Local site conditions, not landscape context, influence restored plant communities within urban contexts

Restoration outcomes are variable, which impairs our ability to plan projects, meet goals, and predict restoration outcomes. Understanding the drivers of this variation is an important research need, especially within urban ecosystems, which support altered abiotic and biotic conditions and face higher rates of loss and degradation than non‐urban areas. Despite the importance of urban areas for restoration, research and practice have largely focused on non‐urban areas. It is unclear if we can extend current knowledge from restoration ecology to urban systems. Here, we surveyed 30 urban prairie restoration plantings across southern Michigan. We collected plant community and site condition data (e.g. soil attributes) and we quantified landscape context as the percentage of urban land surrounding each site. Variation in plant community composition among restorations was related primarily to site‐level factors, such as soil compaction, texture, and water‐holding capacity, rather than landscape context. Non‐prairie species were structured primarily by the local site conditions. There was an increase in non‐prairie richness for sites that experienced warmer local climate conditions, while there was a decrease in non‐prairie richness for sites where soils were less compacted, sandier, and had elevated water‐holding capacity. Prairie species richness responded oppositely. Overall, our study revealed specific factors structuring restoration outcomes in urban contexts and illustrated the importance of local site conditions, not surrounding landscape context, for shaping plant community composition. Restoration practices developed in non‐urban areas should be extended to urban contexts to better understand the impact of local site conditions on plant community development.


Introduction
Restoration outcomes are notoriously variable, which in turn limits our ability to plan restoration projects, reliably meet restoration goals, and predict restoration outcomes across systems (Brudvig & Catano 2021).As a result, restored areas, on average, fall short of achieving the same levels of biodiversity and ecosystem functioning that we observe in intact ecosystems (Rey Benayas et al. 2009;Atkinson et al. 2022).We do not understand which drivers of variation are consistently important across restoration contexts.Consequently, resolving these drivers among restored areas is an important research need for the field generally (Brudvig et al. 2017), so that we can counter widespread declines in biodiversity and ecosystem functioning globally (Suding 2011).However, much of the focus of past research attempting to identify consistent drivers of variation among restored areas has been limited to studies within nonurban contexts, with very little work done to understand the drivers of variation within urban contexts (Rega-Brodsky et al. 2022).
Over the past 20 years, urban land cover has increased by 50% globally (Potapov et al. 2022), resulting in a massive conversion of greenspace to hardscape.Cities are often located in naturally species-rich regions (Cincotta et al. 2000;Kühn et al. 2004;Luck 2007), or biodiversity hotspots, thus the resulting degradation of environmental conditions, alteration of landscape structure, and the introduction of non-native species (Kaye et al. 2006;Vilà et al. 2010;Klaus & Kiehl 2021) as a result of urbanization may have an outsized effect on urban ecosystems.Consequently, there is growing interest in conserving and restoring urban ecosystems to promote native biodiversity and ecosystem functionality (Dearborn & Kark 2010;Nilon et al. 2017).Yet, how to best reach these goals through ecological restoration remains an open question.
The establishment and assembly of both target (desired) and non-target (undesired) species within restored areas is frequently mediated by the abiotic and biotic site conditions (MacDougall et al. 2008;Young et al. 2015;Stuble et al. 2017), landscape context (Woodcock et al. 2010;Crouzeilles et al. 2016), and the details of the ecological restoration practices (Brudvig & Damschen 2011;Grman et al. 2013;Guiden et al. 2021).However, the relative importance of each of these suites of factors is not consistent across systems.In non-urban contexts, ecological restoration of the target plant community within degraded sites, such as grasslands that have been converted for agricultural use, is often limited by the target species' ability to establish, given the altered soil conditions and competition with non-target, weedy species (Gornish & Ambrozio dos Santos 2016).Shifts in the surrounding landscape structure, like the expansion of farmlands in non-urban areas, can impact the assembly of restored plant communities by reducing sources of target species and limiting their dispersal through fragmentation (Minor et al. 2009;Brudvig & Damschen 2011).Ecological restoration practitioners often attempt to ameliorate the impact of altered site conditions and landscape contexts by implementing management practices such as seeding target species to overcome dispersal limitations, removing undesired species, or amending soils, but these approaches vary from site to site, which can also introduce variation in restoration outcomes (Grman et al. 2013;Meissen et al. 2020;Guiden et al. 2021).The current understanding of how local and landscape factors structure restoration outcomes is derived primarily from non-urban systems, with very little work done to examine these drivers in urban systems.
