Introduction

Understanding the spatial distribution of factors that determine ecosystem services is key to managing human dominated landscapes for biodiversity and people (Díaz et al. 2013; Gross et al. 2017). Factors such as habitat loss and fragmentation not only affect the provisioning of services including water supply, timber production or pollination, but also the potential of people to interact and care about nature. While a high number of small fragments can increase interactions between people and nature (Mitchell et al. 2015; Zambrano et al. 2019), they also may have negative effects on biodiversity that can reduce the potential of people to encounter unique ecosystems or charismatic or rare species (Davies et al. 2004; Bełcik et al. 2020; Halstead et al. 2019). Therefore, landscape changes not only shape biodiversity, but also the ecosystem services related to spiritual, educational, and recreational non-material benefits people derive from nature (cultural ecosystem services; Ridding et al. (2018)). Understanding the factors that influence those benefits in human-transformed landscapes is key to fostering positive relationships between people and nature.

Despite the wealth of research demonstrating the positive impact of nature interactions on human health and wellbeing (Sandifer et al. 2015; White et al. 2019), there is still limited empirical evidence concerning how changes in species assemblages influence human-nature engagement. Differences in the diversity of flora and fauna assemblages can capture human interest and curiosity, leading to increased exploration and appreciation of biodiversity (Fairchild et al. 2018; Tribot et al. 2019). On the other hand, biodiversity loss can lead to the “extinction of experience” due to fewer opportunities for people to interact with nature (Soga and Gaston 2016). More attention is needed to strategically plan how to foster connections with nature while protecting biodiversity that provides myriad benefits. To achieve this, it is crucial to operationalize methodologies aimed at uncovering the significance of the spatial, temporal, and socio-economic factors driving human nature interactions in transformed landscapes (Soga and Gaston 2020).

One factor that can affect how people perceive nature is the presence of species with traits that are considered attractive or appealing to humans (Lišková and Frynta 2013; Prokop and Fančovičová 2013; Prokop and Randler 2018). Studies from different regions of the world have demonstrated that certain species traits such as color, size and shape favor positive feelings and perceptions of species as beneficial. Such traits can also promote connectedness of people with a place (Garnett et al. 2018; Prokop et al. 2023; Zhang et al. 2023). “Attractive” traits could even influence human preferences toward conserving particular species and support education and fundraising (Hoyle et al. 2017). While there is some evidence for which traits make a species attractive to humans (Lišková and Frynta 2013; Garnett et al. 2018), how the overall attractiveness of faunal assemblages changes in human dominated landscapes remains unexplored.

Landscape structure (composition and configuration) is a major driver of species richness and the distribution of species traits (Martin et al. 2019; Suárez-Castro et al. 2020), and thus has the potential to affect the overall aesthetic attractiveness of an assemblage. Landscape composition refers to the amount of each land cover type or land use within a landscape, and it filters species that share traits necessary for exploiting certain types of habitats (Evans et al. 2011; Sol et al. 2014; Litteral and Shochat 2017). Landscape configuration refers to the spatial distribution of landscape elements within a given area (e.g., number of fragments, their shape, and spatial arrangement), and it interacts with landscape composition to drive trait diversity in human dominated landscapes (Martin et al. 2019; Suárez-Castro et al. 2022a, b). For example, fragmentation of native vegetation tends to increase the negative effects of built infrastructure on small species with low dispersal capacities (Müller et al. 2013; Litteral and Shochat 2017; Suárez-Castro et al. 2022a, b). Many of these species could display a high variety of colors and shapes that are often thought to be important in determining human preferences (Correia et al. 2016). The loss of species vulnerable to habitat loss and fragmentation may impact the representation of attractive traits in an assemblage.

