Introduction

With more than 50% of the world’s population living in urban areas (United Nations 2019), cities are rapidly expanding worldwide. The associated transformation of wildlife habitats is considered a leading cause of biodiversity loss (Fahrig 2003), but awareness of the need for urban wildlife protection has increased in the last decades (Lin and Chen 2022). Strategies for sustaining biodiversity and related ecosystem services have been rising in cities through the integration of green infrastructures (Aznarez et al. 2022), also mitigating urban warming (Bartesaghi-Koc et al. 2020).

The urban physical environment host a myriad of environmental conditions that represent a multi-dimensional ecological filter often associated with a loss of taxonomic and functional diversities among animal and plant communities (Aronson et al. 2014; McKinney 2006). Biotic and abiotic factors occurring across the urban matrix determine the presence and the abundance of species, based on their ability to cope with urban conditions (Delgado de la flor et al. 2024; MacGregor-Fors et al. 2022). Accordingly, species can be classified as urban dwellers, utilisers or avoiders (Fischer et al. 2015). It depends on whether they can pass through the urban filter, but also establish over a longer term. The resulting community distribution of species across space leads to the emergence of diversity patterns along urbanisation gradients (Aronson et al. 2014; Hahs et al. 2023). To explain these patterns, recent urban ecology studies have aimed to disentangle the roles of specific urban factors (Beninde et al. 2015) and investigated landscape (e.g., connectivity, heterogeneity or the proportion of impervious surface) and local (e.g., vegetation structure, microtopography or management practices) predictors distinctly. Landscape predictors are generally considered as upper-level filters that determine the permeability of the surrounding urban matrix for the immigration of species from rural adjacent areas into cities (Beninde et al. 2015; Faeth et al. 2011). Local predictors are intended to determine habitat suitability and are mainly related to species survival, reproduction success and abundance (Beninde et al. 2015; Proske et al. 2022). Although ecological filters operate at both spatial scales (Christie et al. 2010; Kyrö et al. 2018; Norton et al. 2014; Otoshi et al. 2015; Philpott et al. 2014), the relative weights of landscape and local-scale predictors in shaping diversity patterns differs across taxa, and is strongly linked to the functional traits of species (Croci et al. 2008).

Urban temperature-related predictors are increasingly taken into account, in link with habitat-related factors. The urban heat island (UHI) is characterised by an increase of the mean temperature of urban cores relatively to adjacent rural areas (Oke et al. 2017), mainly resulting from warmer night temperatures (Dubreuil et al. 2020; Foissard et al. 2019; Oke et al. 2017). Atmospheric UHIs depict air warming above the surface (e.g., at a 2-m height, Foissard et al. 2019), and are characterised as large areas of heat at the landscape scale (Oke et al. 2017). While many ecological studies have dealt with the relationship between atmospheric UHIs and vegetation, including in our study area (Mimet et al. 2009), the relationship with arthropod fauna is much less documented. Arthropods are ectothermic species whose metabolism depends on outside temperature. They are particularly interesting ecological targets in cities because of their sensitivity to temperature and consequently to warming (Hamblin et al. 2018; McGlynn et al. 2019; Menke et al. 2011). First results showed that the urban atmospheric UHI measured at the canopy level was associated with shifts in arthropod phenology (Meineke et al. 2014) and community composition (McGlynn et al. 2019). Other studies measuring local temperature within a few meters reported associations between near-ground temperature and changes in arthropod abundance (Hamblin et al. 2018; Piano et al. 2020a, b; Shochat et al. 2004). Although most studies emphasize the physiological implication of near-ground temperature for single taxa (Angilletta et al. 2007; Diamond et al. 2018; Yilmaz et al. 2022), information about community-wide responses to near-ground temperature still remains scarce (McGlynn et al. 2019; Piano et al. 2020a, b).

