Emerging field of human-wildlife conflict management

Get Complete Project Material File(s) Now! »



Globally, human-wildlife conflict (HWC) is an issue of growing concern to conservation biologists (Messmer 2000; Dickman 2010; Manfredo 2015). HWC occurs where the activities and requirements of humans and wildlife overlap, resulting in direct and indirect negative impacts to one or both sides (ICUN 2003). Human population growth and the accompanying decline in ‘wild spaces’, combined with the restoration of some wildlife populations, has led to an increase in the frequency and severity of HWC; a trend that will likely continue to escalate (Madden 2004; Thirgood & Redpath 2008; Dickman 2010). From a conservation perspective, conflicts between humans and wildlife are of particular concern when and where they lead to the harm or death of individuals of threatened or protected species. Taxa subjected to persistent conflict(s) with humans are more vulnerable to population decreases, which ultimately may lead to their (local) extinction (Messmer 2000; Ogada et al. 2003). Increases in mortality as a direct or indirect consequence of human activity not only affect the population viability of endangered species, but also have broader environmental impacts on ecosystem dynamics and the preservation of biodiversity (Woodroffe et al. 2005).
Quantifying the extent of interactions between humans and conflict-prone wildlife is essential for predicting and mitigating future negative encounters (Graham et al. 2009; Gubbi 2012). The type and duration of human-wildlife interactions can vary markedly in space and time because such interactions are the outcome of dynamic abiotic and biotic processes. In Nepal, habitat fragmentation has pushed tigers into close proximity to human areas where they alter their natural activity patterns to avoid periods of high human activity (Carter et al. 2012). One example of how biotic factors influence the likelihood of HWC is reproductive state: during musth, male elephants have heightened aggression levels, which can induce them to attack humans (Das & Chattopadhyay 2011). Effective management of HWC requires understanding how animal populations respond differentially to human presence (e.g., neutral, repulsed, or attracted). The response-continuum to humans is both contextual and scale-dependent, varying among individuals based on their experience and innate tendencies (Dall et al. 2004). Some animals, for example, are strongly attracted to human areas, and this behaviour elevates their susceptibility to conflict with humans (Jäggi 2008). Despite the importance of understanding how animals and humans interact, little is known about the spatio-temporal patterns of these interactions, primarily due to the considerable logistical challenge of quantifying them.
A spatially explicit approach, in which animal locations are used to characterise patterns of movement in human-modified landscapes, has great potential to improve our understanding of the dynamics of wildlife interactions with people. Animal locations provide the basic unit of movement paths and can identify areas where individuals interact with habitats co-used by humans (Cagnacci et al. 2010). For example, information about the movement patterns of tigers provides knowledge of the probability of tiger-human interactions in multiple-use landscapes, which is vital for providing guidance on how to balance competing land-uses (Ahearn et al. 2001). Additionally, location fixes of African elephants provides information about patterns of spatio-temporal overlap with people and provide insight into how elephant movement behaviour influences incidences of crop raiding (Jackson et al. 2008). At present, the most efficient means for obtaining location data from a variety of animal species is through the use of animal-borne tracking technologies.
Animal-borne tracking technologies, such as the Global Positioning System (GPS), can be used with high accuracy to pinpoint the locations of animals. A range of technologies have been used to track wildlife; e.g., radio-telemetry (Millspaugh & Marzluff 2001), geolocators (Egevang et al. 2010), ARGOS satellite tags (Costa et al. 2010), GPS/GSM loggers (Weber et al. 2015), and video-tracking (Moll et al. 2007). These devices have been used to study patterns of animal habitat use and resource selection (Kertson et al. 2011; Nelson et al. 2012), activity patterns (Owen-Smith & Goodall 2014), identification of areas important for conservation (Schofield et al. 2013), and response to anthropogenic disturbances (Panzacchi et al. 2013). The resultant data sets allow researchers to perform a range of analyses that can yield insights into the mechanisms underlying patterns of movement (Wells et al. 2014; Bestley et al. 2015; Gurarie et al. 2015). For example, GPS location fixes can be used to infer behavioural states from fine-grain movement data, which provides a new approach for quantifying animal movement behaviour in human-dominated landscapes (Jonsen et al. 2005; Wall et al. 2014; Zhang et al. 2015). Tracking technology and the associated analytical techniques it facilitates have the potential to provide a detailed picture of how wildlife interacts with humans and human-associated areas.
Here, I describe the approach of adopting a spatially explicit perspective to assess human-wildlife interactions as a proxy of HWC. I used location data obtained with GPS telemetry to quantify the locations and magnitude of interactions between wild kea (Nestor notabilis), and centres of human activity in New Zealand. My aims were to: (1) characterise the spatial distribution of GPS fixes collected, and ascertain whether kea are selecting human areas as preferred habitat; (2) determine how distance to human areas, and therefore conflict probability, varies through time; (3) assess individual-level differences in the locations and magnitude of interactions; and (4) quantify how behavioural patterns differ as a function of increasing distance from human areas. Finally, I discuss the general issues related to using a spatial approach to quantify HWC and the implications for the management of conflict-prone species.


