Study area

Our study was focused on Southern California, with an emphasis on the greater Los Angeles area, situated in the California Floristic Province, one of 36 biodiversity hotspots in the world [49,50]. The region is also one of five Mediterranean ecosystems in the world, which occur only on the western margins of landmasses between 30 and 40° latitude and which are typically characterized by cold and wet winters and warm and dry summers. Southern California has diverse topography, including the transverse mountain ranges to the north and east and the peninsular mountains ranges to the south, as well as diverse habitat types, including chaparral, coastal sage scrub, oak woodlands, coastal dunes and bluffs, riparian woodlands, and a variety of wetland habitats, which host over 2,200 species of vascular plants [51]. While the calculation of a formal CUTI was limited to areas within the city of Los Angeles, for the estimation of species-level UAI, our study area included all land within a 200 km buffer of the Los Angeles City boundary (approximate centroid: 34.031656 N, 118.241716 W), including the cities of Los Angeles, San Diego, Bakersfield, and Santa Barbara. We chose to focus on this broad geographic region because we were interested in creating a metric that could be measured repeatedly over time and would be robust to species that do not currently reside in our focal area of Los Angeles, but could colonize in the future. Furthermore, we treat urban tolerance here (as measured by the UAI) as a species-level trait, which is best estimated using occurrence data from a broader geographic region than just Los Angeles. As such, we aimed to include areas with a wide range of levels of urbanization, including multiple manifestations (i.e., cities) of urban land-use within the broader Southern California region.

Urban intensity

In order to estimate urban associations, we first had to define a continuous spatial layer of the urban intensity of our study region. To do this, we used Principal Component Analysis (PCA), which decomposes multivariate datasets into major axes of variation, to create a single composite index of urban intensity from multiple sources. Following [38], this index included the Visible Infrared Imaging Radiometer Suite (VIIRS) nighttime lights data layer, but we added additional environmental variables related to urbanization as different taxa are likely to respond to different aspects of the urban environment. We initially tested a set of six data layers to depict urban intensity across our study area, but we removed several layers including pollution of fine particles smaller than 2.5 micrometers (PM2.5) [52], Average Traffic Volume [52], and Population Density [53] due to collinearity and coarser resolution. Thus, our PCA index represents a composite of three layers: (1) light pollution from VIIRS Version 4 DMSP-OLS Nighttime Lights Time Series (https://eogdata.mines.edu/products/dmsp/#v4_dmsp_download) [54], (2) "Percent Impervious Surfaces" from National Land Cover Database (NLCD) 2016 Percent Developed Imperviousness (CONUS) (https://www.mrlc.gov/data/nlcd-2016-percent-developed-imperviousness-conus) [55], and (3) "Noise Pollution" from National Park Service (NPS) Geospatial Sound Modeling 2013–2015 (https://irma.nps.gov/DataStore/Reference/Profile/2217356) [56]. All spatial layers were reprojected and resampled, as needed, to a 0.25 x 0.25 mile grid prior to combination. The first axis of the PCA explained 86.5% of the variance in all three layers (S1 Table), indicating as expected that the three layers are all indicative of the same general process (i.e., “urban intensity”) yet individually add unique information. Because PCA axis 1 (“PC1”) explained >70% of the variation, we retained it as our sole spatial index of urban intensity (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Map of urban intensity measured across a broad study region in Southern California.

Our metric of urban intensity was determined as the first PCA axis of three different variables. Warmer colors indicate higher levels of urban intensity. Solid lines detail major roadways within California. Map created using data from [57–61].

