High-resolution quantification of atmospheric CO2 mixing ratios in the Greater Toronto Area, Canada

Abstract

Many stakeholders are seeking methods to reduce carbon dioxide (CO${}_{\mathrm{2}}$) emissions in urban areas, but reliable, high-resolution inventories are required to guide these efforts. We present the development of a high-resolution CO${}_{\mathrm{2}}$ inventory available for the Greater Toronto Area and surrounding region in Southern Ontario, Canada (area of $\sim$ $\mathrm{2.8}\times {\mathrm{10}}^{\mathrm{5}}$ km${}^{\mathrm{2}}$, 26 % of the province of Ontario). The new SOCE (Southern Ontario CO${}_{\mathrm{2}}$ Emissions) inventory is available at the $\mathrm{2.5}\times \mathrm{2.5}$ km spatial and hourly temporal resolution and characterizes emissions from seven sectors: area, residential natural-gas combustion, commercial natural-gas combustion, point, marine, on-road, and off-road. To assess the accuracy of the SOCE inventory, we developed an observation–model framework using the GEM-MACH chemistry–transport model run on a high-resolution grid with 2.5 km grid spacing coupled to the Fossil Fuel Data Assimilation System (FFDAS) v2 inventories for anthropogenic CO${}_{\mathrm{2}}$ emissions and the European Centre for Medium-Range Weather Forecasts (ECMWF) land carbon model C-TESSEL for biogenic fluxes. A run using FFDAS for the Southern Ontario region was compared to a run in which its emissions were replaced by the SOCE inventory. Simulated CO${}_{\mathrm{2}}$ mixing ratios were compared against in situ measurements made at four sites in Southern Ontario – Downsview, Hanlan's Point, Egbert and Turkey Point – in 3 winter months, January–March 2016. Model simulations had better agreement with measurements when using the SOCE inventory emissions versus other inventories, quantified using a variety of statistics such as correlation coefficient, root-mean-square error, and mean bias. Furthermore, when run with the SOCE inventory, the model had improved ability to capture the typical diurnal pattern of CO${}_{\mathrm{2}}$ mixing ratios, particularly at the Downsview, Hanlan's Point, and Egbert sites. In addition to improved model–measurement agreement, the SOCE inventory offers a sectoral breakdown of emissions, allowing estimation of average time-of-day and day-of-week contributions of different sectors. Our results show that at night, emissions from residential and commercial natural-gas combustion and other area sources can contribute > 80 % of the CO${}_{\mathrm{2}}$ enhancement, while during the day emissions from the on-road sector dominate, accounting for > 70 % of the enhancement.

Introduction

Urban areas are sites of dense population and the intensity of human activities (such as transportation, industry, and residential and commercial development) makes them hotspots for anthropogenic carbon dioxide (CO${}_{\mathrm{2}})$ emissions. While occupying only 3 % of the total land area, urban areas are locations of residence for 54 % of the global population and are the source of 53–87 % of anthropogenic CO${}_{\mathrm{2}}$ emissions globally (IPCC-WG3, 2014; WHO, 2015). When considering Canada alone, the urban population accounts for an even larger fraction of the total (81 % in 2011) (Statistics Canada, 2011) while urban areas cover only 0.25 % of the land area (Statistics Canada, 2009). Recognizing their influence on the global carbon budget, many urban areas are seeking methods to reduce their anthropogenic CO${}_{\mathrm{2}}$ emissions. The Greater Toronto Area (GTA) in southeastern Canada, for example, has committed to the “Change is in the Air” initiative and is a part of the C40 Cities Climate Leadership Group, both of which call to reduce CO${}_{\mathrm{2}}$ emissions 30 % below 1990 levels by 2020 (C40 Cities, 2016; Framework for Public Review and Engagement, 2007). However, in order to effectively guide anthropogenic CO${}_{\mathrm{2}}$ mitigation strategies, reliable inventories are needed, particularly at high spatial and temporal resolution, to gain a better understanding of the carbon cycle (Gurney et al., 2009; Patarasuk et al., 2016). To our knowledge, the only spatially disaggregated CO${}_{\mathrm{2}}$ inventories available for use in the GTA are the EDGAR v.4.2 (Emission Database for Global Atmospheric Research) CO${}_{\mathrm{2}}$ inventory (available at annual, 0.1${}^{\circ }$ $\times$ 0.1${}^{\circ }$ resolution) (EDGAR, 2010) and the FFDAS v2 (Fossil Fuel Data Assimilation System) CO${}_{\mathrm{2}}$ inventory (available at hourly, 0.1${}^{\circ }$ $\times$ 0.1${}^{\circ }$ resolution) (FFDAS, 2010), both which are limited in their spatial and/or temporal resolution and therefore are not well-suited for the quantification and understanding of CO${}_{\mathrm{2}}$ emissions at the urban scale. The Canadian national CO${}_{\mathrm{2}}$ inventory, in contrast, is only available at the provincial level (Environment Canada, 2012).

Efforts to develop emission inventories at the fine spatial and temporal resolution required for urban-scale understanding of CO${}_{\mathrm{2}}$ emissions have been driven by both policy- and science-related questions (Gurney et al., 2009; Patarasuk et al., 2016). From a policy perspective, improving CO${}_{\mathrm{2}}$ emission quantification is essential to independently evaluate whether anthropogenic mitigation regulations are being effectively implemented. From a scientific perspective, gaining information about anthropogenic CO${}_{\mathrm{2}}$ emissions from urban areas has been primarily motivated by atmospheric CO${}_{\mathrm{2}}$ inversions, which are used to better understand the global carbon cycle (Gurney et al., 2009; Patarasuk et al., 2016). Regardless of the motivation, quantification of CO${}_{\mathrm{2}}$ source–sink processes currently uses two techniques: the bottom-up approach and the top-down approach. In the bottom-up approach, local-scale activity level information is combined with appropriate emission factors to infer emission rates. This method has been used widely to develop many inventories (EDGAR, 2010; FFDAS, 2010; Gurney et al., 2009) but is limited by the accuracy of the input parameters. Conversely, in the top-down approach, inverse modelling is used to exploit the variability in atmospheric mixing ratios of CO${}_{\mathrm{2}}$ to identify the source–sink distributions and magnitudes; this method is limited by insufficient mixing ratio data and uncertainties in simulating atmospheric transport (Pillai et al., 2011). Given current policy needs, a strategy using solely bottom-up or top-down approaches is likely insufficient to evaluate CO${}_{\mathrm{2}}$ emissions; rather, a synthesis of the two methodologies is required (Miller and Michalak, 2016). Successful examples of high-resolution CO${}_{\mathrm{2}}$ inventory development are available on the urban scale, such as the Airparif inventory in Île-de-France (publicly available at http://www.airparif.asso.fr/en/index/index) and in Indianapolis, Los Angeles, Salt Lake City, and Phoenix through the Hestia project (Gurney et al., 2012); on the national scale, such as in China (Zhao et al., 2012); and on the global scale (Wang et al., 2013). However, to our knowledge, there are currently no studies that have quantified Canadian CO${}_{\mathrm{2}}$ emissions at the fine spatial and temporal resolution required for urban analyses in Canada.

