CO₂ fluxes of vegetation in the Greenbelt of Ontario and increased net ecosystem emissions associated with its removal

Natural ecosystems are an important sink for carbon dioxide (CO₂), which is one of the main drivers of global climate change (Denman et al., 2007; Forster et al., 2007). Worldwide, the terrestrial biosphere absorbs approximately 31% of CO₂ emitted from human activities (Denman et al., 2007; Canadell et al., 2021). Therefore, preserving natural ecosystems has been identified as an important way to mitigate climate change with the additional benefits of preserving biodiversity and ecosystem health (Dinerstein et al., 2019; Convention on Biological Diversity, 2022). Both the Government of Canada and the City of Toronto have committed to reducing their greenhouse gas emissions to net zero by 2050 and 2040, respectively (City of Toronto, Live Green Toronto, 2021; Government of Canada, SC, 2021; Environment and Climate Change Canada, 2022). These plans, however, rely on vegetation to sequester CO₂ (City of Toronto, Live Green Toronto, 2021; Environment and Climate Change Canada, 2022).

Regional CO₂ fluxes are composed of both anthropogenic and biogenic fluxes. Therefore, it is important to assess both when considering a region’s carbon budget (Zhao et al., 2010; Hardiman et al., 2017; Wei et al., 2022). Anthropogenic emissions are frequently estimated using emission inventories based on various factors including statistics, such as population, energy use, and agricultural activity, along with emission factors that determine the amount of emissions associated with these activities (Janssens-Maenhout et al., 2019; Shekarrizfard et al., 2022). Vegetation fluxes in urban regions, however, have been more difficult to estimate.

In 2005, the Greenbelt was established in Southern Ontario, surrounding the Greater Toronto and Hamilton Area (GTHA), to limit urban sprawl and “provide permanent protection” to the surrounding croplands and natural landscape and their ecosystem functions, including carbon sequestration (Ministry of Municipal Affairs and Housing, 2005; Ministry of Municipal Affairs, 2017). However, in December 2022, the proposed changes were made to the Greenbelt of Ontario despite strong opposition (Ministry of Municipal Affairs and Housing, 2022). These changes included the removal of protections from 2,995 ha (7,400 acres) of vegetated land from the inner edge of the Greenbelt in order to create 50,000 new homes (Ministry of Municipal Affairs and Housing, 2022). Although these changes have recently been retracted (Ministry of Municipal Affairs and Housing, 2023b), it is still important that we understand the significance of natural vegetation and croplands in terms of the local CO₂ budget and the impact of land use changes on the ecosystem services they provide. This article focuses on the net amount of CO₂ absorbed by vegetation in the Greenbelt and how changes to the Greenbelt may affect the amount of CO₂ sequestered by these ecosystems.

Ecosystems absorb and emit CO₂ during photosynthesis and respiration, respectively. Gross primary productivity (GPP) quantifies the amount of CO₂ absorbed during photosynthesis. Ecosystem respiration (R_eco) is the amount of CO₂ emitted by plants and soils. R_eco can be further divided into the amount of CO₂ respired by vegetation, including stems, foliage, and roots (autotrophic respiration), and the amount of CO₂ released during the decomposition of organic material both above ground and in the soil (heterotrophic respiration). The net amount of CO₂ exchanged is, therefore, the difference between these 2 fluxes; this is known as net ecosystem exchange (NEE = R_eco − GPP; Janssens et al., 2001). Note that a negative NEE value represents a net removal of CO₂ from the atmosphere.

One of the most promising methods to estimate GPP on regional scales is by using measurements of solar-induced chlorophyll fluorescence (SIF) (Frankenberg et al., 2011b; Mohammed et al., 2019; Magney et al., 2020). Plants absorb solar radiation and use this energy for photosynthesis; however, not all of the absorbed energy is used. Some of the remaining energy is emitted as light in the red and near-infrared (650–800 nm) portions of the spectrum. This signal is known as SIF and can be detected using satellite instruments with fine spectral resolution (Frankenberg et al., 2011a; Mohammed et al., 2019). The relationship between SIF and CO₂ absorption by plants was first discovered by Kautsky and Hirsch (1931), and since then, it has been shown that on regional scales, SIF is linearly related to GPP (Frankenberg et al., 2011b; Wood et al., 2017; Magney et al., 2019; Turner et al., 2020). Several studies have highlighted the nonlinear relationship between SIF and photosynthesis measured at the leaf and canopy levels or on short timescales (He et al., 2020; Marrs et al., 2020). This nonlinear behavior has been ascribed to the nonlinear relationship between SIF yield and photochemistry due to other pathways through which plants can dissipate excess light energy, which may occur due to stress (Marrs et al., 2020) or light-saturation (He et al., 2020). However, these nonlinear behaviors are largely removed when averaging over larger spatial scales or longer temporal scales and through the reabsorption of emitted photons within canopies (He et al., 2020; Magney et al., 2020). The functional relationship between SIF and GPP at regional scales is not fully understood and the linear relationship between SIF and GPP is likely due to common drivers, such as the amount of light absorbed and photochemical regulation (Magney et al., 2020). In addition, while SIF can be measured by satellite instruments when there is minimal to moderate cloud cover, it is not measured under high cloud cover (Köhler et al., 2018). Cheng et al. (2022) suggest that filtering out measurements during moderate and high cloud cover may cause seasonal SIF estimates under frequent cloud cover to be overestimated by up to 25%. However, diffuse, cloudy conditions can also result in higher light use efficiency by vegetation (Law et al., 2002; Knohl and Baldocchi, 2008; Dengel and Grace, 2010; Urban et al., 2012).