We expect different ecological responses in cities depending on the scale considered to characterise temperature, as in the case of land cover and habitat. The landscape-scale atmospheric measurements strongly differs from local-scale near-ground temperatures (Lembrechts et al. 2020). The standardised atmospheric UHI is a reliable metric commonly used to characterise warming patterns at the landscape scale, whereas near-ground temperatures are spatially heterogeneous and commonly referred to as thermal mosaics (McGlynn et al. 2019; Ziter et al. 2019) due to the fine-scale complexity of vegetated and built surfaces within cities. Temperature measured at the ground level matches the scale at which arthropods experience their thermal environment (Pincebourde and Woods 2020). As a result, local thermal heterogeneity provides opportunities for small organisms to buffer adverse thermal conditions through behavioural thermoregulation (Woods et al. 2015). As with habitat-related predictors, it is necessary to consider multiple – local and landscape – scales to study the response of arthropod communities to temperature. Near-ground (local) and atmospheric (landscape) measurements of temperature have rarely been taken into account simultaneously in urban ecology studies because of the lack of sensor networks for monitoring temperature at different spatial scales.

Spiders are ideal biological models for urban ecology studies because they are convenient to collect and can be found abundantly along urbanisation gradients (Piano et al. 2020a, b; Varet et al. 2011; Vergnes et al. 2014). Furthermore, they are highly diverse groups that respond rapidly to urban environmental stressors, and substantial literature is available about landscape and local habitat predictors driving their taxonomic and functional diversities (Buchholz et al. 2018; Piano et al. 2020a, b; Sattler et al. 2010; Varet et al. 2011; Vergnes et al. 2014). Moreover, spiders are prey and predators, making them important links in trophic chains and the basis of many ecosystem services. They are also ectotherms, and this makes them particularly sensitive to urban warming (Johnson et al. 2020; Meineke et al. 2017).

In this study, we aimed to disentangle landscape factors (land cover and the atmospheric UHI) from local factors (habitat and near-ground temperature) to explain diversity patterns among spider communities. The main novelty of our study consists in analysing the role of temperature at the landscape (atmospheric UHI) and local (near-ground temperature) scales, thanks to two independent networks of temperature sensors. We expected taxa to display different responses to temperature at the landscape and local scales. Indeed, in a recent study examining the decoupled effect of urban warming on fly communities, the atmospheric UHI overweighted land-cover factors as a driver of the community composition (McGlynn et al. 2019). In addition, urban near-ground temperature was found closely related to the abundance of Coleoptera and Hemiptera (Piano et al. 2020a, b). We therefore expected near-ground temperature to modulate the number of trapped spider individuals by affecting habitat thermal suitability, as proposed by Shochat et al. (2004), whereas the atmospheric UHI should be more related to taxonomic and functional diversity patterns (McGlynn et al. 2019).

Materials and methods

Study area and sampling design

We conducted our study in Rennes, a city of 223,000 inhabitants in north-western France. City planning is characteristic of European medium-sized cities, with a historical city centre composed of densely built residential and commercial neighbourhoods (> 40% built-up area) surrounded by suburban areas characterised by a lower building density (10–30%) and more vegetation cover (15–50%). The city is surrounded by a peri-urban conurbation where land cover is mainly composed of fields and individual houses. Rennes is located 70 km from the sea and is under the influence of temperate oceanic climate. In 2022, the mean annual temperature was 13.7 °C, the mean winter temperature 7.7 °C and the mean summer temperature 19.6 °C (Saint-Jacques airport station). Despite its mild climate, the city faces regular UHI events. Between 2004 and 2019, strong UHI situations (above 4 °C) represented more than 17% of the total amount of nights, and this proportion reached 30% in specific months (e.g., September) (Dubreuil et al. 2020).

To identify suitable sampling areas where land cover and the atmospheric UHI were not correlated, we first performed a spatial correlation analysis, as described in Cabon et al. (2024). This sampling site selection procedure is an important prerequisite because it determines the opportunity to discriminate individual effects of predictors related to landscape habitat and atmospheric UHI on spider communities.

To characterize the atmospheric UHI, we used the long-term Rennes Urban Network (https://run.letg.cnrs.fr) established in 2004. Since 2020, 30 DAVIS Vantage-Pro-2 automatic weather stations and 93 connected temperature sensors (Rising-HF- RHF1S001) have been deployed across the city centre and the peri-urban and rural areas of Rennes, and collect temperature data at a three meters height, at 60- and 15-min frequencies, respectively. Using this raw temperature data, we measured the daily UHI intensity by calculating the difference between the minimum daily temperature data at a reference cold station located outside the urban area located outside the urban area, about 11 km in the north of Rennes (Melesse, 48°12′18.1″N 1°40′52.8″W) and the minimum daily temperature data recorded by all other sensors. In a second step, we interpolated these discrete measures of daily UHI intensity by inverse distance weighting to obtain maps of estimated values covering the entire city extent at 100-m resolution (Dubreuil et al. 2020). Since the UHI intensity varies over time, we averaged the daily intensity values to the whole study period (from 1st March to 30th September 2022) to obtain a single raster map of mean UHI intensity (Fig. 1).

