peak N 0 2 concentrations. Poor model performance at station T02 reflects the inability of the model to simulate rather unusual local conditions in which observed N 0 2 concentrations monotonically increased throughout the episode. The characteristic diurnal patterns were not evident. Monitor malfunction was ruled out by the local monitoring agency ( G V R D , personal communication). Factors which may have contributed to this result include large and persistent NOx emissions and/or depressed mixing heights in the vicinity of the station, which is located on the roof of a high school in a residential/light commercial area of Vancouver. Spatial patterns of N 0 2 concentrations predicted for 17-19 July 1985 are depicted in Figures 5.11-5.13. For each day, N 0 2 isopleths at 0900 PST, 1200 PST, 1500 PST and 1800 PST are presented along with observed values. Contours are plotted at 10 ppb intervals, beginning at 10 ppb. At 0900 PST on 17 July, the 10 ppb contour extends across the valley in a northwest-southeast fashion. Concentrations in excess of 30 ppb are found to the northeast of Chapter 5. Model Results and Discussion 85 Figure 5.8: Predicted and observed NO2 concentrations at stations T02, T03, T04 and T05 in the L F V , 17-19 July 1985. Solid lines show predicted values. Open circles indicate observed data. Chapter 5. Model Results and Discussion 86 Figure 5.9: Predicted and observed N 0 2 concentrations at stations T07, T09 and T14 in the L F V , 17-19 July 1985. Solid lines show predicted values. Open circles indicate observed data. Chapter 5. Model Results and Discussion 87 Figure 5.10: Predicted and observed N 0 2 concentrations at stations T15, T16, T i l and T12 in the L F V , 17-19 July 1985. Solid lines show predicted values. Open circles indicate observed data. Chapter 5. Model Results and Discussion 88 Figure 5.11: Contours of predicted N 0 2 at 20 ppb intervals on 17 July 1985 at (a) 0900 PST, (b) 1200 PST, (c) 1500 PST and (d) 1800 PST. Large numbers indicate observed concentrations. Chapter 5. Model Results and Discussion 89 Figure 5.11 (Continued). Chapter 5. Model Results and Discussion 90 (a) AVERAGE , 85199/ 9.00 - 85199/10.00, level 1, species N02 5500 , 1 1 1 1 , 1 1 1 1 1 , 1 5490 \\-Figure 5.12: Contours of predicted N 0 2 at 20 ppb intervals on 18 July 1985 at (a) 0900 PST, (b) 1200 PST, (c) 1500 PST and (d) 1800 PST. Large numbers indicate observed concentrations. Chapter 5. Model Results and Discussion 91 (c) AVERAGE , 85199/15.00 - 85199/16.00, level 1, species N02 5500 , 1 1 , 1 , 1 , 1 , 1 1 1 1 5490 -5480 - „•:.,•''. Figure 5.12 (Continued). Chapter 5. Model Results and Discussion 92 (a) (b) AVERAGE , 85200/ 9.00 - 85200/10.00, level 1. species N02 « 5450 520 540 X ( m e t e r s ) AVERAGE , 85200/12.00 - 85200/13.00, level 1, species N02 520 540 X ( m e t e r s ) Figure 5.13: Contours of predicted N 0 2 at 20 ppb intervals on 19 July 1985 at (a) 0900 PST, (b) 1200 PST. (c) 1500 PST and (d) 1800 PST. Large numbers indicate observed concentrations. Chapter 5. Model Results and Discussion 93 (c) . 5450 AVERAGE , 85200/15.00 - 85300/16.00, level 1, species NOZ (d) AVERAGE , 85300/18.00 - 85200/19.00, level 1, species N02 i 1 r - i 1 q _i i i_ I :_j i ^ J u 520 540 X ( m e t e r s ) 1 3 . • • Figure 5.13 (Continued). Chapter 5. Model Results and Discussion 94 Burrard Inlet, toward the Pitt River valley. At 1200 PST, the L-shaped plume extends from the eastern end of Burrard Inlet to the Pitt River, and then spreads up the Pitt River valley to Pitt Lake. However, little change in the maximum concentrations are noted. By 1500 PST, the northern extension of the plume is no longer evident. N 0 2 concentrations in excess of 40 ppb are predicted over the area between Burrard Inlet and the Pitt River. Predicted N 0 2 concentrations at 0900 PST on 18 and 19 July are much higher than predicted on 17 July. Strong concentration gradients are centred over Burrard Inlet and North