Measurement and modeling of traffic-related air pollution in the British Columbia Lower Mainland for use in health risk assessment and epidemiological analysis Final Report June 29, 2005 Sarah Henderson and Michael Brauer School of Occupational and Environmental Hygiene and Centre for Health and Environment Research The University of British Columbia 2206 East Mall, Vancouver, BC V6T 1Z3 www.cher.ubc.ca www.soeh.ubc.ca brauer@interchange.ubc.ca tel 604 822 9585 fax 604 822 9588 1 Summary As part of an ad hoc national initiative to characterize traffic pollution in major urban centres, the School of Occupational and Environmental Hygiene (SOEH) at the University of British Columbia (UBC) was contracted by the Greater Vancouver Regional District, the BC Ministry of Water, land and Air Protection, Environment Canada and Health Canada to develop spatial regression models for Greater Vancouver similar to those developed or underway for the metropolitan areas of Toronto, Montreal and Hamilton. This report briefly introduces the rationale for such models and gives detailed discussion of our methods and results. The objectives and deliverables are outlined below. 1.1 Objectives • To choose sampling locations that optimally reflect the variability in measured NO2 (as an indicator of motor vehicle related air pollution) using a location-allocation modeling algorithm, weighted for population density • To conduct two two-week sampling campaigns over periods that maximize the likelihood of averaged values reflecting the annual mean • To conduct fine particulate matter sampling at a limited subset of study sites • To use a geographic information system (GIS) to plot sites and generate potentially predictive variables based on land use, population density, traffic density, the road network and geography • To identify parsimonious linear regression models and to build ambient exposure surface maps from the resulting regression equations. 1.2 Deliverables • Report describing the study methodology, summarizing the measured data, quantifying the relationship between measured data and potentially descriptive spatial variables, and discussing the significance of our results. • Pollution surfaces/maps for the GVRD showing the estimated distribution of ambient concentrations, available as GIS grid files. 2 Introduction Traffic-related air pollution is a complex mixture of gases and particles, the exact composition of which is variable over time and space. Rather than attempting to measure all constituents of this mixture, this study measured nitrogen oxides (NOX) and fine particle (PM2.5) mass and light absorbing carbon, which are proxies for overall traffic-related air pollution. As discussed in our report on pilot findings for this study1, traffic-related air pollution and its health effects have received considerable attention over recent years. Emission inventories consistently identify motor vehicles as a predominant source of anthropogenic pollution, and epidemiological research has associated residence in proximity to traffic corridors and measurements of traffic-related air pollution with a number of health outcomes, including asthma, low birth weight and mortality. To better understand how traffic-related air pollution affects the health of people living in the Greater Vancouver Regional District (GVRD) it is necessary to understand the distribution of exposure in the population. Because individual exposure is dependent on countless personal factors it is impossible to model at a population-based level, but the spatial distribution of ambient concentrations may be a good proxy for personal exposure and is relatively straightforward to estimate. By measuring NOX and PM2.5 at 116 and 25 sites, respectively, and using linear regression to relate concentrations to relevant and readily available spatial variables, we were able to map estimated traffic pollution over the entire GVRD. These high resolution air pollution maps can then be used in risk assessment or in epidemiological studies to estimate individual, ambient exposure to traffic pollution based on the residential address of study subjects. 2.1 Nitrogen Oxides According to the year 2000 Emissions Inventory for the Canadian portion of the Lower Fraser Valley, vehicle emissions are the dominant source of NOX with 24% and 16% of annual releases being attributed to light- and heavy-duty vehicles, respectively.2 The primary NOX compounds are nitric oxide (NO) and nitrogen dioxide (NO2), both of which are products of fossil fuel combustion. Most traffic-related NOX is emitted as NO and quickly reacts with atmospheric ozone (O3) to produce NO2. Nitric oxide is considered to be indicative of localized traffic impacts, with concentrations decreasing with increasing distance from major roadways. Nitrogen dioxide, a secondary atmospheric reaction product, is spatially more homogeneous and is indicative of wider-scale traffic impacts. By measuring NO and NO2 we were able to model both impacts, and future epidemiological studies might be able to gain further insight into differential health outcomes associated with proximity to heavy traffic. 2.2 Fine Particulate Matter Vehicles are also an important source of PM2.5 (particles with aerodynamic diameter less than 2.5 micrometers) and are reported to be responsible for 8% of annual Lower Fraser Valley emissions by weight, although this may not be the most relevant metric. Previous research has shown that particles in vehicle exhaust follow a bimodal distribution, with a nuclei-mode peak around 15 and an accumulation-mode peak around 50 nanometres 3,4 Without finer speciation of particulate matter in the 1 Brauer M and Henderson S (2003). Diesel Exhaust Particles and Related Air Pollution from Traffic Sources in the Lower Mainland. Report submitted to Health Canada; Vancouver, BC. 23 pages. http://www.cher.ubc.ca/PDFs/diesel02.pdf 2 Greater Vancouver Regional District. 2000 Emission Inventory for the Canadian Portion of the Lower Fraser Valley Airshed. (2003). Burnaby, BC. 119 pages. http://www.gvrd.bc.ca/air/pdfs/2000EmissionInventory-Canada.pdf 3 Abu-Allaban M, Rogers CF, and Gertler AW (2004). A quantitative description of vehicle exhaust particle size distributions in a highway tunnel. Journal of the Air & Waste Management Association, 54(3): p. 360-366. 4 Ristovski ZD, Morawska L, Bofinger ND, et al. (1998). Submicrometer and Supermicrometer Particulate Emission from Spark Ignition Vehicles. Environmental Science and Technology, 32(24): p. 3845-3852. Emissions Inventory it is difficult to know what fraction of ultrafine particulate matter (particles with diameter less than 100 nanometers) in the Lower Fraser Valley is attributable to vehicular sources. Because particles of different size and composition have different respiratory deposition characteristics are posess different toxicity, this is important information for understanding the contribution of mobile sources to the overall public health risk. While the size distribution of particles in heavy duty diesel (HDD) and light duty spark ignition (LDSI) vehicle exhaust is similar, the mass and composition of the particulate matter is not. In a recent analysis of tunnel air quality Abu-Allaban et al. estimated the emission rates of HDD and LDSI vehicles to be 2.8×1014 and 1.8×1013 particles/vehicle-km, respectively.5 Other studies have demonstrated that elemental carbon, trace metals and adsorbed organic material are present in different fractions in HDD and LDSI exhausts6. The composition of particles resulting from incomplete diesel combustion is relatively high in elemental carbon, which serves as a marker to differentiate between HDD and LDSI pollution.7 Light absorption by particulate matter is primarily due to the presence of elemental carbon, so reflectance measurements on sampled filters provide important information about the relative intensity of HDD traffic at each of our sampling sites. More generally, filter light absorption is a good indicator of vehicle exhaust in urban areas and was used in this study as a surrogate for the particle component of vehicle exhaust. 2.3 Spatial Regression Modeling for Exposure Assessment Under ideal circumstances it would be feasible to measure individual exposure to traffic pollution for all residents of the GVRD, but even studies that take such measurements on a few hundred people are prohibitively expensive. However, it remains impossible to quantify the relationship between traffic exposure and human health without somehow estimating exposure. The validity of epidemiological results is dependent on the accuracy of the exposure estimates, which is good is incentive to move beyond crude estimates from discreet air quality monitoring sites locations by developing more sophisticated approaches. Traffic pollution models are one solution that, as discussed in our pilot report, the dispersion of pollution from mobile sources is challenging to model due to the vast amount of location-specific data necessary over the entire area of interest.8 In recent years the development of geographic modeling techniques has yielded promising alternatives for the generation quality, population-based exposure estimates based on readily-available data9. 5 Abu-Allaban M, Rogers CF, and Gertler AW (2004). A quantitative description of vehicle exhaust particle size distributions in a highway tunnel. Journal of the Air & Waste Management Association, 54(3): p. 360-366. 6 Andrew Gray H and Cass GR (1998). Source contributions to atmospheric fine carbon particle concentrations. Atmospheric Environment, 32(22): p. 3805-3825 7 Jeong C-H, Hopke PK, Kim E, et al. (2004). The comparison between thermal-optical transmittance elemental carbon and Aethalometer black carbon measured at multiple monitoring sites. Atmospheric Environment, 38(31): p. 5193-5204. 8 Brauer M and Henderson S (2003). Diesel Exhaust Particles and Related Air Pollution from Traffic Sources in the Lower Mainland. Report submitted to Health Canada; Vancouver, BC. 23 pages. http://www.cher.ubc.ca/PDFs/diesel02.pdf 9 Brauer M, Hoek G, Van Vliet P, Meliefste K, Fischer P, Gehring U, Heinrich J, Cyrys J, Bellander T, Lewne M, Brunekreef B. Prediction of long term average particulate air pollution concentrations by traffic indicators for epidemiological studies. Epidemiology, 2003;14:228–239. 3 Methods 3.1 Site Selection 3.1.1 Location Allocation Location-allocation modeling for optimal sampling site location is a mathematically complex process that has been described elsewhere.10 Briefly, this two-step algorithm involves building a demand surface and then solving a constrained spatial optimization problem to determine optimal locations for a pre-specified number of samplers. In essence the demand surface uses information available prior to sampling to estimate how concentrations of a pollutant are distributed in the study domain. Typical inputs include locations and measurements from regulatory monitors, land use coverage, and population density. Sites are then selected so that they capture the complete range of pollutant values while being placed at maximum distance from one another. Our colleagues at the McMaster Institute for Environment and Health used this approach to identify 100 sampling locations in the GVRD based on an NO2 demand surface. This procedure is described in greater detail in our interim report (Appendix A). 3.1.2 Additional Sites We identified 16 additional special interest sites within the GVRD. Site E1 was located outside of an apartment block in the municipality of Surrey whose residents had issued public complaints about local traffic. Sites E2 through E7 were at intersections that service intense diesel traffic within the municipality of Vancouver. Sites E8 through E10 were placed along the proposed route of a new freeway in south Surrey, near the Fraser River. Sites E11 through E15 were located in the town of Abbotsford, which was not included in the location-allocation model but which is a location for which residents have expressed concerns regarding local and transported air quality. Site E16 was at a busy intersection in Port Moody where a local resident had made repeated complaints about the air quality. 3.1.3 PM2.5 Subset A subset of 25 sites was selected for inclusion in a particle sampling (PM2.5) campaign. Random numbers were assigned to the 100 location-allocation sites and then sorted in ascending order, and the first 25 sites were chosen although some substitutions were made to ensure adequate variability and to accommodate the larger instrumentation. 3.2 Sampling 3.2.1 Nitrogen Oxides Sampling Period Optimization The ambient concentration of nitrogen dioxide (NO2) in the GVRD is known to follow a seasonal cycle with peak concentrations in winter and lower concentrations during the summer period11. We decided a priori to conduct two sampling campaigns timed to best estimate the annual ambient NO2 concentrations at our sampling sites. To identify the optimal sampling periods we studied data from 13 GVRD fixed monitoring stations for the years spanning 1994 through 1998. Running two week averages were calculated for each site starting on January 1st of each year, and then opposing values 10 Kanaroglou PS, Jerrett M, Morrison J, et al. (2005). Establishing an air pollution monitoring network for intra-urban population exposure assessment: A location-allocation approach. Atmospheric Environment, 39(13): p. 2399-2409 11 Brauer M, Brumm J, Ebelt S. Evaluation of ambient air pollution in the Lower Mainland of British Columbia: Public health impacts, spatial variability, and temporal patterns. Report to the Administrative Council of Lower Mainland Medical Health Officers. May 15, 2000. http://www.cher.ubc.ca/PDFs/LMAirQualityReport.pdf (i.e. those 26 weeks apart) were averaged and compared to the annual mean. Results consistently demonstrated that the combination of a period in February and September produced an average within 15% of the annual mean, as is evident from Figure 3.1. Figure 3.1. Range of annual NO2 averages compared to two-week running averages from 13 monitoring stations in the GVRD between 1994 and 1998. Ogawa Samplers These small, cylindrical passive samplers measure 2.5 cm long by 1.5 cm in diameter, with a space in each end to house Ogawa pre-coated pollutant collection pads. In this study we placed an NOX pad in one end, and an NO2 pad in the other, enabling us to ascertain additional concentrations for NO and SO2. The protocol suggested by Ogawa USA was followed for the preparation, storage and transport of all samplers12 Deployment One hundred and thirty-two Ogawa samplers were deployed at 116 sites in the GVRD between February 24-28 and September 8-12 2003. The weather was clear during both periods, minimizing the potential for rainwater contamination of the collection pads. Samplers were installed by UBC technicians on lampposts or hydro poles at heights ranging from 10 to 12 feet above the ground, with the following information recorded at each site: 1. Site and sampler identification numbers 2. Latitude, longitude, elevation and GPS accuracy 3. Ordinal indicator of automobile traffic (1 = very light and 5 = very heavy). 4. Ordinal indicator of diesel truck traffic (1 = very light and 5 = very heavy). 5. Binary indicator of 4- or 5-way intersection (0 = no and 1 = yes). 6. Number of lanes on the closest road. 7. Exact details of sampler location (pole #, etc.) to ensure replicability. Samplers were exposed for 14 days after installation, and were collected in the same order they were deployed between March 10-14 and September 22-26. The weather was overcast and wet during the March period, and special precautions were taken to ensure that samplers remained dry while they were disassembled and stored in preparation for extraction and analysis. 