In an effort to address this gap, this study was focused on quantifying CO${}_{\mathrm{2}}$ emissions at a fine spatial and temporal resolution in the GTA and Southern Ontario (we expanded the inventory beyond the urban area of the GTA so we could exploit information on CO${}_{\mathrm{2}}$ mixing ratios collected at rural areas in Central and Southwestern Ontario, providing additional sites for inventory validation). We present the new high-resolution Southern Ontario CO${}_{\mathrm{2}}$ Emissions (SOCE) inventory, which quantifies CO${}_{\mathrm{2}}$ emissions from seven source sectors (on-road, off-road, area, point, marine, residential, and commercial natural-gas combustion) at 2.5 km $\times$ 2.5 km spatial and hourly temporal resolution for an area covering $\sim$ 26 % of the province of Ontario ($\sim$ 2.8 $\times$ 10${}^{\mathrm{5}}$ km${}^{\mathrm{2}})$. The SOCE inventory was used in combination with the Environment and Climate Change Canada (ECCC) GEM-MACH chemistry–transport model to simulate CO${}_{\mathrm{2}}$ mixing ratios in a domain including southeastern Canada and northeastern USA (hereafter referred to as the “PanAm domain”) for comparison with in situ measurements made by the Southern Ontario Greenhouse Gas Network. Until now, estimates of anthropogenic CO${}_{\mathrm{2}}$ emissions in the GTA were available only from the EDGAR v.4.2 (EDGAR, 2010) and the FFDAS v2 (FFDAS, 2010) inventories, which have very different annual totals for this region (1.36 $\times$ 10${}^{\mathrm{8}}$ vs. 1.05 $\times$ 10${}^{\mathrm{8}}$ t CO${}_{\mathrm{2}}$, respectively). Therefore, we expect the results of this work will improve our ability to quantify the emissions of CO${}_{\mathrm{2}}$ in the entire domain as well as the relative contributions of different sectors, providing a more detailed characterization of the carbon budget in the GTA.

Methods

Geographic domain

The geographic focus of this study was the GTA in Southern Ontario, Canada. The GTA is the largest urban area in Canada; it comprises five municipalities, Toronto, Halton, Durham, Peel, and York, which together have a population exceeding 6 million (Statistics Canada, 2012b). Although the GTA comprises only 0.07 % of Canadian land area, it represents over 17 % of the total population as a result of rapid urbanization over the past few decades (Statistics Canada, 2012b). Therefore, high-resolution characterization of CO${}_{\mathrm{2}}$ emissions can help integrate climate policy with urban planning. This region is home to a network of measurement sites providing long-term, publicly available datasets of atmospheric CO${}_{\mathrm{2}}$ mixing ratio measurements (Sect. 2.2; Environment Canada, 2011) which can be used to evaluate model outputs and inventory estimates. In 2016 the government of Ontario released a Climate Change Action Plan, which includes an endowment given to the Toronto Atmospheric Fund of CAD 17 million to invest in strategies to reduce greenhouse gas pollution in the GTA (Ontario, 2016). Therefore this research can provide timely information on the carbon budget in the GTA and help to implement effective reduction strategies.

Total anthropogenic CO${}_{\mathrm{2}}$ emissions for a weekday in February 2010 estimated by the SOCE inventory for the province of Ontario and by the FFDAS v2 inventory for the remainder of the GEM-MACH PanAm domain. Locations of in situ measurements of CO${}_{\mathrm{2}}$ in the Southern Ontario Greenhouse Gas Network are shown in the inset (Downsview is a square, Egbert is a circle, Hanlan's Point is a triangle, Turkey Point is a diamond). The Downsview and Hanlan's Point sites are both located in the GTA. Unit: g CO${}_{\mathrm{2}}$ s${}^{-\mathrm{1}}$ grid cell${}^{-\mathrm{1}}$.

[Figure omitted. See PDF]

The Southern Ontario Greenhouse Gas Network

Measurements of ambient CO${}_{\mathrm{2}}$ dry air mixing ratios began in 2005 in Southern Ontario at the Egbert station followed by the Downsview station (2007), Turkey Point station (2012), and Hanlan's Point station (2014) (Fig. 1). Measurements were also temporarily made at a site in downtown Toronto, the Toronto Atmospheric Observatory (TAO) (43.7${}^{\circ }$ N, 79.4${}^{\circ }$ W), but the instrument was relocated from this site in January 2016. Egbert is located $\sim$ 75 km north-northwest of Toronto in a rural area, Downsview is located $\sim$ 20 km north of downtown core of the city of Toronto in a populated suburban area, Turkey Point is located to the southwest of the GTA in a rural area on the north shore of Lake Erie, and Hanlan's Point is located on Toronto Island, just south of the city of Toronto on the shore of Lake Ontario. Site details and instrument types used can be found in Table 1. CO${}_{\mathrm{2}}$ measurements are collected as a part of ECCC's Greenhouse Gas Observational Program. The measurement procedure follows a set of established principles and protocols outlined by a number of international agencies through recommendations of the Meeting on Carbon Dioxide, Other Greenhouse Gases, and Related Measurement Techniques, coordinated by the World Meteorological Organization (WMO) every 2 years.

Summary of atmospheric measurement programs in Southern Canada operated by Environment and Climate Change Canada.

			Elevation	Intake	In situ
Start date	Site name	Coordinates	(a.s.l.)	height	instrumentation
March 2005	Egbert	44.231037${}^{\circ }$ N, 79.783834${}^{\circ }$ W	251 m	3 m, 25 m${}^{\ast }$	NDIR
November 2010	Downsview	43.780491${}^{\circ }$ N, 79.468010${}^{\circ }$ W	198 m	20 m	NDIR
November 2012	Turkey Point	42.635368${}^{\circ }$ N, 80.557659${}^{\circ }$ W	231 m	35 m	CRDS
June 2014	Hanlan's Point	43.612201${}^{\circ }$ N, 79.388705${}^{\circ }$ W	87 m	10 m	CRDS

${}^{\ast }$ At Egbert, a 25 m tower was installed in 9 March 2009. NDIR is nondispersive infrared. CRDS is cavity ring-down spectroscopy.

The atmospheric CO${}_{\mathrm{2}}$ observational program at the Egbert site is based on nondispersive infrared (NDIR) methodology and fine-tuned for high-precision measurements (Worthy et al., 2005). A detailed description of the NDIR observational system can be found in Worthy et al. (2005). The atmospheric CO${}_{\mathrm{2}}$ observational programs at Turkey Point, Downsview, and Hanlan's Point are based on cavity ring-down spectroscopy (CRDS). Each Picarro CRDS system is calibrated in the ECCC central calibration facility in Toronto before deployment to the field. The response function of the analyzer is determined against three calibrated standards tanks (low, mid, high). The working (W) and target (T) tanks assigned to the system are also included in the injection sequence and calibrated. At each site, ambient measurements are made using two sample lines placed at the same level. Each sample line has separate dedicated sample pumps and driers ($\sim$ $-$30 ${}^{\circ }$C). Pressurized aluminum 30 L gas cylinders are used for the working and target tanks. The sample flow rate of the ambient and standard tank gases is set at $\sim$ 300 cc min${}^{-\mathrm{1}}$. The injection sequence consists of a target and working tanks sequentially passed through the analyzer for 10 min each every 2 days. The ambient data from line 1 are passed through the analyzer for 18 h followed by line 2 for 6 h. The line 1–line 2 sequence repeats one time before the target and working tanks are again passed through the system. The working and target tanks are calibrated on site at least once per year against a single transfer standard transported between the sites and the central laboratory facility in Toronto. The CO${}_{\mathrm{2}}$ measurements from both the NDIR and CRDS analytical systems have a precision of around 0.1 ppm based on 1 min averages and are accurate to within 0.2 ppm.

