A citizen science approach to evaluating US cities for biotic homogenization

Shared biodiversity between cities

We identified which species were found in the majority of the cities to compare these widespread species with the total pool of observations. We also divided the dataset by major taxa: four plant groups (monocots, dicots, ferns, and conifers) and six animal groups (birds, insects, reptiles, amphibians, mammals, and gastropods), such as to allow for better comparisons between similar taxa. To capture observed species from groups that had insufficient observations on their own (e.g. isopods, fungi, arachnids), we also created a catch-all “other” category.

Biotic homogenization with increasing urban intensification

After seeing how biodiversity was shared between cities, we asked whether the biotic homogenization effect was stronger with increasing urbanization intensity. Based on geographic coordinates, we linked all observations with a NLCD2011 land cover classification from the Multi-Resolution Land Characteristics Consortium (MRLC). Assessed nationwide at a 30 × 30 m resolution, every pixel is assigned one of 16 land cover classifications, four of which are forms of developed land with increasing urbanization intensity (developed-open space, developed-low intensity, developed-medium intensity, developed-high intensity; further details in Table 1). We collapsed the remaining land cover classifications into “water”, “agricultural”, and “natural”. As we were only interested in comparing increasing levels of urbanization against the natural land use type, we excluded any observations that were classified as having occurred within agricultural or water pixels.

We then analyzed the relative influence of level of urban intensification and city on community composition. To do this, we built Bray–Curtis dissimilarity matrices based on the species composition at each level of urbanization within each city, and visualized community composition using non-metric multi-dimensional scaling (NMDS) with 100 restarts. We applied a stress cut-off of 0.20; if stress was >0.20, we considered the NMDS plot to be unreliable (Quinn & Keough, 2002). We visualized groupings both based on land cover type and by city.

As regional location can be an important environmental filter in determining community composition (Williams et al., 2009; Aronson et al., 2014; Pearse et al., 2018), we also created NMDS plots for regional groups in a series of triads of increasing geographic distance. Specifically, we focused on a Texas group (Houston, Dallas, and Austin), Atlantic Coast group (New York City, Boston, and Washington DC), Pacific Coast group (Seattle, Los Angeles, and San Francisco), and a fairly widespread Central United States group (Salt Lake City, Minneapolis, and Chicago) (Fig. 1).

To examine whether community composition becomes more similar with increased urban intensification, we subdivided observations based on their land cover classification (natural, developed-open space, developed-low intensity, developed-medium intensity, developed-high intensity). We then looked for the effect of regional location (with three city “replicates” for each region as above—Raleigh and Miami were excluded from this analysis because they did not fall neatly into one of the other regional categories). We built a PERMANOVA (Permutational Multivariate Analysis of Variance, (Anderson, 2017)) model for each land cover group with 999 iterations based on Bray–Curtis dissimilarity (R package vegan (Oksanen et al., 2015)), then compared the R², p-value and AIC score for each of the models generated by the five different land cover classifications. We would expect that if biotic homogenization were occurring with increased urban intensification, the models built off of the observations from the more developed land cover types would perform less well because the effect of regional location should be reduced.

Shared biodiversity between cities

We analyzed 66,209 citizen science research grade iNaturalist observations across 14 US metropolitan areas. Overall, dicots, the largest plant group, were overwhelmingly the most observed (59.6%) and had the most species (52.4%). The next most observed groups were birds (12.8%), monocots (8.7%), and insects (8%). However, despite making up only 8% of the observations, insects actually made up 18.4% of the total species richness. Birds, on the other hand, made up only 7.8% of species richness, meaning they have a higher proportion of number of observations per species.

Of the 5,209 observed species, exactly 100 were found in the majority (eight or more) of the cities (Table 2), which we hereafter refer to as our “cosmopolitan” species. While the cosmopolitan species were primarily birds and dicots (36 each), and a few mammals (seven), insects (seven), and reptiles (four), there was also one cosmopolitan species each for amphibians, monocots, and conifers, and no representative species for gastropods or ferns. Although only 1.9% of the total species richness, these widespread cosmopolitan species made up 21.4% of the total observations. Two birds, the rock dove and American crow, were the only species observed in each of the 14 cities. Ten additional species were observed in 13 cities each—seven of which were also birds (red-winged blackbird, mallard, great blue heron, turkey vulture, house sparrow, American robin, and mourning dove), but also one dicot (common dandelion), one insect (Asian lady beetle), and one mammal (common raccoon).

Taxa varied in how cosmopolitan (again, here defined as being found in the majority of our cities) they were as a group. Mammals and birds had the highest proportions of cosmopolitan species (10.6% and 10.1% respectively). On the opposite end of the spectrum, insects and dicots had a much smaller proportion of their species observed in the majority of cities (0.83% and 1.5% respectively). Our findings that cities comprise a few cosmopolitan species with a mix of many local species complement other findings that the majority of urban species are still local species (Aronson et al., 2014).

However, these cosmopolitan species accounted for the majority of observations for mammals (55.2%) and birds (64.8%), and even made up a large proportion of observations for insects (25.3%) and dicots (15.7%). While it is possible that these patterns could also be explained by cosmopolitan species being more recognizable to people (and therefore more frequently identified, leading to an inflation in the proportion of observations for these groups), the substantial proportion of cosmopolitan species could also be indicative of a downward trend of the relative abundance of native species populations in cities. Previous multi-city studies of biotic homogenization have relied on species lists (Aronson et al., 2014), which cannot capture shifts in community proportions. With mass species declines in tropical and temperate ecosystems (Hallmann et al., 2017; Lister & Garcia, 2018), such findings of cosmopolitan species making up such a large portion of the community relative to native species merit further investigation.

Biotic homogenization with increasing urban intensification

We next asked whether the effect of biotic homogenization grows stronger as a landscape becomes more developed through urbanization. The clustering in our NMDS plots suggests that urban biodiversity is to some degree city specific but also tied to particular levels of urbanization (Fig. 2). Plants exhibited a slightly different pattern from animals (Fig. S1), with the plant communities observed in the highest levels of urban intensification having the greatest differentiation, opposite to the pattern that would be expected if urban homogenization were occurring. This contrasts with a previous study that found that across cities, cultivated yards tended to be more similar to one another compared to the similarity of their associated natural areas across cities (Pearse et al., 2018), which could be due to being unable to differentiate between cultivated and spontaneous vegetative growth observations, and the iNaturalist platform discourages the recording of cultivated plants and animals.

We found that communities, regardless of level of urban intensification, within the same city were found close together on the NMDS plots—a pattern further reinforced by region (Fig. 2B). For example, all three Texas metropolitan cities (Houston, Dallas, and Austin) were grouped near one another, as were the cities along the Atlantic (Boston, New York City, and Washington D.C.) and Pacific Coasts (Seattle, San Francisco, and Los Angeles). Miami, being more geographically isolated and environmentally distinct than the other cities was relatively far on the plot from the other cities. Such findings complement what we found on the between cities comparison, where urban communities are largely a reflection of the local regional community, with a few cosmopolitan species. This regional clustering was found for both plants and animals. Animal communities overall were more similar between cities than plant communities, perhaps because of their mobility and ability to respond relatively quickly to land cover changes.

In the regional triad NMDS plots (Fig. 3) which peeled away some of the environmental variations between cities, community composition showed overlap between the different levels of urbanization in an ordered way along the urbanization spectrum, in that more similar levels of urbanization also share more similar communities. In all four regional groups, community composition from high-intensity urbanization were more distinct than those from all other land cover types—even more distinct than those from natural were from the least developed areas. For the Atlantic and Pacific Coast cities, there appeared to be a longitudinal gradient, with the cities falling in the geographic middle (New York City and San Francisco, respectively) having all of their land cover community compositions falling between the community compositions of cities that were more north and south. The distinctness of communities from each land cover type was more evident in those triads that have cities that are geographically closer to one another. In other words, as environmental context becomes less variable, levels of urbanization become more important in defining the community composition.

