Inequities in tree density, size, and species diversity in 32 Canadian cities

Amidst increasing urban populations¹, access to urban nature is imperative, supporting physical health, mental wellbeing, and improving nature connectedness². Trees form a principal component of urban nature, providing ecosystem services that mitigate many of the negative effects of dense urban spaces³, including improving air quality⁴, moderating urban heat⁵, and reducing stormwater runoff⁶. Trees also benefit human health, improving physical and mental wellbeing⁷. Thus, trees are critical for the livability and sustainability of urban spaces.

While some ecosystem services of trees, such as carbon sequestration, may provide benefits across a large spatial scale, for most urban ecosystem services, the closer a person's proximity to a tree, the greater the benefits they receive from it^8,9,10. For this reason, a growing number of studies consider if the spatial distribution of urban forests is equal (i.e., even) across a city^11,12,13, ideally incorporating number, size, and taxonomic diversity across space to provide a fulsome understanding of where urban forests and their benefits are located. Because of the potential for distributional inequities (i.e., unfairness) for marginalized populations¹⁴, equity analyses are an important additional step to elucidate commonalities and differences in the distribution of trees in relation to sociodemographics¹².

There are several measures that can be considered in assessing the distributional equality and equity of trees. One common, simple measure is the number of trees per unit area, referred to in this paper as the density of trees. Research has highlighted inequitable associations between tree density and income, population density, racial minority population proportions, and residential instability^15,16,17,18. Distribution of trees can also be measured based on tree size, using either median diameter at breast height (DBH), describing the size of individual trees, or basal area, the cross-sectional trunk area calculated using DBH and summed for all trees within the plot¹⁹. Tree size has been found to be positively associated with income and disadvantaged subpopulations^16,20,21, although there is variation by race²¹.

In addition to tree abundance and size, tree diversity is also a common concern of urban foresters^22,23. Higher tree species diversity can reduce the relative vulnerability of an urban forest to pests and pathogens²⁴ as areas lacking diversity can lose trees quickly under a pest outbreak²⁵. Distributional diversity studies often focus on the “luxury effect”, where species diversity is higher in neighborhoods that are more affluent²⁶ where residents have greater access to resources and political capital²⁷. The luxury effect has been found in the distribution of urban tree species in North America and Europe^15,28; however, contrasting effects have been found in Pakistan²⁹, Beijing³⁰, Chile³¹, and Australia³².

While there are numerous studies focusing on the spatial distribution of urban forests, most are limited to a single city or region. These single-city studies do not allow for inter-city comparisons that could provide crucial insights for understanding why urban forest inequities occur so frequently. Notable exceptions include Locke et al.¹³ and Volin et al.,³³ each of which compares distributional equity in several cities; however, both studies exclusively use canopy cover as their measure of urban forest presence, which has generally low association with several of the ecosystem services urban forests provide³⁴. Other studies that have compared distributional equality or equity across multiple cities used measures that include all vegetation³⁵ and do not consider the size or diversity of trees. By including all vegetation, as often occurs with land cover-based studies, the specific benefits of trees versus shrubs and grasses are ignored³⁶ and the urban forestry management unit (i.e., a tree) is obscured³⁷. Additionally, by not considering the size or diversity of trees, many prior studies do not account for the positive relationship between tree size and ecosystem services³⁸ nor the distribution of monocultures²⁵, which influence the vulnerability of the urban forest to pests and pathogens²⁴.

Our study addresses these gaps by assessing the spatial distribution of street trees in 32 Canadian cities (Fig. 1), examining tree density, size, and species diversity. In Canada, street trees are managed and maintained by municipal urban forestry departments. Unlike privately-owned trees, municipal governments can influence species selection, density, and maintenance, all of which collectively influence the distribution of tree density, size, and species diversity³⁹. Previous localized research has found street trees to be unequally and inequitably distributed^40,41,42, warranting a national-scale comparison of cities to identify consistencies and variabilities in distributional equality and their association with sociodemographics. Through this comparative analysis, we focus on two research questions: (1) Using the Gini Index, what are normative levels of distributional inequality in street tree count and basal area? and (2) Are there consistent associations between street tree density, size, and species diversity with population density and indices of deprivation?