Fig. 1
figure 1

Map of the study area. The mean atmospheric UHI (from 1st March to 30th September 2022) is illustrated by a coloured gradient ranging from blue (low intensity; minimum = 0 °C) to red (high intensity; maximum = 3 °C). Grey areas display impervious surface

Full size image

We selected 36 sampling sites (Fig. 1) located on the identified suitable sampling areas, all located on grasslands. The sampling sites were located in public parks, community gardens, green spaces adjacent to educational buildings, or private gardens.

Community data collection

We sampled arthropods from March to September 2022, using three pitfall traps per sampling site placed five meters from each other to form a triangle and favour trapping efficiency (Ward et al. 2001). The three traps standardly headed north, south-west and south-east. We used plastic cups of 85 mm diameter and 115 mm height buried to the ground level and filled with 150 mL of saline solution at 100 g/L to prevent decomposition, and a drop of neutral soap to avoid floating. The traps were opened during the first two weeks of each seven months of sampling and emptied directly after each trapping session, resulting in 756 sampling units (i.e., three traps on each of the 36 sites and seven sessions). We sorted all spider individuals per site and sampling session and kept them in separate vials containing 70% ethanol. Mature spider individuals were identified to the species level using Roberts’ (1985, 1987) identification keys (nomenclature following the World Spider Catalog version 22; https://wsc.nmbe.ch/) and the number of trapped individuals per species, site and session was assessed. Three functional traits were retrieved from literature data: body size (mm), dispersal (ballooning capacity) and hunting guild (Appendix 1). These traits are regularly considered to cover a broad range of functional attributes and to respond to environmental changes along urbanisation gradients (Buchholz et al. 2018). More precisely, body size affects species physiology and fitness, and determines the ecological niches and spatial distribution of species (Magura and Lövei 2021; Merckx et al. 2018). Considering the sexual dimorphism in most spiders, we considered the body size of females as a proxy for a given species (Merckx et al. 2018). Hunting guild informs on the foraging behaviour, and provides relevant information on the species’ rank in the food web (Cardoso et al. 2011). The dispersal capacity informs on the species’ ability to escape their home range so as to avoid competition or adverse environmental conditions and favour gene flow across populations (Bonte and Dahirel 2017).

Landscape predictors

We calculated habitat and temperature predictors at the landscape scale based on land-cover and atmospheric UHI data, respectively. We used circular buffers of 100-m radius around the sampling sites to calculate the landscape predictors because this extent is (1) an appropriate scale to investigate landscape-scale drivers of arthropod communities (Chatelain et al. 2023; Lafage et al. 2015; McCary et al. 2018), (2) the finest spatial resolution of atmospheric UHI data available in our study area, and (3) the scale at which the correlation between urban land-cover (i.e., the built-up area) and the atmospheric UHI in Rennes remains low (Foissard et al. 2019). As a complement, we calculated habitat and temperature predictors within 200-m and 500-m radius buffers, to compare landscape predictors across multiple scales.

We derived three landscape-habitat predictors based on land-cover data: (1) the proportion of impervious surface (buildings, roads and parking sites), (2) the proportion of grassland (lawns and meadows), and (3) patch isolation, known to be a relevant landscape-scale predictor of spider and carabid communities in urban areas (Buchholz et al. 2018) (Fig. 2). We estimated patch isolation as the mean nearest-neighbour distance of each grassland patch from edge to edge using the ‘lsm_l_enn_cv’ function of the ‘landscapemetrics’ R-package (version 1.5.5 Hesselbarth et al. 2019) in R software. Land-cover data was retrieved at five-meter resolution from the French portal for geographic data (dataset OCSGE 2017; geo.data.gouv.fr; accessed on 1 December 2021).