https://doi.org/10.1371/journal.pone.0295476.g001

iNaturalist records and data quality filtering

We focused our analysis on selected target taxonomic groups, which were picked a priori with expert input to represent 12 taxonomic groups that are generally well-detected and well-surveyed by community scientists on the iNaturalist platform. The 12 groups include: snails and slugs (Order: Stylommatophora); spiders (Order: Araneae); dragonflies and damselflies (Order: Odonata); grasshoppers, locusts, and crickets(Order: Orthoptera); leafhoppers (Family: Cicadellidae); lady beetles (Family: Coccinellidae); hoverflies (Family: Syrphidae); bees and wasps (Family: Apidae and Vespidae); butterflies and moths (Families: Papilionidae, Pieridae, Lycaenidae, Nymphalidae, Sphingidae and Erebidae); herpetofauna (Classes: Amphibia and Reptilia); mammals (Class: Mammalia); and birds (Class: Aves). We queried the iNaturalist API for occurrence data on 13 January 2022 using the ‘rinat’ package version 0.1.8 [62]. We downloaded all iNaturalist records between 2011–2021 for the higher order taxa groups defined above, only limiting records to “research quality” georeferenced occurrences (i.e., those with a consensus taxonomic ID and location coordinates) bound within our broader study region (i.e., 200 km radius around Los Angeles). After downloading, we further filtered our data to remove species that had more than 60% of all observations records marked as “Geoprivacy = obscured” (a situation where iNaturalist provides spatial coordinates of sightings, but these coordinates are randomly offset by up to 22 km from the true location). Although we additionally filtered out all obscured spatial records for all species, we wholly excluded species meeting this arbitrary threshold as we believed that such a widespread degree of geoprivacy indicated species for which remaining iNaturalist data would not likely represent the species’ true distribution within the study area. Finally, we reclassified all records identified to the subspecies level to the species level following other similar work using related datasets [63,64].

Expert review

After downloading iNaturalist records and applying our initial hard filters for data quality, we engaged the LASAN Biodiversity Expert Council, a regional group of scientists and taxonomic specialists who advise the annual biodiversity report, to further assist in data curation and QA/QC. Five experts in areas of specific taxonomic focus for the species included in this study (i.e., arthropods, mollusks, mammals, birds, and herpetofauna) were asked to evaluate occurrence data for the specific higher order taxonomic grouping they specialized in by using the following questions: 1) Is this species native to the study area?; 2) Is the species terrestrial?; 3) Is the natural history of this species so different from others in its taxonomic grouping that records for this particular species should not be used as indicators of search effort for other similar species?; and 4) Is there any other reason why we should exclude this species from this study? The third question refers to the issue that iNaturalist data are presence-only and do not, on their own, provide information on absence or non-detection. Increasingly, however, ecologists are using multi-taxa presence-only surveys to bin species into ‘detection groups’, whereby an observation of one species at a location provides inference on the non-detection of other species [44,65]. This assumption of substitutability is justified as natural history observers are often searching broadly within taxonomic groups; for example, a birdwatcher’s positive record of one bird species says more about the non-detection of another bird species than it does about the non-detection of a butterfly. In the context of the present study, we did not require species’ occurrences to be perfect indices of non-detection for other species, but simply sought taxonomic groups where the presence of one species in that group would serve as a broad index of survey effort for all species in that group. Thus, we sought via expert review to exclude taxa that differed so much from the rest of their grouping (e.g., diurnal versus nocturnal; or identifiable via photography versus identifiable only via microscope) that they should not be treated as survey effort proxies. For the fourth question, some common reasons for excluding species based on expert review included species with extremely limited distributions that would otherwise be uninformative to urban tolerance (e.g., a species of plethodontid salamander limited to a single remaining population on Mt. Baldy, Los Angeles, USA), or misidentifications based on recent taxonomic splits.

Following data review by taxonomic specialists, we curated their responses to make sure that experts interpreted these questions similarly. We filtered observations based on these responses to exclude non-native species, species unlikely to be detected by typical observers, non-terrestrial (i.e. marine or freshwater) species, and species according to additional criteria as determined by the taxonomic group specialists.

