GPS-derived foraging behaviour of red deer and liveweight gain of red deer and wapiti crossbred sire lines on contrasting winter forage types

Experimental overview

An experiment to investigate variation in foraging behaviour and liveweight gain of rising 1-year-old female deer fed on two contrasting winter feed options was undertaken at Invermay Agricultural Research Centre, (Mosgiel, New Zealand 45°58′S, 170°04′E) during July and August 2020.

All experimental procedures had prior approval from the AgResearch Invermay Animal Ethics committee (PRJ0121969), as required in New Zealand by the Animal Welfare Act 1999.

Animals and nutritional management

A cohort of 163 female red deer and wapiti crossbreds was selected from a larger group that had been bred using artificial insemination utilising semen from five red deer sires and two wapiti sires. Progeny were randomly assigned while balancing for the seven sire lines from weaning at 3 months of age, and managed on pasture in two groups (81 and 82 individuals). During the winter period, the groups were randomly designated to one of two winter feeding regimes, an autumn conserved perennial ryegrass (Lolium perenne)-dominant pasture or a monoculture crop of kale (Brassica oleracea cv ‘Kestrel’). Both feeding regimes were supplemented with ad libitum pasture baleage that was replenished on eating. The kale regime group was transitioned onto the diet over a 2-week period following industry standard practice prior to trial commencement. Once transitioned onto their diets the groups were managed for the following 6 weeks on a paddock basis (1–3 ha) with no strip grazing. This allowed natural foraging behaviour to be expressed and enough feed to meet growth requirements of at least 140 g/day, which is a higher growth rate than is commonly achieved by female rising 1-year-old red and wapiti crossbred deer during winter.

Animal liveweight was recorded at the start and finish of the 6-week trial period (excluding the 2-week transition period) to the nearest 0.5 kg by using Tru-test load cells (Datamars, Auckland, New Zealand) attached to a Prattley (Prattley Industries, Temuka, New Zealand) auto drafter and controlled with a Gallagher Tsi2 head unit (Gallagher, Hamilton, New Zealand).

Feed quality was assessed at the start of the 6-week trial period from both treatments pre-grazing. Representative samples were collected, harvesting both pasture and kale to ground level, and drying in a fan-forced oven at 65°C until a constant weight was reached. Samples were sent to Hills Laboratories, Hamilton, New Zealand, for analysis for crude protein, neutral detergent fibre, acid detergent fibre, lipid, soluble sugars and starch, and metabolisable energy by near-infrared spectrometry.

GPS technology and setup

Foraging behaviour was captured using GPS units over a 2-week period at the start of the 6-week trial period on all progeny from four red deer sire lines, except for twin progeny. The selected progeny were from two pairs of sires that had similar 12-month liveweight estimated breeding values. Due to a limited number of GPS units, they were not attached to progeny from wapiti crossbred sires and one red deer sire (unmatched for 12-month liveweight estimated breeding values). The units were developed by DataCarter Ltd (Palmerston North, New Zealand) and comprised a printed circuit board with a microcontroller, a u-blox NEO-M8Q GPS receiver and a SD card holder and were powered by a 3.6 V 19 Ah lithium thionyl chloride battery giving a recording duration of between 14 and 17 days. Units were housed in an injection moulded hard plastic casing (12 cm × 6.5 cm × 5 cm) weighing approximately 265 g and attached to adjustable nylon webbing that is fitted to the animal’s neck by using side-release clip buckles. Each collar was adjusted to optimally fit the individual animal by ensuring four fingers could fit comfortably behind the strap. The GPS units were programmed to record a location when the animal had moved more than 5 m from the previous location, or if the animal was stationary the location was recorded every 30 s. Locations were not recorded if the animal was moving at a greater speed than 20 m/s, because these were considered to be spurious GPS locations.

Behavioural-data classification

Deer foraging behaviour was estimated using two methodologies. The first method used GPS data location to calculate daily distance travelled and to inform diurnal activity patterns. The second method used a classifier algorithm that used data from a previous currently unpublished validation study (subsequently referred to as the validation study) for classifier training. The validation study recorded both GPS and accelerometer data at one observation per second. To create training data comparable to the current study, the GPS data in the validation study were downsampled to match the sampling scheme for the collars in this study (Winter GPS data). A multivariate, non-parametric changepoint algorithm (James and Matteson 2014; James et al. 2019) was applied to the downsampled validation study GPS data, so as to break the data into intervals on the basis of timestep, distance, speed, and turning angle. Within each interval, a confidence ellipse was fitted to the GPS locations, and the radii of the major and minor axes were recorded. These were used to calculate features for each interval, including a measure of (inverse) tortuosity (ratio of minor axis to major axis), a measure of area covered (area of the ellipse), and a measure of rate of area covered (area divided by duration of the interval). Average speed over the interval was also used as a feature.

Observational data from the validation study were used alongside this validation study feature data to train a decision tree (Ripley 2021), allowing each interval to be classified as ‘Foraging’, ‘Walking’, or ‘Resting’.

The same changepoint algorithm and feature engineering steps were applied to the Winter GPS data, and the trained decision tree was applied to obtain predicted behaviours.

Counting of bouts is sensitive to the choice of minimum bout length, without which there is a risk that classification errors or transient behaviours will inflate the count. In this study, we counted ‘sustained bouts’, by applying a smoothing filter to our classification output. The filter replaces the classification of bouts, which are shorter than the minimum length with the most recent sustained behaviour. The filter was run with both a 5-min and a 15-min threshold, to get a sense of the sensitivity of the analysis to the chosen threshold for a sustained bout.