Fig. 2
figure 2

Distribution of landscape and local predictors among four groups: landscape habitat, local habitat, atmospheric UHI, and near-ground temperature

Full size image

We derived one temperature predictor at the landscape scale – the mean atmospheric UHI. To assign UHI intensities to our 36 sampling sites, we extracted the mean pixel (100 × 100 m) value within circular buffers (i.e., 100, 200 and 500-m radius) around the sampling sites from the UHI-raster map, using the ‘exact_extract’ function of the ‘exactextractr’ R-package (version 0.9.1; Baston 2023) in R software.

Local-scale predictors

For local-scale habitat predictors, we measured the percent cover of plant species and the height of the vegetation within the triangle delineated by the three pitfall traps during each sampling session (Fig. 2). We averaged the values to obtain one final value per site and per predictor. These two predictors are regularly used to characterise arthropod habitats (Delgado de la Flor et al. 2017). Since management intensity is known to have a negative effect on the taxonomic diversity of spiders (Buchholz et al. 2018), we also calculated an index of management pressure based on the mowing regime, obtained by summing the total number of cuts observed in the field during the sampling season (ranging from zero to six).

Near-ground temperatures were recorded locally every 15 min during each trapping session with a specific network of temperature loggers (Lascar EL-USB-2+ ; Tinytag Talk 2 TK-4023) placed five centimeters above the ground surface, in the direct vicinity of the north-headed pitfall trap, at each sampling site. In the same way as for the atmospheric UHI, we calculated the differences between the daily minimum temperatures obtained from the coldest station of the near-ground sensor network, used as a reference, and that of each of the other sites. We obtained a single mean value of near-ground temperature by averaging daily differences throughout the study period.

Statistical analyses

Community indices

Pitfall traps overestimate active species. Therefore, we considered the sum of individual counts at each site as a measure of species’ activity density (Topping and Sunderland 1992), but refer to ‘number of trapped individuals’ for the sake of clarity. Taxonomic and functional diversity were quantified using the Hill numbers of orders Q = 0; 1; 2 with the ‘iNEXT.3D’ R-package (version 1.0.1; Chao et al. 2021). This method allows to account for the possible influence of dominance patterns, with indices of order Q0 being the most sensible to rare species (i.e., the actual species richness) and Q2 the less sensible (Hill 1973). For functional indices, a species’ functional distance matrix was calculated on multiple traits using the Gower method (Gower 1971) with the ‘daisy’ function of the ‘cluster’ R-package (version 2.1.4; Maechler et al. 2023). Dispersal and hunting guild were treated as nominal whereas body size was a continuous numerical variable. Based on these functional distances among species, the computed functional diversity indices quantify the effective number of equally distinct virtual functional groups, which can thus be considered as ‘functional species’ (Chao et al. 2021). Prior to analysis, we checked the absence of any significant spatial autocorrelation for all indices using Moran’s test in the ‘spdep’ R-package (version 1.2.8; Bivand et al. 2023), considering the nearest-neighbour distance (N = 3) as the metric and performing 599 permutations. To ensure that diversity indices could reliably be compared between sites, we also checked sample coverages and found more than 90% completeness at all sites. Therefore, we considered the observed in the subsequent analysis and have not rarefied or interpolated individuals counts.

Spider response to landscape and local predictors

To disentangle which of the habitat or temperature and spatial scale (i.e., landscape- or local-scale) predictors mostly explained spider community indicators, we computed generalized linear mixed models (GLMMs) using the ‘lme4’ R-package (Bates et al. 2015). We checked correlations among predictors prior to the analysis, and found only low correlations (r < 0.7; Appendix 2). We built distinct models, structured with a single index of number of trapped individuals, species diversity or functional diversity as the response variable. The taxonomic and functional indices calculated at Q = 0, 1 and 2 orders were tested individually, resulting in a total of seven distinct models. We fitted each model with all predictors listed in the Fig. 2 (i.e., Impervious surface, Open vegetation, Patch isolation, Atmospheric UHI, Vegetation cover, Vegetation height, Mowing regime and Near-ground temperature) and selected the best model among all possible candidate models using the ‘dredge’ function in ‘MuMin’ R-package (Bartoń 2016). Since habitat size, has previously been stressed to modulate the establishment of habitat-specific assemblages in the urban matrix (Beninde et al. 2015), we accounted for its potential effect in each GLMM. Since this variable is not directly related to our “habitat vs. temperature” framework, we added patch area as a random categorical factor with levels ‘small’, ‘medium’ and ‘large’ defined so that the three groups are composed of a balanced number of sites. Secondly, we estimated the relative contribution of each individual predictor to the explained variance of models, using the ‘glmm.hp’ R-package (Lai et al. 2022). All statistical analyses were performed in R version 4.0.3.

Results

The total area suitable for sampling after spatial correlation analysis covered 6000 ha, i.e., 58% of the entire extent of the study area. Moreover, 1685 ha of urban grasslands were located in suitable areas, i.e., 82% of the total urban grasslands of the study area. The 36 selected grasslands ranged from 0.1 to 4.7 ha in size (mean = 0.8 ± 1.0 ha) and the distance between them ranged from 181 m to 11.36 km (mean = 4458 ± 2173 m). Grasslands were characterised by diverse vegetation structures, with an average vegetation height ranging from 5 to 44 cm (mean = 20 ± 11.5 cm) and a vegetation cover ranging from 60 to 100% (mean = 90.5 ± 9.4%). The mowing frequency ranged from zero to six times during the sampling period.

Sampling results

In total, data from four sampling units out of the 756 (i.e., 0.5%) was not available due to damaged traps and was estimated by the averaged data from the two other traps forming the triplet. The data set includes 20,761 mature spider individuals assigned to 21 families and 137 species. Among them, the most represented families in terms of number of trapped individuals were Linyphiidae (N = 9045; 44%), Lycosidae (N = 5462; 26%) and Tetragnathidae (N = 4937; 24%). The species with the highest number of trapped individuals were Pachygnatha degeeri (N = 4913; 24%) and Pardosa cf. tenuipes (N = 2603; 13%). We did not detect any spatial autocorrelation related to spiders’ number of trapped individuals or to spiders’ taxonomic and functional diversity.

Effects of habitat and temperature on spider communities

Our models showed that the number of trapped individuals was significantly negatively related to near-ground temperature (P < 0.001, Table 1, Fig. 3a). Species diversity at Q0 order was negatively related to the atmospheric UHI (P < 0.001, Table 1, Fig. 3b) and to the mowing frequency (P < 0.001, Table 1, Fig. 3c). Functional diversity at Q0 order was significantly positively related to local vegetation height (P < 0.001, Table 1, Fig. 3d) and negatively related to local vegetation cover (P < 0.001, Table 1, Fig. 3e).

Table 1 Relationships estimated by GLMMs (estimate ± standard error) between the number of trapped individuals, species diversity orders or functional diversity orders of spiders, and predictors related to near-ground temperature, landscape habitat measured within 100-m-radius buffers, landscape temperature (atmospheric temperature), local habitat or local temperature (near-ground temperature)
Full size table
Fig. 3
figure 3

Plots showing significant relationships between the number of trapped individuals, species diversity or functional diversity of spiders, and predictors related to near-ground temperature, atmospheric UHI and local habitat (i.e., mowing regime, vegetation height, vegetation cover). Confidence intervals at 95% are shown

Full size image

The hierarchical partitioning analysis showed that for the number of trapped individuals, near-ground temperature alone significantly explained a quarter (24%) of total variation (Fig. 4a). Regarding species diversity at a Q0 order, atmospheric UHI and the local habitat alone explained 20% and 10% of the variance, respectively and these values decreased with increasing orders (Fig. 4b). For functional diversity at a Q0 order, vegetation height and vegetation cover explained 22% and 25% of the variance, respectively and these values decreased with increasing orders (Fig. 4c). The same analysis performed with landscape habitat and atmospheric UHI predictors calculated at larger scales (i.e., 200-m and 500-m radius buffers) showed similar results than those obtained at the 100-m scale (Appendices 3 and 4). Therefore, we only consider results related to landscape predictors calculated at a 100-m scale in the rest of the paper.