GEM-MACH chemistry–transport model

In this project, we used the GEM-MACH (Global Environmental Multi-scale – Modelling Air quality and CHemistry) chemistry–transport model (CTM) (Gong et al., 2015; Moran et al., 2013; Pavlovic et al., 2016; Talbot et al., 2008) to link surface emission estimates and atmospheric mixing ratios. GEM-MACH is an online CTM embedded within the Canadian weather forecast model GEM (Côté et al., 1998a, b). The configuration of GEM-MACH used in our study has 62 vertical levels from the surface to $\sim$ 1.45 hPa on a terrain-following staggered vertical grid for a log-hydrostatic pressure coordinate. The thickness of the lowest layer was 40 m. The PanAm domain used in our simulations, which includes Central and Southern Ontario as well as western Québec and the northeastern USA, is shown in Fig. 1. The PanAm domain has 524 $\times$ 424 grid cells in the horizontal on a rotated latitude–longitude grid with 2.5 km grid spacing and covers an area of approximately 1310 km $\times$ 1060 km (total domain area is 1.39 $\times$ 10${}^{\mathrm{6}}$ km${}^{\mathrm{2}})$. A 24 h forecasting period was used with a 60 s time step for each integration cycle. Meteorological fields (wind, temperature, humidity, etc.) were reinitialized every 24 h (i.e., after each model integration cycle); chemical fields were carried forward from the previous integration cycle (i.e., perpetual forecast). Hourly meteorological and chemical boundary conditions were provided by the ECCC operational 10 km GEM-MACH air quality forecast model (Moran et al., 2015).

In our study, we simulated two scenarios of CO${}_{\mathrm{2}}$ surface fluxes:

Scenario 1 is characterized by the sum of the following:

• anthropogenic fossil fuel CO${}_{\mathrm{2}}$ emissions within the province of Ontario estimated by the SOCE inventory, available at 2.5 km $\times$ 2.5 km spatial and hourly temporal resolution, as described in Sect. 2.4;

• anthropogenic fossil fuel CO${}_{\mathrm{2}}$ emissions estimated by the FFDAS v2 inventory (FFDAS, 2010) outside of the province of Ontario (province of Québec and USA), available at 0.1${}^{\circ }$ $\times$ 0.1${}^{\circ }$ spatial and hourly temporal resolution;

• biogenic CO${}_{\mathrm{2}}$ fluxes from the C-TESSEL land surface model, as described in Sect. 2.5.

Scenario 2 is characterized by the sum of the following:

• anthropogenic fossil fuel CO${}_{\mathrm{2}}$ emissions estimated by the FFDAS v2 inventory (FFDAS, 2010) for the entire domain, available at 0.1${}^{\circ }$ $\times$ 0.1${}^{\circ }$ spatial and hourly temporal resolution;

• biogenic CO${}_{\mathrm{2}}$ fluxes from the C-TESSEL land surface model, as described in Sect. 2.5.

CO${}_{\mathrm{2}}$ is not a usual chemical species considered by GEM-MACH but a set of special inert tracer fields was added to GEM-MACH for this project to account for CO${}_{\mathrm{2}}$ concentration fields associated with difference source sectors and the lateral boundaries. The CO${}_{\mathrm{2}}$ boundary conditions set at the lateral and top edges of the domain were obtained from the Monitoring Atmospheric Composition and Climate (MACC) global inversion, v.10.2 (http://www.gmes-atmosphere.eu/). Model-simulated specific humidity ($q$, kg kg${}^{-\mathrm{1}}$) was used to convert estimated CO${}_{\mathrm{2}}$ mixing ratios to dry air mixing ratios. CO${}_{\mathrm{2}}$ dry air mixing ratios are hereafter referred to CO${}_{\mathrm{2}}$ mixing ratios.

High-resolution SOCE inventory development

The high-resolution SOCE inventory was constructed primarily from a pre-existing carbon monoxide (CO) inventory developed by the Pollutant Inventories and Reporting Division (PIRD) of ECCC as part of the 2010 Canadian Air Pollutant Emissions Inventory (APEI). The CO inventory is a comprehensive national anthropogenic inventory that includes emissions from area sources, point sources, on-road mobile sources, and off-road mobile sources, including aircraft, locomotive, and marine emissions for base year 2010 (Moran et al., 2015). This annual inventory at the provincial level compiled by PIRD was transformed into model-ready emissions files using the Sparse Matrix Operator Kernel Emissions (SMOKE, https://www.cmascenter.org/smoke/) emissions processing system for spatial allocation (distribution of non-point-source emissions to 2.5 km $\times$ 2.5 km (roughly 0.02${}^{\circ }$ $\times$ 0.02${}^{\circ }$ resolution) using spatial surrogate fields) and temporal allocation (conversion of inventory annual emission rates into hourly values) (Moran et al., 2015). Because Ontario CO emissions in the 2010 APEI were processed in separate steps with SMOKE by primary source sector, files of gridded hourly CO emissions fields were available for seven different inventory sectors: area sources, point sources, on-road mobile sources, off-road mobile sources, marine sources, residential natural-gas sources, and commercial natural-gas sources. The spatial and temporal allocations applied to these seven sectors were different, so in effect they constitute a set of spatiotemporal emissions basis functions. More detailed information about the CO inventory compilation and subsequent processing has been provided elsewhere (Environment Canada, 2013; Moran et al., 2015; PIRD, 2016).

The objective of our work was to calculate CO${}_{\mathrm{2}}$ emissions based on this processed, sector-specific, model-ready CO inventory for Ontario grid cells using sector-specific emission ratios estimated by the Canadian National Inventory Report (NIR) (Environment Canada, 2012). The NIR estimates CO${}_{\mathrm{2}}$ and CO emissions using primarily bottom-up estimates; for example, emissions from industrial process are estimated using production data reported directly by facilities whereas emissions from road transport activities are estimated using vehicle population data, fuel consumption ratios, and vehicle kilometres travelled as reported by Environment Canada (2012). In the model-ready Ontario CO inventory, emission sources are classified by SCC (Source Classification Code) and were mapped to NFR (Nomenclature for Reporting) codes for accurate cross-reference with the NIR CO${}_{\mathrm{2}}$ and CO estimates. Provincial totals for CO${}_{\mathrm{2}}$ and CO are estimated based on the NFR sources that are included in the sector, producing the following NIR sector-averaged ${\mathrm{CO}}_{\mathrm{2}}:\mathrm{CO}$ ratio: $${\text{CO}}_{{\mathrm{2}}_{\text{(sector, kt)}}}={\text{CO}}_{\text{(sector, kt)}}\times {\displaystyle \frac{{\text{CO}}_{{\mathrm{2}}_{\text{(Ontario sector, kt)}}}}{{\text{CO}}_{\text{(Ontario sector, kt)}}}}.$$

This sector-averaged ${\mathrm{CO}}_{\mathrm{2}}:\mathrm{CO}$ ratio is used to convert the APEI-based model-ready gridded CO emissions fields into CO${}_{\mathrm{2}}$ emissions fields at the same spatial and temporal resolution. Because the spatial and temporal variability of the sources of CO${}_{\mathrm{2}}$ are similar to those of CO, the fine-resolution gridded CO sectoral emissions fields for Ontario were primarily used as spatial and temporal proxies for CO${}_{\mathrm{2}}$ emissions; the use of NIR-based ${\mathrm{CO}}_{\mathrm{2}}:\mathrm{CO}$ ratios helps to produce realistic emissions estimates of CO${}_{\mathrm{2}}$ despite uncertainties in CO emissions estimates. A detailed outline of this conversion is presented for each of the seven CO emissions sectors in the following subsections. Unless otherwise noted, temporal allocation of emissions in each sector is based on estimates made available by SMOKE.