Study species

The kea is a large, omnivorous parrot (family Strigopidae) found mostly in high-altitude Southern Beech (Nothofagaceae) forest, sub-alpine shrublands, and high-alpine basins and ridges in the South Island of New Zealand. Recently, kea were classified as ‘Nationally Endangered’ by the New Zealand Department of Conservation (‘DoC’) (Robertson et al. 2013) and the species’ current IUCN classification is ‘Vulnerable’, CITES Appendix II (IUCN 2014). Kea populations are believed to be declining and this downward trend is attributed in large part to increasing conflicts with humans (Temple 1996; Edwards & O’Connor 2014). Kea are opportunistic scavengers that often exploit anthropogenic resources in their habitats, potentially causing property damage, stock deaths, economic losses, and disturbance of worksites and businesses (Brejarrt 1994;Reid, McLelland & Gartnell 2011). Conflict between kea and humans can result in intentional (e.g., shooting, poisoning) or accidental (e.g., road accidents, electrocution) mortality events (Orr-Walker & Roberts 2009). Successful mitigation of kea-human conflict will require a thorough knowledge of interactions between the two species.

Study area

My study was undertaken at Arthur’s Pass National Park (42.93˚S, 171.56˚E) in the Southern Alps, near Mounts Rolleston, Temple, and Cassidy (Figure 3.1). Topographic features at the study site include deeply incised glacial valleys, high alpine peaks, and steep scree slopes; elevations range from 300 – 1720 m above mean sea level. In the study site there is a small area of human settlement centered on Arthur’s Pass village (0.6 km2, Dundas 2008) consisting of a small resident population of c. 50 people (Brown 2007) and a large annual influx of tourists (c. 250,000 p.a.; Dundas 2008). There are approximately 70 –100 resident kea in the study area (DoC).
Figure 3.1: Arthur’s Pass National Park, New Zealand, showing GPS fixes from all study birds. Each individual’s locations are indicated by a different colour. Contour lines represent distances from human areas at 500-m intervals with lightest red colour denoting the nearest distance. Letters have been used to highlight the main centres of human activity (A= Otira Village; B= Candy’s Bend Lookout; C= Death’s Corner Lookout; D= Arthur’s Pass Village).

Tracking devices

The GPS loggers used in my study were commercially available 20-channel receivers (Mobile Action Technology; Xindian District, New Taipei City, Taiwan) with integrated data storage and passive ceramic aerial, powered by a 380 mAh 3.7V lithium-polymer rechargeable battery. The receivers were removed from their original plastic housing and sealed in two layers of c. 0.9 mm polyolefin heat-shrink wrap (RNF-100-1, Raychem; Menlo Park/Redwood City, California, USA). Plastic tubes (6 and 4 mm external and internal diameters, respectively) for attachment of harnesses were fixed to the loggers with superglue, before a third layer of shrink wrap was added and sealed. Completed devices weighed approximately 19 g and were c. 60 mm × 27 mm × 12 mm in dimension. I configured the loggers to continuously record position fixes over a 24 h period at a nominal sampling interval of one fix every 3 min; this sampling regime permitted collection of sufficient data to describe, in detail, the birds’ daily patterns of movement and behaviour.

Capture and handling

GPS data-loggers were deployed on 10 adult male kea intermittently between 03 September, 2012 and 08 January, 2014. Locations where kea were captured had a strong human presence. I selected these sites as ‘urban’ birds are the most likely to be interacting with humans, which are the data I required for this study.
Kea are a protected species and I was only granted legal permission to catch a limited number of adult males. In addition, these birds must have been spotted previously around the study site at least three times to give an increased probability of successful recapture. To assess the age of my study birds, I used the DoC database, which provides records of resident birds. Kea were captured either using a leg noose mounted on a 1-m pole, noose lines, (see Bub 2012), or with a net gun that used a 0.32-calibre blank pistol cartridge to propel a 4-m weighted net over the target. GPS loggers were attached to the birds (generally in <15 min) between the wings and above the center of gravity using backpack harnesses (2 g) constructed of 2-mm nylon cord that incorporated a cotton weak link positioned over the keel (Karl & Clout 1987). GPS devices and harnesses ranged in weight between 1.9% and 2.6% of the study birds’ body mass (810 g – 1079 g). Loggers were retrieved by re-capturing the study birds (using the methods described above) after a minimum of seven days – the approximate operational life of the batteries at the scheduled sampling interval. Data recorded in the on-board memory of the loggers then were downloaded to a laptop computer for subsequent analysis. Methods of capture, attachment, and recapture are described in more detail in Kennedy et al. (2015).

Data analysis

Error screening. – The latitude and longitude locations downloaded from the GPS loggers (World Geodetic Survey 1984) were converted to planar eastings and northings in New Zealand Transverse Mercator (NZTM) coordinates using the PROJECT tool in ArcMap v. 10.1 (ESRI 2012). To remove large locational errors, I applied a running median-median smoothing filter (Tukey 1977) to the eastings and northings of each individual’s movement trajectory. Median filters reduce possible signal noise and are considered to be the ‘ideal’ smoothers of spikey time-series data (Evans 1982). Similar to Tukey (1977), my process employed a sampling window of n = 3 sequential observations using the ‘moving’ function in MATLAB v. R2012b (The MathWorks, Natick, MA, USA). A sampling window of three locations is optimal as it excludes single outliers and negates the ‘staircase’ effect that is common in median smoothing (Lind et al. 2005).
Spatial distribution of GPS fixes. – I digitised a 0.4 m resolution aerial photograph of the study site (Land Information New Zealand) to develop a feature class map in ArcMap v. 10.1 (ESRI 2012) and identified the following human infrastructure: buildings, backyards, parks, camping sites, car parks, and tourist lookouts. Roads were not included as human infrastructure as they follow the natural major orientation of the valley (a natural landscape feature) and hence may be misleading when interpreting movement trajectories. To determine the spatial distribution of GPS fixes in relation to the distance to human infrastructure, I calculated the horizontal distance of each GPS fix to the nearest digitised human feature using the Point Distance tool in ArcMap v. 10.1 (ESRI 2012).
Habitat Selection. – Jacobs Index (Jacobs 1974; Eq. 3.1), as calculated in Ranges7v0.81 software (South et al. 2005), was used to assess whether individuals preferentially selected areas of human infrastructure within their 99% kernel density estimate of space use (‘third-order’ selection; Johnson 1980).
Where r is the proportion of habitat type used and p is the proportion of that habitat available. A value of +1 indicates maximum selection, -1 indicates maximum avoidance, and 0 is neutral selection (i.e., selection is proportional to availability). In this evaluation, individuals were considered to be the sample unit, and human infrastructure was determined to be significantly preferred or avoided if the mean value of Jacob’s index was significantly different from zero (Palomeres et al. 2000). The evaluation was accomplished by calculating 95% confidence intervals for all individuals’ D-indices to assess whether zero was inside or outside the interval, as in (Kauhala & Auttila 2010).
Inferring behavioural states. – Models of multiple correlated random walks were fitted to each kea’s movement trajectory using a switching Monte-Carlo Markov-Chain (MCMC) algorithm (as in Morales et al., 2004). MCMCs provide a means to assign a distinct behavioural state to each individual GPS location fix; these are dependent on distributions of step lengths and the absolute values of relative turning angles. This method is commonly used to infer modes of behaviour from movement trajectories (e.g., Patterson et al. 2009; Postlethwaite & Dennis 2013), and it is well suited for analysis of remote animal-tracking data, in which consecutive data points are typically nondependent (Dean et al. 2012). First, I ascertained the number of distinct behavioural states within each kea movement trajectory using a maximum likelihood method, as in (Dean et al. 2012). WinBUGS 1.4 (Spiegelhalter et al. 1999) was used to run the MCMC; parameters were first initialised using k-means clustering (Leggetter & Woodland 1995) and then optimised through unsupervised training using the Baum-Welch algorithm (Rabiner 1989). For detailed descriptions of the MCMC technique as applied to animal-movement data see Morales et al. (2004) and Postlethwaite & Dennis (2013).
Behavioural bouts. – By assigning states of behaviour as inferred from the MCMC model to GPS fixes, I plotted the sequence of behaviour over an individual’s entire movement trajectory, showing where, when, and for how long each kea engaged in different states of behaviour. Behavioural bouts were defined as sub-segments of movement trajectories in which behavioural states for sequential observations were the same (a minimum of two observations were considered sufficient to constitute a bout). The duration of each behavioural bout was calculated by subtracting the time of the first fix from the end time of the last fix of the bout. I used a two-tailed paired t-test to test for differences in the mean bout duration of different behavioural states inferred by the model. As most interactions occur during the day, only behavioural bouts from the time between sunrise and sunset were used.
Temporal and spatial variation in behaviour. – I calculated how distance to human features and total distance travelled varied for the behavioural states over a 24 h period by using each individual’s mean distance to the nearest human feature and distance travelled per hour (sum of distances between locations) over the duration of the tracking period. Here I presented two representative examples of segments of one individual’s (B) movement trajectory to evaluate how behaviour is dependent on distance to Arthurs Pass Village (‘APV’). To evaluate whether there was a correlation between proximity to humans and duration of ‘State 1’ behaviours in each segment, I used linear regression.


