Acceleration data reveal behavioural responses to hunting risk in Scandinavian brown bears
Data files
Jul 18, 2025 version files 2.45 GB
-
bear_activity_patterns_data.csv
2.44 GB
-
bear_daily_distances.csv
34.66 KB
-
random_forest_training_dataset.csv
6.80 MB
-
README.md
5.13 KB
Abstract
Predation may indirectly influence prey’s fitness and population dynamics through behavioural adjustments in response to perceived predation risk. These non-consumptive effects of predation can also arise from hunting by humans, but they remain less documented. Advances in biologging allow detailed assessments of the activity budgets of elusive wildlife, increasing the potential to uncover the non-consumptive effects of human activities on animals. We used tri-axial accelerometry to record the daily activity of 24 Scandinavian brown bears (20 females and four males) from a heavily hunted population in Sweden, for a total of 29 bear-years (2015-2022). We used a random forest algorithm trained with observations of captive brown bears to classify the accelerometry data into four behaviours, running, walking, feeding, and resting, with an overall precision of 95%. We then used these classifications to evaluate changes in bear activity budgets before and during the hunting season. Bears exhibited a bimodal daily activity pattern, being most active at dusk and dawn, and resting around midday and midnight. However, during the hunting season, males became more nocturnal compared to before the hunting season, suggesting a proactive behavioural adjustment to reduce encounters with hunters. Females showed the opposite pattern and had a higher probability of being active during the day, potentially to increase nutritional gains before denning. Additionally, daily number of running bouts did not vary between the pre-hunting and hunting seasons in both sexes, but females’ proportion of running bouts occurring during legal hunting hours was higher during the hunting season than prior to it, which suggest a reactive behavioural adjustment to encounters with hunters. Detailed assessments of wild animal behaviours, allowed through recording of movement data at high frequencies, have the potential to improve our understanding of the impacts of human activity on wildlife.
https://doi.org/10.5061/dryad.pc866t214
Description of the data and file structure
Updated 2025-07-18
Clermont et al. 2025 - Acceleration data reveal behavioural responses to hunting risk in Scandinavian brown bears, Ecology and Evolution, https://doi.org/10.1002/ece3.71489
Code/software
R codes stored at Zenodo: https://doi.org/10.5281/zenodo.14884698
Files and variables
Description of datasets and scripts included, by analysis (see Methods):
1. Behavioural classification algorithm analysis:
Data: random_forest_training_dataset.csv
R Script: random_forest_bear_behaviour.R
Description of variables:
- start_UTC_timestamp: timestamp in UTC at the start of the 3s sequence
- Subject: the name of the bear. Eternity is the young female, Freja is the older one.
- 36 summary statistics used in the random forest analysis, all calculated over a 3s sequence:
- mean_x: mean of raw acceleration of x
- mean_y: mean of raw acceleration of y
- mean_z: mean of raw acceleration of z
- mean_magnitude: mean of magnitude. Magnitude is the sqrt of sums of squares of the acceleration in x, y, z
- std_x: standard deviation of x
- std_y: standard deviation of y
- std_z: standard deviation of z
- std_magnitude: standard deviation of magnitude
- max_x: maximum value of x
- max_y: maximum value of y
- max_z: maximum value of z
- max_magnitude: maximum value of magnitude
- min_x: minimum value of x
- min_y: minimum value of y
- min_z: minimum value of z
- min_magnitude: minimum value of magnitude
- cor_xy: Pearson's correlation coefficient between x and y
- cor_xz: Pearson's correlation coefficient between x and z
- cor_yz: Pearson's correlation coefficient between y and z
- mean_dba_x: mean of DBA of x. DBA is the dynamic body acceleration, i.e. raw acceleration minus static acceleration calculated as a 3s running mean of raw acceleration
- mean_dba_y: mean of DBA of y
- mean_dba_z: mean of DBA of z
- ODBA_total: sum of ODBA. ODBA is the overall dynamic body acceleration, i.e. the sum of absolute DBA over all axes
- mean_ODBA: mean of ODBA
- kurtosis_x: measure of weight of the tails relative to a normal distribution of x
