Introduction

Approximately 3 % of the world’s terrestrial surface is urban (Grimm et al. 2008) and over the next 20 years this area is projected to triple from 2000 levels (Seto et al. 2012). This expansion will result in a variety of environmental changes including increased localized temperatures due to urban heat islands, altered biogeochemical processes, increased pollution levels, natural habitat fragmentation, and an increase in highly managed greenspaces (Pickett et al. 2011). Ultimately, urbanization is a process led by culture, economics, and existing natural structure (Alberti 2005) and varies widely in cities across the world. Generalized principles gained from these studies illustrate that urbanization degrades urban streams and soils, reduces habitat specialists along an urban–rural gradient, and shifts terrestrial trophic dynamics to bottom-up processes as a result of human-introduced resource subsidies (Pickett et al. 2011). Increasing recognition of the large-scale implications of urbanization led to the National Science Foundation funding an additional 20+ exploratory projects (Urban Long-Term Research Areas Exploratory Award Program, or ULTRA-Ex) which continue to study various aspects of urban ecology such as land use, green infrastructure, forestry, watersheds, and the roles of governance and social justice.

The global recession of 2009 altered the urbanization process for many cities in a variety of ways. Many areas in the United States experienced high numbers of housing foreclosures, high rates of job loss, and population decline. Their continuing economic struggles have major implications for future sustainability programs, especially in comparison with more prosperous cities. Termed variously as urban decline, shrinkage, perforation, or deindustrialization, these cities are defined by increased numbers of abandoned sites, demolitions, and decreased proportions of impervious surfaces (Haase 2008). In order to address the various issues surrounding vacant land reclamation an ULTRA-Ex was established in Cleveland, Ohio, in 2009 (Schwarz 2011). Similar to other rust belt cities such as Detroit, Philadelphia, and Buffalo, large-scale population loss has occurred steadily in Cleveland since the 1950s as the economic core of the steel industry waned. Compounded by the recent economic crisis, hundreds of residential and commercial properties have been abandoned and many demolished or foreclosed, leaving an estimated 3,300 acres of vacant land (Cleveland Land Lab 2008). Property maintenance by trash removal and grass mowing costs over $3 M annually (Community Research Partners & ReBuild Ohio 2008). Such social changes have affected not only the city’s economy, but also its greenspace composition. Vacant lands are a signature of these deindustrialized landscapes, but can be considered a resource for urban revitalization (Gardiner et al. 2013). While these parcels are largely managed by the city with monthly mowing, they are dominated by weedy forbs and are generally considered as unsightly. The surrounding community desires more functional land uses, such as parks, sites for stormwater and contamination management, and urban agriculture (Grewal and Grewal 2012; Gardiner et al. 2013, 2014).

Some of these vacant lands will not be immediately converted for alternative uses and maintenance of a portion of vacant land as a form of early successional habitats may be beneficial for urban biodiversity by their serving as greenspace corridors or analogues for more natural environments (Angold et al. 2006; Lundhom and Richardson 2010; Kattwinkel et al. 2011). Urban ecologists should communicate research that supports this ecological land use complementation (Colding 2007), potentially relieving part of the burden experienced by policy decision makers in regards to vacant land reuse. However, cities under duress will benefit most from land reutilization strategies that actively include public participation and meet their social and economic needs. Engaging citizens by supporting the creation of community gardens, park restorations, and other greenspace ‘beautification’ strategies is not an easy task (Gobster 2001; Petts 2007), yet urban land use designs that combine principles of both sociological and ecological sustainability are probably the clearest way forward for vacant land reuse (Hunter and Hunter 2008). These management decisions will alter the habitats that support urban plant and animal biodiversity and their associated ecosystem functions (Zipperer et al. 2000). Highly urbanized landscapes experiencing such changes, such as the core of Cleveland, provide a research site to identify how alteration in greenspace design affects patterns of biodiversity and urban ecosystem functioning.

Urban arthropod communities and their roles as providers of pollination, nutrient cycling, and pest control will be affected by these land use decisions (Isaacs et al. 2009). Arthropods have become a focal group to study urban biodiversity as they are easy to sample, diverse in species and trophic structure, and often respond quickly to habitat change (McIntyre 2000). Spiders are a highly abundant and diverse predator group in natural and urban ecosystems and are known to respond to anthropogenic landscape alterations (Shochat et al. 2004b). We therefore chose spiders as an indicator group for an examination of how vacant land revitalization affects arthropods. In any terrestrial ecosystem, habitat management (Nyffeler and Breene 1990b; Bell et al. 2001) and plant structure and heterogeneity (Hatley and MacMahon 1980; Harwood and Obrycki 2007; Simao et al. 2010) influence the ability of spiders to colonize, hunt, and reproduce within a site. This ultimately structures their abundance, diversity, and composition. Plant communities are generally the only directly controlled biotic component of urban habitats and serve as the ‘template’ for the configuration of other trophic levels (Faeth et al. 2011). The conversion of vacant lots into greenspaces with alternative management systems, such as community gardens or low maintenance native plantings, will alter plant productivity and disturbance regimes, and in turn influence arthropod biodiversity and their provision of ecosystem services.