12 NO/NO2/NOx and SO2 sampling protocol using the Ogawa sampler. Version 4.0, 1998, Ogawa and Co., USA. http://www.ogawausa.com/Prov398.exe Quality Control Three quality control measures were employed to assess the validity of sampler results. First, duplicate samples were taken at 18 sites. Second, Ogawa samplers were co-located at 14 (Table 3-1) air quality monitoring stations in the GVRD so that results could be compared to those from continuous samplers. Third, a total of 25 field blanks were deployed at the 14 GVRD stations. These samplers remained in leak-proof bags and light-safe containers for the duration of the sampling period. Table 3-1. GVRD air quality stations at which Ogawa samplers were co-located Station ID Station Location T1 Robson Street, Downtown Vancouver T2 Kitsilano High School, Vancouver T4 Kensington Park, North Burnaby T6 Second Narrows Bridge, North Vancouver T9 Rocky Point Park, Port Moody T13 116th Street, North Delta T15 72nd Avenue, East Surrey T17 Aragon Road, South Richmond T18 Rumble Street, South Burnaby T20 Meadowlands Elementary School, Pitt Meadows T27 52nd Avenue, Langley T30 Golden Ears Elementary School, Maple Ridge T32 Douglas College, Coquitlam T33 Bevan Avenue, Abbotsford 3.2.2 Particulate Matter (PM2.5) Harvard Impactors These instruments are connected to low-flow pumps that draw air through a series of two impactors to remove all particles greater than 2.5 µm in aerodynamic diameter. Smaller particles are then trapped on a pre-weighed Teflo® filter, which is weighed again after sampling to ascertain the average particle concentration. Instruments were cleaned and calibrated prior to deployment, and filters were treated according to standard methods. Samplers were deployed in environment-proof plastic cases that were adapted to house the impactor (with the intake shaft protruding) and an SKC programmable low-flow pump, as shown in Figure 3.2. Figure 3.2. Harvard impactor and pump in environment-proof case Sampling cases were installed by UBC technicians on the same posts as the Ogawa samplers at a height of 10 feet. The pumps were programmed to sample for a total duration of 1440 minutes (24 hours) over a 9999 minute period (81 minutes short of 7 days) at a flow rate of approximately 4 L/min. Study Sites Particle sampling was conducted in the spring and fall of 2003, though only spring results are presented here due to several technical challenges during the fall period. Due to limited pump availability five impactors were deployed weekly for a total of seven weeks between March 5th and May 8th, 2003 to ensure a complete set of data. The following information was recorded at each site: 1. Site and filter identifiers. 2. Date and time of the installation and collection. 3. Pump flow rate at time of deployment and collection. 4. Elapsed pump time at time of collection (9999 for a complete sample, but less if a battery fault or flow fault occurred during the sampling period). Sampled filters were stored in an environmentally controlled room between collection and analysis. Quality Control Two quality control strategies were used to verify the accuracy of results from the Harvard impactors. First, impactors were installed on the air intake manifolds at GVRD stations T2 (Kitsilano) and T18 (Burnaby South) for the duration of the sampling period. These stations were chosen for their centralized locations; an important consideration when adjusting for temporal effects. Both have continuous PM10 samplers, and T18 also has one of the seven PM2.5 samplers in the region. Like their roaming counterparts, the pumps at these locations were programmed to collect 1440 minutes of sample over a 9999 minute period. Collection filters were changed weekly, on the same day that the roaming pumps were deployed, and pumps were re-calibrated after every filter change. Second, a field blank (a filter loaded into an impactor that was not attached to a pump) was located at station T2 at the beginning of every sampling period. A total of 6 field blanks was collected. 3.3 Sample Analysis 3.3.1 Oxides of Nitrogen and Sulfur Ogawa NOX and NO2 collection pads were extracted into 6 ml of deionized water between March 18- 28 and October 6-10, 2003. To generate sulfate from adsorbed SO2 the spring NO2 extractions were further partitioned into two 3 ml samples, one of which was treated with 5 µl of 3% hydrogen peroxide (H2O2) as per instructions in the Ogawa sampling protocol. Extracted samples were refrigerated until analysis by ion chromatography, which occurred between May 6-22 and October. Nitrite concentrations in the samples were detected by wavelength and suppressed ion conductivity, while sulfate concentrations were detected by conductivity alone. Aqueous concentrations were converted to concentrations in air using the average temperature and relative humidity for the sampling period, the sample duration in minutes, and the appropriate Ogawa constants. Average temperature was calculated from continuous monitoring data collected at 16 of the 17 eligible GVRD air quality stations within the spatial range of the sampling locations. Station T14 (Burnaby Mountain), was excluded due to its elevation. Average relative humidity was calculated from data collected at all 5 eligible GVRD stations within the sampling range. Temperature and relative humidity were calculated for moving two-week averages, and applied depending on the date of sampler deployment. 3.3.2 Particulate Matter Mass concentration The post-sampling weight of all spring filters was measured on two occasions and, due to some instability in the first set, the second set of measurements was used. Average PM2.5 was calculated by dividing the filter weight gain by the total volume of air that passed through it (based on initial and final flow rates). Instability in the weight measurements given the low amount of collected mass indictated that these results could not be considered accurate, and the more objective filter absorbance measurement is presented throughout the remainder of this document. Absorbance Filter reflectance was measured with an M43D Smokestain Reflectometer using standard methods. 13 Results were used to calculate filter absorbance with the following equation: Abs = 100000 ereflectancfilter ereflectancblank field average ln filter through passedair of volume 2 area surfacefilter ××       (Equation 1)14 Previous results have shown that filter absorbance is strongly correlated with the EC concentration15,16, which is estimated here from the regression relationship measured at the GVRD’s South Burnaby station in 2001, shown in Equation 2. EC = 1.196 0.09145 - Absorbance (Equation 2)17 Temporal Adjustments To account for wide-scale temporal variations in pollutant concentrations during the sampling period all results were adjusted to data from the fixed monitoring station at T18 in south Burnaby. Adjustment ratios for PM2.5 concentration, absorbance, and estimated elemental carbon (EC) were calculated for each sampling period using Equation 1. Measured values were then divided by the adjustment ratio, resulting in an increase if the one-week average underrepresented the overall average, or a decrease if the opposite was true. Adjustment Ratio = period sampling entirefor station fixedat ion concentrat average weeksampling onefor station fixedat ion concentrat average (Equation 3) 13Determination of absorption coefficient using reflectometric method. ULTRA “Exposure and risk assessment for fine and ultrafine particles in ambient air”. 16 October 1998. http://www.ktl.fi/ultra/adobe/out/sop-abs.pdf 14 Ambient air determination of a black smoke index. International Organization for Standardization. International Standard 9835-1993 (E), 1993. 15 Rich K. Air pollution and patients with implanted cardiac defibrillators [microform] : an epidemiological analysis and assessment of exposure. University of British Columbia. School of Occupational &Environmental Hygiene. Thesis. M.Sc., 2003. 16 Noullett M. Exposure to fine particulate air pollution in Prince George, British Columbia. University of Northern British Columbia. Natural Resources & Environmental Studies ; 2004. Thesis M.Sc. 2004. 17 Rich K. Air pollution and patients with implanted cardiac defibrillators [microform] : an epidemiological analysis and assessment of exposure. University of British Columbia. School of Occupational &Environmental Hygiene. Thesis. M.Sc., 2003. 3.4 Data Analysis 3.4.1 Variable Generation Overview A total of 98 potentially predictive variables in five categories were generated based on spatial attributes of the 116 monitoring sites. An exhaustive list of variable titles with descriptions is found in Table 3-2. All variables were created in ArcView 3.218 with geographic data projected to the North American Datum 1983 (NAD83) and expressed in units of the Universal Transverse Mercator (UTM). Table 3-2. Complete list of predictor variables generated for regression modeling Variable Name(s) Description Category DIST.RD1 Distance in km to nearest highway Road Length DENS.RD123 Density in km/hectare of road network within a 300 m radius Road Length RD1.100, RD1.200, RD1.300, RD1.500, RD1.750 and RD1.1000 Total length in km of all highways within the 100 to 1000 m radii Road Length RD1.200-100, RD1.500-100, RD1.500- 300, RD1.1000-500 and RD1.1000-750 Total length in km of all highways within the specified annuli (doughnuts) Road Length RD2.100, RD2.200, RD2.300, RD2.500, RD2.750 and RD2.1000 Total length in km of all major roads within the 100 to 1000 m radii Road Length RD2.200-100, RD2-500.100, RD2.500- 300, RD2.1000-500 and RD2.1000-750 Total length in km of all major roads within the specified annuli (doughnuts) Road Length TRK.100, TRK.200, TRK.300, TRK.500, TRK.750 and TRK.1000 Total length in km of all designated truck routes within the 100 to 1000 m radii Road Length TRK.200-100, TRK-500.100, TRK.500- 300, TRK.1000-500 and TRK.1000-750 Total length in km of all designated truck routes within the specified annuli (doughnuts) Road Length AD.100, AD.200, AD.300, AD.500, AD.750 and AD.1000 Density of automobile traffic in vehicles/hectare/hour within the 100 to 1000 m radii Traffic Density AD.200-100, AD.500-100, AD.500-300, AD.1000-500 and AD.1000-750 Density of automobile traffic in vehicles/hectare/hour within the specified annuli (doughnuts) Traffic Density TD.100, TD.200, TD.300, TD.500, TD.750 and TD.1000 Density of diesel traffic in truck /hectare/hour within the 100 to 1000 m radii Traffic Density TD.200-100, TD.500-100, TD.500-300, TD.1000-500 and TD.1000-750 Density of diesel traffic in trucks/hectare/hour within the specified annuli (doughnuts) Traffic Density X.10000 and Y.10000 X and Y coordinates in UTM NAD-83 projection divided by 10,000 Geographic ELEV Elevation in metres Geographic SHORE Distance to the Pacific Ocean in km Geographic EAP.750, EAP.1000, EAP.1250, EAP.1500, EAP.2000 and EAP.2500 Population kernel density calculated within the 750 to 2500 m search radii Population Density COMM.300, COMM.400, COMM.500 and COMM.750 Hectares of commercially zoned land within the 300 to 750 m radii Land Use RES.300, RES.400, RES.500 and RES.750 Hectares of residentially zoned land within the 300 to 750 m radii Land Use GOV.300, GOV.400, GOV.500 and GOV.750 Hectares of government/institutional zoned land within the 300 to 750 m radii Land Use IND.300, IND.400, IND.500 and IND.750 Hectares of industrially/resource zoned land within the 300 to 750 m radii Land Use OPEN.300, OPEN.400, OPEN.500 and OPEN.750 Hectares of open land and/or waterbody within the 300 to 750 m radii Land Use With the exception of ELEV, X.10000 and Y.10000, all covariate grid layers were produced by spatial analysis of vector input layers, as summarized in Table 3-3. The specific methods for each variable group are discussed in the upcoming sections. 18 Environmental Systems Research Institute (1997).ArcView 3.2. Geographic Information System. ESRI Inc. Table 3-3. Description of the predictor variable layers Variable Group (units) Sub-Types (nomenclature) Base File (source, type) Spatial Analyst Feature Used Output Resolution (m) Road Length (km)/ Distance to Road (km) Freeways (RD1) Major Roads (RD2) Truck Routes (TRK) Distance to Freeway Road Network (DMTI, line) Neighborhood Statistics/Find Distance 10 Traffic Density (vehicles/hour/hectare) Automobile Density (AD) Truck Density (TD) Transportation Model (GVRD, line) Calculate Density 10 Population Density (per hectare) Person Density (EAP) Dissemination Area (DMTI, polygon) Calculate Density 5 Land Use Area (hectares) Residential (RES) Commercial (COMM) Open Space (OPEN) Governmental (GOV) Industrial (IND) Land Use (DMTI, polygon) Neighborhood Statistics 10 Geographic Elevation (ELEV) Longitude (X) Latitude (Y) Digital Elevation Map (DMTI, raster) n/a 30 Road Length Variables Road length variables (RD1 and RD2) were generated from a street network file for British Columbia distributed by DMTI Spatial. This file included six road type categories, which we collapsed into three to ensure consistency with the modeling methodology used for the Toronto data19. Road types one (RD1), two (RD2) and three (not used in regression modeling) include all freeways and highways, all major arterial roads, and all minor and residential roads, respectively, as shown in Figure 3.3. The truck route length variable (TRK) was generated from a designated truck route file that we constructed according to the methods outlined in Appendix B, with the final results shown in Figure 3.4. Figure 3.3. Greater Vancouver street network according to road type Buffers with the specified radii were constructed around each of the 116 sampling locations, and the ‘Neighbourhood Statistics’ function of Spatial Analyst20 was used to determine the total length of each 19 Jerrett M, et al., in press, 2005 20 Environmental Systems Research Institute (1997).ArcView 3.2. Geographic Information System. ESRI Inc. road type within each buffer. This function works by converting the line file to a grid file of specified cell size (5 metres in this case), then summing the total area associated with each road type for each buffer. Dividing this value by five gives the approximate total length of the road type within the buffer, though Figure 3.5 demonstrates how error may be introduced when the roads are winding or cross the gridded field on a diagonal. Annulus variables were created by subtracting results for the inner buffer size from results for the outer buffer size. Figure 3.4. Designated truck route network in Greater Vancouver Figure 3.5. Potential for introduction of error during the conversion of a road line file to a grid file (B Beckerman, personal communication,2004). Traffic Density Variables Traffic density variables were generated using output from the 2001 EMME/2 planning model, which was jointly developed by the GVRD, TransLink, and the BC Ministry of Transportation and Highways to improve understanding of transportation-related issues in the district. The correlation between output from this model and traffic count data collected from the GVRD is 0.85 at 20 sites spanning volumes from 700 to 4000 vehicles/hour. Model output estimating automobile and truck volumes during morning rush hour was used for this study. Data were received in a line file and the ‘Poly to Points’ ArcView extension21 was used to convert lines to points at a resolution of one meter with each point having the traffic count attributes of the line segment from which it came. Again, we used the ‘Calculate Density (simple)’ function of Spatial Analyst at search radii of 100, 200, 300, 500, 750 and 1000 meters to generate the variable layers at a 10m grid resolution, then used the ‘Get Grid Value’ extension22 to assign variable values to each sampling point, as illustrated in Figure 3.6. The annulus 21 Huber W (2002).Poly to Points 1.0. Avenue script for ESRI ArcView GIS. Quantitative Decisions. 