Land‐use history causes differences in park nighttime cooling capacity and forest structure

Abstract

Cities are temporally dynamic ecosystems that experience continuous redevelopment over time. Urban parks, which provide critical benefits to resident wellbeing, are developed on different land‐use types and contain natural elements that are susceptible to the effects of historical decision‐making. Thus, understanding the current day functioning of our city's ecosystems and planning for the equitable distribution of park benefits in the future requires incorporation and understanding of the impacts of historical decision‐making. We measured neighborhood sociodemographic composition, forest structure, and the cooling effect of 33 sites in parks across Montreal, each with a past land‐use in one of three classes: agricultural, forested, or industrial. We asked the following questions: (1) what are the effects of historical land‐use on current park forest structure, diversity, and consequently the capacity to provide cooling? (2) how do surrounding communities differ around parks of each historical land‐use type, and what are the implications for equitable access to cooling? We found that forest structure and cooling capacity differed across past land‐use type, and forest structure has complex relationships with park cooling capacity. Our results provide evidence of historical environmental injustice impacting current day cooling capacity for marginalized groups. Previously industrial parks had less cooling capacity at night, while simultaneously being surrounded by communities with higher proportions of immigrants and lower median incomes than Montreal's average. However, daytime cooling capacity was similar across past land‐use type, highlighting the importance of current management decisions to provide a critical ecosystem service, temperature mitigation, regardless of a site's history. Planting areas of small, dense forest stands with trees ≥5 cm dbh within urban parks can help augment daytime cooling benefits in the city but may hinder nighttime cooling. To provide both nighttime and daytime cooling, a mixed management strategy of park trees is required, where small and large trees are incorporated at different densities. Finally, we find evidence of gentrification surrounding all park sites, emphasizing the complex socioecological dynamics of green infrastructure and the need for community‐led greening projects paired with social housing policies.

INTRODUCTION

Urban greening is gaining popularity as a strategy to improve human wellbeing while combatting climate change and biodiversity loss (Haaland & van den Bosch, 2015; Nelson & Bigger, 2022). However, to more fully understand the benefits provided by urban natural ecosystems, we must integrate historical conditions. Urban landscapes experience legacy effects, defined as the present‐day impacts of past decisions or events (Ossola et al., 2021), which can in turn affect ecosystem functioning for decades to millennia (Cramer et al., 2009; Foster et al., 2003). An important driver of legacy effects in cities is land (re)development, the process of changing one land cover type to another, often multiple times across decades (Raciti et al., 2011; Ziter & Turner, 2018). Throughout the evolution of cities, patches of land are continually maintained, redeveloped, and abandoned, creating a complex history (Alig et al., 2004; Koebel, 1996). For example, many cities in North America were deindustrialized in the mid‐20th century, leading to a series of “revitalization” efforts which redeveloped former industrial areas to mixed‐use areas with office buildings, hotels, green spaces, and residences (Koebel, 1996). Urban growth in North American cities also often results in expansion at city edges, where agricultural and forested land surrounding cities are developed into urban land‐use types, including parks and green spaces (Dallimer et al., 2015). This mix of urban intensification and/or land abandonment in urban cores combined with the expansion of suburbs at city edges means that urbanization results in contemporary green spaces existing in a variety of spaces and incurring a variety of legacy effects (Alig et al., 2004; Dallimer et al., 2015).

Land‐use and (re)development patterns in cities are determined by urban planning, a social endeavor with place and period‐specific goals determined by those in power (Pallagst, 2009). Thus, urban planning and land development in post‐colonial North America inevitably and intentionally upholds inequity and injustice (Alexander, 1992; Calmore, 1997; Goetz et al., 2020; Powell, 2008; Ture & Hamilton, 1992), ultimately structuring the ecology of our cities (Schell et al., 2020). Environmental injustice is a documented form of inequity across North American cities, and although urban nature is increasingly considered a type of infrastructure all communities deserve access to, decades of evidence demonstrate that urban planning benefits white, rich communities and endangers low‐income communities of color (Gelobter, 1993; Goetz et al., 2020; Maynard, 2017). Marginalized communities are often relegated to neighborhoods that are poorly designed (Gelobter, 1993), under serviced (Larsen & Gilliland, 2008; Shah et al., 2016), and have environmental hazards that threaten public health (United Church of Christ, 1987; Waldron, 2018). For example, on the island of Montreal, urban heat is correlated with the percentage of low‐income residents and education level across neighborhoods (Fan & Sengupta, 2022). Low‐income and visible minority residents also experience inequity in access to vegetation (Pham et al., 2012), a key factor in combatting urban heat. Further, neighborhoods surrounding industrial areas were historically inhabited by working‐class or impoverished immigrants working in the quarries (Héroux & Martin, 2020; Ndiaye, 2004). Today, populations with higher proportions of visible minorities and low‐income households are more likely to live near highly congested parks in Montreal, reducing access to park benefits such as heat mitigation (Jepson et al., 2022). More broadly, areas of high social vulnerability in several Canadian cities have lower urban forest resilience (Landry et al., 2020), a key factor determining the longevity and health of urban nature.