Our objective was to measure the community response of spiders to potential urban vacant land redesign in Cleveland, Ohio. We measured their activity densities, diversity, and assemblage composition in residential vacant lots, urban gardens, and planted urban prairies within a peri-urban park system. Urban agriculture is a prominent strategy for economic sustainability in Cleveland (Grewal and Grewal 2012), while transformation of vacant lands into wildlife habitats such as planted prairies has become a growing trend for inner city vacant land management (e.g. Earth Day Coalition’s program Naturehood). Within this study system our principal questions were:

  1. (1)

    How do spider assemblages differ among urban habitats in measures of abundance and richness over 3 months of the summer growing season?

  2. (2)

    How does the level of taxonomic resolution achieved affect the interpretation of spider assemblage responses to habitat change?

Materials and methods

Study area

The city of Cleveland, Ohio, USA (41°28′56″N 81°40′11″) is part of a metropolitan area with a population of 393,806 at a density of 5,113 persons per square mile (2010 US Census). Cleveland has a history as a manufacturing center, yet contains 1335+ hectares of vacant land after decades of urban decline and population loss (Cleveland Land Lab 2008). The city is located on the southern coast of Lake Erie and experiences a continental climate (average January temperature −2.6 ° C, July 22.7 °C; NOAA 1981–2010 normals).

Study sites

A total of 23 study sites included eight vacant lots that were previously residential, eight gardens established on formerly vacant lands, and seven planted prairies (Appendix Fig. 4). Prairies were selected to be representative of potential ‘urban prairie’ habitats to be established on vacant lands in the future; although they were part of the city’s Metropark system, these sites were still generally surrounded by high levels of urbanization. Sites were split into 4 sampling plots based on the length of the longest edge. Prairies were on average larger (6833 m2 ± 1615 SE) than both vacant lots and gardens (2857 m2 ± 767 SE, 1736 m2 ± 491 SE, respectively). Area measurements were made in Google Earth Pro and site boundaries were defined by property parcels, tree lines, and/or fences. Prairie management consisted of early spring mowing or burning and occasional removal of invasive plants. Vacant lots were mowed monthly by the City of Cleveland. Garden management consisted of crop planting, harvesting, mulching, weed removal, and frequent irrigation, but no use of broad spectrum insecticides. All gardens were at least in their second year of establishment at the beginning of the study. Study sites were separated by a distance of 450 m to 33 km.

Spider sampling

Within each sampling plot, collections targeting ground-active and vegetation-dwelling spiders were conducted using pitfall traps and vacuum sampling. One pitfall trap per plot was randomly placed and site edges were avoided by a minimum of 10 m. An edge was defined as a tree line, fence, or sidewalk. Holes for the pitfall traps were dug using a golf-course cup cutter to reduce soil disturbance around the traps, which were 0.95 L plastic cups, unbaited, and filled halfway with a 10 % soap solution. Pitfall traps were set for 1 week during 3 months of the growing season (set dates 2011: June 14–16, July 12–14, August 16–18; 2012: June 11–13, July 9–11, August 7–9). Each collection day was structured to reduce traveling time between sites while maintaining equal treatment combinations. Vacuum samples were collected using a reversed leaf blower (STIHL BG 72, Virginia Beach, VA) where samples were collected in an approximately 700 mL tube. A 1 m2 quadrat was placed in the proximity of each pitfall trap and vegetation was vacuumed for foliage-active spiders. Vacuum samplings were conducted once during each pitfall trap collection period either at the time of trap placement or collection, depending on the presence of rain, between 0900 and 1500 h.

Pitfall traps have a variety of limitations and are not always reflective of absolute population abundances (Topping and Sunderland 1992), but they are a suitable method for comparing between-site differences in activity density. Adult and sub-adult spider (hereafter also referred to as adult) specimens were identified using the available taxonomic literature (Dondale and Redner 1978, 1982, 1990; Paquin and Dupérré 2003; Platnick and Dondale 1991; Prentice 2001; Sandlin 2013; Ubick et al. 2005) and nomenclature follows Platnick (2013). All adult, sub-adult, and adult carapace specimens were included in total density counts. Oftentimes adult specimens fall apart in pitfall traps, but the carapace can be used for at least genus identification. Juveniles were counted separately. Specimens that were not easily identifiable to genus were grouped together as one ‘genus’ within their respective family (e.g. ‘other salticids’). These were included as such in analyses. All analyses were conducted at both genus and family level. Some species were the only genus representative and easily discernible, so their full species names are provided.

