Together- in space and time: measuring and interpreting
geo-referenced contacts in animal ecology
Austin, Tx, 11/9-10/2016
Francesca Cagnacci
Federico Ossi
Fondazione Edmund Mach, Italy
Movement ecology take-off: from tools…
… to (re-discovered) research frameworks…
Lagrangian: analysis of individual movement trajectoriesEulerian: expected pattern of space use by an individual or a population over static landscapes (e.g., habitat occupancy modelling)
Mechanistic understanding: process
of individual movement
From Nathan et al. 2008
… including individuals-to-population scaling up.
Mueller and Fagan 2008, Oikos
Ecological and landscape dynamics on population distribution
Relationships among moving individuals (similarity, synchrony, cohesion)
What did I learn?
Migratory behaviour
Movement
mechanisms Movement path
MOVEMENT PATTERNS Population trajectories Population patterns POPULATION DISTRIBUTION Undefined behavioural rules Motion -Navigation capacity Internal state INDIVIDUAL MECHANISMS EXTERNAL CONTEXT
Drake et al. 1995 Mueller & Fagan 2008 Nathan et al. 2008
Migration arena
[natural selection]
Migration syndrome Genetic complex Landscape structure/ dynamic resourcesExternal factors & Dynamics
What did I learn?
Movement is a multi-faceted phenomenon
- Movement emerges on heterogenous landascapes, in
space and time (context of movement)- directly linked to
habitat selection Workshop #2!
Migratory behaviour
Movement
mechanisms Movement path
MOVEMENT PATTERNS Population trajectories Population patterns POPULATION DISTRIBUTION Undefined behavioural rules Motion -Navigation capacity Internal state INDIVIDUAL MECHANISMS EXTERNAL CONTEXT
Drake et al. 1995 Mueller & Fagan 2008 Nathan et al. 2008
Migration arena
[natural selection]
Migration syndrome Genetic complex Landscape structure/ dynamic resourcesExternal factors & Dynamics
What did I learn?
Movement is a multi-faceted phenomenon
- Movement emerges on heterogenous landascapes, in
space and time (context of movement)- directly linked to
habitat selection.
- Movement depends on underling, largely unknown,
individual
mechanisms
(e.g.,
memory,
learning,
physiological state)
Migratory behaviour
Movement
mechanisms Movement path
MOVEMENT PATTERNS Population trajectories Population patterns POPULATION DISTRIBUTION Undefined behavioural rules Motion -Navigation capacity Internal state INDIVIDUAL MECHANISMS EXTERNAL CONTEXT
Drake et al. 1995 Mueller & Fagan 2008 Nathan et al. 2008
Migration arena
[natural selection]
Migration syndrome Genetic complex Landscape structure/ dynamic resourcesExternal factors & Dynamics
What did I learn?
Movement is a multi-faceted phenomenon
- Movement emerges on heterogenous landascapes, in
space and time (context of movement)- directly linked to
habitat selection.
- Movement depends on underlining, largely unknown,
mechanisms (e.g., memory, learning, physiological state)
- Movement is a multi-scale phenomenon:
Movement behavior = applies to individuals
(physiological, behavioral, genetic- movement
process
and pattern)
Movement
ecology=
applies
to
populations
(ecological,
(ecological, evolutionary- movement outcome)
If 1+1= 2, studying animal
interactions
should
be
conceptually easy…
Macroclimate
Local climate
Habitat & food resources
Physiology Individual/collective movement behaviour Demography/Population abundance Population distribution Landuse Species distribution Predation Competition Host-parasite-pathogens Individual Global, landscape Population Species Ecological Scale Communities
Decomposing ecological relationships,
at different scales
Which way
should I
take?
That depends
on where you
want to end
up.
PART I: Studying spatio-temporal patterns in the use of resources…
‘Experimental’ conditions – supplemental feeding and animal movement
Ossi et al. in press Ecosphere
Resource acquisition
Food Resource distribution in winter
Movement Winter supplemental feeding
Winter Supplemental feeding is pervasively applied to target
deer populations
Concentrate resources can alter normal animal space use
patterns AND interactions
Consequences: dominance/competition/social structure/demography and population distribution/disease transmission
‘Experimental’ conditions – supplemental feeding and animal movement
FIRST OBJECTIVE: evaluate the actual use of feeding stations and the effect on
individual space use
SECOND OBJECTIVE: evaluate the effect on interactions (or the definition we used
Model species: the European roe deer
- Most abundant deer species
in Europe
- Major environmental and
climatic gradients
- High ecological plasticity, but
less adapted to snow, and
droughts
- Highly selective shrub/forb
browser
- Cover and protection from
predators (hiding and flushing
anti-predatory tactic)
EURODEER Collaborative Project
49 study areas, ranging 63°N-38°N 0-2000m a.sl. 33 partners, 15 countries > 2500 individual tracks > 8 million GPS locations TOTALLY BOTTOM-UP AND COLLABORATIVEOBJ. 1/HYP. 1: F.S. are mainly used under harsh conditions.
OBJ. 1/HYP. 2: F.S. are attractive points and decrease home range size.
• Temporal variation in use of feeding stations throughout the year should
increase when FS are active AND under severe winter conditions (snow
cover, low temperatures
)• The pattern should be more evident at high altitudes and latitudes (Alps
and Scandinavia)
• The intense use of feeding stations limits home range size of roe deer
(attractive point)
• The use of feeding station decreases in
presence of inter-specific competition
• 9 study areas
• More than 1,000,000 GPS locations from almost 200 roe deer (post –selection)
• Ca. 300 Feeding stations (and ancillary information)
• Snow data from MODIS (satellite) (500 m resolution, presence/absence)
• Temperature data from meteo stations
OBJ. 1: Material and Methods (1)-
• Determination of ‘ weekly use’ of F.S.
Time spent within a buffer of 50 m in a week (based on linear
interpolation) t >1, then use=1
• Identification of available F.S. per animal, based on spatio-temporal overlap: week*animal*feeding station
• Scale of analysis: week
OBJ. 1: Matherial and Methods (2)-
Spatio-temporal resolution
Pr ed ic ti ve pr ob ab ili ty o f u se
Results (yearly pattern)
Legend
Predictive P of use Temperature
Periods of activation feeding stations
OBJ. 1/HYP. 1: F.S. are mainly used under harsh conditions
• Strong temporal effect • Main determinants of use:
low temperature ONLY WHEN feeding stations are active
• No effect of snow
Results
• Use increases with altitude at low latitudes (i.e. on the Alps) and viceversa
P re d ic ti o n o f u se P re d ic ti o n o f u se P re d ic ti o n o f u se P re d ic ti o n o f u se P re d ic ti o n o f u se P re d ic ti o n o f u se P re d ic ti o n o f u se P re d ic ti o n o f u se P re d ic ti o n o f u se
OBJ. 1/HYP. 1: F.S. are mainly used under harsh conditions
Results
• Roe deer that use feeding stations have a smaller home range
06-21/03 22/03-06/04
09-24/05 25/05-09/06
ON ON/OFF
OFF OFF
OBJ. 1/HYP. 2: F.S. are attractive points and decrease home range size.
OBJ. 2/HYP. 1: F.S. increase the probability of encounters
OBJ. 2/HYP. 2: F.S. effect depends on energetic trade-offs and social
traits (sex, age)
• Individuals tend to be closer than normal when they are in
proximity of feeding stations
• There is a temporal pattern in aggregation trends
• Adults and subadults stay closer than
groups of adults
• Grouping increases with winter severity (snow presence, low
temperatures)
• Females are closer each with the
•Alpine range, elevation 500 – 2500 m asl
• 11 couples of animals, selected on the basis of the capture
‘clusters’ they belong to
OBJ. 2: Matherial and Methods (1)-
• Contacts inferred from trajectories, i.e. from GPS locations
(‘indirect encounter mapping’, Krause et al. 2013)
• Temporal resolution: GPS location frequency (3hrs)
Spatiotemporal database for handling moving data
OBJ. 2: Matherial and Methods (2)-
Spatio-temporal resolution
F.S. are moving bools (activation)
Trajectories are moving objects
(by means of linear interpolation)
OBJ. 2: Matherial and Methods (2)-
Spatio-temporal resolution
OBJ. 2: Matherial and Methods (2)-
Spatio-temporal queries: from moving objects to moving reals
Moving object &
Moving object =
Distance between
animals (true or
interpolated)
Moving object &
Moving bool =
Distance of animal
from FS in relation
to FS management
3900 ‘tuples’, i.e. spatio-temporal matches between roe
deer pairs, with their respective Distance and Distance
of both from the common closest FS
50% < 80 m
- Reclassification in binary ‘closeness’ between animals (empirical
distribution-based threshold) binomial distribution
- Generalized additive model (GAM) framework to catch the temporal
pattern
Closeness (0/1) ~ s(day) + s(couple, bs = "re") + Distance FS + snow + sex + age + Temperature
OBJ. 2: Material and Methods (3)-
Ecological statistics (full model )
Parameter Estimate Std. Error Pred. weight
(Intercept) 0.4354 5.7486
Average Distance from the closest
feeding site -0.0089 0.0004 1
Sex of the couple (male-female) -2.4975 9.5992 0.5
Presence of snow 0.3308 0.1154 1
Age of the couple (adult -subadult) 2.052 10.6938 0.5
Daily minimum temperature -0.0256 0.022 0.4
Model averaging
OBJ. 2/HYP. 1:
Probability of encounters decreases with distance from F.S.
• GPS data are acquired PERIODICALLY
• Proximity loggers overcome these limitation by
recording continuous encounter data (but…)
• We have no knowledge about what the monitored individual does in between 2
locations
• We might underestimate the use of specific sites (feeding stations, water holes,
movement corridors), or individual-to-individual encounters
• Three ‘typologies’ of contact:
• Mobile - mobile
• Mobile - fixed
• (Fixed - fixed)
• Whatever the contact type, what data we get?
• ID of loggers
• DURATION of contact (please note! Temporal resolution- epochs)
• NO distance (see later…)!!
IIa- Measuring encounters: proximity loggers
Measuring encounters: proximity loggers vs Periodic GPS acquisition
Case I – Supplemental feeding in roe deer
• Fixed proximity logger deployed at feeding site
• Estimation of HR (90 – 50 – 25 – 10) based on 3h periodic GPS
• One case of contact recorded, but NO spatial overlap between F.S. and home range
• Proximity loggers as a new tool to explore individual use of specific resources
Wesley
Buttercup
Measuring encounters: proximity loggers vs Periodic GPS acquisition
Case II – Urban foxes in Brighton, UK
1 km
Measuring encounters: proximity loggers vs Periodic GPS acquisition
Case II – Urban foxes in Brighton, UK
Contact temporal unit: 20’’ GPS periodic, 30’
Many short contacts (postprocessing) Individual/daily variation
Substantial total time
Contact density vs time of day/ GPS closest periodic to contacts
vs time of day
Proximity sensors provide more dense information
Animals can relocate ‘far’ immediately before or after the contact:
classic spatial data do not always detect use of local resources
Many contacts in inter-fix interval…
- Inter-individual avoidance trough contact detection
Animals can relocate ‘far’ immediately before or after the contact:
classic spatial data do not detect inter-individual contacts
Hr based on ‘contact’
periodic fixes:
no overlap
Contact detection with proximity loggers: what about the spatial
resolution?
When two loggers detect each other,
what is the distance between the
individuals wearing them?
Empirical work on animals in
semi-controlled settings to compare observed
distances with contact occurrence
not as easy as it seems…
Contact detection with proximity loggers: what about the spatial
resolution?
Power 3 Power 7, mean-size mammal Power 7, large-size mammal Power 11 Power 27 Power 3 Power 7, mean-size mammal Power 7, large-size mammal Power 11 Power 27Contact detection with proximity loggers: let’s get the error
components right
TP = True positives = expected and recorded contacts within x’
FN = False Negatives = expected but not recorded contacts within x’
FP = False Positives = unexpected but recorded contacts beyond x’
FN = False Negatives = unexpected and recorded contacts beyond x’
Contact detection with proximity loggers: let’s get the error
components right
Precision TP/(TP+FP) Probability that a recorded contact occurs within x’
False Discovery Rate FP/(TP+FP) = 1- Precision Probability that a recorded contact occurs beyond x’
Sensitivity TP/(TP + FN) Probability to detect a contact occurred within x’
False Negative Rate FN/(TP + FN) = 1 - Sensitivity Probability to miss a contact occurred within x’
Power 3 Power 7, mean-size mammal Power 7, large-size mammal Power 11 Power 27
Contact detection with proximity loggers: how to decide ‘desired’
distance
Probability that a
recorded contact occurs within x’
Power 3 Power 7, mean-size mammal Power 7, large-size mammal Power 11 Power 27
Contact detection with proximity loggers: how to decide ‘distance’
(and power)
Probability to miss an occurred contact within x’
Established technology
High energy consumption
sparse data points
Contacts are
inferred
, with
great uncertainty
13:00 16:00 19:00✔
✖
✖
17:15 20:15 23:15IIb- Measuring encounters: geo-referenced proximity detection
?
Uses the low-power radio as
a contact “sensor”:
direct
contact detection
No location information
✔
✖
Bio-logging devices: proximity-tags
Direct
contact detection
Location
acquired
when
and
where
contacts occur
Spatio-temporal
resolutions
are
setting parameters (see later)
✔
✔
New bio-logging device: geo-referenced proximity detection
Picco et al., Best Paper Award IPSN 2015
✔
ttrigGPS (e.g., 15 min)
• Contacts geo-referenced by the
closest
GPS fix
• Significantly reduces energy/memory consumption
periodic GPStriggered GPS
tGPS (e.g., 3 hours) t
noGPS (e.g., 15 min)
node GPS schedule
What is geo-referenced proximity detection?
What is geo-referenced proximity detection?
3 complementary data types
a b c d e f g h a1 a3a2 b1 c2 b2 e2 f2 g2 b3 c3 d3 e3 f3 c1 d1 e1 f1 g1 g3 Power 3 range Power 7 range Fixed logger
Mobile logger 1 path Mobile logger 2 path Mobile logger 3 path
a1 b2 c2 d2 e2 f2 b1 c1 d1 e1 f1
What is geo-referenced proximity detection?
An empirical set of simulations
Ossi et al. in press Animal
Trial Contact success rate False negative rate Contact-triggered GPS location success rate
Total TP Rate (%) (TP/Total) FN FP Rate (%) (FN/(FN+FP) Total TP Rate (%) (TP/Total) 1L-P3 204 184 90 20 0 100 14 14 100 1L-P7 204 167 81 29 8 78 8 8 100 2L-P3 480 414 86 46 20 70 17 11 60 2L-P7 480 423 88 40 17 70 15 13 98 3L-P3 864 774 90 52 38 58 29 28 99 3L-P7 864 795 92 44 25 62 30 30 100 Total 3096 2757 89 231 108 68 113 104 92
What is geo-referenced proximity detection?
An empirical set of simulations
Is it possible to detect encounters in animal ecology studies?
Is it possible to infer ‘interaction patterns’?
Ecological statistics and hypothesis-based assessment reconcile patterns
& processes.
Proximity loggers convey more information than contact detection
inferred from GPS-based locations, unless a high frequency acquisition schedule can be applied
Definition of ‘encounters’ should be a trade-off between ecological
requirements and technological limitations (errors)
Maria Valent, Risorse estive & invernali, Densità di popolazione Federico Ossi, Risorse invernali Francesca Cagnacci Federico Ossi Johannes de Groeve Nathan Ranc Paola Semenzato Valentina Erculiani Julius Bright Ross Sandro Nicoloso Ralf Gueting Tomas Behr Gianpietro Picco Amy Murphy Davide Molteni