Controlling for differences in sampling effort

We took a number of steps to control for differences in sampling evenness and effort in iNaturalist data. First, to address the inherent biases associated with sampling evenness, we performed a broad spatiotemporal thin [66,67]. We thinned species-specific data to one observation per year within each 0.25 x 0.25 mile grid cell. This produced a database where every species is recorded as present or not detected in each grid cell and for each year between 2011–2021 (i.e. the number of yearly detections out of 11 years). Second, to address the additional bias associated with varied sampling effort, we defined site-specific sampling effort for each of the twelve focal taxonomic groups. We did this because observers may not equally record observations for all taxonomic groups (e.g., an observer may not observe spiders while recording birds). Taxonomic group-specific effort grids were calculated based on the observation effort per year for the corresponding focal taxonomic groups, summing within grid cells the number of years with at least one observation in a year of a species from within a taxonomic group. Thus, taxonomic group-specific effort layers indicate the number of years (0–11) in which observers obtained at least one record of a target group, which serves as the maximum potential number of thinned presences for any given species of that group. As such, the thinned species presence layer and the matching effort layer represent a spatially-varying binomial response, where the number of binomial trials is the effort in a grid cell and the number of binomial successes is the thinned number of species’ presences.

Measuring species-level relationship to urban intensity layer in study area

Species-level indices of urban association were calculated based on thinned occurrence data and taxon-matched effort data across the entire Southern California study region. We calculated a UAI for each species that had at least 25 thinned annual occurrences across our study region. Specifically, using the ‘stats’ package [68], for each species we modeled the number of thinned occurrences, given effort per cell, as a binomial process that varied as a function of a single covariate: the urban intensity (PC1) of the grid cell. This model allowed us to estimate the number of thinned species occurrences as a binomial variable, where the number of successes (i.e., ‘occurrences’) was capped by the number of years with non-zero effort for the taxonomic group in each cell. In this way, our model accounted for taxon-specific sampling effort over time in each grid cell. The resulting logit-linear slope of the trend line, which indicates the relationship between species’ occurrence and urban intensity, was stored as the UAI for a given species. Positive slopes indicate urban tolerance, negative slopes suggest urban intolerance, and slopes of zero indicate no relationship of occurrence to urban intensity.

Calculating a Community Urban Tolerance Index (CUTI)

After calculating a UAI for each species, we quantified a CUTI for each grid cell in Los Angeles by taxonomic group, as well as a composite score for the entire city (i.e., metric 1.2b for the city of Los Angeles). To calculate a taxonomic group-specific CUTI for each grid cell in Los Angeles, we matched species’ UAIs to species occurrences in individual grid cells and calculated a raw CUTI score per cell by taking the mean UAI score of all species within any given cell, weighted by the thinned temporal occurrence of each species (i.e., a value of 1–11 for the number of years that the species occurred in that cell). This resulted in 12 grids (one for each taxonomic group) with a group-specific CUTI score for every cell in the city in which the group was detected. To interpret these results at a broader taxonomic scale, we also calculated the mean CUTI across all pixels for each of the 12 taxonomic groups. Finally, to calculate a composite CUTI across all taxonomic groups, we averaged the 12 taxonomic group grids and city-wide scores. In all cases, the raw CUTI scores were binned into a 5-point scale as follows: -Infinity to -0.5 = 5; -0.5 to -0.25 = 4; -0.25 to 0 = 3; 0 to 0.25 = 2; 0.25 to 0.5 = 1; and 0.5 to Infinity = 0. On this scale, a cell with a CUTI index of 4 or 5 suggests that species in aggregate are more natural-area associated, while a cell with an index of 0 or 1 suggests that species are more urban tolerant. To test for an association between UIA and urban intensity (with the hypothesis that areas of higher urban intensity have lower CUTI scores), we used an ANOVA and Tukey HSD test in the ‘stats’ package [69] with urban intensity as the response variable and categorical binned CUTI scores for each grid cell as an independent variable.

Extrinsic and intrinsic factors leading to differences in UAI

The most urban tolerant species in our study was the slipper snail (C. lubrica). This species is known to be widespread and euryhygric (i.e. able to withstand a broad range of moisture conditions), and it is possible that urbanized areas such as Los Angeles provide year-round access to a variety of moisture regimes through the addition of ornamental landscapes and lawns [70–72]. Other studies have found high native snail abundance in areas of high urban intensity in Tennessee [73]. Given that there were only five species of snails and slugs that remained in our dataset post-filtering, this pattern could largely be due to a small sample size of species, and perhaps more purposeful sampling is required to truly ascertain the affinity of this entire taxonomic group for urban environments.

