Mapping the diversity of street tree inventories across eight cities internationally using open data

Tree diversity, on a species, genus, and family-level, is an important factor in securing healthy urban forests and providing ecosystem services for billions of city dwellers. Using open-source data on global tree inventories, this study examines (1) the diversity of species, genera, and family of urban street trees in eight cities internationally; (2) how they score on diversity benchmarks and indices; and (3) the diversity variation inside and outside of cities ’ centers. We hypothesized most cities would score poorly on diversity benchmarks and spatial patterns in species composition would exist, as illustrated by established relationships between urban density and urban tree diversity. Results indicate city centers were less likely to approach the proposed diversity benchmarks than outside the city center. Overall, both Shannon and Simpson diversity indices show greater diversity outside of the city center, especially at the species-level. Understanding street tree diversity and spatial variation patterns across cities internationally can offer needed evidence to back up heuristic benchmarks. The methodology and open-source data used in this study are intended to enable practitioners better target tree diversity efforts.


Introduction
Tree diversity provides a basis for evaluating a city's ability to support a healthy urban forest (Isbell et al., 2011;Morgenroth et al., 2016).Diverse urban forests, both in terms of structure and species composition, ensure their resilience to the challenges of the urban environment (Ordóñez and Duinker, 2014).Accounting for diversity when managing an urban forest can help support and enhance ecosystem services (i.e., benefits nature provides to human well-being) delivered and minimize management expenses (Conway and Vander Vecht, 2015).There is mounting evidence that species diversity in particular enhances ecosystem resistance to bio-and ecological disturbances (Dobbs et al., 2014;Riley et al., 2018).Urban forests with lower species richness face greater risks of mass mortality and canopy cover loss from the introduction of species-or genus-specific pests and disease-exemplified by the mass mortality events-of elm (Karnosky, 1979;Steenberg et al., 2013), ash (Kovacs et al., 2010(Kovacs et al., , 2014;;Sadof, 2016), chestnut (Rigling and Prospero, 2018), and oak trees (Haight et al., 2011;Juzwik et al., 2011), by Dutch elm disease, emerald ash borer disease, chestnut blight fungus, and oak wilt fungus, respectively, which have collectively killed over 100 million urban trees across North America.An urban forest dominated by monocultures, in the case of invasion events, often subsequently incurs high costs to treat, remove, and/or replace diseased, dying, or dead trees (Laćan and McBride, 2008;Nitoslawski and Duinker, 2016).Greater diversity also provides habitat for a wider range of organisms, especially native species in some cases; when these are well-represented, they can contribute to local biodiversity protection (Alvey, 2006;Almas and Conway, 2016).Altogether, a healthy, diverse urban forest contributes to biodiversity conservation (SDG 15), tackling climate change (SDG 13), improving urban habitats (SDG 11;Endreny, 2018), providing safe drinking water (SDG 6), supporting food security (SDG 2), and reducing inequalities (SDG 10;O'Brien et al., 2010).
Over the years, tree diversity benchmarks have been put forward to measure and monitor species richness (i.e., the number of different tree species) and evenness (i.e., the representation of a given species within a total number of individuals) across the urban forest (Kendal et al., 2014).For example, the 10/20/30 benchmark proposed by Frank Santamour (1990) states that urban forests should comprise no more than 10 % of any particular species, 20 % of any genus or 30 % of any single-family.Santamour's aim was to mitigate risk, and although there is relatively little empirical evidence to support this approach, it has been widely used since in urban forest management plans (Kendal et al., 2014).In more recent years, scholars and practitioners have called for more rigid benchmarks, such as 5/10/15, to further mitigate risks associated with mass mortality events and other environmental stressors that could negatively impact the urban forest canopy (Leff, 2016;Watson, 2018).Other diversity measurements include mathematically computed diversity indices, ubiquitous in ecological research.Of the many species diversity indices used in the literature, the Shannon Index, or Shannon-Wiener Index (Shannon, 1948), is most commonly used.The Simpson Index (Simpson, 1949) is also frequently used and accounts for both species abundance and evenness.
A broader understanding of urban tree diversity and spatial variation patterns across cities internationally can offer needed evidence to back up heuristic benchmarks and help local practitioners better target tree diversity efforts.Smart-city concepts and technologies may provide consistent, reliable, and precise data to foster proactive management and effective policy making (Nitoslawski et al., 2019;Galle et al., 2019).Open data, specifically, by which municipal and other government data is made widely available for public use, provides further opportunity to study urban forest patterns and dynamics across a wide range of cities-which may have previously been much more difficult and laborious (Steenberg et al., 2018).This paper uses data from OpenTrees.org, the world's largest database of collated publicly available municipal tree inventories.To facilitate and stimulate further research and analysis of additional cities, we have used only open-source data and software in this study and we encourage researchers, practitioners, and citizen scientists to quantify their own city's diversity via the Diversitree project website and open-source project repository, which are the companion tools to this paper. 1his study examines (1) the diversity of species, genera, and family of street trees in eight cities internationally; (2) how they score on two tree diversity benchmarks and two diversity indices; and (3) the comparative diversity inside and outside of city centers.We hypothesized relative abundance will be higher than diversity benchmarks at the species-level, but comparable at the genus-and family-level, based on Kendal et al.'s (2014) previous global findings.We also hypothesized that there would be differences in species composition based on location-in this case, inside and outside of city centers.This is the first paper to assess these spatial variations in tree diversity using open-source tree inventory data to draw comparisons across cities internationally.

Understanding and measuring tree diversity
Tree diversity metrics and benchmarks provide concrete figures that are relatively standardized and comparable, which can prove useful for calibrating planting and management targets, as well as carrying out broader diversity analyses across multiple locales.Recognizing that tree diversity is crucial to urban forest resilience, and following studies illustrating that many urban areas fail to reach the 10/20/30 benchmark (Kendal et al., 2014), some urban forest managers have called for higher diversity standards.In 2015, the City of Portland, Oregon, in the United States, adopted a revised 5/10/20 benchmark, reducing the rule that urban forests should comprise no more than 10 % of any particular species to 5%.In 2018, the Morton Arboretum officially proposed a new benchmark: 5/10/15, where urban forests should comprise no more than 5% of any particular species, 10 % of any genus or 15 % of any single-family.While easily put forward in theory, the strictest taxonomic diversity target of 5/10/15 may be difficult to implement in practice due to limited choice at local nurseries, existing traditions and customs, and cost and uncertainty of maintaining different species (Urban, 2008;Morgenroth et al., 2016).When studying the composition, condition, and structure of 12 urban forests across Great Britain, Monteiro et al. (2020) found none of the locations met the 5/10/15 benchmark.In a similar study across 10 Nordic cities, none of the locations met the 5/10/15 benchmark either (Sjöman et al., 2012).In a study of Bangalore's (India) street trees, the most dominant species accounted for 9% of the total population, still above the 5/10/15 benchmark (Nagendra and Gopal, 2010).Bangkok (Thailand) gets closer with no species exceeding 7% of the total (Thaiutsa et al., 2008).Bangalore and Bangkok's findings confer with Kjelgren et al. (2011) which found tropical climates offer a larger selection pool of resilient tree species for increasing tropical urban tree diversity (Jim and Liu, 2001).
When quantifying tree diversity-whether through benchmarks or indices-scale, population, and units of analysis must be considered.Urban forests can be analysed at various levels, including the plot-, neighbourhood-, city-, or regional-level.Differences already exist across North America and Europe, where "urban" forests are generally considered only in cities, or at regional-level, respectively (Konijnendijk et al., 2006;Simpson, 1949).Despite urban forests being defined as "all trees in a city" by Konijnendijk et al. (2006), in some cases, "urban forest" is still used synonymously to mean only street trees, park trees, or all public trees (Pregitzer et al., 2019).The metric on which researchers and municipalities choose to analyse trees also differ.Diversity can be calculated on the basis of stem count (i.e., the number of trees of that specific species) or basal area (BA; i.e., the cross-sectional area of a tree stem measured at breast height, or 1.4 m).Originally, Santamour presented his 10/20/30 benchmark in stem count, which can be useful for planting goals.However, using BA can offer a marker for canopy availability for wildlife, as BA correlates with attributes such as size, leaf surface area, and biomass (Slik et al., 2010).Tree size is also important for calculations of ecosystem services, real estate values, tree removal, and replacement costs (Nowak and Dwyer, 2007;Song et al., 2018).BA is more typically used in forestry (e.g.timber harvesting); however, it is gaining popularity in urban forestry as it complements stem counts for diversity indices.Ideally, both stem count and BA are useful for comprehensive studies of urban forest diversity (Nitoslawski and Duinker, 2016).
It is just as useful to understand how patterns in vegetation diversity change across the urban landscape, in order to quantify the provisioning of ecosystem services to urban dwellers at finer scales (Bourne and Conway, 2014;Dobbs et al., 2014).Cities are complex and heterogeneous landscapes, reflecting differences in resident socio-demographics, land use types, climatic and geographical variables, and governance structures.Urban-rural gradient studies, as an example, have proven useful for assessing degrees of interactions between social (e.g.population, urban form, density) and ecological (e.g.topography, landscape use and history) variables driving urban tree diversity (Ranta and Viljanen, 2011).
It has been suggested that vegetation species richness peaks in suburban and peri-urban landscapes (Alvey, 2006;McKinney, 2008a,b;Turner et al., 2005).The intermediate disturbance hypothesis has been offered as a potentially common pattern manifesting across many urban landscapes, attributed to environmental and habitat heterogeneity created by various actors and jurisdictions, along with individual management practices and preferences occurring at finer scales (Hansen et al., 2005;McKinney, 2002;Pautasso and Gaston, 2006).Indeed, research has shown that species richness might be driven by land cover type and presence of habitat edges-such as the periphery of a subdivision development ( Čepelová and Münzbergová, 2012;McWilliam et al., 2014).Residential properties, particularly single-family homes with gardens and yards, also play an influential role; planting and tree maintenance preferences could depend on historical neighborhood development patterns (Doody et al., 2010;Nitoslawski et al., 2016), culture and ethnicity (Kinzig et al., 2005), age and education level (Kirkpatrick et al., 2012;Meléndez-Ackerman et al., 2014), income (Hope et al., 2003), and ownership status (Kendal et al., 2012).The presence and design of parks and other green spaces found in suburban and peri-urban areas may also influence tree-species richness.Factors include size (Godefroid and Koedam, 2003), shape (Pennington et al., 2010), and type of recreational activity and usership (Vakhlamova et al., 2016).
Most relevant to our study, the diversity of trees along streets, more often than not maintained by the municipality (Steenberg, 2018), may depend on nursery stock availability (Conway and Vander Vecht, 2015), geography and climate (Ramage et al., 2013), operational and management needs (such as snow removal in colder climates and irrigation needs in warmer and drier ones), and contemporary priorities and challenges, such as climate change (Ordóñez and Duinker, 2014) and invasive species (Raupp et al., 2006).More recently, researchers have further explored the role of neighbourhood age and development history (Avolio et al., 2018), resident preferences (Avolio et al., 2018), local governmental planting programs (Falfán and MacGregor-Fors, 2020), and cultural legacies and human-nature relationships (Laurian, 2019).These examples illustrate the varying temporal and spatial scales at play; invariably, despite the fact that some studies have identified common patterns across different cities globally, local context is an important consideration when identifying drivers of urban street tree diversity.

Data collection
We obtained inventories of urban trees from eight cities across five different Köppen climate classifications from OpenTrees.org,which harvests open-source tree inventory data from municipalities.Cities included in the study were Cambridge, MA (USA); Vancouver (Canada); Buenos Aires (Argentina); Bologna (Italy); Amsterdam (The Netherlands); Oslo (Norway); Paris (France); and Melbourne (Australia). 2These cities were selected to offer a diverse breadth across geographies and climate zones; although, at the time of writing, Open-Trees.org lacked tree inventory data on Asian and African cities and thus no cities from those continents were included.
For this study, only street trees were used in the analysis.Street trees were defined as trees within a reasonable catchment of streets or roads (explained below), defined either by existing data tags or proximity to the street centerline relative to its width.Oslo's tree inventory contained only street trees, and the tree inventory of Paris tagged data with the location type.Vancouver and Buenos Aires inventories' were not explicitly street trees only, but all trees inventoried fell within the aforementioned reasonable catchment of streets.For Amsterdam, Bologna, Cambridge, and Melbourne, in OpenStreetMap (OSM), streets (any line/way segment with a "highway" tag) were selected and filtered for only non-pedestrian streets (e.g.cycleways, paths, steps, etc. were irrelevant to this study).A Euclidean buffer to the centerline of each street segment, based on two times the width of the street, estimated from the number of lanes by a typical lane width or typical road of that type or from a width field if present, was used.Tree inventories were then clipped based on this area, excluding trees in parks and areas outside of the urban grid.The trees removed from the inventory represented between ~32 % of the data (Cambridge) to 48 % (Paris).For Paris, the impact of this data cleaning procedure had a major impact on the results; the calculated relative dominance of the London Plane species (Platanus × acerifolia) in the Paris city center reduced from 58.6%-41.6%.
All inventories used were conducted using field surveys and had information at the species-and genus-level, but family-level information was missing from all but two of the inventories.Missing data was filled in using the Catalog of Life's 2019 Checklist release, a comprehensive species taxonomy database.3

Data cleansing and preprocessing
Tree inventories were imported from various geospatial or commaseparated values (CSV) file formats to geo-package (SQLite) files for tree point data and for geographic units of analysis.The data was cleaned by removing any entries that lacked species data or diameter at breast height (DBH) columns.Across several inventories, data integrity issues on diameter measurements persisted, and a decision was made to remove the 99th percentile of trees by DBH from each data set.This removed outlier trees with excessively large sizes, in one case a reported circumference of 2.5 km.After removing incomplete and erroneous data, all inventories had reasonable and comparable mean and maximum DBH measurements, respectively ~20− 30 cm and ~70− 90 cm.Amsterdam and Bologna had classifications instead of absolute measurements for DBH; therefore, the average value for each classification range was selected.
Using Python libraries Pandas and GeoPandas in a Jupyter notebook, the following indices and statistics were calculated for each city center and outlying area: stem count and BA, maximum relative abundance and dominance, Shannon Index, and Simpson Index, at the species-, genus-, and family-level.The results are visualized in the accompanying web tool, along with additional information about the cities. 4 Due to discrepancies in city center boundaries and definitions, each city's center was defined on a case-by-case basis (Table 1; Fig. 1).Generally, the center boundaries reflect the densest historic, administrative, or commercial urban area.
The present study computed the relative abundance of the most common species, genus and family, in addition to the Shannon and Simpson diversity indices using both stem count and BA.Stem count treats small and large trees equally, while BA favours larger trees as they have a larger cross-sectional surface area.

Data analysis
For each city, two diversity indices were computed using relative dominance (i.e., BA) at species-, genus-, and family-level.For each zone, the proportional species composition was also calculated.The BA is the cross-sectional area of a tree stem measured at diameter at breast height (DBH; 1.4 m), according to the following formula (Erdle, 2012):  Relative abundance, calculated with stem counts, was used to score how well each city scored on the two diversity benchmarks (10/20/30 and 5/10/15).The Shannon Index (H1) calculates the proportion of species i relative to the total number of species (n i ), which is then multiplied by the natural logarithm of this proportion (lnn i ), summed across all species, and then multiplied by -1 (Shannon, 1948): where pi represents the count proportion of the ith species.The Shannon Index increases as the community's richness and evenness increase.
Values range from 0 to 5, but in most ecological studies, typically fall between 1.5 and 3.5, and rarely exceed 4 (Magurran, 1988).The Simpson Index (1− D) is less sensitive to species richness and heavily weighted towards the most abundant species.It gives the probability that any two individuals drawn at random from an infinitely large community belong to different species.The index calculates the proportion of species i relative to the total number of species (n i ), sums the squared proportions for all the species, and then takes the reciprocal (Simpson, 1949): Where n i represents the count proportion of the ith species.Since D is a measure of dominance, 1− D estimates species diversity, and therefore, the value of D ranges between 0 and 1. Ecological literature strongly advocates the use of both diversity indices to account for species richness and relative abundance and provide the most complete mathematical description of the data (Gorelick, 2006;Ricotta and Szeidl, 2006;Zahl, 1977).
The open-source Python scripts and Jupyter notebooks used to calculate our results will be made available via GitHub.5Individuals or municipalities may quantify their own street tree inventories and contribute their findings.

Sensitivity analysis
In all of the included study cities, the stem count and BA of trees located in the city center geographies is lower, often substantially lower, than that of the rest of the city.At low numbers, diversity indices can be heavily impacted by a low sample size.To ensure that our metrics describing trees located inside and outside of the city centers are not unduly influenced by a low count of trees and that the results of each city center and outer areas are generalizable across our data, we conducted a sensitivity analysis (Fig. 2).We randomly sampled increasing numbers of trees from each city's dataset to identify a reasonable minimum tree count that must be included in a geographic unit to be generalizable against a larger sample size.On a logarithmic sequence, we sampled values from 1 to 100,000 (or the entire dataset) and calculated diversity indices for each random sample.Each sample size was repeated for 100 iterations and the median Shannon diversity indices were taken to even out variation, particularly at smaller sample sizes.Fig. 2 presents the results of the sensitivity analysis using stem counts for each city center.
Based on this analysis, we find that for cities in the study the diversity indices saturate with data on 500-700 stems.Beyond this point increasing sample sizes have slight but very minor positive impacts on diversity indices.The leanest city center data is present in Oslo's tree inventory, which contains only 828 trees within the city center after data cleaning.While this is sparse, particularly as compared to other datasets (Amsterdam: 7,828 stems, Paris: 13,228 stems, Buenos Aires: 39,942 stems), it is sufficiently large to saturate diversity indices.This means that across all our data we can better isolate differences in diversity index outcomes to differences in spatial distribution, and not of the metrics themselves.The sensitivity analysis was also conducted for the Shannon Index using BA and for the Simpson Index using stem counts and BA (Fig. A9, in the appendix), and mirrored the results in Fig. 2.

How diverse are the world's street tree inventories?
Relative abundance, calculated with BA, was used to determine how well each city scored on the two diversity benchmarks (10/20/30; 5/10/ 15).These results are presented in Table 2. Relative abundance, calculated with stem count, is presented in Table 3.
When the most abundant species, genus, and family was calculated using BA, none of the cities met the 5/10/15 benchmark, or the 10/20/ 30 benchmark.Amsterdam (outside) and Bologna (outside) were close to reaching the 10/20/30 benchmark, 11/19/19 and 16/16/16, respectively.Oslo (inside) stood out with 52/53/53, the most severe benchmark transgression of all the study's included cities.In general, in line with the findings from Kendal et al. (2014), the relative abundance at the species-level was much higher than the proposed 10 % benchmark, but comparable with proposed benchmarks at the genus-(20 %) and family-level (30 %).
However, when the most abundant species, genus, and family was calculated using stem count, once again none of the cities met the 5/10/ 15 benchmark, but Amsterdam (outside) came close with 8/14/14, meeting the 10/20/30 benchmark.No other cities met the 10/20/30 benchmark, although many including Bologna (inside and outside),

Table 2
Diversity benchmarks of street tree inventories across eight cities internationally, calculated using basal area.Data in bold indicates the most abundant species, genus, or family that met the proposed 10/20/30 benchmark.All others failed to meet the proposed benchmark.Cambridge (outside), and Vancouver (outside) came close.As with BA, the stem count findings are aligned with those from Kendal et al. (2014), with the relative abundance at the species-level typically higher than the proposed 10 % benchmark, but also comparable with proposed benchmarks at the genus-(20 %) and family-level (30 %).In general, when the diversity benchmark was calculated using stem count, more cities met    the proposed benchmarks, or came closer to meeting them.It should be reiterated that Santamour (1990) presented his original 10/20/30 benchmark in stem count, not BA.For BA, the top three most abundant families across the eight cities were Platanaceae (i.e., "the plane-tree family"), Sapindaceae (i.e., "the soapberry family"), and Ulmaceae (i.e., "the elm family").These families were represented across all 8 cities.In addition, most of the most abundant species in each city were not native to that city, only Amsterdam with Dutch Elm (Ulmus × hollandica), Oslo with Linden (Tilia spp.; of the genus Tilia, species was not specified), and Cambridge with Honeylocust (Gleditsia triacanthos) had native species as their most abundant, meaning the urban forest would be very vulnerable to their loss.For stem count, the top two most abundant families across the eight cities were Platanaceae and Sapindaceae.These families were represented across 6 of the 8 cities.Similar findings were found in regards to the most abundant species being non-native.
The results from the Shannon and Simpson Index computations at the species-level for BA are presented in Figs. 3 and 4. Results at the specieslevel for stem count are presented in Figs. 5 and 6. Results at the genusand family-level for BA and stem count are in the Appendix.

Spatial variation of tree diversity for basal area
A paired-sample t-test was conducted to compare the Shannon and Simpson Index values between the city center and outside for each city.The BA results are in Table 4. Since the direction of the difference did not matter, a two-tailed hypothesis was used.
For the Shannon Index, on the species-level, for BA, there was a significant difference in the scores for the center 2.48 (±0.47) and outside 3.09 (±0.45) conditions; t(7)=− 3.39, p = 0.01.These results suggest Shannon diversity is lower within the city center.A greater difference in Shannon diversity exists between center and outside on the genus-and family-level, for BA, both relationships are statistically significant.On the genus-level, there was a significant difference in the scores for the center 1.71 (±0.38) and outside 2.04 (±0.25) conditions; t (7)=− 2.46, p = 0.04.On the family-level, there was also a significant difference in the scores for the center 1.88 (±0.45) and outside 2.36 (±0.27) conditions; t(7)=− 3.04, p = 0.02.Since all p-values for Shannon Diversity were less than 0.05, the relationships are statistically significant.It indicates strong evidence against the null hypothesis, as there is less than a 5% probability the null is correct (and the results are random).Therefore, we reject the null hypothesis, and accept the alternative hypothesis that there is a relationship between Shannon Diversity within the city center and outside of it for BA.Across taxonomic ranks, for all cities, city centers had less diversity than outside of the city center.
The value of Simpson's Index (1− D) ranges between 0 and 1.Like the Shannon Index, the greater the value, the greater the sample diversity.In general, the Simpson Index showed a similar pattern as the Shannon Index when contrasting city center and outside, but the relationships were less statistically significant.For the Simpson Index, on the specieslevel, for BA, there was a non-significant difference in the scores for the center 0.82 (±0.08) and outside 0.89 (±0.06) conditions; t(7)=− 2.36, p = 0.05.On the genus-level, for BA, there was also a non-significant difference in the scores for the center 0.70(±0.13)and outside 0.81 (±0.06) conditions; t(7)=− 2.19, p = 0.06.On the family-level, for BA,  there was a significant difference in the scores for the center 0.71 (±0.14) and outside 0.84 (±0.06) conditions; t(7)=− 2.46, p = 0.04.As the computed p-value for the species-and genus-level is greater than the significance level alpha = 0.05, one cannot reject the null hypothesis H0.The family-level relationship was the only relationship with a p-value of less than 0.05, and thus had a significant relationship.

Spatial variation of tree diversity for stem count
The results of the paired-sample t-test for stem count are in Table 5.
For the Shannon Index, on the species-level, for stem count, there was a significant difference in the scores for the center 2.83 (±0.46) and outside 3.60 (±0.43) conditions; t(7)=− 4.28, p = 0.004.These results suggest Shannon Diversity is lower within the city center and higher outside of it.On the genus-level, for stem count, there was a significant difference in the scores for the center 1.87 (±0.37) and outside 2.23 (±0.23) conditions; t(7)=− 3.02, p = 0.02.On the family-level, for stem count, there was also a significant difference in the scores for the center 2.15 (±0.46) and outside 2.74 (±0.27) conditions; t(7)=− 3.65, p = 0.01.Since all p-values for Shannon Diversity were less than 0.05, the relationships are statistically significant.It indicates strong evidence against the null hypothesis, and therefore, we accept the alternative hypothesis that there is a relationship between Shannon Diversity within the city center and outside of it for stem count.Like BA, across taxonomic ranks, city centers always had less diversity than outside of the city center.
The Simpson Index showed a similar pattern as the Shannon Index when contrasting city center and outside.For the Simpson Index, on the species-level, for stem count, there was a significant difference in the scores for the center 0.85 (±0.09) and outside 0.93 (±0.05) conditions; t (7)=− 2.61, p = 0.04.On the genus-level, for stem count, there was also a significant difference in the scores for the center 0.73 (±0.11) and outside 0.84 (±0.06) conditions; t(7)=− 2.5, p = 0.04.On the familylevel, for stem count, there was a significant difference in the scores for the center 0.75 (±0.12) and outside 0.88 (±0.05) conditions; t(7)=− 2.98, p = 0.02.As the computed p-value is lower than the significance level alpha = 0.05, for all taxonomic ranks, one should reject the null hypothesis H0, and accept the alternative hypothesis Ha.Like BA, across taxonomic ranks, city centers always had less diversity than outside of the city center.Our sensitivity analysis illustrates that even our smallest sample size of trees in a city center (Oslo, n = 828) is sufficient to saturate Shannon and Simpson diversity indices to a generalizable finding.The differences we observe reflect a wide number of factors, such as ecological conditions, historical and contemporary urban forestry practices, and characteristics of the urban forms themselves.

Discussion
This study provides insight into the diversity of street tree populations in eight cities internationally.The results presented help understand diversity at a spatial scale by contrasting diversity at the species-, genus-, and family-level both within and outside city centers using two diversity indices (the Shannon Index and the Simpson Index) and two approaches (BA and stem count).The methodology and opensource data used in this study are intended to inspire local tree managers and policymakers to quantify street tree diversity in their own urban forest locales and inform their urban forest management strategies accordingly.

Relation to similar studies
Six of the eight selected cities in the present study rank in the top 100 of the Husqvarna Urban Green Space Index (HUGSI): Oslo (15th), Vancouver (45th), Amsterdam (57th), Paris (84th), Buenos Aires (87th), and Melbourne (88th), meaning they are some of the 'greenest' cities in the world. 6The HUGSI applies computer vision and deep learning techniques to satellite imagery to quantify the size, proportion, distribution and health of green spaces in urban areas, as well as those of trees.However, with the exception of Amsterdam, none of the cities in the HUGSI met the proposed diversity benchmarks.This invites the question of whether these cities simply lack diversity in their otherwise ambitious urban forest management plans or whether the proposed benchmarks are unrealistic.
Zuring and Grootens (2015) examined the feasibility of the 10/20/30 benchmark as a guideline for diversity in Dutch municipal tree inventories.Eight Dutch municipalities, including Amsterdam, were tested against the 10/20/30 benchmark, and none of the eight cities were found to meet the benchmark.In the present study, only street trees were analysed, while Zuring and Grootens (2015) used an inventory of all public trees.The authors' research revealed that the 10/20/30 benchmark was unsuitable for protecting against pests and disease.The 5/10/20 benchmark was shown to be better suited to df = degree of freedom; the number of degrees of freedom is approximated by the Welch-Satterthwaite formula.
N.J.Galle et al. protecting against pests and disease but would require the municipalities included in the study to supplement over 25 % of their existing inventory to meet the benchmark.Kendal et al. (2014) used stem count in their diversity index computations and found similar results to those of the present study: relative abundance was much higher than the 10/20/30 benchmark proposed by Santamour (1990) but was comparable with proposed benchmarks at the genus-and family-level.Kendal et al. (2014) noted mean values for abundance across the 151 datasets analysed of 20 %, 26 % and 32 %.The mean values for abundance across the eight cities using stem count analysed in the present study were 24 %, 32 %, and 34 %, for the species-, genus-, and family-level, respectively, which are not unlike the results of Kendal et al.'s study.Although diversity benchmarks were not frequently met in the present study, other research has shown that urban environments may contain relatively high levels of biodiversity (Alvey, 2006).Alvey (2006) also posits that previous research and attempts to preserve global biodiversity have typically focused on natural habitats rather than urban areas in which little natural habitat remains.Recently, Ossola et al. (2020) compiled urban tree inventories and species lists from 473 cities across 73 countries, called the Global Urban Tree Inventory (GUTI).The authors found 8% of the world's known tree biodiversity within the 4, 734 species identified, one-tenth of which faces conservation threats in the wild.While nearly 5,000 species were represented across cities, the same 79 tree species were planted in over 100 cities, and the same 381 species were found in over 30 cities (Ossola et al., 2020).The most common species in the GUTI, the Ginkgo (Ginkgo biloba), was not the most abundant species in any of the cities selected in the present study.However, species of Acer, which were the GUTI's next most common species, were also the most abundant species in three of the present study's selected cities: Cambridge, Oslo, and Vancouver.Indeed, in the five other selected cities (Amsterdam, Bologna, Buenos Aires, Melbourne, and Paris) the most abundant species was the London Plane (Platanus × acerifolia).In the GUTI, however, the London Plane was only the 67th most common urban tree species.Projections have suggested that one-sixth of tree species could be hosted in cities and towns globally (Ossola et al., 2020).This study's most compelling trend was seen when tree diversity was contrasted between the inside and outside of city centers.Both the Shannon and Simpson diversity indices showed overall greater diversity outside of city centers, particularly at the species-level.Alvey's (2006) finding that urban density of city centers had a direct effect on tree diversity was corroborated by the results of the present study.The present study did not investigate potential explanations for this trend, though it corroborates findings where vegetation species richness is greater in suburban and peri-urban landscapes (McKinney, 2008a,b).

Differences in diversity benchmarks, stem count, and basal area
When calculated with BA, all eight selected cities failed to meet either the 10/20/30 or 5/10/15 diversity benchmark.When calculated according to stem count, none of the cities met the 5/10/15 benchmark, but Amsterdam (outside) did meet the 10/20/30 benchmark, with 8/ 14/14.In general, more cities came closer to meeting the diversity benchmarks when abundant species were calculated with stem count, including Bologna (inside and outside), Cambridge (outside) and Vancouver (outside).Santamour (1990) also presented his original benchmark in stem count, not BA.
Overall, all stem count and BA analyses were very similar, with subtle variations leaning towards stem counts appearing more diverse.The subtle difference in diversity is illustrated in Fig. A9 (in the appendix), in which the stem count curves in the sensitivity analysis level off slightly faster and higher than those of BA.Since stem count treats small and large trees equally, and BA favours larger trees as they have a larger cross-sectional surface area, the difference may be explained by cities opting for smaller, ornamental street tree species like the Street Parade (Malus baccata) and the American sweetgum (Liquidambar styraciflua).Ornamental species, when weighted uniformly, appear more diverse in stem counts, but less diverse when their smaller size is taken into consideration in BA analyses.Ultimately, the use of both stem count and BA complemented one another in the present study and using both allowed the study to come to more representative conclusions.
One possible explanation for why Amsterdam met the 10/20/30 benchmark, over any of the other selected cities, is the city's long history of systematically planting trees along newly dug canals.Beginning in the 1600s, lindens and later elms were used to reinforce the new quaysides, improve air quality, and add aesthetic value.The city was termed 'Elm City' in 2005, and this study has shown that Dutch Elm remains Amsterdam's most abundant street tree.Dutch Elm is a naturally occurring hybrid of Wych elm (Ulmus glabra) and field elm (Ulmus minor) and is often planted in Dutch cities due to its tolerance to pollution, poor soil, and salty sea winds (Postma and Goossen-van de Geijn, 2016;Gibbs, 1978).Damaging trees in Amsterdam has been a crime since 1454, and past punishments have included fines or even the removal of a guilty person's right hand.Amsterdam's systematic planting and strict protections may explain why its street tree networks are potentially more diverse.
The 10/20/30 benchmark is generally more recognised in North American cities-Cambridge, US and Vancouver, Canada, refer to the benchmark in their respective urban forest management plans (City of Cambridge, 2020a, 2020b;City of Vancouver, 2018).Almere, NL, is the only European city found to explicitly refer to the Santamour benchmark in its official policy (Platform i-Tree Nederland et al., 2019).i-Tree's direct reference to the benchmark, in addition to the regional differences in native tree diversity, may explain why the North American cities included in the present study performed better based on this metric.
The Santamour metric is also referred to in Melbourne's policy documents.The city's urban forest strategy specifically cites the 10/20/ 30 benchmark as a priority, second only to a 40 % overall canopy goal (City of Melbourne, 2012).When cities prioritise different aspects of urban forestry, different outcomes emerge, such as increased equity, canopy coverage, urban heat island mitigation, storm water permeability, biodiversity and citizen engagement.Cities that mention a diversity benchmark in their urban forest strategies (i.e., Cambridge, Vancouver, Melbourne) also tend to mention canopy coverage targets.
The review of the literature found that the other selected cities-Bologna, Buenos Aires, Paris, and Oslo-did not cite a specific diversity benchmark, but rather focused on other urban forest functions, such as urban food production and carbon sequestration, as well as increasing biodiversity more broadly.The biodiversity plan for Paris emphasises tree selection within the context of other biodiversity efforts, specifically those relating to insects and birds, rather than a specific metric or benchmark (City of Paris, 2018).
In contrast to Amsterdam, which showed the highest diversity, Oslo fared worse based on the diversity benchmarks.The center of Oslo registered 56/59/59 for stem count, in contrast to outside the center, which registered 15/25/26.This poor diversity result is corroborated by previous research examining the diversity of urban tree populations in Nordic cities, which found that 70 % of all newly planted trees in street environments in Oslo belonged to one clone of Tilia (Sjöman et al., 2012).In addition, Sjöman and Östberg (2019) found lindens to comprise 40 % of the total tree stock in Oslo.Using stem count, when the present study's authors averaged the city's stock of lindens both inside and outside the city center, lindens was found to comprise 36 % of Oslo's total tree stock.Sjöman et al. (2012) also noted that Oslo's tree inventory does not specify the lindens' species beyond Tilia spp., though Pauleit et al. (2002) concluded the lindens are likely to be Tilia × europaea, which is generally known as the common linden.While common lindens are largely resistant to disease, they can be susceptible to fungal diseases and aphid infestation.The lack of diversity and relative monoculture of Oslo's street tree network raises concerns of a significant risk of tree loss to disease.N.J.Galle et al.

Differences in urban development patterns
The urban development patterns of the selected cities may also have played a role in differences in tree diversity observed.For example, across the Shannon and Simpson indices and BA and stem count approaches, Amsterdam, Melbourne, Oslo and Paris showed the most significant spatial variation in tree diversity within and outside city centers.This may be partially explained by the circular and radialconcentric urban design adopted in these cities.In contrast, the spatial variation observed in Melbourne could be explained by its general development pattern of urban sprawl, whereby the city center is historic and the outskirts were developed more recently (Woodcock et al., 2010).Both of these types of urban development may lend themselves to suburban development with larger planting areas to allow for a wider palette of tree species (Nitoslawski et al., 2016).Conversely, these spatial variation patterns may be explained by the relatively recent increased focus on street tree diversity (Cowett and Bassuk, 2017), whereby cities have been encouraged to plant more diverse species in available areas outside of city centers, increasing species richness outside, but not inside, the centers (Alvey, 2006).This is particularly true for street trees; typically, streets are wider and have more planting space outside of city centers (Southworth and Parthasarathy, 1996).Schwartz et al. (2006) have highlighted the threat of biotic homogenization in urban forests, whereby non-native species may out-compete native species for resources, leading to reduced numbers of native species or potential local extinctions.This concern is echoed by Alvey (2006) and Trentanovi et al. (2013), who warn of urbanisation as a driver of biotic homogenisation.While the increase in non-native species may increase diversity and even performance on diversity benchmarks, it also has the potential to result in overall biodiversity losses on a global scale (McKinney, 2006).Of the eight cities included in the present study, a native species was the most abundant species in only one city's (Amsterdam) street tree inventory.

Research limitations
The interpretation of the results of the present study has some limitations.First, the computations relied on the availability of open-source tree inventory data.The quality of tree inventory data, especially when collected by citizen scientists and is openly-accessible, can be irregular (Roman et al., 2017).However, the citizen scientists in Roman et al.'s (2017) study were approximately 90 % consistent with experts for genus identification, and within the correct genera, correctly recorded species for 84.8 % of trees.Tree species and location must be accurate to calculate relative abundance and diversity indices based on stem count.To calculate relative abundance and diversity indices based on BA, the DBH must also be accurate.Multiple data inputs increase the chance of deviating data outputs.
As two of the eight cities had street tree inventories only, the rest of the inventories were clipped to only include street trees.Similarly, any data missing for DBH or species were not included.These steps were critical to normalise variables to allow for a better interpretation of the results.However, the steps taken also mean that the results do not represent urban forests in their entirety.In addition, while attempts were made to include cities from several Köppen climate classifications, the eight cities selected primarily represent North America, Europe and Australia.This is a common bias that is also discussed by Kendal et al. (2014), who analysed 108 cities worldwide and experienced a lack of (open-source) tree inventory data from South America, Africa and Asia.
Overall, it is important to consider the goal behind enhancing tree diversity in urban areas.If it is to improve resilience to future threats, it is important that diversification does not result in planting more species that are unsuitable for urban environments, or that a greater diversity does not lead to an increased percentage of the urban forest being threatened by, for example, polyphagous pests.While diversity can be a positive policy goal in and of itself, it is critical to nuance that increasing tree diversity alone may not be helpful for urban forest resilience in all cases.

Future research
The present study offered a unique approach to examining spatial variations within and outside city centers to support or potentially redefine heuristic benchmarks and help local practitioners improve the targeting of their tree diversity efforts.This study was the first of its kind to contrast tree diversity within and outside of city centers using opensource tree inventory data to draw comparisons across cities internationally.
Furthermore, while most cities did not meet diversity benchmarks, other metrics such as the HUGSI and city-specific goals suggest the need for a broader, more inclusive performance index for urban forests.Future research could consider relationships between other urban forest variables and drivers of biodiversity such as equity, tree diversity, canopy cover, stormwater mitigation, health and resilience, and aesthetic factors.This study's international and comparative approach used only open-source tree inventory data to encourage researchers, practitioners, and citizen scientists to quantify their own city's urban forest diversity via the Diversitree project website and open-source project repository.As discussed in the section on sensitivity analysis (section 4.1), we find that a sample of 500-700 trees is sufficient to saturate these measures.

Fig. 1 .
Fig. 1.City center boundaries in the selected cities.

Fig. 2 .
Fig. 2. Results of the sensitivity analysis for all cities.The x-axis represents the increasing sample size count randomly sampled from the datasets, and the y-axis represents the Shannon Index of the species of the datasets.Each color represents the result from one city.Tree city center stem counts are marked with vertical lines.

Fig. 5 .
Fig. 5. Shannon Index (H1) of street tree inventories at species-level in the city center and outside of the center across eight cities internationally, calculated using stem count.*Higher numbers indicate a more diverse community.

Fig. 3 .
Fig. 3. Shannon Index (H1) of street tree inventories at species-level in the city center and outside of the center across eight cities internationally, calculated using the basal area.*Higher numbers indicate a more diverse community.

Fig. 4 .
Fig. 4. Simpson Index (1− D) of street tree inventories at species-level in the city center and outside of the center across eight cities internationally, calculated using the basal area.**Higher numbers indicate a more diverse community.

Fig. 6 .
Fig. 6. Simpson Index (1− D) of street tree inventories at species-level in the city center and outside of the center across eight cities internationally, calculated using stem count.**Higher numbers indicate a more diverse community.

Fig. A2 .
Fig. A2.Shannon Index (H1) of street tree inventories at family-level in the city center and outside of the center across eight cities internationally, calculated using the basal area.*Higher numbers indicate a more diverse community.

Fig. A4 .
Fig. A4. Simpson Index (1− D) of street tree inventories at family-level in the city center and outside of the center across eight cities internationally, calculated using the basal area.**Higher numbers indicate a more diverse community.

Fig. A1 .
Fig. A1.Shannon Index (H1) of street tree inventories at genus-level in the city center and outside of the center across eight cities internationally, calculated using the basal area.*Higher numbers indicate a more diverse community.

Fig. A7 .
Fig. A7. Simpson Index (1− D) of street tree inventories at genus-level in the city center and outside of the center across eight cities internationally, calculated using stem count.**Higher numbers indicate a more diverse community.

Fig. A8 .
Fig. A8. Simpson Index (1− D) of street tree inventories at family-level in the city center and outside of the center across eight cities internationally, calculated using stem count.**Higher numbers indicate a more diverse community.

Fig. A6 .
Fig. A6.Shannon Index (H1) of street tree inventories at family-level in the city center and outside of the center across eight cities internationally, calculated using stem count.*Higher numbers indicate a more diverse community.

Fig. A5 .
Fig. A5.Shannon Index (H1) of street tree inventories at genus-level in the city center and outside of the across eight cities internationally, calculated using stem count.*Higher numbers indicate a more diverse community.

Fig. A9 .
Fig. A9.Sensitivity analysis of Shannon Index (top row) and Simpson Index (bottom row) for basal area sums (left column) and stem counts (right column).Each color represents a city, and each symbol (x, +, and -) represent the species-level, genus-level, or family-level taxonomy and diversity indices.As discussed in the section on sensitivity analysis (section 4.1), we find that a sample of 500-700 trees is sufficient to saturate these measures.

40000 Table 1
Definitions of city center boundaries.

Table 3
Diversity benchmarks of street tree inventories across eight cities internationally, calculated using stem count.Data in bold indicates the most abundant species, genus, or family that met the proposed 10/20/30 benchmark.All others failed to meet the proposed benchmark.

Table 4
Paired-sample t-test results for the Shannon and Simpson Index values comparing street tree diversity inside and outside of city centers, calculated using basal area.
df = degree of freedom; the number of degrees of freedom is approximated by the Welch-Satterthwaite formula.N.J.Galle et al.

Table 5
Paired-sample t-test results for the Shannon and Simpson Index values comparing street tree diversity inside and outside of city centers, calculated using stem count.