Fig. 1: Location of the 32 Canadian cities with tree inventories included in this study.

Red dots with black outlines indicate the study cities.

To examine potential inequities, we used population density and the four variables included in the Canadian Index of Multiple Deprivation (CIMD): i) residential instability, a measure of resident fluctuation, ii) economic dependency, a measure of a population’s reliance on the workforce, iii) ethno-cultural composition, a measure of immigrant and racialized group populations, and iv) situational vulnerability, a measure of additional demographic characteristics such as education levels and home ownership⁴³.

For urban foresters, landscape planners, and allied professionals, this study benchmarks street tree distributional inequalities and inequities in Canada. This helps identify how one city compares to others through a quantitative measure of distributional inequality, calculated using the Gini Index, and the association of street tree distribution to sociodemographic variables. For researchers, this study highlights consistencies and variations in sociodemographic associations with - and distributional inequities in—street tree density, size, and species diversity. The study results identify issues with extrapolating locally identified associations to the national and international scale.

Across the 32 Canadian cities examined in this study, street tree density ranged from as few as 183 trees per km² in Kelowna, BC to 1283 trees per km² in Vancouver, BC (median for all study cities (x͂) = 569 trees) (Fig. 2). Street tree density varied within and across the eight provinces included in the study. Median street tree DBH, an indicator of where larger trees are located, ranged from 8 cm in Ajax, ON to 35 cm in Regina, SK. Like street tree density, median DBH varied within and across provinces. Basal area of street trees per km² of included census dissemination areas (DAs) ranged from 20.82 m²/km² in Moncton, NB to 144.60 m²/km² in Vancouver, BC (x͂ = 43 m²/km²). Cities closer to the Rocky Mountains and Atlantic Ocean generally had lower basal area.

Fig. 2: Street tree statistics for the 32 study cities.

Street tree density (a) is reported for all cities. Median diameter at breast height (DBH) (b) and basal area (c) are reported for 27 study cities as five study cities did not collect DBH measurements.

Street tree species diversity was measured at the DA-level using Shannon’s Diversity Index (H). Typical H values range from 1, indicating low diversity, to 3.5, indicating high diversity⁴⁴. When considering the median H across a city, species diversity ranged from 1.17 in Strathcona County, AB to 3.12 in Toronto, ON.

Presence of inequalities

Moderate distributional inequalities were present across the study cities, as indicated by the Gini Index (G). The Gini Index considers the cumulative distribution of a good (e.g., percent of street tree population) per a cumulative proportion of the population with Gini Index results from 0, indicating perfect equality, to 1, indicating perfect inequality⁴⁵. Because the variables must be cumulative, we use raw counts (i.e., not standardized by land area) at the census dissemination area (DA) level.

Based on the Gini Index, the distribution of street tree count was most equal in Fredericton, NB (G = 0.259) and least equal in Maple Ridge, BC (G = 0.681). The median Gini Index for street tree count was 0.401 with quartiles one and three of 0.350 and 0.490, respectively. Showing greater levels of inequality, the distribution of street tree basal area was most equal in Windsor, ON (G = 0.324) and least equal in Lethbridge, AB (G = 0.709). The median Gini Index for street tree basal area was 0.469 with quartiles one and three of 0.417 and 0.532, respectively. The Gini Index values for each city for tree count and basal area are included in Supplementary Table 1.