Area emissions

Area emissions are mostly small stationary sources that represent diffuse emissions that are not inventoried at the facility level. In the APEI CO inventory, the major emission sources in the area sector include emissions from public electricity and heat production (1A1a), residential and commercial plants (1A4a and 1A4b), stationary agriculture–forestry–fishing (1A4c), iron and steel production (2C1), and pulp and paper (2D1). The NIR estimates an Ontario total from these (and other minor sources) of 2.3 $\times$ 10${}^{\mathrm{4}}$ kt CO${}_{\mathrm{2}}$ and 219 kt CO, producing a ${\mathrm{CO}}_{\mathrm{2}}:\mathrm{CO}$ ratio of 107 kt ${\mathrm{CO}}_{\mathrm{2}}/\mathrm{kt}$ CO. This ratio was applied to every area sector grid cell belonging to Ontario in the domain to convert sector CO emissions to CO${}_{\mathrm{2}}$ emissions.

Point emissions

Point emissions are stationary sources in which emissions exit through a stack or identified exhaust. In the APEI CO inventory, the major emission sources in the point sector include public electricity and heat production (1A1a), stationary combustion in manufacturing industries and construction (1A2f), chemical industry (2B5a), pulp and paper (2D1), iron and steel production (2C1), and other metal production (2C5). Unlike the area sector, we found that applying a single ${\mathrm{CO}}_{\mathrm{2}}:\mathrm{CO}$ ratio to every facility did not produce realistic CO${}_{\mathrm{2}}$ emissions due to the negligible emissions of CO and therefore highly variable ${\mathrm{CO}}_{\mathrm{2}}:\mathrm{CO}$ ratios (because of a small denominator). Therefore, we used ECCC Facility Reported Data (Environment Canada, 2015) to identify the geocoded location and annual average ${\mathrm{CO}}_{\mathrm{2}}:\mathrm{CO}$ for 48 individual facilities in Ontario (Table S1 in the Supplement) and applied the specific ${\mathrm{CO}}_{\mathrm{2}}:\mathrm{CO}$ ratios to the grid cells where the facilities were located. In addition, stack height of individual facilities were included in the emission model to optimize plume rise. All other point sources (minor facilities) were scaled by a sector average ${\mathrm{CO}}_{\mathrm{2}}:\mathrm{CO}$ ratio of 313 kt ${\mathrm{CO}}_{\mathrm{2}}/\mathrm{kt}$ CO, calculated from Ontario total CO${}_{\mathrm{2}}$ and CO point-source emissions from the NIR. Temporal allocation of emissions in the point sector is based on facility level operating schedule data collected by ECCC.

On-road mobile emissions

On-road emissions include the emissions from any on-road vehicles (quantified by the Statistics Canada Canadian Vehicle Survey) (Environment Canada, 2013). In the APEI CO inventory, the major emission sources in the on-road sector includes gasoline- and diesel-powered light- and heavy-duty vehicles (1A3b). The NIR estimates an Ontario total from these (and other minor on-road sources) of 4.4 $\times$ 10${}^{\mathrm{4}}$ kt CO${}_{\mathrm{2}}$ and 1.5 $\times$ 10${}^{\mathrm{3}}$ kt CO, producing a ${\mathrm{CO}}_{\mathrm{2}}:\mathrm{CO}$ ratio of 29 kt ${\mathrm{CO}}_{\mathrm{2}}/\mathrm{kt}$ CO. This ratio was applied to every on-road grid cell belonging to Ontario in the domain to convert sector CO emissions to CO${}_{\mathrm{2}}$. Temporal allocation of emissions in the on-road sector is estimated using data collected in the FEVER (Fast Evolution of Vehicle Emissions from Roadways) campaign in 2010 (Gordon et al., 2012a, b; Zhang et al., 2012). There are challenges associated with using a single ${\mathrm{CO}}_{\mathrm{2}}:\mathrm{CO}$ ratio for all on-road vehicles (both gasoline and diesel powered) as well as for all hours of the day (e.g., cold-start emissions from vehicles are different than running emissions). Therefore, the CO${}_{\mathrm{2}}$ on-road emissions estimated in this study are an approximation of a more complex reality.

Off-road mobile emissions

Off-road emissions include the emissions from any off-road vehicles that do not travel on designated roadways, including aircraft, all off-road engines (such as chainsaws, lawn mowers, snow blowers, and snowmobiles), and locomotives. In the APEI CO inventory, the major emission sources in the off-road sector include civil aviation (1A3a), railways (1A3c), and agriculture–forestry–fishing: off-road vehicles and other machinery (1A4c). Similar to the point sector, we found that applying a single ${\mathrm{CO}}_{\mathrm{2}}:\mathrm{CO}$ ratio to every grid cell did not produce realistic CO${}_{\mathrm{2}}$ emissions for the two major airports in the GTA, Pearson International Airport (hereafter referred to as Pearson Airport) and Billy Bishop Toronto City Airport (hereafter referred to as Billy Bishop Airport). Therefore, we used air quality assessment reports compiled for each airport (RWDI AIR Inc., 2009, 2013) to identify the geocoded location and facility-specific annual average ${\mathrm{CO}}_{\mathrm{2}}:\mathrm{CO}$ ratio. Sources of emissions from each airport include aircraft (landing and take-off cycles), auxiliary power units, ground support equipment, roadways, airside vehicles, parking lots, stationary sources, and training fires; note that emissions from aircrafts in-transit between airports, which are injected in the free troposphere, have not been included in this inventory (Moran et al., 2015; RWDI AIR Inc., 2009). Based on these two reports, we applied a ratio of 175 kt ${\mathrm{CO}}_{\mathrm{2}}/\mathrm{kt}$ CO to the grid cell containing Pearson Airport and a ratio of 20 kt ${\mathrm{CO}}_{\mathrm{2}}/\mathrm{kt}$ CO to the grid cell containing Billy Bishop Airport. All other off-road sources belonging to Ontario grid cells were scaled by a sector average ${\mathrm{CO}}_{\mathrm{2}}:\mathrm{CO}$ ratio of 7 kt ${\mathrm{CO}}_{\mathrm{2}}/\mathrm{kt}$ CO, calculated from NIR-reported Ontario total CO${}_{\mathrm{2}}$ and CO emissions. Similar to on-road emissions, there are challenges associated with using a single ${\mathrm{CO}}_{\mathrm{2}}:\mathrm{CO}$ ratio for all vehicles (both gasoline and diesel powered) as well as for all hours of the day. Therefore, the CO${}_{\mathrm{2}}$ off-road emissions estimated in this study are an approximation for a very complex sector.

Marine emissions

Commercial marine emissions include the emissions from any marine vessels travelling on the Great Lakes (quantified by Statistics Canada, Shipping in Canada) (Environment Canada, 2013). In the APEI CO inventory, the major emission source in the marine sector is national navigation (1A3d). The NIR estimates an Ontario total from this source of 729 CO${}_{\mathrm{2}}$ and 0.86 kt CO, producing a ${\mathrm{CO}}_{\mathrm{2}}:\mathrm{CO}$ ratio of 844 kt ${\mathrm{CO}}_{\mathrm{2}}/\mathrm{kt}$ CO. This ratio was applied to every marine grid cell in the domain to convert sector CO emissions to CO${}_{\mathrm{2}}$. Note that inclusion of this source sector was desirable because two of the CO${}_{\mathrm{2}}$ measurement stations considered in this study (Turkey Point and Hanlan's Point) are near-shore stations.