Analyzing diversity patterns of trait diversity in urban areas can provide key insights about how landscape change affects the potential of people to interact with species considered attractive. Human-induced disturbances associated with urbanization, including light pollution and turbidity of water bodies, can have significant effects on the production, perception, and function of coloration in animal species that produce color-based signals (Koneru and Caro 2022). Additionally, habitat loss and fragmentation may expose colorful animals to increased competition and predation risks (Spaniol et al. 2019, 2020), or they could favor species with aggressive behaviors (Beggs et al. 2020; Westgate et al. 2021). In turn, loss of “attractive” traits could result in people perceiving negative experiences with nature if they are left with an assemblage perceived to be unattractive (Prokop and Randler 2018). Despite this, there is a lack of tests of how landscape structure affects the distribution of traits that are considered attractive. This information could be used to foster landscape planning for retaining species assemblages that increase positive experiences between people and nature.

Here, we identify how landscape structure (landscape composition and configuration) affects patterns of trait diversity associated with bird attractiveness. We used data from standardized surveys of birds and urban landscapes in Brisbane, Australia, to explore how changes in landscape structure affect the potential of people to encounter species that are considered attractive. First, we analyze how overall trait attractiveness changes in landscapes that vary in their composition and configuration. Then, we analyze how landscape structure drives the occurrence of traits related to color, song and behavior characteristics. Based on these analyses, we provide new information about the effects of landscape structure on the distribution of bird traits that can influence the value people give to natural spaces within the city. This study generates new insights into how landscapes can be managed to protect species assemblages that maximize attractiveness and thereby help to connect people to nature in urban environments.

Methods

Study area

We conducted structured surveys of bird species present in the Brisbane Local Government Area (hereafter ‘Brisbane’) located on the eastern coast of Queensland, Australia (27° 28′ S, 153° 07′ E; Fig. 1). Approximately half of the city (49%) has tree canopy coverage (Jacobs et al 2014). Brisbane is also located in one of Australia’s most biodiverse regions, with about 400 bird species recorded (eBird. 2023). In recent decades, the city has experienced extensive land clearing and fragmentation of native vegetation as a consequence of rapid population growth, which in turn has promoted declines among woodland birds (Catterall et al. 2010; Shanahan et al. 2011).

Fig. 1
figure 1

Brisbane Local Government Area showing the landscape units at 1 km and the process to select survey points at the 1 ha scale

Full size image

Survey sites

Birds were surveyed across 42 landscapes of 1 × 1 km, and each landscape varied in tree cover area and extent of fragmentation following the stratified random sampling design described below. Although many bird species use areas larger than 1 km2, studies on the effects of urbanization on bird diversity have shown that the scales at which predictor variables relate significantly to local bird assemblages are often between 500 and 2000 m (Litteral and Shochat 2017).

To characterize landscape structure, we selected environmental variables that have been associated with the impact of habitat loss and fragmentation on urban bird diversity, at both local and landscape scales (Catterall et al. 2010; Litteral and Shochat 2017; Batáry et al. 2018). These variables were the proportion of impervious surface and built infrastructure, the proportion of tree cover, and the level of aggregation of vegetation patches (as an inverse measure of fragmentation). Tree cover was selected as the main vegetation variable to describe landscape structure as it is highly correlated with other metrics such as percentage of urbanization (Mitchell et al. 2016).

Tree cover was characterized using a 5 m resolution tree cover layer (presence/absence of vegetation ≥ 2 m in height) from 2009 obtained from Brisbane City Council (BCC) that was based on LiDAR and satellite image-derived foliage projective cover. Built infrastructure was derived from a 1:50,000 land use map of South-east Queensland (25 to 100 m horizontal resolution) based on Landsat and SPOT5 satellite imagery, aerial photography, ancillary data sets, and field observations from 2011, 2012, and 2013 (Mitchell et al. 2016). To measure fragmentation, we used a “clumpiness” metric (CLUMPY, Fragstats) that measures spatial aggregation within a set area (McGarigal et al. 2002). This metric is a reliable measurement for habitat fragmentation because it is relatively insensitive to variation in habitat extent (Wang et al. 2014).

To select the landscapes where we carried out the surveys, we divided the Brisbane Local Government Area into landscape units of 1 km. Each landscape unit was assigned one of the following tree cover categories: low (< 30%), medium (30–60%), and high (> 60%). For each tree cover category (low, medium and high), we selected 16 landscape units stratified across different levels of fragmentation at the 1-km scale, ranging from low (clumpiness index > 0.8) to high (clumpiness index < 0.5) levels of fragmentation (Fig. 1). These categories were defined based on the quantile distribution of clumpiness values for Brisbane. In addition, we calculated the median foliage height diversity (FHD) in each landscape from remote sensing data processed by Caynes et al. (2016). FHD is a relative density measure that describes how evenly vegetation is distributed across different vertical strata (Caynes et al. 2016). A detailed description of the methodology to select landscape units and survey points, as well as the data used to quantify landscape structure and vertical height diversity can be found in Suárez-Castro et al. (2022a, b), Caynes et al. (2016) and Mitchell et al. (2016).

Bird surveys

We surveyed birds in three random locations within each 1-km2 landscape unit. These locations were selected considering different percentages of tree cover (< 30%, 35–60%, and > 60%) when possible at the 1-ha scale. This process resulted in the selection of 126 survey points distributed across 42 landscape units. Within each 1-ha grid (three in total per landscape unit), we chose a random point that could be located on a street, park, backyard or inside a bushland reserve. If a survey point was inaccessible, we chose another random point inside the 1-ha grid until we could survey the area. The choice of 1-ha grids enabled the capture of heterogeneity in vegetation and impervious surfaces inside each landscape unit (Mitchell et al. 2016).

The same observer (AFSC) carried out 5-min point counts between dawn and 10 am to estimate bird density in each landscape unit. We did not aim to sample the entire 1-ha grid, and we only considered birds that were seen from a fixed position. Each sampling point was visited four times—twice per each of the two seasons of summer (January–March 2016) and winter (mid-May–August 2016). Each bird was recorded as seen, heard or flying over, while the distance between the bird and observer was recorded using a rangefinder TruPulse 360B. Birds recorded as flying over were omitted in the statistical analyzes to avoid biases in recording species at local scales. Each bird detection was assigned to one of five distance classes from the site (0–10 m, 10–20 m, 20–40 m, 40–50 m, and > 50 m). All survey data can be found in Suarez-Castro et al. (2022b).

We aimed to record a heterogeneous assemblage of species and detect relative richness and trait diversity among landscape units. Our study design (same observer, stratified random sampling) aimed to ensure that detection errors should be the same among the surveyed sites, and that our estimates of relative richness and diversity should be unbiased. We focused on terrestrial forest or bushland dependent birds, thus excluding aquatic and coastal species. We acknowledge that aquatic species that inhabit blue spaces may play a crucial role connecting people with nature. However, our focus on bushland species associated with urban green spaces facilitates the quantification of effects of landscape composition and landscape configuration variables, and avoids biases produced by local-scale habitat variables such as the presence of water bodies.

Species traits and attractiveness metrics

Data on traits that may affect human preferences for birds were obtained from Garnett et al. (2018). These authors used a choice model to explore which aesthetic features of birds were considered most attractive by members of the Australian public while controlling for biases associated with knowledge of iconic species. The traits analyzed in that study include appearance [colorful, boldly marked (i.e. pattern), and dull (i.e. grey-brown)], size, song (melodious, quiet, harsh) and behavior (confiding, spectacular, secretive). A definition of the traits used in this study can be found in Fig. 2. Other traits related to attractiveness such as presence of elongated rump feathers were not considered as they are uncommon in the studied bird assemblage. In addition, morphometric traits (e.g. overall shape) are not included in the attractiveness score. Despite this, the traits used in this study should account for overall differences in the appearance of birds from the studied assemblage. We also obtained data on the body size for each species (Garnett et al. 2015).

Fig. 2
figure 2

Traits related to color, song and behavioral characteristics considered in this study. All trait values and definitions have been taken from Garnett et al. (2018). Species from top left to bottom right: Malurus cyaneus, Dacelo novaeguinae, Rhipidura leucophrys, Grallina cyanoleuca, Cacomantis flabelliformis, Meliphaga lewinii, Pachycephala pectoralis, Cracticus nigrogularis, Cacatua galerita, Trichoglossus haematodus, Pardalotus punctuatus, Eopsaltria australis, Todiramphus sanctus, Merops ornatus, Rhipidura albiscapa, Entomyzon cyanotis, Petroica rosea, Megalurus timoriensis. All photos taken by AFSC except M. timoriensis, G. cyanoleuca and R. albiscapa used under Creative Commons License 2.0

Full size image

We computed a mean attractiveness metric for birds present in each landscape unit using the following equation provided by Garnett et al. (2018) for the attractiveness score (A) of each species i:

$${A}_{i}= {x}_{i}\widehat{\beta }$$

where \(\hat{\beta }\) is a vector of coefficients derived using a latent class model to account for preference variation across respondents, and xi is the coefficient for each trait x derived from a choice experiment (Garnett et al. 2018). The overall attractiveness metric M for each landscape unit represents a weighted measure of traits characteristics people valued most:

$$M= \sum_{i=1}^{n}{p}_{i}{A}_{i}$$

where \({p}_{i}\) and \({A}_{i}\) is the relative abundance and attractiveness score of species i, respectively.

To test if species are more likely to occur in areas where their trait values are more dissimilar to the local community-weighted mean, we also calculated attractiveness dispersion as a measure of the mean distance of each species score to the mean attractiveness value in each landscape. It is calculated as (Laliberté and Legendre 2010):

$$FDis= \frac{\sum {a}_{j}{z}_{j}}{\sum {a}_{j}}$$

where ai is the abundance of species i and zj is the distance of species i to a weighted centroid c:

$$c= \frac{\sum {a}_{j}{x}_{ij}}{\sum {a}_{j}}$$

where ai is the abundance of species i, and xij the attribute of species i for trait j. This metric allows to test how landscape structure may be filtering the occurrence of species with extreme attractive trait values. All metrics were computed using the package FD (Laliberté et al. 2014) in R version 3.6.2.

Data analysis

Species abundance estimates

To calculate species abundance and to account for the effect of environmental factors for detection probability, we used multiple-covariate distance sampling (Thomas et al. 2010). For each species, we fitted two alternative models to the observed distribution of detections using half-normal or hazard key functions. Covariates for each model included season, detection type (i.e., if the bird was seen or heard only), and percentage of tree cover at the 1-ha resolution. For species recorded infrequently (< 30 records), a common detection function was generated by grouping infrequently recorded species with more common species expected to have similar detectability (Alldredge et al., 2007). These groupings were based on similarities in diet (i.e., frugivores, nectarivores, insectivores) and main foraging stratum (canopy, understory, ground). For each of these groups, we used species as a covariate to model the scale parameter. For each landscape, we then calculated density estimates and analytical 95% confidence intervals for each species using the model with the lowest Akaike information criteria (AIC). The density of each species at the 1-km scale was estimated as the average density across the three sites surveyed in each landscape unit. Results of these analyses are presented in Suárez-Castro et al. (2022a, b).”

Effects of landscape structure on overall bird attractiveness.

To evaluate the relationship between landscape structure and the overall mean attractiveness score in each landscape weighted by species abundance, we used structural equation models.. Following Suarez-Castro et al. Suárez-Castro et al. (2022a, b), we specified a model where landscape composition variables (proportion of built infrastructure [impervious surface + buildings] and tree cover), landscape configuration variables (clumpiness index) and foliage height diversity have a direct relationship to weighted mean attractiveness scores. We also included a mediating effect of fragmentation on the effects of tree cover and built infrastructure by assuming a causal relationship between these variables and fragmentation. In addition, we included species richness in the models to test if there is a correlation between the overall attractiveness score and the number of species in each landscape. Since bird guilds respond in different ways to landscape structure (e.g. insectivores vs frugivores; Suarez-Castro et al. Suárez-Castro et al. (2022a, b), we conducted these analyzes for all the species, and then separately for insectivores and frugivores/nectarivores.

Following the recommendations by Weston and Gore (2006), we evaluated the overall model fit of the proposed models using two parsimony-based fit indices, the Tucker–Lewis index (TLI) and the root mean square error of approximation (RMSEA). Specifically, a TLI value of 0.95 or larger and RMSEA values < 0.06 indicate a good model fit (Weston and Gore 2006). All response and explanatory variables were standardized (mean = 0, SD = 1) prior to fitting the models. Standard errors for the residuals models were estimated using a bootstrapping approach with 100 iterations Analyzes were performed using the lavaan package (Rosseel 2012) in R Version 3.6.2. We then plotted the relationship between each of the landscape configuration and landscape composition variables and the attractiveness metric obtained for each landscape.

Effects of landscape structure on specific traits

We used multinomial logistic regression to examine the effect of landscape structure variables on each of the following categorical traits: Song (melodious, quiet, harsh), Personality (confiding, secretive, spectacular) and Color (colorful, pattern and dull). For each trait (i.e. response variable), we first built a full model that incorporated all the components of landscape structure as predictors. The full model included built infrastructure, fragmentation, tree cover and foliage height diversity as independent variables. We also specified interaction terms between fragmentation and landscape composition (built infrastructure and tree cover), and foliage height diversity and landscape composition. Previous work has shown that body size is a key determinant of species responses in urban environments (Batáry et al. 2018; Suarez-Castro et al. Suárez-Castro et al. (2022a, b, so we included this trait as a predictor in the models. This way, we assumed that landscape structure drives the distribution of attractive traits in different ways depending on the species size.

To account for differences in species population sizes across landscapes, we used species counts in each landscape as weightings. We then employed a stepwise approach procedure to select models with the lowest Akaike information criterion (AIC) values and to identify the best model. Since some highly dominant species in Brisbane (Rainbow Lorikeet Trichoglossus haematodus and Noisy Miner Manorina melanocephala) may bias the observed patterns, we evaluated how consistent were the models with and without these species. Analyzes were performed using the mclogit package (Croissant 2020) in R Version 3.6.2.

Results

We recorded a total of 82 species. A list of all species and their associated traits can be found in the Supplementary Information 1 (Table S1). Categories of traits were evenly represented for color (colorful = 31 (38%), dull = 27 (33%), pattern = 24 (29%)), and song (melodious = 26 (31%), harsh = 23 (28%), quiet = 33 (40%)). Traits were unevenly represented for personality, with the number of species classified as spectacular lower (n = 14) than confiding (n = 39) and secretive (n = 29). The proportion of individuals recorded across all landscapes was also evenly distributed for color and song categories, whereas for personality, 63% of the individuals recorded were classified as confiding (readily approachable), 27% as spectacular and only 10% as secretive.

Associations between overall attractiveness score and landscape attributes

The SEM models for all species had a good overall fit (TLI = 1, RMSEA = 0.038), and included direct pathways for tree cover, percentage of built infrastructure, and fragmentation, as well as mediating effects of fragmentation on landscape composition effects (Fig. 3). Foliage height diversity did not significantly improve overall model fit (TLI = 1, RMSEA = 0.545), suggesting that the vertical structure of the vegetation did not help to explain the effects of landscape structure on overall bird attractiveness at the 1 km2 resolution. SEMs showed that landscape structure filters species with extreme trait values, as seen by the negative effects of built infrastructure on attractiveness community weighted mean (Figs. 3a, 4). Despite this, weak effects of built infrastructure on functional dispersion show that highly urbanized areas can maintain species with extreme trait values, even if the community weighted mean declines (Fig. 3b). Although fragmentation does not have a direct effect on mean attractiveness, this variable has a negative effect on functional dispersion (Fig. 3c). We did not find a clear direct relationship between tree cover and overall attractiveness when all the species were considered (Fig. 3a, b). Our results also show that the number of species present in a landscape unit is more correlated to average attractiveness compared to attractiveness dispersion (Fig. 3a, b).

Fig. 3
figure 3

Structural equation models showing the direct effects of landscape structure on attractiveness community weighted mean (CWM) and attractiveness dispersion. A one-headed arrow represents a proposed, direct relationship between two variables. A two-headed arrow indicates an unanalyzed relationship, where a covariance between the two variables is specified and has no implied direction of effect. Numbers between arrows indicate standardized path coefficients, which allow for a direct comparison of effect sizes between variables; the thickness of the line represents the strength of the relationship

Full size image
Fig. 4
figure 4

Relationship between landscape configuration (a), landscape composition (b, c) and overall attractiveness metric of bird species present in landscapes from Brisbane, Australia

Full size image

Models developed for each guild (Insectivores, Frugivores/Nectarivores; Fig. 3c–f) included the same pathways as those developed for the whole assemblage. However, landscape structure variables affected overall attractiveness metrics for each guild in distinct ways (Fig. 3, Supplementary Information 2). Built infrastructure did not affect the attractiveness community weighted mean when each guild was evaluated separately. In addition, the proportion of tree cover was associated with an increase in overall community weighted mean attractiveness of insectivores (Fig. 3c). We did not find clear effects of fragmentation on the attractiveness metrics for individual guilds, except for the community weighted mean for frugivores/nectarivores, where overall attractiveness decreased in more fragmented landscapes. The correlation between species richness and attractiveness metrics is lower at the guild level compared to the whole assemblage and is particularly weak for the attractiveness community weighted mean in both guilds (Fig. 3c, d).

Influence of landscape attributes on the distribution of species traits

The models with the lowest AIC for the distribution of categorical traits included the same set of variables: built infrastructure, tree cover, fragmentation, vegetation vertical structure, body size, and their interactions (Table S3). The best model for song characteristics had an explained deviance of 37.2%, whereas for personality and color the explained deviance was 41.1 and 24.4% respectively (Fig. 5). Although the overall explained deviance decreased when models were performed excluding the most dominant species (Rainbow Lorikeet T. haematodus and Noisy Miner M. melanocephala) (SM Table S3, Figure S1), we found similar patterns for most of the landscape structure variables and the analyzed traits. Therefore, the description of the results below is based on the models that included all species, unless stated otherwise.

Fig. 5
figure 5

Multinomial logit models showing the estimated effects of landscape structure variables on each categorical trait (Song, Color and Personality) associated with bird attractiveness as defined by Garnett et al. (2018). Frag landscape fragmentation, size species body size, Tr tree cover, Veg vegetation vertical structure, Built built infrastructure, RL reference level

Full size image

For all models, body size moderates the effects of landscape structure on the evaluated traits. Compared to quiet species, high levels of vegetation structure, low levels of fragmentation and low levels of built infrastructure promote melodious species and affect negatively harsh species (Fig. 5a). However, these relationships change when the interaction with body size is included, and effects become positive on harsh species. High levels of built infrastructure have a high negative effect on colorful species, but positive interactions between body size and built infrastructure promote large, colorful species in areas with high levels of urbanization (Fig. 5b). On the other hand, when personality traits are considered, both tree cover and built infrastructure have positive effects on secretive and spectacular species compared to confiding species. However, when body size is considered, this relationship changes (Fig. 5c) and spectacular species are affected negatively. Although fragmentation tends to negatively affect melodious, secretive and spectacular species compared to quiet and confiding species, we did not find any interaction between this variable and body size to drive patterns of traits associated with attractiveness (Fig. 5c).

Discussion

Despite the wealth of research demonstrating the positive impact of nature interactions on human health and wellbeing (Sandifer et al. 2015; White et al. 2019), there is still a lack of empirical evidence concerning how changes in species assemblages influence the potential of people to interact with species considered attractive. Our approach shows how a trait-based approach may assist in revealing the importance of spatial factors driving how people may engage with particular species in urban landscapes (Soga and Gaston 2020). By combining information about landscape structure and species’ morphological and behavioral attributes, we show that landscape composition and configuration affect the ability of people to encounter species considered attractive. This information can support planning to protect species and assemblages that may prevent the extinction of experiential connections with nature.

Effects of landscape structure on bird attractiveness metrics

Our models demonstrate that the relative amount of built infrastructure in a landscape interacts with fragmentation and tree cover to drive the overall attractiveness of the landscape’s bird assemblage. Areas with high built infrastructure tend to have low community weighted mean attractiveness for birds in Brisbane. These findings complement evidence from other regions showing that urbanization negatively affects species that provide specific cultural services (Cox et al. 2018; Valente-Neto et al. 2021). However, weak effects of built infrastructure on functional dispersion suggest that even at low levels of species richness, landscapes can support some species with high attractiveness scores (e.g. Rainbow Bee-eater Merops ortus Willie Wagtail Rhipidura leucophrys). Negative effects of fragmentation on functional dispersion suggest that maintaining these species with extreme trait values in areas with high built infrastructure could be achieved if fragmentation is low. At the assemblage level, we did not find a clear direct relationship between tree cover and overall attractiveness. It is thus possible to maintain high levels of bird assemblage attractiveness even at relatively low levels of tree cover, possibly because remnant green spaces with high levels of vegetation structure can attenuate negative effects of built infrastructure (Callaghan et al. 2018; Mbiba et al. 2021).