Behaviour and performance analysis

Linear mixed models (Bates et al. 2015) were fit with the response variables, including daily distance, number of foraging bouts, foraging bout duration, foraging bout speed, and foraging bout distance. For the foraging bout responses, models were fitted using values calculated with both the 5-min and 15-min bout cut-off. All of these models used treatment, sire, and their interaction as fixed effects, as well as a fixed contrast for the two pasture paddocks used. Random effects for animal, for day, and for treatment nested within day were included. For one model (number of foraging bouts with 5 min cut-off) the random effect for day was removed to achieve numerical convergence.

Liveweight data were used to calculate growth rates for the 6-week mid-winter period (23 June – 7 August) after the 2-week transition period. Growth rates were analysed with a linear model for each winter forage, with sire line and starting winter liveweight as covariates. Starting and ending winter liveweights were also compared among sire lines with linear models.

Growth rates were compared to behaviours. Per-animal means were calculated for daily distance travelled, and foraging bout number, duration, speed, and distance (with both the 5-min and 15-min cut-off) for a total of nine behaviour variables. For each behaviour variable, a linear model was fit, with growth rate as response and behaviour, winter forage, sire, and all interactions as covariates. None of these nine regressions was significant, with P-values from F-test ranging from 0.14 to 0.74, so they have not been presented in the results section.

All analyses was performed in R 4.2.1 (R Core Team 2022). ANOVA was performed using the car package (Fox and Weisberg 2019), marginal means and their standard errors were calculated using the emmeans package (Lenth 2022), and post hoc comparisons among groups used the multcomp package (Hothorn et al. 2008).

GPS location data

The average daily distance travelled as estimated using the GPS location on the kale treatment was 6119 m/day (±283.2 m), and 6699 m/day (±278.4 m) for the pasture treatment (Table 1). The diurnal pattern of the kale treatment group indicated two activity peaks during daylight hours at 10 h and 18 h after midnight (Fig. 1). The morning activity peak was extended over a 3-h period, with a peak of over 600 m travelled in a single hour on average per day. The evening peak was shorter but had a higher peak of travel at over 800 m per hour on average. The pasture treatment group had four activity peaks during daylight hours at 8, 12, 14, and 18 h after midnight (Fig. 1). The activity periods lasted approximately an hour on average each day, except for the midday peak, which had a 2-h average duration. The average hourly distances per day travelled during these activity periods were under 500 m, each except for the evening activity period which peaked at close to 800 m.

Fig. 1.

Average hourly distance travelled (±s.e.) as measured by GPS by two groups of farmed red deer managed on either kale (Brassica oleracea) or pasture forage over a 2-week period in winter.

Behaviour classifier

The kale and pasture treatment groups were predicted to have 5.99 (±0.148) and 6.69 (±0.141) 15-min or longer foraging bouts respectively, per 24-h period (Table 1). These bouts had an average length of 95 (±3.6) and 86 (±3.4) min each for the kale and pasture group respectively, equating to total foraging time of 570.5 and 620.7 min per 24-h period. The distances travelled during these 15-min or longer bouts were 679 (±54.0) and 620 (±53.0) m respectively, for the kale and pasture groups per foraging bout with an average foraging speed of 1.9 km/h. Average daily foraging distances calculated using the classifier data were 3741 and 3955 m for 5-min sustained bouts and 4068 and 4150 m for 15-min sustained bouts for the kale and pasture groups respectively.

Foraging bout duration was different among sires for bouts greater than 5 min (P = 0.025, Fig. 2) and 15 min (P = 0.024), with the Red 1 progeny having the shortest duration (48.2 and 86.9 min respectively) and Red 2 progeny the longest duration (54.8 and 98.1 min respectively). The number of foraging bouts were also different among sires for foraging bouts greater than 5 min (P = 0.016) in duration but there were no differences for foraging bouts greater than 15 min. There were no other sire effects detected in predicted foraging behaviours.

Fig. 2.

Average foraging duration of individual progeny from four different red deer sires for foraging bouts longer than 5 min in duration.

Liveweight and liveweight gain

A total of 161 deer had complete liveweight records that could be used for liveweight analysis. Starting liveweights averaged 79.7 (±1.1) and 80.7 (±1.1) kg respectively, for the kale and pasture groups. End-of-trial weights averaged 83.0 (±1.2) and 82.6 (±1.2) kg respectively, for kale and pasture groups, equating to an average group daily growth rates of 72.4 (±6.5) and 41.7 (±6.5) g/day respectively.

Starting liveweight was different among the sire lines (P < 0.001), with Wapiti 1 progeny being (P < 0.05) heavier than any other sire progeny, with an average weight of 94.2 kg (Table 2). Red 2 progeny were the lightest, with an average of 73.9 kg, but were not significantly different from other red deer sire progeny.

Liveweight gain over the 6-week trial period was higher for the Wapiti 1 (171.5 g/day) progeny than for the red deer sire progeny (≤67 g/day) on the kale treatment (P < 0.001) but not on the pasture treatment (P = 0.074; Table 2, Fig. 3). Starting and finishing liveweights were higher for the Wapiti 1 sire progeny (94.2 and 99.0 kg respectively) than for all red deer sire progenies (average 76.92 and 78.82 kg, respectively), but did not differ from those for Wapiti 2 sire progeny (81.8 and 86.2 kg respectively).

Fig. 3.

Mid-winter (23 June – 7 August) growth rate (g/day) of progeny from seven sire lines grazing either kale crop or pasture compared against start of period (23 June) liveweight.

Feed quality of the kale crop averaged 13.3 MJME pre-grazing, compared with an average of 10.5 MJME of the pasture pre-grazing.