A Bentho-Pelagic Observatory

1
A Bentho-Pelagic Observatory to Support
Fisheries and an Ecosystem Approach to
Management NEBO http//nebo.whoi.edu
2
Northeast Bentho-pelagic Observatory (NEBO)
Northeastern Regional Coastal Ocean Observing
System
1. Stellwagen Bank Marine Sanctuary 2. Northeast
Peak, Closed Area II, Habitat of Particular
Concern 3. Great South Channel, HAPC? 4. Hudson
Canyon Closed Area
Selection of 4 Sentinel Sites
3
Who we are Scott Gallager, Amber York (WHOI
Biology Department) Steve Lerner (WHOI Applied
Ocean Physics Engineering) Hauke Kit-Powell
(WHOI Marine Policy Center) Richard Taylor and
Norman Vine (Fishing Community, Geo-Spatial
Specialists) Larry Mayer (Center for Coastal
Ocean Mapping, University of New Hampshire) Yuri
Rzhanov (UNH) image processing Amy Holt Cline
(UNH) Education and Outreach Coordinator Peter
Auster (National Undersea Research Center,
University of Connecticut) Lakshman Prasad (Los
Alamos National Laboratory) Kathryn Ford ,
Massachusetts Department of Marine Fisheries,

4
Partners and End Users Stellwagen Bank National
Marine Sanctuary, David Wiley, Director New
England Fishery Management Council, Sally
McGee Massachusetts Coastal Zone Management
Program, Leslie Ann McGee, Director NOAA
National Marine Fisheries Service, Mike Fogarty,
Thomas Noji, Robert Reid, Vince Guida. Dvora
Hart, Paul Rago United States Geological Survey,
Page Valentine, Walter Barnhardt Gulf of Maine
Census of Marine Life (CoML), Lew Incze Gulf of
Maine Mapping Initiative, Sara Ellis, Megan
Tyrrell Massachusetts Fishermans Partnership,
David Bergeron, Olivia Free.
5
NOAA
6
Bentho-pelagic coupling (highly simplified)
wind
CO2
Carbon pump
Convergent front
zooplankton
phytoplankton
fish
turbulence
detritus
Boundary layer shear
concentration deposition
anoxia
Structuring elements of Benthic communities
nutrients
7
Thermal Fronts and the Distribution of
Scallops on Georges Bank
An example of strong bentho-pelagic
coupling adult scallops are most abundant
directly under persistent fronts where larvae
are concentrated
8
Integration of pelagic ecosystem components
Video Plankton Recorder transect across Great
South Channel
abundance of Calanus finmarchicus
9
with benthic components
10

• What is an Ecosystem Approach to Management?
• To understand an ecosystem sufficiently well to
  allow predicting change of one component in light
  of change in another.
• Important holistic concepts
• sustainable yield, species-specific total
  allowable catch, recovery time, resilience
• bentho-pelagic coupling (energy flow), habitat
  functional value.
• oceanographic drivers (nutrients, currents,
  fronts, eddies)

Variables species abundance, distribution,
substrate composition, water column properties
(temperature, salinity, nutrients, phytoplankton,
zooplankton). over time scales of
hours-months-years-decades
Indices patchiness indices, inter-species
and inter-substrate type relationship ,
species diversity in relation to substrate at
multiple spatial scales, water column
processes, external impacts (storms, fishing,
oil spills, etc.) assess community composition
including dominant and rare species across
habitat types and over time, assess habitat
associations of key fish taxa based on variations
in size and density, habitat classifications
that include attributes of grain size and
biogenic components
11
Objectives of NEBO

• (1) To establish four key locations along
  northeast coastline
• benthic community structure
• coupling between the water column and benthic
  community
• system change over time scales of days to years
  will be quantified
• (2) To develop tools necessary to integrate and
  fuse disparate fisheries-relevant
• data sets
• segment and classify seafloor targets and
  substrate
• visualize the results in near real-time
• fusion of optics and acoustics
• optical and acoustic mosaics
• Plankton distributions (Video Plankton Recorder)
• Water column measurements (NERACOOS, GoMOOS,
  etc. buoys and gliders)
• (3) To establish metrics for quantifying change
• ecosystem dynamics
• benthic community structure
• organism abundance and size distributions
• diversity of a wide range of benthic and
  demersal taxa