Although SIF can be used to estimate GPP, what we are most interested in for this work is the net biogenic CO₂ fluxes (NEE). Both R_eco and GPP can be estimated using empirical vegetation models. These models use statistically derived relationships, including, for example, the linear relationship between SIF and GPP to estimate ecosystem fluxes (Wu et al., 2021). There also exist more complicated process-based models, which use theory-based mathematical descriptions of physical processes rather than observed relationships (Schaefer et al., 2008; Adams et al., 2013). Process-based models have many advantages over empirical models, including more thorough descriptions of the processes and sources behind biogenic fluxes. For example, many process-based models take into account pools of accumulated carbon both above ground as debris and in the soil (Schaefer et al., 2008), while this is typically not accounted for in empirical models (Gourdji et al., 2022). However, these process-based models tend to require more site-specific data, such as soil texture and soil carbon content (Potter et al., 1993; Schaefer et al., 2008), which is often only available at select locations or at coarse resolutions rather than across entire regions (Adams et al., 2013). By using an empirical model, we can use more widely available data including meteorological and satellite-based observations such as SIF to estimate the fluxes of CO₂ to and from vegetation on local and regional scales (Hardiman et al., 2017; Wu et al., 2021).

The carbon sequestration of the Greenbelt of Ontario has been studied before, but previous estimates of biogenic CO₂ fluxes were based on land cover data and approximated stand age alone (Tomalty and Fair, 2012). This work provides a more up-to-date, data-driven approach using a modified version of the SIF for Modeling Urban biogenic Fluxes (SMUrF) model, which takes into account important factors including air and soil temperatures, impervious surface area, and land cover type as well as direct measurements of SIF, a by-product of photosynthesis (Wu et al., 2021). We use this modified version of the SMUrF vegetation model to estimate fluxes of CO₂ to and from vegetation in the Greenbelt of Ontario. To address recent proposals for the expansion of urban areas into the Greenbelt, we seek to estimate the differences in ecosystem fluxes associated with the removal of sections of the Greenbelt and their replacement with various urban intensities. In doing this work, we aim to determine the importance of the Greenbelt in the local carbon budget, which we will highlight by comparing the magnitude of the Greenbelt’s biogenic CO₂ fluxes to inventory-estimated anthropogenic emissions from the GTHA.

To estimate the biogenic fluxes of CO₂, we use a modified version of the SMUrF model (Wu et al., 2021), adjusted to use downscaled SIF data from the Tropospheric Monitoring Instrument (TROPOMI, Köhler et al., 2018). The original SMUrF model developed by Wu et al. uses the Contiguous SIF (CSIF) product. The CSIF product uses a neural network trained on SIF from the Orbiting Carbon Observatory version 2 (OCO-2) and surface reflectance from the Moderate Resolution Imaging Spectrometer (MODIS) (Zhang et al., 2018). The resolution of CSIF and the subsequent resolution of the SMUrF model output is 0.05 × 0.05° (roughly 5 × 5 km² at 43°N, Zhang et al., 2018; Wu et al., 2021). This resolution, however, is moderately coarse for our study as sections of the Greenbelt of Ontario, including many of those proposed for removal, have widths on the order of hundreds of meters (Ministry of Municipal Affairs and Housing, 2023a). In addition, the CSIF data were trained on SIF measured using OCO-2’s nadir observation mode (Zhang et al., 2018), which does not pass over the GTHA (Haynes, n.d.) and thus may not properly represent SIF in this region. Therefore, to increase the spatial resolution and to directly use SIF measurements over the region, we modified the SMUrF model to use SIF measured from TROPOMI (Köhler et al., 2018), downscaled to a 500 m × 500 m resolution using the Near Infrared Reflectance of Vegetation Index (NIR_v), which we calculate from MODIS reflectance data (Schaaf and Wang, 2015), following the method described in Turner et al. (2020). When the downscaled TROPOMI SIF data were missing or had large fractional uncertainty, we replaced the downscaled TROPOMI SIF with the CSIF product, downscaled using MODIS NIR_v following the same method described above. Lastly, vegetation largely stops photosynthesizing in the winter months either by shedding its leaves or by downregulating photosynthesis in the cold season (Chang et al., 2021). Therefore, to avoid noise in our model while vegetation is dormant and photosynthesis rates are negligible, we set SIF to 0 mW m⁻² nm⁻¹ sr⁻¹ during winter. The beginning and end of the growing season were estimated as the dates before and after which GPP was approximately 0 µmol m⁻² s⁻¹ at 3 eddy-covariance flux towers in the region (Arain, 2018a, 2018b; Staebler, 2021).

