Transcript Wayne Getz
EVERYTHING DISPERSES TO MIAMI
December 14 - December 16, 2012
T-LoCoH:
A Spatiotemporal Method for
Analyzing Movement Data
Andy Lyons, Wendy Turner & Wayne Getz
UC Berkeley, 2012
Outline and Take Home Message
Quick
review of methods
to analyze movement and
construct home range and
utilization distributions
Discuss spatio-temporal
issues
Present T-LoCoH as an
extension of LoCoH
methods to include time
Worton 1989
These data
are more
interesting
than mere
step-size,
turning-angle
and CRW
statistics or
home range
boundaries
and UD plots
Classic Home Range Methods
Aggregate Summaries
Classic Home Range Methods
Aggregate Summaries
Minimum
Convex Polygon
easy
to understand
and compute
point peeling
algorithms can
produce UDs
sensitive to outliers
and point geometry
Classic Home Range Methods
Aggregate Summaries
Alpha
Hull
similar to MCP, can
model concave
geometries
Classic Home Range Methods
Local Probability Functions
Kernel Density
Estimator
most common HR estimator
widely implemented
impose a Gaussian or
compact kernels
“h” parameter controls
width of kernels
smoothing
output: raster surface
Classic Home Range Methods
Local Probability Functions
Kernel Density
Estimator
most common HR estimator
widely implemented
impose a Gaussian or
compact kernels
“h” parameter controls
width of kernels
smoothing
output: raster surface
Classic Home Range Methods
Local Probability Functions
Kernel Density
Estimator
most common HR estimator
widely implemented
impose a Gaussian or
compact kernels
“h” parameter controls
width of kernels
smoothing
output: raster surface
Home Range Hull Methods
Local Polygons
Characteristic
Hull
create Delaunay triangles
start peeling them off,
longest perimeter first
pause when N% of
points are enclosed, call
that the N% utilization
distribution
output: polygons
Hull Home Range Methods
Local Convex Hulls
Local Convex Hull (LoCoH)
create a little MCP or hull
around each point
sort those smallest to largest
start merging
pause when N% of points
are enclosed, call that the
N% utilization distribution
output: polygons
New Home Range Methods
Local Probability Functions
Brownian Bridge
New Home Range Methods
Local Probability Functions
Brownian Bridge
output: raster probability
surface
Recent
Improvements
Trade-offs among methods
hugs the data, defines boundaries
omission errors
tailored parameters
smoothed: obscures boundaries
commission errors
‘automatic’
LoCoH Algorithm
1. Loop through points
• For each point,
calculate distances to
nearby points
• Pick a set of nearest
neighbors
• k-method
• r-method
• a-method
• Draw local hulls
around all points
• Sort hulls in a
meaningful way
• Start merging hulls
• When merged hull
encompasses x% of
points, pause and call
that an isopleth
• Visualize & analyze
LoCoH =Local Convex Hull
LoCoH Algorithm
1. Loop through points
• For each point,
calculate distances to
nearby points
• Pick a set of nearest
neighbors
• k-method
• r-method
• a-method
• Draw local hulls
around all points
• Sort hulls in a
meaningful way
• Start merging hulls
• When merged hull
encompasses x% of
points, pause and call
that an isopleth
• Visualize & analyze
LoCoH =Local Convex Hull
LoCoH Algorithm
1. Loop through points
• For each point,
calculate distances to
nearby points
• Pick a set of nearest
neighbors
• k-method
• r-method
• a-method
• Draw local hulls
around all points
• Sort hulls in a
meaningful way
• Start merging hulls
• When merged hull
encompasses x% of
points, pause and call
that an isopleth
• Visualize & analyze
LoCoH =Local Convex Hull
6
5
4
3
2
7
1
LoCoH Algorithm
1. Loop through points
• For each point,
calculate distances to
nearby points
• Pick a set of nearest
neighbors
• k-method
• r-method
• a-method
• Draw local hulls
around all points
• Sort hulls in a
meaningful way
• Start merging hulls
• When merged hull
encompasses x% of
points, pause and call
that an isopleth
• Visualize & analyze
LoCoH =Local Convex Hull
LoCoH Algorithm
1. Loop through points
• For each point,
calculate distances to
nearby points
• Pick a set of nearest
neighbors
• k-method
• r-method
• a-method
• Draw local hulls
around all points
• Sort hulls in a
meaningful way
• Start merging hulls
• When merged hull
encompasses x% of
points, pause and call
that an isopleth
• Visualize & analyze
LoCoH =Local Convex Hull
d a
LoCoH Algorithm
1. Loop through points
• For each point,
calculate distances to
nearby points
• Pick a set of nearest
neighbors
• k-method
• r-method
• a-method
• Draw local hulls
around all points
• Sort hulls in a
meaningful way
• Start merging hulls
• When merged hull
encompasses x% of
points, pause and call
that an isopleth
• Visualize & analyze
LoCoH =Local Convex Hull
LoCoH Algorithm
1. Loop through points
• For each point,
calculate distances to
nearby points
• Pick a set of nearest
neighbors
• k-method
• r-method
• a-method
• Draw local hulls
around all points
• Sort hulls in a
meaningful way
• Start merging hulls
• When merged hull
encompasses x% of
points, pause and call
that an isopleth
• Visualize & analyze
LoCoH =Local Convex Hull
7.
5.
8.
3.
6.
2.
1.
4.
LoCoH Algorithm
1. Loop through points
• For each point,
calculate distances to
nearby points
• Pick a set of nearest
neighbors
• k-method
• r-method
• a-method
• Draw local hulls
around all points
• Sort hulls in a
meaningful way
• Start merging hulls
• When merged hull
encompasses x% of
points, pause and call
that an isopleth
• Visualize & analyze
LoCoH =Local Convex Hull
LoCoH Algorithm
1. Loop through points
• For each point,
calculate distances to
nearby points
• Pick a set of nearest
neighbors
• k-method
• r-method
• a-method
• Draw local hulls
around all points
• Sort hulls in a
meaningful way
• Start merging hulls
• When merged hull
encompasses x% of
points, pause and call
that an isopleth
• Visualize & analyze
LoCoH =Local Convex Hull
20th% isopleth
LoCoH Algorithm
1. Loop through points
• For each point,
calculate distances to
nearby points
• Pick a set of nearest
neighbors
• k-method
• r-method
• a-method
• Draw local hulls
around all points
• Sort hulls in a
meaningful way
• Start merging hulls
• When merged hull
encompasses x% of
points, pause and call
that an isopleth
• Visualize & analyze
T-LoCoH Algorithm
1. Loop through points
• For each point,
calculate distances to
nearby points
• Pick a set of nearest
neighbors
• k-method
• r-method
• a-method
• Draw local hulls
around all points
• Sort hulls in a
meaningful way
• Start merging hulls
• When merged hull
encompasses x% of
points, pause and call
that an isopleth
• Visualize & analyze
T-LoCoH Approach
Euclidean Distance “Time Scaled Distance”
7.
5.
8.
3.
6.
2.
1.
4.
Sort hulls by a time-dependent metric: elongation,
revisitation index, duration / intensity of use
New visualization tools
Time Scaled Distance
Want the “distance” to reflect both how far apart two
points are in space as well as time
We transform the time difference
between two points to spatial
units by asking:
time
how far would the animal have
traveled had it been moving at
maximum speed in same direction?