Burnaby. However, comparisons with observed values at station T05 in Burnaby (100 ppb) indicate that N 0 2 concentrations in this region were still greatly underpredicted during the morning hours. The afternoon hours are characterized by lower concentrations and weaker gradients, with predicted values in excess of 30 ppb found to the east of Burrard Inlet. These observations are further reflected in Table 5.4, which lists the daily maximum predicted and observed N 0 2 concentrations, and the respective times of these maxima, for each station. Maximum concentrations are greatly underpredicted at stations T04, T05 and T12 on 18 July, and at stations T05 and T12 on 19 July. The timing of observed maximum concentrations occurs during the late morning/early afternoon at most stations. However, at the downwind station T12 and the elevated station T14, the model predicts maxima to occur in the late evening. Predicted concentrations and the associated residuals (predicted - observed) are com-pared against observations in Figure 5.14. It is clear from this figure that the model has a tendency to underpredict N 0 2 concentrations above 30 ppb. Similar model performance statistics as calculated for O 3 were also calculated for N 0 2 , and the results are shown for individual stations in Table 5.5 and overall in Table 5.6. Paired predictions and observations were only included where observations exceeded 20 Chapter 5. Model Results and Discussion 95 Figure 5.14: (a) Comparison of predicted vs. observed 0 3 concentrations, and (b) com-parison of residual (predicted-observed) vs. observed 0 3 concentrations, all sites 17-19 July 1985. Chapter 5. Model Results and Discussion 96 Table 5.4: Observed (Obs.)and predicted (Pred.) maximum one-hour N 0 2 concentrations in ppb at monitoring stations in the L F V , 17-19 July 1985, along with the respective time of day (Hr). 17 July 1985 18 July 1985 19 July 1985 Station Obs. Hr Pred. Hr Obs. Hr Pred. Hr Obs. Hr Pred. Hr 1 2 21.0 21 14.6 23 43.0 21 52.5 9 58.0 24 47.1 9 3 23.0 21 17.6 17 30.0 8 35.0 8 38.0 7 35.0 1 4 48.0 10 37.3 18 86.0 10 48.4 11 62.0 9 55.3 11 5 40.0 22 32.7 24 100.0 9 53.0 11 86.0 9 52.4 11 6 - - - - - - - - 66.0 20 47.2 10 7 Q 28.0 12 20.4 10 46.0 10 32.2 12 22.0 12 34.8 12 O 9 35.0 22 36.7 18 59.0 11 38.6 12 41.0 19 44.5 12 10 38.0 12 32.7 19 55.0 10 35.0 13 42.0 8 40.8 12 11 - - - - - - - - - - - -12 - - - - 48.0 21 8.0 7 39.0 21 6.3 7 13 48.0 11 17.3 21 55.0 11 32.7 24 44.0 11 27.6 1 14 40.0 23 40.2 18 62.0 23 44.3 12 52.0 1 46.3 12 15 24.0 1 17.4 21 - 24 - - 30.0 21 11.0 8 16 - - - - - - - - - - -ppb. From Table 5.5, it is seen that the lowest mean bias and error is reported for stations T03 and T04. The highest mean bias and error values are reported for stations T02, T12, T13 and T15. However, poor data availability at stations like T15 prevent any further discussion of these results. On each day of the episode, maximum observed N 0 2 concentrations were recorded in North Burnaby, at monitoring stations in close proximity to existing oil refineries located along Burrard Inlet: 48.0 ppb (T04) on the 17th, 100.0 ppb (T05) on the 18th, and 86.0 ppb (T05) on the 19th. A station in North Delta, T13, also recorded a maximum N 0 2 Chapter 5. Model Results and Discussion 97 Table 5.5: Model performance based on observed and predicted maximum one-hour N 0 2 concentrations at stations in the L F V , 17-19 July 1985. Station 17 July 1985 18 July 1985 19 July 1985 Bias Error No. Bias Error No. Bias Error No. 2 -0.606 0.606 3 -0.274 0.682 13 -0.485 0.556 21 3 -0.373 0.373 3 -0.094 0.295 10 -0.156 0.292 11 4 -0.038 0.207 16 -0.194 0.248 17 -0.069 0.335 17 5 -0.114 0.183 19 -0.365 0.365 23 -0.418 0.430 21 6 - - - - - - -0.201 0.312 7 7 -0.307 0.307 2 -0.214 0.258 5 0.522 0.522 2 9 0.038 0.188 12 -0.209 0.326 19 -0.282 0.390 21 10 -0.048 0.185 11 -0.247 0.338 12 -0.210 0.341 17 12 - - - -0.247 0.338 12 -0.931 0.931 3 13 -0.612 0.612 9 -0.429 0.480 10 -0.487 