The SMUrF model converts these SIF products to GPP using land cover-specific empirical linear relationships between SIF and GPP derived from CO₂ flux observations at 86 eddy-covariance towers globally (Wu et al., 2021). Respiration is estimated using land cover-specific neural networks that depend on soil and air temperatures, from the European Centre for Medium-Range Weather Forecasts' Reanalysis version 5 product (Copernicus Climate Change Service, 2017), as well as GPP (Wu et al., 2021). The estimated ecosystem respiration was further amended by splitting it equally into autotrophic and heterotrophic respiration and reducing the soil respiration using each pixel’s impervious surface fraction following the method described in Hardiman et al. (2017) and Winbourne et al. (2022). Impervious surface fraction was determined using a combination of the City of Toronto’s (2019) Impermeable Surface dataset, the Global Man-made Impervious Surface dataset (Brown de Colstoun et al., 2017), the Annual Crop Inventory (Agriculture and Agri-Food Canada, 2020), and the Southern Ontario Land Resource Information System (Ontario Ministry of Natural Resources and Forestry, 2019) (see Section S1, Figure S1, and Tables S1–S3 of Supplemental Material for details). Our modifications to the SMUrF model improved the spatial resolution from 0.05° × 0.05° to 500 m × 500 m. Uncertainties attributed to the 8-day and daily estimates of GPP and R_eco, respectively, from the SMUrF model are given by the coefficients of variation of the modeled fluxes compared to eddy-covariance flux data (Wu et al., 2021). Uncertainty in NEE is estimated by adding the errors in R_eco and GPP in quadrature; this results in a median error of approximately 13% per pixel for a given 8-day period.

To verify the validity of the updated SMUrF model’s biogenic flux estimates, we compare them to fluxes from 3 ground-based eddy-covariance flux towers in the region (Arain, 2018a, 2018b; Staebler, 2021). We also compare the SMUrF vegetation fluxes to those estimated from another model: the most recent version of the Vegetation Photosynthesis and Respiration Model (VPRM, Gourdji et al., 2022) to which we also apply the urban adjustments outlined in Hardiman et al. (2017) and Winbourne et al. (2022). We find the updated SMUrF model’s 8-day NEE agrees well with the eddy-covariance flux towers as well as with VPRM’s estimates over a subsection of the Greenbelt in 2019 (Figure S2). Although these agree well at 8-day timescales, summing to compute annual NEE is most relevant here. Therefore, to estimate a more realistic uncertainty for the annual sums from the SMUrF model, we take the mean percentage difference between the annual sums of NEE estimated by SMUrF and those of the eddy-covariance flux towers (Arain, 2018a, 2018b; Staebler, 2021). That is, we estimate the uncertainty of the annual net biogenic fluxes as 45% (for more details, see Section S3 and Table S4 of the Supplemental Material). We also compare our results to the annual sum of NEE estimated by VPRM over a subsection of the Greenbelt in 2019; this results in a percentage difference of 36% between NEE estimated by the 2 models (Figure S2).

Since the SMUrF model agrees fairly well with both the eddy-covariance flux towers and the VPRM, it was applied to areas within the Greenbelt, which were identified using shapefiles provided by the Ontario Ministry of Municipal Affairs and Housing (2023a; personal communication, 14/02/2023). SMUrF was run for 3 years from 2018 through 2020. Daily R_eco was averaged to 8-day resolution to match that of GPP and then summed over the year to obtain an annual NEE. We then averaged the 3 years to identify the average yearly fluxes of CO₂ from vegetation in the Greenbelt, as well as the regions proposed for removal (Figure 2). Variability in the multiyear average, ΔNEE, is estimated as half of the range of yearly values, including the uncertainty of each year, δNEE, between all 3 years. That is,

where NEE_max/min and δNEE_max/min are the maximum and minimum yearly NEE between 2018 and 2020 and their associated uncertainties, respectively.

We identify the average NEE of different land cover types in the Greenbelt using the MODIS MCD12Q1 Land Cover data (Friedl and Sulla-Menashe, 2019). This allows us to determine how the ecosystem fluxes from the areas proposed for removal from the Greenbelt, which are predominantly croplands, might compare to fluxes from other regions in the Greenbelt with different land cover types.