Fig. 4
figure 4

Relative variance explained independently by landscape- and local-scale predictors according to GLMM models for spiders’ a number of trapped individuals, b species diversity (orders Q = 0, 1 and 2) and c functional diversity indices (orders Q = 0, 1 and 2)

Full size image

Discussion

Understanding the factors that shape biodiversity patterns in environments facing global changes is of prime importance for predicting and mitigating the adverse effects of global warming. Cities and their related UHIs provide an excellent framework to conduct climate-related ecological studies. However, atmospheric UHIs and land-cover artificialisation tend to co-vary, so that disentangling the predictors from each other is a challenge. The present study uses a reproductible approach that partially de-correlates urbanisation from the UHI (Cabon et al. 2024). Near-ground temperature and the atmospheric UHI outperformed landscape and local habitat predictors in explaining variations in arthropod communities. In addition, we went beyond this result by comparing for the first time the relative weights of two temperature predictors (near-ground temperature and the atmospheric UHI) that vary differently across the urbanisation gradient at two independent spatial scales (100 m vs. metric).

Based on the ecological characteristics of ground-dwelling spiders, our results show that 100 m is a valuable landscape scale to decouple the effect of land cover from that of the atmospheric UHI. This result is partially supported by recent ecological studies. McGlynn et al. (2019) found proportions of impervious surface (within 50-m buffers) decoupled from temperature across the Los Angeles Basin. Similar results were obtained by Hamblin et al. (2018) with 200-m buffers. However, these two studies only considered near-ground temperature when measuring correlations with land cover. Our study finally provides first evidence that atmospheric UHI patterns can also be partially decoupled from land cover data.

Effects of temperature at the landscape and local scales

An increasing number of studies at the community level not only focus on habitat-related predictors to identify urban drivers, but also consider temperature (Hamblin et al. 2018; McGlynn et al. 2019; Meineke et al. 2017; Piano et al. 2020a, b). Yet, unlike land cover predictors that are almost systematically assessed at the landscape and local scales, temperature remains mainly considered at single spatial scales. However, near-ground temperature deviates from the atmospheric UHI (Oke et al. 2017), as confirmed by the low correlation observed in our study (r = 0.46). The main difficulty in studying temperature at different scales lies in the complexity of setting up devices at the scale of the entire city and at each sampling site. Two sensor networks need to be deployed – one spread evenly across the city to capture the atmospheric UHI, the other consisting of one (or more) logger per sampling site to capture near-ground temperature. Last but not least, all this represents a large number of devices and measurements associated with a cumbersome data analysis process. We approximated near-ground temperature with a single sensor per site that provided a significant predictor. However, an interesting approach to go further would be to upgrade the local sensor network by multiplying the number of measurement points at each sampling site to obtain a measure of local temperature heterogeneity.

To the best of our knowledge, this study is the first that attempts to specify the effects of temperature simultaneously measured at the landscape and local scales. At the landscape scale, our results are in accordance with our first expectation: variations in spider taxonomic diversity were associated to the atmospheric UHI, explaining 20% of the variance alone. The observed shift in taxonomic diversity reflects changes in community compositions suggesting the existence of a filtering process in response to the atmospheric UHI. Species may be filtered by large areas of heat acting as thermal barriers at the city’s edge, where only species physiologically able to cope with urban warming persist (e.g., urban-adapted heat-tolerant species; see Hamblin et al. 2017; Piano et al. 2017), whereas others are extirpated. The analysis confirmed that atmospheric UHI is an important driver constraining the species diversity of spiders. Different mechanisms can lead to the observed decrease in species diversity with increased temperature. Heat-sensitive native species could be replaced by a lower number of non-native heat-tolerant ones, suggesting a decrease in species diversity along with a high turnover rate (McKinney 2008). Alternatively, the reduced number of species may result from the persistence of only a subset of the species pool able to cope with warming conditions (Piano et al. 2017). For example, ant communities are better represented by heat- and dry-tolerant species in urbanised habitats than in natural adjacent areas (Menke et al. 2011). Along the same urbanisation gradient as in the present study, Cabon et al. (2024) examined changes in spider community functional composition and found a community-wide shift toward thermophilic species, explained by the exclusion of heat-sensitive species in urban areas. Here however, the absence of relationship between functional indices and the atmospheric UHI could be explained by the lack of thermal attributes in the set of traits selected to compute the functional diversity indices.

At the local scale, our results also confirm our theoretical framework. Near-ground temperature had no effect on taxonomic or functional diversity, but was negatively associated with the number of trapped individuals (24%). This result is in line with several recent studies that identified urban near-ground temperature as a prevailing driver of arthropod abundance (Hamblin et al. 2018; McGlynn et al. 2019; Meineke et al. 2013). Yet, unlike several studies reporting an increased abundance of phytophagous arthropods with temperature increase (Dale and Frank 2014, 2018; Meineke et al. 2013), our results show a negative relationship between the number of trapped individuals and near-ground temperature. Although phytophagous arthropods and predator spiders display opposite patterns to urban thermal conditions, our result is supported by recent studies on flying arthropods (Hamblin et al. 2018; McGlynn et al. 2019), showing that temperature is associated with decreased abundance (but see Geppert et al. 2023). In this context, warming is hypothesized to affect arthropods through direct effects (e.g., habitat temperature limits survival or fitness through physiological processes), or indirect effects (e.g., via limited access to feeding resources) (Hamblin et al. 2018).

Since near-ground temperature was not associated with changes in taxonomic or functional diversities, it might not be directly involved in a filtering process, sorting species according to their ability to persist in urban environments. Instead, near-ground temperature may determine the abundance patterns of the locally present species. Near-ground temperatures are heterogeneous, and this variability matches the local perception range of most arthropods, i.e., about 20 cm (Pincebourde and Salle 2020). For example, the measured mean atmospheric UHI spanned 3 °C over a distance of 5 km, whereas our temperature loggers recorded near-ground temperatures spanning up to 5 °C. Since these temperature-related predictors are based on daily minimum temperature records, they mostly depict nocturnal thermal situations. However, a daytime experiment conducted by Pincebourde and Salle (2020) showed that temperatures can range up to 16 °C within less than 1 m distance on leaf litter. Therefore, mobile ground-dwelling arthropods might have plenty of opportunities to behaviourally thermoregulate over a local spatial scale in grassland vegetation during daytime. Many ground-dwelling spiders are known to behaviourally thermoregulate to avoid deleterious thermal conditions (Cabon et al. 2023; Humphreys 1974). Therefore, species-specific thermoregulatory behaviour can lead to community-wide avoidance of hot deleterious conditions. This idea is supported by previous results obtained by Shochat et al. (2004), who hypothesized that local deficits of spider abundance resulted from behavioural reactions (i.e., local emigration) in response to warmer microenvironments.

Effect of habitat

As already found in previous ecological studies on urban environments (Alaruikka et al. 2002; Buchholz et al. 2018; Magura et al. 2010), our results show that landscape habitat predictors did not affect spider taxonomic and functional diversities. This suggests that habitats located in highly urbanised areas can host diversified communities. Yet, it is worth pointing out that the scale selected here to describe the landscape (100-m radius buffers) corresponds to the optimal scale allowing a maximized local decorrelation of UHI and land cover, and can be interpreted as matching the home range of mobile ground-dwelling spiders. Although this scale is the most suitable to answer our research questions, relationships between spider community indices and landscape habitat predictors may also be observed at larger scales, yet without opportunity to dissociate the effects of temperature from those of habitat.

Local habitat predictors outperformed landscape habitat predictors in explaining spider taxonomic diversity. If we refer to the usual dichotomous framework opposing the local and landscape scales (Beninde et al. 2015; Buchholz et al. 2020; Egerer et al. 2017), the urban landscape should filter species according to their capacity to disperse into urbanised areas. Therefore, landscape descriptors are mainly referred to as drivers of the taxonomic and functional diversities of urban communities. By contrast, local habitats should determine the availability of resources needed for survival and reproduction, and control the establishment of viable populations. Our results confirm the importance of local habitat for spiders, but suggest that this predictor rather drive taxonomic diversity. Other studies have reported similar results, e.g., in semi-natural habitats (Lafage et al. 2015; Sarthou et al. 2014), where local habitat features have been found to be more significant drivers of spider species diversity than landscape predictors. Similarly, spider species diversity was mainly related to local habitat characteristics such as the bare-ground cover or the vegetation structure, in both non-urban (Pétillon et al. 2008) and urban habitats (Argañaraz et al. 2018).