The most urban intolerant species in our study was the greenish blue butterfly (Icaricia saepiolus). Compared to other North American butterfly families, Lycaenidae is overrepresented in terms of number of species proposed for listing [74]. This is largely due to host plant specificity of Lycaenids, primarily for plants in the genera Lupinus and Eriogonum, and the fact that these plants are adapted to disturbance regimes that are infrequent in the urban context [74–76]. Conservation of many of the special status butterfly species including several Lycaenids therefore relies on maintaining and expanding critical segments of habitat that contain host plant species within urban settings and maintaining habitat fragments of varying sizes through deliberate disturbance as a management tool [76]. For example, the Palos Verdes blue butterfly (Glaucopsyche lygdamus palosverdesensis) is an endangered subspecies of Lycaenid butterfly in Los Angeles County, and findings from the US Fish and Wildlife Service demonstrate that the species appears to be establishing in reintroduction sites due primarily to efforts to remediate historical habitat through mechanical disturbance, non-native plant removal, and planting of successional host plants [77]. Based on the findings from this study, the City of Los Angeles could use existing butterfly observations to identify target areas for conservation for the Palos Verdes blue butterfly and other imperiled butterfly species. In areas where observations overlap with property owned and managed by the City of Los Angeles, the city can focus restoration efforts on increasing host plant abundance through direct plantings of early successional plants and targeted mechanical disturbance to create conditions necessary for early successional habitat needed by the butterflies. Following restoration activities like these, city managers can evaluate success through a reevaluation of this metric on an annual basis.

All other higher order taxonomic groups had higher presence in less urbanized regions within the city. Previous research in Los Angeles, research has demonstrated that the presence of many taxonomic groups is negatively affected by increased levels of urbanization. For example, coyotes and bobcats in mixed urban/natural areas have home ranges that primarily utilize natural areas [78], and some regionally common amphibian species are markedly absent from streams within urbanized areas of the city [79], but within taxonomic groupings there was high levels of variation in individual species-level responses. Many factors contribute to these varied responses of species to different levels of urban intensity, including species-specific functional traits. Specifically, in mammals and birds, functional traits help explain urban tolerance of species, including body size, dietary breadth, clutch size, and nesting strategy, among other predictors in birds [64] and litter size in mammals [80]. While our study does not seek to evaluate all individual species-level traits and how they relate to values of the UAI, some previously noted relationships between species’ functional traits and responses to urbanization are recapitulated in our findings. For example, Cooper et al. [81] found that urban raptors in Los Angeles including Cooper’s hawk (Accipiter cooperii) and red-shouldered hawk (Buteo lineatus) responded to increases in urban cover within their home ranges by increasing nesting in urban sites, while other species of urban raptors such as American kestrels (Falco sparverius) were more likely to nest in less-urban areas within their home ranges. They concluded that nesting strategy may play a role in the response of urban raptors to levels of urbanization, as American kestrels are cavity nesters and dead trees are likely to be removed by homeowners [81]. Notably, these previously-reported responses to urbanization for Cooper’s hawks, red-shouldered hawks, and American kestrel are confirmed in our estimated values of UAI (0.11, 0.19, and -0.03 respectively). Future studies could further explore the relationships between species-specific functional traits and UAI for the 510 native species studied here using existing databases of functional traits such as AVONET [82], AnimalTraits [83], or COMBINE [84].