22 Davies J and Elmquist M (2000).Get Grid Value 2.0. Avenue script for ArcView GIS. density variables were calculated by the following equation, where X and Y are the large and small radii (in meters), respectively, and D is the density: 22 22 ).().(.. YX YDYXDXYXD ππ ππ − − = (Equation 4) Figure 3.6. The underlying grid value is assigned to the overlying sampling location for the TD.200 Population Density Variables Population density was estimated using 2001 census data from the dissemination area level. The total population in each polygonal area was assigned to its centroid, and the ‘Calculate Density (kernel)’ function of Spatial Analyst was run at search radii of 750, 1000, 1250, 1500, 2000 and 2500 meters. These steps are outlined in Figure 3.7. The ‘Get Grid Value’ extension was used again to assign values for the population density variables to each monitoring location. Dwelling density was originally calculated in the same way, but dropped from further analyses when population density clearly emerged as the more predictive variable category. TD.200 at site E7 = 71.7 Figure 3.7. Progression from a base polygon file to an intermediate centroids to a predictive variable density layer. Land Use Variables Commercial, residential, industrial, government and open land use variables were calculated from a GVRD land use map available from DMTI at radii of 300, 400, 500 and 750 meters. The original file included a sixth category entitled ‘waterbody’ which was appended to the ‘open’ category for the purposes of this work. The file was converted to a grid with 5 metre cell spacing and the ‘Neighbourhood Statistics’ function of Spatial Analyst was used to tally the number of hectares of each zoning type within the specified buffer sizes. Geographic Variables The X and Y variables were generated using an ‘Add XY Centroid’ ArcView extension23 then divided by 10,000 to reduce the UTM units to a workable range. A gridded digital elevation map (DEM) for the region was compiled using DMTI data for region 92G, parts 1-3 and 6-8, and elevation was assigned using the ‘Get Grid Value’ extension. Distance to the Pacific shore line was ascertained with a DMTI water map and the ‘Nearest Feature’ ArcView extension24. 23 Hare T (2003).Add XY Centroid 2.0. Avenue Script for ESRI ArcView GIS. United States Geological Survey. 24 Fox TJ (1998).Nearest Feature 1.0. Avenue script for ArcView GIS. 3.4.2 Model Building Model Generation Multiple linear regression models were constructed in S-PLUS25 for four response variables: average (spring and fall) NO2 concentration, logarithms of the average NO and NOX, and springtime filter absorbance. In all cases we built two separate models, first using the ‘traffic density’ and then the ‘road length’ variables. Traffic flow data are not always available, and it is important to understand how models developed with these variables differ from those that rely on universally available road length estimates. We also constructed ‘small’ and ‘large’ versions of all models, where the former included only those variables with buffer sizes of 500m or less and the latter included all variables. We started each model by calculating the correlation between the response variable and all the covariates. The results were then ranked and the most highly correlated variable was retained while all collinear variables (defined as a correlation coefficient greater than 0.6) within the same variable group were omitted from further consideration regardless of their ranking. For example, assume that RD.100 is the top-ranked variable. Table 3-4 demonstrates how other the 200-750 and 200.100 variables in the RD group would be automatically eliminated from the analyses based on their correlations with RD.100. Likewise, if RD1.1000 was the top variable, all other RD1 covariates would be omitted, with the exception of RD1.100. This process was continued down the variable ranking, resulting in a list of independent, potentially predictive variables which were entered into a bi-directional stepwise process to produce the final models. Table 3-4. Table of correlation coefficients for the RD1 group of variables R D 1. 10 0 R D 1. 20 0 R D 1. 30 0 R D 1. 50 0 R D 1. 75 0 R D 1. 10 00 R D 1. 20 0. 10 0 R D 1. 50 0. 10 0 R D 1. 50 0. 30 00 R D 1. 10 00 .5 00 R D 1. 10 00 .7 50 RD1.100 1.00 0.87 0.72 0.63 0.60 0.54 0.72 0.58 0.52 0.40 0.26 RD1.200 0.87 1.00 0.92 0.82 0.78 0.70 0.97 0.79 0.70 0.51 0.31 RD1.300 0.72 0.92 1.00 0.95 0.88 0.79 0.93 0.94 0.86 0.57 0.37 RD1.500 0.63 0.82 0.95 1.00 0.96 0.87 0.84 0.99 0.97 0.65 0.40 RD1.750 0.60 0.78 0.88 0.96 1.00 0.93 0.79 0.96 0.96 0.77 0.49 RD1.1000 0.54 0.70 0.79 0.87 0.93 1.00 0.71 0.86 0.86 0.94 0.78 RD1.200.100 0.72 0.97 0.93 0.84 0.79 0.71 1.00 0.82 0.72 0.51 0.32 RD1.500.100 0.58 0.79 0.94 0.99 0.96 0.86 0.82 1.00 0.98 0.65 0.40 RD1.500.300 0.52 0.70 0.86 0.97 0.96 0.86 0.72 0.98 1.00 0.66 0.40 RD1.1000.500 0.40 0.51 0.57 0.65 0.77 0.94 0.51 0.65 0.66 1.00 0.92 RD1.1000.750 0.26 0.31 0.37 0.40 0.49 0.78 0.32 0.40 0.40 0.92 1.00 Model Validation Two approaches were used to assess the validity of the NOX surfaces. First, annual average NO, NO2 and NOX concentration from 17 continuous samplers in the GVRD were compared to the annual average predicted by the respective surfaces at the same locations. Although such validation provides 25 Insightful S-PLUS (2003).6.2 for Windows. Exploratory data modeling and statistical analysis software. Insightful Corporation. valuable insight into the fit of the model, air quality monitoring sites are rarely located near to heavy traffic and their measurements do not reflect the true distribution NOX concentrations. To check model performance at more heavily impacted sites we also performed a leave-one-out validation where model coefficients are calculated using data from all but one of the sites, and then applied to predict the value of the response at the omitted site. These fitted values were compared to the measured values at each site, and a regression line was calculated. Because the absorbance of particulate matter is not routinely measured by the GVRD, the absorbance models were only validated by the leave-one-out approach. Regression Surfaces All regression surfaces grids were built in ArcView 3.2 using Spatial Analyst 2.0. We multiplied coefficients from the final linear regression models against each grid cell in their corresponding variable layers. Next, we summed all layers (including a constant layer for the intercept) yield the fitted surface. A simple example of these arithmetic grid operations is shown in Figure 3.8. In the cases of NO and NOX the fitted surfaces were exponentiated to yield the final pollution surface, and in all cases the fitted values were constrained within the range of measured values. All regression surfaces were produced at a resolution of 10 meters. =× 09.0 + = =−× 32.0 Figure 3.8. Demonstration of the arithmetic grid operations used to build pollution surfaces from linear regression models 4 Results 4.1 Site Selection The location allocation model was first implemented based GIS layers for land use, road networks, and regulatory air quality monitoring in the GVRD. Figure 4.1 shows the sampling locations identified by this model clustered along the freeways and highways, where the residential population is relatively sparse. To improve the distribution of sampling sites the location allocation model was weighted by population density and run a second time. Figure 4.2 shows the final sampling locations which were manually transferred to a city map, and readings taken by GPS show that 99 of the actual sites were within 25 meters of their modeled locations. The remaining site relocated 250 meters southeast of its modeled location due to zoning restrictions. Samplers were strapped to lampposts (n=70), hydro poles (n=25) and street signs (n=5). Figure 4.1. Results of the first location allocation modeling run Figure 4.2. Results of the second location allocation modeling run With the addition of 16 special interest locations, we sampled for nitrogen oxides at 116 sites and particulate matter at 25 sites. Appendix C is a detailed list of the sites including study ID, specific location, subjective indicators of car and truck traffic, location type (NOX only or NOX and PM2.5), and data completeness. 4.2 Sampling 4.2.1 Period Optimization The 2003 annual average concentrations of NO, NO2 and NOX at 16 air quality sites in the GVRD were compared to the average of those measured during the spring and fall campaigns. Figure 4.3 shows the strong one-to-one relationships between the annual and campaign averages for all NOX compounds. Figure 4.3. Shows the accuracy with which measurements during the spring and fall campaigns reflect 2003 annual averages at 16 air quality monitoring sites in the GVRD 4.2.2 Campaign Results Complete data for the spring and fall sampling campaigns can be found in Appendix D, and are summarized in Table 4-1. Due to sampler damage, loss and some analytic errors, complete NOX data are available for 105 of the 116 sites. Of the remaining 11 sites, spring data are available for 2, fall data are available for 7, and no data are available for the remaining 2. Although we attempted to analyse the spring campaign samples for SO2, we had limited confidence in the erratic results and this pollutant was excluded from further analysis. Similarly, due to inconsistency in microbalance measurements we chose to analyse filtered particulate matter based on absorbance rather than concentration, as the former is independent of filter weight. Table 4-1. Summary statistics for pollutant concentrations measured during the spring and fall 2003 sampling campaigns. Spring Mean (SD) Fall Mean (SD) Estimated Annual Mean (SD) Estimated Annual Inter- Quartile Range Distribution NO (ppb) 38.6 (24.2) 24.2 (15.9) 31.1 (18.7) 17.1 – 40.2 Log Normal NO2 (ppb) 19.6 (4.9) 12.9 (4.1) 16.0 (4.1) 13.3 – 18.1 Normal NOX (ppb) 57.8 (28.0) 32.2 (18.8) 47.0 (21.9) 31.3 – 58.1 Log Normal SO2* (ppb) 5.1 (16.8) - - - Log Normal Absorbance (m-1) 0.84 (0.47) - 0.84 (0.47) 0.55 – 1.07 Normal PM2.5* (µg/m3) 4.1 (2.0) - - - Normal *These measurements were dropped from further analysis due to lack of confidence in the values obtained. Although the two campaigns were designed to capture an estimate of the annual mean, the spring concentrations were significantly higher than the fall concentrations for NO (paired t=8.09, p=0), NO2 (paired t=16.65, p=0) and NOX (paired t=10.39, p=0). This is surprising given the historical trends illustrated in Figure 3.1, and it suggests that our values may underestimate the true annual mean due to unusually low concentrations during the fall sampling period. 4.2.3 Quality Control Co-located Samplers Ogawa samplers were co-located with 14 continuous NO/NO2 monitors (chemiluminescence) in the GVRD during both campaigns to gauge the accuracy of our sampling equipment. Logged data from the GVRD monitors were averaged over the two-week period of co-location and compared to concentrations derived from the Ogawa monitors. In general, the two methods agreed very well for the spring campaign (Figure 4.4), but less well in the fall (Figure 4.5). Note that in the fall one sampler was lost from the downtown location that yielded the second highest NO, NO2 and NOX concentrations during the spring campaign. UBC vs. GVRD Concentrations for Oxides of Nitrogen (spring) y = 0.9389x - 2.3867 R2 = 0.9694 y = 1.0886x - 2.629 R2 = 0.8543 y = 0.8816x - 1.3874 R2 = 0.9343 0 10 20 30 40 50 60 0 10 20 30 40 50 60 70 UBC Concentrations (ppb) G VR D C on ce nt ra tio ns (p pb ) NOX NO2 NO Figure 4.4. Relationship between oxides of nitrogen measured by Ogawa passive samplers (UBC) and continuous samplers (GVRD) based on two-week averages during the spring sampling campaign. UBC vs. GVRD Concentrations for Oxides of Nitrogen (fall) y = 1.0201x + 0.1424 R2 = 0.8266 y = 1.2049x + 2.8336 R2 = 0.7332 y = 0.8674x + 0.7123 R2 = 0.5914 0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35 UBC Concentrations (ppb) G VR D C on ce nt ra tio ns (p pb ) NOX NO2 NO Figure 4.5.Relationship between UBC and GVRD measurements for the fall campaign. A single Harvard impactor was co-located with a continuously logging PM2.5 Tapered Element Oscillating Microbalance (TEOM) monitor at Burnaby South high school in the GVRD. Particle samples were collected over six weeks, giving only six points of comparison between the two methods, which yielded poor agreement as seen in Figure 4.6. Unfortunately these UBC measurements were subject to the same analytic uncertainties that lead us to exclude PM2.5 concentration from the regression analyses, so the validity of their apparent relationship with the GVRD measurements is suspect. Without GVRD filters from which to measure absorbance, there is limited evidence by which to access the accuracy of the impactors. UBC vs. GVRD Concentrations for Particulate Matter y = 0.2371x + 2.7975 R2 = 0.1949 2.5 3.0 3.5 4.0 4.5 5.0 0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 PM2.5 from UBC Harvard Impactor (µg/m3) PM 2. 5 f ro m G VR D C on tin uo us S am pl er ( µ g/ m 3 ) Figure 4.6. Relationship between PM2.5 measured with UBC Harvard impactor compared to GVRD continuous monitors Duplicate Samples We collected duplicate samples at 18 and 20 sites during the spring and fall campaigns, respectively. Once again the spring measurements were more reliable with Pearson’s correlation coefficients of 0.91, 0.98 and 0.99 for NO, NO2 and NOX, respectively. The equivalent fall correlations were 0.80, 0.93 and 0.96. In all cases the slopes were close to 1, suggesting no systematic under- or over-measurement by the Ogawa samplers. Due to our limited number of sampling apparatuses for particulate matter, duplicate samples were not collected at any of the locations. Field Blanks Twenty-six Ogawa field blanks were distributed between the 14 co-location sites during both the spring and fall campaigns. The mean nitrite (µg/ml) concentrations in the NOX extract from blanks were 0.032 in spring and 0.038 in fall, as compared to 2.92 and 1.95, respectively, in the samples. Filter field blanks were deployed at Burnaby South high school during each particle sampling period. They had an average weight gain of 0.018 mg as compared to 0.039 in the samples, which resulted in four samples below the limit of detection. The blank filters had an average reflectance of 101.8% compared to the reference filter (100 by definition) and 90.8% in the sampled filters. 