Data analyses

Trap catches were pooled for each site during each sampling month for analysis with pitfalls and vacuum samples analyzed separately. All statistical analyses were carried out using the open statistical environment R version 3.0.0 (Development Core Team 2013). We compared differences in activity densities of total spiders and select families and genera using generalized linear models (GLMs) for both sampling years separately. Activity density models were constructed using a negative binomial error distribution fitting for overdispersed count data (O’Hara and Kotze 2010) using a log-link function and maximum likelihood Laplace approximation. In models of pitfall trap activity densities, an offset variable for trap number was also included to account for lost traps or entire sites during sampling (losses of entire sites included one garden in August 2011 and the same garden plus one prairie in August 2012).

In lieu of the traditional null hypothesis testing (e.g. t-tests and analysis of variance) for determining the influence of predictive factors, the information-theoretic (I-T) approach allows for a more informative probability test of the null hypothesis along with any other alternative hypothesis/es, as opposed to simply testing one hypothesis against the null (Burnham and Anderson 2002; Burnham et al. 2011). Essentially, I-T methods allow us to infer the probability of a hypothesis given the data, whereas traditional null hypothesis testing provides the probability of the collected data given the null (Lukacs et al. 2007). To examine the differences among greenspace type (habitat), period of the growing season (month), and the possible interactions of these factors on activity density and diversity responses, we constructed the same alternative models for each response variable (including a null model; Appendix Table 3).

According to the I-T Akaike’s information criterion corrected for small sample sizes (AICc), models were ranked and compared. Models with AICc differences <2 were considered competing models and Akaike weights were used to determine the strength of evidence for each competing model. For best and competing models we reported the maximum likelihood pseudo-R 2 for GLMs. In figures, means are presented as ± SE. Because we were interested in comparing activity densities within greenspaces throughout the growing season, and the potential implications for ecosystem functioning, for clarity of comparisons post hoc pairwise comparisons of responses within habitat during each sampling month were determined using the appropriate distribution.

The following analyses on spider assemblage richness and compositions were conducted using only pitfall trap data, which were identified to genus level unlike spiders collected by vacuum sampling. To account for the fact that not all species can be sampled with one method and therefore true richness was not observed within each habitat (Colwell and Coddington 1994; Gotelli and Colwell 2001), we extrapolated richness estimates using three incidence-based richness estimators (Chao1, first order jackknife, and bootstrap). Extrapolated estimates are useful when time and site availability for sampling are limited (Colwell et al. 2004), as the case would be for many urban studies that rely on acquiring collection permission from a variety of landowners. For this analysis, we pooled specimens within each site across sampling months and again examined results separately for both years.

Finally, we used nonmetric multidimensional scaling (NMDS) to visualize the spider assemblages within each habitat. Ordinations were conducted using Bray-Curtis distances in three dimensions. We visualized these differences for both sampling years with total pooled spiders per site, and also for each sampling month to see how these assemblage compositions potentially changed throughout the growing season. Significant differences were tested using permutational multivariate analysis of variance (PERMANOVA) which allows for any type of distance measure (Anderson 2001). We again used Bray-Curtis distances with 9999 permutations. These ordinations are not based on log-likelihoods and therefore we did not utilize I-T methods for these results.

Results

Activity density: Pitfall traps

Pitfall traps collected 21 families and 66 identified genera out of a possible 38 spider families and 220 genera within those collected families that are recorded in Ohio (Bradley 2013; Table 1). Fewer spiders were captured in 2012 (3330) compared to 2011 (4028). Lycosidae (wolf spiders) and Linyphiidae (sheet-web weavers) were the most commonly collected families in all habitats, constituting 74.0 and 13.7 % of the collections in 2011 and 66.0 and 21.4 % in 2012, respectively. Juveniles were at least 90 % Lycosidae due to the frequent capture of reproductive females carrying offspring.

Table 1 Spiders collected from pitfall traps in prairies, vacant lots, and community gardens in Cleveland, OH
Full size table

In general, the best fit models for activity densities were the habitat only and the additive habitat and month, indicating that significant differences among habitats remained consistent even if activity densities differed among sampling periods (Table 2). In 2011, total spiders were most active in vacant lots compared to prairies and gardens, except in June when activity densities were similar among vacant lots and prairies (Fig. 1a). Activity densities in gardens were consistently low, while activity densities in prairies declined throughout the summer and became equivalent to garden levels in August. However, the habitat only model was identified as the best fit (model weight 0.715). In 2012, spiders were again most active in vacant lots and did not differ among gardens and prairies during any month; overall numbers generally decreased in all habitats during the growing season (Fig. 1f).