Residential and commercial emissions

Residential and commercial CO${}_{\mathrm{2}}$ emissions reflect on-site combustion of natural gas for electricity and heating, a source that we found was not included in the APEI CO inventory because of the high efficiency of the furnaces and resulting low CO emissions. To include the CO${}_{\mathrm{2}}$ emissions from these on-site furnaces, we used the Statistics Canada 2012 Report on Energy Supply and Demand to quantify the amount of natural gas consumed by residential and commercial buildings in Ontario, 7.9 $\times$ 10${}^{\mathrm{3}}$ gigalitres (GL) and 4.9 $\times$ 10${}^{\mathrm{3}}$ GL respectively (Statistics Canada, 2012a). The Canadian NIR estimated 1879 g CO${}_{\mathrm{2}}$ m${}^{-\mathrm{3}}$ natural gas as the CO${}_{\mathrm{2}}$ emission factor specific to the province of Ontario, based on data from a chemical analysis of representative natural-gas samples and an assumed fuel combustion efficiency of 99.5 % (Environment Canada, 2012). Using this emission factor, CO${}_{\mathrm{2}}$ emissions from residential and commercial on-site furnaces in Ontario were estimated to be 1.5 $\times$ 10${}^{\mathrm{7}}$ t and 9.2 $\times$ 10${}^{\mathrm{6}}$ t, respectively. These two emission totals were spatially allocated using a “capped-total dwelling” spatial surrogate developed by ECCC and temporally allocated using the SMOKE emissions processing system (Moran et al., 2015).

Biogenic fluxes

The net ecosystem exchange fluxes used in our simulations were provided by the land surface component of the ECMWF forecasting system, C-TESSEL (Bousetta et al., 2013). Fluxes are extracted at the highest available resolutions, 15 $\times$ 15 km and 3 h for January and February 2016 and 9 $\times$ 9 km and 3 h for March. These data are interpolated in space and time to be consistent with our model resolution. With our main priority being understanding anthropogenic emissions in the GTA, we chose to analyze a period where the biogenic CO${}_{\mathrm{2}}$ flux is minimized and therefore this paper focuses on 3 winter months in 2016, January to March inclusive.

Results and discussion

The SOCE inventory

Figure 1 displays the PanAm domain total anthropogenic CO${}_{\mathrm{2}}$ emissions estimated by the SOCE inventory for the province of Ontario portion ($\sim$ 0.02${}^{\circ }$ $\times$ 0.02${}^{\circ }$) and by the FFDAS v2 inventory (0.1${}^{\circ }$ $\times$ 0.1${}^{\circ }$) (FFDAS, 2010) for the remainder of the domain. Regions of high emissions typically correspond to population centres, for example the GTA in Ontario, Montréal and Québec City in Québec, and Chicago, Boston, and New York City (amongst others) in the USA. Emissions from highways and major roadways are only clear in the province of Ontario (at higher spatial resolution) but industrial and large-scale area sources are evident across the entire domain.

Anthropogenic CO${}_{\mathrm{2}}$ emissions for the year 2010 in the black-box area (shown in Fig. 2a) by sector. Values in parentheses indicate the percentage contribution of the sector to the total CO${}_{\mathrm{2}}$ emissions in the black-box area.

	FFDAS v2${}^{\mathrm{3}}$	EDGAR v4.2${}^{\mathrm{2}}$	SOCE CO${}_{\mathrm{2}}$
	CO${}_{\mathrm{2}}$ inventory	CO${}_{\mathrm{2}}$ inventory	inventory
Sector	(Mt CO${}_{\mathrm{2}}$ yr${}^{-\mathrm{1}}$)	(Mt CO${}_{\mathrm{2}}$ yr${}^{-\mathrm{1}}$)	(Mt CO${}_{\mathrm{2}}$ yr${}^{-\mathrm{1}}$)
Area${}^{\mathrm{1}}$	–	46 (33.9 %)	42 (43.9 %)
Point	–	46 (33.7 %)	24 (25.7 %)
Marine	–	0.1 (0.10 %)	0.1 (0.10 %)
On-road	–	41 (30.2 %)	24 (25.0 %)
Off-road	–	3 (2.2 %)	5 (5.3 %)
Total	105	136	95

${}^{\mathrm{1}}$ Area sector represents the summation of area $+$ residential $+$ commercial natural-gas combustion. ${}^{\mathrm{2}}$ The EDGAR inventory v4.2 can be found at http://edgar.jrc.ec.europa.eu. ${}^{\mathrm{3}}$ The FFDAS v2 inventory can be found at http://hpcg.purdue.edu/FFDAS/.

The total CO${}_{\mathrm{2}}$ emissions can be separated into contributions from the seven sectors in the province of Ontario described in Sect. 2.4. Figure 2 shows the anthropogenic CO${}_{\mathrm{2}}$ contributions from the area sector, residential and commercial sector, point sector, marine sector, on-road sector, and off-road sector, focused on Southern Ontario and the GTA. If we consider emissions from a domain including the area solely around the GTA (indicated by the black-box in Fig. 2a), the total CO${}_{\mathrm{2}}$ emissions estimated by the SOCE inventory is 94.8 Mt CO${}_{\mathrm{2}}$ per year (Table 2). Figure 2a and b display the CO${}_{\mathrm{2}}$ emissions from the area sector and from residential and commercial natural-gas combustion in Southern Ontario. These two sectors combined represent the largest source of CO${}_{\mathrm{2}}$ in the black-box area (41.6 Mt CO${}_{\mathrm{2}}$ yr${}^{-\mathrm{1}}$, contributing 43.9 % of the total). The majority of these emissions are concentrated in the GTA and surrounding urban areas as a result of a significant portion of the population (64 %) being reliant on natural gas for heat production (Statistics Canada, 2007, 2012a). Figure 2c represents emissions from the point sector, contributing 24.4 Mt CO${}_{\mathrm{2}}$ yr${}^{-\mathrm{1}}$, 25.7 % of the total. The largest point-source emitters are located on the western shore of Lake Ontario (Hamilton and surrounding areas) as this area is one of the most industrialized regions of the country with intensive metal production activities. Figure 2d, e, and f display CO${}_{\mathrm{2}}$ emissions from various transportation sectors, marine, on-road, and off-road respectively, which together contribute more than 30 % of total CO${}_{\mathrm{2}}$ emissions in the area within the black box. While emissions from marine activity are minimal, those from on-road and off-road sources are significant (25.0 and 5.3 %, respectively), concentrating on the major highways connecting the various population centres of the GTA to the downtown core, as well as at Pearson Airport located within the city.

Anthropogenic CO${}_{\mathrm{2}}$ emissions for a weekday in February 2010 in Southern Ontario. Emissions are estimated by the SOCE inventory for the (a) area sector, (b) sum of the residential and commercial sectors, (c) point sector, (d) marine sector, (e) on-road sector, and (f) off-road sector. Unit: log${}_{\mathrm{10}}$ (g CO${}_{\mathrm{2}}$ s grid cell${}^{-\mathrm{1}}$).

