Evidence for a by-product mutualism in a group hunter depends on prey movement state
Data files
Jul 30, 2024 version files 33.40 MB
Abstract
Group-hunting animals have been shown to engage their prey in long pursuits (kms) to increase capture success, but the evidence for this is limited to a few terrestrial species. This predation strategy may be effective in the open-ocean, where group-hunts are characterised by large predator-prey size ratios and there are few places for prey to hide. However, recording the attack rates, capture rates and capture success probabilities of grouping predators during long-distance pursuits through the open ocean, remains a significant challenge. Here we use a combination of underwater and aerial video to assess the attack and capture behaviour of striped marlin, Kajikia audax, in relation to the movement of their schooling prey (‘stationary’ and ‘mobile’ movement states) during group hunts. We found that the attack rate was higher when the prey school was ‘stationary’ and, during these times, individual marlin could make longer sequences of uninterrupted attacks on the prey school, effectively monopolising the prey resource for short periods over conspecifics. When the prey school was ‘mobile’, the attack pattern changed and individual marlin alternated their attacks on the prey school more frequently. Capture success probability (likelihood of capture per attack) was not affected by prey movement state. However, attacks on ‘mobile’ schools resulted in more prey fish becoming isolated from the school, and these isolated prey were easily captured - not by the attacker but by conspecifics swimming nearby. The likelihood of a prey becoming isolated from a ‘mobile’ school was predicted by prey school speed, with a 0.1 m.s-1 increase leading to a 23.5% increase in the chance of isolation. Altogether this meant the capture rate at the group level was highest when the prey school was ‘mobile’ rather than ‘stationary’ because the predators that were pursuing the ‘mobile’ prey school had an increased availability of isolated and easy to capture prey. This mechanism provides evidence for a by-product mutualism selecting for conspecific group formation and we discuss our findings in consideration of the social structure of the hunting group and the type of prey.
Hansen, Matthew (Forthcoming 2024)
https://doi.org/10.5061/dryad.5x69p8dc9
Description of the data and file structure for the 4 sets of analysis conducted in the manuscript:
1.) The relationship between predator behaviour and differences in prey
2.) Predator positions during attacks isolate analysis
3.) Predator positions during attacks stat mob analysis
4.) Drivers of isolation events
email:mjhansen.sci@gmail.com
1.) the_relationship_between_predator_behaviour_and_differences_in_prey.csv
Data set contains time stamped behavioural data extracted from video recordings of group hunting striped marlin attacking schools of prey fish in the open ocean.
Every row within the provided data frame relates to a predator attacking the prey school or capture of an isolated prey item
BB: Prey school ID.
Time: Timestamp of predator attack in seconds.
BB_2: Sequences of consistent prey motion state within individual prey schools.
Time2: Timestamp of predator attack in seconds within BB_2.
Year: Year of video recording.
Size: Estimated prey school size.
NewSeq: Binary variable indicating if a predator attack is happening within a continuous sequence of attacks by the same individual (1=New sequence by new individual).
Lost_visual: Binary variable indicating if video footage was removed between events due to bad visibility or obstruction (1 = yes).
Time_lost: Relates to Lost_visual. Indicates numerical value of seconds removed between two events.
State: Prey motion state (stat = stationary, mov = mobile).
TimebwAttacks: Numerical. Time difference in seconds between two consecutive predator attacks.
Capture: Binary variable indicating if attack resulted in a capture by the focal individual (1 = yes).
Isolated: Binary variable indicating if the row relates to the capture of a isolated prey fish (1 = yes).
DashNum: Count data of predator attack within a continuous sequence of attacks by the same individual.
CaptureDashNum: Count data. Combination of DasNum and Capture. Indicates on which attack in a sequence a prey fish was captured.
EndDashNum: Count on the last attack within an attack sequence.
CapRecord: Binary variable indicating if data was used in analysis related to predator capture performance (1 = yes).
CaptureCum: Cumulative count of captures within BB_2.
DashCum: Cumulative count of predator attacks within BB_2.
reapproach: Binary variable indiucating if a predator succesfully reapproaches the prey after the current attack (1 = yes).
size_cat: Factorial variable of prey size category (3lvl: small, medium, large).
recording_mode: Factorial variable of footage type used for video analysis (3lvl: underwater = uw, ae = aerial, uw_ae = both types)
missing data code: NA
Code/Software
Behavioural data from video recordings was timestamped with BORIS (https://www.boris.unito.it/; v. 7.13.9). All analysis was performed in Rstudio (v.4.1.2). For specific libraries and repositories used for the analysis see method section of main manuscript. Further information and code can be supplied by the corresponding author upon reasonable request.
2.) predator_positions_during_attacks_isolate_analysis.csv
Data set contains data of individual striped marlins attacking schooling prey in the open ocean. Data was extracted from still frames of videos recorded with unmanned aerial vehicles. This data set can be used to investigate differences in orientation and distances to the prey school of individual predators.
Individual rows relate to positions of individual predators during an attack event.
BB: Prey school ID.
BBatk: Attack event ID.
type: Factorial (3 lvl). Future role of the focal predator during the analysed attack. Next_attacker: Focal individual will gain prey access in the next attack. Random: Individual is a random group member. Scrounger: Individual gains access to an isolated prey item produced by the focal attack.
distance_m: Distance of focal predator to the prey school in meter.
Orientation: Heading in degrees of the individual predator.
State: Factorial (2 lvl). Motion state of the prey school.
distance_cm: Distance of focal predator to the prey school in centimeter.
Code/Software
Coordinate data was extracted from still frames with Biotracker (https://arxiv.org/abs/1803.07985). All analysis was performed with Rstudio (v.4.1.2). For specific libraries and repositories used for the analysis see method section of main manuscript. Further information and code can be supplied by the corresponding author upon reasonable request.