This time-distance becomes
a third axis in “space time”
y
x
Time-Scaled Distance (TSD)
space-selection
s=0
time-selection
s≈1
points from
other visits
to this area
Sorting Hulls in a Meaningful
Way: Time-Use
revisitation rate
duration or intensity of use
duration of use
important
seasonal
resources
infrequently
used resources
year
- long
resources
revisitation index
Sorting Hulls in a Meaningful Way:
Identify Canonical Activity Modes
Sorting Hulls in a Meaningful
Way: Elongation
eccentricity
of bounding ellipsoid
perimeter : area ratio
Sorting Hulls in a Meaningful
Way: Hull Metrics
Density
area
number of nearest neighbors
number of enclosed points
Time Use
revisitation rates
mean visit duration
Time (parent point)
hour of day
month
date
Elongation / Movement Phase
eccentricity of ellipsoid bounding
the hull
perimeter / area ratio
average speed of nearest
neighbors
standard deviation of nearest
neighbor speeds
Ancillary
Variables
ancillary variables associated
with hulls
proportion of enclosed points
that have property X
Simulated Data
• Single virtual animal moves
between 9 patches
• constant step size and sampling
interval
• unbounded random walk within
1. spatially overlapping
each patch for a predetermined
but temporally separate
# steps
resource edges
• directional movement
to the
next patch
• duration and frequency
of patch use varied
Patch
Visits
Total Pts
p1
2 x 120
240
p2
4 x 60
240
p3
1 x 240
240
p4
6 x 40
240
p5
12 x 20
240
p6
4 x 60
240
p7
6 x 40
240
p8
4 x 60
240
p9
2 x 120
240
2. gradient of
directionality
3. varied frequency
of use
T-LoCoH General Workflow
1. Select a value of s based on the time scale of
interest
2. Create density isopleths that do a “good job”
representing the home range
e.g., no spurious crossovers
3. Compute hull metrics for elongation and/or timeuse
4. Visualize isopleths and/or hull points
5. Interpret and/or plot against environmental
variables
With Time
k=3
Without Time
s = 0.1
s=0
Isopleth level indicates the proportion of total points enclosed along a
gradient of point density (red highest density, light blue lowest).
With Time
k=7
Without Time
s = 0.1
s=0
Isopleth level indicates the proportion of total points enclosed along a
gradient of point density (red highest density, light blue lowest).
With Time
k = 15
Without Time
s = 0.1
s=0
Isopleth level indicates the proportion of total points enclosed along a
gradient of point density (red highest density, light blue lowest).
Simulated Data:
Density Isopleths
Hulls sorted from most number of points per unit area (red) to
least (blue)
Simulated Data:
Elongation Isopleths
Hulls sorted by eccentricity of bounding ellipse (left) or
perimeter/area ratio (right) from most (red) to least (blue)
elongated.
Simulated Data:
Revisitation Isopleths
Hulls sorted by number of separate visits
(inter-visit gap = 24 time steps)
Simulated Data:
Duration Isopleths
Hulls sorted by mean number of locations per visit
(inter-visit gap = 24 time steps).
Etosha
National
Park,
Namibia
Female springbok
Female springbok: density isopleths
Text
Female Springbok:
Hull revisitation rate and duration over time
Female Springbok:
Directional Routes
Map of directional routes formed by identifying hulls with a
perimeter area ratio value in the top 15%. Blue dots are
known water points.
Hour of day
speed
1
0
0
hour
24
Hour of day
vs
Avg. Speed
Territorial
male
a = 3700
Male Springbok: Hulls
in Time-Use Space
Male Springbok: Hulls
in Time-Use Space
Next step to include
Environmental Variables
Association
Hull Metrics
count of spatially
overlapping hulls
for two
individuals
number of
separate visits in
overlapping hulls
time lag of
overlapping hulls
T-LoCoH for R
Pre-processing
remove bursts
sub-sample
animations
Feature Creation
hulls
isopleths
directional routes
Hull metric creation
time use
elongation
Plotting
hull and isopleth maps
pair-wise hull metric
scatterplots
hull-scatter plots
support for shapefiles &
imagery
Export formats
R format
csv
shapefiles
http://locoh.cnr.berkeley.edu/tlocoh
Acknowledgements
Andy Lyons
Scott Fortmann-Roe
Wendy Turner
Chris Wilmers
George Wittemyer
Sadie Ryan
Werner Kilian
Namibian Ministry of
Environment and
Tourism
staff of the Etosha
Ecological Institute
Berkeley Initiative in
Global Change Biology
NIH Grant GM83863
http://locoh.cnr.berkeley.edu/tlocoh