Street tree density

Based on Lagrange multiplier (LM) tests, multivariate regressions for street tree density were best modelled with spatial lag models (SLM) in 19 cities and spatial error models (SEM) in 4 cities. Nine cities could not be fit to models. All but three cities had between two and five significant model terms (Fig. 3). Two of the three cities, Lethbridge, AB, and Longueuil, QB, had no significant associations between tree density and any sociodemographic variables. The third city, Maple Ridge, BC, had only one significant model term (situational vulnerability). The models for each city are included in Supplementary Table 2.

Fig. 3: Positive (blue), negative (red), and nonsignificant (dashed lines) associations between street tree measures and sociodemographic variables.

Blank cells indicate models that could not be fit, including cities without diameter at breast height (DBH) data. Census demographics are indicated by codes PD population density, RI residential instability, ECC ethno-cultural composition, ED economic dependency, and SV situational vulnerability.

In comparing the positive versus negative associations, population density was consistently positively associated with street tree density in 19 cities (Table 1, Fig. 3). Residential instability and economic dependency were consistently negatively associated with street tree density in 16 and nine cities, respectively. Ethno-cultural composition was negatively associated with street tree density in eleven cities but positively associated in Burlington and Mississauga, ON. Street tree density associations with situational vulnerability were sparse and varied, with two positive associations (Regina, SK, and Winnipeg, MB) and two negative associations (Maple Ridge, BC, and Fredericton, NB).

Apart from Mississauga, ON, every city with negative associations between residential instability and street tree density also had positive associations between population density and street tree density (Fig. 3). Two patterns—positive associations with population density and negative associations with residential instability—were present across the provinces of Ontario (n = 11), Manitoba (n = 1), and Saskatchewan (n = 1). No other census variables showed the same provincial patterns.

Street tree size

In evaluating street tree size, we considered two variables: median DBH, which is a proxy for where the largest trees are located, and basal area per km², which is a proxy for the volume of trees and ecosystem services provided. Based on LM tests, median DBH was best modelled with SLM in 24 cities and SEM in one city—Quebec City, QC. Of the seven remaining cities, five did not collect DBH measurements, while models could not be fit for the sixth and seventh cities—Kingston and Welland, ON. Across the 25 cities where models could be fit, 20 cities had at least two significant model terms, two had one significant model term, and three cities had no significant associations between median DBH and any sociodemographic variables (Fig. 3). The models for each city are included in Supplementary Table 3.

In comparing the positive versus negative associations, residential instability was consistently negatively associated with median DBH, occurring in 13 cities, while economic dependency was consistently positively associated with median DBH, occurring in 10 cities (Table 1, Fig. 3). Ethno-cultural composition had the greatest number of significant associations. Median DBH was negatively associated with ethno-cultural composition in 19 cities but positively associated in Ajax, ON. Situational vulnerability had the opposite direction with twelve positive associations with median DBH and only one negative association (Toronto, ON). Population density had five positive associations and one negative association, totalling only six cities while 19 cities had non-significant associations.

Commonly, if situational vulnerability was positively associated with median DBH, ethno-cultural composition was also negatively associated with median DBH (n = 11) (Fig. 3). An exception to this was Toronto, ON where both were negatively associated. Provincial-level patterns were visible for median DBH. Across the prairie provinces (Alberta, Saskatchewan, Manitoba), situational vulnerability was consistently positive and ethno-cultural composition was negative (apart from Strathcona County, AB, which was non-significant). Ethno-cultural composition was almost consistently negatively associated with median DBH in Ontario. Quebec had the greatest consistency in significant associations, which held the pattern of greater median DBH in areas of higher population density and situational vulnerability but lower median DBH in areas of higher residential instability and ethno-cultural composition, indicative of distributional inequities.

Basal area was best modelled with SLM in 18 cities and SEM in two cities. Of the 12 remaining cities, five did not collect DBH measurements, and models could not be fit for the other seven cities. Across the 20 cities where models could be fit, 17 cities had at least two significant model terms, two had one significant model term, and only one city, Kingston, ON, had no significant associations between basal area and any sociodemographic variables (Fig. 3). The models for each city are included in Supplementary Table 4.

