The early bird gets the worm: foraging strategies of wild songbirds lead to

elbowcheepΤεχνίτη Νοημοσύνη και Ρομποτική

15 Οκτ 2013 (πριν από 3 χρόνια και 8 μήνες)

74 εμφανίσεις



The early bird gets the worm:
foraging strategies of wild songbirds
lead to

the early discovery of food sources


Damien R. Farine and Stephen D. Lang



SUPPLEMENTARY METHODS



Deployment times


Deployments of feeders were designed to fit equally between dawn and dusk.
Due to travel times between each deployed feeder, morning sites were deployed
between 7am and 8am, with sunrise occurring at 7am at the start of our study.
These were then removed

between 12pm and 1pm, exactly five hours after being
deployed. Afternoon treatments were set out between 12pm and 1pm (after
being collected from morning sites), and removed between 5pm and 6pm.
Sunset at the start of our study was 5:23pm.


Permutatio
n test


Due to the large number of undiscovered sites, and the potential for social effects
in the number of individuals that arrived at a given site, we used a permutation
test to examine the difference between morning and afternoon treatments.
The
same
permutation test was conducted on each half
-
hour segment (from 0 to 5
hours) by comparing observed difference in the mean number of individuals that
arrived at morning treatments and afternoon treatments with the posterior
distribution of differences. For

each permutation, we
randomly swapped

feeder
s

between

morning and afternoon treatments and calculated the

resulting

difference in the means between
these two resulting groups
. This process was
repeated 1000 times to generate the posterior distribution of

random
ised
differences. The reported
P
-
values

are

the proportion of
times
the

difference

between the am and pm means from a permutation

was

greater than or equal to
the observed difference. Given that discoveries may not have been independent,
we random
ised all individuals at a feeder
in a given half
-
hour
as one
observation

rather than re
-
distributing each individual independently.



Group size over the day


In order to test for the

potential conf
ounding effect that larger group sizes may
result in a g
reater number of individuals discovering sites in the morning rather
than the afternoon, we examined the group size distribution for each half
-
hour
period in the day from the permanent grid of feeders. In order to infer groups in
this stream of data, we u
sed Gaussian mixture models

[
1
]
. This approach has
previously been used for inferring social network structure in this population
[
2
]
.

It determines regions of temporally dense activity in the visitation profile in
order to infer the best
-
fitting number of
clusters or 'gathering events'.
It then
finds the best
-
fitting model for each visit cluster based on the power
-
law
distribution of the difference in arrival times between consecutive individuals.
Finally
, it calculates
the
membership of individuals in gro
ups based on their visit
time
.
Thus, t
his machine
-
learning method automatically estimates the best
fitting group co
-
membership in the population by reducing overall entropy in the
system. Groups ranged in duration from 2 to 10 minutes in length, and from

2 to
29 individuals.
In order to assess the pattern of group size over the course of the
day, w
e then fit a smooth spline to the group size distribu
tion using the central
time of each group.



Underlying movement pattern


Our hypothesis takes two parts:
i) that individuals should search more actively
for food in the early parts of the day, and then ii) that these individuals should
exploit the best food patch at the end of the day. In order to test if this could be
happening, we isolated all within
-
day m
ovements by individuals between known
food patches from the permanent grid of feeders, comparing the distribution of
these movements over the course of the day to the activity pattern from the
overall visits. The hypothesis suggests that movements between

these known
sites should abruptly end as the birds shift to exploiting the site they decide is
best.





SUPPLEMENTARY RESULTS


Discovery by species


Species

Coefficient

SE

P

Blue tit

3.863



Coal tit

0.099

0.496

0.841

Great tit

-
0.463

0.351

0.189

Marsh tit

-
0.257

0.437

0.556

Nuthatch

-
0.419

0.794

0.598


Table S1.

Model of arrival times by species using data from Figure 2c shows that
there is no significant difference between species using blue tits as the reference
category.









Group size



Figure S1.

Observations from the permanent grid of 65 feeders suggest
increasing group size over the course of the day, until a fission event late in the
afternoon. The greater number of discoveries observed in this study were
unlikely to have resulte
d from larger groups early in the day.



Movements to optimal feeders




Figure S2.

The pattern of movements by individuals between feeders in the
permanent grid supports the hypotheses that individuals settled on their optimal
food source at the end of the day. There was little deviation of movements
between feeders during the first
half of the day, suggesting that the searching
behaviour exhibited by birds at that time
of the day was independent of
movements between known food sources. The number of movements in the
second
-
half of the day were lower than expected, with the exception

of a peak
approximately two hours before dusk that could be due to individuals moving to
their preferred food source for exploitation.







Figure S3
.

Odds of discovery
of an experimental

site (data from Figure 2a)

given
the pattern of movements by individuals between known sites
. This

shows that
the large number of early morning arrivals was not explained by
movement

rate
in the underlying population.

Solid line is the ratio of new arrivals

(discoveries)

at novel f
ood resources and the movement between known sites from the
permanent grid of feeders. Grey shading is the 95% confidence range of the odds

calculated using bootstraps per Figure 2b
.




SUPPLEMENTARY MATERIAL REFERENCES


1.

Psorakis I.,
Roberts S.J., Rezek I., Sheldon B.C. 2012 Inferring social
network structure in ecological systems from spatio
-
temporal data streams.
J R
Soc Interface

9
, 3055
-
3066. (doi:doi:10.1098/Rsif.2012.0223).

2.

Farine D.R., Garroway C.J., Sheldon B.C. 2012 Social
network analysis of
mixed
-
species flocks: exploring the structure and evolution of interspecific social
behaviour.
Anim. Behav.

84
, 1271
-
1277. (doi:10.1016/J.Anbehav.2012.08.008).