Introduction

Human-wildlife conflict occurs when wildlife and people come into contact with negative impacts to one or both of them. It is a pervasive problem, especially when top predators threaten the food, safety, and economic resources of local stakeholders [1]. Unlike wild prey, domestic animals are highly vulnerable to predation due to low vigilance rates [23–4]. Moreover, they frequently occur in predictable high densities and are easier to catch than wild prey species [5,6]. Livestock consumption by predators and resulting retaliatory or preventive killing are a worldwide conservation concern [789–10]. Persecution of predators on the one hand, and the need to conserve them on the other, create human-carnivore conflict [11].

The most widely studied of such conflicts have involved carnivore mammals, while similar research on other predators is scant, due largely to a series of biogeographical coincidences [12], as the developed nations – where most of the research on human-wildlife conflict is done [11] – are in the northern hemisphere. Because most top predators in northern ecosystems are mammals of the order Carnivora – with a peripheral role for raptors – most published research deals with conflict involving these species. Although this research focus does not reflect the realities of faunal composition in the tropics, it has nevertheless been followed by researchers in tropical regions.

Combine this with taxonomic xenophobia and the disproportionate attention received by charismatic species [13,14], and the result is a near-total lack of research on the role of non-mammalian predators in human-wildlife conflict. For instance, the CENAP (Brazilian government authority responsible for predation conflict) [15] deals solely with carnivorous mammals. However, in tropical regions top predator guilds have different compositions from those in the northern hemisphere, with eagles [16], crocodilians [17], giant snakes [18], and lizards [19] also occupying this role, demanding research on a much broader array of taxa than is currently the case [20].

Conservationists typically assume that people kill predators to reduce predation of domestic animals and/or life-threatening risks. Interviewing local people has been widely used to assess community perceptions of predators and to gather detailed information on human-predator interactions [21222324–25]. Although ecological data are important in conservation decisions, two other aspects are also vital, and often neglected: the attitudes of local people toward wildlife, and the social context. Questionnaires and interviews are frequently used to measure perceptions and attitudes, but their application is limited in space and their wide-scale use would be prohibitively expensive. Socially contextualized information – such as the Human Development Index – can reveal the social variables of such conflicts [26,27], with the advantage that it can be retrieved from published sources.

Although regarded as a nontraditional scientific data source, the internet is creating a revolution in human communication, with effects ranging from massive information spreading on major conflicts [28] to broad availability of biologically useful data [293031–32]. Internet sources can be standardized and used to gather information on human-predator conflicts with hundreds of professional and amateur internet videos containing information on foraging incidents, in detail and on a geographic scale that hardly could be attained by any other method. Analyses of these nontraditional data sources offer an empirical background to human-wildlife conflicts for which there are no published quantitative field data [33].

Boidae and Pythonidae snakes are known to prey on domestic animals [25,3435–36]. Boids, which occupy a variety of habitats and substrates in the Neotropics, are large sized, non-venomous, constricting species whose females are much larger than males [37]. In South America, the two most widely distributed and best-studied semiaquatic species are Eunectes notaeus (Cope 1887), the Yellow Anaconda, reaching nearly 3.7m and 29kg [38], and Eunectes murinus (Linnaeus 1758), the Green Anaconda, reaching 5.17m and 97.5kg [39]. Anacondas are killed in retaliation for livestock consumption and perceived risk to humans, and their fat is widely used for its supposed curative properties [40]. Anaconda leather has commercial value, but only in Argentina is there a sustainable management program for the Yellow Anaconda [41]. Snakes in the genus Eunectes are also wild-collected for the pet trade [42]. All anaconda species are included in CITES-Appendix II and have not yet been assessed under IUCN Red List categories and criteria.

We used internet videos to investigate attitudes of local people toward anacondas and to determine the extent, nature, and effects of human-anaconda conflict, based on the following hypotheses: (i) people are more inclined to retaliatory kill anacondas that preyed on pets or livestock; (ii) people are less inclined to kill anacondas that have eaten recently, assuming they pose less risk; (iii) people are more inclined to kill large-sized anacondas that are perceived to present greater risk; and (iv) anaconda killing is negatively correlated with Human Development Index (HDI). With these hypotheses we characterized the relationship between people and anacondas, and tested whether HDI affects human-wildlife conflict.