Table 2 Results for model comparisons of GLMs measuring activity density in regards to greenspace habitat and sampling month
Full size table
Fig. 1
figure 1

Activity density per month (mean ± SE) of total adult spiders, Lycosidae, Pardosa, Linyphiidae, and Grammonota in 2011 (a–e) and 2012 (f–j). Best and competing models (with R 2) are indicated for each plot. Post hoc comparisons are illustrated within each month: in comparison to a black dot, a white dot indicates significantly lower, a grey dot indicates intermediate and equal to both other values, and * indicates all values are significantly different

Full size image

Patterns of total spider activity were largely driven by the most commonly collected family, Lycosidae (Fig. 1b and g). Pardosa was the most common genus collected (53.9 % overall catch, 76.6 % of Lycosidae). In 2011, Pardosa activity nearly quadrupled in numbers in vacant lots across the summer (Fig. 1c). Their densities in prairies declined over time, while in gardens they were consistently low. The habitat only model was selected as the best model, but the explained variance of this single factor was generally low (R 2 = 0.189). In 2012, Pardosa activity was again equivalent in all habitats in June, but increased in vacant lots while remaining steady in prairies and gardens (Fig. 1h). Explained variance of models were again low for the best fit models (R 2 = 0.153 and 0.208), indicating that Pardosa activity was quite variable among sites.

Other common lycosids were also most often found in vacant lots, including Allocosa funerea and Trochosa ruricola (6.6 and 13.3 % of Lycosidae, respectively), both of which generally declined in activity during the summer in all habitats. These two species along with Pardosa were found in all habitats. Seven additional identified lycosid genera were collected only in prairies, with the exceptions of two Pirata in vacant lots in June 2012 and one Rabidosa rabida specimen in a garden in July 2012 (Table 1).

In 2011, Linyphiidae were also most active in vacant lots, but numbers generally declined in all habitats throughout the summer (Fig. 1d). According to pairwise comparisons, their activities were equivalent in all habitats in June, but became increasingly different in vacant lots and gardens, with prairies as intermediate. In 2012, Linyphiidae remained highest in vacant lots during all months with little fluctuation (Fig. 1i). Grammonota was the most commonly caught linyphiid genus (38.1 %), and throughout the summer remained consistently active in vacant lots, drastically more so than in gardens and prairies, in 2011 (Fig. 1e) and 2012 (Fig. 1j), where collections in the latter two habitats were close to zero. The collections of Grammonota in vacant lots 2012 were double that seen in 2011.

Estimates of richness

Extrapolated richness estimates were similar during both collection years (Fig. 2). Due to the high incidences of rare genera and families, both prairies and gardens were expected to be richer than what was observed. Estimated values of richness in vacant lots were generally consistent with the observed values during both collection years, indicating that the ground active spider assemblages found in this habitat were well represented in our collections. When analyzed at the family level gardens had slightly higher estimated values compared to vacant lots, indicating potentially richer assemblages than represented by collections (Fig. 2c–d). However, genus-level richness estimates within vacant lots and gardens were similar, and therefore these habitats likely contained similar numbers of genera (Fig. 2a–b).

Fig. 2
figure 2

Incidence-based estimates of richness (mean ± SE) at genus and family levels in 2011 and 2012

Full size image

Spider assemblage composition: Pitfall traps

Spiders differed in overall genus composition among habitats (PERMANOVA for pooled specimens per site; 2011: F2,22 = 8.1621, R 2 = 0.4494, P < 0.05; 2012: F2,22 = 4.0028, R 2 = 0.2860, P < 0.05), although vacant lot assemblages considerably overlapped the other two habitats during both 2011 (Fig. 3a) and 2012 (Fig. 3b). The most abundant genera were concentrated within vacant lots, but appeared in all habitats (ranked top 10 abundances pooled over both years: Pardosa, Trochosa, Grammonota, Allocosa, Tennesseellum, Glenognatha, Xysticus, Erigone, Pachygnatha, Zodarion; Table 1). Within individual prairies, assemblages were highly variable as visualized by the large NMDS spread (Fig. 3a–b). Prairies were most different from gardens (NMDS axis 1) which were also quite variable. The tight clustering of vacant lots indicates low variation within this habitat. These significant differences were also observed at the family level, but with more apparent habitat overlap compared to genus level analysis (PERMANOVA 2011: F2,22 = 11.213, R 2 = 0.5286, P < 0.05, Fig. 2; 2012: F2,22 = 5.3612, R 2 = 0.3490, P < 0.05; Fig. 3c–d).

