Large herbivore populations can be highly affected by resource availability (i.e. bottom-up) and the predator presence (i.e. top-down). Assessing the relative effect of these forces has been a major topic of interest in ecological studies of this group (e.g. Anderson et al., 2010; Hopcraft et al., 2010; Owen-Smith, 2010; Riginos, 2015). As both resource availability and predator presence may commonly vary in space, the local abundance of large herbivores is expected to be heterogeneous in landscapes. Furthermore, individuals can avail of this heterogeneity to access high-quality resources while avoiding predation (Armstrong et al., 2016; Hebblewhite & Merrill, 2009). Understanding these effects is important for wildlife conservation and management since herbivores are considered the trophic group at the highest extinction risk, with consequences for entire ecosystems (Atwood et al., 2020). The knowledge about bottom-up and top-down effects in large herbivores can be used to predict consequences of global climate changes in forage availability, to foresee possible effects of large herbivore or predator extinctions or reintroductions on ecosystem functions, and to manage exotic invasive herbivores or game species.
Beyond primary productivity itself, a major factor influencing large herbivore populations is the distribution of high-quality forage. Large herbivores usually seek vegetation with high nutritional quality, that is, young and fresh plants, easily digestible and rich in protein. As the plant biomass increases, forage quality reduces, thus herbivores tend to optimally choose intermediate forage biomass to maximize energy intake (Fryxell, 1991; Hebblewhite et al., 2008; Hobbs & Gordon, 2010). In heterogeneous landscapes, the variation in vegetation phenology can sustain high herbivore densities. Especially in seasonal environments, high-quality forage is considered the major factor influencing large herbivores migrations, for example, in African savannas (Boone et al., 2006; Purdon et al., 2018) and mountainous regions of North America (Aikens et al., 2017; Hebblewhite et al., 2008; Jakes et al., 2018; Merkle et al., 2016; Sawyer & Kauffman, 2011) and Eurasia (Bischof et al., 2012; Rivrud et al., 2016). The spatial distribution of herbivores in a specific time would be then strongly determined by plants’ phenological stage, with higher densities in areas showing higher vegetation greenness values.
Predation is also an important process shaping large herbivore populations, usually modulating the effects of optimal foraging preferences (Hebblewhite et al., 2008; Hopcraft et al., 2010; Rivrud et al., 2018). Predation can be a potential cause of local extinction for large herbivores, even when kill rates are low (Festa-Bianchet et al., 2006). Beyond the direct effect of predation in removing individuals from the population (i.e. consumptive effects), prey species can show behavioural responses as they perceive the risk of predation (Lima & Dill, 1990; Say-Sallaz et al., 2019). When exposed to predation risk, large herbivores can increase vigilance behaviour or herd size, and even change used areas or time of activity (Creel et al., 2014). These non-consumptive effects may have different – and sometimes larger – effects in prey dynamics than direct predation (Creel & Christianson, 2008). Thus, the resulting spatial distribution pattern of large herbivore individuals would be a trade-off between foraging gains and predation effects.
Studying large herbivore populations is challenging, especially at large spatial scales and in remote areas. The influence of forage distribution and predators on herbivore individuals has usually been studied using biotelemetry, by evaluating how landscape variables associated with forage availability and predation affect individuals' movements and space use (Hopcraft et al., 2014; Merkle et al., 2016; Owen-Smith, 2010). Population-level studies of large herbivores are commonly carried out using aerial surveys with manned aircraft (e.g. Caughley, 1974; Vucetich & Peterson, 2004), which is often financially prohibitive and can prevent the necessary amount of spatial or temporal replications for accurate and frequent estimates needed for decisions in conservation (Ferreira & Aarde, 2009; Fritsch & Downs, 2020). Recently, drones (Unmanned Aerial Systems or Remotely Piloted Aircraft) have emerged as an accessible, safe, and cost-effective alternative for aerial surveys and have been tested to sample large herbivore populations (e.g. Barasona et al., 2014; Chrétien et al., 2016; Gentle et al., 2018; Linchant et al., 2018; Rey et al., 2017; Vermeulen et al., 2013).
As for other wildlife survey methods, aerial count data obtained from drone-based surveys are susceptible to imperfect detection. Some individuals present in the sampling area may be unavailable for detection (e.g. under a tree) or, even when available, an observer can fail to detect them when reviewing imagery (Brack et al., 2018). Addressing these sources of error in count data with a robust framework is imperative to obtain unbiased abundance estimates. Given the potential of hierarchical N-mixture models to estimate abundance accounting for imperfect detection and without the need of marking individuals (Royle, 2004), they have been proposed as a feasible approach for aerial surveys (Brack et al., 2018; Williams et al., 2017), especially for modelling abundance at large spatial scales (e.g. Martin et al., 2015). Moreover, the typical sampling design for N-mixture models matches the characteristics of drone surveys, that is, multiple and short flights in spatiotemporally replicated surveys.
Our objective here was two-fold: i) explore the potential of drone surveys and hierarchical N-mixture models to estimate large herbivore abundance and ii) use this innovative approach to evaluate the relative influence of bottom-up (forage and water) and top-down (predators) factors on the local abundance of marsh deer (Blastocerus dichotomus). The marsh deer is the largest cervid in the Neotropics (up to 150 kg), highly adapted to use marshes and swamps, living in low densities in South America’s wetlands and savannas. Because of its association with wetlands, marsh deer’s current distribution is highly fragmented and the species is listed as Vulnerable with extinction (IUCN; Duarte et al., 2016). Its main predator is the jaguar (Panthera onca) and, although marsh deer are considered a valuable prey for jaguars (Hayward et al., 2016), they only compose ≤ 10 % of the jaguar’s diet (Cavalcanti & Gese, 2010; Azevedo & Murray, 2007; Perilli et al., 2016). Here, we specifically tested a positive influence of (1) vegetation greenness (as a proxy for high-quality forage availability) and (2) water bodies on the local abundance of marsh deer, and a negative effect of (3) jaguar density. We also tested a modulatory effect of jaguar density on the relation of deer local abundance with vegetation greenness, expecting higher jaguar densities to be associated with weaker positive effects of forage. We assessed these predictions during the dry season of the Pantanal wetland of Brazil. Additionally, we discuss the feasibility of spatiotemporally replicated drone surveys applied here, and compare two hierarchical N-mixture modelling approaches (single and double observer protocol) for estimating the abundance of large herbivore populations, which can also be employed for other wildlife species in different contexts.