Increasing the presence of urban forests and parks is often regarded as universally positive (Roman et al., 2021). However, a concern of urban greening is the potential perpetuation of societal inequities (Kiani et al., 2023; Sax et al., 2022). This is particularly true for parks developed on industrial land in historically marginalized neighborhoods. Without strong social housing policy, urban greening can lead to gentrification and displacement of residents (Anguelovski, 2016; Anguelovski et al., 2019). In addition, legacies of historical inequity in neighborhoods through various mechanisms such as environmental racism and the presence of industrial lands have been shown to continue to impact key providers of ecosystem services such as current tree canopy and growth (Healy et al., 2022; Nix et al., 2023). The influence of historical land‐uses and inequity on the current distribution of tree canopy has other important implications which have not yet been thoroughly studied, notably the distribution of urban heat.

Urban trees and forests are a critical component to consider in understanding the link between historical legacies and the benefits of green spaces. Park trees promote wellbeing in diverse ways, from physical relief through heat mitigation to mental health improvements via stress reduction and promotion of physical activity provided by exposing the brain to nature (Remme et al., 2021). Benefits of urban trees are often determined by the composition of tree species, size, and density (Qiu et al., 2018; Villamagna et al., 2013; Ziter et al., 2013). The long‐lived nature of trees means their current state and the functioning of forest ecosystems are strongly affected by legacy effects of past management decisions (Roman et al., 2018, 2021). Legacy effects may influence tree size and diversity due to legacies in soil properties and nutrient levels that continue to influence tree growth and functioning today (Aznarez et al., 2023; Pregitzer & Bradford, 2023). For example, agricultural land may be depleted of nutrients following its use as farmland, influencing the soil structure and development over time (Pabst et al., 2016). Indeed, agricultural land often accumulates carbon in the soil following development, whereas former forest remnants do not, which has the potential to bring about long‐term changes in tree growth and associated ecosystem services (Raciti et al., 2011). Legacy effects from past land‐use types can also determine which species may or may not be present in an area (Munteanu et al., 2015). For example, in residential areas in Haikou, China, that were previously woodlands, there are seven species of trees that are not found in residential areas that were previously open land (Yang et al., 2017).

An urban park is thus a complex web of past and current land uses, and understanding the physical influences of historical land use on vegetation will allow us to tease out a piece of the puzzle when assessing the differences in our parks' benefits across space. Many cities are interested in pursuing restoration and altered management of urban green spaces. Quantifying the legacy effects of past land use types on the benefits of trees in current urban parks will facilitate improved planning for equitable distribution of park benefits in the future. Here, we ask two research questions. First, what are the effects of historical land use on current park forest structure, diversity, and consequently the capacity to provide cooling? Second, how do surrounding communities differ around parks of each historical land use type, and what are the implications for equitable access to cooling?

We hypothesize that the historical land‐use type will influence the quantity of temperature mitigation currently delivered by urban trees in Montreal's parks via differences in tree size, tree basal area, and tree species richness. Tree size and basal area are both factors that drive cooling in urban settings (Helletsgruber et al., 2020; Rahman et al., 2020). Additionally, tree species is a driving determinant in cooling efficacy (Helletsgruber et al., 2020) and higher species richness of trees can lead to an increase in cooling through an increase in canopy density (Gillerot et al., 2024). We predict that previously industrial sites will have the lowest levels of temperature mitigation due to smaller trees, lower basal area, and less species‐rich tree stands. We expect these differences in tree communities at previously industrial sites because of their association with high levels of heavy metals and low levels of carbon and nitrogen stocks as a result of soil contamination, removal, and treatment (Ashwood et al., 2022; Clarke et al., 2015; Goyal et al., 2020; Rooney et al., 2012). We predict previously agricultural sites will have a higher capacity for temperature mitigation than previously industrial sites but less than previously forested sites. Agricultural lands typically have higher levels of heavy metals and lower carbon stocks than forested sites, which can cause a decrease in woody plant species richness (Arao et al., 2010; McLauchlan, 2006; Shumi et al., 2018; Zwolak et al., 2019) However, they can have high levels of nitrogen stocks, which increases plant growth (McLauchlan, 2006). Finally, we predict that previously forested sites will have the largest, densest trees and highest species richness, leading to the highest capacity for temperature mitigation due to their relatively less intense management (see Figure 1 for hypothesized causal relationships).

FIGURE 1.

Directed acyclic graph depicting our hypothesized causal relationships between variables in our study system. Arrows in this graph indicate that the variable at the blunt end of the arrow directly affects the variable at the sharp end of the arrow (i.e., a causal relationship). For example, changing the past land use will result in a change in nitrogen stocks. Temperature, tree density, tree size, tree diversity, past land use, and park age were all included in our models.

Regarding question two, we hypothesize that due in part to historical environmental injustice there will be differences in resident communities surrounding parks characterized by each past land‐use type. We predict that proportions of marginalized populations, such as recent immigrants, people of color, and residents with low incomes will be present at higher rates surrounding parks that have industrial land‐use legacies (Gelobter, 1993; Gerrish & Watkins, 2018; Jolivet & Carré, 2017; Landry et al., 2020). Due to the complex nature of these relationships, as well as historical data limitations, we do not have the ability to test the causal relationships between past land‐use and current sociodemographic profile. However, we assess current differences among areas with different past land‐use types using an exploratory data analysis so that we can better contextualize our results in terms of the capacity of cities to work towards green equity.

METHODS

Study area

We sampled 33 sites spread across the island of Montreal (hereafter referred to as “Montreal”) in the summers (May–August) of 2022 and 2023. Montreal houses Canada's second largest city, has 2.04 million residents (Gouvernement du Québec, 2022) and is approximately 4022 km². Montreal is composed of the 19 boroughs of the city of Montreal, along with 14 independent municipalities (Association des municipalités de banlieue, 2018; Ville de Montréal, 2023a), and the level of urbanization, population density, socioeconomic status, and development varies greatly within and across these areas. The parks that we sampled spanned the island (Figure 2), including 13 boroughs and three municipalities.

FIGURE 2.

(a) Our study sites across the island of Montreal (n = 33, n = 11 per past land‐use type). Different past land‐uses were categorized into three main types, shown in different color shades: Industrial, agricultural, or forested. (b) Park Angrignon, one of our parks that had multiple previous land‐use types. Each previous land‐use type was considered a different study site and was sampled separately. Each study site had 15 potential sampling points (9 true sampling points), five in low canopy areas, five in medium canopy areas, and five in high canopy areas. For visual depiction of the sampling plots, see Appendix S1: Figure S1. Maps were created using “sf,” “ggplot2,” “MetBrewer,” and “osmdata” packages in R 4.3.1 (Mills, 2022; Padgham et al., 2023; Pebesma, Bivand, et al., 2023; R Core Team, 2023; Wickham et al., 2023).

Explanatory variables

Legacy effects

To test how legacy effects influence park ecosystem service capacity, we identified the historical land‐use type of 27 parks. We defined historical land‐use type as the category of land‐use that the park was developed on and is therefore dependent on the year the park was developed, which ranges in our study from 1887 to 2016 (Appendix S1: Table S2). Through extensive searching of historical city archives, satellite imagery, and gray literature published by historical societies, municipalities, and historical enthusiasts, we reconstructed the history of each park to identify (1) the age of the park and (2) the land‐use type immediately prior to park development (Appendix S1: Table S1). We sorted historical land‐uses into three categories: agricultural, industrial, and forested. We sampled 27 parks; however, within parks, each different historical land‐use was treated as a different study site, resulting in 33 study sites total (n = 11 of agricultural, industrial, and forested historical land‐uses each, see Appendix S1: Table S2 for historical land‐use(s) of all parks). For parks with multiple historical land‐uses, we digitized the different historical land‐use boundaries using satellite imagery (Division de la géomatique and Ville de Montréal, 2023) and historical map layers from the City of Montreal (Division Soutien Technique, Infrastructures, CESM and Ville de Montréal, 2023; Ville de Montréal, 2023b) using QGIS v3.16 (QGIS Development Team, 2020).

Forest structure

We assessed the forest structure of our study sites in the summer of 2022 by assessing tree basal area, median size, and species richness. To sample each site representatively, we sampled plots in areas of low, medium, and high canopy. Prior to entering the field, we calculated the canopy cover across the parks in 0.04 ha pixels (Communauté métropolitaine de Montreal, 2019). We then selected three low (10%–30%), medium (31%–80%), and high (81%–100%) plots in each site to sample. If there were not enough low, medium, or high canopy areas to sample, we did as many as possible (n = 9 for study sites, with some exceptions when not possible, n = 279 sampling plots total). For each of the sampling plots, we conducted a forest inventory. Once arriving at the sampling plot, we walked 20 m in each cardinal direction and flagged the edge of the plot. The four cardinal points were then connected diagonally, creating a 0.08 ha square plot (Appendix S1: Figure S1). We measured and identified every tree with a dbh >5.0 cm. For the first half of the field season, we also did a complete inventory of the 0.08 ha plot for trees with a dbh >1.0 cm. Due to time constraints, we changed our methodology to a “mini‐plot” of 0.005 ha for trees between 1 and 5 cm dbh halfway through the season, consistent with the Canadian Forest Inventory (Canadian Forestry Inventory Committee, 2008). The mini‐plot was placed in the most representative area of the larger sampling plot, and every tree under 5 cm dbh was identified and classified into a size category (1–3 cm dbh or 3–5 cm dbh, for more information see Appendix S1: Figure S1). For trees with multiple stems, if the split in stems was above 30 cm, we measured the dbh just below the split. If the split was below 30 cm, we measured each stem individually at breast height (Magarik et al., 2020). To identify trees to species level, we used the National Audubon Society Field Guide to Trees: Eastern Region (Little, 1980), Trees of Canada (Farrar, 1995), and Seek (iNaturalist, 2022). If identifying to species level was not possible, we identified to the genus level (<0.1% of trees).

Response variables

Temperature regulation

We collected temperature data the summer following the vegetative data collection. We deployed sensors during the hottest months of the year, collecting data between June 1 and July 31, 2023, recording a data point every 15 min. Sensors were CredoSense CSL‐T0.5 temperature loggers, with an accuracy range of ±0.1–0.5°C (Appendix S1: Figure S2). We placed temperature sensors in three plots within each site—one for each canopy level (Appendix S1: Table S2). We installed sensors at approximately 1.5 m in height, in a location representative of the canopy cover of the plot, typically near the plot center, attached to a tree branch or trunk if the tree was too large to reach the branches.

