Urban ecological communities are shaped by an array of environmental and physical factors that include climate, biogeography, species interactions, dispersal, and ecological filters at the habitat and landscape scales. In addition, in urban systems, many social dynamics, decision-making processes, and other social characteristics, such as development, policy, government actions, as well as socioeconomic status of residents – the so-called ‘luxury’ effect – may also play strong roles in shaping which species occur within the urban context. This interaction between ecological and social drivers may impact the species richness of communities, and may also influence their functional traits and functional diversity, with important implications for ecosystem services provided within urban spaces. Urban agroecosystems provide food and other well-being benefits to urban residents and are valuable green spaces in the city that provide refuges for biodiversity. Despite this fact, one of the biggest risks to gardens is insecure land tenure. While plant communities within gardens may be shaped by gardener demographics, food cultures, and ecological processes, little is known about how factors such as luxury and land tenure security may impact plant diversity, plant functional traits, and functional diversity, as well as species and trait composition. In this study, we ask how garden physical features, luxury, and land tenure security influence plant species richness, functional diversity, and species and trait composition within gardens in the California central coast. We found that land tenure secure gardens had higher plant and crop richness. Variables associated with garden luxury boosted ornamental plant richness, flower abundance and height, lowered crop richness, and strongly impacted plant species composition. Garden plot size negatively correlated with plant species and functional richness and relative abundance of trees. Weed species richness was not impacted by any social or physical feature examined. Thus overall, in community gardens, cultivated plants, and their traits, are strongly shaped by the socioeconomic factors of land tenure security and luxury as well as the spatial distribution and size of garden plots, with important implications for both conservation and ecosystem services provided by garden habitats within cities.

Description of study sites

We collected vegetation data in 2021 in a network of 23 urban community gardens in the California central coast in Monterey, Santa Clara, and Santa Cruz Counties (Fig. S1) where the research team has worked since 2013 (see Jha et al. 2023). The gardens were originally selected for their variation in management and landscape surroundings. All gardens are community gardens where vegetables, fruit trees, and ornamental plants are grown. The gardens are managed either collectively by groups of volunteer or paid managers, or in individual plots or allotments. Gardens range in size from 444 m² to 15,400 m² and have been cultivated for between 5–54 years. Each garden was separated from other garden sites by at least 2 km and up to 90 km.

Vegetation survey methods

In each garden, we sampled vegetation to describe the plant community. At the center of each garden, we established a 20 m x 20 m plot within which we counted the number of tree and shrub individuals and identified all trees and shrubs to species. Then, within eight randomly located 1 m x 1 m plots within the 20 m x 20 m plot, we identified and estimated the percent cover from each herbaceous plant species. We surveyed vegetation three times over the summer of 2021 (June 21–25, July 6–11, and August 2–5).  

Plant classification

Because of differences in the way the gardeners may use or perceive different types of plants, following identification of plants (herbaceous, trees, and shrubs) to species, we also classified each plant species according to whether they are commonly referred to as crops (e.g., vegetables, fruits, herbs, medicinal plants), ornamentals (e.g., flowering plants planted on purpose), weeds (e.g., spontaneous herbs that although may be used by gardeners for food are not intentionally planted), or grass. For herbaceous plants, we focused species identification and trait measurement effort on forbs (herbaceous non-grasses) as in other urban garden systems (e.g., Ballare et al., 2019) as grasses comprised less than < 5% of the herbaceous cover. We encountered 18 herbaceous plant morphospecies (representing just 0.9% of all plant cover surveyed) and nine tree or shrub morphospecies (representing only 1.2% of all tree and shrub individuals counted) that we were not able to identify to family or genus. These were excluded from the analysis.

Plant functional trait measurements 

Using data from the same standardized vegetation surveys as those described, but conducted from 2016–2018 in 30 gardens within the same study region, we compiled a full plant species list, and calculated the relative abundance of each species based on either raw abundance data (for trees and shrubs) or percent cover data (for herbaceous plants). We then identified for trees and shrubs, and separately for herbaceous plants, the most abundant species that collectively represent 70–80% of the total relative abundance for each plant group, as this range is recommended to effectively characterize plant traits within a community (Cornelissen et al., 2003). This resulted in a list of 88 plant species from 34 families: 73 herbaceous plant species (covering 77% of percent cover) and 15 tree and shrub species (or 70% of individuals counted) (Table S1). For each of these 88 species, we collected a suite of traits from individual plants during summer 2021, targeting traits important for beneficial insects, including pollinators and natural enemies (e.g., Perović et al., 2018). 