4.3 Linear Regression Modeling 4.3.1 Final Models The final models for LogNO, NO2, LogNOX and Absorbance are presented in Table 4-2 with their coefficients, associated p-values, and R-squared values. The coefficients for all variables in all models are significantly different from zero at the 95% level. Although we can obtain slightly better R-squared values (i.e. we can explain more variability in the response) by including additional variables, the models presented here are parsimonious in that each variable contributes at least 3% to the R-squared. This criterion ensures that models remain relatively simple to build and interpret. The validation R2 shown in Table 4-2 presents results from a linear regression on the measured values against the bootstrap fitted values. Table 4-2. Summary of final linear regression models used to build pollution surfaces Response N Traffic Variable Type Coefficients Variables Model R 2 Validation R2s Length 5.5818 2.4354 -0.0039 0.0070 0.7316 -0.0484 -0.0454 Intercept RD2.100 ELEV EAP.2500 X.10000 RD1.200 TRK.750 0.62 0.49 (GVRD*) 0.57 (LOO†) LogNO 114 Density 116.3505 0.1323 0.0007 -0.0024 -0.1286 -0.1959 Intercept TD.1000 AD.100 ELEV X.10000 Y.10000 0.57 0.65 (GVRD) 0.51 (LOO) Length 40.7918 0.0847 -0.5534 4.0414 -0.0198 14.7619 0.1144 Intercept EAP.2500 X.10000 RD2.200 ELEV RD1.100 COMM.750 0.53 0.69 (GVRD) 0.47 (LOO) NO2 114 Density 41.1287 0.6030 0.0679 0.0021 0.1605 -0.5637 -0.0173 0.1158 Intercept TD.1000 EAP.2500 AD.100 TD.200 X.10000 ELEV COMM.750 0.60 0.79 (GVRD) 0.54 (LOO) Length 5.8200 1.7883 0.5840 -0.0439 -0.0029 0.0067 -0.0349 Intercept RD2.100 RD1.200 X.10000 ELEV EAP.2500 TRK.750 0.63 0.57 (GVRD) 0.58 (LOO) LogNOX 114 Density 86.8320 0.1101 0.0005 -0.1087 -0.1427 -0.0017 Intercept TD.1000 AD.100 X.10000 Y.10000 ELEV 0.59 0.71 (GVRD) 0.52 (LOO) Length 1.1356 1.1470 -0.1760 -0.0047 -0.1561 Intercept RD2.100 DIST.RD1 TRK.500 ELEV 0.49 0.30 (LOO) Absorbance 25 Density 0.8450 0.2825 -0.0021 -0.0985 -0.0033 Intercept TD.1000 AD.1000 TD.300 ELEV 0.57 0.35 (LOO) *GVRD = R2 for the regression on annual averages estimated from the modeled surfaces vs. those measured at 17 GVRD monitoring sites †LOO = R2 for the regression line on annual averages estimated from leave-one-out models compared to those measured during two sampling campaigns at 114 sites. 4.3.2 Trials and Errors Before deciding on the final methodologies discussed in section 3.4 and models presented above, we took multiple approaches to variable generation and model building, a few of which deserve special attention here. Vector vs. Grid Variables One key concept in GIS is the distinction between vector and grid data. Point, polygon and line files are all stored in vector form, which is efficient because it requires only the locations of vertices and feature attributes. However, what is gained in efficiency is lost in spatial information, which makes it impossible to perform the arithmetic operations shown in Figure 3.8 with vector layers. On the other hand, when vector data is converted to grid it is associated with a cost in accuracy, as was demonstrated in Figure 3.5. Because of the vector-to-grid operations transformed to generate our road length and traffic density variables, we can be sure of inaccuracies that are likely to increase with decreasing size of the variable radii. To address this problem we used vector operations to generate accurate values for both variable types (computationally possible for 114 points, but not for a whole city) to compare them to the less accurate values, and to assess how this inaccuracy affected the final models. Figure 4.7 shows how the grid method produces inaccurate but well-correlated variable values for AD.100, yet accurate values for AD.1000. When the accurate, vector-derived values for the road length and traffic density variables were used in the final NO, NO2 and NOX models we observed greater improvement in the fit of the density models. When these values were used for the Absorbance model the fit of the traffic density model was improved by ~40%, but no difference was seen in the road length model. Considering the already problematic nature of the Absorbance models (as will be seen in section 5) this is neither surprising nor worrying. AD.100 Values Generated by Vector and Grid Operations y = 0.2751x + 11.797 R2 = 0.7853 0 200 400 600 800 1000 1200 1400 0 500 1000 1500 2000 2500 3000 3500 4000 Vector Estimate G rid E st im at e AD.1000 Values Generated by Vector and Grid Operations y = 0.8844x - 0.5699 R2 = 0.9712 0 100 200 300 400 500 600 0 100 200 300 400 500 600 Vector Estimate G rid E st im at e Figure 4.7. Grid versus vector estimates for the variables AD.100 and AD.1000, demonstrating greater accuracy of the grid method with greater variable radii. Small Buffer vs. Large Buffer To examine the possibility that variables with small radii produce smoother, more well-fitted surfaces than those with large radii, we conducted our linear regression modeling once with all variables, and then again with those variables with radii less than or equal to 500 meters. We found only minimal differences between the two sets of models and the regression surfaces they produced. Annuli Variables Annulus or ‘doughnut’ variables are included in this type of modeling to more accurately characterize those sites that are not surrounded by a uniform road network. For instance, a rural site may have very few roads within a 200 m radius, but may still be surrounded by a dense network at a 500 m radius, as shown in Figure 4.8. When averaged over the area of the entire buffer the density of the network is underestimated, but when averaged over an annulus it is accurate. Figure 4.8. Example of a site where the 500-200 annulus buffer contains important information about the pollution received by the site. There are no roads within the 200 m radius, but several within the 500 m radius. 4.4 Regression Surfaces Figure 4.9 through Figure 4.16 on the following pages show the final pollution surfaces based on the regression results presented in Table 4-2. In general, those models built with the road length variables give a clear image of the contributing road network, and they assume that traffic is evenly distributed along the available roads in each category. From experience we know that traffic behaves quite differently, which is what the models based on density are intended to capture. In the NO, NO2 and NOX models we can see that pollution concentrations are elevated hot spots around the district; many of these are at the entrances to bridges and tunnels. Although the traffic density models tend to be less well-fitted to the data than the road length versions, this increased error could be attributable to error in the TransLink EMME/2 model rather than in the fitted coefficients. Either way, it is interesting that both model types provide essentially equivalent fit statistics and similar regression surfaces. Note that, since these surfaces are based upon the regression models described in Table 4.2, they are estimates of the spatial distribution of these traffic related air pollutants in this region. The models clearly indicate that a substantial fraction of the variability in the measured concentrations of these air pollutants remain unexplained. As discussed in the Discussion section below, the concentrations predicted by these models and depicted in the surfaces provide a better indication of the average concentrations (for example, those measured by the GVRD monitoring network) than of higher concentrations experienced in the region. Accordingly, these surfaces, while depicting the general spatial patterns of air pollution, reflect relative differences within the region but should be used cautiously for specific comparisons or to quantitatively predict concentrations at specific locations. Fi gu re 4 .9 . E xp on en tia te d re su lt of th e fin al re gr es si on s ur fa ce fo r L og (N O ) b ui lt w ith ro ad le ng th tr af fic v ar ia bl es Fi gu re 4 .1 0. E xp on en tia te d re su lt of th e fin al re gr es si on s ur fa ce fo r L og (N O ) b ui lt w ith tr af fic d en si ty v ar ia bl es Fi gu re 4 .1 1. R es ul t o f t he fi na l r eg re ss io n su rfa ce fo r N O 2 b ui lt w ith ro ad le ng th tr af fic v ar ia bl es Fi gu re 4 .1 2. R es ul t o f t he fi na l r eg re ss io n su rfa ce fo r N O 2 b ui lt w ith tr af fic d en si ty v ar ia bl es Fi gu re 4 .1 3. E xp on en tia te d re su lt of th e fin al re gr es si on s ur fa ce fo r N O X bu ilt w ith ro ad le ng th tr af fic v ar ia bl es Fi gu re 4 .1 4. E xp on en tia te d re su lt of th e fin al re gr es si on s ur fa ce fo r N O X bu ilt w ith tr af fic d en si ty v ar ia bl es Fi gu re 4 .1 5. R es ul t o f t he fi na l r eg re ss io n su rfa ce fo r A bs or ba nc e bu ilt w ith ro ad le ng th tr af fic v ar ia bl es Fi gu re 4 .1 6. R es ul t o f t he fi na l r eg re ss io n su rfa ce fo r A bs or ba nc e bu ilt w ith tr af fic d en si ty v ar ia bl es 5 Discussion The atypical meteorology and topography of the Fraser Valley play an important role in pollution generation and dispersion within the GVRD. Prevailing westerly winds produce long-range pollutant transport up the valley, rainfall scrubs particulate matter from the ambient air, and elevation affects the dynamic relationship between nitrogen oxides and ground-level ozone. In combination these factors differentiate the GVRD from other regions to which land use regression techniques have been successfully applied. Table 5-1 shows how the fitted GVRD models compare (by multiple R2) to those fitted elsewhere in Canada and Europe. Table 5-1. Comparison of land use regression model fit across different study locations Study Location Investigator, Year Range NOX R2 Range PM R2 Vancouver, BC Brauer, 2003 0.53 – 0.63 0.49 – 0.57 Toronto, ON Jerrett, in press 0.65 - Montreal, QC Gilbert, 200426 0.39 - Netherlands, Stockholm, Munich Hoek, 200127 0.62 – 0.85 0.66 – 0.81 El Paso Gonzales, 200528 0.81 - Although our models do not describe as much variability as those developed for some other cities, they perform relatively well given how the GVRD differs from other settings, and land-use regression provides a feasible approach for estimating individual level exposure to pollution in this region. 26 Gilbert NL, Goldberg MS, Beckerman B, Jerrett M, Brook JR. Predicting spatial variability of ambient nitrogen dioxide in Montréal, Canada, with a land use regression model. Epidemiology: Volume 15(4) July 2004 p S200 27 Hoek G, Meliefste K, Brauer M, van Vliet P, Brunekreef B, Fischer P. et al. 2001. Risk assessment of exposure to traffic- related air pollution for the development of inhalant allergy, asthma and other chronic respiratory conditions in children (TRAPCA). Final Report. IRAS, University, Utrecht, April 2001 28 Gonzales M, Qualls C, Hudgens E, Neas L. Characterization of a spatial gradient of nitrogen dioxide across a United States-Mexico border city during winter. Sci Total Environ. 2005 Jan 20;337(1-3):163-73. Table 4-1 gives some information about the distribution of pollutant concentrations from the UBC sampling locations, and Figure 5.1 shows how the variability in these measurements compares to that in NOX measurements from 16 GVRD air quality stations during the same periods. This relatively dense regulatory network fails to capture the true variability in concentration of traffic-related pollutants and our models explain more than GVRD monitoring about the intra-urban distribution of these pollutants, despite having lesser R2 values than others in Table 5-1. As such, these models can provide epidemiologists and risk assessors with a valuable tool for examining relationships between health and individual level exposure estimates in large populations. Figure 5.1. Box plots showing the distribution in NO2 and NO concentrations from 16 GVRD air quality monitoring sites and 114 UBC sampling sites during the two campaigns. Within the City of Vancouver, DMTI provides a block-face population file, which enumerates the number of residents living on every side of every city block. The resolution of these data is much finer than that of the Dissemination Area population file we used to generate population density variables, so we used this file to examine the estimated distribution of exposure within the 2001 Vancouver population. Figures 5.2 through 5.4 show the histograms for models developed with the traffic density variables, though results are similar for the road length models. Figure 5.2. Distribution of estimated NO exposures for 1.9 million residents of the City of Vancouver Figure 5.3. Distribution of estimated NO2 exposures for 1.9 million residents of the City of Vancouver Figure 5.4. Distribution of estimated NOX exposures for 1.9 million residents of the City of Vancouver Figure 5.5. Distribution of estimated PM2.5 absorbance for 1.9 million residents of the City of Vancouver The unusual pattern seen in Figure 5.5 is a direct result of the unusual buffering effect shown in Figure 4.16, where areas near to major roads actually have lower absorbance estimates than those slightly removed from the roads. This may be driven by the fact that only three sites in our particulate matter subset were more than 300 meters from a major road or freeway so that the background levels were not adequately characterized. Because absorbance was strongly associated with traffic density at 1000 and 300 meters, residual collinearity in the variables may have caused this, even though the condition for inclusion (Pearson coefficient less than 0.6) was met. We experimented by eliminating a) up to three extreme values from the data set and b) either the TD.1000 or TD.300 variable, but the first approach did not change the form of the fitted equation and the second reduced the R2 value to below 0.3. Although this anomaly is somewhat remedied by using road length instead of traffic density variables, Figure 4.15 lacks the smoothness of the other surfaces. A “leave-one-out” validation further demonstrates that the absorbance models based on 25 sites are not as robust as those for the NOX compounds based on 114 sites, although previous studies in Europe have derived successful models with 40+ locations29. As mentioned in Section 3.1.3, some substitutions to the randomly selected particulate matter subset were made to ensure that the upper end of the distribution was captured, and we may have oversampled high traffic sites for particulate matter. In addition, these sampling locations were identified to satisfy an NO2 demand surface and, given that the relationship between NO2 and absorbance is weak (r = 0.5), they may not be optimal for particulate matter monitoring. In the cases of NO, NO2 and NOX the road length and traffic density variables produced similar exposure surfaces, suggesting that land use regression can be successfully applied in areas where traffic counts and/or traffic flow models are unavailable. For logNO and logNOX the road length surfaces have higher R2 values than their traffic density equivalents, but validation against measured GVRD values yielded considerably the highest R2 values for the traffic density surfaces in all cases (see Figure 5.6). This suggests that the traffic density models perform better in the lower parts of the concentration distributions where, according to Figure 5.1, most of the GVRD values lie. One likely explanation is described in Section 4.3.2. Figure 4.7 shows how the relationship between vector and grid estimates for automobile density in a 100 meter radius is stronger at lower values. Because most regulatory monitors are deliberately distanced from high traffic areas, the GVRD traffic density values fall within 29 Brauer M, Hoek G, Van Vliet P, Meliefste K, Fischer P, Gehring U, Heinrich J, Cyrys J, Bellander T, Lewne M, Brunekreef B. Prediction of long term average particulate air pollution concentrations by traffic indicators for epidemiological studies. Epidemiology, 2003;14:228–239. the range where the surfaces are most accurately estimated. However, this goodness of fit is not reflected in the model R2 because the error in AD.100 increases with the