Physiological and behavioural responses of moose to hunting with dogs

Biologging devices (sensors collecting data) were used to investigate the physiological and behavioural responses of moose to hunting with baying dogs. Moose increased their heart rate, body temperature and movement rate at the day of the approach and compensated by increasing their resting time the day after the approach.

and length of the exposure (McLaren et al., 2007, Reeder and Kramer, 2005, Rehbinder, 1990. A stressor is a stimulus causing stress, and the stressor could be physical, psychological or both (Reeder and Kramer, 2005). As a response to a stressor, the sympathetic nervous system is activated immediately, also known as the fight-or-flight response, while the hypothalamic-pituitary-adrenal axis is activated more slowly and lasts longer (Reeder andKramer, 2005, Rehbinder, 1990). Stress results in physiological (i.e. metabolic, neuroendocrine and immunological) and behavioural responses of the animal, and objective and quantitative measurements of stress are often used to assess animal welfare (McLaren et al., 2007). Both immediate and long-term responses to stressful situations could result in life-threatening consequences, including hyperthermia in acute stress situations and immune suppression, loss of body weight and decreased reproductive rates in longterm situations (Moberg, 2000).
In Sweden, more than 80 000 moose (Alces alces) are harvested annually. The hunting season lasts for 5 to 6 months from the beginning of September or October until January or February, depending on local regulations (Lavsund et al., 2003;Länsstyrelserna, 2020). Hunting with dogs, especially baying dogs, is a common practice (Ball et al., 1999;Heberlein, 2000). Concerns about animal welfare and stress related to hunting have been raised (Nelson et al., 2005, Paquet andDarimont, 2010), but so far, data are lacking to assess these issues. Different physiological and behavioural variables help to measure and to determine if hunting disturbances result in increased energy expenditure.
Stress can be measured through physiological variables like body temperature (T b ) and heart rate (HR) (McLaren et al., 2007). However, the identification of changes during potentially stressful situations requires a baseline for these variables. Continuous measurements of HR and T b could indirectly assess the sympathetic nervous system and the acute stress response as the activation results in the release of catecholamines (adrenaline and noradrenaline), which increases HR and T b . Increased HR (>200 bpm) and T b (>42.0 • C) have been recorded in brown bears (Ursus arctos) chased by baying dogs and black bears (Ursus americanus) hunted using bait (Laske et al., 2011;Støen et al., 2018). Extremely high T b is cytotoxic, and even short periods of exposure to high T b can be lethal (Lepock, 2003, Tansey andJohnson, 2015). Increased HR during hunting situations could be due to the combination of both fleeing and stress, and it is difficult to differentiate the role of each as both are normal physiological reactions. Persistently elevated HR can be dangerous when lasting for a long time, and even fatal, due to arrhythmias, infarction or myopathy of the heart (McLaren et al., 2007;Stephenson, 2007).
Behavioural responses to stress caused by hunting, also known as anti-predator behaviour, have been studied in different game species including moose (Baskin et al., 2004;Neumann et al., 2009;Ericsson et al., 2015b;Sand et al., 2016), red deer (Jarnemo andWikenros, 2014, Sunde et al., 2009), roe deer (Capreolus capreolus) (Benhaiem et al., 2008, Cederlund andKjellander, 1991) and brown bears (Ordiz et al., 2013;Hertel et al., 2016). In ungulates, displacement from typical home range, increased home range area and nocturnal behaviour and stay, fight or flee when attacked are examples of anti-predator strategies (Benhaiem et al., 2008, Cederlund and Kjellander, 1991, Jarnemo and Wikenros, 2014, Lingle and Pellis, 2002, Stankowich, 2008, Sunde et al., 2009. Predation risk can drive prey to be more vigilant and to shift to safer habitats, which can result in lower fitness due to poorer forage quality and less time spent on feeding (Brown et al., 1999). How individuals respond varies among habitat characteristics, hunting pressure and hunting method (Jarnemo and Wikenros, 2014). Hunting disturbance can result in increased resting time to compensate for increased energy consumption or to recover from the exhaustion (Le Grand et al., 2019).
While several studies have evaluated behavioural responses (Ericsson et al., 2015b;Sand et al., 2016), none have studied the physiological responses of moose to baying dogs. Behavioural studies found that exposure to baying dogs resulted in increased activity, i.e. increased maximum speed and flight distance, and in moose leaving the area after disturbance (Ericsson et al., 2015b, Sand et al., 2016. Sand et al. (2016) found differences in the flight patterns between females with and without calves, as well as differences in flight patterns and reactions of moose in response to different dogs. Behavioural impacts, together with physiological responses, have been assessed in simulated hunts with baying dogs in brown bears (Le Grand et al., 2019). During experimental days, compared to control periods, the authors noted higher HRs and T b , with the bears also travelling longer distances and at higher speeds.
Here, we investigated the behavioural and physiological responses of moose to simulated hunting situations with baying dogs (hereafter approaches). We tested for changes in five variables; maximum T b , HR, Euclidean distance travelled, maximum speed of movement and time allocated to resting versus active behaviour. We measured moose response during three periods (reference period (14-20 August), day of the approach and day after the approach). Based on previous studies of the effect of hunting on moose and other game species, we tested the following predictions: Effects on physiology; maximum T b and mean HR will be higher during the day of the approach compared to the reference period and the day after the approach.
Effects of behaviour: 1. Total distance travelled (per day) will be longer and the total time spent resting (minutes inactive/day) will be lower on the day of the approach compared to the day after. 2. Maximum speed will be higher during the day of the approach compared to the day after.