We would expect similar types of factors to drive variation in restoration outcomes in urban systems as in non-urban, such as the biotic and abiotic site conditions, landscape context, and restoration practices implemented.However, the ways in which these factors are modified by humans in the urban environment may differ, and consequently, how they influence restoration outcomes may also differ.Previous work has found that urbanization-often quantified as impervious surface covermodifies biodiversity.High levels of urbanization, or greater than 50% impervious surface cover, result in a decrease in overall species richness (McKinney 2008;Rega-Brodsky et al. 2022).Moderate levels of urbanization, between 20 and 50% impervious surface cover, tend to support non-native species assemblages of varying diversity among different taxonomic groups (McKinney 2008;Rega-Brodsky et al. 2022).Novel, non-target species assemblages found in urban environments may compete with target species during restoration and reduce target species establishment (Shochat et al. 2010;Johnson & Handel 2016).Further, urbanization alters local abiotic site conditions, which reduces the capacity of the urban environment to support target native plant communities (Walker et al. 2009;Threlfall et al. 2016).Abiotic site conditions, including local climate and soil conditions, are altered by urban development (Kaye et al. 2006).For instance, the urban climate is modified by the presence of impervious surfaces like pavement, buildings, and other infrastructure, which can lead to consistently higher air temperatures in cities (i.e. the urban heat island effect; Oke 1982;Li et al. 2020).These high temperatures can increase heat stress-related mortality of urban plants (Czaja et al. 2020); this, in turn, may reduce plant establishment or persistence during restoration.Urban soils may be impacted by a legacy of urban development through damage to the soil structure via compaction from construction machinery (Pavao-Zuckerman 2008) and increases in heavy metal concentration from pollution deposition, which can alter soil chemistry (Pouyat & McDonnell 1991).Studies of the impact of local site conditions on restoration outcomes within urban contexts are limited.However, soil compaction has been found to lead to lower target species richness in urban restoration plantings (e.g.Sullivan et al. 2009).
Urban landscapes are composed of mosaics of different land uses (Niemelä 1999), including housing developments, transportation infrastructure, highly managed green spaces such as parks or golf courses, and remnant natural areas.Surrounding landscape context of urban greenspaces can affect local biodiversity by including impervious surfaces like roads that can facilitate the movement of invasive species (Skultety & Matthews 2017) and reduce the dispersal and recruitment abilities of native species (Overdyck & Clarkson 2012) and by influencing human recreational use of urban greenspaces (Talal & Santelmann 2019).However, the ways in which cities develop, and prioritize green space within a city, varies from city to city.Further, there is a paucity of studies examining how the urban landscape context impacts community assembly within urban greenspaces (Aronson et al. 2016), and even less work has been done to understand how urban landscape context influences urban ecological restoration efforts (Rega-Brodsky et al. 2022).
If the ways in which local and landscape factors impact restoration outcomes differ between urban and non-urban areas, but the same general types of factors explain variation, we may be able to translate our current understanding of restoration ecology to urban systems.For instance, the abiotic site conditions, specifically highly compacted soils due to construction with heavy machinery, may impair target species establishment at a site.Though the cause of compaction may be different, the ecological restoration practices developed to cope with similar conditions in former agricultural fields or mining sites may be transferable to the urban environment.However, if new types of factors are important, a new urban ecological restoration framework may be needed.For example if urban areas are experiencing consistently higher temperatures and related climate modifications due to the urban heat island effect, this may result in a mismatch between the regional climate conditions that species are adapted to and urban climate conditions, impairing target species establishment at a site.This may suggest the need for new restoration approaches in urban areas, such as the use of seed mixes containing novel species assemblages or genotypes.
