1. Animal movements towards goals or targets are based upon either maximization of resources or risk avoidance, and the way animals move can reveal information about their motivation for movement. 2. We use Bayesian movement models and hourly GPS-fixes to distinguish animal movements into movement states and analyse the influence of environmental variables on being in and switching to a state. Specifically, we apply our models to understand elephant movement decisions surrounding agricultural fields and crop consumption. As it is unclear what the role of habitat features are on this complex issue, we analyse whether elephants target agricultural crops for consumption, or simply pass through them in search of water. 3. Our Hidden-Markov models divide elephant movements into two states: exploratory movements that are fast and directional, and encamped movements that are slow and meandering. For each elephant, we ran 16 models with each possible combination of habitat features (river, elephant corridor, agricultural field, trees), and repeated these analyses including interaction effects with both season and time of day. We used cross-validation to select the best performing model, and GLMMs to analyse the influence of habitat features on being in and switching to a state. 4. Our results show that in corridors, exploratory movements are dominant. Elephants mainly showed encamped movements at the river during the dry season, when temporary water sources have dried out and elephants rely on this permanent water source. In fields, males most often exhibited exploratory movements to and from the river, while females showed an increase in the frequency of encamped movements at night –when most crop consumption and movements through fields occur- and during the dry season. 5. The predation-risk hypothesis could explain this behaviour, since foraging in fields might be less risky under the cover of darkness and during the dry season when farmers are absent from fields. This sexual segregation in elephant movement decisions highlights the importance of risk in movement patterns, while the increase in encamped movements in the dry season suggests the importance of agricultural timing. Taking this into account could increase efficiency of elephant crop consumption mitigation. 08-Jan-2020

GPS location data

We collected GPS data from April 2014 to July 2016 for 6 male and 5 female elephants, each in different herds, collared with Vectronic GPS collars. The collars produced hourly fixes of the elephants’ locations, resulting in ca.19,000 fixes per elephant. We downloaded data using the Vectronic GPS PLUS X Collar Manager. When there were GPS reception issues resulting in missing data, we coded these as NA values. Each elephant included in the study produced data with less than 25% missing values.

Remote sensing data collection

To create habitat feature shape files, we used Landsat 8 band combinations (Data available from the U.S. Geological Survey) and the Semi-automatic Classification Plugin (Congedo, 2014) available in the open source Quantum GIS Geographic Information System program (Quantum GIS Development Team, 2016). We calculated overlaps between downloaded GPS coordinates and habitat features using the ‘NNJoin’ QGIS Plugin (Tveite, 2015), resulting in dummy variables for each of the covariates per hourly GPS fix. We created shape files for the Okavango river (‘river’), elephant corridors (determined by Songhurst et al., 2015), agricultural fields and tree agglomerations. For both of the river and field variables, we also calculated a buffer zone of each 300m, since research has shown that elephant movements change within this distance of these landscape features (Saj et al., 2001; Naughton-Treves & Treves, 2005; Chamaillé-Jammes et al., 2013). As some of the habitat features may overlap (e.g. corridor and field), an elephant can be in multiple habitats at once. We excluded other vegetation, such as grasses and herbs in the area, as they are highly dependent on rainfall and non-permanent.

Preparing the data

We combined the data from all elephants and converted the GPS coordinates into step sizes and turning angles using the adeHabitatLT package for R (Calenge et al. 2016; R Core Team, 2017). We discarded 11,731 fixes (of the total 213,301 fixes for all elephants) because these located the elephants to the same anomalous position, which was >4000 km outside of their home ranges. We constructed a ‘season’ dummy variable considering the wet season to range from November-April and the dry season from May-October and a ‘time of day’ dummy with night occurring from 20.00-6.00, including dusk and dawn.