traffic density. Furthermore, these variables are based on output from the imperfect EMME/2 traffic flow model which, according to our previous analysis, yielded an R2 of 0.72 when modeled estimates were compared to measured values at 20 intersections.1 Continued validation and refinement of the traffic flow model will provide better insight into strategies for improving the accuracy of traffic density variables. Figure 5.6. Traffic density and road length surface estimates validated against annual NOX concentrations at 16 monitoring sites in the GVRD. There are a few other ways in which the fit of our models could be improved. First, incorporation of prevailing wind directions would suggest that oblongs based on wind vectors rather than circles would provide more appropriate buffer zones for traffic-related study variables. The Border Air Quality Study30 team is currently exploring the addition of street canyon descriptions and wind information in these models. On the same theme, the X and Y coordinates that appear consistently throughout these models are likely to be surrogates for some other (yet undefined) variables that are important in pollution dispersion. The inclusion of wind data might eliminate the need to retain X and Y, or other variables may still need to be identified and defined. We also examined the possibility that the X variable might be acting as a proxy for distance to coastal shoreline, but when we included this variable in the place of X the R2 for all models was decreased. Although all models are relatively well-fitted and most of the surfaces are intuitively acceptable, epidemiologists who plan to use these results for research purposes must be aware of their limitations. While these estimates may be useful for characterizing the variability in residential exposure to ambient traffic pollution, they do not account for the many sources of variability in personal exposure, resulting from mobility, activity and occupational factors. Given this limitations, these surfaces are most useful in providing estimates of exposure variability in large cohort studies utilizing administrative health data. 30 www.cher.ubc.ca/baqs.htm Measurement and modeling of traffic-related air pollution in the British Columbia Lower Mainland for use in health risk assessment and epidemiological analysis. Interim Report Michael Brauer and Sarah Henderson School of Occupational and Environmental Hygiene The University of British Columbia 2206 East Mall Vancouver, BC V6T 1Z3 brauer@interchange.ubc.ca tel 604 822 9585 fax 604 822 9588 March 28, 2003 Overview This interim report presents the general study methodology and an overview of the progress completed to date, including the general land use regression modeling of air pollution based on existing monitoring data, the location-allocation procedure used to identify sampling sites and the performance of the first round of sampling using passive monitors. Background There has been increasing attention devoted to the direct health risks associated with air pollution from local traffic sources. This interest stems in part from the fact that mobile sources are the major contributor to emissions in many urban areas. Further, recent research has demonstrated associations between traffic-related air pollution and a range of health outcomes including increased mortality, cancer incidence, asthma and allergy prevalence, lung function, chronic respiratory symptoms, and adverse birth outcomes. Geographic modeling approaches allow for individual exposures to be modeled based upon regression of measured air pollutant concentrations against surrogate variables in a Geographic Information System (GIS) framework. The specific use of traffic-related surrogate variables allows these methods to develop exposure estimates for epidemiological studies and/or risk assessment that are specific to traffic-related pollutants. While limited information from North American studies suggests that traffic-related within-city spatial variability that is not reflected by ambient monitoring networks exists, the geographic modeling approach has not yet been applied in the US or Canada. The outcome of this project will be a model by which concentrations of traffic-related air pollutants can be predicted at any location with in the Lower Mainland. This approach is a simpler and much less expensive alternative to dispersion modelling, and will allow us to generate a continuous surface of estimated concentrations for the entire region to be used in risk assessment and epidemiological analyses. Objectives 1. To use the land-use regression and the GIS-based locations/allocation approach to identify 100 target measurement sites. 2. To conduct (roadside and background) monitoring of NO2 using passive samplers at the multiple measurement locations within the Lower Mainland identified by location/allocation models. 3. For these measurement sites, to collect geographic information on traffic counts, population density, altitude, land use, etc. 4. To model the relationship between the geographic variables and the long-term average concentrations of NO2. This would enable the use of geographic data to estimate measured ambient concentrations for future epidemiological studies and risk assessment. 1 Methods The overall goal is to replicate the current measurement and modelling program being conducted in Toronto, to extend the approach to develop long-term average air pollution exposure estimates for the BC Lower Mainland. This approach extends our pilot work conducted in summer 2002 to additional locations, while limiting the focus to a single pollutant, NO2, as an indicator of traffic emissions. Previous work in Europe has demonstrated that NO2 measurements are highly correlated with PM measurements and are valid predictors for traffic-related pollution models. Measurements will be collected at 100 sites for 2-week periods. This measurement procedure will be repeated during 3 additional periods during a total one-year period, to allow for the determination of long-term average concentrations. These discountinuous measurements will be adjusted by co-locationg passive samplers at 5 continuous measuremement sites (GVRD network sites) throughout the entire sampling period to allow for the correction of the discontinuous measurements for temporal trends (ex. due to meteorology). 1. We have obtained digital street maps and land-use data that has been combined with GVRD monitoring network data to estimate the pollution surface. This estimated surface will be used to optimally locate 100 air pollution monitoring sites within the region to maximize the amount of explained variance in the measured concentrations. 2. At these 100 target monitoring sites we have located passive monitors for a 2-week period in March 2003. This will replicate the protocol employed in Toronto and Hamilton and will provide an extremely dense monitoring network to support the regression modelling. We have used Ogawa passive samplers to measure SO2, NO2 as well as NO, the primary NOx emission, which may serve as a superior indicator of traffic exhaust. At a subset of locations we will conduct PM measurements using the portable battery- operated monitors for PM2.5 and elemental carbon. 3. At 5 GVRD monitoring locations we have deployed deploy identical passive monitors to calibrate the passive samplers to the GVRD network data. PM2.5 samplers have been co- located at two GVRD PM2.5 monitoring sites to collect a continuous record of 2-week pollution averages. These measurements will be used to calibrate the discontinuous measurements for temporal variation. Ideally, we would have preferred to conduct both NO2 and particle monitoring at all locations, although the costs of such an approach would be considerably higher (approximately 250,000 for the entire project period). The NO2–only approach described above will be linked with the data we have already gathered in our pilot study and with limited PM monitoring at a subset of locations (approximately 25) to determine the relationship between particles and NO/NO2 in the framework of the geographic modeling. 2 The general approach is illustrated below: Location/Allocation analysis to identify measurement sites based on land-use regression model approach average Measurements at 100 selected locations GIS variables Traffic intensity Road distance Population density Altitude Land use Measurements at 100 selected locations Air quality (long-term averages) NO2 (PM components at a subset of ~ 25 locations) Repeat for additional season and calibrate to generate long-term Application Apply regression model to GIS data to predict long-term average air pollution concentrations for risk assessment and epidemiological studies Regression Model Predicting air concentrations with GIS variables for measurement sites 3 Preliminary results Land use regression This section describes the development of a general land use regression model that was used to be used in location-allocation analysis to identify sampling sites. The model development and location-allocation was conducted under the direction of Dr. Michael Jerret of the School of Geography and Geology and McMaster Institute of Environment and Health, McMaster University using DMTI land use and transportation data and GVRD monitoring network data for the 2001 calendar year (correct) that was pre-processed at UBC. This group also conducted the Toronto and Hamilton studies. This multivariate regression model was developed to identify which general land uses may be best suited to predicting NO2 concentrations in the metropolitan Vancouver area based upon existing GVRD network monitoring data. Initially, each of 13 different variables (Table 1) were computed through individual bivariate regression models using annual average NO2 (for 2001) concentrations from all 17 GVRD network sampling locations where NO2 is monitored: (T1, T2, T4, T6, T9, T13, T14, T15, T17, T18, T20, T26, T27, T30, T31, T32 and T33) as the dependent variable. Land use variables were calculated in hectares and road lengths in kilometres. Land use parameters were measured under a two circular buffers that extends from each monitoring location out to a radius of 200 m. Road length parameters were measured under two separate buffers. A first circular buffer extends from each monitoring point out to a radius of 50 m. A second buffer is in the shape of an annulus (donut), with the inner edge at a radius of 50 m and the outer at 200 m. All area and length calculations were performed using ArcView 3.2 software’s Spatial Analyst extension (ESRI Corp., Redlands, CA). Table 1: Land use regression model variable names and descriptions Land Variables Buffered Around NO2 Monitoring Sites Transportation Variables Buffered as Distances around NO2 Monitoring Sites name description name description Open_100 Open space/park land (100m) Rd2_50 Major roads & primary arterials (0-50m) Res_100 Residential (100m) Rd2_200 Major Roads & primary arterials (50-200m) Gov_100 Government and institutional (100m) Rd_Dist Distance to Expressways Open_200 Open space/park land (200m) Rd3_50 Local Roads (0-50m) Res_200 Residential (200m) Rd3_200 Local Roads (50-200m) Gov_200 Government and institutional (200m) Rd_Dens Density of Road Network within 500m Ind_200 Resource and industrial (200m) The variable that represented the presence of roads located within a 50-200m buffer (Rd3_200) was most strongly associated with the measured NO2 concentrations (t = 2.935, p= 0.010). Next, 4 the Rd3_200 variable was paired with each of the other variables other variable in a series of trivariate regression models again using annual average NO2 as the dependent variable. This series of analyses identified the pairing of Rd3_200 with Rd_Dens as the most successful combination for predicting ambient NO2 concentrations. Model results (manual forward stepwise ordinary least-squares regression) are presented in Tables 2 and 3. Table 2: Descriptive statistics for land use regression variables Variable Mean Std. Deviation N Mean No2 15.837059 4.2153822 17 Rd3_200 .984706 .6063762 17 Rd_Dens 1.405235 1.7759631 17 Table 3: Land use regression results for model predicting NO2 concentrations Variable Coefficient Std. Error t-value Prob. > t VIF Constant 10.678 1.426 7.488 0.000 Rd3_200 3.631 1.217 2.984 0.010 1.031 Rd_Dens 1.127 .416 2.712 0.017 1.031 Adj. R-sq 52.4% As demonstrated in Table 3, density of roads and the presence of secondary roads located between 50 and 200m of a given site are significant land-use variables for predicting NO2 concentrations at the 17 GVRD monitoring sites. The coefficient for the first variable suggests that each standard deviation increase (0.606 km) in secondary roads located within 50 – 200m of a given site represents an increase of 2.202 ppb in NO2 exposure (0.606 * 3.631 = 2.202 ppb). Similarly, each standard deviation increase (1.776 Ha) in the second variable represents an increase of 2.002 ppb in NO2 exposure (1.776 * 1.127 = 2.002 ppb). Figure 1 is a scatter plot of measured vs predicted values and indicates that there are no significant outliers among the predicted values. Unstandardized Predicted Value 262422201816141210 M ea n N O 2 30 20 10 0 Figure 1: Scatter plot of measured and predicted values from land-use regression model. 