Study area and animals
The study was conducted in north-eastern Sweden, in the county of Västerbotten in Nordmaling, Vännäs and Umeå municipalities. During the study years of 2017 and 2018, a total of 675 and 782 moose were shot in the three municipalities, respectively (SvenskaJägareforbundet, 2019). The study area is characterized by boreal forest, dominated by Scots pine (Pinus silvestris), Norway spruce (Picea abies) and birch (Betula spp.) in addition to agricultural land and marshland. The elevation level in the area ranges from 0 to 300 m.a.s.l., and the total land area of the three municipalities are 4076.2 km 2 (SCB, 2019). During the management period 2016-2018, the estimated moose density was on average 8.2 moose per 1000 ha in the winter (i.e. after hunting) (CAB, 2019). We studied the responses of eight adult female moose (>1.5 years old) (not each of the eight females was studied every year) and 12 baying dogs of different breeds frequently used for moose hunting (Norwegian Elkhound: n = 4, Jämthund: n = 3, other breeds: n = 5). In 2017, all five females had calves in spring, and in 2018, six had calves and one was without calf in spring.

Biologgers, programming and data collection
Collars included a GPS receiver, an ambient temperature recorder, a triaxial acceleration sensor, a mortality sensor, a very high frequency (VHF) transmitter and a global system for mobile communication modem (VectronicAerospace, 2019a). In addition, each collar was linked to external sensors, i.e. the MIT and proximity sensors UHF-ID tags (VectronicAerospace, 2019b; VectronicAerospace, 2019c). We adjusted collar-derived ambient temperatures according to the offsets described by Ericsson et al. (2015a) to represent a reliable index for the actual ambient temperatures. The acceleration sensor integrated in the collar measures activity as back-forward and right-left movement over two axes (X and Y) on a scale from 0-255 at 6 to 8 Hz, and stores average values over 5-min recording intervals. Overall activity was presented as the sum of the activity data on the X-and Y-axes, ranging from 0 to 510 (Gervasi et al., 2006). GPS positions were recorded every 3 h during the study period, and we changed it manually for the day of the approach to 10-min positions. This schedule was kept until the end of the study period for that year. Additionally, in 2017, the schedule switched to 1-min positions (n = 2) when in contact with a proximity sensor (attached to a dog). All GPS positions with associated information (ambient temperature, the most recently stored ruminal temperature and proximity contact) were sent to the wireless remote animal movement (WRAM) database for storage (Dettki et al., 2014). At recapture of the moose, the remaining data stored on board the collar were manually downloaded and sent to the WRAM database.
To record moose physiological responses, we deployed a MIT in the rumen as described earlier (Minicucci et al., 2018;Graesli et al., 2020). MITs record the ruminal temperature at 5-min intervals and transmit the information for storage to the collar unit (VectronicAerospace, 2019b). In addition, moose were fitted with surgically implanted subcutaneous HR loggers (Reveal XT and DST centi HRT). Reveal XTs continuously monitor the HR using an electrocardiogram (ECG), which converts the mean R-R interval (rate of a ventricular cycle) into HR, and store 2-min average values (Medtronic, 2017). DST centi HRT calculates a mean HR from a 4-s ECG strip with a 150 Hz measurement frequency and stores the HR with a quality index of the signal clarity and the R-R interval regularity (StarOddi, 2017). Prior to the surgery, an analgesic, meloxicam (Metacam ® , Boehringer Ingelheim Vetmedica GmBH, Germany), was given subcutaneously at a dose of 0.5 mg/kg. The DST centi HRT was surgically implanted as earlier described (Graesli et al., 2020), and the Reveal XT was surgically implanted subcutaneously at the left side of the most rostral part of the sternum, according to the same procedure. We programmed and activated the Reveal XT after implantation, prior to the anaesthetic reversal.