Fig. 3
figure 3

Nonmetric multidimensional scaling (NMDS) plots at genus level (2011 a, 2012 b) and family level (2011 c, 2012 d). Ordinations based on Bray-Curtis distances. Taxonomic names not included for clarity. Ordination stress values a–d: 0.093, 0.095, 0.122, 0.130. All plots in 3 dimensions

Full size image

Spider assemblage composition among all greenspace types was also significantly different when analyzed during each month separately at genus level composition (Appendix Figs. 5 and 6). When comparing only vacant lots to gardens (which had the most similar surrounding landscapes) assemblage differences remained significant at the genus level overall (PERMANOVA for pooled specimens per site; 2011: F1,15 = 8.4969, R 2 = 0.3777, P < 0.05; 2012: F1,15 = 4.4017, R 2 = 0.2392, P < 0.05) during every sampling month (Appendix Table 4). At the family level, vacant lots and gardens were not significantly different during June 2012, but were for the remaining sampling months (June 2012 PERMANOVA: F1,15 = 1.94934, R 2 = 0.1219, P = 0.1234; Appendix Table 5).

Activity density: Vacuum samples

We collected 13 and 14 spider families in 2011 and 2012, respectively, by vacuum sampling. The three most abundant families during both years were Linyphiidae (43.0 %, 48.6 %), Salticidae (16.3 %, 11.2 %), and Lycosidae (11.1 %, 15.0 %). More juveniles and adults were collected with vacuum sampling in 2012 (2011: 454 juveniles and 135 adults; 2012: 1278 juveniles and 214 adults). Juveniles were most commonly captured in prairies, especially later in the season (2011 best model habitat×month, ΔAICc = 0.0, K = 10, R 2 = 0.569; 2012 best model habitat+month, Δ AICc = 0.0, K = 6, R 2 = 0.747). Adult numbers were similar in prairies and vacant lots, lowest in gardens, and highest in June in 2011 (best model habitat, Δ AICc = 0.0, K = 4, R 2 = 0.247; next best model habitat+month, Δ AICc = 0.1, K = 6, R 2 = 0.296) and August in 2012 (best model habitat+month, Δ AICc = 0.0, K = 6, R 2 = 0.380; next best model habitat×month, ΔAICc = 1.3, K = 10, R 2 = 0.460). During specimen identification we decided that the genus composition did not appear different enough from pitfall traps to warrant additional analysis.

Discussion

The reutilization of vacant lands in Cleveland and other urban areas will impart social, economic, and ecological impacts. When the ecological goals of greenspace management include invertebrates the focus is often on butterflies and other pollinators (New et al. 1995; Connor et al. 2002). Indeed, one of the surveyed prairies in our study is a monitoring point for butterflies that was seeded with plants specifically for their benefit (Monarch Waystation at Brookside Reservation, Cleveland Metroparks). Urban community gardens also often include plantings set aside for pollinators in order to bolster this service within crops. In this study we demonstrate how another ubiquitous and beneficial invertebrate group responds to urban habitat management. Spiders as a key predatory guild have been examined in other cities, such as Phoenix (Shochat et al. 2004b; Bang and Faeth 2011). However, Cleveland provides a unique and informative arena for studying how deindustrialization impacts generalist predators and ecosystem functioning. Within this urban ecosystem, we examined how spider assemblages differed among vacant lots, urban gardens, and peri-urban planted prairies in measures of abundance and richness across the 2011–12 growing seasons. We also quantified how the level of taxonomic resolution achieved affects the interpretation of spider assemblage responses to habitat change. We predicted that: (a) strong environmental filtering would result in a dominance of disturbance-tolerant species within vacant lots, able to rapidly colonize a patch following mowing (Swan et al. 2011); (b) prairies would contain the richest spider assemblages including both habitat specialists and urban exploiters, given that these peri-urban parklands were surrounded by more contiguous vegetation and experienced the lowest level of vegetative disturbance (McKinney 2006; Dearborn and Kark 2010); and (c) gardens would support the lowest spider activity density due to frequent and intense management disturbances (e.g. Gardiner et al. 2014). Because gardens and vacant lots were located more within the urban core and had similar surrounding landscapes, we further predicted (d) that these two habitats would be more similar in spider assemblages compared to prairie sites. Although an urban ecosystem is complex and potentially rapidly changing in greenspace composition (Sattler et al. 2010; Ramalho and Hobbs 2012), we also predicted (e) that detected patterns would be similar between sampling years. Our second research question considered how the level of taxonomic resolution achieved affected the interpretation of spider assemblage responses to habitat change. Here, we predicted (f) that a lower resolution would demonstrate that habitats were compositionally different, even if similarly rich.