To assess the cooling effect of park vegetation as an ecosystem service, instead of raw values of temperature, we deployed control sensors in the neighborhoods surrounding each study site and compared the temperature differences between the park and the surrounding built‐up area. We selected control sensors to be at least 500 m away from study sites or other large greenspaces (Hamada & Ohta, 2010; Vaz Monteiro et al., 2016). Control sensors were attached to poles in parking lots. If two study sites were very close together, and therefore had the same surrounding neighborhood, only one control sensor was deployed for both sites. In total, we deployed 97 park sensors and 18 control sensors, for a total of 115 sensors. After collection, three sensors were malfunctioning and 14 were no longer at the site when we went to collect them, resulting in 98 full sensor datasets (see Appendix S1: Table S2 for sensor location details). One agricultural site had all three sensors stolen, resulting in 11 sites with temperature data for previously forested and previously industrial land‐use legacies, and 10 sites for previously agricultural. To deal with missing control sensors, we supplemented our data using a dataset collected in the same summer, using the same type of sensor, with similar criteria to us or used the data from the next closest control site. For the dataset provided to us by a similar project, the control sensor was placed on a roadside, not in a parking lot. We conducted a sensitivity analysis to ensure that the data followed the same trends as other control sensors in our study (Appendix S1: Figure S3).

Neighborhood demographics

We calculated the neighborhood demographics surrounding each study site using census data. Based on previous work on park access in Quebec (Apparicio et al., 2010; Robillard et al., 2023), we selected 1000 m as our buffer distance to include residents who may use the park regularly. Using census variables, we calculated population density, median income, proportion of people with low income, recent immigrants, visible minorities, and those who have a bachelor's degree or higher for the neighborhoods within 1000 m surrounding each of our study parks at the dissemination area level (Government of Canada, 2021; Pham et al., 2013; Statistics Canada, 2022). When dissemination areas crossed buffer boundaries, we calculated the area that intersected the buffer and used a spatially weighted mean for calculations. For each past land use type, we calculated the mean of each of our variables of interest within the industrial, agricultural, and forested groups. We then collected the data for each of our variables of interest at the broader island level (Statistics Canada, 2022). To obtain the difference between communities surrounding parks of each previous land use type and the island of Montreal averages, we subtracted the values for the island from the mean values for the communities surrounding previously industrial, agricultural, and forested areas.

Statistical analysis

We calculated tree species richness, basal area, and size composition at the plot level. We calculated all forest structure indices separately for trees <5.0 cm dbh and trees >5.0 cm dbh. Basal area was calculated by summing the cross‐sectional area of each tree at breast height and then dividing the total by the plot size for a final unit of square meters per hectare. By calculating basal area instead of stem abundance, we were able to compare the density of trees <5.0 cm even when plot size differed as they shared a common unit (in square meters per hectare). Size composition was calculated by assessing the median dbh across all trees within a plot. For mini‐plots, where trees <5.0 cm dbh were classified into bins of 1–3 cm and 3–5 cm, for each recorded tree we randomly sampled values within the size categories to produce a dataset of real numbers. We only calculated species richness for trees ≥5 cm dbh due to the differences in sampling methodologies across the season for trees <5 cm dbh. Since all plots were sampled at 100% coverage, we calculated species richness by summing the number of unique species found at each site.

To assess the cooling benefits provided by each plot, we calculated the average temperature of daytime and nighttime hours per date for each sensor. Using the date, latitude, and longitude of each sensor, sunrise and sunset times were calculated via the “suncalc” package (Thieurmel & Elmarhraoui, 2022). Sensors logged temperature every 15 min, and we classified each logged point as daytime if it was between sunrise and sunset or nighttime if it was before sunrise and after sunset. We then grouped the data by plot and date and calculated a mean temperature for daytime and nighttime. We did the same for each of the control sensors. We then calculated the cooling effect, separated by day and night for each date of sampling (n = 64 dates total). The cooling effect was calculated by taking the average temperature of each plot per day and subtracting the average temperature of the park's control site for that day. We built our models using the causal framework outlined in Figure 1. We used five Bayesian general linear mixed‐effects models to assess the effects that we were interested in (full model equations found in Appendix S1: Figure S4). We opted for Bayesian models due to their flexibility, generative nature, and production of probability distributions for effect sizes and comparisons instead of point estimates. To assess the total effect of past land‐use on cooling effect, we modeled cooling effect as the response, with past land‐use type, time of day, and age of park as fixed effects. We also had an interaction effect between time of day, that is, nighttime or daytime, and past land‐use type. Plot and park identification code (ID) were varying effects. To assess the total effects of past land‐use type on forest structure, we produced three models with tree basal area, tree size, and tree diversity as response variables and past land‐use type as a fixed explanatory variable for each. Then, to test the direct effects of forest structure on cooling, we modeled cooling effect as the response, with tree basal area, tree size, and tree diversity as fixed effects. We also included time of day, that is, nighttime or daytime, as an interacting effect with each of the tree variables. Plot and park ID were included as varying effects. For all models, numeric variables were centered and scaled before modeling, and weakly informative priors were selected. We define weakly informative priors as priors that increase the probability of biologically possible values for our variables while still allowing the possibility of extremely strong relationships. Setting weakly informative priors can improve chain convergence without affecting model performance or interpretation (McElreath, 2020). For our models with tree structure as response variables, we computed the global mean contrasts between each group (i.e., agricultural, forested, industrial) to assess differences between past land‐uses using the posterior distributions of each model. For temperature models, the continuous explanatory variables in the model were set to their average values when computing the contrasts. We performed prior predictive checks using “tidybayes” (Kay & Mastny, 2023) and “ggplot2” (Wickham et al., 2023) packages to ensure that priors were appropriate (Appendix S1: Figure S5). We assessed model convergence by ensuring that all Rhat values were below 1.01, verifying that Markov chain Monte Carlo (MCMC) chains appeared well mixed through inspection of trace plots, and confirmed that there were minimal divergent transitions (<0.5%). To check if models were well specified, posterior predictive checks were completed. All model diagnostics can be found in Appendix S1: Figure S6. Data preparation and cleaning, spatial analysis, and modeling was done using R 4.3.0 (Bürkner et al., 2023; Pebesma, Bivand, et al., 2023; Pebesma, Sumner, et al., 2023; R Core Team, 2023).

RESULTS

Overview

We sampled temperature across 61 days from June 1 to July 31, 2023. The mean temperature difference between park and control sites averaged across all days was 1.36 ± 1.0°C, with park plots nearly always cooler than control sites (98% of measurements). The cooling effect per day across all plots ranged from −0.8 to +7.8°C, with a positive cooling effect indicating park plots providing more cooling than control sites. For forest structure variables, we measured a total of 11,053 trees across all parks. For trees ≥5 cm dbh at the plot scale, mean basal area ranged from 0.29 to 55.0 m²/ha, median size ranged from 6.4 cm dbh to 64.6 cm dbh, and mean species richness ranged from 1 to 11 species per park. Trees <5 cm dbh ranged in basal area from 0.0 to 7.9 m²/ha and median tree size ranged from 1.1 to 4.6 cm dbh.

Total effect of past land‐use on cooling

We found some differences in cooling effect across land‐use legacies (Figure 3). All past land‐use types cooled more during the day than at night (Figure 3). Previously forested plots cooled the most overall, with a mean daytime cooling effect of 1.5 ± 0.08°C and a mean nighttime cooling effect of 1.4 ± 0.08°C. Previously agricultural sites were similar, with a mean daytime cooling effect of 1.4 ± 0.08°C and a mean nighttime cooling effect of 1.3 ± 0.08°C. Previously industrial sites had the largest difference between daytime and nighttime cooling effects. Their daytime cooling effects were comparable to previously forested and agricultural sites, with a mean of 1.4 ± 0.09°C. However, previously industrial sites had the lowest nighttime cooling effects with a mean of 1.0 ± 0.09°C (Figure 3).

FIGURE 3.

Visualization of the posterior draws for the model assessing the total effect of past land use on the cooling effect of parks. The y‐axis indicates the density of the probability distributions, and the x‐axis indicates the relative cooling effect, which is the cooling effect relative to the average cooling effect across the entire dataset. A dot represents the mean, and linear intervals indicate the 66% and 95% credible intervals. Night and daytime averages were calculated by categorizing each temperature measurement into day or night categories based on the sunrise and sunset times of that day. Temperature was collected over 61 days in 2023.

Direct effect of past land‐use on forest structure

We observed some differences in forest structure across different past land‐use types (Figure 4). For large trees (dbh ≥ 5 cm), previously forested sites had the highest tree basal area. Previously forested sites had basal areas that were on average 0.72 ± 2.46 m²/ha higher than previously agricultural sites and 2.72 ± 2.80 m²/ha higher than previously industrial sites. Previously industrial sites had the largest trees ≥5 cm dbh, with a median dbh 8.8 ± 3.6 cm higher on average than previously forested sites and 3.9 ± 3.4 cm higher on average than previously agricultural sites. Previously agricultural sites were 4.8 ± 3.4 cm larger on average than previously forested sites (Figure 4). Similar to large tree basal area, species richness was highest in previously forested sites, but differences were small overall (Figure 4) Previously forested sites had on average 0.8 ± 0.5 more species than previously agricultural sites and 1.1 ± 0.5 more species than industrial sites. Previously agricultural sites had on average 0.3 ± 0.5 more species than previously industrial sites. Regarding small (dbh <5 cm) basal area, previously agricultural sites had the highest basal area, with on average 0.56 ± 0.49 m²/ha higher basal area than previously forested sites and 0.22 ± 0.55 m²/ha higher basal area than previously industrial sites. Previously industrial sites had higher small tree basal area than previously forested sites, on average 0.33 ± 0.67 m²/ha higher. Consistent with large trees, we found that previously industrial sites had the largest small trees (Figure 4). Previously industrial sites had on average 0.5 ± 0.2 cm larger trees than previously forested sites and 0.2 ± 0.2 cm larger trees than previously agricultural sites. Previously agricultural sites had 0.3 ± 0.2 cm larger trees than previously forested sites on average.

FIGURE 4.

Visualization of the pairwise contrasts of the posterior draws for each combination of past land‐use type to assess the direct effects of past land use on forest structure of parks. The top three panels show the relationships between cooling and forest structure of large trees (trees with a dbh of ≥5 cm). Bottom two panels show the relationships between cooling and forest structure of small trees (trees with dbh <5 cm). Dot represents the global mean of the contrast; linear intervals indicate 66% and 95% credible intervals. Contrasts can be interpreted as the effect of switching from the state on the right, for example, forested, to the state on the left, for example, agricultural.

Direct effect of forest structure on cooling

We found evidence that differences in forest structure alter the levels of cooling provided by parks (Figure 5). For trees ≥5 cm dbh, we found a positive effect between tree basal area and cooling during the day and a negative effect between tree basal area and cooling at night. Large tree size and species richness had little to no effect on cooling. For trees <5 cm dbh, the relationship between basal area and cooling switches—we found a positive relationship between small tree basal area and cooling during nighttime and a slightly negative effect between small tree basal area and cooling during the day. Small tree size had a weak positive relationship with cooling during the day and a weak negative relationship with cooling during the night (Figure 5).

FIGURE 5.

Visualization of the predicted relationships between forest structure and relative cooling effect, calculated while holding all other variables in the model constant and using time of day as an interaction effect. Predicted draws were calculated 1000 times, and then extracted and plotted. Colored lines indicate the conditional expectation of the model to the data, with gray credible intervals at 0.50, 0.8, and 0.95. Top three panels show the relationships between cooling and forest structure of large trees (trees with a dbh of ≥5 cm). Bottom two panels show the relationships between cooling and forest structure of small trees (trees with dbh <5 cm). Points represent raw data, colored by time of day.

Neighborhood sociodemographic composition

Due to data limitations, we were not able to causally test the relationships between historical land use and current community demographics. Instead, we offer descriptive summaries of the communities surrounding our sites using current census data. Using these data, we did find differences among communities surrounding past land use types. However, we also found evidence for general systemic inequity in the immediate areas surrounding our parks, consistent with previous literature (Table 1). There was a higher proportion of residents with bachelor's degrees or higher, and a lower proportion of visible minorities in the areas immediately surrounding parks of all historical legacies than we would expect given the island of Montreal's averages (Table 1). Using census data, we found trends that support what we would expect regarding legacy effects of environmental injustice when looking at the areas surrounding previously industrial parks. Areas surrounding previously industrial parks had a lower median income and prevalence of single detached homes, and the highest population density when compared to Montreal's averages and other land use types.

TABLE 1.

Mean difference in socioeconomic variables between neighborhoods surrounding each park type and island of Montreal's averages, for parks with each of three different historical land uses.

Past land use	Population density (per km2)	Median total income ($)	Single detached homes (% residents)	Visible minorities (% residents)	Recent immigrants (% residents)	Attained a bachelor's degree or higher (% residents)
Industrial	7683 ± 2576	−1849 ± 2440	−7.3 ± 7.5	−10.4 ± 10.4	3.5 ± 2.0	6.8 ± 13.6
Agricultural	2537 ± 6041	1230 ± 3855	28.3 ± 39.6	−8.6 ± 9.9	2.1 ± 2.9	14.4 ± 11.0
Forested	1571 ± 3498	698 ± 2908	21.2 ± 28.6	−3.4 ± 11.0	3 ± 3.1	5.1 ± 12.2

Note: Positive values indicate a value higher than the island's average and negative values indicate a value lower than the island's average. SDs are presented beside means.

DISCUSSION

We leveraged historical archives, urban forestry fieldwork, and sensor data to answer two research questions in Montreal, QC. For our first question, we found support for our hypothesis that capacity for cooling differs across previous land‐use types, most notably the cooling effect of previously industrial sites at nighttime is lower than that at previously agricultural and previously forested sites. Further, we found that forest structure had some differences across land‐use types. Previously forested parks had smaller trees with a higher basal area and a slight increase (~1 species on average) in species richness. In contrast, previously industrial sites had lower basal area but larger trees on average, while previously agricultural sites had the highest density of trees <5 cm dbh. We also found partial evidence to support our hypothesis for our second question. We found that communities surrounding previously industrial parks had lower median incomes and higher population densities when compared to previously agricultural and forested parks and to the Montreal area average, indicating persistent effects of environmental injustice.