12
Image acquisition Manual data extraction Image
test set Automated Image processing Light field
, color correction Segmentation ROI
extraction Feature extraction Classification
Texture energy analysis Fusion of images with
GIS Data Visualization Data Products Morphology
Diversity
13
0.68 inch fiber optic tow cable
Machine vision digital stereo video cameras
Fiber optic telemetry bottle
Benthos Altimiter
SBE 37 CTD
2 Imagenix side scan sonars
RDI 1200 kHz ADCP
Strobe
Strobe
x4
3 m altitude typical
HABCAM Habitat Mapping Camera System
Stereo 2 Mpixel system
1m Field of View 5 10 frames per second gt50
overlap _at_ 5kts
14
SM-3002 multibeam
Imagenix sidescan on vehicle
300m
15
Data Flow Wire Diagram
Field Operation
Real-Time Data Products
Multi-Resolution Synthesis Products
display
display
HabCam
species-specific abundance
image acquisition
patch dynamics
GIS acquisition
time, date, altitude, depth, lat, lon, roll,
pitch, yaw, temp, salinity, heading, speed
raw 16 bit tiff full res jpegs
habitat distribution
species-specific distribution
2TB/day
classification results
RAID array
Virtual Index card (VIEW) images metadata
QGIS-Mapserver Interface
RAID array
multi-species management models
manual classification
automated classification
real-time mosaics
population dynamics models
Image processing lightfield correction color
correction segmentation region of interest
extraction feature extraction
Index Card Viewer
ship navigation
adaptive sampling
16
Image acquisition Manual data extraction Image
test set Automated Image processing Light field
, color correction Segmentation ROI
extraction Feature extraction Classification
Texture energy analysis Fusion of images with
GIS Data Visualization Mosaics Data
Products Morphology Diversity
5
4
3
2
1
1. Elephant trunk 2. Nantucket Light Ship Closed
Area 3. Western Great South Channel 4. Closed
area I 5. Closed Area II- HAPC

• 5 distinct areas
• 40,000 images each
• Manually classified
• Training set

http//habcam.whoi.edu
17
Automated mosaicing post process now.real-time
soon with Yuri Rzhanov (UNH) (see poster)
18
Invasive tunicate Didemnum sp. has been found in
large (gt20nm2) areas on Georges Bank
1m Field of view
19
Image acquisition Manual data extraction Image
test set Automated Image processing Light
field, color correction Segmentation ROI
extraction Feature extraction Classification
Texture energy analysis Fusion of images with
GIS Data Visualization Data Products Morphology
Diversity

• .
• strobed light ( 2 usec )
• eliminate motion artifacts
• homomorphic filtering
• (low pass filter)
• white balance correction
• to achieve true color
• adaptive homogeneity-directed demosaicing
• direct in situ measurements of
• optical water characteristics,
• linear application of Beer's Law

20
Image acquisition Manual data extraction Image
test set Automated Image processing Light field
, color correction Segmentation ROI
extraction Feature extraction Classification
Texture energy analysis Fusion of images with
GIS Data Visualization Data Products Morphology
Diversity
Separation of targets from a complex (noisy)
background Is THE most difficult challenge
21
(No Transcript)
22
Image acquisition Manual data extraction Image
test set Automated Image processing Light field
, color correction Segmentation ROI
extraction Feature extraction Classification
Texture energy analysis Fusion of images with
GIS Data Visualization Data Products Morphology
Diversity
scallop
skate
flounder
23
(No Transcript)
24
Feature extraction
Target Classification
X
Gabor matrix
Gabor transform
automatically extracted ROI
Feature Vectors
X, SD of Gabor transform 4 scales 8
orientations RGB 192 elements
3 color angles 3 color-edge angles 3 SD of
dot products of the three color bands for
each image.
1113 dimensional feature space for randomly
oriented images using very low Gabor center
frequencies (less than 0.05).
Support Vector Machine Soft-margin, Gaussian RBF
kernel. Gaussian RBF parameter gamma 2.0 SVM
regularization parameter 70.0
LOO Cross Validation
25
Automated Segmentation and Classification Results
for Closed Area II HAPC accuracy number
of scallops counted in 9.5 km by
hand 8,172 number of scallop ROIs segmented by
edge flow algorithm 7,099 87 number of scallops
correctly identified by SVM 5,984 84 Overall
accuracy 73
Confusion Matrix
26
Image acquisition Manual data extraction Image
test set Automated Image processing Light field
, color correction Segmentation ROI
extraction Feature extraction Classification
Texture energy analysis Fusion of images with
GIS Data Visualization Data Products Morphology
Diversity