0.533 11 14 0.071 0.283 5 -0.334 0.463 17 -0.217 0.458 12 15 - - - - - - -0.758 0.758 3 Total -0.144 0.269 80 -0.295 0.405 131 -0.307 0.433 146 concentration of 48.0 ppb on the 17th. Maximum predicted concentrations on each of the three days were 40.2 ppb at T14, 53.0 ppb at T05 and 55.3 ppb at T04, respectively. Peak concentrations were underpre-dicted by 49.0%, 63.3% and 53.9% on the three days, when paired in space and time. Relaxing spatial and temporal requirements, peak concentrations were underestimated by 16.3%, 47.8% and 37.1%. This was well within the acceptable N 0 2 criteria used by C A R B (1990). The time of peak concentrations was a minimum of 7 hours late on the 17th and 2 hours late on the 18th, when compared to observations. Poor performance on the 17th may be attributed to model start-up effects. Peak modelled concentrations Chapter 5. Model Results and Discussion 98 Table 5.6: Model N 0 2 performance statistics for the L F V , 17-19 July 1985. Performance measure 17 July 1985 18 July 1985 19 July 1985 Maximum observed N 0 2 concentration (ppb) 48.0 (T04/T13) 100.0 (T05) 86.0 (T05) Maximum predicted N 0 2 concentration (ppb) 40.2 (T14) 53.0 (T05) 55.3 (T04) Peak accuracy paired in space and time unpaired in space and time -49.0% -16.3% -63.3% -47.8% -53.9% -37.1% Time of observed N 0 2 maximum 0900-1100 PST 0800-0900 PST 0800-1100 PST Time of predicted N 0 2 maximum 1700-1800 PST 1000-1100 PST 0800-0900 PST Mean normalized bias -14.4% -29.5% -30.7% Mean normalized gross error 26.9% 40.5% 43.3% were within the period of maximum observed concentrations on the 19th. Mean normal-ized bias increased from -14.4% to -30.7% over the three-day simulation. This compares favourably with the performance goal of +/- 30%. Mean normalized gross error similarly increased, from 26.9% to 43.3%, but were well within the performance goal of 50%. Chapter 5. Model Results and Discussion 99 5.4 Discussion Although the general features of the 17-19 July 1985 0 3 episode were reproduced, it is clear that the model tended to underpredict both 0 3 and N 0 2 concentrations. Due to the number of input requirements by the model, it is difficult to isolate just one factor which may have contributed to these findings. However, some of the more important factors are discussed in the following. Firstly, the evaluation of model performance was based on comparisons between ob-servations made at a single point and model results volume-averaged over a grid cell with dimensions of 5 km by 5 km in the horizontal and 100 m or greater in the vertical. Hence, some discrepancies are expected between observed and predicted values. Previous studies have indicated that predicted peak 0 3 concentrations decrease with increasing grid size (Seigneur et al., 1981). Where sub-grid-scale effects are masked by grid cell averages, the discrepancy between predicted and observed concentrations will be larger. This may explain the poor characterization of NO2 concentrations at station T02 in Vancouver, as shown in Figure 5.8. A comparison of predicted wind fields and observations in Section 4.2.1 indicated that modelled wind fields had a greater southerly component than observed winds during the morning hours. This suggests that the influence of the Coast Mountains on local wind fields may have been overestimated. As a result, precursors emissions occurring in the Vancouver/Burnaby area during the morning rush hour would have been transported northwards, as reflected by the isopleths of predicted NO2 concentrations at 0900 PST shown in Figures 5.11-5.13. This may partly explain the underpredicted 0 3 concentra-tions at sites located further up the valley. However, this does not discount the finding that elevated 0 3 concentrations may occur along the valley walls. In fact, lidar data collected by McKendry et al. (1997) during Pacific'93 (Steyn et al., 1997) support this Chapter 5. Model Results and Discussion 100 finding, showing