We observed a positive effect of vegetation height on spider functional diversity. This result was quite expected because the occurrence of a species within a specific microhabitat is intimately linked to the hunting method practiced by the foraging guild it belongs to (Barriga et al. 2010; Diehl et al. 2013; Gibson et al. 1992; Rao 2017). As a result, different hunting guilds need different microhabitat structures (Bell et al. 2001; Morris 2000; Uetz 1979, 1991). The vertical structure of herbaceous vegetation can vary drastically among grasslands, depending in particular on the management regime (Bell et al. 2001; Chollet et al. 2018). A high herbaceous vegetation provides more opportunities than a low vegetation for the establishment of weaver species, because elements for fixing webs (i.e., anchorage points) are more available (Diehl et al. 2013; Langellotto and Denno 2004; Mcnett and Rypstra 2000; Uetz 1991). Nevertheless, the presence of high herbaceous vegetation does not rule out the maintenance of a surface structure suitable for the establishment of wandering species (Maelfait and De Keer 1990; Uetz et al. 1999). In addition, high vegetation should provide increased potential for niche differentiation through the diversification of phytophagous prey (Morris 2000; Woodcock et al. 2009) and indirectly favour the co-occurrence of diverse hunting guilds (Lyons et al. 2018), which could explain the observed enhanced functional diversity.

The local proportion of vegetation cover was significantly negatively related to spider functional diversity. In other words, the sites with a heterogeneous vegetation cover supported more functionally diverse spider communities. Similar outcomes have been found in extensively managed habitats, which hosted functionally diverse communities composed of more web-building and ambush-hunting species than intensively managed ones in urban (Buchholz et al. 2018) and agricultural landscapes (Diehl et al. 2013). As for vertical structures, horizontal heterogeneity (e.g., vegetation cover, bare soil or litter) tends to support functionally more diverse communities compared to homogeneous environments, in particular when habitat conditions provide opportunities for multiple foraging guilds to co-occur (Uetz et al. 1999).

Although spider communities can be structured according to inter-annual climatic fluctuations, in terms of activity (Bowden et al. 2018; Muster et al. 2020) and composition (Relys et al. 2002), the results discussed here were obtained after a single year of sampling. Longer-term studies are therefore needed to determine whether the effects we observed fluctuate over time or not. However, in view of the particularly high temperatures for the region measured in 2022 and the particularly intense UHI conditions for Rennes (HCBC 2023), we believe that these results should be representative of the future climate, during which this type of heatwave is likely to be repeated and intensified (HCBC 2023).

Implications for urban planning

The factors promoting urban biodiversity and the efficient scales of management action need to be clearly identified in order to guide urban planning. Ecological studies of urban green spaces are needed to identify the landscape- and local-scale factors that drive floristic and faunistic diversity and to guide urban planning in the development of efficient conservation or restoration measures (Lepczyk et al. 2017). In light of conservation measures, our results show that arthropod communities can be enhanced by developing strategies at both local and landscape scales. At the landscape scale, the atmospheric UHI was a prevailing filter of spider assemblages that sorted species according to their ability to adapt and persist in warm urban habitats (Cabon et al. 2024; Merckx and Van Dyck 2019). Therefore, increasing biodiversity requires that additional species can pass through the urban thermal filter. Implementing green infrastructures is a widely used strategy that efficiently generates cool islands (Manso et al. 2021; Norton et al. 2014). However, present knowledge highlights the importance of sustaining connectivity between urban vegetated areas such as grasslands, and thermal connectivity within cities to facilitate immigration and the establishment and maintenance of species coming from rural surroundings. By highlighting the climate-biodiversity co-benefits of green spaces, our results can be used to design planning solutions that meet the ambitions set out in the roadmaps developed by local authorities (Rennes Métropole 2019). Our results also show that local characteristics (temperature and habitat) are particularly promising levers to enhance arthropod conservation in urban grasslands because they can promote diversity locally (e.g., through the diversification of vegetation structures). In particular, the present study shows that decreasing near-ground temperature and favouring high herbaceous vegetation coupled with heterogeneous ground coverage maintained at a low frequency would increase the number of trapped individuals and diversity of spiders in urban grasslands. As previously demonstrated for grassland vegetation in our study area (Chollet et al. 2018), the local improvement of grassland management practices is a first step towards arthropod conservation. Furthermore, our results show that prioritising grassland management measures could not only mitigate the loss of taxonomic diversity induced by the intensification of the atmospheric UHI in cities, but could also promote the establishment of more functionally diverse species pools.