In comparing the positive versus negative associations, population density was consistently positively associated with basal area, occurring across 12 cities, while residential instability was consistently negatively associated with basal area, occurring across 14 cities (Table 1, Fig. 3). The three remaining variables had small inconsistencies in their patterns. Ethno-cultural composition was frequently negatively associated with basal area (n = 18) but positively associated in Strathcona County, AB. Economic dependency was typically negatively associated with basal area (n = 6), but positively associated in Strathcona County, AB, and Mississauga, ON. Lastly, situational vulnerability was often positively associated with basal area (n = 7), but negatively associated in Toronto, ON.

There were few consistent patterns across cities and provinces (Fig. 3). However, 13 cities had negative associations between basal area and both residential instability and ethno-cultural composition. Negative associations between basal area and residential instability and ethno-cultural composition were common for Saskatchewan, Manitoba, Ontario, and Quebec. Similarly, positive associations with basal area and population density were common for Saskatchewan, Manitoba, and Quebec.

Street tree species diversity

Street tree species diversity was best modelled by SLM in 14 cities, SEM in eight cities, and OLS in one city. Models could not be fit for the remaining nine cities. Across the 23 cities where models could be fit, 20 cities had at least two significant model terms, while three cities had only one significant model term. In contrast to the other three tree variables, no cities lacked significant association between species diversity and any sociodemographic variables (Fig. 3). The models for each city are included in Supplementary Table 5.

In comparing the positive versus negative associations, population density was universally negatively associated with species diversity, occurring across 23 cities with no city having a non-significant or negative result (Table 1, Fig. 3). Species diversity was also consistently negatively associated with economic dependency (n = 9) and situational vulnerability (n = 12). Species diversity was mixed in its association with residential instability and ethno-cultural composition. Residential instability was negatively associated with species diversity in seven cities but positively associated in two cities. Ethno-cultural composition was negatively associated with species diversity in five cities but positively associated in three cities.

Because of the many significant negative associations between species diversity and population density, the pattern was found across the country (Fig. 3). The only other identifiable pattern was the negative association between species diversity and situational vulnerability in New Brunswick.

Through a national-level examination of Canadian urban tree inventories, we found inequalities in the distribution of street trees in all 32 study cities. Yet, despite the existence of these distributional inequalities, we found nuanced patterns in the association between street tree variables—density, size, and species diversity—and sociodemographic variables. Most cities displayed a combination of some inequitable and some non-significant associations between distribution of street tree density and size across different sociodemographic variables.

The Gini Indices identified in our study suggest varying levels of distributional equality exist across Canadian cities. While there are different interpretations of Gini Indices in urban greening, the general heuristic is that Gini Indices under 0.2 indicate equality, 0.2 to 0.3 indicate relative equality, 0.3 to 0.4 indicate relative fairness, 0.4 to 0.5 indicate concerning disparity, and above 0.5 indicate significant disparity⁴⁶. None of the 32 cities had equal distributions of trees (i.e., > 0.2), although greater equality was found in the distribution of street trees by count than basal area. The most equal distribution was found in the street tree count of Fredericton, NB (G = 0.259), which could reflect the city’s low population and long history of proactive approaches to tree planting and management⁴⁷. Conversely, the least equal distribution of street tree count was in Maple Ridge, BC (G = 0.681), which may be due to the city’s more diverse land use and recent rapid urban growth, which could limit opportunities for street tree establishment, or the prioritization of tree planting in parks and on private property⁴⁸, which would not be reflected in our study of street trees.

There are limited comparable studies of the Gini Index for street trees. However, a canopy cover-based assessment of distributional equality in 40 US cities found Gini Indices ranging from 0.15 to 0.47³³, indicating potentially greater distributional inequality in Canadian cities’ street tree populations. Similarly, when basal area is considered, inequalities in size distribution could reflect different stages of urban development or tree planting³⁸. In some cases, a moderate Gini coefficient may even be preferable given that it could represent a larger diversity in tree sizes and thus a more resilient future urban forest^20,49.