Spatial distribution of GPS fixes

Data obtained from the GPS loggers demonstrate the degree of association of the kea’s locations to areas of human settlement. Six of the ten study birds regularly visited two tourist areas beside a primary road (Death’s Corner and Candy’s Bend), while areas of activity for the four remaining birds were concentrated near APV (Figure 3.1). One individual (D) repeatedly travelled between the two tourist areas and a nearby village (Otira) c. 6 km distance, where he recurrently visited several residences. Recorded distances of kea locations from the nearest centres of human activity varied between 0 and 3749 m, with a mean of 676 ± 3 m (Figure 3.2). More than 50% of all GPS fixes were recorded within 700 m of human infrastructure for eight of the kea, and within 1055 m for all birds (Figure 3.2). All kea clearly preferentially selected human areas. Jacob’s Index values were all positive, ranging from 0.33 to 0.98 (= 0.81 ± 0.15, 95% confidence intervals; Table 3.1), demonstrating that habitat use of human areas was disproportionately higher than expected on the basis of availability.

Behavioural model

All of the kea exhibited two distinct modes of behaviour in their movement trajectories. ‘State one’ behaviour as inferred by the MCMC model is characterised by low movement rates (6.6 ± 0.7 m/min; mean + 1 SEM) and high relative turning angles (81.8 ± 2.1˚), while ‘State 2’ behaviours exhibited the reverse pattern (42.8 ± 5.3 m/min and 54.5 ± 1.7˚). Visual inspection of the movement trajectory of kea A (Figure 3.3) indicates that location observations classified as ‘State 1’ are representative of ‘ground-based’ behaviours such as resting (example 1a) and walking (example 1b), whereas location observations classified as ‘State 2’ behaviour are representative of flight, including shorter-distance ‘hop’ flights around single focal areas (example 2a) and longer-distance commuting trips between focal areas (example 2b).

1.1 Human-wildlife conflict
1.2 Increasing magnitude of human-wildlife conflict
1.3 Emerging field of human-wildlife conflict management
1.4 Animal movement patterns
1.5 Influence of HWC on population dynamics
1.6 The kea (Nestor notabilis)
1.7 Human-kea conflict
1.7 Thesis objective
1.8 Thesis structure
2.1 Introduction
2.2 Methods
2.3 Results
2.4 Discussion
3.1 Introduction
3.2 Methods
3.3 Results
3.4 Discussion
4.1 Introduction
4.2 Methods
4.3 Results
4.4 Discussion
5.1 Summary of findings
5.2 Characterising the movement and behavioural patterns of kea
5.3 Quantifying the nature and extent of kea interactions with anthropogenic infrastructure
5.4 The effect of human activity on Kea
5.5 Qualitatively assess the impact of HWC on kea population dynamics
5.6 Informing Management examples
5.7 Potential of approach for other species
5.8 Future directions
5.9 Concluding remarks
References .
A spatio-temporal approach for exploring human-wildlife conflict using the kea (Nestor notabilis) as a case study

Related Posts