5 Creation of the demand surface This model was then used to generate a predicted NO2 pollution surface for the entire region based on the transportation variables found to be predictive in the model via a focal sum statistic. The focal sum statistic computes the area or the length parameters of the land use or transportation criteria within the same neighbourhood used for measuring the independent variables in the regression analysis. This was computed for every 5 m grid cell in the study region. The 5 m resolution was chosen to allow a reasonable approximation of the vector dataset into raster format. This predicted pollution surface is shown in Figure 2. Next, the model described above was used in the location-allocation procedure to identify the core 100 target sampling locations for the deployment of passive samplers. The procedure is described in more detail elsewhere (Kanaroglou et al, 2003). Briefly, the predicted NO2 surface identified from the existing monitoring network was supplied to an algorithm that identified the optimal locations for 100 samplers following the basic criteria that more monitors should be located where the pollution surface is expected to exhibit higher spatial variability. Spatial variability is determined by the semi-variance at a large number of locations in the study area, based on the assumption that air pollution is spatially continuous and that the distance between two separate sampling locations correlates with their measurement of the pollutant. The measurement of the variation between predicted concentrations at two locations at a defined distance apart (in this case 300m) is the semi-variance. Further, given that the correlation of a random variable between two locations is inversely proportional to the semi- variance of the variable at those locations, there is no benefit from further monitoring stations if the distance between stations and their neighbours is reasonably small compared to the semi-variances within those same neighbours. A grid was imposed on the study area, and the semi-variance determined at all the nodes of the grid. Location-allocation for identification of target sampling sites To compute the actual location of target sampling sites the locational demand is first calculated via a grid at the 500 m resolution, resulting in a lattice of “candidate locations”. To select the target locations ARC/INFO software and the Attendance Maximizing Problem (AMP) algorithm was used. To describe the AMP, we assume that the monitoring sites are stadiums. Each of the candidate locations contains the number of people equal to the “demand” at that location. In this analogy, the people travel to the closest stadium, but the probability of them attending a given event decreases linearly as the distance from the stadium increases. The essence of AMP is to maximize the total attendance of all stadiums (i.e., monitoring stations). Analogously, we seek to maximize attendance of pollution variability at our 100 sampling stations. The remaining part of the computation is performed using the location-allocation function of the ArcPlot module of ArcGIS. Specifically, the demand locations were set as the locations of the GVRD monitoring stations that are assigned to the closest 500 x 500 m grid cell centroid and the candidates for the 100 sampling locations. Each 500 x 500 m grid cell represents the sum of the 5 m resolution demand surface that intersects the 500 x 500 m cell area. Figures 3 and 4 depict the sampler locations identified by this procedure (Figures 3 and 4). To arrive at the final set of target sampling locations, the demand surface was then weighted for population density and supplied to the location-allocation procedure (Figure 5). Briefly, a census tract population surface was aggregated at a 1000 m resolution to produce a surface with each grid cell containing 6 the total population inside the cell boundary. The weighting was implemented as a bivariate linear rescaling of the pollution semi-variance. Specific core target sampler locations are listed in Appendix 1, along with additional sampling locations that were included based upon specific local interest. Sampler deployment and descriptive characteristics A total of 132 samplers, including 18 duplicate samplers, were deployed for a 2-week during in March 2003. This total represents the 100 core sampling sites identified in the location-allocation procedure as well as an additional 16 samples that were located in areas of specific local interest: 5 (plus 1 duplicate) in Abbotsford, 4 in Surrey along a proposed transit corridor and at the home of a resident with air quality complaints, 6 in Vancouver along high traffic corridors (Knight Street, Cambie Street, SW Marine Drive, Granville Street, Clark Drive) (Figures 6 and 7). In addition, 14 samplers and 20 field blanks were located at 14 GVRD monitoring network sites (T27, T33, T30, T20, T32, T9, T13, T15, T18, T4, T6, T1, T2, T17) for comparison with monitoring network data. Of this total of 146 samplers that were deployed, 7 (5%) were not successfully retrieved due to damage/theft (3 samples) or missing sampler covers (4 samples). When samplers were deployed locations were subjectively surveyed for traffic levels by counting the number of lanes, identifying if measurements were made at an intersection, and ranking the adjacent road for total and truck traffic according to the following classifications. Percentages are of all sampling sites are given in parentheses. Lanes: 1 = A small or residential road with no defined lanes (i.e. no painted lines) (67/132 = 51%) 2 = 1 lane in each direction with traffic divided by a central, painted line (32/132 = 24%) 3 = 1 lane in each direction with a turning lane. (0/132 = 0%) 4 = 2 lanes in each direction (27/132 = 20.5%) 5 = 2 lanes in each direction with a turning lane (0/132 = 0%) 6 = 3 lanes in each direction (6/132 = 4.5%) Intersection: 1 = Sampler was located at a 4 (or more) way intersection (21/132 = 15.9%) 0 = Sampler was located on a straight stretch or road, or at a 3-way intersection (111/132 = 84.1%) Traffic: 1 = One or two cars passed us while we were on-site. (36/132 = 27.3%) 2 = Several cars passed us while we were on-site, but not a steady stream of traffic.(44/132 = 33.3%) 3 = A steady stream of traffic passed us while we were on-site (23/132 = 17.4%) 4 = Heavy traffic (only found on major roads) (25/132 = 19%) 5 = Very heavy traffic (freeway etc.) (4/132 = 3%) Trucks: 0 = We didn't see any trucks (67/132 = 51%) 1 = We saw one truck (6/132 = 20%) 2 = We saw a couple of trucks (15/132 = 11.3%) 7 3 = Light truck traffic (15/132 = 11.3%) 4 = Steady truck traffic (6/132 = 4.4%) 5 = Heavy truck traffic (3/132 = 2%) A complete list of all sampling locations is provided in Appendix 1. Additional work Samples are currently being extracted and will be analyzed by ion chromatography for NO, NO2 and SO2 concentrations. In addition, particle samples are currently being collected at a subset of 25 locations (Appendix 2) using battery-operated pumps and Harvard Impactors with a PM2.5 size cut. Particle samples are collected at 5 locations each week and at two GVRD sampling sites T2 – Kitsilano and T18 – South Burnaby. The samplers are then rotated to a new set of 5 sites. The samplers located at the GVRD sampling sites will be used to adjust the discontinuous measurements at the 25 locations for temporal variability, as conducted in our pilot study. All particle samples will be analyzed for PM2.5 mass and for filter optical absorbance, which we have demonstrated previously to be highly correlated with elemental carbon. Additionally, all passive and particle sampling will be repeated in September 2003 to produce more stable estimates of long-term average concentrations. These mean of the two measurements will then be used for geographic modeling to construct predicted air pollution (NO, NO2, SO2, PM2.5, elemental carbon) surfaces for the entire region. References Kanaroglou PS, Jerrett M, Morrison J, Beckerman B, Arain A, Gilbert NL, Brook JR. Establishing an air pollution monitoring network for intra-urban population exposure assessment: a location-allocation approach. Submitted to Science of the Total Environment, 2003. 8 Figure 2. Predicted NO2 surface based on multivariate regression model for 17 GVRD monitoring network sites. 9 Figure 3. Target sampler locations predicted from location-allocation model, presented by land use categories. 10 Figure 4. Target sampler locations predicted from location-allocation model, presented by predicted NO2 surface shown in Figure 2. 11 Figure 5. Target sampler locations predicted from location-allocation model, weighted by population density. 12 Figure 6. Regional map indicating actual locations of all passive samplers that were deployed (not including samplers co- located at GVRD monitoring network sites. 13 Figure 7. Example map of City of Vancouver indicating actual locations of all passive samplers that were deployed (not including samplers co-located at GVRD monitoring network sites. 14 Appendix 1. List of sampling locations and characteristics. Additional (non-core) sampling locations are indicated in bold text and samplers that were not recovered successfully are highlighted in yellow. Duplicates are indicated by a 1 in the Dup column. Dup Municipality ID Traffic Lanes Intersection Trucks Location Notes Notes 0 Abbotsford 39 1 1 0 0 North end of bend in Marble Hill Dr 0 Abbotsford 36 2 1 0 0 West side of Princess St just North of Conrad Ave Sampler Missing 0 Abbotsford 38 4 2 1 5 North-West corner of Valley Rd and Abbotsford/Mission Hwy 0 Abbotsford 34 5 2 1 4 South-West corner of Fraser Hwy and Mt Lehman Rd 1 Abbotsford 35 5 2 1 4 South-West corner of Fraser Hwy and Mt Lehman Rd 0 Abbotsford 37 3 4 0 2 East side of Gladwin Rd just North of Haida Dr 0 Burnaby 92 1 1 0 0 South side of Southlawn Dr between Beta Ave & Delta Ave 0 Burnaby 88 1 1 0 0 Far South end of Hatton Ave 0 Burnaby 102 2 1 0 0 North side of Balmoral St between Griffiths Ave & Salisbury Ave 0 Burnaby 112 1 2 0 0 South side of Hammarskjold Dr midway between Kensington & Hastings 0 Burnaby 84 2 2 1 1 North-West corner of Eton St & Gilmore Ave 0 Burnaby 90 3 2 0 2 East (South?) side of 16th Ave at bend South of 4th St 0 Burnaby 106 3 2 0 1 West side of Patterson Ave between Beresford St & Wilson Ave 0 Burnaby 120 3 2 0 1 South-East corner of Moscrop St & Alderwood Cres 0 Burnaby 81 4 4 0 3 Far South end of Bell Ave near Lougheed Hwy 0 Burnaby 87 4 4 0 2 South-East corner of Nelson Ave & Bennett St 0 Coquitlam 69 1 1 0 0 Far West end of Magnolia Pl on the North Side 0 Coquitlam 105 1 1 0 0 North of intersection of Bowron St & Nicola Ave 0 Coquitlam 72 2 1 0 0 North-West corner of Admiral Crt & Palmdale St 0 Coquitlam 93 2 2 0 0 North side of LeClair Dr just North of Lorraine Ave 0 Coquitlam 70 3 2 0 1 North side of Town Centre Blvd 1 Coquitlam 71 3 2 0 1 North side of Town Centre Blvd 0 Coquitlam 111 4 4 0 3 West side of Blue Mountain St just North of King Albert Ave 1 Coquitlam 110 4 4 0 3 West side of Blue Mountain St just North of King Albert Ave 0 Delta 18 1 1 0 0 West of intersection of 84B Ave & 115th St 0 Delta 17 2 1 1 0 South-East corner of 115th St & 72A Ave 0 Delta 13 2 2 0 1 South side of 6th Ave 200m W of 52nd St 1 Delta 14 2 2 0 1 South side of 6th Ave 200m W of 52nd St 0 Delta 12 4 4 1 3 South-East corner of 53rd St & Ladner Trunk Rd 0 Langley 32 1 1 0 0 North-East corner of 204th St & 45A Ave Cover Missing 0 Langley 33 2 2 0 0 East side of 236th St just North of 50th Ave 0 Langley 31 4 4 1 3 North-West corner of Fraser Hwy & 201A St 0 Maple Ridge 64 2 2 0 0 West side of 231B St just North of 117 Ave 0 Maple Ridge 65 3 2 0 0 South-East corner of Dunbar St & 122 Ave 0 New Westminster 83 1 1 0 0 South side of Hamilton St midway between 16th St & 14th St 15 1 New Westminster 86 1 1 0 0 South side of Hamilton St midway between 16th St & 14th St 0 New Westminster 114 2 1 0 0 East side of 3rd St just North of 3rd Ave 0 New Westminster 89 5 6 0 5 West side of Brunette Ave between Hwy 1 and Braid St 0 North Vancouver 57 1 1 0 0 South side of Pinewood Cr (East/West) East of Redwood St 0 North Vancouver 58 1 1 0 0 Intersection of Wavertree Rd & Skyline Dr 0 North Vancouver 62 2 1 0 0 North side of E 21st St between Anita Cr & Casano Dr 1 North Vancouver 63 2 1 0 0 North side of E 21st St between Anita Cr & Casano Dr 0 North Vancouver 61 1 2 0 0 North-West corner of intersection of Haywood St & E 4th St 0 North Vancouver 59 3 2 0 1 Far West end of W 21st St, West of Jones 0 North Vancouver 60 3 4 0 1 Lonsdale Ave between W 5th St & W 4th St 0 Pitt Meadows 66 2 1 0 1 East side of Reichenbach Rd 300m North of Dewdney Trunk Rd 0 Port Coquitlam 67 1 1 0 0 Far West end of Lincoln Ave 0 Port Coquitlam 68 1 1 0 0 North side of Fraser Ave between Shaughnessy St & Flint St 0 Port Coquitlam 73 1 1 0 0 North side of Celeste Cres at intersection with Delia Dr 0 Richmond 3 1 1 0 0 South-West corner of Beecham Rd & Lockhart Rd 0 Richmond 5 1 1 0 0 South side of Georgia St between 3rd Ave & 2nd Ave 0 Richmond 8 1 1 0 0 West side of Ruskin Rd 120m North of Ryan Rd 0 Richmond 9 1 1 1 0 South corner of Aquila Rd & Dennis Cres 0 Richmond 11 1 1 0 0 Far East end of Livingstone Pl on the East side of the North/South bend 0 Richmond 4 3 4 0 1 South side of Francis Rd just West of Craigflower Gate 0 Richmond 6 3 4 0 1 West side of Gilbert Rd between Petts Rd & Bamberton Dr 1 Richmond 7 3 4 0 1 West side of Gilbert Rd between Petts Rd & Bamberton Dr 0 Richmond 2 4 4 1 2 North-East corner of Westminster Hwy & Cooney Rd 0 Richmond 10 4 4 0 2 West side of #3 Rd midway between Granville Ave & Bennett Rd 0 Surrey 16 1 1 0 0 North side of 70th Ave between 126A St & 127A St 0 Surrey 27 1 1 0 0 North side of Poplar Dr just East of 154th St 1 Surrey 28 1 1 0 0 North side of Poplar Dr just East of 154th St 0 Surrey 41 1 1 0 0 North of intersection of 149A St & 95th Ave Sampler Missing 1 Surrey 42 1 1 0 0 North of intersection of 149A St & 95th Ave Cover Missing 0 Surrey 45 1 1 0 0 South side of Campbell Pl just East of 127th St 0 Surrey 46 1 1 0 0 South-West corner of 97A Ave & 117B St 0 Surrey 50 1 1 0 0 West end of 117th Ave just East of 138 St 0 Surrey 113 1 1 0 0 North end of 164 St on the West side 0 Surrey 40 2 1 1 0 South-West corner of 162nd St & 90th Ave 0 Surrey 47 2 1 0 2 North/East side of intersection of Faulkner Rd and Tannery Rd 0 Surrey 48 1 2 0 0 North side of Melrose Dr between Grosvenor Rd & Coventry Rd 1 Surrey 49 1 2 0 0 North side of Melrose Dr between Grosvenor Rd & Coventry Rd 0 Surrey 44 2 2 0 0 East side of Old Yale Rd between 105th Ave & 104A Ave 0 Surrey 15 3 2 0 0 North side of 70th Ave 220m East of 138th St Cover Missing 16 0 Surrey 19 3 2 1 1 North-West corner of 122A St & 82nd Ave 0 Surrey 24 3 2 0 2 North side of 16th Ave just West of 130th St 0 Surrey 29 3 2 0 2 Outside of 17660 60th Ave 0 Surrey 30 3 2 1 2 North-West corner of 192nd St & 72nd Ave 0 Surrey 43 4 2 0 2 South side of 100th Ave 100m West of 140th St 0 Surrey 25 3 4 0 2 North side of 16th Ave midway between 142nd St & Bishop Rd Sampler Missing 0 Surrey 20 4 4 0 3 West side of 140th St 100m South of intersection with 88th Ave 1 Surrey 21 4 4 0 3 West side of 140th St 100m South of intersection with 88th Ave 0 Surrey 51 4 4 0 3 Far South end of 152nd St South of 108th Ave 0 Vancouver 22 1 1 0 0 South side of Fraserview Dr midway between Burquitlam Dr & Nanaimo St 0 Vancouver 118 1 1 0 0 North-West of intersection of Northumberland Ave & Beadnell Cres 0 Vancouver 121 1 1 0 0 North side of Talisman Ave between Yukon St & Dinmont Ave 0 Vancouver 103 1 1 0 0 East of intersection of Skeena St & E 5th Ave 0 Vancouver 1 2 1 1 1 South-East corner of Cartier St & W 71st Ave 0 Vancouver 53 2 1 0 1 North side of Hamilton St just East of Drake St 0 Vancouver 54 2 1 1 0 North-West corner of Broughton St & Burnaby St 0 Vancouver 52 2 1 0 1 East side of Bayshore Dr between Bidwell St & Cardero St 0 Vancouver 95 2 1 0 1 East side of McHardy St between Clive Ave & Austrey Ave 1 Vancouver 96 2 1 0 1 East side of McHardy St between Clive Ave & Austrey Ave 0 Vancouver 126 2 1 0 1 South side of E 21st Ave between Windsor St & Glen Dr 0 Vancouver 75 2 1 0 0 North side of St Lawrence St between Nanaimo St & Claredon St 0 Vancouver 119 2 1 0 0 South-West corner of Elliot St & E 45th Ave 1 Vancouver 127 2 1 0 0 South-West corner of Elliot St & E 45th Ave 0 Vancouver 76 2 1 0 0 East side of Lanark St just South of E 49th Ave 0 Vancouver 94 2 1 0 0 East side of Prince Edward St between E 40th Ave & E Woodstock Ave 0 Vancouver 99 2 1 0 0 East side of Cambie St between W 60th Ave & W 61st Ave 1 Vancouver 125 2 1 0 0 East side of Cambie St between W 60th Ave & W 61st Ave 0 Vancouver 122 2 1 0 0 West side of The Crescent between McRae Ave & Angus Dr 0 Vancouver 100 2 1 0 0 North-West corner of Trafalgar St & W 13th Ave 0 Vancouver 97 2 1 1 0 North-West corner of Bayswater St & W 3rd Ave 1 Vancouver 104 2 1 1 0 North-West corner of Bayswater St & W 3rd Ave 0 Vancouver 80 2 1 0 0 North-East corner of Crown St & W 19th Ave 0 Vancouver 171 2 1 0 1 South-East corner of Yew St & W 37th Ave 0 Vancouver 77 2 1 0 1 South side of W 2nd Ave between Burrard St & Pine St 0 Vancouver 108 2 1 0 0 West of the intersection of Eton St & Wall St 0 Vancouver 79 2 1 0 0 East side of Nootka St between Turner St & Georgia St 0 Vancouver 117 2 2 0 1 North-West corner of Fraser St & E 11th Ave 0 Vancouver 78 3 2 0 2 West side of Rupert St between E 20th Ave & E 21st Ave 0 Vancouver 101 3 2 1 0 West side of Trimble St just South of W 3rd Ave 17 0 Vancouver 109 3 2 0 1 North-West corner of Dunbar St & W 33rd Ave Cover Missing 0 Vancouver 85 3 4 0 3 North-West corner of Copley St & E 15th Ave 0 Vancouver 169 3 4 1 4 East side of intersection of Gore Ave & Georgia St 0 Vancouver 23 4 4 1 4 North-East corner of Chester St & Marine Dr 0 Vancouver 115 4 4 0 3 North-West corner of Columbia St & W 12th Ave 0 Vancouver 168 4 4 0 2 South side of W 41st Ave just East of Montgomery St 1 Vancouver 170 4 4 0 2 South side of W 41st Ave just East of Montgomery St 0 Vancouver 91 4 4 0 3 East side of Granville St about 150 ft North of Broadway 0 Vancouver 98 4 4 0 4 North-West corner of Powell St and Clark Dr 0 Vancouver 116 4 4 0 3 North side of E 1st Ave between Commercial Dr & Salsbury Dr 1 Vancouver 74 4 4 0 3 North side of E 1st Ave between Commercial Dr & Salsbury Dr 0 Vancouver 123 4 6 0 4 East side of Knight St between E 35th Ave & East 36th Ave 0 Vancouver 124 4 6 1 3 North-West corner of Cambie St and SW Marine Dr 0 Vancouver 82 4 6 0 3 East side of Cambie St midway between W 40th Ave and W 41st Ave 0 Vancouver 167 4 6 0 2 West side of Granville St about 200 ft North of W 70th Ave 0 Vancouver 128 5 6 0 5 South-East corner of Knight St and E 58th Ave 0 West Vancouver 55 2 1 0 1 Far East end of Clyde Ave, East of 6th St 1 West Vancouver 56 2 1 0 1 Far East end of Clyde Ave, East of 6th St 0 White Rock 26 2 2 1 0 North-East corner of Pacific Ave & Dolphin St 18 19 APPENDIX 2 – PARTICLE SAMPLING LOCATIONS MUNICIPALITY ID LOCATION Burnaby 24 South side of Southlawn Dr between Beta Ave & Delta Ave Burnaby 38 Far South end of Bell Ave near Lougheed Hwy Burnaby 44 South-East corner of Moscrop St & Alderwood Cres Coquitlam 26 North of intersection of Bowron St & Nicola Ave Delta 87 South-East corner of 115th St & 72A Ave Delta 94 South-East corner of 53rd St & Ladner Trunk Rd New Westminster 67 South side of Hamilton St midway between 16th St & 14th St North Vancouver 1 Intersection of Wavertree Rd & Skyline Dr Port Coquitlam 28 North side of Fraser Ave between Shaughnessy St & Flint St Richmond 74 North-East corner of Westminster Hwy & Cooney Rd Richmond 84 West side of Gilbert Rd between Petts Rd & Bamberton Dr Richmond 86 South corner of Aquila Rd & Dennis Cres Surrey 69 Far South end of 152nd St South of 108th Ave Surrey 71 South side of 100th Ave 100m West of 140th St Surrey 73 North of intersection of 149A St & 95th Ave Surrey 75 South side of Campbell Pl just East of 127th St Surrey 80 West side of 140th St 100m South of intersection with 88th Ave Surrey 88 North-West corner of 192nd St & 72nd Ave Vancouver 13 North-West corner of Broughton St & Burnaby St Vancouver 14 East side of intersection of Gore Ave & Georgia St Vancouver 15 East side of Nootka St between Turner St & Georgia St Vancouver 23 North side of E 1st Ave between Commercial Dr & Salsbury Dr Vancouver 29 East of intersection of Skeena St & E 5th Ave Vancouver 31 North-West corner of Columbia St & W 12th Ave Vancouver 47 South-East corner of Yew St & W 37th Ave Appendix A – Truck Route File The truck route shape file was manually created from the DMTI road network using information from 16 GVRD municipalities. The following municipalities have designated truck networks, maps of which are available online or through contact with the appropriate engineering/operations departments: 1. Vancouver 2. Burnaby 3. New Westminster 4. Coquitlam 5. Port Coquitlam 6. Surrey 7. Langley 8. Delta 9. White Rock The following municipalities do not have officially designated truck routes, and the drivers of heavy vehicles are advised to use the arterial road network. For the purposes of this file ‘arterial roads’ are defined as road types 1 and 2 (RD1 and RD2). 10. North Vancouver 11. Richmond 12. Port Moody 13. Maple Ridge 14. Pitt Meadows 15. Abbotsford The remaining municipality of West Vancouver has a list rather than a map of designated routes. There are only three streets on which trucks are allowed in the downhill direction and, for the purposes of this file, these streets have been retained as the truck routes, while those without downhill-only restrictions have been deleted. It is also important to note that all highway and freeway ramps have been removed from the street file, and that divided streets and highways are shown as a single line. Appendix B – Sampling Location Details Site ID Municipality Location Carsa Trucksb PM Subset? Complete Sample?c 1 North Vancouver Wavertree Rd. & Skyline Dr. 1 0 Y Y 2 West Vancouver East end of Clyde Ave. 2 1 Y 3 North Vancouver West end of West 21st St. 3 1 Y 4 North Vancouver East 21st St. b/w Anita & Casano 2 0 Y 5 North Vancouver Pinewood Cr. & Redwood St. 1 0 Y 6 North Vancouver Lonsdale Ave. b/w West 5th & West 4th 3 1 Y 7 North Vancouver Haywood St. & East 4th St. 1 0 Y 8 Coquitlam West end of Magnolia Place 1 0 Y 9 Vancouver Bayshore Dr. b/w Bidwell & Cardero 2 1 Y 10 Vancouver Eton St. & Wall St. 2 0 Y 11 Burnaby Eton St. & Gilmore Ave. 2 1 Y 12 Coquitlam North side of Town Centre Boulevard 3 1 Y 13 Vancouver Broughton St. & Burnaby St. 2 0 Y Y 14 Vancouver Gore Ave. & Georgia St. 3 4 Y Y 15 Vancouver Nootka St. b/w Turner & Georgia 2 0 Y Y 16 Burnaby Hammarskjold Dr. b/w Kensington & Hastings 1 0 Y 17 Port Moody Murray St & Moody St 3 1 Fall Only 18 Port Coquitlam West end of Lincoln Ave. 1 0 Y 19 Vancouver Hamilton St. & Drake St. 2 1 Y 20 Vancouver Trimble St. & West 3rd Ave. 3 0 Y 21 Vancouver Bayswater St. & West 3rd Ave. 2 0 Y 22 Vancouver West 2nd Ave. b/w Burrard & Pine 2 1 Y 23 Vancouver East 1st Ave. b/w Commercial & Salsbury 4 3 Y Y 24 Burnaby Southlawn Dr. b/w Beta & Delta 1 0 Y Y 25 Burnaby South end of Hatton Ave. 1 0 Y 26 Coquitlam Bowron St. & Nicola Ave. 1 0 Y Y 27 Coquitlam Admiral Ct. & Palmdale St. 2 0 Y 28 Port Coquitlam Fraser Ave. b/w Shaughnessy & Flint 1 0 Y Y 29 Vancouver Skeena St. & East 5th Ave. 1 0 Y Y 30 Vancouver Trafalgar St. & West 13th Ave. 2 0 Y 31 Vancouver Columbia St & West 12th Ave. 4 3 Y Y 32 Vancouver Fraser St. & East 11th Ave. 2 1 Y 33 Vancouver Crown St. & West 19th Ave. 2 0 Y 34 Vancouver The Crescent b/w McRae & Angus 2 0 Y 35 Vancouver Copley St. & East 15th Ave. 3 3 Y 36 Vancouver East 21st Ave. b/w Windsor & Glen 2 1 Y 37 Vancouver Rupert St. b/w 20th & 21st 3 2 Fall Only 38 Burnaby Bell Ave. & Lougheed Highway 4 3 Y Y 39 Coquitlam Blue Mountain St. & King Albert Ave. 4 3 Y 40 Vancouver Talisman Ave. b/w Yukon & Dinmont 1 0 Y 41 Pitt Meadows Reichenbach Rd. 300m N of Dewdney Trunk 2 1 Y 42 Vancouver Dunbar St & West 33rd Ave 3 1 Fall Only 43 Vancouver St. Lawrence St. b/w Nanaimo & Claredon 2 0 Y 44 Burnaby Moscrop St. & Alderwood Cr. 3 1 Y Y 45 Coquitlam LeClair Dr. & Lorraine Ave. 2 0 Y 46 Port Coquitlam Celeste Cr. & Delia Dr. 1 0 Y 47 Vancouver Yew St. & West 37th Ave. 2 1 Y Y 48 Vancouver Knight St. b/w 35th & 36th 4 4 Y 49 Vancouver McHardy St. b/w Clive & Austrey 2 1 Y 50 Vancouver West 41st Ave. & Montgomery St. 4 2 Y 51 Vancouver Prince Edward St. b/w 40th & Woodstock 2 0 Y 52 New Westminster Brunette Ave. b/w Highway 1 & Braid 5 5 Y 53 Vancouver Elliot St. & East 45th Ave. 2 0 Y a Subjective measurement of automobile traffic on an ordinal scale, where 1 = very light and 5 = very heavy b Subjective measurement of truck traffic on an ordinal scale where 0 = no trucks observed and 5 = constant diesel traffic c A complete sample is defined as one that averages the spring and fall concentrations for the observed NOX values (all Spring PM2.5 samples were completed). For some locations only Spring or Fall concentrations are available (due to sampler damage or analytic error) and no data are available for two locations. 54 Burnaby Patterson Ave. b/w Beresford & Wilson 3 1 Y 55 Vancouver Lanark St. & East 49th Ave. 2 0 Y 56 Burnaby Nelson Ave. & Bennett St. 4 2 Y 57 Burnaby 16th Ave. at bend South of 4th St. 3 2 Y 58 Maple Ridge Dunbar St. & 122 Ave. 3 0 Y 59 Burnaby Balmoral St. b/w Griffiths & Salisbury 2 0 Y 60 Vancouver Cambie St. b/w 60th & 61st 2 0 Y 61 Vancouver Northumberland Ave. & Beadnell Cr. 1 0 Y 62 Maple Ridge 231B St. & 117 Ave. 2 0 Spring Only 63 Vancouver Chester St. & Marine Dr. 4 4 Y 64 Vancouver Fraserview Dr. b/w Burquitlam & Nanaimo 1 0 Y 65 New Westminster 3rd St. & 3rd Ave. 2 0 Y 66 Vancouver Cartier St. & West 71st Ave. 2 1 Y 67 New Westminster Hamilton St. b/w 16th & 14th 1 0 Y Y 68 Surrey Melrose Dr. b/w Grosvenor & Coventry 1 0 Y 69 Surrey South end of 152nd St. 4 3 Y Y 70 Surrey Old Yale Rd. b/w 105th & 104A 2 0 Y 71 Surrey 100th Ave. 100m West of 140th St. 4 2 Y Y 72 Surrey 97A Ave. & 117B St. 1 0 Y 73 Surrey 149A St & 95th Ave 1 0 Y No Sample 74 Richmond Westminster Highway & Cooney Rd. 4 2 Y Y 75 Surrey Campbell Place & 127th St. 1 0 Y Y 76 Surrey 162nd St. & 90th Ave. 2 0 Y 77 Richmond Beecham Rd. & Lockhart Rd. 1 0 Y 78 Richmond East end of Livingstone Place 1 0 Y 79 Richmond #3 Rd. b/w Granville & Bennett 4 2 Y 80 Surrey 140th St. 100m South of 88th Ave. 4 3 Y Spring Only 81 Delta 84B Ave. & 115th St. 1 0 Y 82 Surrey 122A St. & 82nd Ave. 3 1 Y 83 Richmond Francis Rd. & Craigflower Gate 3 1 Y 84 Richmond Gilbert Rd. b/w Petts & Bamberton 3 1 Y Y 85 Richmond Ruskin Rd. 120m North of Ryan Rd. 1 0 Y 86 Richmond Aquila Rd. & Dennis Cr. 1 0 Y Y 87 Delta 115th St. & 72A Ave. 2 0 Y Y 88 Surrey 192nd St. & 72nd Ave. 3 2 Y Y 89 Richmond Georgia St. b/w 3rd & 2nd 1 0 Y 90 Surrey 70th Ave. b/w 126A & 127A 1 0 Y 91 Surrey 70th Ave 220m East of 138th St 3 0 Fall Only 92 Langley Fraser Highway & 201A St. 4 3 Y 93 Langley 236th St. & 50th Ave. 2 0 Y 94 Delta 53rd St. & Ladner Trunk Rd. 4 3 Y Y 95 Langley 204th St & 45A Ave 1 0 No Sample 96 Surrey Poplar Dr. & 154th St. 1 0 Y 97 Surrey 16th Ave. & 130th St. 3 2 Y 98 Surrey 16th Ave b/w 142nd & Bishop 3 2 Fall Only 99 White Rock Pacific Ave. & Dolphin St. 2 0 Y 100 Delta 6th Ave. 200m West of 52nd St. 2 1 Y E1 Surrey Outside of 17660 60th Ave. 3 2 Y E2 Vancouver West 70th Ave. & Granville St. 4 2 Y E3 Vancouver BRd.way Ave. & Granville St. 4 3 Y E4 Vancouver Cambie St. & South West Marine Dr. 4 3 Y E5 Vancouver Cambie St. & West 41st Ave. 4 3 Y E6 Vancouver East 57th Ave. & Knight St. 5 5 Y E7 Vancouver Clark St. & Pender Ave. 4 4 Y E8 Surrey North end of 164th St. 1 0 Y E9 Surrey 117th Ave. & King Rd. 1 0 Y E10 Surrey North end of Tannery Rd. 2 2 Y E11 Abbotsford Fraser Highway & Mount Lehman Rd. 5 4 Y E12 Abbotsford Gladwin Rd. & Haida Dr. 3 2 Y E13 Abbotsford Evergreen St & Conrad Ave 2 0 Fall Only E14 Abbotsford Abbotsford-Mission Highway & Valley Rd. 4 5 Y E15 Abbotsford North end of Marble Hill Dr. 1 0 Y E16 Port Moody Clark St. & Saint Johns St. 4 3 Fall Only Appendix C – Results Table 1 – Results for the February Ogawa sampler deployment in the GVRD Site_ID Municipality Location NOX NO2 NO SO2 1 North Vancouver Wavertree Rd. & Skyline Dr. 21.09 9.29 11.81 2.11 2 West Vancouver East end of Clyde Ave. 34.41 15.02 19.39 2.00 3 North Vancouver West end of West 21st St. 38.19 15.84 22.35 2.91 4 North Vancouver East 21st St. b/w Anita & Casano 34.18 13.37 20.82 3.12 5 North Vancouver Pinewood Cr. & Redwood St. 31.47 11.54 19.93 2.90 6 North Vancouver Lonsdale Ave. b/w West 5th & West 4th 58.58 20.19 38.39 2.91 7 North Vancouver Haywood St. & East 4th St. 40.53 16.96 23.57 3.19 8 Coquitlam West end of Magnolia Place 21.24 12.82 8.43 2.76 9 Vancouver Bayshore Dr. b/w Bidwell & Cardero 43.33 21.17 22.16 2.97 10 Vancouver Eton St. & Wall St. 45.67 19.53 26.14 2.44 11 Burnaby Eton St. & Gilmore Ave. 36.60 15.03 21.58 3.51 12 Coquitlam North side of Town Centre Boulevard 33.08 14.29 18.78 2.53 13 Vancouver Broughton St. & Burnaby St. 57.37 22.85 34.52 2.52 14 Vancouver Gore Ave. & Georgia St. 84.71 24.84 59.86 3.07 15 Vancouver Nootka St. b/w Turner & Georgia 39.56 18.24 21.33 2.44 16 Burnaby Hammarskjold Dr. b/w Kensington & Hastings 26.68 14.29 12.40 1.61 17 Port Moody Murray St. & Moody St. Sampler Damaged 18 Port Coquitlam West end of Lincoln Ave. 21.09 11.88 9.21 2.69 19 Vancouver Hamilton St. & Drake St. 64.24 26.02 38.22 3.28 20 Vancouver Trimble St. & West 3rd Ave. 59.89 21.15 38.73 3.57 21 Vancouver Bayswater St. & West 3rd Ave. 52.77 20.27 32.50 2.81 22 Vancouver West 2nd Ave. b/w Burrard & Pine 60.67 22.85 37.82 132.08 23 Vancouver East 1st Ave. b/w Commercial & Salsbury 103.19 24.63 78.56 2.88 24 Burnaby Southlawn Dr. b/w Beta & Delta 37.50 17.77 19.73 2.31 25 Burnaby South end of Hatton Ave. 30.01 15.13 14.88 0.00 26 Coquitlam Bowron St. & Nicola Ave. 46.14 18.25 27.89 0.00 27 Coquitlam Admiral Ct. & Palmdale St. 32.84 14.01 18.83 2.97 28 Port Coquitlam Fraser Ave. b/w Shaughnessy & Flint 40.94 15.47 25.46 2.94 29 Vancouver Skeena St. & East 5th Ave. 76.64 23.69 52.95 0.00 30 Vancouver Trafalgar St. & West 13th Ave. 46.10 18.63 27.48 2.66 31 Vancouver Columbia St & West 12th Ave. 77.14 24.61 52.53 0.00 32 Vancouver Fraser St. & East 11th Ave. 80.28 22.98 57.29 3.88 33 Vancouver Crown St. & West 19th Ave. 37.45 17.10 20.35 2.55 34 Vancouver The Crescent b/w McRae & Angus 40.35 19.04 21.31 0.00 35 Vancouver Copley St. & East 15th Ave. 74.97 21.56 53.41 3.85 36 Vancouver East 21st Ave. b/w Windsor & Glen 47.12 19.90 27.22 0.00 37 Vancouver Rupert St. b/w 20th & 21st Data Lost 83.68 0.00 38 Burnaby Bell Ave. & Lougheed Highway 89.17 22.95 66.22 0.00 39 Coquitlam Blue Mountain St. & King Albert Ave. 63.08 19.31 43.77 0.75 40 Vancouver Talisman Ave. b/w Yukon & Dinmont 39.16 17.29 21.87 0.00 41 Pitt Meadows Reichenbach Rd. 300m N of Dewdney Trunk 27.15 10.22 16.93 2.57 42 Vancouver Dunbar St & West 33rd Ave Sampler Damaged 43 Vancouver St. Lawrence St. b/w Nanaimo & Claredon 45.32 19.14 26.18 3.21 44 Burnaby Moscrop St. & Alderwood Cr. 55.57 17.53 38.03 0.00 45 Coquitlam LeClair Dr. & Lorraine Ave. 30.86 17.41 13.44 0.00 46 Port Coquitlam Celeste Cr. & Delia Dr. 33.10 16.11 17.00 2.77 47 Vancouver Yew St. & West 37th Ave. 43.61 18.06 25.55 0.00 48 Vancouver Knight St. b/w 35th & 36th 72.80 20.86 51.94 1.68 49 Vancouver McHardy St. b/w Clive & Austrey 44.91 17.70 27.21 1.19 50 Vancouver West 41st Ave. & Montgomery St. 53.51 19.53 33.98 3.20 51 Vancouver Prince Edward St. b/w 40th & Woodstock 40.16 20.46 19.70 0.00 52 New Westminster Brunette Ave. b/w Highway 1 & Braid 105.67 28.33 77.34 97.49 53 Vancouver Elliot St. & East 45th Ave. 44.92 17.00 27.91 1.82 54 Burnaby Patterson Ave. b/w Beresford & Wilson 30.33 17.17 13.16 1.62 55 Vancouver Lanark St. & East 49th Ave. 43.81 18.29 25.52 3.68 56 Burnaby Nelson Ave. & Bennett St. 54.43 21.59 32.84 2.35 57 Burnaby 16th Ave at bend South of 4th St 45 68 20 48 25 20 0 00 58 Maple Ridge Dunbar St. & 122 Ave. 33.82 13.41 20.41 3.06 59 Burnaby Balmoral St. b/w Griffiths & Salisbury 42.47 20.52 21.94 0.00 60 Vancouver Cambie St. b/w 60th & 61st 41.79 18.44 23.35 1.80 61 Vancouver Northumberland Ave. & Beadnell Cr. 30.77 17.53 13.25 1.30 62 Maple Ridge 231B St. & 117 Ave. 26.64 11.47 15.18 3.07 63 Vancouver Chester St. & Marine Dr. 173.39 35.72 137.67 1.95 64 Vancouver Fraserview Dr. b/w Burquitlam & Nanaimo 65.25 25.79 39.46 2.91 65 New Westminster 3rd St. & 3rd Ave. 40.68 19.48 21.21 0.00 66 Vancouver Cartier St. & West 71st Ave. 76.46 31.80 44.66 2.28 67 New Westminster Hamilton St. b/w 16th & 14th 42.54 20.19 22.34 1.38 68 Surrey Melrose Dr. b/w Grosvenor & Coventry 36.99 16.31 20.68 0.64 69 Surrey South end of 152nd St. 108.84 23.89 84.95 1.52 70 Surrey Old Yale Rd. b/w 105th & 104A 38.53 18.46 20.07 1.70 71 Surrey 100th Ave. 100m West of 140th St. 92.52 24.02 68.50 2.09 72 Surrey 97A Ave. & 117B St. 41.15 16.05 25.10 1.64 73 Surrey 149A St & 95th Ave Sampler Damaged 74 Richmond Westminster Highway & Cooney Rd. 153.36 35.51 117.85 2.25 75 Surrey Campbell Place & 127th St. 46.50 15.22 31.28 1.72 76 Surrey 162nd St. & 90th Ave. 52.86 18.78 34.08 2.63 77 Richmond Beecham Rd. & Lockhart Rd. 67.64 23.29 44.35 4.96 78 Richmond East end of Livingstone Place 84.02 22.00 62.02 3.95 79 Richmond #3 Rd. b/w Granville & Bennett 103.45 27.47 75.98 2.62 80 Surrey 140th St. 100m South of 88th Ave. 