• Defining Habitat using Texture
• Habitat Categories
• mud/sand without emergent biological structure
• mud/sand with emergent biological structure
• small gravel (lt 2cm) without emergent/attached
  biological structure
• small gravel (lt 2cm) with emergent/attached
  biological structure
• shell aggregations and/or reefs without
  emergent/attached biological structure
• shell aggregations and/or reefs with
  emergent/attached biological structure
• cobble/boulder without emergent/attached
  biological structure
• cobble/boulder with emergent/attached biological
  structure

27
Image acquisition Manual data extraction Image
test set Automated Image processing Light field
, color correction Segmentation ROI
extraction Feature extraction Classification
Texture energy analysis Data Visualization Fusio
n of images with GIS Mosaics Data
Products Morphology Diversity
http//nebo.whoi.edu
Click on URL
28
Establish training sets for automated
classification using novel point and click
routine to sort through large image data base
over the web
29
(No Transcript)
30
Data Products Morphometry Abundance Distribution
Patchiness indices Species diversity relationship
to substrate Survey design Real-time adaptive
sampling
CV
Strong patchiness makes discontinuous sampling
difficult
31
Data Products Morphometry Abundance Distribution
Patchiness indices Species diversity relationship
to substrate Survey design Real-time adaptive
sampling
Power Spectrum of Abundance as a Function of
Spatial Elephant Trunk scallops
1000 900 800 700 600 500
400 300 200 100
Spatial Scale (m)
32
Understanding Patchiness Grain and Intensity
320 m
Neighbor k analysis second-order moment k(t)
EN(t) / X expected number of individuals within
distance t of any individual. Or average number
of individuals within distance t of any given
individual EN(t) N-1 ? ? It (Uij) Uij is
distance between individual i and j L(t)
EN(t) - mean EN(t) of 999 iterations
33
Data Products Morphometry Abundance Distribution
Patchiness indices Species diversity relationship
to substrate Survey design Real-time adaptive
sampling
Placopecten
34
Data Products Morphometry Abundance Distribution
Patchiness indices Species diversity Relationship
to substrate Survey design Real-time adaptive
sampling
Substrate composition
Taxon abundance
35
Data Products Morphometry Abundance Distribution
Patchiness indices Species diversity relationship
to substrate Survey design Real-time adaptive
sampling
36
NEBO Data products Indices for use in Ecosystem
Based Management
1. Taxon- specific abundance plots of
macrobenthos and macrophytes 2. Spatial plots
of Species Diversity Indices (e.g.,
Shannon-Weiner) as a function of binning scale)
3. Species-specific patch size estimates using
point process statistics 4. Organism-substrate
association indices using cluster analysis 5.
Georeferenced species and substrate
distributional maps, 6. Water column properties-
hydrography, chlorophyll, and zooplankton
abundance in relation to habitat structure, 7.
Temporal information including seasonal and inter
annual variation in species and substrate
composition, organism-substrate associations,
water column properties, and food
availability, 8. Imagery data sets and mosaics
for each of the sentinel sites to be available
over the internet, 9. Estimates of annual
variation in larval/juvenile recruitment for
selected species, 10. Compile data to support
characterization and designation of EFH,
37
NEBO will provide
1. Multi-scale (time and space) processes
affecting habitat and species composition 2.
Recovery rates and dynamics of impacted habitats
3. Scale dependent distribution of micro
habitats within regional areas
38
(No Transcript)
39
NOAA
40
Real-time adaptive sampling using knowledge of
community patchiness
Poisson simulated scallop distribution with
even allocation of sample grid spacing
yields convergence of mean density and CV
after 10 transects
41
Poisson simulated scallop distribution with
allocation of sample grid spacing based on
Neighbor K calculated in real-time
yields convergence of mean density and CV
after 2-3 transects
42
click
43
(No Transcript)
44
Data Products Morphometry Abundance Distribution
Patchiness indices Species diversity relationship
to substrate Survey design Real-time adaptive
sampling
Shell height measurements (mean) Automated
141.3 mm Manual 146.4 mm
45
(No Transcript)
46
IOOS Core Variables

• Bathymetry
• Bottom Character
• