Large-scale assessment of commensalistic–mutualistic associations between African birds and herbivorous mammals using internet photos

Data searching

To collect a large dataset of spatially and taxonomically distributed data on bird–mammal associations, we did an extensive internet search for photos on Google Images. Jarić et al. (2016) pointed out that internet searching based on English common names produced more results than when scientific names were used. Moreover, since results of searches using English and scientific names were highly correlated, we decided to search only for English names of birds and mammals, although this could restrict the geographical coverage and decrease the use of records from non-English-speaking countries. To capture relative frequency of each bird–mammal association, our search phrase typically contained the name of the bird species combined with the name of the species/genus of larger-bodied African mammal herbivore (>10 kg). Our list of taxon names was based mainly on association reports reviewed by Dean & MacDonald (1981), but we also included searches for associations not reported by them; for the complete index (see Appendix S1). If no bird visitors were recorded for some mammal species, we repeated the search using a more general “bird” or “birds” term to find whether, at least in some cases, these mammals were visited by birds. We also used this searching phrase for species where few interactions had already been found by a word combination search. However, we used only results revealing new, typically rare, bird–mammal associations, hence avoiding significant bias in the search in favour of common or well-recognized associations.

The Google searches for photos for each combination of bird and mammal taxa were conducted separately, and for each combination we aimed to collect as many photos as possible until the search produced only a small proportion of photos with relevant content. For common species it is virtually impossible to collect all available photos, so this solution represented a trade-off between the number of available relevant photos and the time spent searching for new photos; however, we consider that the proportion of available photos sampled was similar in all species. This procedure standardized our data for analysis, increasing possibility that variations in the frequency of internet photos in our dataset may reflect proportional differences in real animal abundance and/or extent of spatial distribution. However, the potential influence of the “charisma” of a species on its appearance on the internet requires cautious interpretation. We only analyzed photos in which birds were in direct contact with the bodies of a mammal, excluding cases where the birds were only feeding or flying near the mammal. We did not include photos of mammals without birds in our data set, even if such individuals were visible in photos.

We focused only on free-living, non-domesticated mammal species in sub-Saharan Africa. We also excluded photos where birds were observed on captive African mammals outside Africa (e.g., zoos). When photos were part of a series, we chose the one showing the highest number of associated birds/mammals. Paintings and photos which were suspected to be photomontages were ignored (<1% of all photos). To limit other sources of bias, photos suspected to be shared by multiple sources were briefly checked to see whether they had already been included (all photos were collected exclusively by one author, PM, enabling us to do this consistently). We were particularly careful when working with unusual photos that people might prefer to share, e.g., a mammal individual covered by a large number of birds or “cute” or interesting animal species and scenes. However, it is still possible that a small number of duplicates remained undetected because we did not cross-check all possible combinations of photos. On the other hand, such cases were quite rare, and it seems that, for the volume of photos we collected, online sharing of photos would not substantially bias results obtained from a Google search compared to field data. Similarly, Leighton et al. (2016) found that the proportion of black colour morphs in black bear subspecies collected from a Google search was highly correlated with that from fieldwork, suggesting that online photos do not substantially over-represent bears with atypical colouration in particular subspecies.

Although there could be an inherent bias toward photos with larger numbers of birds on mammals caused by a preference for photographers to publish such photos, this should apply to all species and not affect relative differences. Even if only some photographers did this, we again do not expect any consistent bias in our data because photos originated from numerous authors. Furthermore, the probability of photographing a bird on a mammal may be related to the habitat structure (e.g., dense versus open habitats) and may underestimate involvement of species living in more closed habitats. Because a substantial proportion of the photos came from amateur photographers we expected less of a bias toward rare species or other unusual occurrences, as compared to professional photographers or researchers (see also Dylewski et al., 2017).

To investigate the spatial patterns in our dataset, we included only photos with the location given to at least country level. For each record, we specified the geographical location as accurately as possible. When the location was only at the country level, coordinates were taken as the centre of the mammal and/or bird species distribution in that country (accessed on http://www.iucnredlist.org and http://www.hbw.com).

