Science - USA (2020-07-10)

87% of localizations occurred (Fig. 1A) ( 15 ).
Wedelineated individual tracks by defining

1 min stops as either start (“origin”) or end
(“goal”) points, each classified as either cave,
tree, or out of range type ( 15 ). In this way,

9218 individual movement tracks were iden-

tified (Fig. 1B). The overall ATLAS coverage

area, where bats were successfully tracked

at the Hula Valley and its surroundings, was

88,200 ha.

The bats exhibited typical, repetitive forag-

ing movements characterized by straight flights

(Fig. 1, B to D), with mean track straightness

close to 1 regardless of origin or goal type (Fig. 2A

and table S1) or of travel distance (linear re-

gression,r^2 < 0.01,P= 0.81). Turning angles

and headings were tightly distributed around

zero (Fig. 2B and table S1), even for goals more

than 21 km away. These parameters were con-

sistent along the entire length of the tracks

(Fig. 2C), implying that the bats, on average,

depart already in the direction of their goal

and stay on course for the duration of the flight.

Straightness did not significantly differ between

juveniles (3 to 8 months old) and adults (ttest,

t=−0.19,P= 0.85), nor was it affected by

forearm length, a proxy for age (linear regres-

sion,r^2 = 0.01,P=0.25).Althoughthese

straight tracks are substantially different from

tortuous tracks expected from random searches,

straight tracks alone might be insufficient

to indicate goal-directed movement, because

moving straight in a random direction is con-

sidered an efficient random search strategy

( 16 ). However, flight direction was far from

random, as each bat flew every night to a few

(mean ± SD, 4.13 ± 1.04) specific favorite fruit

trees to which they returned on different nights

[Pianka’s modified niche overlap index ( 17 ),

P= 0.005; Fig. 2D] and arrived from multiple

directions (Fig. 2E). Altogether, our analyses

show that Egyptian fruit bats seldom exhibit

random search but rather traverse their home

range in repeated, straight goal-directed flights

to a few selected favored goals. Although such

movements in the shortest or least-effort

paths in multiple directions match predictions

arising from spatial knowledge congruent with

a cognitive map ( 1 ), further evidence is needed

to support this conjecture ( 18 ).

The ability to take novel shortcuts be-

tween any two known locations that are not

within detectable range of each other is con-

sidered the hallmark of a cognitive map and

the strongest evidence for its existence ( 1 , 18 ).

Lacking prior lifetime movement information

on our captured wild bats, we defined a short-

cut by a given bat as the first track between two

previously recorded goals, and an unrecorded

route as the first track to a previously un-

recorded goal, that have not been traversed by

this bat at least 10 nights after tagging; the

mean number of nights was in fact much

higher: 37.1 ± 40, median 18, maximum 158

nights. In total, 397 of the 9218 tracks (4.3%)

were defined as shortcuts or unrecorded routes

(Figs. 1, C and D, and 3A), observed in 70 of the

172 bats (41%), in equal proportions among age

groups (Fisher’s exact test,P= 1.00). Notably,

though being somewhat longer, shortcuts and

unrecorded routes did not otherwise differ

significantly from regular tracks in straight-

ness, turning angles, and headings (Fig. 3C).

Furthermore, track straightness did not change

190 10 JULY 2020•VOL 369 ISSUE 6500 sciencemag.org SCIENCE

A B

C D

E

20

15

10

5

0

Track straightness

Track length (km)

Tracks >10 km

Turning angles

(313,664)

Goal headings

(313,664)

Turning anglesGoal headings

120°

60°

0°

Turning anglesGoal headings

Overlap index

1

0.8

0.6

0.4

0.2

0

Interval (num. nights)

between compared nights

0 10 20 30 40 50

Random tree selection

0 0.2 0.4 0.6 0.8 1

1

0.8

0.6

0.4

0.2

0

P-value (Rayleigh test)

of origins distribution

P

-value (Rayleigh test) of

arrival headings distribution

1

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0

120°

60°

0°

Tracks <10 km

Fig. 2. Characteristics of Egyptian fruit bat flight tracks.(A)Boxplots of length (blue, left axis) and
straightness index (red, right axis) of different track categories (number of tracks in parentheses).“Unclassified”
could not be confidently categorized because of missing data at the track’s beginning or end. (B) Distribution
of (top) turning angles (angle relative to previous step) and (bottom) headings (angle relative to goal) of all
nonstationary localizations. (C) Mean turning angles (red line) and headings (blue line) with circular SD (red and
blue areas, respectively) for different track percentiles, for tracks shorter (top) or longer (bottom) than 10 km.
(D) Pianka’s overlap index ( 17 ) for mean ± SEM (across bats) overlap in tree visitations (within each bat) between
different nights. Dotted line is the result of 10,000 bootstraps of random tree selections in each night, using
only trees visited by bats during the same season as the tested bat. Compared to random, bats show significant
fidelity to favorite fruit trees up to ~50 nights apart. (E) Scatterplot ofPvalues of Rayleigh’s test for circular
nonuniformity, for all 239 target trees visited from multiple origins by the same bat. Theyandxaxes account for
arrival headings and origins’angles to the target tree, respectively. Dotted lines show the 0.05 significance
level for each axis. In red are 21 (9%) significant deviations from uniform for theyaxis. The Hodges-Ajne
nonparametric test found no significant deviations from uniform (fig. S1).

RESEARCH | REPORTS