Effect of past land‐use on cooling

Studies have shown canopy cooling effects are typically stronger during the day (Wujeska‐Klause & Pfautsch, 2020; Ziter et al., 2019). Accordingly, our findings revealed that all park types showed a stronger capacity for cooling during the daytime than during nighttime. The similar levels of daytime cooling across all park types are likely attributed to current park management and suggest that parks, regardless of history, can provide critical ecosystem services. However, we observed a difference in cooling capacity across past land‐use types at night. This difference may reflect differences in the broader urban heat island effect, which is often felt most intensely at night (Iungman et al., 2023). Historical industrial sites are generally located in denser urban areas and transformed into smaller parks, which have lower transpiration rates and reduced shade provision, ultimately leading to less cooling at night (Ziter et al., 2019). As such, these parks may show a reduced ability to combat urban heat island effects. This is further supported by the nighttime cooling effect observed in previously forested parks, which are larger in size compared to previously industrial and agricultural parks and maintained the highest levels of nighttime cooling. Additionally, previously forested parks were more often located in lower density areas, which show less of an urban heat island effect, resulting in cooler nighttime temperatures. Thus, we concluded that land‐use legacy effects on cooling capacity may manifest not only via changes to forest structure but also the landscape context in which we typically find parks with different historical land‐uses. These nighttime differences in cooling capacity are especially important as nighttime temperature is often considered a better predictor of mortality and morbidity (Sarofim et al., 2016). Although our study demonstrated differences in park cooling capacity by historical land use, there are also microclimatic variables that impact people's perception of, and response to, heat such as relative humidity and wind. Investigating how land‐use legacies, along with the influences of other microclimatic variables, would be a useful addition to the literature.

Effect of past land‐use on forest structure

Forest structure also differed across past land‐use types, especially in relation to basal area and tree size. Consistent with other studies that found temporal distance from anthropogenic disturbance is positively correlated with basal area, previously forested sites had smaller trees with higher plot‐level basal area (Pregitzer & Bradford, 2023; Thompson et al., 2002). One limitation when working with historical data is that it is often not possible to find the complete history of a site. As such, we were unable to determine the logging and management history of the previously forested sites, though the small, dense stands suggest that these areas are relatively young forests (Thompson et al., 2002). Previously industrial sites were characterized by lower basal area but higher median tree size when compared to other past land‐use types. Higher tree size in previously industrial sites may be due to the approach taken during park establishment. Industrial sites have the topsoil replaced upon restoration and conversion to a park, and their woodlands often start with intentional planting of saplings due to low natural recruitment (Ashwood et al., 2022; Robinson et al., 1992). Newly planted saplings in Montreal are generally required to be between 5 and 7 cm dbh, which are larger than many of the naturally establishing saplings in the previously agricultural and forested sites (Anjou, 2019; Rivière‐des‐Prairies–Pointe‐aux‐Trembles, 2013; Saint‐Laurent, 2018). These results suggest that there is a density–size trade‐off present in urban parks that is driving differences across past land‐use types. This trade‐off has been previously documented, as large trees can be difficult to maintain and manage, especially in high volumes (Endreny et al., 2017; McPherson et al., 1997; Soto et al., 2018). Therefore, urban forest stands often alternate between two options: smaller, dense trees or larger, sparser trees, as a strategy to balance size, density, and cover.

One similarity in forest structure that we found across past land‐use types was species richness, where previously forested sites had slightly higher levels than previously industrial and agricultural sites but only by an average of ~1 species. This finding is in contrast to a study that found species composition of New York City forests is dependent on land‐use history (Pregitzer & Bradford, 2023). Our finding is a testament to the power of park management and shows that maintaining tree diversity is possible across a city with varying conditions through cohesive management (Aznarez et al., 2022). However, we recognize that species richness is not always the most meaningful measure of diversity and has no effect on cooling in our study. Future studies looking at how legacy effects influence functional diversity and the presence of native versus invasive species (Pregitzer & Bradford, 2023) would be an interesting extension of this research.

Effect of forest structure on cooling

For trees ≥5 cm dbh, cooling is driven mainly by plot basal area. We observed a positive relationship between cooling and basal area during the day, and a negative relationship between cooling and basal area during the night. This finding aligns with literature that shows higher tree density typically results in more tree cover and thus higher rates of transpiration and shade provision, which increases cooling during the day (He et al., 2021; Zhao et al., 2023; Ziter et al., 2019). However, we would expect this relationship to hold with tree size, and it does not. We found no relationship between tree size for trees ≥5 cm dbh and cooling. This is contrary to our predictions, as there are many studies that demonstrate that larger size of trees in parks results in greater cooling (Helletsgruber et al., 2020; Rahman et al., 2020). It is plausible that our contrary results are due to the density‐size trade‐off. Due to planting and maintenance strategies in urban park management, development patterns, and resource constraints, it is rare to find a park that has both large and dense trees (e.g., see Appendix S1: Figure S8). Further, the largest trees in urban parks are often surrounded by mowed turfgrass, leading to a mitigation of the cooling effect, especially on hot days (Kraemer & Kabisch, 2022; Rahman et al., 2020). In contrast, more “natural” or forest‐like conditions in our study parks tended to have denser stands of small trees, surrounded by herbaceous vegetation. Our findings suggest that large trees alone are not enough to provide cooling in urban parks, with high tree density and/or cover providing stronger daytime cooling. Although we were not able to measure understory structure in our study, literature suggests that it can also impact cooling (Fung & Jim, 2019; Stickley & Fraterrigo, 2021) and should be considered in the future as a potentially important component of urban park cooling capacity.

Increasing the basal area of trees in parks or creating more “naturalized” or rewilded areas may be good strategies to provide temperature regulation during the day. However, dense, forested parks may provide daytime cooling at the expense of nighttime cooling and the many other needs that parks serve (Gaston et al., 2013; Kremer et al., 2016). At night, high tree cover can actually trap heat, decreasing the cooling effect (Wujeska‐Klause & Pfautsch, 2020). Additionally, parks with only dense forest stands lose the opportunity to provide open space for recreation, community gardens for food provisioning, wildflower beds for pollinators, etc. To further examine the impact that forest structure can have, we suggest looking at the effects of forest structure within a single park to help inform the role of management on the capacity for cooling and balance this information with strategies to provide other desired ecosystem services.

We found that as the basal area of small trees increases, the nighttime cooling effect increases, and the daytime cooling effect is limited. This directly conflicts with what we would expect based on the literature and our results for trees ≥5 cm dbh. We hypothesize that this relationship is due to a trade‐off between large and small trees in our system. Areas with a high basal area of trees <5 cm dbh—dense, shrubby stands—are less likely to contain a large shade tree in the parks we studied; therefore, heat can escape at night and results in a positive relationship between the basal area of trees <5 cm dbh and nighttime cooling. However, we do see the expected relationship with tree size given the literature, where larger trees increase daytime cooling because of increased shade provision but trap heat at night (Wujeska‐Klause & Pfautsch, 2020). We found that an increase in median tree size causes an increase in daytime cooling and a decrease in nighttime cooling.

Communities surrounding past land‐use types

Our second hypothesis was partially supported. Industrial areas in Montreal have a history of systemic inequity, with neighborhoods surrounding areas that were used for quarrying, garbage dumping, and/or other industrial activities being inhabited by working‐class or low‐income people, who are often recently immigrated, and are majority Black or Brown (Kikano et al., 2023; Schell et al., 2020; Waldron, 2018). Our results provide evidence that park establishment in these areas can provide daytime cooling, though the critical service of nighttime cooling is lacking. With differentiated management that incorporates strategies to improve nighttime cooling while also maintaining daytime cooling, parks established on previously industrial land have the potential to counter a history of inequity in these neighborhoods, as they are providing important natural benefits to people with lower median incomes living in denser populations.

A natural conclusion to draw from our results is that previously industrial areas should be priority sites for the development of new urban green spaces, to increase nighttime cooling capacity and combat the urban heat island effect. However, we are wary of making that recommendation, as we found that communities immediately surrounding parks in all three of our past land‐use types have lower proportions of visible minorities and higher proportions of highly educated people than average for the island of Montreal. This finding supports the luxury effect—an established relationship across North America between wealth/privilege and plant diversity (Hope et al., 2003) that has been expanded to include vegetation metrics beyond diversity such as greenness (Mennis, 2006), vegetation cover (Clarke et al., 2013), and more (Schell et al., 2020). This indicates to us that efforts to provide urban nature, involving the establishment or restoration of urban green spaces, may fail in their mission and result in gentrification (Anguelovski, 2016; Sax et al., 2022). Green gentrification is an issue that we commonly see in Canada and around the world, and which can only be solved by stronger housing regulations and community‐led policies (Goossens et al., 2020; Parish, 2020; Triguero‐Mas et al., 2021). Further investigation into the connections between historical injustice, ecological capacity of greenspaces, and gentrification is warranted. We recommend that moving forward, urban greening is done in collaboration with housing and urban planning departments, and driven by community‐led greening initiatives.

In conclusion, we found that incorporating historical context can be a useful way to understand differences in contemporary ecological and societal conditions. Building on this study by incorporating other key ecosystem services such as air pollution mitigation, recreation provision, cultural importance, and habitat provisioning is integral to discovering how our histories influence our current day service provision, as well as informing how our current decision‐making will influence our future service provision. We see through this work that the restoration of industrial sites has the potential to provide critical ecosystem services to equity‐seeking groups and suggest this as a promising strategy to provide urban nature as long as it is done intentionally and led by the community it is serving. Finally, our results support that building greenspaces on all different types of sites, from forest remnants to quarries, results in the capacity for daytime cooling and comparable tree diversity, critical benefits to our cities.

AUTHOR CONTRIBUTIONS

Isabella C. Richmond contributed to the conceptualization, data collection, analysis, writing, and editing. Michael A. Paulauskas, Erica Padvaiskas, and Laura Carolina Gonzàlez Sinisterra contributed to the conceptualization, data collection, and editing. Kayleigh Hutt‐Taylor contributed to the conceptualization, analysis, and editing. Alec L. Robitaille contributed to the analysis. Carly D. Ziter contributed to the conceptualization and editing.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Appendix S1.

Richmond, Isabella C. , Paulauskas Michael A., Padvaiskas Erica, Gonzàlez Sinisterra Laura Carolina, Hutt‐Taylor Kayleigh, Robitaille Alec L., and Ziter Carly D.. 2025. “Land‐Use History Causes Differences in Park Nighttime Cooling Capacity and Forest Structure.” Ecological Applications 35(5): e70082. 10.1002/eap.70082

DATA AVAILABILITY STATEMENT

Data and code (Richmond, 2023) are available from GitHub at https://github.com/zule-lab/parklegacies and Zenodo at https://doi.org/10.5281/zenodo.10406628.