To compute the differences in cumulative fluxes of CO₂ associated with the removal of the designated Greenbelt areas during the construction of the proposed neighborhoods, we require an estimate of the typical duration of construction of new residential areas. Using Google Earth Pro historical imagery over the construction of 4 housing developments in the GTHA, we estimate that the construction of housing developments typically takes approximately 5 years. The Google Earth imagery also suggests that portions of these housing developments are completed with new vegetation in place before other sections have even begun construction (Google, 2022). Based on these observations, it appears that roughly 50% of the areas are completed after about 3 years. We therefore assume that for the first 3 years, the entire area will only have heterotrophic soil respiration and there will be no sequestration. We calculate a rough estimate of the soil CO₂ respiration during this period by running the SMUrF model and setting photosynthetic fluxes (GPP) and autotrophic respiration to zero, thus computing only the heterotrophic portion of the respiration (Panel 2 of Figure 1). The development proposal for the 2,995-ha section of the Greenbelt to be removed specifies that these areas will be replaced by roughly 50,000 homes resulting in a housing density of roughly 16.7 units per hectare (Ministry of Municipal Affairs and Housing, 2022). This dwelling density matches that of typical detached single-family homes (Toronto City Planning, 2021). We identified a region elsewhere in the GTHA, which has detached single-family homes that was, roughly 20 years ago, predominantly croplands before its gradual development into neighborhoods (Google, 2022) and, using the SMUrF model, estimated its current average net annual biogenic fluxes of CO₂. Thus, for the remaining 2 years of construction, we assume that roughly 50% of these areas will only have heterotrophic soil respiration, while the remainder will have the same net ecosystem fluxes as this low-density neighborhood (Panel 3 of Figure 1). After construction, we assume fluxes will be the same as that of the low-density neighborhood (Panel 4 of Figure 1). Finally, we use the SMUrF model to estimate biogenic CO₂ fluxes in the downtown core of Toronto. We use the detached residential neighborhood and downtown Toronto as proxies for the biogenic fluxes in suburban and urban areas, respectively. Thus, by replacing the average fluxes to and from vegetation in the Greenbelt with the average fluxes in urban and suburban areas, we estimate the net difference in biogenic CO₂ sequestration if the Greenbelt were to be converted to each of these land use types.

To put these biogenic fluxes into context, we frame them in terms of local anthropogenic emissions. We do this by comparing the Greenbelt’s annual average biogenic CO₂ fluxes to Scopes 1 and 2 anthropogenic emissions estimated from The Atmospheric Fund for the GTHA (Shekarrizfard and Sotes, 2021a, 2021b; Shekarrizfard et al., 2022). Scope 1 regional anthropogenic emissions encompass emissions associated with activity within the region’s boundaries including ground transportation, fossil fuel combustion, agriculture, industry, and the disposal of waste and wastewater (Chen et al., 2020; Shekarrizfard and Sotes, 2021a). Scope 2 emissions account for emissions related to the amount of electricity used by the region, regardless of whether the electricity is produced inside or outside of the region’s boundaries (Chen et al., 2020; Shekarrizfard and Sotes, 2021a). Finally, Scope 3 emissions, which are not captured in The Atmospheric Fund’s inventory (Shekarrizfard and Sotes, 2021a, 2021b), refer to emissions associated with the region but which take place outside of its boundaries. This includes, for example, emissions from waste produced inside the region’s boundaries but treated outside the region, transportation outside of the boundary, and life-cycle emissions associated with production that occurs outside of the region’s boundaries (Chen et al., 2020; Shekarrizfard and Sotes, 2021a).

The Greenbelt of Ontario, illustrated in Figure 2, is a swath of croplands and natural vegetation surrounding the GTHA. Here, we present the CO₂ fluxes we calculate using the modified SMUrF model for the entire Greenbelt and estimate the change in biogenic fluxes in areas of the Greenbelt that were considered for housing construction. We then estimate the carbon sequestration impacts of potential further development and finally, compare these fluxes to reported anthropogenic emissions.

Applying the SMUrF model to the entire Greenbelt, with the exception of the most northwestern portion along Lake Huron, we determine the average annual net CO₂ absorbed by its vegetation for 2018 through 2020 to be 9.9 ± 6.4 TgCO₂ (8.9 ± 5.7 MgCO₂ ha⁻¹ yr⁻¹) (Figure 2, Table 1). Although there is a significant amount of interannual variability, the Greenbelt is a persistent and significant net annual sink of CO₂, which is within the range of net annual uptake of CO₂ reported for other mid-latitude forests (3.7–19.7 MgCO₂ ha⁻¹ yr⁻¹, Wofsy et al., 1993; Urbanski et al., 2007; Hardiman et al., 2017; Arain et al., 2022). The areas proposed for removal represent 0.4% of the total Greenbelt area and therefore are expected to have a proportionally smaller net sink of CO₂ compared to the entirety of the Greenbelt. According to the SMUrF model, they sequester between 0.0061 and 0.036 TgCO₂ per year, including the uncertainties in annual NEE. Using Equation 1, this results in an average of 0.0154 ± 0.0152 TgCO₂ (15.4 ± 15.2 GgCO₂) per year (3.31 ± 3.26 MgCO₂ ha⁻¹ yr⁻¹) (Table 1), which represents about 0.15% of the total net sequestration of the Greenbelt, suggesting that the areas identified for development are less productive than the Greenbelt average. This estimate of biogenic fluxes from the sections proposed for removal, which are predominantly croplands, falls within literature values. However, the reported cropland NEE can range from significant sinks of CO₂ to moderate sources (sink of 22 MgCO₂ ha⁻¹ yr⁻¹ to source of 7.7 MgCO₂ ha⁻¹ yr⁻¹, Béziat et al., 2009, and sources within).

During the time it takes to convert the designated areas from the Greenbelt into housing, these areas will be devoid of vegetation. However, the soil will continue to release CO₂ during this time resulting in net CO₂ emissions (Peng et al., 2008; Halim et al., 2023). During the first 3 years of construction, we assume little to no vegetation on the construction sites (Panel 2 of Figure 1), based on our observations of historical imagery. Using only the heterotrophic portion of soil respiration, we therefore estimate a soil respiration of 0.084 ± 0.041 TgCO₂/yr in the areas proposed for removal. This agrees well with a similar analysis performed using VPRM over the same area (Figure S2). This estimate was also within the range of heterotrophic soil respiration from the literature (Mayer et al., 2017; Hoffmann et al., 2018; Kosunen et al., 2019; Halim et al., 2023). For the remaining 2 years of construction, we assume that roughly 50% of the area is vegetated. Accounting for the autotrophic respiration and photosynthesis in half of the area proposed for development, in addition to the heterotrophic respiration across the entire site (Panel 3 of Figure 1), we estimate that the area is a source of 0.031 ± 0.022 TgCO₂/yr during the final 2 years of construction. The sum of the overall soil respiration and the small amount of uptake in these areas during the 5-year period gives a cumulative source of 0.314 ± 0.078 TgCO₂. In contrast, if these sections of the Greenbelt remain undisturbed (Panel 1 of Figure 1), they would be a net biogenic sink of 0.077 ± 0.035 TgCO₂ over the course of the 5 years. Thus, if we take into account both the sequestration potential lost from the removal of these sections of the Greenbelt and the continued soil respiration after the disturbance, the net difference in fluxes would be 0.390 ± 0.083 TgCO₂ (Figure 3). In addition, although these areas are unlikely to remain completely barren of vegetation after construction is complete, the net amount of CO₂ absorbed is likely to be similar to the net sequestration of the existing, undisturbed section of Greenbelt (Figure 4).

Any further development of the Greenbelt beyond that proposed will continue to decrease the amount of sequestration by vegetation in the Greenbelt. For example, if 10% of the Greenbelt were removed, the region would lose approximately 0.99 ± 0.64 TgCO₂ of net sequestration for each year that these lands remain barren. In addition, if vegetation is removed, soil respiration will persist, thus creating a net source of 2.02 ± 0.99 TgCO₂ each year. Therefore, the difference between the net fluxes that would occur if the Greenbelt vegetation remained unchanged and those that would occur if the vegetation was removed and the land remained barren results in a net increase in atmospheric CO₂ of 3.0 ± 1.2 TgCO₂ per year. This estimate is based on the average net CO₂ sequestration of the Greenbelt; however, the entire region does not sequester the same quantity of CO₂ annually (Figure 4, top panel). We find that forests (defined as having > 60% tree cover) and savannas (having 10%–60% tree cover) tend to sequester more CO₂ annually compared to cropland areas in the Ontario Greenbelt (Figure S3), and because the areas proposed for removal are predominantly croplands, they provide less net sequestration (3.31 ± 3.27 MgCO₂ ha⁻¹ yr⁻¹) compared to the average of the entirety of the Greenbelt (8.9 ± 5.7 MgCO₂ ha⁻¹ yr⁻¹; Figure 4). Therefore, conversions of the Greenbelt from more productive forests and savannas to urban areas are likely to result in even larger losses in sequestration than those reported for the areas currently proposed for removal in Sections 3.1 and 3.2.

To determine the importance of the Greenbelt of Ontario in terms of carbon sequestration, we compare the total net sequestration of the Greenbelt and that of the parcels planned for removal to a bottom-up inventory of anthropogenic greenhouse gas emissions in the GTHA. The Atmospheric Fund reports total anthropogenic CO₂ equivalent emissions (CO₂ eq.) in the GTHA to be, on average, 52.9 TgCO₂ eq. yr⁻¹ from 2015 through 2021 (Shekarrizfard et al., 2022), where anthropogenic methane emissions, based on data from the Greater Toronto Area, represent a relatively small 6% of these emissions (Mostafavi Pak et al., 2021). Therefore, the Greenbelt annual net sequestration (9.9 ± 6.4 TgCO₂ yr⁻¹) represents approximately a fifth of the CO₂ eq. emissions reported by the Atmospheric Fund. This highlights the importance of the Greenbelt in the GTHA’s local carbon budget.

The net biogenic CO₂ emissions associated with converting the proposed 2,995 ha of the Greenbelt to housing over the 5 years estimated for construction (0.390 ± 0.083 TgCO₂) is smaller than the magnitude of the net sink of the total Greenbelt, as expected, because of the smaller area and the current land use type. However, written in terms of annual average CO₂ emissions from gasoline-burning passenger vehicles, which we assume to be 4.6 MgCO₂ per vehicle per year (Office of Transportation and Air Quality, 2018), the net difference in biogenic emissions associated with the conversion of this section of Greenbelt land is equivalent to adding the emissions of roughly 17,000 passenger vehicles each year over the 5 years of construction.

In addition to biogenic emissions, there will be anthropogenic emissions associated with the building of new homes in the Greenbelt. The embodied emissions associated with building single-family homes in Toronto is 137 ± 46 kgCO₂ eq. m⁻² of living space (Arceo et al., 2023). We estimate the construction of the proposed 50,000 low-density single-family homes (Ministry of Municipal Affairs and Housing, 2022), using the average living space for detached single-family homes in Ontario (171 m², Statistics Canada, 2019), to result in an additional 1.17 ± 0.39 TgCO₂ eq. emissions. This only considers Scope 3 emissions associated with the extraction, manufacturing, and transportation of building materials (Arceo et al., 2023) as the emissions associated with the building process (Scopes 1 and 2) have been shown to be small in comparison (Pacheco-Torres et al., 2014).

In this research, we determine that the biogenic CO₂ fluxes in the Greenbelt of Ontario (net sink of 9.9 ± 6.4 TgCO₂) play an important role in the annual carbon budget of the region by offsetting roughly a fifth of the GTHA’s anthropogenic emissions. If the areas proposed for development were to proceed with the single-family housing development, converting sections of the Greenbelt to housing will result in an overall loss of sequestered CO₂ even if postconstruction biogenic CO₂ sequestration were to recover to preconstruction levels. Therefore, if the regions proposed for removal are converted to housing, this results in a cumulative difference of 0.390 ± 0.083 TgCO₂ in biogenic fluxes, roughly equivalent to the emissions of 17,000 passenger vehicles each year over the 5 year construction, when considering both the continued soil respiration and the net sequestration lost over the course of a 5-year construction period (Figure 3). This net sequestration loss is largely due to our estimates of the net emissions of CO₂ from soils during the construction period.

In addition to biogenic fluxes, the building of housing itself has associated emissions, which we estimate to be 1.17 ± 0.39 TgCO₂ eq. for the construction of the proposed 50,000 houses. These emissions will add to the net CO₂ released to the atmosphere from this conversion of land. Adding housing will further increase anthropogenic emissions from heating as well as from vehicle use in these areas. In particular, lower density residential areas have been shown to be associated with higher per capita emissions compared to high-density areas in Toronto (Norman et al., 2006) and around the globe (Wu et al., 2020), thus increasing the net emissions even further. For example, denser areas tend to have fewer per capita transportation emissions because of less frequent personal vehicle use, shorter commutes to amenities, and access to public transit (Norman et al., 2006; Perumal and Timmons, 2017).

Toronto is in a housing crisis, and experts agree that additional housing is required (Kalinowski and Gibson, 2021). However, our results suggest that creating single-family low-density neighborhoods on the inner edge of the Greenbelt is more likely to result in higher net CO₂ emissions compared to higher density housing. Higher density housing would reduce the net amount of biogenic CO₂ sequestration lost from converting land in the Greenbelt because the total amount of land required would be significantly smaller. Within the City of Toronto, for example, higher density residences, including duplexes, townhouses, triplexes, fourplexes, and apartment buildings, house on average 113 people, or 48 units, per hectare, with population density potentially being even higher if considering only apartment buildings (Toronto City Planning, 2021). Therefore, building higher density residences can decrease the amount of land required to build 50,000 homes from 2,995 hectares to roughly 1,000 hectares. It is evident from our study that building higher density housing would decrease net biogenic efflux associated with removals of existing vegetation from the Greenbelt. We note, however, that we have not considered all aspects associated with different housing scenarios in this region, such as access to public transit, distances to amenities, decreased local food production from the removal of Greenbelt croplands, and building and maintaining roads and other infrastructure.

While these results show the significant role that the Greenbelt of Ontario plays in the regional carbon budget, it is important to note that we have made several important assumptions. Here, we describe the possible impacts of those assumptions. First, we use the SMUrF model to estimate annual biogenic fluxes, when it was designed to diagnose diurnal and seasonal fluxes of biogenic CO₂ (Wu et al., 2021). As an empirical model, SMUrF does not contain carbon pools, whereas more complicated process-based models track the amount of biomass in soils, on the ground as litter, and in living vegetation (Schaefer et al., 2008). Despite its lack of carbon pools, the SMUrF-modeled fluxes agree relatively well with eddy-covariance flux tower estimates of annual NEE and with VPRM (Figure S2), which has been shown to compare well with annual fluxes estimated using process-based inventories (Winbourne et al., 2022). The SMUrF model and VPRM, however, deviate in the shoulder seasons (Figure S2), and this affects the annual sums of NEE. The SMUrF GPP calculation is driven by measurements of SIF, which is expected to be a better proxy for photosynthesis over longer timescales (Magney et al., 2020). Furthermore, since the SMUrF model’s neural networks were trained using eddy-covariance flux tower estimates and they contain information about GPP derived from SIF, these should capture the average expected respiration, including that associated with changes in rates of photosynthesis. However, longer term accumulations of biomass may not be captured, particularly when considering disturbances, as we do here. This could potentially be improved by examining the time dependence of respiration on GPP and training the neural networks to include information about past fluxes.

The areas being removed from the Greenbelt are predominantly croplands; thus, it is important to consider management practices as these can strongly affect biogenic fluxes (Decina et al., 2016; Qiu et al., 2020; Hundertmark et al., 2021). Photosynthesis in croplands is estimated by the SMUrF model using C3/C4 specific relationships between measured SIF and GPP, and therefore, management affecting productivity should be constrained by changes in SIF measurements. However, the respiration model is based solely on land use type, GPP estimated from SIF, and air and soil temperatures. This does not account for the effects of management practices of individual farms, such as tillage or biomass removed during harvest or added as mulch or fertilizer, on R_eco (Decina et al., 2016; Hundertmark et al., 2021). The neural network was, however, trained at 11 croplands across the United States and Europe (Wu et al., 2021), including pasture and fields rotated through soy and corn, which are 3 of the most common crops grown in the region proposed for removal and the Greenbelt as a whole (Agriculture and Agri-Food Canada, 2020). The neural network, in principle, should capture the average of the practices at all of these training sites. Therefore, while the individual practices of each farm will not be captured, the model should give an estimate based on the average practices from the sites on which it was trained. We therefore make the assumption that the agricultural management in the Greenbelt is similar to that of the United States and European sites used for training the SMUrF model.

The soil CO₂ respiration reported during the construction period is only a rough estimate as the SMUrF model is not designed to capture disturbances such as land-use change. In this model, we assume that roughly 50% of the respiration estimated by the neural network is heterotrophic (Hardiman et al., 2017; Winbourne et al., 2022) and we set the autotrophic portion to 0 μmol m⁻² s⁻¹. However, ratios of heterotrophic to autotrophic respiration have been shown to vary both seasonally and between sites with literature values ranging between 10% and 40% (Suleau et al., 2011), and 37% and 55% if only considering the soil component of respiration (Hanson et al., 2000; Carbone et al., 2016; Fan and Han, 2018), which is, on average, lower than the portion we estimated. Conversely, some studies have shown that heterotrophic soil respiration tends to increase in the years following disturbances (e.g., Ball et al., 2007; Mayer et al., 2017; Halim et al., 2023). Additionally, we assume the same soil and air temperatures in these areas with and without vegetation. Vegetation tends to have a cooling effect (Kurn et al., 1994; Yu and Hien, 2006), and therefore, its removal may cause higher surface temperatures and, since respiration increases with temperature (Schimel et al., 1994), higher respiration rates. Therefore, while our results show that this area, if left barren, would become a biogenic source of CO₂, the exact amount of CO₂ emitted remains uncertain.

Finally, we assume that the regions that were proposed to be converted from Greenbelt areas to neighborhoods will have similar fluxes to that of an existing neighborhood. However, previous land use may also affect the soil and vegetation fluxes after land conversion (Zenone et al., 2013). While the neighborhood we used as an estimate for biogenic fluxes after construction was developed from areas that were once croplands, most of the areas were converted over 10 years ago (Google, 2022), and since soil organic carbon content has been seen to vary with time after disturbances (Zhang et al., 2010), this land may not have the same carbon fluxes as when it was first converted.

