Arthropod abundance modulates bird community responses to urbanization

We analysed the role of species interactions in wildlife community responses to urbanization. Specifically, we investigated non‐trophic associations within a bird community and the role of trophic interactions in the responses of bird species to the urbanization gradient.


| INTRODUC TI ON
Urban areas are increasing and expanding throughout the world, with an expected 70% of the human population residing in cities by 2050(United Nations, 2018. Urbanization is one of the most drastic anthropogenic modifications of the natural landscape and habitats (Sala et al., 2000;Shochat et al., 2010), and the sprawl of urban areas results in biotic homogenization across large areas and biodiversity loss (Leveau et al., 2017;McKinney, 2008;Morelli et al., 2016;Sol et al., 2014). Yet, urban areas still host high numbers of wildlife species and even some endangered species (Aronson et al., 2014;Ferenc et al., 2014;Ives et al., 2016). Given the rapid increase in urbanized areas worldwide on the one side and the current biodiversity crisis on the other side, it is crucial to understand how the special conditions provided by cities as novel ecosystems (Kowarik, 2011) affect biodiversity.
However, the knowledge on how urban environmental drivers shape biodiversity is insufficient, because in addition to climate and habitat, biodiversity is affected by species interactions (HilleRisLambers et al., 2012). Studying these species interactions might be one of the pieces that help us to comprehend the puzzle of biodiversity response to urbanization. Two broad types of species interactions may be distinguished: trophic (i.e. predation) and non-trophic (e.g. competition, facilitation and mutualism) ones (Bronstein, 1994). Species interactions play a crucial role in shaping wildlife communities (Cavieres et al., 2014;Martin et al., 2018;Yodzis, 1981). For example, competition for limited resources modifies the community by favouring a higher abundance of the successful competitors, which will displace other species (Goldshtein et al., 2018;Robertson et al., 2013). Regarding trophic interactions, the ability of some species to exploit food resources under disturbance affects community composition (Galbraith et al., 2015;Planillo et al., 2015; K. E. Plummer et al., 2019). Availability of invertebrate prey, mainly arthropods, also determines the reproductive success of many bird species inhabiting anthropogenic habitats and shapes the long-term viability of the populations (Peach et al., 2008;Seress et al., 2012Seress et al., , 2018. Therefore, it is essential to consider species interactions in addition to environmental drivers, to gain a complete understanding of urban drivers of biodiversity. Importantly, species interactions may be modified by urban environmental drivers, as demonstrated by recent studies (Chamberlain et al., 2014;Gelmi-Candusso & Hämäläinen, 2019;Harrison & Winfree, 2015). Environmental factors were previously shown to interact in their effects on biodiversity and ecosystem functioning, for example modifying predator-prey relationships or detritivore species dynamics (Crain et al., 2008;Galic et al., 2017;Garnier et al., 2017;Karakoç et al., 2018). It is therefore likely that urban environmental drivers will be moderated by species interactions in their effects on wildlife community composition in cities. For example, below a certain threshold of prey availability, the diversity of predators is expected to be zero irrespective of suitability and connectivity of the urban landscape. In this context, investigating how urban community composition is shaped by either environmental drivers or species interactions separately will result in an incomplete, if not biased, understanding. Therefore, we studied how both environmental factors and species interactions shape bird community composition in the city-state of Berlin, a metropolitan area in Central Europe.
To which extent species can respond to urban environmental drivers partly depends on their traits. Among them, body mass, migratory status and diet are important traits for bird species occurrence in cities (Evans et al., 2011;Jokimäki et al., 2016). Body mass is positively correlated with urbanization, especially within clades (Callaghan et al., 2019;Croci et al., 2008). Migrant birds are expected to be scarcer in urban areas (Croci et al., 2008;Evans et al., 2011;Jokimäki et al., 2016;Kark et al., 2007), as their shift in phenology to earlier dates is much slower than that in resident species (Chamberlain et al., 2009;Samplonius et al., 2018), and thus, suitable nesting sites might be already occupied by resident species when migrants arrive to the urban area (Jokimäki & Suhonen, 1998).
Regarding diet preferences, bird species with narrow diets, for example insectivorous, are expected to decrease and omnivorous species with wider diets are expected to increase with urbanization, as they can take advantage of resources provided by human activities, such as garbage (Callaghan et al., 2019;Croci et al., 2008).
To better understand the complex interactions of urbanization, bird communities and their food resources, here we investigated how bird community composition in the urban area is affected by three aspects: (a) environmental factors reflecting the urbanization gradient, of which we distinguish nature-like variables and anthropogenic disturbance, (b) trophic species interactions as measured K E Y W O R D S arthropod abundance, community composition, joint species distribution models, species interactions, urban ecology, urbanization gradient, wildlife diversity by arthropod abundance and (c) species associations (proxy for non-trophic interactions) within the community. We focus on the bird community and their invertebrate prey (arthropods), both of which were extensively surveyed in citywide regular and standardized monitoring schemes of Berlin (Abraham et al., 2019;Möller et al., 2019;Südbeck et al., 2005). To achieve our goal, we use the modern technique of hierarchical multiresponse models, joint species distribution models (JSDMs), to analyse species responses to environmental conditions while simultaneously accounting for associations between species within the community Pollock et al., 2014;Warton et al., 2015). That is, after removing the effects of environmental covariates, the remaining correlation in the residual variance that is no longer explained by the covariates is termed "species association" (Dormann et al., 2018;Ovaskainen et al., 2017;Warton et al., 2015).
Our main objective is to disentangle the relative importance of environmental conditions and species interactions for bird community composition in urban areas. We predict that: (a) global species abundance will be higher in less urbanized sites with high arthropod abundance; (b) arthropod abundance, as an essential resource, will explain more variability in bird species abundance than the urbanization gradient by itself; (c) species inhabiting sites with high anthropogenic disturbance will use more anthropogenic resources and show lower or no response to arthropod abundance; and (d) resident species with broader diets and larger body mass will displace other species and be more abundant in more urbanized areas.

| Study area
We studied the bird community in the city of Berlin, capital of Germany (52°31' N, 13°24' E). Berlin is the largest city in Germany and constitutes a federal state, with a population of nearly 3.65 million people and an area of 892 km 2 (Amt für Statistik, 2017).
The city is subject to a moderate continental climate, with average precipitation of 568mm and mean annual temperatures from 10.5°C in the city centre to around 8°C in the surrounding forests

| Bird community data
We used data on bird species abundance from the Berlin breeding bird monitoring survey in the year 2017, provided by the Berlin Senate Department for Environment, Transport and Climate Protection (SenUVK). The survey consists of 30 pre-established transects, located within one-km 2 grids distributed across the city ( Figure 1). Grids were previously selected to capture the widest possible range of habitats. To ensure species detection, each transect was visited four times during the bird breeding season from mid-March to mid-June, approximately once per month whenever possible and leaving at least a week between visits (Südbeck et al., 2005). Transects were walked at or shortly after sunrise for 2-4 hr, in days without precipitation or strong winds. All birds seen and heard within the transects were recorded. For a detailed description and information on standards for breeding bird surveys, see Südbeck et al. (2005) and the website of the Swiss Ornithological Institute (https://www.vogel warte.ch/de/proje kte/ monit oring/). One transect was visited only 3 times and therefore was removed from further analyses.
In the 29 monitoring transects, a total of 97 breeding bird species were recorded. Aquatic species, two non-native species and species that were present in only 2 or less sites were removed to avoid confounding effects and ensure convergence of the statistical model, resulting in 66 species in total (Appendix S1: Table S1.1). As transects differed in length (range: 2.8-6.4 km), we divided each bird species abundance by transect length to obtain an index of relative bird abundance per km (hereafter termed "bird index") that was used as response variable in the analyses.
We selected three traits that are related to wildlife responses to urbanization: body mass, diet and migratory status (Callaghan et al., 2019;Croci et al., 2008;Evans et al., 2011). Body mass values and diet information were obtained from the ELTON database (Wilman et al., 2014). Body mass values were log-transformed to favour linearity. Diet included four categories: invertebrates (species feeding mainly on invertebrates or insectivorous), herbivores (species feeding mainly on plants and seeds), omnivores (species with a mixed diet) and scavengers (species feeding on carcasses and leftovers). Migratory status was assigned following the Checklist of the Birds of Germany (Barthel & Helbig, 2006) (Appendix S1: Table   S1.1). We used three categories: "migrant" for species that breed in Germany and winter in other countries, "resident" for species that spend the whole year in Germany and "partial-migrant" for species with mixed behaviour, in which some of the individuals are migrant and others are resident.
Carabids and spiders were sampled by pitfall traps (Brown & Matthews, 2016;Hill et al., 2005). We used a nested design (Boetzl et al., 2018), with four pitfall traps in each sampling site in a square with a distance of 6 m to each other. Pitfall traps consisted of plastic cups (10 cm diameter and 16.0 cm depth) filled with a 1% formalin-detergent solution. A grid was placed on each pitfall trap to keep organic material outside of the traps. Sampling took place in 2017 from May to July and from September to October, and pitfall traps were emptied every four weeks. Both taxa were sorted and afterwards preserved in ethanol. Standard keys were used for identification of adult carabid beetles (Müller-Motzfeld, 2006) and spiders (Almquist, 2005(Almquist, , 2006Roberts, 1987Roberts, , 1998. Grasshopper data were collected using a box quadrat (ground area of 2 m 2 and gauze-covered sides of 0.8 m height) (Gardiner et al., 2005), haphazardly set up 10 times per site in August 2017. All individuals found inside the box quadrat were identified to be species using keys provided by Bellman (2006), counted and afterwards released.
The field sampling identified a total of 104 carabid species, 20 grasshopper species and 182 spider species. Prior to the statistical analyses, rare species, defined as those present in less than 5% of the sites, were removed from the dataset to avoid convergence problems. The final datasets contained 73 carabid species, 18 grasshopper species and 112 spider species (Appendix S1: Table S1.2). We used the number of individuals of each species as a relative abundance index.

| Environmental data
We selected variables relevant to the distribution of wildlife species in urban areas, as identified by other studies (Batáry et al., 2018;Beninde et al., 2015;Ferenc et al., 2014). We divided our environmental variables into two groups: variables related to natural conditions (e.g. tree cover and open green area) and variables related to anthropogenic disturbance in the urban areas (e.g. noise level) ( Table 1). All environmental layers were rasterized at 20 x 20 m resolution, but their importance was tested at a 100 m, 500 m, 1 km and 5 km scale using a moving window approach (further information in Appendix S2: Figures S2.1 and S2.2).

| Data analysis
We modelled urban bird community response to environmental drivers and arthropod abundance, the species associations within  Tikhonov et al., 2019). Community composition was studied by using the relative abundance of each species as response variables for all the models (for modelling details, see below).
Our modelling workflow consisted of the following steps ( Figure 2): as JSDM fitting is computationally demanding, precluding fitting all possible models, we first selected the optimal spatial resolution and the relevant environmental variables for the JSDMs.
Variables and their spatial resolution were selected using Akaike's information criterion for small sample sizes (AICc) in multiresponse models, after testing for multicollinearity (see Appendix S3: Figure   S3.3, Tables S3.3 and S3.4). Second, for each arthropod group (carabids, spiders and grasshoppers), we ran a JSDM with the respective arthropod abundances as model responses and the environmental variables selected in the previous step as predictors. These models were used to estimate arthropod abundance in the bird transects, because invertebrate sampling plots and bird transects were not spatially overlapping. We predicted arthropod abundance in the area of the transects, which was then used as a covariate in the bird community model (see below).
Last, we ran the hierarchical JSDM for the bird community with

| Modelling and extrapolating arthropod communities
For each arthropod group, we run a hierarchical JSDM in a Bayesian framework, using as response variables the abundance of each species within the respective arthropod group (carabids, spiders and grasshoppers). We included the previously selected environmental variables as explanatory variables (see above), a spatially explicit random effect with the location of each sampling site to control for spatial effects, and used Poisson error distribution with a log-link function.
TA B L E 1 Environmental variables considered for the hierarchical joint species distribution models. Variables are categorized into two groups ("anthropogenic disturbance" and "nature-like") that will be used for variable selection (see Methods) value gives information about how accurate the model prediction will be in each site for the full arthropod community, that is, how accurate the prediction will be in the bird transects. All the explanatory power values were relatively high (Species R 2 carabids = 0.71; R 2 grasshoppers = 0.6; R 2 spiders = 0.72; Site R 2 carabids = 0.94; R 2 grasshoppers = 0.92; R 2 spiders = 0.95); thus, we confidently used these JSDM for extrapolating arthropod abundance to the bird monitoring transects creating raster maps with a resolution of 100 m, which was the best scale for predicting arthropod abundances. Then, we averaged the predicted values per arthropod group of all the raster cells overlapping the bird transects. Finally, total arthropod abundance was calculated as the sum of the predicted abundances of the three arthropod groups at the bird transects.

| Modelling the bird community
We modelled bird community using a hierarchical JSDM in a Bayesian framework  that models each of the observed species accounting for the potential associations among them and provides results for individual species and the global community, as the sum of the species responses. We used two different types of response variables: each bird species relative abundance index (bird index) and species traits. The explanatory variables were the selected environmental variables (Table S3.4) and total predicted arthropod abundance in each transect. We did not use the different arthropod groups as separate covariates because their relative abundances were highly correlated (spiders-grasshopper: Pearson's r = .927; spiders-carabids: r = .917; and grasshoppers-carabids: r = .924). Finally, the spatial locations of the bird monitoring transects were included as spatially explicit random effect to account for a potential spatial structure in the data (Appendix S4: Table S4.5).
We ran three chains of 780,000 iterations with a burn-in of the first 30,000 iterations and a thinning rate of 50 to reach model convergence, retaining 45,000 (3 × 15,000) samples for further analyses. Model convergence was assessed using chain trace plots and the Gelman-Rubin statistic (Gelman & Rubin 1992). We calculated the 95% and the 75% posterior credible intervals (CI) for all the parameters. Following other publications (Mata et al., 2014;Ribeiro et al., 2018), we considered that 95% CI not overlapping zero showed a strong effect and that 75% CI not overlapping zero were enough to F I G U R E 2 Representation of the modelling workflow followed in this study to disentangle effects of environment and trophic interactions, as well as explore species associations (lower left pictogram). The workflow has been divided into three sections: arthropod community models (JSDMs; left square), bird community model (JSDM; central-rigth square), which is used to derive the final results (right). Green boxes and light grey arrows represent input data in the models. Orange boxes and dark grey arrows represent outputs of the statistical models. Grey boxes represent relevant modelling steps

Results
show the existence of an effect. Consequently, we considered that when the 75% CI overlapped zero, the support from the data was weak to demonstrate an effect.
We examined the individual species responses to search for common trends and patterns. By carefully considering the negative, positive or lack of response of each species to each of the environmental variables, and whether this response was modulated by the arthropod abundance, we defined species groups with similar response patterns to the urbanization gradient.
We grouped the explanatory variables into four groups (Table 1) Analyses were done in R v 3.5.2 (R Core Team, 2019).

| RE SULTS
Bird relative abundance was highly impacted by arthropod abundance, which modulated the community response to environmental variables ( Figure 3). As a general response, relative bird abundance was higher in areas with higher arthropod abundance, especially in areas with high tree cover (Figure 3a). When arthropod abundance was low, the relative abundance of birds was negligibly influenced by anthropogenic disturbance variables (human population and noise levels) and tree cover. When arthropod abundance was high, relative bird abundance decreased sharply with increasing anthropogenic disturbance (Figure 3c When extrapolating the responses to the area of the whole city, spatial patterns became apparent, with species from group 2, woodland group, showing a distribution pattern very similar to that of the whole community (Figure 4; for environmental variables maps, see Appendix S2: Figures S2.1 and S2.2).
The variance partitioning analysis of the explanatory variable groups in the JSDM showed that the variable interactions with arthropod abundance explained almost 40% of the variance ( Figure 5).
Regarding the other groups, variance partitioning is not straightforward to interpret because of the different group sizes; "nature-like variables" and "disturbance variables" are each represented by two variables, while "arthropod abundance" is represented by one variable, and the (statistical) "interactions with arthropod abundance" are four variables. Arthropod abundance as single effect additionally explained on average 13% of the variation, as much as either nature-like (14%) or disturbance variables (16%). Taking that into account, we can confidently say that arthropod abundance and its statistical interactions with environmental variables were most important for bird community composition.
The analyses based on trait values showed fewer clear responses, although some potential trends are worth mentioning (Appendix S7: Figure S7.8). Regarding diet, species feeding on invertebrates showed a positive trend in relation to arthropod abundance and tree cover and negative trend in relation to noise. No trend in species body mass was apparent. Regarding migratory status, the most pronounced responses were observed for partial migrants, which were negatively affected by noise and positively affected by arthropod abundance and tree cover. No trend was identified in body mass.
The associations among bird species within the community identified in the residual variance of the JSDM were all positive ( Figure 6).
These non-trophic associations occur mainly among bird species with high prevalence in more urbanized areas (group 1, urban group).
Some positive associations were also found among species belonging to different response groups, which could point to facilitation or neutral co-occurrence between those species. No competitive interactions resulting in spatial exclusion were found, as shown by the lack of negative associations.

