Antipredator responses could affect nutrient intake, which could lead to nutritional deficits. However, little is known about the antipredator response of small herbivores because most are nocturnal or crepuscular and therefore very difficult to study in the field. Therefore, we experimentally assessed the effect of a reactive response to predation risk on the nutrient (i.e., phosphorous) intake of the European hare (Lepus europaeus) using three different playback sounds. Additionally, we studied the time spent being costly vigilant, the time spent foraging, and the vegetation height in which the hares were present using accelerometers and GPS. Our results showed that elevated predation risk from our playback experiment did not affect the (1) phosphorus intake, (2) time spent being costly vigilant, and (3) time spent in tall vegetation. However, elevated predation risk did increase the time spent foraging. Possibly hares spent more time foraging with an increased predation risk because hares cannot seek refuge from predators. Additionally, the effect on phosphorus intake could be weak because phosphorous intake does not benefit a flight escape, while the reactive response acts late in the predation sequence limiting the effect on hare ecology. Prey anti-predator responses seem strongly related to the escape tactics of prey species that can differ between different habitats and the time of the day. More detailed field studies are necessary to get a better insight into species’ anti-predator-food tactics.

Study site

All data used in this study were collected by Weterings et al. (2018) in 2014–2015 in a coastal dune landscape (52°33′N, 4°38′E) in the Netherlands. In this dune landscape, we focused on two study sites (275 and 50 ha) with a population of European hares (±15 hares/km²). The areas consisted of patches of grass, thicket, brushwood, and forest.

Research design and data collection

In October 2014, nine hares were caught with Speedset static hare nets (height 45 cm, with 13 cm full mesh; JB’s Nets, Alexandria, UK), blindfolded (Paci et al. 2012), and kept in darkened boxes temporarily to reduce stress. Five hares were caught from the Koningsbos area and four hares from the Vennewater area. Hares were equipped with a lightweight GPS-ACC collar (69 g, 1.8% ± 0.2 SD of a hare’s body weight) that included a radio link for wireless communication (Type A, E-obs GmBH, Gruenwald, Germany) to minimize disturbance of the hares. Sex and life stage were determined by Stroh's method (juvenile < 1 year/adult > 1 year). Healthy hares (weight 2981–4400 g) were tagged without sedation (Gerritsmann et al. 2012). All handling of the hares was approved by the Wageningen University Animal Experiment Committee (no. 2014034.b) and followed the EU Directive 2010/63 on the protection of animals used for scientific purposes.

To investigate the costs of a predator–prey encounter on European hare nutrient intake, we conducted a playback experiment between 16th of December 2014 and 21st of January 2015. Even though different cues can be used to trigger a response to increased predation risk (Prugh et al. 2019), playbacks are often used in field studies to investigate immediate responses to predator–prey encounters (Clinchy et al. 2012). Moreover, for species that rely more on sound than sight and smell, such as hares (Łopucki et al. 2017), playbacks are generally more meaningful in assessing prey response to encounters (Jarvis 2004). Furthermore, playbacks may often be more alarming than visual cues (Cohen et al. 2009). Hares that participated in the playback experiment were selected based on their spatial distribution to maximize the distance between individual hares treated. Based on the GPS locations of individual hares, hares within 300 m of each other were assigned the same treatment. We used playbacks of conspecific alarm calls of hares, instead of playbacks of predators, because prey often responds more strongly to conspecific alarm calls than to playbacks of predators (Schmidt et al. 2008; Magrath et al. 2014). Conspecific alarm calls may warn conspecifics of predators (Smith 1965; Sherman 1977; Zuberbühler et al. 1999; Blumstein 2007), but may also communicate directly to predators that have been detected (Digweed and Rendall 2009a, b; Hasson 1991; Sherman 1985; Woodland et al. 1980).

The playback experiment consisted of three different treatments (1) playbacks of conspecific alarm calls of hares, (2) playbacks of sheep (control playback), and (3) no sound. The playback experiment consisted of three blocks of four days. In every treatment block, hares were either exposed to the treatments from audio boxes (Foxpro Fury2, FOXPRO inc. Lewistown) or to no sound at all (Supplementary materials 1, Table S1). Different treatments occurred within the same block; however, hare and sheep playbacks were never used within the same block. The three treatment blocks were chosen to control for changes in daylight and weather conditions. Weather data were collected from the weather station in IJmuiden (approximately 10 km from Castricum) (Koninklijk Nederlands Meteorologisch Instituut 2020). After each treatment block, there were at least five days without playbacks to avoid carry-over effects (e.g., Petrovan et al. 2012). Ten different combinations of three playback fragments of each 40 s were placed in random order. To avoid habituation, fragments were never used more than two times (McGregor et al. 1992). Playbacks were played for 40 s at 20:00 h (CET), because hare activity and foraging behavior peaked during that time (Hansen 1996), thereby increasing the chance of triggering anti-predator responses during foraging. Audio boxes were placed 50 m south from the core location of GPS activity of a given hare at 20:00 h on previous days (mean distance between boxes = 1117 ± 1882 m), with the largest speaker directed towards the north to standardize distribution of playbacks in different directions.

Costly vigilance and foraging time

To investigate time spent costly vigilant and time spent foraging, we recorded accelerometer (ACC) data of collared hares in three axes, every two minutes for 24 h a day with a frequency of 10.54 Hz per axis. To interpret the accelerometer data, a hand-held video camera was used to record behavior of collared hares in the field. We video-recorded the behavior of eight hares to account for individual variation between hares (see Brivio et al. 2021); six hares from this study site and two hares from another coastal dune habitat on Schiermonnikoog island. Behavioral observations of the latter two hares in a comparable habitat were only added to improve the classification of the accelerometer data of this study, not to explain our hypothesis, because the GPS collars used and the sampling design were exactly the same. The six hares in our two study sites were observed when hares were expected to be the most visible and active (7:00–10:00 and 13:00–16:00). We recorded a total of 9225 s of behavior (mean 1153 s ± 1509 SD per observation).

Vegetation height

To investigate the vegetation height, we measured the vegetation height at five orthogonal locations in six random 2 × 2 m quadrants in each of the 20 vegetation types (n = 120) (Agriculture; flower-rich grasslands; bulb fields; dune grasslands; Burnet rose, creeping willow-, blackberry thicket; bare sand; calcareous dune grassland; calcareous dune valleys; deciduous forest; coniferous forest; former agriculture; other; other forests; reed swamp; reed swamp communities; herbaceous, fault, and mantle communities; thickets; nutrient-rich grasslands; nutrient-rich pioneer communities, flood meadows, and pace vegetation; near-shore communities). Next, we recorded GPS locations of individual hares every 12 min for 24 h a day, and used ArcGIS (version 10.7) to link the GPS location of the hares with the average height of the vegetation type at that location.

Nutrient intake

To test the effect of predation risk on nutrient intake, the available food quantity (i.e., edible biomass) and the nutrient concentration of each vegetation type were measured. We collected samples of edible biomass for seven of the most important plant species in the diet of hares (i.e., Festuca rubra, Agrostis capillaris, Poa pratensis, Holcus lanatus, Poa trivialis, Taraxacum officinale, Rubus caesisus; Kuijper et al. 2008; Weterings et al. 2018) and a commercial flower bulb species using the hand-pluck method (de Vries and Schippers 1994). Edible biomass (i.e., the green plant parts that have a high nutritional value and are selected by hares; Homolka 1987) were collected by the hand-pluck method in six randomly placed circular plots (10 m radius) up to 50 cm in height in each vegetation type (n = 120). Plant parts were air-dried, stored, and chemically analyzed for the concentration of phosphorus (P). We chose phosphorus to investigate nutrient intake, because phosphorus plays an important role in the body of animals, involving the skeletal formation, energy storage, metabolism, nerve impulse transmission and muscle contraction (Barboza et al. 2009) that could facilitate flight from predators. Furthermore, phosphorus is considered one of the most important nutrients for hares (Miller 1968).

Data preparation

Costly vigilance and foraging time

The video recordings of hare behavior were used to label one-second segments of ACC data that only consisted of the same behavior. Hare behavior was classified into six postures (i.e., sitting, sitting alert, standing, standing on hind legs, movement, and jumping), and six activities (i.e., chewing, cropping, grooming, scratching, shaking, and stretching) using the software Avidemux (2.6.6). Labeled segments of ACC data (training data) were used to classify unlabeled segments of ACC data into behaviors using Decision Tree (accuracy 80.96% ± 0.75 SD) in the AcceleRater software (Resheff et al. 2014). We used the time sitting alert as a proxy for the time spent on costly vigilance, and cropping time as a proxy for foraging time. We chose cropping time instead of chewing time to determine the foraging time, because cropping time was classified with higher precision and recall than chewing time.

Even though there is a trade-off between foraging time and chewing time (Spalinger and Hobbs 1992), nutrient intake increases when chewing time as well as foraging time increases (Gross et al. 1993).

Vegetation height

To calculate the fraction of time hares spent in a certain vegetation height in an hour, the GPS location of the hares was linked to a high-resolution GIS map (1:5000) of the different vegetation types (Everts et al. 2008, 2009). However, whenever hares were present in multiple vegetation types within an hour, we calculated the weighted vegetation height.

Nutrient intake

We calculated the relative nutrient intake of hares by multiplying the time spent foraging by the phosphorus concentration in the edible biomass (Fig. 1).

The average edible biomass (g/m²) was calculated for each vegetation type by summing the amount of edible biomass (g) of all plant species in one square meter of a vegetation type up to 50 cm in height. The average content of phosphorus in every vegetation type was calculated by averaging the percentage of phosphorus in the edible biomass present in the vegetation type, weighted by their volume per square meter up to 50 cm in height (see Weterings et al. 2018).

Data analysis

We explored the data using the protocol of Zuur et al. (2010) to identify potential statistical problems. Because all males were juveniles, we could not investigate the effect of life stage in our analysis. We used Generalized Additive Mixed Models (GAMMs) in R (R Core Team 2021; R package ‘mgcv’ version 1.9-0 (Wood 2017)) to test the effect of the treatment on the fraction of time spent costly vigilant by hares (n = 1390) (i.e., beta distribution), the average vegetation height (n = 1227) in which the hares were present (i.e., Gaussian distribution), the fraction of time spent foraging (n = 1342) (i.e., beta distribution) and on the phosphorous intake by hares (n = 1342) (i.e., Gaussian distribution). GAMMs describe highly nonlinear relationships between response and explanatory variables using smoothing functions (Guisan et al. 2002). In total, we investigated 168 h (7 days times 24 h) of response by the hares. All four global models included the treatment, the control variables sex, body weight, temperature, wind speed, rainfall, prior treatment, time of day and the interaction treatment*time of day. Because hares shift between short and tall vegetation during a day at dusk and dawn (Schai-Braun et al. 2012), we included time of day and the interaction treatment*time of day in the analysis. The variable ‘prior treatment’ was added to control for any carryover effects by the treatment the day before. The prior treatment on the first day of a treatment block was categorized as no treatment. Additionally, we transformed (1) the amount of rainfall into presence-absence data, because the data mainly showed zeros, and (2) phosphorous intake (log(x + 1)) because the data were right-skewed. We included vegetation height as an explanatory variable in the models that investigated time spent on costly vigilance (Hopewell et al. 2005; Riginos and Grace 2008), foraging time (Heuermann et al. 2011) and phosphorous intake (Shipley 2007). However, vegetation height was excluded from the foraging time model and the phosphorous intake model due to multicollinearity. We found no multicollinearity between the other control variables. All continuous covariates were standardized to compare the effect size within and between models. Hare ID and treatment day block within hare ID were considered random factors. We excluded one hare from the analysis because we did not identify its sex. A GAMM with cyclic smoother was used to model the effects during the time of day, to avoid discontinuity between subsequent days. Temporal autocorrelation among subsequent hours within a time block was addressed by including an autocorrelation structure, modeling a decreasing degree of autocorrelation with increasing temporal distance between data points. We chose an autoregressive (AR(1)) covariance type for individual time blocks at each site as this resulted in the best fit. The Akaike Information Criterion (AIC) was used to select the final model using the ‘base’ R-package (version 3.6.1). We validated the final model using the ‘MuMIn’ R-package (version 1.43.17) (Bartoń and Bartoń 2020) to plot the residuals against the predicted value and all the covariates.