Ontario’s Greenbelt is an area designed to prevent urban sprawl and protect agriculturally and ecologically important lands around the GTHA. Using a modified version of the SMUrF vegetation model, we find that the Greenbelt is a significant net CO₂ sink of, on average, 9.9 ± 6.4 TgCO₂/year, offsetting roughly one fifth of the reported Scopes 1 and 2 anthropogenic emissions of CO₂ in the GTHA. Recently, protections had been removed from approximately 2,995 ha of the Greenbelt to construct housing. We find that removing portions of the Greenbelt to expand urban sprawl will decrease the net amount of CO₂ that can be absorbed by the region’s vegetation. This land is likely to remain barren during part of the construction process, when it will lose the ability to sequester CO₂ from vegetation, and the soils will continue to act as a source of CO₂. Once housing has been developed, the region will be unable to regain this lost net CO₂ sequestration as urban areas typically have similar net CO₂ sequestration to the Greenbelt land use they will be replacing. These new housing developments also have embodied and operational emissions associated with them, which are higher, per-capita, than those associated with higher density housing, thus increasing the overall CO₂ emissions of the area. Therefore, any further development into the Greenbelt will continue to strip away its potential to sequester CO₂ and increase net CO₂ emissions, which will result in the region being further from reaching its net-zero emission goals.

The unmodified SMUrF model is available at https://doi.org/10.5281/zenodo.4018123. SIF data from the TROPOspheric Monitoring Instrument (TROPOMI) is available at ftp://fluo.gps.caltech.edu/data/tropomi/ungridded/SIF740nm/, paper doi: https://doi.org/10.1029/2018GL079031. Reflectance data (MCD43A4) and land cover data (MCD12Q1) from the MODerate resolution Imaging Spectrometer (MODIS) are available at https://doi.org/10.5067/MODIS/MCD43A4.006 and https://doi.org/10.5067/MODIS/MCD12Q1.006, respectively. The Contiguous SIF (CSIF) data set is available at https://data.tpdc.ac.cn/en/data/d7cccf31-9bb5-4356-88a7-38c5458f052b/, paper doi: https://doi.org/10.5194/bg-15-5779-2018. Soil and air temperature are available from the European Centre for Medium-Range Weather Forecasts Reanalysis v5 product (ERA5) at https://doi.org/10.24381/cds.bd0915c6. Global Man-made Impervious Surface (GMIS) data is available at https://doi.org/10.7927/H4P55KKF. The Annual Crop Inventory (ACI) data sets are available at https://open.canada.ca/data/en/dataset/ba2645d5-4458-414d-b196-6303ac06c1c9. The City of Toronto’s Impermeable Surface data set is available at https://open.toronto.ca/dataset/topographic-mapping-impermeable-surface/. The Southern Ontario Land Resource Information System data set is available at https://www.arcgis.com/home/item.html?id=0279f65b82314121b5b5ec93d76bc6ba. The shapefile for the Greenbelt of Ontario is available at https://data.ontario.ca/dataset/greenbelt-plan-mapping.

The supplemental files for this article can be found as follows:

The supplemental material (PDF) for this research includes:

Figure S1. Impervious surface area fraction flowchart.

Figure S2. Correlation between SMUrF and VPRM ecosystem fluxes.

Figure S3. Greenbelt NEE fluxes by land cover type.

Table S1. Modeled annual biogenic CO₂ fluxes from 2018–2020.

Table S2. Modeled annual biogenic CO₂ fluxes from 2018–2020 using Toronto-ACI-SOLRIS.

Table S3. Modeled annual biogenic CO₂ fluxes from 2018–2020 using GMIS.

Table S4. Annually summed NEE from SMUrF and eddy-covariance flux towers.

The authors thank Dr. M. Altaf Arain and Dr. Ralf Staebler for the use of eddy-covariance flux tower data at the Turkey Point and Borden Forest flux tower sites, respectively, and for helpful feedback on the comparison of annual NEE between the model and the eddy-covariance flux towers. Figures 2, 3, 4, S2, and S3 were created using Python. Figures 1 and S1 were created using Wondershare EdrawMax. They also thank the two anonymous reviewers for their helpful comments that substantially improved the manuscript.

The authors acknowledge the support of the Natural Sciences and Engineering Research Council of Canada (NSERC), (funding reference number PGSD3-559076-2021 and CGS M 2020). They would also like to acknowledge the support of Climate Positive Energy at the University of Toronto. All above funding sources are associated with Sabrina Madsen.

The authors declare no competing interests.

Sabrina Madsen and Debra Wunch developed the concept and designed the study. Sabrina Madsen performed the analysis and wrote the manuscript. Dien Wu developed the SMUrF model and assisted with the modifications to the model. Md Abdul Halim provided expertise on soil respiration. All authors read and provided comments on the manuscript.

How to cite this article: Madsen, S, Wu, D, Halim, MA, Wunch, D. 2024. CO₂ fluxes of vegetation in the Greenbelt of Ontario and increased net ecosystem emissions associated with its removal. Elementa: Science of the Anthropocene 12(1). DOI: https://doi.org/10.1525/elementa.2023.00102

Domain Editor-in-Chief: Detlev Helmig, Boulder AIR LLC, Boulder, CO, USA

Associate Editor: Daniel Liptzin, University of Colorado Boulder, Boulder, CO, USA

Knowledge Domain: Atmospheric Science