Data recorded by the collars, MITs and Reveal XTs were downloaded in the field during recaptures in February 2018 and in February 2019. The recaptures were carried out in the same way as the initial captures. During the recaptures in 2019, we surgically removed the Reveal XTs according to the same procedure as used for implantation.
The dogs included in the real hunts belonged to local hunting teams, while the experimental hunts were conducted by trained field personnel (experienced hunters) with trained hunting dogs. All dogs were equipped with Garmin T5 or DC50 Dog collars (Garmin Ltd, Olathe, Kansas, USA) and could be tracked directly via a hand-held GPS (Garmin Astro 320) (Garmin, 2019). In addition, the dogs of the local hunting teams were equipped with UHF proximity tags (Vectronic Aerospace GmbH, Berlin, Germany), which send a signal to the moose collar when in close proximity (Vectron-icAerospace, 2019c).

Experimental protocol
The experimental hunts were performed after the national leash law (i.e. dogs should be kept on leash/under supervision from 1 March to 20 August) was lifted (21 August) and until the beginning of the annual moose hunt (first Monday in September, i.e. 4 September in 2017 and 3 September in 2018, Länsstyrelserna, 2020). Real hunting situations were part of the regular moose hunt in the area from 4-24 September 2017. None of the real hunts included in our study resulted in a moose being killed. The experiments were approved by the Animal Care Committee for Northern Sweden in Umeå (application numbers: Dnr A 3-16, Dnr A 28-17). Experienced hunters, hunting dogs and field personnel carried out the real and experimental hunts, while experienced field personnel, pilots and veterinarians carried out captures, handling and surgeries.

Experimental hunts
For each experimental hunt, a dog handler walked with the leashed dog in an upwind direction to a known moose position given by its latest GPS position. The current position of the moose was determined using a VHF receiver, aiming to release the dog close enough to the moose to avoid that the dog would pick up the scent of other, unmarked animals, while at the same time avoiding that the target moose would become aware of the human presence. Within a distance of 200 ± 50 m the dog was let off the leash and started tracking the moose while the handler monitored the dog using a handheld GPS, thus simulating a real hunting situation (see Fig. 1). Once the dog had been in close contact with moose or was not able to find the moose, the handler leashed the dog and walked back to the car. Dog handlers participating in experimental hunts filled out a field protocol noting date and time, coordinates, moose ID, dog ID, hunting situation and weather. The dog's GPS track was later downloaded from the hand-held GPS.

Real hunts
When near, the UHF-ID tag of a hunting dog triggered the proximity sensor on the moose collar, and a message with the ID tag number of the dog was sent together with the GPS data from the moose. The signal strength depends on distance and topography between the UHF-ID tags and the moose collar and is expected to be between 300 and 400 meters (VectronicAerospace, 2019c). We then contacted the corresponding dog handler and met to download the tracklog of the dog, using a laptop computer and the basecamp software (Garmin Ltd, Olathe, Kansas, USA), and to collect information about the hunting situation. The field protocol was the same for both experimental and real hunts. The dog collars were programmed to collect a GPS fix every 1 s.

Data preparation and analyses
Twenty-three dog-moose encounters (hereafter, called 'approaches') occurred during real (n = 6) and experimental (n = 17) hunting situations on eight individual female moose (overview in the Supplementary material (Appendix 1)). Based on the distance between the moose and the dog as well as the duration of the contact, we classified approaches as successful (moose and dog within 240 meters for at least 10 min), disturbed (moose and dog were within 240 meters for less than 10 min) and not disturbed (moose and dog were never within 240 meters). The 240-meter distance threshold was based on maximum flight initiation distance reported from a study of the same moose approached by a skier in wintertime (Viljanen, 2019), and the 10-min threshold was based on the 10-min resolution of GPS positions. We tested for differences in moose behaviour (i.e. activity) and physiology (i.e. T b , HR) between the day of the approach (24 h) and a 7-day reference period before the approach and to 1 day after the approach using thereby each moose as its own control. We used 1 week prior to the beginning of the dog-training period (14-20 August) as the before-approach reference period, assuming moose under non-disturbed conditions, but within similar seasonal conditions. To test for difference in moose movement, we excluded the reference period, because only GPS data at 3-h intervals were available for this period. The distance travelled per day and the speed would, therefore, have been underestimated and would have led to a bias in our model. HR, T b and activity values from the reference period were generally higher than values from later in the hunting season, due to seasonal variations (Graesli et al., 2020).
We only included approaches classified as successful or disturbed in the analyses and excluded approaches that had missing data (n = 4), were classified as not disturbed (n = 2), proximity sensor triggered by a dog in a car (n = 1) and approaches carried out the day before another approach on the same moose (n = 2). This resulted in data from 14 approaches (successful n = 10, disturbed n = 4) on eight individual moose. In three of these approaches, the same moose was approached twice the same day because the field personnel considered the first approach of that day as unsuccessful. Consequently, this day was considered as one approach event.

Body temperature (T b )
We used a linear model from the lme4 package (Bates et al., 2015) to test for difference in the maximum daily T b (response variable), considering the explanatory variables Period (factor with three levels (reference period, day of approach and day after approach)) and Success (factor with two levels (disturbed and successful)). Maximum T b instead of mean T b was selected as the response variable to minimize the influence of drinking periods on our results, as drinking cold water drastically decreases rumen-measured T b (Herberg et al., 2018). Not every moose was experimentally approached repeatedly, which resulted in too few data points to include the  individual moose ID as a random structure. We based model selection on Akaike's information criterion corrected for small sample size (AICc) and carried it out with the function lCtab from the bbmle package (Bolker and R Core Team, 2017). We picked the highest ranked model with the highest AICc weight within AICc ≤2 and applied a post-hoc test with estimated marginal means from the emmeans package (Lenth et al., 2019) to estimate which levels of the categorical variables were significantly different.

Movement of the moose
We used GPS locations collected every 10 min when analysing the movement data. In the case of missing positions or delays in switching to 10-min positions (0.08% of the positions), we used the 'na.approx' function from the zoo package in R to linearly interpolate the longitude and latitude values (Zeileis and Grothendieck, 2005). We excluded one of the approaches because the GPS position frequency was not appropriate as the GPS failed to switch to 10-min intervals. In total, we analysed movement data from 13 approaches on eight moose.
We used the AdehabitatLT R package to calculate the Euclidean distance between consecutive GPS positions and calculated the total distance travelled per day (m) and the maximum speed (m/min) per day (Calenge, 2006). We modelled the maximum speed and the total distance travelled per day (response variables), using a gamma-distributed generalized linear model with the identity link function. We applied Period (factor with two levels; day of approach and day after approach) and Success (factor with two levels; disturbed and successful) as explanatory variables. A random structure for individual moose ID was not included (see the explanation for the body temperature model), and the model selection followed the same approach as for body temperature.

Heart rate
The Reveal XT calculated the HR based on recognition of R peaks in the ECG, and the ECGs were deleted (because of the storage limit) while the HR data were stored in the memory of the logger (Medtronic, 2017). The algorithm used for detection for R peaks was not always correct, likely because the Reveal XT is designed for use in human medicine, and heart anatomy and physiology of humans and moose differ. Because of that, some of the heartbeats were not detected and sometimes one heartbeat was detected as two (double counting). The absolute minimum HR of a moose resting during winter was found to be 37.5 bpm (Graesli et al., 2020), so all values lower than this were removed. Based on comparisons of the HR data from the Reveal XT with manually calculated HRs from ECGs obtained by another HR logger (DST centi HRT, which was deployed additionally as part of a separate study) and accelerometer data from the same moose, we found no correlation between activity level and HR accuracy and concluded that the algorithm performed similarly during both active and passive phases. Consequently, we used the obtained raw HR data and removed unrealistically low HRs. By doing this, we were able to calculate differences in HR and identify significant changes in HR over time. However, we were unable to account for potential double counting and can therefore not report specific mean or maximum HR values.
All experiments were carried out between 08.00 and 14.00 local time, and we therefore only assessed the HR measurements during this period of the day. To compare the HR data between the different periods, we built a linear mixed-effects model (nlme package) with the raw HR data and included the variable period (reference time, day of approach and day after approach) as a fixed factor (Pinheiro et al., 2019). To account for the autocorrelation within the HR data, we included the autocorrelation structure corCAR1. Moose ID was included as a random factor to account for repeated measures of the same moose. Due to the small sample size, the model could not include the explanatory variable Success and was therefore only built using the data of successful approaches.

Resting time (active versus inactive behaviour)
To classify moose behaviour into (1) inactive (i.e. resting) and (2) active, we fitted separate hidden Markov models (HMMs) to each individuals' observed activity data (i.e. summed acceleration of X-and Y-axes). HMMs are time-series models that assume animals' observed movement or activity patterns to be determined by an underlying 'hidden' finite state sequence, where the states can be interpreted as proxies for the unobserved behavioural modes of an animal (Patterson et al., 2009;Langrock et al., 2012). We modelled activity using a state-dependent gamma distribution. To account for potential effects of temporal patterns and temperature on moose behaviour, the state transition probabilities were expressed as functions of the time of day, light conditions (dark, light or twilight at time of observation), day of the year and collarderived ambient temperature using a logit link function (i.e. with the categories representing the two different states the process might switch to). We used forward selection based on Aikaike's Information Criterion (AIC) to determine the influence of these covariates considered in each of the individual HMMs. An overview of the individuals and models chosen are presented in the Supplementary material (Appendix 2). For cyclic covariates (day of year and time of day), we included sine and cosine terms. All HMMs were fitted via numerical likelihood maximization using the momentuHMM R package (McClintock and Michelot, 2018). Each model was run with 30 sets of random starting values to avoid local maxima (always choosing the model with the highest loglikelihood value). We then determined the most likely state sequence from each HMM using the Viterbi algorithm and calculated the total time spent resting per day (min) based on these results. Subsequently, we ran a linear model with resting time as the response variable and performed the model selection, following the same approach as for body temperature.

Behavioural responses
In order to detect flight initiation distances (FID-i.e. how close the dog was when the moose started to react behaviourally (based on collar activity (accelerometer) data) or physiologically (based on body temperature) and to estimate for how long collar activity levels and T b were affected by the approaches (i.e. how long it takes for both metrics to return back to the 'normal' pre-approach state), we applied change point analysis (R package changepoint), (Killick and Eckley, 2014). We applied the analysis over the course of the day of the approach and used collar activity data because it was recorded at a higher resolution than the GPS data (5-vs 10-min resolution). We identified two changepoints based on the variance in the data, which should ideally correspond to the start (time of FID) and end (settling down of the moose) of the approach.
We then calculated the time the moose was affected (hereafter: time affected = time between the two changepoints), compared the times of the changepoints with the GPS data of the dogs and calculated FID (Euclidean distance). In addition, we calculated the time affected after an approach ended (the difference between the time of the second changepoint and the time the approach ended (found in the protocol)). We included approaches carried out the day before another approach on the same moose (n = 2) in the changepoint analysis. We excluded those approaches where the moose was moving before the dog found the moose (n = 2) or where data were missing (n = 1). In total, we performed changepoint analysis on 16 approaches (disturbed n = 6, successful n = 10).
All the data were prepared and analysed using R version 3.6.1 (R Core Team, 2019). We considered P values <0.05 as significant.

Body temperature (T b ), movement of the moose, heart rate and resting time
All successfully approached moose fled away from the dogs and increased their HR, T b and movement when approached (see Fig. 2). None of the moose stood its ground or confronted the hunting dogs. Hunting dogs followed moose tracks closely (e.g. Fig. 3). The highest ranked model for each response variable included period for the movement models and the interaction between period and success for T b and resting (see Table 1).
We observed a significantly higher maximum T b (0.88 • C higher-SE 0.15 • C, P value <0.001), at the day of the approach compared to the day after for the successful approaches (see Table 2 and Fig. 4). In addition, the maximum T b on the day of the approach was 0.47 • C higher (SE 0.15 • C, P value 0.031) compared to the reference time. Highest T b measured in the study was 40.8 • C" which was recorded during an approach. We did not detect any differences in T b between the reference period, the day of the approach and the day after the approach among the moose that had been in contact with the dog <10 min (approaches classified as disturbed).
We found a higher maximum speed (mean ± SE: 76.5 ± 14.5 m/min faster; P value <0.001) and a longer distance travelled (mean ± SE: 4.1 ± 1.  at the day of the approach compared to the day after the approach for all moose (see Table 2 and Fig. 5).
HR values for the day of the approach were significantly higher compared to the reference period and the day after the approach (22 bpm (SE 2 bpm, P value <0.001) and 24 bpm (SE 3 bpm, P value <0.001), respectively) (see Table 3 and Fig. 6). The highest HR measured, which could be confirmed from stored ECG episodes, was 195 bpm.
Moose rested significantly longer (94.5 min (SE 20.6 min, P value <0.001)) the day after a successful approach compared to the day of the approach. In addition, we demonstrated that the successfully approached moose rested 96 min longer (SE 30.3 min, P value 0.036) the day after the approach than the moose in approaches characterized as disturbed (moose in contact with the dog <10 min) (see Table 2 and Fig. 7). The mean resting time per day in the reference period was 13.5 h.

Behavioural responses
Based on collar activity (accelerometer) data, we detected changepoints in 10 of the 16 approaches (disturbed n = 1, successful n = 9). Changepoints on T b data were detected in three successful approaches.   Moose T b was affected 14.5 min longer than the moose activity (median total time T b 132 min; range 124-160 min, versus collar activity 117.5 min; range 25-180 min). Similarly, we found that moose T b was on average affected 8 min longer than activity, after an approach ended (median time

Physiological and behavioural responses
This study, combining different types of biologgers, provides novel insight in the correlation between physiological and behavioural responses of moose to hunting with dogs. Combining T b , HR and movement data from biologgers uncovers the underlying mechanisms behind stress responses in moose and adds to the field of conservation medicine. We documented significant physiological and behavioural responses in approaches where the moose and dog had been in contact for >10 min. In contrast, we found minimal changes in behavioural and physiological parameters in approaches where the moose and dog had been in contact for <10 min (approaches classified as disturbed). Our results thus indicate that significant changes in moose behaviour and physiology dependent on the time the dog have been in close contact with the moose. However, we are aware of the small sample size of our study and that this limits the interpretation of our results.
We demonstrated T b in moose exceeding 40 • C in the majority of the successful hunts, and the highest T b measured was 40.8 • C. Brown bears hunted with baying dogs are documented to have an increased T b of 4.7 • C from their baseline (mean) levels to the maximum T b during hunting (37.5 • C versus 42.2 • C) (Evans et al., 2016;Støen et al., 2018). In moose, we found an increase of 2.3 • C from the mean T b baseline levels (Graesli et al., 2020) to the maximum T b (38.5 • C versus 40.8 • C). Long-time exposure to high T b could be life-threatening because of cytotoxicity. Damage of mammalian cells due to hyperthermia starts after a relatively short period with temperatures >40-41 • C, and the degree of damage depends on the exposure time and other stress factors (Lepock, 2003). It results in protein denaturation and impairment of the DNA synthesis in the cell, and longterm exposure leads to organ failure and death (Tansey and Johnson, 2015). Occasionally, we found a rapid decrease in

10
rumen temperature shortly after the end of the approaches (similar to the pattern visualized by Herberg et al. (2018)), suggesting that some of the moose drank water when reaching the highest T b levels. Drinking can, therefore, act as a behavioural thermoregulation strategy to avoid overheating (Tansey and Johnson, 2015).
We found a significant increase in HR, indicating increased energy consumption during the day of the approach. A significant correlation between HR and metabolic rate is demonstrated in a variety of animals, including moose (Green, 2011, Renecker andHudson, 1985). The lack of validated HR measurements (not possible to validate the exact HR values) and the non-linear relationship between HR and metabolic rate in moose (Renecker and Hudson, 1985) prevented us from making a valid estimate on the increased energy consumption in this study. Even though it was not quantified, our findings (22-24 bpm difference) suggest a considerably increased energy expenditure due to hunting related disturbances when comparing our findings with seasonal differences found in moose (Graesli et al., 2020). More specifically, the difference of 31.4 bpm in daily mean HR from the highest levels in summer (71.9 bpm) to the lowest daily mean in winter (40.5 bpm) represents a 60% decrease in metabolic rate from summer to winter (Graesli et al., 2020). Moose might compensate for the increased energy consumption from hunting by increasing the resting time the day after, as we demonstrated. Another reason for the increased resting time is likely to be recovery from exhaustion as suggested in bears (Le Grand et al., 2019). We recommend further studies to determine the effect of these events on the moose's time spent foraging the day after a hunting situation.
Moose populations have been declining along the edges of the moose range, and increasing ambient temperatures and climate change are suggested as a reason for that (Lenarz et al., 2009;Ruprecht et al., 2016;Allen et al., 2017). Moose are easily heat stressed with increasing ambient temperatures, resulting in increased respiratory rate, HR and body temperature (McCann et al., 2013, Renecker and Hudson, 1986, Thompson et al., 2019. When evaluating heat stress in moose one should consider core body temperature and the daily variations, body condition, solar radiation, vapour pressure and season in addition to ambient temperature (Thompson et al., 2019;Graesli et al., 2020). High ambient temperatures during the hunting season might, therefore, result in extra energy consumption in hunted moose. Increasing ambient temperatures from climate change can further result in changes in food resources and increased infection risk in addition to increased energy consumption, and all of these can result in lower survival and reproductive rates (McCann et al., 2013, van Beest and Milner, 2013, van Beest et al., 2012. We therefore suggest that the negative effects of hunting disturbance might be more dramatic in moose in the edges of the moose range, because of the negative effects of climate change, especially on warm hunting days. None of the moose stayed and confronted the attacker (dog or human)-which is in line with earlier findings (Ericsson et al., 2015b). In North American moose, confronting the attacker is a common anti-predator strategy towards fight off attacks from wolves (Ballard and van Ballenberghe, 2007). In Sweden, the main source of mortality for adult moose is hunting (Ericsson and Wallin, 2001). Ericsson et al. (2015b) suggested that female moose may have altered their antipredator behaviour towards hunting. According to Swedish hunting laws any accompanying calf has to be shot before the female (SOU, 2009), providing a learning experience for longliving female moose. As a result, Swedish female moose might be more prone to flee compared to their North American conspecifics. Our results support previous research that hunting disturbances increase travel distance and higher maximum speed on the day of the approach compared to the day after disturbance in moose (Ericsson et al., 2015b;Sand et al., 2016), red deer (Jarnemo andWikenros, 2014, Sunde et al., 2009) and brown bear (Le Grand et al., 2019). In red deer, it is documented that escape strategies are linked to habitat type: red deer in fragmented and more open landscapes fled more often for longer distances, and at a higher speed, than red deer in homogenous forest landscape (Jarnemo andWikenros, 2014, Sunde et al., 2009). It is likely that moose have the same type of response, yet, this aspect was beyond the scope of our study, and we suggest further research on the interplay between habitat type and physiological responses.
Moving in a sinuous pattern is possibly an antipredator strategy by moose, increasing the chance that the dog loses the track or switches to the track of another moose (Baskin et al., 2004, Cederlund and Kjellander, 1991, Ericsson et al., 2015b. Even if not quantified by our GPS analysis, the tracks of the moose and the hunting dog indicated that moose often moved in a sinuous pattern when approached. Our results might, therefore, have been more precise if we used one instead of 10-min GPS intervals when calculating the FID. Including wind as a parameter is valuable in explaining the antipredator strategy of moose and other prey species, as a variable to sense the location of the predator. Sometimes the hunting dogs picked up non-target moose instead; in these cases, the target moose was within 100 meters but did not move away even though a hunting dog was in close vicinity. These results are in contrast to the results of a study on the effects of hunting with dogs on roe deer as a non-target species, demonstrating that roe deer behaviour was significantly affected (Grignolio et al., 2011). The differences between the studies might be due to different hunting strategies such as the use of different types and numbers of dogs, former experiences with hunting, species-specific behaviour and anti-predator strategies, and that some of the roe deer actually were pursued by the dogs (even if non-targeted) (Grignolio et al., 2011).
About 45% of the moose observed by hunters in our study area were shot, meaning that >50% survived a hunting approach (Länsstyrelserna, 2019). There is not any data available on the number of surviving moose that have been

11
chased within a given area. Yes, we assume that the amount of stress from the hunting approaches, for a given surviving moose, will vary and depends on the length of the chase, if moose were aware of hunters/dogs, their previous hunting experience, and if any in their company were shot. In spite of the documented overlap in oestrus and moose hunt in southern Sweden (Malmsten et al., 2014), moose in our study area had a fairly high and stable (same level the past 20 years) reproductive rate in 2017 (>0.7) Wallin, 1999, SvenskaJägareforbundet, 2019). The slaughter weights of the moose in our study area are also stable over time (Svenska-Jägareforbundet, 2019), suggesting that the hunting disturbances must be stable. Most harvest occurs in the beginning of the hunting season (Singh et al., 2014). Yet, to improve moose management reducing moose density in areas with high browsing damages in winter, hunting at the end of the hunting period (in winter) is increasingly discussed. Further studies should include evaluation of the effects of hunting during wintertime when the moose display hypometabolism and compare the physiological and behavioural responses during autumn and winter.

Conclusion
Based on the physiological and behavioural results, we can conclude that hunting with baying dogs represents a notable stress event for the individual moose. Moose might compensate for the increased energy consumption by increasing the resting time on the day after being chased by a hunting dog. Yet, increased frequency of hunting disturbances and higher ambient temperature due to climate change may likely increase the energy consumption and thereby enhance the risk of negative effects on reproduction and body condition. As hunting methods and breeds of hunting dogs used are continuously evolving, managers need to continue to consider the physiological consequences for moose.

Author contributions
J.M.A., A.L.E., G.E., W.N., N.J.S., B.F., F.S. and A.R.G. designed and initiated the study. Data collection was performed by A.R.G., A.L.E., F.S., J.M.A. and B.F. T.G.L. contributed with input for data collection and provided equipment. Data management was done by A.T., B.F., L.G. and A.R.G. A.T., L.G. and A.R.G. did the statistical analysis with advice from B.F., O.D. and L.T.B. A.R.G. led the writing of the manuscript. All authors reviewed the manuscript and approved the submitted version.

Funding
This work was supported by Miljødirektoratet (the Norwegian Environmental Agency, grant numbers 10040125 and 16040078). Inland Norway University of Applied Sciences provided salary for A.R.G. The study was in collaboration with the research program 'Beyond Mooseecology and management of multispecies ungulate systems'. Beyond Moose is financially supported by the Swedish Environmental Protection Agency, Kempestiftelserna, the Swedish Association for Hunting and Wildlife Management (Västerbotten county's älgvårdsfonden) and SLU's Faculty of Forest Sciences.