[Figure omitted. See PDF]

Comparison of the SOCE inventory with other inventories

The EDGAR v4.2 inventory estimates CO${}_{\mathrm{2}}$ emissions on an annual basis and by sector based on Selected Nomenclature for Air Pollution (SNAP) subsectors while FFDAS v2 provides hourly mean grid cell totals. Table 2 shows a comparison between the sectoral CO${}_{\mathrm{2}}$ estimates of the SOCE and EDGAR v4.2 inventories (SNAP sectors were grouped to correspond to SOCE sectors; Table S2) as well as the domain total estimated by the FFDAS v2 inventory for the area surrounding the GTA (the black-box area outlined in Fig. 2a). There is a significant discrepancy between the CO${}_{\mathrm{2}}$ emissions estimated by the SOCE and EDGAR v4.2, inventories both in the relative sectoral contributions as well as domain total (percent difference > 35 %). The largest sectoral discrepancies are in the point and the on-road sectors, where the EDGAR v4.2 inventory estimates a contribution 1.9 and 1.7 times larger than that of the SOCE inventory, respectively. Figure 3 shows a comparison of the spatial distribution of the CO${}_{\mathrm{2}}$ inventory predicted by (a) FFDAS v2, (b) EDGAR v4.2, and (c) SOCE (aggregated to 0.1${}^{\circ }$ $\times$ 0.1${}^{\circ }$ to match the resolution of EDGAR v4.2 and FFDAS v2) for the GTA area. Figure 3 reveals that the largest differences between the SOCE inventory and the EDGAR v4.2 inventory is the CO${}_{\mathrm{2}}$ emissions in the GTA; EDGAR v4.2 predicts much higher emissions in the GTA (in some grid cells, differences are an order of magnitude), particularly in the downtown core relative to the SOCE inventory.

Comparison of spatial distribution of annual CO${}_{\mathrm{2}}$ emissions inventories for the black-box area (shown in Fig. 2a) at 0.1${}^{\circ }$ $\times$ 0.1${}^{\circ }$ resolution. Panel (a) shows the FFDAS v2 inventory estimate, panel (b) shows the EDGAR v4.2 inventory estimate, and panel (c) shows the SOCE inventory estimate. Domain totals are shown on top of each panel and locations of in situ measurements of CO${}_{\mathrm{2}}$ for three stations in the Southern Ontario Greenhouse Gas Network are shown in panel (a) (Downsview is a square, Hanlan's Point is a triangle, TAO is a pentagon). The other two stations, Egbert and Turkey Point, are located outside this area.

[Figure omitted. See PDF]

Although there is no sectoral breakdown in the FFDAS v2 inventory, the domain total around the GTA can be compared to that of the SOCE inventory, Table 2. Unlike the comparison with the EDGAR v4.2 inventory, there is a closer agreement between the FFDAS v2 inventory and the SOCE inventory (difference of $\sim$ 10 %). The comparison plots in Fig. 3 show a good agreement of the spatial variability of emissions in the GTA between the FFDAS v2 and SOCE inventories; however, the gradient between urban and rural areas is not as sharp in the SOCE inventory as it is in the FFDAS v2 inventory. Furthermore, emissions along the western shore of Lake Ontario (Hamilton and the surrounding areas) are predicted to be larger in the SOCE inventory relative to FFDAS v2. The FFDAS v2 inventory was interpolated to 0.02${}^{\circ }$ $\times$ 0.02${}^{\circ }$ using a mass conservative interpolation scheme to allow the production of a difference plot of the two inventories, SOCE minus FFDAS v2, shown in Fig. S1. The difference plot reveals the largest divergence between the inventories occurs in the GTA and Ottawa, with the FFDAS v2 inventory estimating > 1000 g CO${}_{\mathrm{2}}$ s${}^{-\mathrm{1}}$ ($\sim$ 30 kt CO${}_{\mathrm{2}}$ yr${}^{-\mathrm{1}}$) more than the SOCE inventory in some grid cells. In addition to similar spatial variability, the FFDAS v2 and SOCE inventories also have similar temporal variability. Figure S2 shows the diurnal profile of estimated emissions from January to March for both the FFDAS v2 and SOCE inventories for the black-box area in the PanAm domain. Both inventories allocate the highest emissions between 08:00 and 18:00 EST and the lowest emission between 00:00 and 05:00, but the amplitude of the diurnal cycle is higher in SOCE, and emissions in the morning are as high as in the afternoon. FFDAS allocates a relatively larger proportion of the emissions to the 15:00–19:00 period.

Preliminary analyses using the SOCE, FFDAS v2 and EDGAR v4.2 inventories with FLEXPART

To investigate the impact of the different inventories on ambient mixing ratios, preliminary analyses were run with footprints generated for every third hour of the day (e.g., 00:00, 03:00, 06:00, 09:00) by the FLEXPART Lagrangian particle dispersion model (Stohl et al., 2005) driven by GEM meteorology for two sites, Downsview and TAO. Footprint areas were multiplied by the inventory estimates to arrive at mixing ratio enhancements and then compared against the measured CO${}_{\mathrm{2}}$ gradient between the Downsview and TAO stations for the year 2014. Gradients were used to capture the CO${}_{\mathrm{2}}$ mixing ratios in the downtown core of the city (since Downsview and TAO are situated just north and south of downtown Toronto, respectively). Observed gradients ranged from $+$20 to $-$10 ppm. Figure S3 displays the measured and modelled CO${}_{\mathrm{2}}$ gradients. These results show that when the EDGAR v4.2 inventory was used, simulated CO${}_{\mathrm{2}}$ gradients were consistently overestimated by $\sim$ 10–60 ppm relative to observations. Conversely, when the SOCE inventory was used, a higher level of agreement was obtained between simulated mixing ratios and measurements; however, none of the model simulations were able to capture times when the gradient was negative (CO${}_{\mathrm{2},\text{TAO}}$ > CO${}_{\mathrm{2},\text{Downsview}})$, an effect we believe to be due to the TAO inlet being $\sim$ 60 m above ground level and surrounded by many high-rise buildings, creating canyon flows and turbulence which are not properly accounted for in GEM at this resolution. These factors contributed to the decommissioning of TAO in January 2016. The poor performance of our model system when using the EDGAR v4.2 inventory to simulate CO${}_{\mathrm{2}}$ mixing ratios was also found by a study quantifying on-road CO${}_{\mathrm{2}}$ emissions in Massachusetts, USA (Gately et al., 2013). In this study, EDGAR emission estimates were found to be significantly larger than any other inventory by as much as 9.3 million tons, or > 33 %. The difference in estimates between the EDGAR v4.2 and the SOCE inventories is likely explained by their underlying differences in methodology. Being a global product and not specifically designed for mesoscale applications, the EDGAR v4.2 inventory estimates CO${}_{\mathrm{2}}$ emissions based on country-specific activity data and emission factors, but the spatial proxies used to disaggregate the data are not always optimal. A study performed by McDonald et al. (2014) showed that the use of road density as a spatial proxy for vehicle emissions in EDGAR v4.2 causes an overestimation of emissions in population centres (McDonald et al., 2014). Given the much larger emission estimates for on-road CO${}_{\mathrm{2}}$ from EDGAR v4.2 (Table 2), this also seems to be an issue in the GTA. Based on this large discrepancy, the EDGAR v4.2 inventory was not further used in this study and we focussed on the inventories developed for regional-scale studies.

When similar preliminary analyses were run with FLEXPART footprints using the FFDAS v2 inventory, Fig. S3, good agreement was observed with CO${}_{\mathrm{2}}$ gradients measured between the Downsview and TAO stations. We are confident that the enhanced measurement agreement between the FFDAS v2 and SOCE relative to EDGAR v4.2 is due to improved methodology; spatial allocation of emissions in FFDAS v2 is achieved through the use of satellite observations of nightlights from human settlements from the US Defense Meteorological Satellite Program Operational Linescan System (DMSP-OLS).

Beyond the differences in methodology for estimating and allocating emissions, it is important to note that the emissions reported in Table 2 by the FFDAS v2, SOCE, and EDGAR v4.2 inventories also fundamentally differ in time period quantified. The emissions reported for both FFDAS v2 and the SOCE are based on emissions from 3 winter months (January–March 2010) extrapolated for the entire year. However, emissions from EDGAR v4.2 are annual averages of all 12 months of 2010. Since CO${}_{\mathrm{2}}$ emissions in the GTA are higher in the winter months relative to the summer months because of increased building and home heating, it is likely that the average annual estimates of SOCE and FFDAS v2 are slightly overestimated. This does not affect the relative agreement between SOCE and FFDAS v2 but it does further increase the divergence between the EDGAR v4.2 and SOCE and FFDAS v2 inventories. Following this and the improved agreement with observations, the FFDAS v2 inventory was used with the SOCE inventory for all subsequent modelling analyses.

Time series of measured (blue) and modelled February afternoon (12:00–16:00 EST) CO${}_{\mathrm{2}}$ mixing ratios for the four sites used in this study. The red and gold markers are the modelled mixing ratios when using the SOCE CO${}_{\mathrm{2}}$ inventory and the FFDAS v2 inventory, respectively.

[Figure omitted. See PDF]

Simulation of CO${}_{\mathrm{2}}$ mixing ratios in the Greater Toronto Area

We used the GEM-MACH chemistry–transport model and the SOCE and FFDAS v2 inventories to simulate hourly CO${}_{\mathrm{2}}$ mixing ratios in the PanAm domain. The model framework was evaluated for a continuous 3-month period, January–March 2016, using four sampling locations in the GTA (Fig. 1; note that measurements for the Hanlan's Point site were not available until 14 January 2016). Figure 4 displays afternoon (12:00–16:00 EST) measured and simulated CO${}_{\mathrm{2}}$ mixing ratios produced with the SOCE and FFDAS v2 inventories for the two emissions scenarios described in Sect. 2.3 for the month of February (Figs. S4 and S5 show the same figure for other months). We chose to present only afternoon data as this is the time of day when the mixed layer is likely to be the most well-developed; nighttime and morning data showed largest variations in observations as a result of the shallow boundary layer causing surface emissions to accumulate within the lowest atmospheric layers (Breon et al., 2015; Chan et al., 2008; Gerbig et al., 2008). During the night, atmospheric mixing ratios are most sensitive to vertical mixing, an atmospheric process that is difficult to model for stable boundary layers.

The time series comparisons at all four sites demonstrate the model's general ability to capture variability in observations of CO${}_{\mathrm{2}}$, albeit with better skill for the Downsview and Egbert sites (this is particularly clear when we look at model–measurement difference plots; Fig. S6). The model is able to capture many extreme events of mixing ratio increases and decreases, such as 11–14 February 2016 at the Downsview site; however, some short time periods are poorly simulated, such as 21–23 January 2016 at Hanlan's Point, when the model significantly overestimated measured CO${}_{\mathrm{2}}$. Generally, mixing ratios simulated by the FFDAS v2 inventory are similar or larger than those produced when the SOCE inventory is used, with differences most noticeable at the Downsview and Hanlan's Point sites. This was expected as the difference plot shown in Fig. S1 reveals that the SOCE and FFDAS v2 inventories diverge the most in the GTA (where the Downsview and Hanlan's Point sites are located) and are more similar in rural areas (where the Turkey Point and Egbert sites are located).

Time series of mean measured (blue) and modelled diurnal CO${}_{\mathrm{2}}$ mixing ratios at the four sites considered in this study for January–March 2016. The red and gold markers are the modelled diurnal mixing ratios when using the SOCE CO${}_{\mathrm{2}}$ inventory and the FFDAS v2 inventory, respectively. Note the difference in scale for urban and rural sites. Error bars represent standard error of the mean.

[Figure omitted. See PDF]

Measured CO${}_{\mathrm{2}}$ mixing ratios have a typical diurnal pattern, in which mixing ratios are higher at night and lower during the day, despite higher emissions during the day. This results from the daily cycle of the mixed layer, which is shallow at night due to thermal stratification and deepens during the day due to solar heating of the surface. Figure 5 displays the measured and modelled mean diurnal profile of CO${}_{\mathrm{2}}$ at the four sites in our network using data from January to March 2016 (note difference in $y$-axis scale for urban vs. rural sites). At all four sites, the shapes of the modelled and measured mixing ratios throughout the day agree very well, suggesting that the GEM meteorology in our framework is capturing the diurnal variation in emissions and the boundary layer evolution. At the Downsview site, there is a very strong agreement between the modelled and measured diurnal profiles when using the SOCE inventory, whereas the FFDAS v2 simulated profile largely overestimates mixing ratios, particularly at nighttime. This is consistent with the FFDAS inventory having larger emissions than the SOCE inventory during the night (Fig. S2). At the Hanlan's Point site, a difference of $\sim$ 5 ppm CO${}_{\mathrm{2}}$ is observed when using the SOCE inventory relative to measurements; however, similar to the Downsview site, the FFDAS v2 simulated profile has a larger difference of $\sim$ 10 ppm CO${}_{\mathrm{2}}$. At both the Egbert and Turkey Point sites, the use of both inventories similarly overestimates the diurnal pattern of CO${}_{\mathrm{2}}$ mixing ratios by $\sim$ 3–5 ppm, again likely a result of the similarities of these two inventories at these sites (Fig. S1). At all four sites, it is possible that some of the biases that are observed in simulated CO${}_{\mathrm{2}}$ mixing ratios may arise from inaccuracies in the boundary CO${}_{\mathrm{2}}$ provided by MACC; this aspect was not, however, further explored in this study.

Scatter plot of the modelled and measured afternoon (12:00–16:00 EST) CO${}_{\mathrm{2}}$ mixing ratios from January to March 2016 at the four monitoring stations used in this study. The top and bottom panels show measurement–model correlation when the SOCE inventory and the FFDAS v2 inventory were used, respectively. The model vs. measurement correlation coefficient ($R$), root-mean-square error (RMSE), and mean bias (MB) (unit: ppm) are provided within each panel. Solid lines are the standard major axis regression lines and dashed lines are $\mathrm{1}:\mathrm{1}$ lines shown for reference.

[Figure omitted. See PDF]

Quantifying model–measurement agreement

Figure 6 shows scatter plots of afternoon (12:00–16:00 EST) modelled versus measured CO${}_{\mathrm{2}}$ mixing ratios from January to March 2016 at the four sites used in this study. The top row shows the correlation between measured and modelled mixing ratios using the SOCE inventory and the bottom row shows the correlation using the FFDAS v2 inventory. It is immediately clear that there is a stronger model–measurement correlation at the Downsview and Egbert sites ($R>\mathrm{0.75}$) relative to that of Hanlan's Point or Turkey Point ($R<\mathrm{0.53}$). The difficulty with accurately simulating CO${}_{\mathrm{2}}$ mixing ratios at Hanlan's Point and Turkey Point may arise from their proximity to shorelines, Hanlan's Point to Lake Ontario, and Turkey Point to Lake Erie (see Fig. 1). Differential heating of land versus water near these lakes creates pressure gradients driving unique circulation patterns (Burrows, 1991; Sills et al., 2011). These circulation patterns are difficult for models to capture and therefore may contribute to the relatively poor correlation observed at Hanlan's Point and Turkey Point.

It is also clear from Fig. 6 that simulating CO${}_{\mathrm{2}}$ mixing ratios at the Egbert and Turkey Point sites using either the FFDAS v2 or the SOCE inventory results in similar performance, likely because the emissions estimated by the two inventories are similar in the vicinity of these two rural sites (see also Fig. 5). However, at both the Downsview and Hanlan's Point sites, using the SOCE inventory provided a slightly higher correlation and reduced root-mean-square error (RMSE) and mean bias relative to using the FFDAS v2 inventory. The improvement by using the SOCE inventory is likely a result of both the improved spatial resolution (2.5 km vs. 10 km), and therefore more accurate allocation of emissions to grid cells and also a better estimation of emission magnitudes, as large differences are shown in Figs. 3 and S1.

Modelled sectoral percent contributions to diurnal local CO${}_{\mathrm{2}}$ enhancement for February 2016 at Downsview averaged by day of week. Note: area $=$ area $+$ residential natural-gas combustion $+$ commercial natural-gas combustion. (Time zone is EST).

[Figure omitted. See PDF]

Sectoral contributions to simulated CO${}_{\mathrm{2}}$ mixing ratios

One of the major advantages of the SOCE inventory over the FFDAS v2 inventory is the availability of sectoral emission estimates. Figure 7 displays the sectoral percent contributions to diurnal CO${}_{\mathrm{2}}$ mixing ratio enhancements (calculated as local CO${}_{\mathrm{2}}$ mixing ratios above the MACC-estimated background) for the Downsview station in February 2016 averaged by the day of week (Figs. S7 and S8 displays the same for other months). This figure clearly demonstrates the importance of area emissions (defined here as the sum of the area $+$ residential natural-gas combustion $+$ commercial natural-gas combustion) to simulated CO${}_{\mathrm{2}}$ mixing ratios, reaching $\sim$ 80 % contribution in the early morning and late evening, consistent with times when emissions from home heating are the dominant source of CO${}_{\mathrm{2}}$. Contributions from area emissions decrease to $\sim$ 35 % midday, which coincides with when emissions from other sources, such as on-road, gain importance. In the midday, emissions from the on-road sector can contribute $\sim$ 50 %, which is consistent with transportation patterns of the times when the population is travelling to and from work and other activities. The relative contributions to CO${}_{\mathrm{2}}$ mixing ratios from point-source emissions are quite variable during the course of a day and week, but they generally seem to increase in the early morning and evening and can contribute a significant portion of total CO${}_{\mathrm{2}}$ emissions (up to $\sim$ 20 %). Figure 7 indicates that biogenic sources of CO${}_{\mathrm{2}}$ play a negligible role during January–March in the GTA (the biogenic line is not visible because it is located on the zero line underneath the marine line). Recent studies, however, have shown the importance of the biospheric contribution (up to $\sim$ 132–308 g CO${}_{\mathrm{2}}$ km${}^{-\mathrm{2}}$ s${}^{-\mathrm{1}})$ to measured CO${}_{\mathrm{2}}$ in urban environments during the growing season (Decina et al., 2016). Therefore, this finding supports the importance of modelling CO${}_{\mathrm{2}}$ in the wintertime in cities like the GTA to avoid complications associated with biospheric contributions. The new ability to understand the sectoral contributions to CO${}_{\mathrm{2}}$ mixing ratios in the GTA and Southern Ontario has implications from a policy perspective; with recent initiatives to curb CO${}_{\mathrm{2}}$ emissions, understanding from which sector the CO${}_{\mathrm{2}}$ is being emitted could be useful to assess how effective applied mitigation efforts have been or where to target future efforts. These efforts could be complemented by running simulations with additional tracers, such as CO, nitrogen oxides (NO${}_x$), or stable carbon isotopes (${}^{\mathrm{12}}$C and ${}^{\mathrm{13}}$C) to gain further insight.

Conclusions

We presented the SOCE inventory for Southern Ontario and the GTA, the first, to our knowledge, high-resolution CO${}_{\mathrm{2}}$ inventory for Southern Ontario and for a Canadian metropolitan region. The SOCE inventory provides CO${}_{\mathrm{2}}$ emissions estimates at 2.5 km $\times$ 2.5 km spatial and hourly temporal resolution for seven sectors: area, residential natural-gas combustion, commercial natural-gas combustion, point, marine, on-road, and off-road. When compared against two existing CO${}_{\mathrm{2}}$ inventories available for Southern Ontario, the EDGAR v4.2 and the FFDAS v2 inventories, using FLEXPART footprints, the SOCE inventory had improved model–measurement agreement; FFDAS v2 agreed well with in situ measurements, but the EDGAR v4.2 inventory systematically overestimated mixing ratios. We developed a model framework using the GEM-MACH chemistry–transport model on a high-resolution 2.5 km $\times$ 2.5 km grid coupled to the SOCE and FFDAS v2 inventories for anthropogenic CO${}_{\mathrm{2}}$ emissions and C-TESSEL for biogenic CO${}_{\mathrm{2}}$ fluxes. We compared output simulations to observations made at four stations across Southern Ontario and for 3 winter months, January–March 2016. Model–measurement agreement was strong in the afternoon using both anthropogenic inventories, particularly at the Downsview and Egbert sites. Difficulty in capturing mixing ratios at the Hanlan's Point and Turkey Point sites was hypothesized to be a result of their close proximity to shorelines (Lake Ontario and Lake Erie, respectively) and the model's inability to capture the unique circulation patterns that occur in those environments. Generally, across all stations and months, simulations using the SOCE inventory resulted in higher model–measurement agreement than those using the FFDAS v2 inventory, quantified using $R$, RMSE, and mean bias. In addition to improved agreement, the primary advantage of the SOCE inventory over the FFDAS v2 inventory is the sectoral breakdown of emissions; using average day-of-week diurnal mixing ratio enhancements, we were able to demonstrate that emissions from area sources can contribute > 80 % of CO${}_{\mathrm{2}}$ mixing ratio enhancements in the early morning and evening with on-road sources contributing > 50 % midday. The applications of the SOCE inventory will likely show future utility in understanding the impacts of CO${}_{\mathrm{2}}$ reduction efforts in Southern Ontario and identify target areas requiring further improvement.

The data can be found in Pugliese (2018; doi:10.5683/SP/GOQGHD).

The Supplement related to this article is available online at https://doi.org/10.5194/acp-18-3387-2018-supplement.

The SOCE inventory was prepared by SCP, with critical input from FV and JM. The CO inventory, which the SOCE inventory is based upon, and the CO${}_{\mathrm{2}}$ emissions fields for residential and commercial natural-gas combustion were provided by MDM, JZ, and QZ. The GEM-MACH modelling analyses were performed by SR and CS. The ambient CO${}_{\mathrm{2}}$ data were collected by DW and his team at Environment and Climate Change Canada. The MACC and C-TESSEL products used in our model simulations were provided by GB. The data were analyzed and interpreted for publication by SCP. This paper was written by SCP, with critical input from JM, FV, and MM.

The authors declare that they have no conflict of interest.

Acknowledgements

The authors are thankful to Robert Kessler, Michelle Ernst, Lauriant Giroux, Senen Racki and Lin Huang for their efforts collecting the ${}^{\mathrm{12}}$CO${}_{\mathrm{2}}$ and ${}^{\mathrm{13}}$CO${}_{\mathrm{2}}$ measurements at Environment and Climate Change Canada. They would also like to thank Elton Chan for providing the FLEXPART footprints and for Pegah Baratzadeh for her help creating the SOCE inventory. Edited by: Robert Harley Reviewed by: two anonymous referees