98.29 22.03 76.26 3.57 81 Delta 84B Ave. & 115th St. 54.82 20.28 34.54 1.30 82 Surrey 122A St. & 82nd Ave. 71.31 23.31 48.01 0.97 83 Richmond Francis Rd. & Craigflower Gate 75.23 24.08 51.15 2.22 84 Richmond Gilbert Rd. b/w Petts & Bamberton 110.42 23.34 87.09 1.69 85 Richmond Ruskin Rd. 120m North of Ryan Rd. 74.35 17.85 56.50 2.47 86 Richmond Aquila Rd. & Dennis Cr. 60.72 21.87 38.85 1.91 87 Delta 115th St. & 72A Ave. 73.31 22.79 50.53 5.04 88 Surrey 192nd St. & 72nd Ave. 49.90 13.90 36.00 2.25 89 Richmond Georgia St. b/w 3rd & 2nd 68.19 22.06 46.13 2.12 90 Surrey 70th Ave. b/w 126A & 127A 56.25 21.06 35.20 1.01 91 Surrey 70th Ave 220m East of 138th St Sampler Damaged 92 Langley Fraser Highway & 201A St. 77.38 21.46 55.92 1.32 93 Langley 236th St. & 50th Ave. 24.21 12.21 11.99 1.97 94 Delta 53rd St. & Ladner Trunk Rd. 103.56 24.33 79.22 5.61 95 Langley 204th St & 45A Ave Sampler Damaged 96 Surrey Poplar Dr. & 154th St. 35.10 16.24 18.86 1.33 97 Surrey 16th Ave. & 130th St. 36.50 12.75 23.75 32.98 98 Surrey 16th Ave b/w 142nd & Bishop Sampler Damaged 99 White Rock Pacific Ave. & Dolphin St. 28.49 11.91 16.59 1.58 100 Delta 6th Ave. 200m West of 52nd St. 37.45 16.29 21.15 1.23 E1 Surrey Outside of 17660 60th Ave. 92.45 21.23 71.22 1 71 E2 Vancouver West 70th Ave. & Granville St. 81.29 22.29 59.00 2.68 E3 Vancouver BRd.way Ave. & Granville St. 121.84 28.63 93.21 4.21 E4 Vancouver Cambie St. & South West Marine Dr. 83.76 23.73 60.03 66.78 E5 Vancouver Cambie St. & West 41st Ave. 81.78 23.33 58.45 2.61 E6 Vancouver East 57th Ave. & Knight St. 72.00 19.71 52.28 9.09 E7 Vancouver Clark St. & Pender Ave. 93.51 28.02 65.50 3.47 E8 Surrey North end of 164th St. 30.25 13.37 16.87 0.00 E9 Surrey 117th Ave. & King Rd. 47.15 17.27 29.88 1.81 E10 Surrey North end of Tannery Rd. 73.99 21.91 52.08 1.82 E11 Abbotsford Fraser Highway & Mount Lehman Rd. 98.73 21.68 77.05 1.74 E12 Abbotsford Gladwin Rd. & Haida Dr. 59.98 19.17 40.80 2.34 E13 Abbotsford Evergreen St & Conrad Ave Sampler Damaged E14 Abbotsford Abbotsford-Mission Highway & Valley Rd. 40.64 14.12 26.52 4.16 E15 Abbotsford North end of Marble Hill Dr. 29.88 16.18 13.71 2.43 E16 Port Moody Clark St. and Saint Johns St. Not Deployed Table 2 – Results for spring particle sampling (PM2.5) in the GVRD Site ID Week Adjusted Concentration (µg/m3) Adjusted Absorbance (10-6/m) 1 30-Apr 1.16 0.24 13 30-Apr 2.90 0.40 14 14-Apr 4.63 0.90 15 5-Mar 4.61 1.14 23 5-Mar 8.29 2.36 24 5-Mar 3.68 1.02 26 4-Apr 2.80 0.55 28 4-Apr 3.57 0.55 29 14-Apr 4.93 1.00 31 14-Apr 6.08 0.84 38 4-Apr 4.29 1.12 44 22-Apr 3.09 0.47 47 14-Apr 4.70 0.60 67 4-Apr 0.92 0.62 69 30-Apr 3.06 1.60 71 19-Mar 3.62 1.07 73 22-Apr 2.52 0.34 74 27-Mar 4.39 1.21 75 19-Mar 6.51 0.36 80 19-Mar 1.35 0.64 83 30-Apr 6.04 0.32 86 27-Mar 2.91 0.86 87 27-Mar 4.35 0.77 88 22-Apr 2.64 0.64 94 27-Mar 8.91 1.31 Table 3 – Results for the September Ogawa sampler deployment in the GVRD Site_ID Municipality Location NO NO2 NOX 1 North Vancouver Wavertree Rd. & Skyline Dr. 15.66 6.13 9.52 2 West Vancouver East end of Clyde Ave. 27.87 11.92 15.95 3 North Vancouver West end of West 21st St. 38.22 13.58 24.64 4 North Vancouver East 21st St. b/w Anita & Casano 20.74 10.70 10.03 5 North Vancouver Pinewood Cr. & Redwood St. 22.05 9.32 12.74 6 North Vancouver Lonsdale Ave. b/w West 5th & West 4th 46.04 15.75 30.29 7 North Vancouver Haywood St. & East 4th St. 32.99 17.14 15.85 8 Coquitlam West end of Magnolia Place 17.12 5.97 11.15 9 Vancouver Bayshore Dr. b/w Bidwell & Cardero 45.79 17.82 27.97 10 Vancouver Eton St. & Wall St. 31.93 16.52 15.42 11 Burnaby Eton St. & Gilmore Ave. 25.21 16.88 8.32 12 Coquitlam North side of Town Centre Boulevard 23.12 8.86 14.26 13 Vancouver Broughton St. & Burnaby St. 51.87 18.67 33.20 14 Vancouver Gore Ave. & Georgia St. 46.75 17.14 29.61 15 Vancouver Nootka St. b/w Turner & Georgia 33.37 14.17 19.21 16 Burnaby Hammarskjold Dr. b/w Kensington & Hastings 24.83 13.84 10.99 17 Port Moody Murray St. & Moody St. 39.31 13.41 25.90 18 Port Coquitlam West end of Lincoln Ave. 15.36 7.73 7.63 19 Vancouver Hamilton St. & Drake St. 70.63 23.37 47.25 20 Vancouver Trimble St. & West 3rd Ave. 42.18 15.23 26.95 21 Vancouver Bayswater St. & West 3rd Ave. 55.66 16.35 39.32 22 Vancouver West 2nd Ave. b/w Burrard & Pine 65.18 18.82 46.36 23 Vancouver East 1st Ave. b/w Commercial & Salsbury 71.28 14.00 57.28 24 Burnaby Southlawn Dr. b/w Beta & Delta 27.88 13.17 14.72 25 Burnaby South end of Hatton Ave. 24.26 7.35 16.90 26 Coquitlam Bowron St. & Nicola Ave. 24.95 12.82 12.13 27 Coquitlam Admiral Ct. & Palmdale St. 25.12 10.95 14.17 28 Port Coquitlam Fraser Ave. b/w Shaughnessy & Flint 23.85 8.53 15.32 29 Vancouver Skeena St. & East 5th Ave. 50.28 14.56 35.71 30 Vancouver Trafalgar St. & West 13th Ave. 30.93 12.44 18.49 31 Vancouver Columbia St & West 12th Ave. 55.81 18.64 37.17 32 Vancouver Fraser St. & East 11th Ave. 70.53 17.54 52.99 33 Vancouver Crown St. & West 19th Ave. 27.55 12.27 15.28 34 Vancouver The Crescent b/w McRae & Angus 30.49 15.82 14.67 35 Vancouver Copley St. & East 15th Ave. 45.89 16.60 29.29 36 Vancouver East 21st Ave. b/w Windsor & Glen 48.82 15.60 33.22 37 Vancouver Rupert St. b/w 20th & 21st 58.57 17.54 41.03 38 Burnaby Bell Ave. & Lougheed Highway 57.41 15.66 41.75 39 Coquitlam Blue Mountain St. & King Albert Ave. 41.83 14.71 27.12 40 Vancouver Talisman Ave. b/w Yukon & Dinmont 29.88 9.85 20.03 41 Pitt Meadows Reichenbach Rd. 300m N of Dewdney Trunk 17.08 7.97 9.11 42 Vancouver Dunbar St & West 33rd Ave 50.85 15.46 35.39 43 Vancouver St. Lawrence St. b/w Nanaimo & Claredon 31.11 15.39 15.72 44 Burnaby Moscrop St. & Alderwood Cr. 31.37 12.26 19.11 45 Coquitlam LeClair Dr. & Lorraine Ave. 26.78 8.07 18.72 46 Port Coquitlam Celeste Cr. & Delia Dr. 22.73 9.79 12.94 47 Vancouver Yew St. & West 37th Ave. 28.74 13.57 15.17 48 Vancouver Knight St. b/w 35th & 36th 69.26 19.49 49.77 49 Vancouver McHardy St. b/w Clive & Austrey 31.60 14.06 17.54 50 Vancouver West 41st Ave. & Montgomery St. 43.01 7.31 35.70 51 Vancouver Prince Edward St. b/w 40th & Woodstock 34.31 12.92 21.39 52 New Westminster Brunette Ave. b/w Highway 1 & Braid 92.27 20.26 72.01 53 Vancouver Elliot St. & East 45th Ave. 30.18 13.69 16.48 54 Burnaby Patterson Ave. b/w Beresford & Wilson 30.58 17.58 13.00 55 Vancouver Lanark St. & East 49th Ave. 32.94 13.20 19.73 56 Burnaby Nelson Ave. & Bennett St. 42.84 11.97 30.87 57 Burnaby 16th Ave. at bend South of 4th St. 37.10 11.28 25.83 58 Maple Ridge Dunbar St. & 122 Ave. 16.46 8.36 8.10 59 Burnaby Balmoral St. b/w Griffiths & Salisbury 40.41 16.70 23.71 60 Vancouver Cambie St b/w 60th & 61st 32 39 8 43 23 95 61 Vancouver Northumberland Ave. & Beadnell Cr. 42.40 16.54 25.85 62 Maple Ridge 231B St. & 117 Ave. Sampler Damaged 63 Vancouver Chester St. & Marine Dr. 78.49 19.24 59.25 64 Vancouver Fraserview Dr. b/w Burquitlam & Nanaimo 29.22 13.76 15.46 65 New Westminster 3rd St. & 3rd Ave. 14.23 8.69 5.54 66 Vancouver Cartier St. & West 71st Ave. 40.46 14.25 26.21 67 New Westminster Hamilton St. b/w 16th & 14th 34.17 13.66 20.52 68 Surrey Melrose Dr. b/w Grosvenor & Coventry 22.75 11.96 10.79 69 Surrey South end of 152nd St. 74.85 14.13 60.72 70 Surrey Old Yale Rd. b/w 105th & 104A 27.06 14.28 12.78 71 Surrey 100th Ave. 100m West of 140th St. 42.59 9.65 32.94 72 Surrey 97A Ave. & 117B St. 26.06 12.09 13.97 73 Surrey 149A St & 95th Ave Sampler Damaged 74 Richmond Westminster Highway & Cooney Rd. 53.30 20.50 32.81 75 Surrey Campbell Place & 127th St. 28.09 8.84 19.25 76 Surrey 162nd St. & 90th Ave. 26.95 10.29 16.65 77 Richmond Beecham Rd. & Lockhart Rd. 21.31 8.27 13.04 78 Richmond East end of Livingstone Place 21.25 10.59 10.66 79 Richmond #3 Rd. b/w Granville & Bennett 58.68 13.66 45.02 80 Surrey 140th St. 100m South of 88th Ave. Sampler Damaged 81 Delta 84B Ave. & 115th St. 23.08 9.42 13.66 82 Surrey 122A St. & 82nd Ave. 31.89 12.01 19.88 83 Richmond Francis Rd. & Craigflower Gate 25.00 11.38 13.62 84 Richmond Gilbert Rd. b/w Petts & Bamberton 39.67 11.19 28.49 85 Richmond Ruskin Rd. 120m North of Ryan Rd. 19.86 8.74 11.12 86 Richmond Aquila Rd. & Dennis Cr. 26.96 7.44 19.52 87 Delta 115th St. & 72A Ave. 27.70 8.58 19.11 88 Surrey 192nd St. & 72nd Ave. 23.23 8.14 15.08 89 Richmond Georgia St. b/w 3rd & 2nd 19.70 10.23 9.47 90 Surrey 70th Ave. b/w 126A & 127A 23.21 6.94 16.27 91 Surrey 70th Ave 220m East of 138th St 21.52 10.77 10.75 92 Langley Fraser Highway & 201A St. 35.38 9.94 25.44 93 Langley 236th St. & 50th Ave. 14.41 7.07 7.35 94 Delta 53rd St. & Ladner Trunk Rd. 49.09 9.96 39.13 95 Langley 204th St & 45A Ave Sampler Damaged 96 Surrey Poplar Dr. & 154th St. 21.76 11.31 10.45 97 Surrey 16th Ave. & 130th St. 21.17 10.72 10.44 98 Surrey 16th Ave b/w 142nd & Bishop 26.39 9.81 16.58 99 White Rock Pacific Ave. & Dolphin St. 18.10 5.90 12.20 100 Delta 6th Ave. 200m West of 52nd St. 18.02 7.18 10.84 E1 Surrey Outside of 17660 60th Ave. 29.66 11.90 17.76 E2 Vancouver West 70th Ave. & Granville St. 51.28 20.10 31.18 E3 Vancouver BRd.way Ave. & Granville St. 128.61 24.41 104.19 E4 Vancouver Cambie St. & South West Marine Dr. 83.43 17.35 66.08 E5 Vancouver Cambie St. & West 41st Ave. 51.38 19.81 31.57 E6 Vancouver East 57th Ave. & Knight St. 42.42 17.03 25.39 E7 Vancouver Clark St. & Pender Ave. 78.33 22.50 55.82 E8 Surrey North end of 164th St. 23.53 12.06 11.46 E9 Surrey 117th Ave. & King Rd. 41.48 15.46 26.02 E10 Surrey North end of Tannery Rd. 43.21 12.97 30.24 E11 Abbotsford Fraser Highway & Mount Lehman Rd. 45.64 13.75 31.89 E12 Abbotsford Gladwin Rd. & Haida Dr. 34.55 12.54 22.01 E13 Abbotsford Evergreen St & Conrad Ave 21.15 9.59 11.56 E14 Abbotsford Abbotsford-Mission Highway & Valley Rd. 22.66 7.71 14.96 E15 Abbotsford North end of Marble Hill Dr. 14.68 6.84 7.85 E16 Port Moody Clark St. and Saint Johns St. 37.96 12.77 25.19 Appendix D – Quality Control 1 Oxides of Nitrogen 1.1 Co-Located Samplers Table 1 – Results from Ogawa samplers co-located with continuous samplers at GVRD stations GVRD Station Location Ogawa NOX GVRD NOX Ogawa NO2 GVRD NO2 Ogawa NO GVRD NO T1 Downtown Vancouver 61.38 52.65 25.40 24.52 35.98 28.13 T2 Kitsilano 56.84 54.63 23.68 24.28 33.15 30.35 T4 North Burnaby 32.82 30.02 18.09 18.68 14.72 11.33 T6 North Vancouver 38.87 32.65 19.59 18.60 19.28 14.05 T9 Port Moody 44.38 36.21 21.12 17.82 23.26 18.39 T13 North Delta 41.31 36.64 19.77 18.55 21.54 18.09 T15 East Surrey 23.29 18.07 13.86 12.64 9.44 5.43 T17 South Richmond 50.10 47.72 18.18 17.14 31.92 30.58 T18 South Burnaby 40.81 35.00 21.03 21.98 19.78 13.02 T20 Pitt Meadows 35.45 29.80 16.79 14.42 18.66 15.37 T27 Langley 17.45 14.99 14.21 9.55 3.24 5.43 T30 Maple Ridge 26.00 23.08 13.50 14.25 12.50 8.83 T32 Coquitlam 38.41 32.28 18.56 17.24 19.85 15.02 T33 Abbotsford 34.58 31.44 16.26 16.58 18.32 14.86 Figure 1 – Relationship between results from passive (Ogawa) and continuous (GVRD) samplers 1.2 Duplicate Samples Table 2 – Results from duplicate samples for oxides of nitrogen Site ID Location NOX Duplicate NOX NO2 Duplicate NO2 NO Duplicate NO 2 West Vancouver 36.32 32.50 15.66 14.37 20.66 18.13 4 North Vancouver 33.78 34.59 13.23 13.50 20.54 21.09 12 Coquitlam 33.25 32.90 14.42 14.17 18.83 18.74 21 Vancouver 51.43 54.12 20.05 20.50 31.38 33.62 23 Vancouver 98.39 108.00 25.11 24.14 73.27 83.86 39 Coquitlam 65.03 61.13 18.81 19.81 46.21 41.32 49 Vancouver 44.74 45.09 18.45 16.95 26.29 28.14 50 Vancouver 52.79 54.24 19.09 19.98 33.70 34.26 53 Vancouver 44.20 45.63 16.75 17.26 27.45 28.37 60 Vancouver 40.64 42.94 18.23 18.65 22.41 24.29 67 New Westminster 41.74 43.34 20.38 20.01 21.36 23.33 68 Surrey 36.26 37.72 14.62 18.00 21.65 19.71 80 Surrey 90.99 105.60 23.67 20.40 67.33 85.19 84 Richmond 110.71 110.14 23.88 22.79 86.83 87.34 96 Surrey 34.42 35.78 15.39 17.09 19.02 18.69 100 Delta 38.10 36.79 16.44 16.15 21.66 20.64 E11 Abbotsford 99.61 97.85 21.16 22.21 78.45 75.64 Figure 2 – Relationship between duplicate sampler concentrations for oxides of nitrogen 1.3 Blank Samples Table 3 – Aqueous nitrite concentrations for NOX and NO2 field and lab blanks Sample ID Field Blank/ Lab Blank NOX Nitrite (ppm) NO2 Nitrite (ppm) 107 L 0.000 0.000 132 L 0.059 0.062 133 F 0.060 0.000 134 F 0.070 0.000 135 F 0.057 0.049 136 F 0.050 0.030 137 F 0.000 0.022 138 F 0.061 0.000 139 F 0.076 0.000 141 L 0.061 0.000 147 F 0.000 0.000 148 F 0.000 0.000 151 F 0.000 0.000 152 F 0.062 0.055 153 F 0.000 0.000 154 F 0.058 0.000 155 F 0.000 0.000 157 F 0.062 0.000 158 F 0.000 0.000 159 F 0.000 0.000 160 F 0.000 0.000 162 F 0.061 0.000 163 F 0.000 0.000 164 L 0.058 0.000 165 L 0.000 0.000 AVERAGE 0.032 0.009 Standard Deviation 0.031 0.019 The aqueous limit of detection (LOD) for NOX and NO2 can then be calculated as follows: LOD = Average of Blanks + 3(Standard Deviation of Blanks) NOX = 0.032 + 3*0.031 = 0.125 ppm of nitrite NO2 = 0.009 + 3*0.019 = 0.066 ppm of nitrite The minimum nitrite concentration in solutions resulting from the extraction of NOX collection pads was 1.085 ppm, and the minimum from NO2 collection pads was 0.530 ppm. 2 Sulfur Dioxide 2.1 Co-Located Samplers Table 4 – SO2 Results from Ogawa samplers co-located with continuous samplers at GVRD stations GVRD Station Location Ogawa SO2 GVRD SO2 T1 Downtown Vancouver 3.03 4.25 T2 Kitsilano 0.00 2.78 T4 North Burnaby 2.68 1.24 T6 North Vancouver 2.32 1.74 T9 Port Moody 2.19 2.20 T18 South Burnaby 0.00 1.00 T20 Pitt Meadows 0.00 1.19 T27 Langley 0.00 1.11 T33 Abbotsford 0.00 0.85 Figure 3 – Relationship between results from passive (Ogawa) and continuous (GVRD) samplers 2.2 Duplicate Samples Table 5 – Results from duplicate samples for sulfur dioxide Site ID Location SO2 Duplicate SO2 2 West Vancouver 1.04 2.95 4 North Vancouver 3.61 2.63 12 Coquitlam 3.03 2.03 21 Vancouver 2.44 3.18 23 Vancouver 3.58 2.17 39 Coquitlam 0.00 1.51 49 Vancouver 2.39 0.00 50 Vancouver 4.01 2.40 53 Vancouver 1.49 2.14 60 Vancouver 2.12 1.49 67 New Westminster 2.76 0.00 68 Surrey 1.28 0.00 80 Surrey 5.08 2.07 84 Richmond 1.41 1.97 96 Surrey 1.24 1.42 100 Delta 1.09 1.37 E11 Abbotsford 1.98 1.50 Figure 4 -- Relationship between duplicate sampler concentrations for oxides of nitrogen 2.3 Blank Samples Table 6 -- Aqueous sulfate concentrations for SO2 field and lab blanks Sample Field Blank/ Lab Blank SO2 Sulfate (ppm) 107 L 1.997 132 L 0.000 133 F 0.000 134 F 0.000 135 F 0.063 136 F 0.000 137 F 0.000 138 F 0.000 139 F 0.000 141 L 0.135 147 F 0.000 148 F 0.138 151 F 0.000 152 F 0.000 153 F 0.000 154 F 0.000 155 F 0.000 157 F 0.139 158 F 0.000 159 F 0.000 160 F 0.000 162 F 0.162 163 F 0.000 164 L 0.158 165 L 0.292 AVERAGE 0.123 Standard Deviation 0.398 The aqueous limit of detection (LOD) for SO2 can then be calculated as follows: LOD = Average of Blanks + 3(Standard Deviation of Blanks) = 0.123 + 3*0.398 = 1.317 ppm of nitrite 3 Particulate Matter 3.1 Co-Located Samplers Table 7 – PM2.5 Concentrations at GVRD station T18 in South Burnaby Sampling Period Adjusted Concentrations from Harvard Impactor (µg/m3) Concentrations from GVRD Continuous Particle Sampler (µg/m3) March 5th – March 12th 4.75 2.73 March 27th – April 3rd 4.51 2.98 April 4th – April 11th 2.54 3.21 April 14th – April 21st 1.49 4.08 April 22nd – April 29th 2.72 3.90 April 30th – May 7th 4.24 4.90 Figure 5 – Relationship between concentrations from Harvard impactor and continuous sampler at T18 3.2 Blank Samples Table 8 – Change in weight for blank filters Filter ID Change in Weight (mg) 4119 0.028 4121 0.028 4122 0.022 4124 0.019 4136 0.010 4148 -0.001 Average 0.018 Standard Deviation 0.011 The limit of detection (LOD) for PM2.5 can then be calculated at follows: LOD = Average of Blanks + 3(Standard Deviation of Blanks) = 0.018 + 3*0.011 = 0.052