Scott Bergen

1
Counting Critters from Space
Monitoring Large Wildlife Directly through High
Spatial Resolution Remote Sensing

• Scott Bergen
• Eric Sanderson
• Pete Coppolillo (Tanzania)
• Joel Berger (USA)
• Claudio Campagna (Argentina)

2
Proof of concept

• Can we use high spatial resolution to count
  wildlife directly from space?
• If so, what animals and under what circumstances?
• Measuring abundance (and species type)
• Utility of hyper-spatial imager for biodiversity

Wildlife Management
Mgmt. Goal
3
Hyperspatial Satellites in Operation
Satellite Band Wavelength Resolution Nadir Resolution Off-Nadir Bit Depth Extent Return Nadir Return Off Nadir
(at 26 degrees)
Ikonos 2 B .45 - .52 3.2 m 4.0 m 11 13 x 13 km 144 days 3-5 days
G .51 - .60 3.2 m 4.0 m 11 11.3 X 11.3 km
R .63 - .70 3.2 m 4.0 m 11
IR .76 - .85 3.2 m 4.0 m 11
  Panchromatic .45 - .90 0.82 m 1.0 m 11      
(at 26 degrees)
Quickbird 2 B .45 - .52 2.4 m 2.88 m 11 16.5 x 16.5 km 11 days 3- 6 days
G .52 - .60 2.4 m 2.88 m 11 16.5 x 165 km
R .63 - .69 2.4 m 2.88 m 11
IR .76 - .89 2.4 m 2.88 m 11
  Panchromatic .45 - .90 0.6 m 0.72 m 11      
(at 30 degrees)
Orbview 3 B .45 - .52 4.0 m 5.2 m 11 8 x 8 km n/a lt 3 days
G .52 - .60 4.0 m 5.2 m 11
R .62 - .69 4.0 m 5.2 m 11
IR .76 - .90 4.0 m 5.2 m 11
  Panchromatic .45 - .90 1.0 m 1.3 m 11      
(at 25 degrees)
Worldview-1 Panchromatic na 0.5 m 0.59 m 11 17.4 x 110 km lt 4.6 days 4.6 days
4
1, Bronx Zoo Experiment

• An experiment that would quantify hyper-spatial
  imagerys ability to identify wildlife in a
  semi-natural environment
• Test species and habitat characteristics
• Build a predictive model to guide and assess
  potential field sites for further experiments

5
Bronx Zoo Experimental Framework

• Two experiments
• Fake Fur Targets of known size and color
• White, Brown, Black
• Small (gazelle), Medium (buck deer), Large
  (bison)
• Animal targets animals in the Zoo
• 27 species in known enclosures
• Number of animals
• Position noted at time of acquisition

6
Logit (identified) -3.666 0.019(Color)
0.970(Size) - 0.230(VegHt) - 0.421(Shade)
Note green, grassy background
7
Hyperspatial Imagery is not for the Birds
8
Information Shadow
9
Shadow Information
10
Important Initial Findings

• Some animals are too small to be identified
  (resolution of imagery), even in aggregate
• Vegatation and shade (cover) significantly
  degrades detectability
• Shadows cast by individual animals can amply
  reflectance signature and provide important
  additional information (height, shape)

11
2, Ruaha National Park, Tanzania

• Large wildlife species elephant, Cape buffalo,
  giraffe, eland, etc.
• Heterogenenous landscape with miombo forest,
  shrublands, and grasslands
• Initially tried to develop 3-d model using 2 sat.
  images _at_ apposing Nadir angles
• Opportunity to perfect ground survey collection

12
(No Transcript)
13
(No Transcript)
14
3, National Elk Refuge, Wyoming
15
Techniques developed here allowed us to be able
to verify the location of over 1,500 individuals
within 10 seconds Allowed us to investigate how
individual animals sex, color, position relative
to sun angle affect reflectance spectra and
composition
16
Adventures in Object Orientation
17
Identifying Noise
18
Deciphered Noise
19
Counting Wildlife using hyperspatial imagery with
different methods
20
4, Peninsula Valdes, Argentina
21
(No Transcript)
22
5, Dumb Luck
23
(No Transcript)
24
(No Transcript)
25
Conclusions

• Can do Big contrasty animals in the open
• Caribou, Musk Ox, Bison, Elk, Seal/Sea Lion,
  Wildebeest, Elephant on grass, etc.
• Contextual and spectral clues enhance automated
  classification
• But old-fashioned manual interpretation faster
  and cheaper, with comparable accuracy
• Costs and accuracy equivalent to standard field
  methods, for situations where suitable

26
Solution Science
Today
Science to explain patterns
Remote Sensing Observations
Describe patterns on Earth
How?
Tomorrow
Remote Sensing Observations
Science to suggest desired patterns
Observed patterns verify we are on track
Less vulnerable, more resilient Provides societal
benefits
27
Thank you
28
(No Transcript)