Habitat generalists and specialists

Species assembly within a particular location will be driven by a combination of factors, including site disturbance and plant productivity, along with species-specific competitive and dispersal abilities. The dominance of Lycosidae and Linyphiidae we detected in our pitfall traps is common and reflective of their epigeal activity and ease of capture (Topping and Sunderland 1992). Activity densities of these families was highest in the vacant lots throughout the summer sampling months, a pattern driven largely by the presence of disturbance-tolerant genera supporting our first prediction (a). Swan et al. (2011) notes that vacant lots and other abandoned properties are more likely to host plants that are highly dispersive colonizers and generally weak competitors; we found these patterns can also be descriptive of their spider assemblages. Lycosidae have been found to dominate mesic yard habitats in desert cities (Shochat et al. 2004b), including Pardosa and Allocosa (Bang and Faeth 2011) as was the case in our study, and also within Toledo, Ohio vacant lots (Moorhead and Philpott 2013). Allocosa funerea and P. milvina (the dominant Pardosa species collected) are also often the most abundant lycosids in regularly maintained Midwestern lawns (Cockfield and Potter 1984). Pardosa are highly mobile foragers adapted to ephemeral habitats (Samu et al. 2003), especially agricultural fields (Marshall et al. 2000), and other open and barren habitats (Mallis and Hurd 2005). Trochosa ruricola is also commonly found in agroecosystems in North America (e.g. Bolduc et al. 2005) and Europe (e.g. Schmidt et al. 2008). Indeed, many species collected within these vacant lots and gardens are frequently found in agroecosystems (Schmidt et al. 2008; Chapman et al. 2013), a commonality indicative of the similarity in environmental pressures imposed by urbanization and agricultural management.

The web-building linyphiids most often collected within the highly urbanized vacant lots are also known habitat generalists found frequently in agroecosystems (Schmidt et al. 2008; Chapman et al. 2013), specifically the species within the subfamily Erigoninae (e.g. Grammonota inornata, Erigone atra, E. autumnalis, E. dentipalpis, [identified, but not analyzed within this paper at the species level] and Tennesseellum formica). Vegetation cutting immediately decreases web densities (Birkhofer et al. 2007), but the speed of web reconstruction will be species-specific due to differing web structuring requirements. Species within the subfamily Linyphinnae generally build webs higher in the vegetative canopy (Harwood and Obrycki 2005) and sit within their webs for multiple days (Janetos 1982), potentially explaining their more frequent occurrence in prairies (e.g. Bathyphantes, Diplostyla concolor, Neriene) which are less-disturbed habitats with increasingly taller vegetation throughout the growing season. Comparatively, Erigoninae are more likely to leave their webs in pursuit of prey (Harwood and Obrycki 2005). However, some Linyphinnae were collected in vacant lots and gardens (e.g. Bahtyphantes, Lepthyphantes, Porrhomma terrestre), reflective of the fact that all sites contained microhabitats with taller, less-disturbed vegetation. These so-called ‘wild’ or unmanaged areas within urban community gardens have been found to be important for the support of diverse assemblages of jumping spiders and other beneficial organisms such as pollinators (Cumming and Wesołowska 2004; Matteson and Langellotto 2010). Our findings lend more support for the inclusion of undisturbed vegetation by gardeners and garden managers in order to provide beneficial habitat for diverse spider assemblages and their associated ecosystem services.

Our second prediction (b) that prairies would host a more diverse spider assemblage, composed of both habitat specialists and urban generalists, was also supported. As clearly evidenced by their richer lycosid assemblage, the prairie assemblages consisted of genera that have more specialized habitat requirements (e.g. Hogna, Pirata, Rabidosa, Schizocosa; Mallis and Hurd 2005). For example, S. ocreata is most frequently found in woody areas with high leaf litter (Wagner and Wise 1996) and Pirata occur frequently in areas close to bodies of water (Graham et al. 2003). In our survey, both of these microhabitats were generally only available in prairies located with a contiguous urban park system. Habitat generalists (e.g. Pardosa) are able to colonize these prairies, but may face stronger and more complex competition pressures there (Marshall et al. 2000). Other ground-active wandering hunters were also more often present in prairies, including multiple species of Corinnidae and Gnaphosidae. These differences in spider abundances and diversity may further influence community trophic dynamics in urban areas. For example, foraging birds, which are often under consideration in urban habitat conservation plans (Melles et al. 2003), utilize a diverse spider assemblage for adult and nestling diets (Sekov and Averenskii 2011; Ostrand and Bollinger 2012), and are influential in top-down trophic dynamics within cities (Shochat et al. 2004a).

The generally higher levels of disturbance within gardens due to mulching, crop planting, weeding, and harvesting may have contributed to the consistently lower activity densities of spiders in this habitat, which supported our third prediction (c). Spiders have been found to be less abundant in traditional agroecosystems compared to more natural ecosystems (Nyffeler and Benz 1987), which has influenced ongoing studies on landscape-scale habitat conservation to support more abundant and diverse spider (Samu et al. 1999) and other natural enemy communities. Curiously, the activity densities within our study prairies were similarly low, but this may have more to do with detectability issues than a reflection of actual abundances. Pitfall traps surrounded by dense vegetation, as was the case in prairies, influences the ability of a spider to escape even after encountering a trap (Topping 1993). Capture efficiency is species-specific regardless of trap surroundings (Topping 1993). Thus, low prairie activity densities might not be fully reflective of actual low abundances for all species. Regarding richness, the extrapolated estimates indicate that both prairies and gardens may be richer than what was estimated by observed numbers. However, the spread of the NMDS plots highlight that these habitats are still generally compositionally different.