Specifically, we measured 11 traits for each plant species relating to four categories: 1) plant structure (plant growth, plant volume), 2) plant defense (specific leaf area, spines, trichomes, extrafloral nectaries), 3) floral attraction (flower color, maximum flower height, number of flowers), and 4) floral access (flower shape, flower volume). To capture traits across multiple gardens, we surveyed plant traits from 3 individual plants for each of the 88 selected species. We chose to sample 1 individual from each plant species at each of three sites, each in a different county, that all harbor high herbaceous plant species richness. Specifically, we sampled plant traits at the Trescony Garden in Santa Cruz County, Pacific Grove Community Garden in Monterey County, and Charles Street Garden in Santa Clara County. Some species were not present in one or more of these sites, and so we visited additional sites (Aptos Community Garden, Beach Flats Community Garden, Homeless Garden Project, Pajaro Garden, Senior Center Garden, Valle Verde Garden, and Mi Jardin Verde) to capture data for the 2^nd or 3^rd plant individual of each species. Due to low abundance, for two plant species, Equisitum sp. (horsetail) and Alstromeria aurea (Peruvian Lily), we measured two individuals at a single site and a third individual at a second site, and for three plant species, Chenopodium album (lamb’s quarters), Lepidium latifolium (pepper grass), and Helminthotheca echiodes (warty dandelion), we only sampled 2 individuals total, at two different sites. There were 18 of the 88 plant species that were not in flower, and thus floral traits were not assessed for those species (Table S1). 

Plant Structure

We measured plant growth form following the categorizations of the Australian National Botanic Gardens (2021) and Pérez-Harguindeguy et al. (2013) where trees are woody plants with fewer than 3 stems and more than 5m tall; shrubs are woody plants less than 8 m tall and with branching at or near soil level; sub-shrubs have stems that are herbaceous in their upper parts; herbs or forbs are plants with no woody tissue present; vines or lianas are climbers rooted in the ground; graminoids are tussock or tufted plants; and stem-succulent shrubs include genera such as Sarcostemma, Opuntia and some Euphorbia. We measured plant volume by measuring plant height (cm) at the tallest point, not including any flowering or fruiting structures, and plant width (cm) at the largest point, and multiplying these two values. 

Plant Defense

We measured specific leaf area (SLA) by collecting one leaf from each plant surveyed (including the petiole), storing the plant (up to 72 h) in a cooler or refrigerator until photographing with Leafscan (Version 1.3. 2) (Anderson and Rosas-Anderson, 2017) to measure leaf area. We then dried leaves in a drying oven at 40 degrees C for 72 h, or until leaves no longer lost any more mass. We then weighed leaves and calculated SLA as the leaf area divided by dry mass. We measured spines according to Callis-Duehl et al. (2017) and Pérez-Harguindeguy et al. (2013). We included all ‘spines’ (sharp, modified leaves, leaf parts or stipules), ‘thorns’ (sharp, modified twigs or branches) and ‘prickles’ (modified epidermis or cork), and scored each plant according to presence or absence of spines on leaves, stems, and flowers. We then scored plants as 0 (no spines, thorns or prickles on any structures), 1 (presence on leaves, stems, or flowers), 2 (presence on two structures), or 3 (presence on all three structures). We sampled plants for trichome presence according to Callis-Duehl et al. (2017) and Bar and Shtein (2019). We inspected for trichomes on leaves, stems, and flowers with a hand lens, and also noted the presence of simple, star-shaped, or glandular trichomes. We then scored plants as 0 (no trichomes on any structure) to 9 (presence of all three types of trichomes on all three structures). We used the World List of Plants with Extrafloral Nectaries (Weber et al., 2015) to determine for all of the plant species whether or not they are known to have extrafloral nectaries (EFNs).

Floral attraction

For flower color, we recorded the primary and secondary colors (e.g., pink, red, orange, yellow, green, blue, violet, mauve) according to a standardized printed color wheel and then collapsed the flower colors into 16 different color combination categories (as in Rosas-Guerrero et al., 2014). We measured maximum flower height (cm) as the height of the tallest flower on the plant above the ground. We counted the number of flowers on each plant. For plants with >100 flowers or with multiple inflorescences, we counted flowers on a particular stem or inflorescence, and then counted total stems or inflorescences to estimate the total flowers per plant. 

Floral access

We measured flower shape according to the ‘Blossom Classes’ in Ollerton et al. (2009) and Rosas-Guerrero et al. (2014). Specifically, flowers were classified as either a) dish - radially symmetrical, mostly flat access to pollen and nectar (not bell, gullet, brush, or trap shaped); b) gullet - bilaterally symmetrical, rough conical shape with upper and lower ‘lip’ where pollen is always placed dorsally (near upper lip) on pollinators; c) flag - bilaterally symmetrical, rough cylindrical shape with upper ‘banner’ and lower ‘keel’ and pollen is always placed ventrally (near keel) on pollinators; or d) tube - radially symmetrical, narrow tube-shaped corolla, often behind the more conspicuous petals. Four flower shape categories (bell, brush, inconspicuous, and trap) were not present in gardens.  We measured flower volume following Hegland and Totland (2005) and Lázaro and Totland (2014). For each plant individual, we haphazardly selected one fully open flower, and measured the corolla height (mm), corolla width (mm), and corolla tube length (mm) (for tube-shaped flowers only). We then multiplied the two (or three values) to calculate floral volume. 

Plant Trait Analysis

We used 2021 vegetation survey data to calculate relative abundance of herbaceous plants as well as trees and shrubs in each garden. For herbaceous plants, we summed total percent cover from each plant species across 24 plots (eight different plots from three time periods), summed the total percent cover from the herbaceous species from which we measured plant traits, and calculated the relative abundance for each species as a fraction of total cover of plants for which we measured plant traits. For trees and shrubs, we calculated the relative abundance as the number of individuals of a species as a fraction of total individuals of trees and shrubs from which we collected trait data. We averaged trait values for each species (for quantitative variables) (3 measurements per trait for 86 plant species and 2 measurements per trait for 2 plant species) and confirmed that categorical variables (e.g., growth form, flower shape) were the same for each species. We natural log (+1) transformed plant volume, number of flowers, flower volume, and the spine indices to meet conditions of normality. 

We used the ‘dbFD’ function in the FD package in R to calculate several functional diversity metrics. Functional richness (FRic) describes how much functional trait space is filled by the community; functional evenness (FEve) describes how evenly the abundance distribution is spread across the functional trait space; and functional divergence (FDiv) describes whether the distribution of species along a trait axis is clumped or spread out, and is useful for assessing species complementarity (Villéger et al., 2008; Woodcock et al., 2019). We also measured functional dispersion (FDis) (Laliberté and Legendre, 2010), and Rao’s quadratic entropy (Q) (Botta-Dukát, 2005), two abundance-weighted measures of trait spread. FRic, FDis, and Q are positive, unbounded values representing trait space. FEve is a value bounded between 0 and 1 where values of 0 occur when most species (and species abundances) represent a small fraction of the trait space, and values of 1 occur when species (and species abundances) are regularly spread across the trait space. FDiv is a value bounded between 0 and 1 where values of 0 indicate the most abundant species have traits close to the community centroid, and values of 1 occur when the most abundant species have extreme trait values. Thus, high FDiv represents high niche complementarity (Villéger et al., 2008). We used the ‘gowdis’ function to calculate a species x species distance matrix for all continuous (i.e., plant volume, trichomes, spines, number of flowers, flower height, SLA), binary (i.e., presence or absence of EFNs), and categorical (i.e., growth form, flower shape) variables. All numerical valuables were automatically scaled to mean 0 and unit variance. For flower color, we used a ‘fuzzy coding’ approach (De Bello et al., 2020) such that a plant with multiple flower colors might have an intermediate trait distance to plants that share only one of multiple flower colors. We then calculated species distances weighted by variable type. Continuous variables and flower color (which had 16 levels) distances were multiplied by 0.67 and binary and all other categorical variables were multiplied by 0.33 to give a larger weight to those variables that do not indicate absolute similarity (0) or difference (1) between trait states (as per De Bellow et al., 2020). We weighted the FEve, FDiv, FDis, and Rao’s Q calculations by species relative abundance. Finally, we used the ‘funtcomp’ function in FD to calculate the community weighted means (CWM) for each trait variable. We used ‘type all’ to return values for each value for binary and categorical variables, rather than returning the predominant trait state type for each community.