In addition to intrinsic characteristics of species, many extrinsic factors may contribute to species-level responses to varying levels of urban intensity. For example, research on urban insect populations in Los Angeles found that diurnal temperature range had a consistent negative effect on occurrence of all arthropod species studied [85]. We found a negative association with urbanization across all native arthropods, although we also observed high levels of variation within higher order taxonomic groupings. It is possible that spatial variation in water resources across the city can partly explain this within-group variation, as has been demonstrated for desiccation-sensitive invasive insect species in Los Angeles [86]. The heterogeneous presence of water on the landscape also strongly influences the vegetation community present, which in turn strongly influences arthropod community composition. Additional research could build off our findings to investigate the relationship between our measures of UAI and both spatial variation in abiotic gradients and functional traits of individual species, such as desiccation tolerances of insects. The large amount of variation within groups can perhaps also be explained by different species-level responses to the anthropogenic stressors that come with urbanization, namely light pollution and noise pollution. Because spatial datasets of artificial light at night (ALAN) and anthropogenic noise were used in the creation of our urban intensity layer for this study, we know that these measures of anthropogenic stressors explain much of the variation in our dataset when considered in aggregate (S1 Table). This is unsurprising given the broad impacts of ALAN [87] and anthropogenic noise [88] for many different groups of organisms, including many of the higher-order taxonomic groupings studied here, such as birds (light: [89]; noise:[90]), mammals (light: [87]; noise: [91]), and invertebrates (light: [92]; noise: [93]). These stressors can lead to a variety of responses that are specific to individual species or communities, including some positive responses (e.g., some bats species have increased foraging opportunities as a results of ALAN; [94]). Finally, large within-group variation in UAI values may also be related to the variable, and relatively artificial, taxonomic levels that we used to aggregate data; future studies with more fine resolution groupings (e.g., at the genus level) may reveal more taxonomically conserved and biologically relevant relationships.

Limitations of and future directions for using crowd-sourced data

While this study includes over 500 native species observed within the study area, approximately 59% of the total community science records were excluded from the analyses because they were considered undetectable to the general public (e.g. small insects, nocturnal mammals, etc.), the data were not at the “research grade” level, or the records for a given species were geoobscured due to species status or user preference. Other studies have noted similar data quality issues and biases in iNaturalist records [95–97]. In order to circumvent data quality issues, this study relied on expert review to identify higher order taxonomic groupings that could be reliably identified by the general public, but in doing so, may have increased ascertainment bias and decreased the overall scope of the data. In an effort to reduce the amount of data lost, future assessments of this metric may benefit from developing a relationship with iNaturalist in order to obtain user-obscured data en masse, which is not currently possible without requesting thousands of individual records from each iNaturalist user. This study may also be limited by bias within higher order taxonomic groups for species that are common or more easily observed, as has been reported previously [98,99]). While we were unable to confirm whether any given species is present within our dataset more or less often than “true” occurrence, due to a relationship with their abundance on the landscape or other factors, we believe that this bias should act randomly across the study area and therefore not impact the overall interpretability of our findings. Additional work could greatly benefit the field by investigating the potential over- or underrepresentation of common species within community science datasets.

While we present several limitations to the available crowd-sourced species presence dataset within our sampling area, these data limitations also provide targets for local environmental managers to improve these datasets and therefore biodiversity monitoring in their regions. Based solely on the number of assessed grid cells across the City of Los Angeles in this study, it is clear that there needs to be substantial effort placed on bolstering community science projects that focus on underrepresented taxonomic groups (e.g. snails and slugs and leafhoppers). Findings from other community science projects indicate that local city residents are underrepresented contributors to community science datasets [100], yet for community science-based biodiversity monitoring to be successful, it must be built from a bottom-up approach that includes both participatory and contributory opportunities for the communities where biodiversity monitoring is to take place [101]. While some efforts in Los Angeles to involve residents in taxonomically focused community science projects have led to increased knowledge of urban biodiversity for these taxa groups (e.g. the BioSCAN project, see https://nhm.org/community-science-nhm/bioscan; [102]), these projects are limited in geographic scope, are often short-term assessments, and the primary role of residents is that of data collection, which may not lead to sustained participation by community members in the future [103]. Moreover, efforts to educate the public on specific indicator species is underway in Los Angeles [104], and these efforts could be directed at underrepresented taxonomic groups highlighted in this study. It has been demonstrated that community science datasets can match if not surpass traditional biodiversity assessment methods in data quantity, and do so in a fraction of the time [105,106]. Therefore, developing long-term and mutually beneficial partnerships with local communities to assess urban biodiversity should be a primary focus of city managers who plan to use large unstructured community science datasets to measure the efficacy of city-wide biodiversity measures.