We found that relationships between bird attractiveness and landscape structure change when they are analyzed at the guild level. For frugivores/nectarivores, built infrastructure does not affect the bird attractiveness community weighted mean of individual guilds as much as it affects attractiveness dispersion. This suggests that even if some species with unique attractiveness scores are lost, the presence of a few attractive species can maintain high mean attractiveness in a landscape unit. This is most likely the case for some species such as the Rainbow Lorikeet (T. haematodus), a species with a high mean attractiveness score (> 50 quantile of attractiveness for the guild), and that tends to be abundant in urban areas (Taylor et al. 2013). However, negative effects of built infrastructure on the dispersion of attractiveness traits for frugivores suggests that some species considered highly attractive due to coloration and/or song characteristics (e.g. the Scarlet Honeyeater Myzomela sanguinolenta, Yellow-faced Honeyeater Lichenostomus chrysops) are lost in highly urbanized environments.

Our results show that loss of tree cover in urban areas leads to insectivore assemblages that are perceived to be less attractive than the insectivore assemblages in more intact areas, due to changes in assemblage composition. Losses in tree cover can thus promote insectivore guilds made up of birds with very similar characteristics of song, personality and color. This corresponds with the loss of species with high attractiveness values, such as small-bodied forest dependent species that tend to be more vulnerable to urbanization (e.g. White-throated Gerygone Gerygone olivacea, Australian Golden Whistler Pachycephala pectoralis). Negative to weak effects of landscape structure variables on the dispersion of insectivore attractiveness values show that even when mean attractiveness declines, remnant species can have very dissimilar trait values.

Even though we cannot establish a causal relationship between landscape structure variables and traits that determine attractiveness of species to humans, we show that body size moderates the associations between landscape structure and song complexity, personality, and color. Most (30 out of 31) colorful species recorded in this study are small to medium size species (Table S1), and 85% of species with melodious calls are small-bodied (< 65 g). In Australia, both high fragmentation and the presence of aggressive birds have negative impacts on low dispersal, small-bodied species (Shanahan et al. 2011; Beggs et al. 2020; Westgate et al. 2021). These negative impacts are reflected in the reduction of mean bird attractiveness in highly urbanized areas, despite the ability for larger-bodied species in Australia to cohabit with built infrastructure (Callaghan et al. 2019; Suárez-Castro et al. 2022a, b). We also found complex relationships between landscape structure and other traits associated with harsh-sounding species (e.g. Australian King-Parrot Alisterus scapularis, Sulphur-crested Cockatoo Cacatua galerita, Leaden Flycatcher Myiagra rubecula), or species that show bold pattens of color (Pied Butcherbird Cracticus nigrogularis, Welcome Swallow, Hirundo neoxena, Double-barred Finch Taeniopygia bichenovii). These traits are shared by species with a wide range of body sizes (Table S1), which could reflect why effects of tree cover and built infrastructure on species with bold patterns of colour range from positive to negative. Testing the association between other traits that determine species responses to landscape change (dispersal, flock behavior, foraging behavior) and song, personality and color traits constitutes an important area of research to understand how to preserve attractive species in urban environments.

Understanding attractiveness metrics in broader contexts

Our approach deals with a crucial unexplored dimension of how people can interact with nature in cities. It moves away from the focus on metrics such as the amount or accessibility to green spaces, or on diversity metrics for one taxonomic group (e.g. bird species richness), and it considers potential associations with assemblages that differ in the distribution of species traits. Analyzing whether human-nature interactions benefit from a diverse set of attractiveness traits is an important area of research. We relied on the published results of a national survey to infer which traits are considered more attractive (Garnett et al. 2018), and this might not be representative of the residents in Brisbane. We recognize that the way a person experiences and responds to distinctive characteristics of nature/biodiversity will be influenced by personal, societal and cultural associations, as well as previous experiences (Soga and Gaston 2016, 2020).