Methods

Data acquisition – Internet searches were made on Google Videos between August 2014 and May 2015. To find the videos we used the keywords boiuna, curiyú, mata-toros, sicuri, sucruiú, sucuri, sucuriju, sucurijuba, and yacumama, the vernacular English name anaconda, and the genus Eunectes. The common names used by Spanish, French, English and Portuguese speakers in South America allowed us to recover videos from different South American countries. For each keyword, we searched the first 100 results and collected data from the videos, together with site addresses. We did not include in our sample nature documentaries or videos produced in private captivity, zoological gardens, studios, or wildlife management facilities. We also checked our final database for duplicate videos (e.g., when a conflict event was depicted in a professional video and also in amateur videos). All site addresses are listed in the Appendix. Some of them contain scenes that may be disturbing for some viewers, as they involve animal cruelty.

Data standardization – For each video, we collected information on year, state-level location (Provincia, Departamento, Estado, etc.), and country. We noted the anaconda species, and recorded the prey species if the snake regurgitated or its stomach was cut open. Identification of most prey species was made directly from the videos, by the authors or by consulting taxon experts. If there was a bulge in the anaconda's stomach but prey could not be identified, it was labeled as “unknown.” Anaconda's body size was recorded in one out of five categories, chosen to minimize the error of estimation: less than 1 m; between 2 - 3 m; between 3 - 4 m, between 4 - 5 m, and more than 5 m. To check the error in our estimates, we asked two experienced biologists (Juan Draque and Mariano Barros, who had handled and measured hundreds of Yellow anacondas during an Argentinian sustainable management program) to produce two independent body size estimates of 30 randomly sampled videos. We then collated their estimates with ours as a test of our accuracy.

Since differently sized snakes could have differential chances of being videoed and posted on the internet (i.e., the “macho” effect increases the posting rate for larger snakes) we performed a two-sided Kolmogorov-Smirnov test between our estimated size distribution for both anaconda species and the size distribution of nearly seven hundred Green anacondas from Rivas [39]. We discovered that our distribution is close to reality, under the null hypothesis that the samples are drawn from different distributions (P = 0.38). Finally, we recorded whether or not the snake was intentionally killed in the video. Additionally, we obtained the Human Development Index (HDI) for each State or equivalent, in the same year of the video (or nearest, when there were no statistics for a specific year), using the reports of United Nations Development Program [42434445–46]. We chose the HDI as a proxy for general education level at each location, since it is measured using the same methods all over the world. Besides education, HDI is a statistic composed of life expectancy and per capita income. For political reasons, Venezuela does not allow evaluations by the United Nations and so lacks data on HDI. Therefore, videos produced in that country were excluded from that analysis.

Statistical analyses –We used Fisher's exact test to investigate whether the proportion of anacondas killed when preying on livestock or pets differed from the proportion of anacondas killed when preying on wild prey [hypothesis (i)]. The same test was used to investigate whether the number of anacondas killed with obvious stomach contents (prey item identified or not) differed from the number of anacondas killed while lacking obvious stomach contents [hypothesis (ii)]. To test hypothesis (iii) we used a logistic regression, in which the dependent variable was binary (killed or not), and the independent variable was the body size estimate (categories 1 - 5). For hypothesis (iv) we also used a logistic regression, in which the dependent variable was binary (killed or not), and the independent variable was the HDI for the location of the video. We tested the accuracy of the anaconda's size estimate through a Spearman correlation between our own estimates and each independent series of estimates provided by the specialists. In order to model the stochastic correlation between the two estimates, we measured 10,000 Monte Carlo simulated Spearman correlations between our data and random samples of the specialists' data. All statistical analyses were made on R software using the Vegan package [48], and significance levels were established at 0.05.

Results

We sampled 330 videos filmed between 2002 and 2015, of which 274 were made by amateurs and 56 by professionals, generally from the news media. The majority of the videos were filmed in Brazil (269 videos), followed by Venezuela (14), Paraguay (seven), French Guyana (six), Colombia, Peru and Ecuador (three videos each), Argentina (two), and Trinidad & Tobago and Bolivia (one video each). We failed to establish the country for 19 videos. We were able to determine more precise location information (state level or equivalent) for 196 videos. Samples came from urban areas in 26.97% of the videos and from rural in 6.67%, while the degree of anthropization was not detectable in the remaining samples.

Body size estimates made by the specialist biologists were correlated with our own data, and there was also a correlation between the two independent specialists' estimations (Spearman correlation test, n = 30, ρ = 0.68, 0.63, and 0.57 for correlation between ours and each specialist's estimation, and between the two specialists, respectively). Monte Carlo randomization showed that this is not a product of chance (P < 0.01). We therefore consider that our size distribution estimation is close to reality.

The species depicted in the videos were Green anacondas in 297 samples and Yellow anacondas in 33 samples. No videos of the remaining members of the genus Eunectes – the Dark-spotted Anaconda Eunectes deschauenseei Dunn and Conant 1936, and the Bolivian Anaconda Eunectes beniensis Dirksen 2002 – were retrieved. In 78 videos, the snakes were killing, ingesting, digesting or regurgitating prey after being harassed, and only four of these were Yellow anacondas. In 27 of the videos that involved incidents of predation, prey was already inside the snake's digestive tract or highly digested and could not be identified. Among the prey items figured in the remaining 51 videos there were domestic animals in 23 of them (11 pets and 12 livestock), and wild animals in 28 (Table 1). Fifty-two videos showed people killing anacondas.

Table 1.

Descriptive data from Yellow and Green anacondas (Eunectes notaeus and Eunectes murinus) feeding habits as depicted in 78 internet videos. The percentage for each prey species is represented by % and the number of samples by n. Samples can be found by code on Appendix 1.

The rate of anacondas killed when preying on domestic animals was 21.7% compared to a 28.5% rate of anacondas killed when taking wild prey, but there was no statistical difference (two tailed Fisher's exact test, 50 d.f., P = 0.749). Snakes that had eaten recently (19.23%) were killed more frequently than those that had not (14.68%), but this difference has no statistical significance (two tailed Fisher's exact test, 329 d.f., P = 0.374). The Anaconda's size affected positively the probability of the snake being killed (logistic regression, 329 d.f., z-value = 3.11, P < 0.001). Anaconda killing following an encounter with humans is negatively and significantly related to changes in the Human Development Index of the location where the encounter occurred (logistic regression, 193 d.f., z-value = 2.91, P < 0.001; Figure 1).

Fig. 1.

A – Effect of body size (size classes' estimates, from one to more than five meters; see text for details) on the probability of an anaconda being killed, according to 330 internet videos; B – Effect of Human Development Index (HDI) on anacondas killing probability, in 196 locations for which we obtained information (state or equivalent).

Discussion

Our study shows for the first time that in South America, anacondas can have high rates of predation on domestic animals, which comprised 41.82% of videoed anaconda prey. Capybaras (Hydrochoerus hydrochaeris) were the most common prey (17.8%), followed by domestic dogs (Canis lupus familiaris; 12.3%). Dogs as main prey have been noted in other studies of domestic animal predation by carnivores [49], as well as by crocodilians [50]. Several other domestic animals, such as cattle calves (Bos taurus), chickens (Gallus gallus), ducks (Cairina moschata), and cats (Felis catus) were also recorded being killed by anacondas. However, caution must be used in prey composition interpretation, because our method is likely to under-represent small prey and might also be biased toward domestic prey.

Although predation on domestic animals can lead to conflicts between their owners and predators [51], the rate of snakes killed after foraging on such food items in our sample of videos (15.7%) cannot be explained by food type, since it was lower than the rate at which anacondas were killed while eating wild animals. This phenomenon has been recorded elsewhere, when financial rewards have been given for not killing animals, even when there were confirmed instances of their having consumed livestock [52]. Although domestic animals are frequent items in the diet of anacondas, low retaliatory killing rates do not indicate persecution based solely on livestock consumption. Therefore, could rare predatory attacks on humans (a single occurrence in our sample, besides two defensive attacks) be the underlying cause of anaconda killing?