Greenspace management and landscape connectivity

Because gardens and vacant lots were located within the urban core and had more similar surrounding landscapes, we predicted (d) that these habitats would have more similar spider assemblages when compared with prairies. We found partial support for this prediction as there was considerable overlap among vacant lot and garden assemblages; however, significant differences were apparent. Vacant lot spider assemblages were very similar, with collections overwhelmingly dominated by genera such as Pardosa, Erigone, and Grammonota in the majority of sites (Table 1). The most common genera in vacant lots were also generally the most commonly found in gardens. However, singleton and doubletons not collected in vacant lots were often collected in gardens. Therefore, garden spider assemblages, like in prairies, had a generally larger spread among collection sites compared to vacant lots. The latter is an example of a self-assembled urban plant community (Swan et al. 2011) where management of particular species is not a focus (at least after turf grass seeding is completed). Alternatively, urban prairies and gardens are facilitated communities where plant species are selected by people for particular traits, such as native status, hosting particular blooms to attract pollinators, or agricultural crop production. These urban habitats receive a much higher socio-economic feedback compared to self-assembled communities (Swan et al. 2011), where specific goals for conservation, aesthetics, and agriculture control the decision making for each site generally independently. Planting composition can be quite similar, as seen in private gardens across five UK cities (Loram et al. 2008), or exhibit significant variation in small-scale structuring (Thompson et al. 2003; Matteson and Langellotto 2010). These dissimilarities between facilitated plant communities may be ultimately reflected in higher trophic level structuring (e.g. spiders) at the patch scale.

Along with the plant community productivity and disturbance level differences between these three examined habitats, the issue of landscape structure has major implications. Within traditional agroecosystems, spiders are generally more diverse and abundant within fields surrounded by a more heterogeneous landscape which provides a large pool of habitats (e.g. Schmidt et al. 2008; Gardiner et al. 2010). However, colonizing patches with lower greenspace connectivity is probably not a major obstacle for many urbanized and agrobiont spider species as they are already adapted to utilizing ephemeral sites and have high dispersal capacities (Samu and Szinetár 2002). For example, Vergnes et al. (2012) found that spider richness was similar in Parisian private gardens either connected or disconnected from woodland habitats. However, the propensity of a particular spider to disperse will depend on its evolutionary adaptations and individual physiology (Weyman et al. 2002). Habitat specialists are less likely to balloon due to selection pressures against risking the chance of landing within an unsuitable location (Bonte et al. 2003). We did find evidence that some habitat specialists more common in prairies were also able to colonize inner city vacant lots and gardens. For example, Oxypoidae juveniles, but not adults, were also frequently collected in vacant lots by vacuum sampling illustrating that they were able to colonize but perhaps not survive to adulthood within this habitat. Whether these taller-vegetation foragers and other habitat specialists are able to thrive within inner city greenspaces with changes in ecosystem management, such as the establishment of small prairie plantings on former vacant lots, remains to be seen.

Supporting our fifth prediction (e), the observed and estimated levels of richness were generally similar between sampling years. However, there was evidence of variation in abundances. Lower activity densities of spiders collected by pitfall traps in vacant lots and grasslands during the second year may be attributable to variations in precipitation and temperature. Shochat et al. (2004b) recorded higher abundances of spiders in multiple urban habitats during a wet season after an El Niño year compared to a subsequent drier year. In Cleveland, 2011 had higher precipitation levels compared to 2012, particularly in April and May (approximate precipitation highs in April and May 2011: 4.45 cm, 3.18 cm, respectively; in 2012: 1.27 cm, 1.91 cm, respectively; NOAA recordings, Cleveland Hopkins Station). Additionally, higher abundance of spiders collected by vacuums in 2012 could be due to a higher propensity of spiders to balloon out of stressful environments (Weyman et al. 2002). The relatively similar activity densities in gardens between years may be explained by the use of regulated irrigation, creating a potentially more stable microclimate (Steiner et al. 1983). These changes are probably not directly due to precipitation, but instead to the indirect effects on foliage growth and prey availability (Shochat et al. 2004b). Responses to these environmental changes will also be species-specific. Less precipitation may have depressed vegetation growth in vacant lots, creating more bare patches useful for web-construction by Grammonota and Tennesseellum formicum (Welch 2013). Both of these species utilize epigeal and basal sites and were doubly abundant in vacant lots during the drier 2012. Alternatively, Glenognatha foxi build webs among foliage (Welch 2013) and were much less abundant in both prairies and vacant lots in 2012. Best fit models for activity densities were also often less explanatory in 2012, indicating high variability during that year. The generally low to moderate R 2 values overall (~0.2–0.5 for activity density models) further supports the idea that environmental stochasticity particularly defines urban spider communities (Sattler et al. 2010). A simple definition of habitat and month component does not fully encapsulate the fine-scale structuring due to climatic changes and ‘randomness’ inherent to human habitat management. A multitude of complicated sociological factors such as infrastructural layout and neighborhood economics will continuously influence the dynamics of highly fragmented urban ecosystems (Collins et al. 2000). Finally, as evidenced in the NMDS plots, the prairies and gardens were quite compositionally variable within themselves, as the case would be for facilitated communities which are independently managed (Swan et al. 2011).