Land tenure security, ‘luxury’, and garden variables

We collected information on a large set of potential predictor variables related to land tenure security, ‘luxury’, and garden physical characteristics. First, we examined land ownership of the gardens by examining the California Multi-Source Land Ownership layer of the California government GIS database (https://gis.data.ca.gov/), web pages maintained by the organizations that support the gardens (https://www.mcsc-capitola.org/, https://www.sanjoseca.gov/, https://www.aptoschurch.org/aptos-community-garden,  https://sites.google.com/site/liveoakgrange/about-the-green-grange/history/sclo-grange-503, https://www.middlebury.edu/institute/news, https://www.cityofsantacruz.com/), and news articles (Meyberg Guzman and Kathan, 2021) and then classified garden properties as either owned publicly (by a state or local government or school district) or privately (by a church, senior or farming organization, or individual). Second, we created a land tenure security index based on the tenure status decision process in Arnold and Roge (2018). We classified gardens where the gardeners (or their organizations) own the property as 4 (highest land tenure security), gardens with a long-term lease with the property owner as 3 (second highest land tenure security), gardens with short-term memorandum of understanding as 2 (third highest land tenure security), and gardens without any of these features, or with imminent risk of eviction as 1 (lowest land tenure security). Third, we examined whether gardens had paid employees (an indicator of economic support for the gardens, and thus higher land tenure security, Arnold and Roge, 2018) and if gardens were affiliated with a school (a positive indicator of social support and of higher land tenure security, Arnold and Roge, 2018). Each of these three variables (land tenure security index, presence of paid employees, and affiliation with a school) were determined through informal conversations with gardeners, garden managers, and paid employees over the past decade, and by consulting information on the garden webpages listed above. Fourth, we documented whether the gardens were managed in an allotment style (with each family or individual gardener making decisions about individual plots) or by a single or small group of garden managers making management decisions about the entire garden area. Garden management style (allotment vs single-manager) substantially overlapped with presence of paid gardeners and with affiliation with a school. For example, all four gardens in our sample with paid gardeners were also the only gardens with a single-manager style and, three of those four gardens were affiliated with a school. No other gardens were affiliated with schools or used a single-manager style. Because of this, of these three variables, we chose to maintain only garden management style in our analysis, and assume that single-manager style is associated with higher land tenure security. 

As measures of garden luxury, we gathered information on median property value, percent owner occupied housing units, median household income, and percent of residents below the poverty line for each zip code containing a garden from the 2021 US Census (United States Census Bureau, 2023). We also downloaded Zillow Home Values (Zillow, 2023) for all months between June 2020–July 2021 (the 12 months leading up to the sampling) and averaged values across the year to obtain one value for the zip code for each garden. Property value can indicate both economic status and economic support for garden security (Arnold and Roge, 2018). The percent of owner-occupied homes can indicate motivation for maintaining a garden (Glennie, 2020). Household income (and conversely percent of residents below the poverty line) can often influence aspects of garden plantings associated with the luxury effect, or indicate fewer financial resources to support garden tenure (Clarke and Jenerette, 2015; Gregory et al., 2015; Baldock et al., 2019; Arnold and Roge, 2018). 

Finally, we included three variables related to the garden physical characteristics -- garden size (ha), garden age (years), and mean plot size (m²) -- all of which may influence biodiversity of plants and other garden organisms (e.g., Goddard et al., 2013; Clarke and Jenerette, 2015; Quistberg et al., 2016). Garden size was measured from Google Earth Pro (2021) using the ruler function. Garden age was determined from speaking with garden managers and by extracting data from city or other historical documents. Mean plot size was calculated by measuring the area of eight haphazardly selected plots varying in size in each garden and calculating the mean area from those values. 

Data analysis

To examine how plant species richness is related to land tenure, luxury, and garden physical features, we built four generalized linear models (GLMs) with the glm function in R (R Core Team, 2022), with total plant species richness, crop species richness, ornamental plant species richness, and weed species richness as the four dependent variables. We used a Gaussian distribution (default identity link). While the Poisson distribution provided a minimally better model fit, it had very similar residual deviance and df values and only slightly lower AIC values (<2 units), as well as identical significant factors as with the Gaussian models, which is easier to interpret in graphical form; thus, we selected the Gaussian models further analysis. We calculated a variance inflation factor (VIF) for each model set using the car package in R (Fox and Weisberg, 2018) and ensured all VIF scores were below 3. Using this process, we ended up with a final model that included garden size, garden age, plot size, land ownership, land tenure security index, management type, median property value, percent owner occupied housing units, and median household income, the latter three extracted from US census data. We included natural log (LN) transformed values for median property value to improve model fit. We include information on the correlations between the continuous predictor variables in Fig. S2. For GLMs, we tested all combinations of the 9 selected explanatory variables with the ‘glmulti’ package (Calcagno and de Mazancourt, 2010). We selected the top model based on the AICc values. If other models were within 2 AICc points of the best model, we averaged all models within 2 AICc points with the ‘MuMIn’ package (Barton, 2012) and reported conditional averages for significant model factors (as in Coux et al., 2016). We graphed all significant predictors of total, crop, ornamental, and weed species richness with the ‘visreg’ package in R; where we report average models and visualized graphs comprised of the same variables included in the final averaged model (Breheny and Burchett, 2013). 

We similarly wanted to explore the relationships between land tenure, luxury, and garden physical features on plant functional traits, and thus conducted five additional GLMs using the exact same predictor variables and approach described above. As response variables we used the five metrics related to functional trait diversity including functional richness (FRic), functional evenness (FEve), functional divergence (FDiv), functional dispersion (FDis), and Rao’s Quadradic Entropy (Q). To conduct an in-depth examination of variation in individual plant traits, we used the community weighted mean (CWM) for trees, shrubs, plant volume, number of flowers, and flower height as additional dependent variables. We chose these traits because we suspected (more than variables related to flower color, shape or size) that they would be influenced by land tenure, luxury, and hence gardener planting choices. For functional richness (FRic, FEve, FDiv, FDis, Rao’s Q) metrics, and for some of the CWM values (number of flowers, flower height, plant volume) we used a Gaussian distribution (default identity link) as this provided the best fit and lowest AIC values. We used a negative binomial (theta set to 1) for the CWM values for trees and shrubs as this provided the best fit and lowest AIC values. 

We assessed differences in composition for five different communities - total plant species, crop species, ornamental plant species, weed species, and plant trait values as measured with CWM values, in three ways. First, we used a permutational multivariate analysis of variance (PERMANOVA) using the “adonis2” function in the “vegan” package version 2.5–4 (Oksanen et al., 2018) to determine if garden size, garden age, plot size, the land tenure security index, management type, land ownership, median property value, percent owner-occupied housing units, and median household income are significant predictors of dissimilarity. We used Bray-Curtis distances and conducted 999 permutations for each of the five communities. Second, we created nonmetric multidimensional scaling (NMDS) plots in “ggplot2” (Wickham, 2016) to visualize differences in composition and added arrows or colors for the significant predictors to plots using the “envfit” function in the “vegan” package version 2.5–4 (Oksanen et al., 2018). Third, to determine which individual species or traits varied with significant categorical variables (e.g., management type, land ownership), we used the “multipatt” function in the “indicspecies” package version 1.7.12 (De Caceres and Legendre, 2009). To examine which species varied with significant continuous variables (e.g., garden age, median property value, percent owner occupied housing units, and median household income), we ran fractional logistic regression with the quasibinomial family and logit link. To examine which CWM trait values differed with significant continuous variables (e.g., percent owner occupied housing units) we used GLM as described above with the negative binomial family, and theta = 1. A flow chart of all field methods, data collection, variable preparation, and data analysis is shown in Fig. S3.