that pollutants are vented along the valley walls during the daytime. The mixed layer depth defines the vertical extent of mixing in the atmosphere. As such, an overestimate of Zj will yield an underestimate of predicted concentrations. It was shown in Figure 4.4) that values of z; were slightly underpredicted at sites in Van-couver, Delta and Surrey. While it is possible that Zj values were overpredicted at sites further inland, maximum predicted values for Abbotsford on 18 and 19 July (660 and 750 m) compared well with measurements made during Pacific'93 (Hayden et al., 1997). However, it is possible that the influence of marine bodies such as Burrard Inlet was underestimated, resulting in overpredicted values of z; near sites such as T07 and T09. The evaluation of model performance has been limited to an assessment of surface level concentrations, due in large part to the lack of measurements aloft. Station T14 on Burnaby Mountain provides an opportunity to investigate how well the vertical profiles were resolved. While predicted O3 concentrations agreed well with observations, NO2 concentrations were poorly characterized, particularly on 18 July when the predicted maximum was approximately 12 hours out of phase with the observed maximum. A number of factors in addition to the vertical structure may have contributed to this finding, including a failure to accurately characterize wind trajectories and local emissions near this site. The chemical mechanism represents a condensed version of the hundreds, if not thou-sands of reactions taking place in the atmosphere. Hence, it is accepted that the chemical mechanism introduces some degree of uncertainty in the predicted values. In a compari-son of C B M - I V (Whitten et al., 1980; Gery et al., 1988, 1989) and the C A L (Carter et a l , 1986; Lurmann et al., 1987) and Stockwell mechanisms (1990), Dodge (1989) found that the three mechanisms generally yielded similar results. Although this does not imply that the mechanisms were therefore correct, they do include the most up-to-date kinetic data and therefore, the \"best science\" of the time. Chapter 5. Model Results and Discussion 101 The emissions inventory is typically one of the largest sources of uncertainty in any modelling study due to the fact that it is impossible to characterize all sources con-tributing to NOx and V O C emissions. Depending on whether conditions are NOx- or VOC-sensitive, underestimated NOx or V O C emissions may respectively result in under-predicted O3 concentrations. It has been acknowledged that the 1985 inventory used in this simulation provided a rough first estimate (GVRD, 1988) and further refinements have since been made to this inventory (GVRD, 1994). As noted in Section 4.3, although overall changes were not large, changes within specific sectors were significant. They included a 23.5% reduction in NOx emissions from point sources and a 34.1% increase in V O C emissions from area sources. Mobile emission estimates used in the simulation were based on Mobile3c and Mobile4.1c estimates. The Cassiar Tunnel study conducted during Pacific'93 (Gertler et al., 1997) showed that Mobile4.1c underpredicted observed NOx and V O C concentrations by 29 and 23%, respectively. However, it was noted that the observed data were obtained under a best case scenario in which vehicles operated under hot-stabilized conditions and at constant speeds. Hence, higher emissions would be expected from vehicles operating under more normal conditions (i.e. variable speed and acceleration). As a means of explaining how errors in V O C or NOx emissions may have effected predictions, sensitivity indicators were considered. Sensitivity indicators are used to determine whether regions are V O C - or NOx-sensitive (Sillman, 1994; Milford et al., 1994; Chameides et al., 1992; Kleinman, 1986; Kleinman, 1991). They provide a means of estimating airshed characteristics without the use of