Implications for urban greenspace management

Our second research question regarding taxonomic resolution stemmed from the fact that spiders are often included in urban ecological studies but can be difficult to identify to species level without years of training (Cardoso et al. 2004). We investigated the taxonomic sufficiency of family-level resolution compared to genus on overall patterns of richness and composition. Although species-level resolution is most appropriate for investigations regarding particular trophic interactions or invasions (Timms et al. 2013), our aim was not to create a biological inventory for the city of Cleveland, but instead to quantify how spider communities are generally influenced by urban greenspace management and for their continued incorporation into rapid biodiversity assessments for urban management and conservation purposes. We found that genus-level resolution compared to family provides a more informative (and complex) picture of spider responses to urbanization (e.g. Cardoso et al. 2004), particularly due to genus and species-specific responses to greenspace management throughout the growing season, supporting our prediction (f). When comparing activity densities and assemblage compositions, the effect of month was often dampened at the family level. Genus-level resolution has been found to be reflective of species-level response to forest disturbance regimes for spiders and other arthropods (Timms et al. 2013), and we propose that their identification to this level in our assessment of greenspace management will be sufficient.

An examination of spider assemblage response to potential vacant land reutilization can be informative for land managers interested in both urban agriculture and conservation. However, regarding urban agriculture in particular, an assessment of spiders’ contribution to ecosystem function will have to go beyond a description of merely richness and abundance. Pardosa, while highly active in gardens, have generally low feeding rates (Nyffeler and Breene 1990a), but Linyphiidae and other web-builders often capture more prey items than they actually feed upon (Riechert and Maupin 1998). Maximizing the availability of microhabitats for both ground-dwelling erigonines and linyphiines higher in vegetation would also allow for a wider capture of prey items (Harwood and Obrycki 2007). Spider assemblage fluctuations throughout the growing season will also have impacts on their biocontrol potential (Welch et al. 2011).

Regarding urban conservation, biotic homogenization (McKinney 2006) is an issue as many species collected are worldwide agrobionts or exotics potentially displacing native species. For example, T. ruricola is originally a Palearctic species and has been shown to be displacing the native T. terricola in some Canadian agricultural sites (Lalongé et al. 1997; Bolduc et al. 2005). The mechanisms of displacement may be caused by reproductive potential and timing differences (e.g. Hann 1990; Eichenberger et al. 2009), but the resilience of particular native populations will be influenced by the presence of structurally complex, undisturbed habitats (e.g. Nyffeler et al. 1986; Burger et al. 2001; Hogg and Daane 2012). The growing trend of replanting vacant lands as prairies may provide inner city locations for native spider species, but also new spaces for competition. Therefore, smaller prairies planted on vacant lands may not hold the same spider assemblages found within the peri-urban parks. The habitat specialists we collected will be more reliant on greenspace connectivity between peri-urban parks and inner city prairies compared to habitat generalists, similar to butterflies that differ in habitat specialization and dispersal abilities (Snep et al. 2006). Therefore, planning for larger and less isolated prairies (Knapp et al. 2008) may be required for future plantings in order to support a higher diversity of plant and animals within the city.

Early education on the conservation value of spiders and other invertebrates can be aided by both these planting programs and urban community gardens as long as the appropriate outreach steps are taken by ecologists, community planners, and land managers. Many of the workers and volunteers in these sites are inner city youth who are most at risk for never gaining the opportunity to value nature, instead become increasingly isolated from and apathetic about such experiences, a scenario described deftly by Miller (2005). Conservation even within peri-urban areas will not be enough due to various economic disparities. Poorer children may not have economic access to transportation allowing them to experience the larger park systems generally concentrated in more affluent areas, a situation found in many cities such as Los Angeles (Joassart-Marcelli 2010) and Cleveland (Scott and Munson 1994). Demonstrating both the usefulness and beauty of spiders, butterflies, bees, and other arthropods in education programs should continue to be a focus for ecological outreach. An integration of goals useful for both urban dwellers in and conservation purposes (Hunter and Hunter 2008) will maximize human well-being, natural education, and biodiversity in both growing and deindustrializing cities.