| D ISCUSS I ON
Our results highlight the key role of trophic species interactions in wildlife community composition under anthropogenic disturbance. Arthropod abundance was the main variable driving bird community response across the urbanization gradient, and it modulated the effect of anthropogenic disturbance on bird species abundance, hence bird community composition. This is a novel finding, as most studies still focus on environmental variables and neglect biotic interactions (Batáry et al., 2018;Beninde et al., 2015;Callaghan et al., 2018).
As a general pattern, both total bird abundance and species richness were higher in low disturbance areas and decreased towards more urbanized areas. Arthropod abundance increased bird abundance in low to moderately disturbed areas, but had little influence on species inhabiting the most urbanized areas. This variable was also the most important factor explaining bird abundance variance along the urbanization gradient ( Figure 5). Although some species have diets that rely mainly on plants or seeds, invertebrates are an important resource during breeding season for many birds (Peach et al., 2008;Seress et al., 2012). Therefore, arthropod abundance represents a key resource that allows bird diversity to increase in less urbanized areas. Above certain disturbance thresholds, our results show that only a few bird species can tolerate the anthropogenic disturbance and overall bird community abundance clearly decreases. A meta-study on urban biodiversity patterns showed no clear overall trend of bird abundance in response to urbanization as both negative and positive trends were detected in a similar amount of studies (Faeth et al., 2011).
Our results demonstrate that failing to explicitly consider prey availability may result in different relations of bird abundance with the urbanization gradient.
Additionally, our analyses of species traits showed that anthropogenic disturbance negatively affected species feeding on invertebrates and partial-migrant species, as corroborated by previous studies, although no clear response was found for fully migrant species (Croci et al., 2008;Evans et al., 2011;Kark et al., 2007).
We identified three groups of birds regarding their responses to the urbanization gradient ( Figure 6). We believe our groups 1, 2 and 3 (birds of urban, woodland and natural areas) resemble and support the classification of wildlife species into urban exploiters, urban adapters and urban avoiders, respectively (Blair, 1996;McKinney, 2002;Shochat et al., 2006). The urban species, or urban exploiters, persisted at high abundance under high disturbance levels. These species, in agreement with other studies (Callaghan et al., 2019;Croci et al., 2008;Kark et al., 2007), are small-to medium-sized, have mainly scavenging or omnivorous diets and are resident species (Appendix S1: Table S1.1). Species in our dataset that were classified as scavengers belong mainly to generalist and opportunistic species and, therefore, they are  Jokimäki et al., 2016). Interestingly, this pattern is parallel to that identified for mammals, where the most opportunistic mesocarnivores can colonize the urban areas (Bateman & Fleming, 2012;Gras et al., 2018).  Hoyo, Elliot, & Sargata, 2004). Although some of these species could perhaps increase their abundance in dis- turbed areas, open habitats in an urban context are usually exposed to direct contact with humans and pets (unpublished data), which might drive bird species away from more populated areas.
An interesting finding of our study is the inclusion of the house sparrow in the group of urban birds, which means a lack of relationship with arthropod abundance. Although its presence in the cities cannot directly be interpreted as adjustment, sparrows are declining across Europe and identifying the conditions under which the species survives is of vital importance for its conservation (BirdLife International, 2018). In our study area, some big areas with low anthropogenic disturbance still persist near the city centre, and we hypothesize that the availability of these quiet areas next to areas with high anthropogenic resources might mitigate the negative effects of areas with high disturbance, thus allowing the species to persist in the city. Another interesting species that is usually sensitive is the nightingale, which also appears at high abundances in areas with low disturbance and high arthropod abundance, highlighting again the key role of prey abundance for bird diversity.
There is some evidence for the highest species abundance at intermediate urbanization levels (Batáry et al., 2018), an effect that conforms with the intermediate disturbance hypothesis (Chace & Walsh, 2006;Connell, 1978;Hacker & Gaines, 1997;Marzluff, 2017;McKinney, 2008). In this context, the findings of how the size of green areas in cities affects bird abundance or diversity are sometimes contradictorily discussed. Larger green areas in urban context are expected to have low disturbance and higher bird diversity, but some results show small green patches with a diversity that is comparable or higher than that of the big areas (Callaghan et al., 2018;Matthies et al., 2017), which sometimes has been related to the intermediate levels of disturbance in such small areas. Our findings suggest that high biodiversity at intermediate urbanization levels or differences in bird abundance between green areas of similar size may be due to missing influential variables in the analyses (Fox, 2013). If we included arthropod abundance or a similar measure of prey availability in the analyses, for the same relative prey abundance, larger green areas should contain higher bird diversity. However, if there is a difference in prey abundance, we would expect small green areas F I G U R E 5 Results of the variance partitioning analysis. Variables were grouped in four groups: (anthropogenic) disturbance variables, nature-like variables, arthropod abundance and the interactions between arthropod abundance and the environmental variables (Arthropodenvir. interaction). The variance explained by the spatial location of the sites (random factor) is also shown. Species names are colour-coded based on the group they were classified: urban birds (grey), woodland birds (orange) and natural areas birds (green) with high prey abundance to support higher bird diversity than large green areas with low prey abundance. Indeed, failing to consider prey availability in the analyses may result in the apparent support of the intermediate disturbance hypothesis.
Regarding the non-trophic interactions, we found only positive associations among bird species. From a statistical point of view, the associations or co-occurrence patterns detected by the JSDM might refer either to real biotic interactions or represent a shared species response to a missing covariate in the analysis Dormann et al., 2018;Warton et al., 2015). In our case, all the positively interacting species detected by our model were present at least at 20 sites (except the nightingale, recorded in 19 sites). This result points to species that can cope well with the urban disturbances and, probably, are benefiting from anthropogenic resources.
We speculate that the presence of important resources for urban birds, such as feeders or artificial nesting sites in highly urbanized areas (personal observation) (Kark et al., 2007;Plummer et al., 2019), might explain this pattern for most of the species.
We did not find any negative association among bird species.
Community responses to disturbance are scale-dependent (Chase et al., 2018), and some responses might be detectable only when using broad spatial scales (Planillo et al., 2015). While competitive interactions may influence species responses to urbanization at broad scales , we focused on the community inhabiting the city. Another possible explanation for the lack of competitive interactions is the potential stress caused by anthropogenic disturbance (Beaugeard et al., 2019;Strasser & Heath, 2013). It has been shown that communities under stress are characterized by positive rather than negative interactions (Callaway et al., 2002;He et al., 2013).
Our results are also subject to some caveats. We used data on ground-dwelling arthropods as a proxy for arthropod diversity. The survey of canopy-dwelling arthropods would be interesting in further studies. Our data come from observations under field conditions, and as such, they are correlational. However, we firmly believe that the responses we found in the data are F I G U R E 6 Associations among bird species in urban areas identified by the JSDM, whose 95 CI did not overlap zero. All associations were positive. Species have been categorized based on their response group (see results). The size of the dot represents the number of sites where the species was recorded (total 29 sites) reliable, as the identified responses are coherent with other studies (Aronson et al., 2014;Beninde et al., 2015;Callaghan et al., 2018;Sol et al., 2014). Caution is also needed when inferring whether a species is successful in a disturbed environment.
Therefore, our study refers only to the observed abundance and cannot be extrapolated to population viability.
We conclude that high invertebrate prey abundance (here included as arthropod abundance) is a key variable for bird community composition in urban areas, and high levels of prey abundance can counteract, to some degree, negative effects of anthropogenic disturbance. In the last 30 years, insect abundance has declined by up to 70% (Hallmann et al., 2017), coupled with a insectivorous bird populations in Europe declining by 13% in abundance, and bird populations in the United States show a decrease of 30% in abundance since the last decades of last century (Bowler et al., 2019;Rosenberg et al., 2019). Against this background, our results have crucial implications for sustainable urban planning if we want to avoid a "silent spring" (Carson, 2002) in cities. Keeping areas with high prey abundance in the city will help maintaining bird diversity and thus decrease the homogenization process that urbanized areas currently undergo (Evans et al., 2018;Ferenc et al., 2014;McKinney, 2006).
In order to preserve bird biodiversity in urban areas, native and diverse arthropod communities should be encouraged in green spaces.
Additionally, a healthy bird community will help in the biotic control of insect pests, preventing damage to vegetation. We suggest some management actions: increase invertebrate abundance in urban parks through the installation or maintenance of structures for arthropod survival by an appropriate habitat management, for example extensive or reduced mowing, leave dead wood and stones, walls as nesting substrates, preserve wastelands and decrease or avoid the use of pesticides; increase the habitat diversity for birds by providing both forested (including dead wood) and open green areas in urban parks; and decrease anthropogenic disturbance, such as noise or human density in some designated core areas to allow the regeneration of sensitive species. (funding number 01LC1501). Open access funding enabled and organized by ProjektDEAL.

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/ddi.13169.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data used in this study are provided in the Supplementary Material files. Stephanie Kramer-Schadt https://orcid. org/0000-0002-9269-4446