Bird and mammal characteristics

Each individual host mammal with birds observed on it represents a basic unit which was assessed separately. In the case of photos showing several mammals, each with one or several birds, each mammal individual was scored as a separate case (such cases represented <15% of all cases). Altogether we collected information on three response variables and three predictor variables:

(a) Response variables

Mean number of birds per mammal individual (hereafter referred to as “number of birds”): We counted the number of individual birds sitting on each individual mammal. For each mammal species, we then calculated the mean number of birds sitting on them.

Mean total mass of birds per mammal individual (hereafter referred to as “mass of birds”): The mean body mass of each bird species associated with individual mammals was obtained from the online edition of “Handbook of the Birds of the World” (HBW Alive, 2016). We used mean body mass for nominate subspecies and did not distinguish between sexes. For each mammal species, we then calculated the mean mass of birds sitting on them.

Total number of bird species per mammal species (hereafter referred to as “number of bird species”): We also collected information on the total number of bird species on each mammal species across the entire pool of photos in order to assess overall bird species richness hosted by them.

(b) Predictor variables

Mammal body mass: Information on body mass (in kilograms) of each mammal species was taken from the online “Encyclopedia of Life” (EOL, 2016). We used mean body mass for nominate subspecies and did not distinguish between sexes.

Mammal herd size: Herd size was estimated as the number of visible mammal individuals (conspecific or heterospecific) in each analysed photo. For each mammal species, we then calculated mean herd size.

Habitat openness: We defined the surrounding habitat for each photo and subjectively scored it into four distinct classes of openness, from 1 for most open habitats to 4 for very closed habitats: (1) near water (e.g., swamps, lakes and other water sources with typically very open vegetation cover), (2) open habitats (e.g., grassland, semi-deserts, open savannahs), (3) higher mosaic vegetation cover (e.g., woodland savannah, shrubland and bush) and (4) higher dense vegetation cover (e.g., woodland and forest). Then, we calculated mean values of habitat openness for each mammal species. We did not assess habitat from highly magnified (zoomed) photos because the identification would not be reliable.

Photo zoom

Because we used several variables which are expected to be strongly influenced by photo zoom, all analyzed photos were scored according to their zoom on a three-point scale: (1) very zoomed photos (only part of mammal body was visible, e.g., head or hind legs with part of the belly), (2) medium zoomed photos (complete or almost complete mammal body was visible and free space on the photo constituted less than one mammal body length on each side), and (3) unzoomed photos (complete mammal body was visible and free space of more than one body length on each side was present).

Statistical analysis

Taxonomic diversity and spatial distribution of bird–mammal associations

In total, we collected information on 2,169 bird–mammal associations of 4,840 individual birds, belonging to at least 48 bird species of 21 families, with 31 species of wild-living African mammals of seven families. This dataset contains records from regions across sub-Saharan Africa with the majority of records from the open and relatively well studied areas of East and Southern Africa (Fig. 1). Only a small number of records came from West and Central Africa (<2%).

Bird–mammal association web

We included 2,147 associations in web analyses where both bird and mammal interactors were identified to species level (data in Appendix S2). These associations represent 123 different association types (i.e., combinations of different bird and mammal species). Of these, 66 (53.7%) association types were not reported by Dean & MacDonald (1981) and some may represent new associations, previously not reported in literature (see Appendix  S2).

Of all cases, 672 cases (31.3%) included birds other than oxpeckers, detected on 18 species of mammals (Fig. 2A). These included cattle egret Bubulcus ibis (51.5% of non-oxpecker cases), wattled starling Creatophora cinerea (8.9%) and piapiac Ptilostomus afer (4.6%). In all records, the most visited mammals were common hippopotamus (31.0%), followed by plains zebra Equus quagga (13.1%) and African elephant Loxodonta africana (11.5%) (Fig. 2A).

We also found 1,475 cases (68.7%) that included oxpeckers: yellow-billed oxpecker (407 cases, 27.5% of oxpecker records) was observed on 16 mammal species whereas red-billed oxpecker (1,068 cases, 72.4%) was observed on 24 species of mammals (Fig. 2B). Yellow-billed oxpecker was most often associated with African buffalo Syncerus cafer (35.9% of all species-specific cases), giraffe Giraffa camelopardalis (13.5%) and hippopotamus (11.3%). The mammal species most often visited by red-billed oxpecker was impala Aepyceros melampus (18.6%), followed by giraffe (13.9%) and white rhinoceros Ceratotherium simum (13.3%) (Fig. 2B).

When all species were analysed together, we found that the association web between birds and their mammal hosts had rather low level of nestedness (NODF = 24.66) and did not differ significantly from values expected under the null model (p = 0.77). When separate analyses for oxpeckers and for the remaining species were carried out, we found that web nestedness was higher in oxpeckers than in the other species and differed significantly only for oxpeckers (NODF = 32.55, p = 0.017; other species NODF = 10.63, p = 0.999).

Relationship between bird, mammal and environmental characteristics

In the full set of species (N = 31 species of mammals), the PGLS model analysing relationships between the number of birds and mammal and environmental characteristics revealed a positive correlation only with mammal body mass (model log likelihood = −3.977, λ = 0.663) (for complete results see Table 1; data in Appendix S3). The mass of birds was positively correlated with mammal body mass and herd size, and a higher mass of birds was also associated with more open areas (full model statistics: log likelihood = 5.653, λ =  − 0.967). The number of bird species was positively correlated with mammal herd size, and more species were recorded in open areas (log likelihood = −42.769, λ = 0.501).

In non-oxpecker bird species (N = 19 species of mammals), we found that the number of birds was higher in closed habitats (log likelihood = −1.406, λ = 0.974). The mass of birds was positively correlated with mammal body mass and herd size, and a higher mass of birds was also associated with more open areas (log likelihood = 2.920, λ = −0.539). A higher number of bird species was also associated with more open areas (log likelihood = −22.630, λ = 0.634). In oxpeckers (N = 26 species of mammals), we found a significant relationship between the mass of oxpeckers and mammal body mass (log likelihood = −123.685, λ =  − 0.852).

Structure of bird–mammal association web

Visualization of the bird–mammal association web revealed that the majority of bird species were associated with larger-bodied herbivores, such as common hippopotamus, African buffalo, plains zebra and African elephant, that occupy mainly open habitats (see mammal body and habitat openness scores in Appendix S3). We found that ∼60% of bird species in the data set were associated with water or were near water ecosystems with many of them recorded sitting exclusively on common hippopotamus. The most common bird interactor was cattle egret which is renowned for its widespread cooperative behaviour with African mammals (Dean & MacDonald, 1981; Ruggiero & Eves, 1988; Kioko et al., 2016; Goodale, Beauchamp & Ruxton, 2017). It is possible that the higher diversity of birds on larger-bodied mammals may be linked to the higher load-carry capacity of these mammals (hence, they can carry both smaller and larger bird species), and they can also provide more feeding opportunities, e.g., by flushing more prey (Wahungu, Mumia & Manoa, 2003; Kioko et al., 2016). Obviously, habitat type where particular bird and mammal species co-occur also has an important effect on bird–mammal association webs, facilitating some associations while limiting others (Heymann & Hsia, 2015; Kioko et al., 2016).

We found that obligate mammal mutuals, oxpeckers, visited a higher number of mammal species compared with the rather casual mammal-associated bird mutuals in the Neotropical region (Sazima et al., 2012). In agreement with previous field studies, the analysis of internet photos showed that oxpeckers were very often associated with larger-bodied mammals (Mooring & Mundy, 1996; Koenig, 1997; Nunn et al., 2011; Ndlovu & Combrink, 2015). Yellow-billed oxpecker was most often associated with African buffalo, giraffe and hippopotamus, while red-billed oxpecker was most often associated with impala, giraffe and white rhinoceros. With the exception of impala, the species mentioned are among the largest herbivores in Africa (Ripple et al., 2015). Although smaller mammals have a higher tick number to body mass ratio (Hart et al., 1990), potentially increasing the efficiency of tick harvesting, it seems that the absolute number of ticks and their abundance plays a more important role (Mooring & Mundy, 1996; Koenig, 1997; Nunn et al., 2011). Oxpeckers were identified as very effective tick removers from body parts that are inaccessible to self-grooming by host mammals, suggesting that they play an important role in tick control in the hosts (Mooring et al., 2000; Bezuidenhout & Stutterheim, 2009; Nunn et al., 2011; Ndlovu & Combrink, 2015). Furthermore, Ndlovu & Combrink (2015) suggested that oxpeckers may prefer larger-bodied ungulates because larger mammals can provide a more stable platform upon which to forage, or are just large enough to support simultaneous feeding of more oxpecker individuals. However, it is still possible that some variability in this pattern may be explained by human preferences for large mammals, resulting in an overrepresentation of photos of large animals on online platforms (Hausmann et al., 2017). Our results also indicate that yellow-billed oxpecker visits a lower number of mammal hosts than red-billed oxpecker. This is in agreement with field studies showing that, even in localities where the distribution of both species overlaps (i.e., the spectrum of potential mammal hosts should be the same for both species), red-billed oxpecker has a wider range of hosts (Ndlovu & Combrink, 2015). However, we collected more photos of red-billed than yellow-billed oxpecker; thus, observed large-scale differences may be due to differences in sample sizes.

The nested structure of mutualistic networks is usually interpreted as an asymmetric specialization (i.e., specialists are associated with generalists rather than other specialists). The concept of nestedness has important consequences for ecological (Bastolla et al., 2009) and evolutionary (Bascompte et al., 2003) principles of biodiversity maintenance. In contrast to earlier results on a similar bird–mammal system in a Neotropical region (Sazima et al., 2012), our web structure for African birds and mammals from internet photos was only weakly nested. A relatively weak nested web structure, even for mutualistic relationships between oxpeckers and African mammals, compared with that between Neotropical birds and mammals (Sazima et al., 2012), could be explained by differences in diversity and spatial distribution patterns of large herbivores between both regions. In Neotropical region, only limited diversity of large herbivores is available, with few dominant and widely distributed species (e.g., domestic animals; see also Sazima et al., 2012); hence, bird communities visiting rarer species typically represent only a subset of bird communities visiting common species. In contrast, in the Afrotropical region, diversity of large herbivores is still relatively well preserved, with proportionally more widely distributed and abundant species than in Neotropical region. Therefore, it is possible that each mammal species is associated with own bird fauna, which is unique and species-specific, rather than being a subset of the bird species that can be found on some other mammal species, resulting in decreased level of observed web nestedness. However, it is noteworthy that commensalistic and mutualistic associations of birds with African mammals are even more common and include many more bird taxa than we reported (Dean & MacDonald, 1981). Our study was limited by the use of photos as information sources which enabled us to classify the exact activity of birds on mammal bodies, e.g., whether they are only sitting or also collecting some organic debris from the host. In mainly commensalistic birds, the level of nestedness was even lower and, moreover, nestedness did not differ significantly from a null model indicating that birds do not prefer mammal species which are visited by a large number of bird species. In oxpeckers, however, difference from a null model were found, suggesting that a nested structure may be the result of a non-random assignment of birds to their mammal hosts.

Mammal and environmental correlates of bird visitation

We found that mammal body mass positively influenced the number and mass of birds observed sitting on mammals in the full set of species (i.e., taking oxpeckers together with other bird species). This may be simply explained by the fact that the size of larger mammals can support a larger number of birds and/or birds with a higher mass which can drive a positive relationship between bird and mammal mass (Kioko et al., 2016). The latter explanation seems to be relevant, especially for the non-oxpecker set of species where mammal body mass was only significantly associated with the mass of birds. The majority of non-oxpecker birds associated with mammals were waterbirds, often heavy species when compared with, for instance, songbirds (Maurer, 1998). Finally, we found a positive correlation between mammal body mass and mass of oxpeckers. Oxpeckers live in small family groups (Van Someren, 1951) and large mammals may support the simultaneous feeding of larger family groups or other unfamiliar individuals on the same mammal individual.

Furthermore, our analysis revealed that larger mammal herd sizes were associated with a higher mass of birds and more bird species in the full set of species, and with bird mass in the set of non-oxpecker species. This may suggest that mammals in larger herds provide more feeding opportunities by flushing more potential prey, such as insects or small vertebrates, and, hence, attract a wider community of birds (Wahungu, Mumia & Manoa, 2003; Kioko et al., 2016). However, our finding of a relationship between mammal herd size and the number of associated bird species must be interpreted carefully. Despite weighting our regression analysis by the number of available photos per mammal species, such a relationship may be still mainly a function of a strong correlation between the number of bird species and sample size.

Habitat openness influenced the mass of birds and the number of bird species sitting on mammals; both were higher in more open habitats in the full set of species as well as in non-oxpecker species (see also Kioko et al., 2016). As mentioned above, the majority of diversity of bird species included in our study was represented by larger-bodied birds associated with water sources and relatively open country. We suggest that such patterns are most probably driven by habitat overlap between water- and open-country-associated larger-bodied birds and their mammal host because, similarly to birds, large African herbivores are more common in open areas than in closed habitats such as forests or woodlands (Anderson et al., 2016). Interestingly, the number of non-oxpecker birds was higher in more closed areas, probably because such habitats can provide more perching opportunities.

Advantages and limitations of an internet search on the diversity of bird–mammal associations

We highlight the potential role of information technologies and new media, such as internet search engines, for further studies in ecology and evolution. Google Images represents one such resource that could facilitate a rapid collection of information on various aspects of ecological systems at large spatial and taxonomical scales. Results obtained from the analysis of internet images can be in good agreement with data from fieldwork, and such an approach could therefore, at least in some cases, supplement or replace less effective and time- and money-consuming fieldwork (e.g., Leighton et al., 2016; Mikula et al., 2016). While we revealed many previously described associations, our approach also identified many novel associations, e.g., that birds were also occasionally associated with sitatunga Tragelaphus spekii and waterbuck Kobus ellipsiprymnus, and that red-billed oxpecker also feeds on much smaller hosts than previously reported (including Thomson’s gazelle Eudorcas thomsonii weighing ∼20 kg) (for some details see Appendix S2; Dean & MacDonald, 1981; Hart et al., 1990; Feare & Craig, 2010; Ndlovu & Combrink, 2015). Moreover, similarly to Dean & MacDonald (1981), we did not find any associated birds for several mammal species that we investigated including, e.g., beira Dorcatragus megalotis.

On the other hand, we were not able to find any evidence for several previously described associations, e.g., between chorister robin-chat Cossypha dichroa and nyala Tragelaphus angasi and bushbuck Tragelaphus scriptus (for other cases see also Dean & MacDonald, 1981). This may indicate limitations of our approach which may be the result of species- and region-specific public bias towards common or “charismatic” species of birds and mammals and to regions with a good infrastructure (Clucas, McHugh & Caro, 2008; Hugo & Altwegg, 2017; Troudet et al., 2017). Alternatively, the absence of some associations among the results of the internet search may be caused by our searching method because only English genus names were involved in the search, hence, overestimating associations for more common species. Finally, a sharp population decline or even extinction of large herbivorous mammals, mainly because of human-induced pressure, in many world regions, including Africa (Ripple et al., 2015), disrupts and reshapes present ecosystem services and ecological associations, including bird–mammal associations (Galetti et al., 2017; Hempson, Archibald & Bond, 2017). This may cause several bird–mammal associations to have been lost over time, resulting in decreased diversity of mutual associations. On the other hand, it is possible that other species which used to interact with herbivorous mammals have established new associations with mammals which are still common (Galetti et al., 2017). For instance, oxpeckers are able to behave plastically and can shift their host selection to other available herbivores, including domestic mammals in regions where they are common (Dale, 1992; Feare & Craig, 2010).