As predicted, the PERMANOVA models (for all observations, plants only, and animals only) built from observations from high-intensity land cover performed the poorest (Table 3), meaning that regional group membership was less able to predict community composition in higher intensity land cover than it could in the other land cover types. This is consistent with what we would expect to occur if biotic homogenization increases with urbanization intensification. However, the effect of regional group is still significant (p < 0.001) even for high-intensity land cover observations. Additionally, while the PERMANOVA models built from the high-intensity land cover observations appear the weakest (based on AIC and R²), the models based on observations from natural and the other developed land cover types did not appear to decrease in strength in an ordered way with increasing urban intensification.

Additional observations

Many species demonstrated a preferential association for either natural or high-intensity urban areas across all the cities they were found in. In general, we found that those species that favored higher intensity urban land cover tended to be non-natives, having origins in Europe, North Africa, and South Africa (e.g. common dandelion, white clover, common ivy, house sparrow, rock dove, common starling). Conversely (and expectedly), those that were found to favor more natural sites are native to North America (e.g. poison ivy, Virginia creeper, northern cardinal). However, it was difficult to identify specific ecological traits that urban specialists shared, as has been a similar finding in other urban ecology studies (Duncan et al., 2011).

Among the widespread cosmopolitan species we identified in the between cities comparison, we expected there to be a preferential association for the higher intensity land use types. There were in fact several species that showed this pattern—such as the house sparrow and rock dove. However, just as many widespread species favored the less disturbed natural land cover types—such as the white-tailed deer. It seems there are multiple human-associated mechanisms that act at different scales. Human transportation networks, as well as agriculture and other human-directed habitat shifts have facilitated species introductions and expanded species ranges, while urbanization has created unique habitats that allow particular species to thrive. While humans are a common denominator, species that benefit from range expansions do not necessarily also benefit from urbanization.

The western honey bee is an example of a species that varied greatly in which land cover type it favored—it was most frequently observed in the highest intensity urban land cover types in Washington DC and Los Angeles, the natural land cover types for Austin, and somewhere along the urbanization spectrum for everywhere else. The honey bee was found in every city except Minneapolis and Seattle, and was most frequently observed in cities in Texas and California. Pollinators, and honey bees in particular, have been shown to be sensitive to climatic differences (Gordo & Sanz, 2006; Bartomeus et al., 2011), and the varying environmental conditions between cities in April could explain why the honey bee was not found in the two northernmost cities and most abundant in the more southern ones. Further, the “snapshot” approach of the City Nature Challenge captures cities at different points in their seasonal progression, as bee abundance phenology is known to vary between land cover types (Leong et al., 2016).

Many frequently observed species are also invasive species—such as garlic mustard. While originally introduced to North America from Europe, it thrives in the forest understory (Stinson et al., 2006). It was particularly abundant in Boston, New York, and Washington D.C., where it was found across all land cover types. Because there are many ongoing efforts to control this species (Nuzzo, 1999; Blossey et al., 2001), it will be important for land managers to consider that urban landscapes could also act as reservoirs maintaining sizeable populations of this species.

Our methodology utilizes within city and land cover type community composition and ranking metrics to avoid biases based on “collecting effort.” However, there remain other challenges in teasing apart patterns reflecting ecological dynamics and natural history versus artifacts associated with data collected opportunistically by members of the public that currently limit ways in which we can interpret our findings. For example, species with the most observations are often not truly the most abundant species in cities, rather they are the easiest to photograph and identify (hence, the “overrepresentation” of bird taxa). Insects and other small taxa that are more difficult to photograph and identify are almost certainly under-recorded. Many species were rarely observed—2,435 of the 5,209 total species included in the dataset were singleton/doubletons, meaning they were only observed once or twice. Although we can assume that most species should be relatively equally photographable and identifiable across land cover types, we recommend using multiple approaches to make comparisons “within the biases,” such as focusing on community composition and nonparametric statistical methods as we have done here.