The Gini Index is solely a measure of equality and does not capture distributional inequities related to sociodemographics⁴⁵. Therefore, we used multivariate regression to assess how DA-level sociodemographics were associated with street tree density, size, and species diversity. While all cities exhibit distributional inequalities in street tree measures, there are notable variations in sociodemographic associations both within and between cities. This variability underscores the complexity of urban tree equity and necessitates a nuanced exploration of intra- and inter-city comparisons.

Street tree density was frequently negatively associated with residential instability (a measure of resident fluctuation), ethno-cultural composition (a measure of immigrant and racialized group populations), and economic dependency (a measure of a population’s reliance on the workforce). A similar pattern was observed for median DBH and basal area with residential instability and ethno-cultural composition. These negative associations mean that as these three measures of deprivation—residential instability, ethno-cultural composition, and economic dependency—increase, tree density, median DBH, and basal area decrease. These are indicative of prevalent inequities that align with previous studies that identified income, race-, ethnicity-, and homeownership-based inequities in urban forest or street tree distribution^11,40,50. The fact that there are consistent patterns at the provincial level for many of these inequities implies similar patterns of tree planting and land use planning that result in inequities.

In contrast to the other Canadian Index of Multiple Deprivation (CIMD) variables, inequitable associations were less frequent between situational vulnerability and tree density, with often non-significant associations, and tree size, which often has positive associations. In the case of positive associations, this means that as situational vulnerability increases, median DBH and, less commonly, basal area, also increase. This was surprising given that situational vulnerability includes median dollar value of the dwelling and education attainment⁴³, which have previously been found to be associated with urban forest inequities⁵¹. Situational vulnerability also includes the proportion of the population that identifies as Indigenous⁴³. Indigenous Peoples have been found to be under-engaged in urban forest management planning (an issue of recognitional inequity)⁵², but the positive associations for situational vulnerability suggest that distributional inequity is not as prevalent an issue as under-engagement. Interestingly, only one city, Mississauga, ON, displayed negative associations between population density and street tree density and size. The positive associations found in the other study cities may reflect common municipal goals to prioritize street tree planting and maintenance in residential areas⁵³ and policies protecting street trees during development⁵⁴.

In contrast to street tree density and size, street tree diversity was universally negatively associated with population density, likely due to challenging growing conditions in densely populated areas, limiting street tree species suitability^25,55. Similarly, street tree diversity was often negatively associated with CIMD variables, perhaps reflecting differences in residents’ access to resources and political capital²⁷ or underlying associations between CIMD variables and challenging growing sites, including areas with greater impervious surface area or air pollution²⁵. Diversifying street tree species in these areas can help mitigate the potential impact of invasive pests and pathogens that would otherwise cause localized canopy cover loss and drive distributional inequalities due to streets being composed entirely of one genus^42,56.

While there were consistencies in patterns of inequities, there were 10 instances where associations between tree variables and sociodemographic variables had both positive and negative associations (Fig. 3). For example, species diversity was positively associated with ethno-cultural composition in three cities, yet negatively associated in another five cities. Relatedly, there were several cities that deviated from otherwise consistent patterns of positive or negative associations. There were also few instances of patterns consistent across provinces or large regions like British Columbia’s Lower Mainland or Ontario’s Greater Golden Horseshoe. Further, there were seldom instances where one socio-demographic variable was generally negatively or positively associated across all four tree variables. Residential instability is the noteworthy exception with 50 negative associations across the four tree variables and only two positive associations (species diversity in Maple Ridge, BC and St. Catharines, ON).