To identify potential drivers of variation in urban restoration outcomes, which may be structuring plant community variation, we posed a series of questions: How do local and landscape factors, specifically local site conditions, landscape context, and restoration practices structure restored plant communities?How do prairie and non-prairie species respond to these local and landscape factors?And how do these findings in urban contexts align with our understanding gained from studies conducted in non-urban contexts?To do this, we surveyed 30 urban prairie restoration plantings across three cities in southern Michigan, U.S.A.We hypothesized that local-scale abiotic site conditions such as soil attributes (e.g.soil texture, nutrient content, soil water-holding capacity), canopy cover, local climate conditions (e.g.seasonal and annual variation in temperature and precipitation), restoration practices, and landscape context quantified as the percentage of urban land surrounding each site, contribute to variation in plant communities among sites undergoing restoration in urban contexts.We explored their roles across the plant communities of these planting sites, with a focus on plant species that are the target of restoration (prairie species) and those that are not (nonprairie species).

Study System
We conducted our study within urban, restored prairie plant communities.Prairie plant communities are typically defined by their dominance of grasses and forbs and the scarcity of woody plant species (Weaver 1954).Prairies were once a dominant community type throughout the Midwestern U.S., covering nearly 70 million ha, and were once abundant in southern Michigan, where our study was situated (Chapman & Brewer 2008;Cohen et al. 2021).However, prairies have now been reduced to less than 4% of their historic range within the United States due to the conversion of land for agriculture, cattle grazing, and urban development (Kindscher & Tieszen 1998;U.S. National Park Service 2022).Prairie plant communities have become regionally rare as a result and are a common target for restoration efforts (Lenhart & Smiley 2018).Further, prairie plant communities have been proposed as a candidate plant community type for restoration efforts in urban areas as urban grasslands like parks, lawns, and similar urban greenspace types have the potential to improve landscape connectivity, increase biodiversity, and enhance ecosystem services in urban centers through ecological restoration (Klaus 2013).

Study Design
Our study assessed the restoration outcomes for each prairie planting site, considering community composition, richness of the entire plant community, as well as prairie and non-prairie species richness and cover.We focused our study on three urban centers in southern Michigan, Ann Arbor, Grand Rapids, and Kalamazoo.We chose these three cities because they are within the top 20 largest in Michigan based on population size, ranging from 196,908 in Grand Rapids to 72,873 in Kalamazoo (QuickFacts 2020) and because each has a notable number of urban prairie restoration plantings.For each city, we first populated a list of potential prairie restoration study sites based on discussions with local restoration practitioners and park managers.From this list, we included sites in our study if we were able to identify a contact person for site access, the site was reconstructed through seed sowing (i.e.not remnants undergoing management), the site resembled a prairie based on a ground-truthing visit (i.e.there was no significant woody encroachment), and if land managers retained information about planting site age and total seeded area.We did not impose a minimum distance between sites to maximize the number of planting sites that could be included in our study and sampled every site on our list that met these criteria.The minimum distance between plots was 91 m, and the average study site was approximately 8 acres (site size ranged from 0.05 to 14.5 ha).In total, we included 30 prairie plantings in our study: 12 in Ann Arbor, 12 in Grand Rapids, and 6 in Kalamazoo (Fig. 1).We collected data on local and landscape factors within each site in addition to plant community composition data within five, 1 Â 1-m subplots along a central, randomly oriented 20 m transect in each of the 30 planting sites June-August 2020.We recorded the cover of each plant species as the percentage of the subplot occupied by each species present.

Site Conditions, Landscape Context, and Restoration Practices Data Collection
Site conditions in our study included site age, tree canopy cover, soil conditions, and bioclimatic conditions.We defined site age as the number of growing seasons since restoration was initiated by sowing seeds of native prairie species.We collected canopy cover and soil condition data at each of the five subplots within each planting site.The canopy cover was measured at the same corner of each of the five subplots within each planting site using a spherical crown densiometer.Values were then averaged across the subplots to generate one value for each planting site.We determined soil conditions by measuring soil texture (the percent of sand, silt, and clay) using the LaMotte Soil Texture test kit, depth to soil compaction (depth in cm to 300 PSI) using a Dickey-John Soil Compaction Tester, soil water-holding capacity (calculated as the proportional difference between the wet and dry weight of the soil samples using the methods laid out in Brudvig & Damschen 2011), and soil nutrient composition.We sent soil samples to Brookside Laboratory, Inc. for soil nutrient testing (Soil Test: Standard Soil with Bray I P).We defined bioclimatic conditions as seasonal and annual temperature and precipitation metrics that influence local-scale prairie plant community establishment and abundance (Groves et al. 2020), such as mean summer and winter precipitation and temperature as well as the fluctuation between summer and annual temperatures.To do this, we generated 19 bioclimatic variables (Hijmans et al. 2005) from 800 m resolution PRISM (PRISM Climate Group n.d.) monthly temperature and precipitation data ranging from the year of planting site establishment until December 2020, the year of our survey.