Several next steps are thus needed to understand how this information can promote nature conservation and benefits to human health and wellbeing simultaneously. First, we do not know how the perceptions of species traits by humans relate to different domains of human wellbeing, including physical, emotional, cognitive, social and spiritual (Linton et al. 2016). These perceptions may vary among countries, regions, or even socioeconomic sectors within a single city. More research and methods (e.g. participatory mapping, customized surveys) are needed to connect traits considered attractive to domains of human health and wellbeing.

We also do not know how different traits may interact, resulting in synergistic or antagonistic outcomes and complicating management of landscape for faunal attractiveness. For example, while the Pied Butcherbird (C. nigrogularis) has a melodious call that could elicit joyful emotions (thereby improving a person’s mood), they can also be highly aggressive during the breeding season, which could result in overall negative responses to the bird’s presence despite its attractiveness rating. Finally, it is also unclear how human-nature interactions may change depending on whether bird attractiveness is driven by a few species, or if there is a cumulative effect when multiple species interact (melodious sounds from multiple species). Studies on how trait diversity may impact health and wellbeing outcomes must extend beyond single traits (e.g. color) and instead consider combinations of traits within an ecological landscape. This would be a more accurate reflection of a person's real-world experience of interactions with biodiversity.

Implications for management

Our results are consistent with previous research showing that urban built infrastructure negatively affects species richness and homogenizes species traits diversity (Cox et al. 2018, Sol et al. 2020). However, patterns of trait diversity are more variable, with urban infrastructure having a larger negative effect on species richness compared to functional diversity (Suarez-Castro et al. Suárez-Castro et al. 2022a, b) or attractiveness metrics (this study). Since trait diversity is not necessarily correlated with species richness, local strategies focused on maintaining habitat heterogeneity can reduce the negative impacts of built infrastructure on specific traits, even when species richness is low (Mbiba et al. 2021; Suárez-Castro et al. 2022a, b). As demonstrated for other metrics of bird diversity (Callaghan et al. 2018; Mbiba et al. 2021), maintaining high levels of bird attractiveness is feasible through practicing differential landscape management approaches. These approaches could include reducing fragmentation in highly built-up areas by preserving well-connected, medium-sized patches (Catterall et al. 2010; Shanahan et al. 2011).

Despite the potential for strategies to mitigate and/or reduce the negative effects of fragmentation and built infrastructure on mean attractiveness, urban environments could lose unique species considered highly attractive in their coloration and/or song characteristics. This is shown in our study by consistently stronger effects of landscape structure on attractiveness dispersion compared to community-weighted mean attractiveness. Our findings indicate that assemblages in areas with high infrastructure tend to be less diverse in color and song characteristics, with small-bodied species being more impacted by high levels of built infrastructure and fragmentation. To enhance species richness and protect unique species at the landscape scale, efforts should focus on reducing the negative effects of built infrastructure by promoting urban reserves and sensitive urban design practices (Garrard et al. 2018) that increase connectivity and protect key habitat at local scales (Wintle et al. 2019). Further research is needed to understand the implications of losing species with unique trait combinations on overall human-nature relationships.

Using a trait-based approach can help to track success of initiatives that seek to achieve wins for both biodiversity and human wellbeing. Recent research has shown that urban assemblages can suffer cultural losses as a product of changes in species assemblages in urban areas (Valente-Neto et al. 2021). Understanding how landscape structure affects the distribution of these species and their associated traits is thus an urgent area of research in human dominated landscapes around the world. This information is key to promote urban landscapes that support attractive species, and as such, could enhance greater human-nature connection and engagement, and could lead to greater support for conservation and positive human health outcomes. Existing and future restoration and conservation strategies and actions can concomitantly benefit both nature and humans if they additionally account for which species’ traits benefit people through the ecosystem functions and services that they support.