The variation in associations between sociodemographics and street tree variables supports other multi-city studies that identify unique city-level patterns of distributional inequity of urban trees^13,33,57,58. Given these differing patterns of inequity, researchers must be wary of extrapolating patterns of inequity observed in studies examining one or only a few cities. Distributional patterns of street tree density, size, and diversity are influenced by municipal socio-political and biophysical factors, both historical and current^59,60,61. Considering these localized factors is important in interpreting distributional inequalities and inequities.

Because we specifically examined street trees, we quantified spatial distribution of a subset of the total tree population in Canadian cities. Private trees, park trees, and riparian vegetation can all contribute to canopy cover⁶²; however, inventories of private, park, and riparian trees are not routinely collected⁶³. Private trees are also subject to management based on the preference of homeowners, particularly in the absence of stringent legislation preventing unpermitted removal⁶². Testing our findings for other land use types may highlight counter relationships or alignment with street tree distributions. While ground-based inventories do not usually include privately-owned trees and sporadically include park and riparian trees, the increasing use of light detection and ranging (LiDAR) in urban environments will provide an opportunity to advance this research to include all trees within cities.

Urban tree inventories can also be subject to measurement error and incorrect species identification. While we removed trees with a DBH above 350 cm, which suggests measurement or recording errors, erroneous values under this threshold may influence the results⁴⁹. However, urban forest managers must make decisions based on their inventories²⁵, and this study uses the information available to managers in examining street tree distribution. Street trees represent an accessible source of urban nature and an asset that is provided and maintained by cities for the betterment of their residents³⁹. With recent recommendations that people be able to see three trees from their residence¹⁰, street trees provide a means for cities to accomplish this imperative.

While trees provide many benefits for residents, some residents may not want trees near their homes^64,65. Given such preferences, some residents may not be concerned about the lack of trees around their residence. Likewise, some may wish to have streets lined with the same species⁶⁶, resulting in low local diversity. Others may prefer small ornamental street trees that result in a smaller median DBH⁶⁷. Our study did not identify such preferences. Rather, we considered spatial distribution because equitable distribution of tree density, size, and diversity promotes equity in urban forest ecosystem services and sustainability^8,9,10,24. Yet, urban forest equity does not simply mean that all streets or neighborhoods have the exact same assemblage of trees. Distributional justice is only the first step in creating an equitable urban forest⁶⁸. Localized efforts to increase recognitional and procedural equity⁶⁹, especially considering residents’ preferences for treed streets, will only further Canadian urban environmental justice.

Study cities, collection of inventories, and data cleaning

This study examined 32 Canadian cities (Fig. 1), all of which were part of the Canadian Institute of Forestry / Institut forestier du Canada (CIF-IFC) Open Urban Forests Project (OUFP) repository, which aims to include all street tree inventories across Canada⁷⁰. The 32 cities span five ecozones⁷¹ and eight provinces, although nearly half were in Ontario (n = 15, 47%), which had 10 more cities than the second most common province, British Columbia (n = 5, 16%).

The OUFP repository does not include all variables collected in the inventories, so the inventory data were sourced directly from the open data catalogues of the cities themselves (Supplementary References 1). The inventories were downloaded on May 24th, 2024. Inventories were included if tree species identification (ID) or diameter at breast height (DBH) was available. In 28 inventories, both species ID and DBH were reported. Three inventories collected species ID but were excluded from DBH analyses because they lacked this measure (New Westminster, BC and Peterborough, ON) or collected DBH in ranked bins (Halifax, NS). We also excluded Maple Ridge, BC, and Waterloo, ON from the DBH analysis as DBH measurements were only collected for 9% and 63% of their street trees, respectively. Only one city’s available tree inventory (London, ON) contained neither variable, so it was excluded from this study. Another tree inventory identified in the OUFP repository (Markham, ON) only includes trees on major thoroughfares, so it was excluded from this study.

The downloaded inventories were cleaned, including removing dead trees and stumps, setting 0 cm DBH tree measurements to null, removing trees outside city limits, and removing non-street trees. Based on histograms of the DBH measures, an outlier threshold of 350 cm DBH was established per Martin et al.⁴⁹. Any DBH measure above 350 cm was set to null. Diseased, dying, or trees in poor condition were not removed from the inventories. Trees with missing species identification were set to “unknown”.