3.) predator_positions_during_attacks_stat_mob_analysis.csv
Data set contains coordinate data of individual striped marlins attacking schooling prey in the open ocean. Data was extracted from still frames of videos recorded with unmanned aerial vehicles. This data set can be used to investigate differences in orientation and distances to the prey school between a focal attacking individual and the current position of the individual that will gain prey access in the future.
BB_state: Prey motion state (stationary/mobile).
Pred_X: x-coordinate in px of the focal attacking individual.
Pred_Y: y-coordinate in px of the focal attacking individual.
Pred_Angle: Heading in degrees for focal individual.
Next_X: x-coordinate in px of the next attacking individual.
Next_Y: y-coordinate in px of the next attacking individual.
Next_Angle: Heading in degrees for next attacking individual.
Diff_angle: Angular difference in degree between heading of the focal and the next attacking individual.
Dist_BB: Distance to the prey school of the next attacking individual.
Code/Software
Coordinate data was extracted from still frames with Biotracker (https://arxiv.org/abs/1803.07985). All analysis was performed with Rstudio (v.4.1.2). For specific libraries and repositories used for the analysis see method section of main manuscript. Further information and code can be supplied by the corresponding author upon reasonable request.
4.) drivers_of_isolation_events.CSV
Data contains output of a custom build computer vision pipeline to gain trajectories from free ranging fish schools filmed by unmanned aerial vehicles in the open ocean. Data was pre-filtered to include only location data predicted with a confidence > 100. Predicted locations of prey schools were synchronized with timestamped predator behaviour (see file: # the_relationship_between_predator_behaviour_and_differences_in_prey.csv in this repository) to investigate the relationship between a prey schools trajectory and the occurrence of isolated prey fish.
avg_pos_y: Y-Coordinate value of prey school location in meter.
avg_pos_x: X-Coordinate value of prey school location in meter.
avg_vel_y: Velocity of prey school in y direction [m/s].
avg_vel_x: Velocity of prey school in x direction [m/s].
avg_vel: Velocity of prey school [m/s].
avg_pixel_pos_y: Y-Coordinate value of prey school location in pixel.
avg_pixel_pos_x: X-Coordinate value of prey school location in pixel.
vel_pixel_pos_y: Velocity of prey school in y direction in pixel.
vel_pixel_pos_x: Velocity of prey school in x direction in pixel.
angle: Heading of the prey school in degrees (-180 | 180). |
bb_video: Combination of prey school ID and individual video.
bb: Prey school ID.
video: individual video ID
frame: frame in video
confidence: Confidence of prediction by automated prey school tracking algorithm (200 = maximum confidence).
fps: fps rate of video
time_s.x: Timestamp calculated from frame and fps column.
id: See column bb_video
SheetName: See column bb_video
Time: Timestamp exported from BORIS annotation of predator behavioural data
Media.file.path: Individual file path
Total.lenght: Video length in seconds.
Behavior: Type of recorded predator behaviour.
state: Motion state of prey school (data set contains only trajectories of mobile prey schools).
ime_s.y: Timestamp calculated from frame and fps column.
success: Binary variable indicating if attack resulted in a capture by the focal individual (1 = yes).
attack: Binary variable indicating if behavioural event is classified as predator attack on the prey school (1 = yes).
inSeq:Binary variable indicatig a predator attack within a continuous sequence of attacks by the same individual (1 = yes).
atk_success: Binary variable indicating if attack resulted in a capture by the focal individual (1 = yes).
lengthSeq: Count data of predator attack within a continuous sequence of attacks by the same individual.
angle360: Heading in degrees (360).
angle_diff: Difference in heading to preceeding row.
time_s: Difference in seconds to preceeding row.
dist_m: Difference in distance travelled to preceeding row in meter.
Iso: Cumulative count of isolation events.
Iso_id: Unique ID of individual isolation events.
cdist15: Cumulative distance in meter travelled by the prey school in a preceeding time window of 15 seconds.
ldist15: Linear distance in meter travelled by the prey school in a preceeding time window of 15 seconds.
cangle15: Cumulative sum of difference in heading by the prey school in a preceeding time window of 15 seconds.
attacks15: Cumulative of predator attacks on the prey school in a preceeding time window of 15 seconds.
Iso_treatment: Binary variable indicating if the values in row are attributed to an isolation event (1 = yes).
x_smooth: Smoothed X-Coordinate value of prey school location in meter.
y_smooth: Smoothed Ý-Coordinate value of prey school location in meter.
dist_m_smooth: Difference in distance travelled to preceeding row in meter based on smoothed coordinates.
cdist_smooth_15: Cumulative distance in meter travelled by the prey school in a preceeding time window of 15 seconds based on smoothed trajectories.
vel_15: Average velocity [m/s] of the prey schoo in a preceeding time window of 15 seconds based on smoothed trajectories.
speed_15: Average [m/s] of the prey schoo in a preceeding time window of 15 seconds.
speed_smooth_15: Average [m/s] of the prey school in a preceeding time window of 15 seconds based on smoothed trajectories.
radangle: Difference in heading angle in radiants to the preceeding row.
tort_15: Average tortuosity (calculated as the ratio between cumulative distance and linear distance) in a preceeding time window of 15 seconds based on smoothed trajectories.
Empty cells are all “NA”. These are cells with no corresponding timestamped predator behaviour from file: # the_relationship_between_predator_behaviour_and_differences_in_prey.csv (in this repository)
Code/Software
All analysis was performed in Rstudio (v.4.1.2). Code to reproduce statistical analysis of drivers of isolation event in R is included in this repository: 240719_suppMat.html