- kurtosis_y: measure of weight of the tails relative to a normal distribution of y
- kurtosis_z: measure of weight of the tails relative to a normal distribution of z
- kurtosis_magnitude: measure of weight of the tails relative to a normal distribution of magnitude
- skew_x: skewness, measure of symmetry of the distribution of x
- skew_y: skewness, measure of symmetry of the distribution of y
- skew_z: skewness, measure of symmetry of the distribution of z
- skew_magnitude: skewness, measure of symmetry of the distribution of magnitude
- dominant_power_spectrum_X: maximum power spectral density of x
- dominant_power_spectrum_Y: maximum power spectral density of y
- dominant_power_spectrum_Z: maximum power spectral density of z
- dominant_power_spectrum_magnitude: maximum power spectral density of magnitude
- behavior: behavior of the bear during that 3s sequence, either resting, feeding, walking, running, based on videos of captive bears
2. Wild brown bear activity patterns analysis:
Data: bear_activity_patterns_data.csv and bear_daily_distances.csv
R Script: analyses_bear_activity_patterns.R
Description of variables for bear_activity_patterns_data.csv:
- bear_ID : unique bear identifier
- start_timestamp_UTC : timestamp in UTC at the start of the 3s sequence
- year
- day: day of August, local time (Sweden, UTC+2)
- ToD: numeric time of day, local time (Sweden, UTC+2)
- hunting_period: either pre-hunting or during hunting
- demo_group: demographic group, either female with offspring, subadult solitary female, adult solitary female, male
- resting: whether the bear is resting during that 3s sequence (0,1), as determined by the trained random forest
- feeding: whether the bear is feeding during that 3s sequence (0,1), as determined by the trained random forest
- walking: whether the bear is walking during that 3s sequence (0,1), as determined by the trained random forest
- running: whether the bear is running during that 3s sequence (0,1), as determined by the trained random forest
- feedwalking: whether the bear is feeding or walking during that 3s sequence (0,1)
- behavior: behavior of the bear during that 3s sequence, either resting, feeding, walking, running, as determined by the trained random forest
Description of variables for bear_daily_distances.csv:
- bear_ID : unique bear identifier
- year
- day: day of August, local time (Sweden, UTC+2)
- hunting_period: either pre-hunting or during hunting
- demo_group: demographic group, either female with offspring, subadult solitary female, adult solitary female, male
- daily_distance : daily distance travelled (m)
There are two main analyses associated with the datasets and scripts provided in this submission. Here, we provide an overview of both analyses but invite readers to read the associated manuscript for further details.
Behavioural classification algorithm
We classified the behaviours of captive brown bears using a random forest algorithm. To do so we used raw acceleration data on three axes recorded at a frequency of 8Hz. Raw acceleration data were divided into 3s samples, and for each sample, we calculated summary statistics (dataset "random_forest_algorithm_dataset.csv"). These statistics included measures like mean, standard deviation, minimum, maximum, skewness, kurtosis, correlations between axes, and dominant power spectrum. We then trained the random forest using known behaviour labels, obtained through visual observations of the captive bears. We build multiple decision trees and combined their results (script "random_forest_bear_behaviour.R").
Wild brown bear activity patterns
We then used the trained random forest algorithm to determine the behaviours of wild bears. We used raw acceleration data on three axes recorded at a frequency of 8Hz, divided into 3s samples, for which we calculated the same summary statistics. From this we obtained the dataset "bear_activity_patterns_data.csv" with, for each individual, a behaviour every 3s during the study period. This was used to evaluate patterns in bear activity (script "analyses_bear_activity_patterns.R"). Our main focus was to evaluate how daily activity patterns differ between two distinct periods, i.e. the prehunting and the hunting seasons, and across different demographic groups. We also evaluated how the daily distance travelled by bears (obtained by adding linear distances between GPS locations) varied between periods (dataset "bear_daily_distances.csv").