There were 31 inventories that included both street and non-street trees. In 18 of them, non-street trees were identified in a variable that was used to subset inventories into those that only contained street trees. In the remaining 13 cities where non-street trees were not identified in the inventory itself, the proximity of trees to the city’s road network (Supplementary References 2) was used to isolate street trees following methodology used by Galle et al.⁷² While Galle et al.⁷² included all trees within a distance of two-times the lane width multiplied by the number of lanes based on road centerlines, we used a larger catchment area. This was largely owing to many Canadian residential streets that are classified as one lane of traffic (i.e., one lane in either direction), but have room for three or more car widths (e.g., parking plus one lane in either direction), thus a larger buffer area from the road centerline is required. To account for parking, in the four cities with road data that included lane counts, the street tree catchment was set as four meters times the number of lanes plus 12 meters, measured from the centerline. For the remaining nine cities without lane count data, a catchment area of 20 meters from the centerline was used, equivalent to a two-lane roadway. Because no inventory included privately owned trees, the main source of error was potential inclusion of park trees that border streets. The proportion of street trees to total inventoried tree count is similar between the column-based and street buffer-based methods at 72.48% and 75.01%, respectively, suggesting the catchment method is appropriate.

Measurements of trees

This study examines four measurements common to urban forestry: tree density (number of trees per square kilometer), median DBH, basal area per square kilometer, and Shannon’s Diversity Index. Each of these measurements was calculated per dissemination area (DA), the smallest spatial unit of the Canadian Census for which data is routinely reported by Statistics Canada.

Tree density was calculated as the number of trees per DA, divided by the size of the DA in square kilometers. Median DBH was calculated in centimeters for each DA. Basal area was calculated by converting the DBH measurements to a circular area for each tree (cm²), then summing across the DA and dividing by the size of the DA. The Shannon’s Diversity Index was calculated for species diversity measures because it includes both richness and evenness and better accounts for rare or underrepresented species than the similar Simpson’s Index⁷². Given the high portion of exotic species in urban areas, many tree species may be represented by only a few individuals³⁷.

Dissemination areas and census variables

Human demographic associations with urban trees were analyzed within DAs. DAs have populations of 400–700 people⁷³. DA boundaries are delineated by physical landscape features (e.g., rivers, major streets) and are informed by local experts to be sociodemographically homogenous⁷³. While some studies have used the larger Census Tracts (CTs), smaller municipalities in our study had too few CTs to conduct meaningful analyses. In cities surrounded by rural areas, such as in the prairie and Atlantic regions of Canada, CTs also cover low population density agricultural or exurban spaces within the city boundaries. Thus, the use of DAs, above a population density minimum, more accurately reflects urban tree distribution in Canadian cities.

In our study, DAs were included if they were within the cities’ boundaries and had population densities above 400 people per square kilometer. This minimum threshold is based on Statistics Canada’s classification of urban areas⁷⁴ and excluded agricultural and exurban areas.

Population density and the Canadian Index of Multiple Deprivation (CIMD) by DA were used to assess the sociodemographic correlates of street tree metrics. The CIMD, developed by Statistics Canada to provide a longitudinal measure of Canadian marginalization using 2021 census data, provides four indicators of population marginalization at the DA level: i) residential instability, which measures fluctuations in neighborhood residents, ii) economic dependency, which addresses a population’s reliance on the workforce, iii) ethno-cultural composition, which measures the immigrant and visible minority populations, and iv) situational vulnerability, which describes additional demographic characteristics such as education levels and home value⁴³. All indicators were assessed as factor scores, where higher values indicate higher levels of marginalization. The national-scale CIMD values, which use the same variables across all cities for each of the four indicators, were used for all cities. The CIMD was chosen to represent these metrics of marginalization because its breadth allows for comparisons to a wide range of existing studies on urban forest equity, and representation as factors allows for the incorporation of many census variables without introducing multicollinearity issues into multivariate analyses. The CIMD was downloaded on April 4, 2025. We supplemented the CIMD variables with population density (people per square kilometer) as a proxy for built density, which may limit space for street trees.