We quantified landscape context as the percentage of developed land within a 500 m radius of each site by using ArcGIS Pro software and land use/landcover maps produced by the National Land Cover Database (NLCD) Landcover & Imperviousness for 2019 (Dewitz & U.S. Geological Survey 2021) The NLCD Landcover & Imperviousness maps include four categories for developed land: open space, low, medium, and high intensity.We used the sum of low, medium, and high intensity developed land to calculate the percentage of developed land surrounding each site.The percentage of developed land within the 500 m radius of each site ranged from 0.1 to 99% (see Fig. S4).
We additionally attempted to acquire management records for each site, including the number and identities of prairie species seeded to initiate restoration, prescribed fire records, and other aspects of management history, like the number of times sites were seeded or mowed.However, this information was poorly retained.Since seed mix design, both the richness and density of species, is a strong determinant of prairie restoration outcomes (Meissen et al. 2020;Glidden et al. 2022), we identified species that were likely to have been seeded into each restoration site.To do this, we first compiled a list of all species that were observed during plant community surveys across all sites and a compiled list of all species that were known to have been included in the seed mixes for the plantings surveyed.Seed mix information for the initial seed application was available for 15 out of the 30 sites included in this study.For species not included in seed mixes, we searched Michigan Flora (2023) to determine if each species observed was native to Michigan and known to be associated with prairies or prairie-like areas.Species that were either included in the seed mix lists provided or are known to be native to Michigan and associated with prairies and prairie-like areas were considered prairie species.
We also collected data related to management practices such as weeding, mowing, and prescribed burning when available.Most of the sites that had some form of management data had a record of whether the planting had ever been burned (n = 26), and this factor was included in preliminary models.However, burning was correlated with the age of the planting and was also a non-significant predictor when included in models for all response variables, thus it was not included in the final set of models.

Data Analysis
We performed all analyses in R studio using R version 4.2.3 (R Core Team 2023).We constructed separate models to assess the roles of site and landscape-level factors for each response variable: plant community composition, prairie and non-prairie species richness, and prairie and non-prairie species cover.
We used a principal components analysis (PCA) to summarize the soil and bioclimatic condition measures (Figs.S1 & S2), as these measures both consisted of several highly correlated factors.The first two axes of the soil attributes PCA explained about 40% of the total variation, hereafter referred to as soil attributes PC1 and soil attributes PC2.Soil attributes PC1 were correlated with clay content, soil pH, and heavy metals like zinc and copper.Soil attributes PC2 were primarily correlated with soil texture and water-holding capacity.The value along each PCA axis for each site was extracted and used in both the multivariate and univariate models.Additionally, we used the R package "dismo" (version 1.3-9; Hijmans et al. 2022) to generate 19 bioclimatic variables (Hijmans et al. 2005) from the PRISM Climate Data (PRISM Climate Group n.d.) monthly climate data for each of the planting sites spanning from the year the site was established until the year of survey.The first two loading axes of the bioclimatic attributes PCA explained about 54% of the total variation, hereafter referred to as bioclimatic attributes PC1 and bioclimatic attributes PC2.Bioclimatic attributes PC1 was correlated with precipitation-related indicators like drier springs and winters as well as temperature-related metrics like diurnal fluctuations.Bioclimatic attributes PC2 was primarily correlated with temperatures like warmer annual, spring, and winter temperatures.The value along each PCA axis for each site was extracted and used in both the multivariate and univariate models.