photochemical air quality models. Sensitivity indicators can similarly be applied to the analyses of modelling results as tools for model performance evaluation. The following sensitivity indicators were used: NOy (Milford et al., 1989; Rao et al., 1993; Sillman et al., 1993; Milford et al., 1994), HCHO-to-NOy ratio (Sillman, 1995), Chapter 5. Model Results and Discussion 102 and 03 - t o -NOz ratio (Sillman, 1995). VOC-sensitive conditions were indicated by the following values: NOy concentration >20 ppb; 0 3 - t o - N O z ratio <7; and HCHO-to-NOy ratio <0.28. The indicators were calculated for daytime conditions only, when photochemical activity is occurring. NOy estimates are for the period from 0700-1800 PST, while the other indicators are for the period from 1200-1800 PST. Only those values from 18-19 July 1985 are shown, thereby excluding values which may have been influenced by model start-up effects on 17 July. Sensitivity indicators calculated for stations T02, T03 and T05 are shown in Fig-ure 5.15. Each of these stations is located slightly upwind of or in close proximity to the main source region, and so VOC-limited conditions would be expected. None of the three stations had consistent results for all three indicators, but both NOy concentrations and the HCHO-to-NOy ratio indicated that conditions near station T05 were VOC-limited. As shown in Figure 5.1, O 3 concentrations are underpredicted at this site. Hence, the indicator results suggest that V O C emissions or at least the reactivity of V O C emissions may have been underestimated near T05. Sensitivity indicators for stations T07, T09 and T14 are depicted in Figure 5.16. While the 03 - t o -NOz ratio indicated NOx-limited conditions, both NOy and the HCHO-to-NOy ratio indicated VOC-limited conditions at all three sites. This finding is also supported by the fact that N 0 2 concentrations were underestimated at these sites, sug-gesting insufficient radicals to drive N 0 2 formation, as depicted in Equations 1.7-1.9. Uncertainties with respect to the vertical profile of pollutants at station T14, an elevated site, also likely contributed to the poor performance at this site. Finally, indicators for stations T15, T i l and T12 are presented in Figure 5.17. A l l three indicators show NOx-limited conditions for most hours of the afternoon. This is as expected for an aged air mass at such downwind sites. Underpredicted peak O 3 concentrations (Figure 5.3) at the three sites suggests that NOx emissions, either locally Chapter 5. Model Results and Discussion 103 or downwind, were underestimated. The fact that the model could not reproduce the very low O3 concentrations observed at night also suggests that local NOx emissions were underestimated. &80 o 6 0 z 40 ! 1 . 100 1 1 1 1 1 T02 -80 T02 A - - IE 60 Q. -- g 40 A 20 0 — i • • i i 12 Hour (PST) 15 18 1 2 3 4 NOz (ppb) 40 60 NOy (ppb) 100 9 I c 140 120 100 (pp 80 o 60 z 40 20 0 AT03 A •+ -ft- * * 12 15 Hour (PST) 18 100 80 1[ 60 CL co O T03 1 1 • A * * ' ' ' A ; > 3 4 NOz (ppb) 40 60 NOy (ppb) 100 s co o c CO CO o\" S3 12 15 Hour (PST) 100r 80 -.a Q . 60 -a . CO 40 -O 20 -o L T05 • A A • . • A 2 3 4 5 NOz (ppb) 40 60 NOy (ppb) Figure 5.15: Sensitivity indicators based on predictions at stations T02, T03 and T05. Triangles show values for 18 July, diamonds for 19 July. £ 12 15 Hour (PST) 18 1 2 3 4 NOz (ppb) 5 6 10 S 6 0 1 4 0 1 2 T07 i 1 • A / * A , * » ^ A * -20 40 60 NOy (ppb) 80 100 100. 12 Hour (PST) 15 18 20 40 60 80 NOy (ppb) 100 12 15 Hour (PST) 18 10 .a C L B 6 0 1 4 0 1 2 T14 ~ i 1 1 * A A - / • / • 20 40 60 NOy (ppb) 100 Figure 5.16: Sensitivity indicators based on predictions at stations T07, T09 and T14. Triangles show values for 18 July, diamonds for 19 July. Chapter 5. Model Results and Discussion 106 oo (qdd) OHOH CO CD (qdd) OHOH ~ CD CO (qdd) OHOH (qdd) eo (qdd) CO (qdd) eo 1 1 1 30 <• - <»• a* - « « « - • • m > J-• i i e I-(73 , 0-3 O X | l CO (D 4 (qdd) A Q N 1 1 1 1 I