Statistical analyses

To examine the distributional equality within each city, the Gini Index was used. The Gini Index was originally developed as a measure of income inequality but has since been employed in examining the distribution of hospitals, grocery stores, parks, and trees⁴⁵. Its primary benefit is its simplicity, measuring the proportion of the asset—in this case, trees—available to a proportion of the population which is divided into groups—in this case, DAs. The tree count and basal area per DA were sorted in ascending order for each city and the Gini Index (G) was calculated using the equation

where Ei is the cumulative proportion of the city’s trees up to the ith DA, Pi is the cumulative proportion of the population up to the ith DA, and n is the number of DAs within the city's boundary. The Gini Index (G) is bound by 0 ≤ G ≤ 1 with values closer to 0 indicating no inequality and 1 indicating complete inequality.

Linear regression models were estimated using ordinary least square (OLS) to assess the relationship between street tree measurements and sociodemographic variables. The validity of each model was assessed, including for the presence of multicollinearity, heteroskedasticity, and non-normal distribution of model error terms. When heteroskedasticity or non-normally distributed error terms were present, spatial autoregressive (SAR) models—namely spatial lag (SLM) and error (SEM) models—were used⁷⁵. SLMs incorporate a spatially lagged dependent variable (WY) to account for instances where the dependent variable in one DA is influenced by the dependent variable in neighboring DAs^76,77. The general form is

where Y is the vector of dependent variables, X is the vector of explanatory variables, β is the vector of regression coefficients, ρ is the spatial lag coefficient, W is the spatial weights matrix, and ε is the vector of independently distributed error terms.

By contrast, SEMs incorporate a spatially lagged error term (Wu) to account for spatial dependence in the error terms⁷⁷, capturing unobserved spatial processes not captured in model variables. The general form is

where Y is the vector of dependent variables, X is the vector of explanatory variables, β is the vector of regression coefficients, and u is the vector of spatially correlated error terms, calculated with the formula

where λ is the spatial error coefficient, W is the spatial weights matrix, u is the vector of spatially correlated error terms, and ε is the vector of independent error terms.

First order queen contiguity was used in the SAR modelling, and Lagrange multiplier (LM) lag and error tests were used to determine whether SLM or SEM should be used⁷⁷. When only one LM value was significant, the applicable SAR model was applied. Where both LM tests were significant, the robust LM tests and their magnitude were used to select the model. In cases where neither robust test statistics were significant, the model was reassessed at the OLS level for fit issues⁷⁸. All multivariate analyses, including spatial structure assessment and modelling, were done in R 4.4.0⁷⁹.

In 36 models, associating each street tree variable with sociodemographic variables could not be modelled using OLS, SLM, or SEM. These models exhibited issues with heteroskedasticity or non-normally distributed residuals, and the corresponding SAR models were not statistically significant. Variable transformations were explored, but did not address model fit issues. To avoid compromising interpretability through more complex or iterative model specifications, these models were omitted from the results, which present the remaining 91 models.

References

Authors and Affiliations

1. Department of Geography, Geomatics and Environment, University of Toronto Mississauga, Mississauga, ON, Canada

   Alexander J. F. Martin, Ashlynn Fleming & Tenley M. Conway

Contributions

A.J.F.M. and A.F. wrote the manuscript text, and T.M.C. edited the manuscript, prepared the tables, and provided supervision and resources. A.J.F.M. created the figures. All authors reviewed the manuscript.

Corresponding author

Correspondence to Alexander J. F. Martin.

Competing interests

The authors declare no competing interests.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions