Between February and mid-March (2018-2021), we immobilized 39 adult
female moose from a helicopter using a dart gun injecting etorphine and
xylazine (Sandegren et al., 1987). The handling protocols adhered to the
ethical requirements for research on wild animals in both Sweden
(decision C281/6 and C315/6) and Norway (decision id 15370). Each female
moose was fitted with a GPS collar (Vectronics Aerospace GmbH, Berlin,
Germany) programmed to record a position every 2 hours.
During the calving season (May-July), we monitored potential calving
events for each collared female using a rolling minimum convex polygon
(rMCP) method described by Nicholson et al. (2019). We employed a
12-point moving window to calculate the area of the rMCP within a
24-hour period. This timeframe allowed us to include both daytime and
nighttime positions and detect changes in movement and space use.
Potential parturition events were defined as the earliest date when the
mean of overlapping rMCPs remained below 1.7 hectares for approximately
72 hours. This criterion was based on the behaviour observed by McGraw
et al. (2014) in female moose in Minnesota, where extensive movements
over large distances occurred just before parturition, followed by a
period of minimal movement (presumably just after parturition; 1.72 ±
0.48 hectares for 7 days; McGraw et al., 2014). We plotted the
overlapping rMCPs for every day from the beginning of May and estimated
calving events by identifying spikes in movement prior to periods of
little movement (area < 1.7 ha) (Fig. 1 in Supplementary
Material). To identify the spike in movement before parturition, we set
a 25-ha threshold after which a search algorithm was initiated to detect
a 72-hour period where the average area of the 24-hour rMCPs was below
1.7 ha (see Nicholson et al., 2019 for details).
Once a calving event was identified, we located the female moose and
approached her silently to visually confirm the presence and number of
newborn calves. The approach was carried out on foot and using a
handheld VHF receiver to locate the female (RX98, Followit, Lindesberg,
Sweden). In some cases, drones were employed, flying to the last known
GPS position of the female moose and hovering over the area searching
for the moose. We waited a minimum of 2 days and a maximum of 7 days
from the assumed calving event before each approach. This interval
allowed time for the females to bond with their calves, reducing the
risk of calf abandonment. Additionally, it enabled us to collect calf
survival data within the first few days after parturition, when calf
mortality is typically high.
Once we had visual contact with the female moose, we waited until we had
visual confirmation of the number of calves present. In cases where the
reproductive status of a female was uncertain during the initial
approach, a second approach was conducted 2 to 7 days later to gather
additional information. Females that did not exhibit movement behavior
indicating calving were approached between the end of June and mid-July
to confirm the absence of calf/calves. We are aware that this method
does not determine calving success with absolute certainty, since
neonates could have died just after birth and prior to surveys;
nonetheless, it represents a reasonably unbiased method to measure the
relative production of calves in early summer. Throughout the study,
there were only two instances where female moose displayed calving
behaviour, but subsequent field checks revealed no presence of calves.
Within one month prior to the onset of the hunting season, each female
moose was re-approached to count the number of calves again using the
same procedure as in early summer. The hunting season in Norway starts
the 25th of September, whereas it differs within the
Swedish part of the study area, starting either in the beginning of
September or on the second Monday of October. Accordingly, our approach
timing was strategically planned to coincide with the hunting initiation
in each respective area, ensuring that the estimates remained unaffected
by the onset of hunting activities. Lastly, a final verification of the
number of calves accompanying each female was performed before natal
dispersal, specifically in April of the following year. This procedure,
encompassing calf-checks after birth, prior to hunting, and prior to
dispersal, was carried out for the study years 2019-2020, 2020-2021, and
2021-2022.
Moose migratory strategy and home
ranges
To classify the migratory strategies of each female moose, we employed
the Net Squared Displacement method (NSD) (Bunnefeld et al., 2011; Singh
et al., 2016). This approach enables the differentiation of various
movement strategies, such as migration and residency (Bunnefeld et al.,
2011; Singh et al., 2016; Börger and Fryxell, 2012), by analysing the
displacement patterns of individual animals using non-linear mixed
effects models (Singh et al., 2016). The NSD method characterizes
migration as a double sigmoid or s-shaped function, which repeats within
a year and involves the animal returning to its departure location
(Singh et al., 2016).
We used GPS positions to estimate seasonal home ranges for each female
moose using 95% minimum convex polygons (MCP). The MCPs were calculated
separately for the summer (1st of May – 31st of August; mean = 39
km2, range = 15 – 88 km2) and
autumn-winter (1st of September until 30th of April; mean = 42
km2, range = 13 – 100 km2) periods
with the amt package in R (Signer et al., 2019). Using the start and end
date of migration (based on NSD), we excluded GPS positions during
migration from our MCP analysis.
Calf survival
Harvest density
We used harvest data from both Norway and Sweden at the moose management
area level (see Wikenros et al., 2020; for details regarding the moose
management systems in Norway and Sweden). We calculated harvest density
as the number of harvested moose per km2 (range = 0.11
– 0.44 moose/km2) and extracted the average harvest
density within the autumn-winter home range for each female moose.
Large carnivores
Wolves
Wolves belonging to four packs, all having cross border territories
along the Swedish-Norwegian border, were located on snow, and
immobilized by darting from helicopter (see Sand et al. 2006; Arnemo and
Evans 2017). Handling protocols fulfilled the ethical requirements for
research on wild animals in Sweden (decision C281/6 and C315/6) and
Norway (The Norwegian Food Safety Authority, decision id 15370). The
collars were programmed to acquire one position every four hours. We
created 95% minimum convex polygons (MCPs) using GPS positions from the
scent-marking adult breeders to represent the territory of each wolf
pack during summer (May – August) and autumn-winter (September –
April) with the package amt in R (Signer et al., 2019). In cases
where multiple adult wolves were collared within a wolf pack during our
study period, we prioritized data from the individual with the most
extensive collar operation during the specific time interval (referred
to as the ”main individual”). For periods when location data from the
main individual were unavailable, we supplemented the dataset with
relocations from the other collared adult. We checked for overlap
between moose and wolf home ranges during both summer and autumn-winter
and categorized each moose home range to be overlapping (entirely or
partially) or not overlapping with the wolf home range (0=no overlap;
1=overlap), depending on whether they were inside or outside a wolf
territory.
Bears
To estimate bear density within the study area, we used density raster
maps provided by Bischof et al. (2020). These maps provided estimates of
bear density as the number of bears per square kilometer. By overlaying
the bear density raster maps with the summer home ranges of each female
moose, we extracted the average bear density for each moose’s respective
summer home range.
Environmental covariates
We obtained data on young forests from the Corine Land Cover (CLC)
inventory (Copernicus Land Monitoring, 2018). Using this inventory, we
calculated the proportion of each female moose’s home range covered by
young forests for each season (summer and autumn/winter) and year (mean
= 15 ± 7; range = 0 – 36%).
We used cumulative average winter snow depth from October to March of
each study year as a proxy for winter severity. The data was obtained
from the Norwegian Water Resources and Energy Directorate (NVE) (for
more information on the interpolation method used see Saloranta 2012).
For each female moose home range in autumn-winter in each year we then
extracted average snow depth.
Normalized Difference Vegetation Index (NDVI) is a measure of
photosynthetic activity at landscape scales that is often used to as a
proxy for plant productivity and nutritional status (Pettorelli et al.
2011). We obtained data on NDVI from Copernicus Global Land Service (300
m resolution raster) and used weekly NDVI cell values to calculate the
mean cumulative summer NDVI for each moose home range (mean = 0.73 ±
0.08; range = 0.53 – 0.86).
Hunting and wolf predation risk
metrics
We used previously estimated relative risks of human hunting and wolf
predation for moose in our study area, based on locations of wolf-killed
and hunter-killed moose during autumn-winter (September-April; Ausilio
et al., 2022). Ausilio et al. (2022) modelled hunting and wolf predation
risk separately using logistic regressions with the relative probability
of a location being a kill site or a random location as a function of
different landscape features (distance to bogs, young forests, main and
secondary roads, elevation, building density and terrain ruggedness).
The estimates represent the odds ratio (relative risk) of being killed
by wolves or hunters (hereafter, relative hunting risk and relative wolf
predation risk). The odds ratios for each given location within the
study area were then plotted as raster layers (25x25 m). For more
details on the methods used to estimate wolf predation risk and human
hunting risk, see Ausilio et al., (2022). We extracted the odds ratio of
hunting and wolf predation risk for a subset (n = 31 females) of moose
home ranges that overlapped with the risk maps from Ausilio et al.
(2022).
Survival analyses
To investigate the factors influencing calf survival during summer and
autumn-winter, we used Cox Proportional-Hazard models. These models
allowed us to analyse the probability of survival (0 = survived; 1 =
died) while considering the repeated checks of individual moose by
including moose ID as a covariate. For each season, we initially
constructed a full model that included all predictors and interactions
based on our hypotheses. To refine the models, we employed stepwise
backwards selection, eliminating interactions with p > 0.10
and individual predictors with p > 0.05.
The summer model included the additive and linear effects of the
following predictors: year (3-level factor), wolf presence, bear
density, proportion of young forests, migratory strategy of the mother
(migratory versus stationary) and NDVI.
The autumn-winter model included the additive and linear effects of year
(3-level factor), harvest density, wolf presence, snow depth, proportion
of clear-cuts/young forests and the migratory strategy. We included the
interaction between wolf presence and snow depth to test whether calf
survival was lower with deeper snow in the presence of wolves compared
to the absence of wolves.
Lastly, we used a subset of female moose to test the effect of hunting
and wolf predation risk on autumn-winter calf survival. This
supplementary analysis allowed us to evaluate whether the mortality risk
metric developed by Ausilio et al. (2022) correlates with calf survival,
but also to account for smaller spatial-scale variation in hunting risk
compared to overall harvest density, which might be too coarse to detect
changes in survival at the individual-level. For this subset of the
data, we assumed that calf survival was a function of the additive and
linear effects of hunting risk, wolf predation risk and year (3-level
factor). All continuous variables were scaled prior to analysis to have
mean = 0 and standard deviation = 1. We could not use all female moose
for this analysis because we lacked data about wolf predation risk for
2021/22, hence why we selected only a subset of moose overlapping in
time with our previously estimated metric of wolf predation risk
(Ausilio et al., 2022).