Due to the number of sites in our study (n = 30), for each of our final models, we limited to five additive predictor variables: site age, soil attributes PC2, canopy cover, bioclimatic attributes PC2, and the percentage of developed land surrounding each site.Here, we considered site age, soil attributes PC2, canopy cover, and bioclimatic attributes PC2 as site-level or local factors and the percentage of developed land surrounding each site as a landscape-level factor.We retained only soil attributes PC2 because it was structured primarily by soil texture and waterholding capacity, for which we have stronger hypotheses for how these impact plant community establishment than we did for the variables associated with soil attributes PC1.Similarly, we retained only bioclimatic attributes PC2 because it was structured by temperature whereas PC1 was structured more by precipitation, this trend likely being driven by proximity to Lake Michigan and not a reflection of urban site conditions.Of the five predictor variables, only bioclimatic attributes PC2 and the percentage of developed land within a 500 m radius buffer of each site were correlated (Pearson's R = 0.37), however, the variance inflation factors of the models were less than 1.5, indicating multicollinearity was low, so all five predictor variables were included in the models.Further, city identity was included in preliminary models, however, it was never a significant predictor of the plant community composition, nor prairie and non-prairie species richness or cover, among the prairie planting sites and therefore was not included in the final models.
We used a permutational analysis of variance using the "ado-nis2" function of the "vegan" package (version 2.6-4; Oksanen et al. 2022) to assess community composition and multiple linear regressions using the "stats" package (R Core Team 2023) to model the roles of local site conditions and landscape context for prairie and non-prairie species richness and cover variables.We examined the residual variance of each model and deemed the normal distribution appropriate for all univariate response variables.Additionally, we visualized the plant community composition data using NMDS ordinations and we utilized the "envfit" function of the "vegan" package (version 2.6-4;Oksanen et al. 2022) to fit the site condition and landscape context vectors, as well as vectors of species which may be driving the dissimilarity in plant community composition between sites.Only species that were found to be significant by the "envfit" function ( p < 0.05) were displayed in the ordination figure.We visualized the conditional effects of the significant model factors using the "ggpredit" function of the "ggeffects" package (version 1.2.0;Lüdecke 2018) to understand their effect on each response variable.
We tested the effect of spatial autocorrelation among our univariate response variables using a generalized least squares model for each response variable using the "gls" function from the "nlme" package (version 3. 1-163;Pinheiro & Bate 2023).Models were constructed with and without spatial autocorrelation corrections (exponential, gaussian, linear, rational quadratic, and spherical) and compared using AIC.For all response variables, aside from non-prairie species richness, the most parsimonious model did not include a spatial autocorrelation correction.For the response variable non-prairie richness, the models with and without the spatial autocorrelation correction were equivalently parsimonious (within 1 AIC point).As there is not a straightforward equivalent test for the multivariate models, we relied on the findings from our linear modeling approach to feel comfortable not including a spatial component in the multivariate models.

Results
Variation in plant community composition across the urban prairie plantings was related primarily to site-level factors, such as soil compaction, texture, and water-holding capacity (F = 1.963,R 2 = 0.063, p = 0.003; Table S1).Additionally, the NMDS ordination (Stress = 0.255, k = 2) revealed that the bioclimatic variables related to warmer annual, summer, and winter temperatures (Bioclimatic Attributes PC2), the percentage of developed land cover, canopy openness, and planting site age appeared to form a gradient that restored planting sites tended to fall along (Fig. 2A).The soil attributes PC axis (Soil Attributes PC2) appeared to form a largely orthogonal axis, relative to the other local and landscape factors plotted within the ordination space (Fig. 2A).Further, at the species level, it appeared that many of the species that were significantly driving the distribution of restored planting sites within the ordination space were non-prairie, and non-native, species such as Fallopia convolvulus (black bindweed), Chenopodiun album (white goosefoot), and Festuca rubra (red fescue) (Fig. 2B).Overall, the non-prairie species that may have been driving the distribution of prairie planting sites were aligning with younger, less open canopy sites while prairie species like Symphyotrichum novae-angliae (New England aster), Solidago speciosa (showy goldenrod), and Panicum virgatum (switchgrass) were aligning with older, more open canopy sites that also exhibited higher levels of surrounding urban land (Fig. 2B).
Non-prairie species richness and cover were structured primarily by the age of the restoration plantings; there was a decrease in both richness ( p < 0.05; Table S2; Fig. 3) and cover of non-prairie species with age ( p < 0.05; Table S3; Fig. S3).Non-prairie species richness was also negatively correlated with soil conditions such as soil compaction, texture, and waterholding capacity ( p < 0.05; Table S2; Fig. 3).Further, there was some evidence that more open canopy conditions reduced non-prairie species richness ( p = 0.087; Table S2; Fig. 3) while warmer annual, summer, and winter temperatures increased non-prairie species richness ( p = 0.054; Table S2; Fig. 3).Although there were no significant predictors of prairie species richness or cover, there was some evidence that sites with warmer annual, summer, and winter temperatures supported lower richness of prairie species ( p = 0.055; Table S2; Fig. 3).

Discussion
Overall, we were able to identify specific factors structuring restoration outcomes in urban contexts and illustrated the importance of local site conditions, not surrounding landscape context, for shaping plant community composition.Site-level factors such as the local climate and the soil conditions were the most significant factors driving variation in the plant community composition among the restored prairie planting sites, while differences in urban land cover in the surrounding landscape explained little variation.However, target, prairie species and non-target, non-prairie species, did respond to the site-level factors differently.
Our findings build on previous work showing how target and non-target plant species tend to respond differently to local-scale factors during ecological restoration.In non-urban contexts, soil type and soil resource availability can impact the cover, dominance, and persistence of target and non-target plant species differently, with non-target plant species often able to become dominant across a wider range of soil conditions than target plant species (Daehler 2003; Gornish & Ambrozio dos Santos 2016).In contrast, we found that both the richness and cover of target prairie plant species were not correlated with local soil conditions surveyed, but that non-target, non-prairie species did decline in richness as the soil conditions within a planting site became less compacted, had greater water-holding capacity, or shifted in soil texture from silty to sandy.The insensitivity of prairie species richness and cover to soil conditions in our study could be due to the species-rich nature of prairie plant communities (Risser 1988) coupled with efficient species sorting along this gradient (e.g.Foster et al. 2011), allowing for the maintenance of community-level richness and abundance across the gradient.
Additionally, at the local scale, we found that bioclimatic conditions, such as warmer summers, winters, and annual temperatures, led to an increase in the richness of non-target, nonprairie species and-to some extent-a decrease in target prairie species.While local climate conditions have been relatively understudied in non-urban contexts within the field of restoration ecology, it is well known that urbanization can have major impacts on the local climate within urban centers, also referred to as the urban heat island effect (Oke 1982;Li et al. 2020).We found some evidence for this effect in our study, as there was a correlation (r = 0.374; p < 0.05) between the amount of surrounding urban area at a planting site and warmer temperatures.Although this correlation was weak (and did not rise to a problematic level for collinearity in our models), this suggests that at least some variation in temperature was structured by urban landscape context.The finding that warmer summer, winter, and annual temperatures led to an increase in non-prairie species is surprising as prairie species are typically thought to be particularly heat and drought-tolerant due to adaptations such as dense root systems to access soil water resources during drought conditions and sun avoiding foliage to reduce heat and light stress in the open canopy conditions of grasslands (Tucker et al. 2011).If the urban heat island effect is detrimental to prairie plant community diversity, restoration practitioners may want to include more heat and drought-adapted prairie species or genotypes in urban restoration plantings expected to be most influenced by the urban heat island phenomena and invest in greater weed control in these plantings to aid in target plant establishment.However, further work is needed to better characterize how local climate within urban settings influences the establishment and assembly of restored plant communities, including effects on individual species, which likely vary in their sensitivities.
The surrounding landscape structure was not an important factor in shaping urban plant community composition, richness, or cover in our study.Previous work has found that in both urban and non-urban contexts, the surrounding landscape can pose a barrier to plant community assembly by facilitating the movement of invasive species (Skultety & Matthews 2017), reducing the dispersal of native species due to habitat fragmentation (Overdyck & Clarkson 2012), and reducing sources of native biodiversity on the landscape through habitat loss (Minor et al. 2009;Brudvig & Damschen 2011).It seems likely that restoration practices, such as assembling plant communities via seed addition at the planting site, may have overcome the potential barriers to dispersal posed by the surrounding urban land cover within our study.Many of the plant species observed most frequently across prairie planting sites were prairie species, indicating that the seeded prairie species were able to establish and persist once dispersed to the planting site.This lends further support for the ability of restoration practices to overcome potential dispersal limitations created by the urban landscape structure, and points to ecological restoration as a tool to enhance the biodiversity of target, native plant species within urban areas.However, the richness and the cover of non-target, non-prairie species were also not impacted by the surrounding landscape structure either.It could be that the non-target, non-prairie species were not dispersal limited in these urban landscapes, or these species were recruiting from a soil seed bank and were not reliant on dispersal.More work needs to be done to understand if and when the surrounding landscape structure influences urban restoration projects (e.g.Mitchell et al. 2016).
Restoration practices and management decisions can play a large role in restoration outcomes in non-urban areas (Brudvig & Damschen 2011;Grman et al. 2013;Glidden et al. 2022), however, we found little evidence in support of this within our study.This finding could stem from a lack of management records, such as the identities of sown species and the timing of key management events, like prescribed burns, invasive species removals, and inter-seeding.Such record keeping can open opportunities for future assessment of the role of restoration practices and management decisions within the urban context and research-management partnerships (e.g.Bach & Kleiman 2021).
Overall, we found that many of the types of local site factors that influence plant community composition in non-urban restoration contexts also play a role in shaping plant community composition in urban contexts.However, the ways in which these factors have been modified by urban development may require both the extension of knowledge from non-urban contexts and the development of new frameworks for urban areas.For instance, within our study, sites that had contrasting soil attributes, like increased soil water-holding capacity and increased sand content, tended to support similar plant communities (i.e.grouped together within ordination space).More work needs to be done to understand how the modified soils in urban areas influence plant-soil interactions within urban settings (sensu Kotze et al. 2021) to minimize the establishment of non-target plant species within the planting area before beginning ecological restoration (Dighton & Krumins 2014).Additionally, we found that local climate conditions, such as elevated seasonal and annual temperatures, led to an increase in non-target, non-prairie species richness and a corresponding decrease in target, prairie species.This may indicate a mismatch between the regional climate conditions that target prairie species are adapted to and urban climate conditions, impairing their establishment at a planting site.This suggests that new frameworks need to be developed for urban areas to identify appropriate target species beyond selecting from a regional species pool.Recent work examining trait-environment relationships at the seed mix design stage (Balazs et al. 2020) could provide a relevant framework to build upon in urban systems.Our findings do suggest that more work needs to be done in both non-urban and urban areas to understand the impact of local climate on plant community development (e.g.Salinitro et al. 2019) and to identify if current ecological restoration techniques are sufficient to address these impacts (Maxwell et al. 2019;Frietsch et al. 2023).
Our findings indicate that we can extend generalities of what we know about restoration ecology from non-urban to urban systems, but that specifics of urban restoration efforts should be tailored and may also require new tools.Similar sorts of site-level factors, such as the soil and the local climate conditions, are important to take into consideration in both non-urban and urban settings, although the ways in which those conditions have been modified may differ.Further, it seems that tried and tested ecological restoration practices, such as sowing the target plant community into the planting site, can overcome limitations to dispersal, which may be posed by the surrounding landscape context (Grman et al. 2013).However, the impact of local climate conditions in urban areas, potentially stemming from the urban heat island effect, may require new practices to establish resilient plant communities (Maxwell et al. 2019;Frietsch et al. 2023).Overall, urban green spaces hold promise as candidate areas for ecological restoration to increase both the biodiversity of urban areas and the ecosystem services provided by urban green spaces.

Figure 1 .
Figure 1.Map of the 30 restored prairie plantings within the cities of Ann Arbor, Grand Rapids, and Kalamazoo, in southern Michigan, U.S.A., which were surveyed as part of this study.

Figure 2 .
Figure 2. (A) NMDS ordination with all site condition and landscape context vectors (B) NMDS ordination with significant species vectors fit.Stress = 0.255.Prairie planting sites located in Ann Arbor, MI are indicated with red circles, sites located in Grand Rapids, MI are indicated with green triangles, and sites located in Kalamazoo, MI are indicated with blue squares.

Figure 3 .
Figure 3. Conditional effects of local level factors on non-prairie (A-C) and prairie species (D-F) richness within each of the 30 urban planting sites.(A) Conditional effect of soil attributes PC2 on non-prairie species richness.(B) Conditional effect of site age on non-prairie species richness.(C) Conditional effect of bioclimatic attributes PC2 on non-prairie species richness.(D) Conditional effect of soil attributes PC2 on prairie species richness.(E) Conditional effect of site age on prairie species richness.(F) Conditional effect of bioclimatic attributes PC2 on prairie species richness.Each black dot represents one of the 30 urban planting sites surveyed.The dark line indicates the regressions of the local-scale factor and the non-prairie or prairie species richness, and the grayshaded region represents the 95% CI.