Current knowledge on human-predator conflict predicts that most killings are retaliatory or preventive [53]. Human-anaconda interactions in South America appear to be more complex, with causal factors normally unconsidered when dealing with mammalian predators. Although the videos did not provide a direct measure of body size of the anacondas, they demonstrated that large (> 5 m) snakes were more likely to be killed, and these are exclusively Green anacondas. While direct attacks on humans are rare, the increase in the killing of larger anacondas suggests that preventive or punitive killing is due to fear and could be one of the main reasons why people selectively kill the larger snakes (see an interesting discussion on the subject in 56).

Fear-based killing is corroborated by the strong influence of the HDI on killing probability. A non-feeding anaconda in a location with a HDI of 0.6 or less is six times more likely to be killed than one that is eating dogs in a location with a HDI of 0.7 or more. This suggests that, despite attacks on humans being rare events, the risk posed by anacondas still occurs in the collective imagination of less developed communities, resulting in the killing of these snakes at any given opportunity. Interestingly, this is contrary to current knowledge on the effects of HDI in conflict, since human-wildlife conflict is commonly associated with food security in less developed nations [54]. Furthermore, as reptiles, anacondas do not share with large cats or other carnivore mammals the same public concerns capable of preventing killing as a result of conflict [55]. Our results therefore add a social component to differences in predator killing probabilities, indicating a strong effect of human development on human-anaconda conflict.

Data for the current study were collected using a method that, although open to further improvements, allows low-cost access to information from a wide geographic range. Additionally, it allows obtaining data that were not biased by conservationist-observer presence. We emphasize that this is not the first time that non-traditional data spontaneously generated by its agents have been used to advance knowledge on conservation subjects [56]. The relationship between predators and people, especially considering the current widespread decline of predators, is a broad topic ready for online investigation.

Rather than definitively answering the questions related to human-anaconda conflict, our discoveries open a field of research rarely explored by tropical biologists. Here we have shown that the perceived risk a reptilian predator imposes on human lives can result in it being killed, even if the actual risk is slim. We encourage future researchers to try to quantify real and perceived losses inflicted on stakeholders by anaconda predation. We also suggest that this should be done for several other tropical predators that are frequently ignored by biologists working in the human-predator conflict field, such as crocodilians, eagles, and other large-sized predators that prey on domestic animals. Since anacondas are charismatic animals frequently depicted in popular books [57], infamous movies, music and mockumentaries, but rarely subjected to research, we expect our work to stimulate further investigation into other aspects of these iconic species.

Implications for conservation

Although lacking evaluation by IUCN, we believe that both anaconda species treated here (the Green and the Yellow anacondas) are not of conservation concern. Both species are flexible in habits, distributed over large areas that include several conservation units, and are capable of sustaining high harvest rates (see historical data from Argentina in 38). Their preventive killing in the context considered here probably has a negligible overall impact on range-wide populations. Our findings should be useful for conflict-related questions, since quantitative data on stakeholders' losses are unknown and conflict with non-mammalian predators is under-researched. Deliberate killing should also be investigated in the other two anaconda species – the Bolivian Anaconda and the Dark-spotted Anaconda – as these have much smaller ranges and are probably exposed to the same kind of persecution described in this study. Their occurrence in still remote areas (Beni river valley in Bolivia and Marajó region in northern Brazil, respectively; 48, 49) certainly accounts for the lack of videos depicting these two species. Since the largest animals are more likely to be killed – which in large constrictors are the females with highest reproductive output [38,39,60,61] – this could generate a conservation problem for these two smaller-ranged anaconda species.

Although many videos depicted anacondas either being killed or being harassed to a degree that led Brazilian authorities to prosecute their authors, several videos show the animals from an ecotourism perspective (see Appendix 1. samples 6, 126, 143, and 187 for instance). This points to a high tourism potential for these large snakes, already noted elsewhere [62]. Although restricted to a few areas of extremely clear water (e.g., 63 for Bonito and 64 for Nobres, both in Midwestern Brazil), ecotourism could be a way to positively incorporate the anacondas into local economies, as the management plan for skin trade is doing for the species in Argentina.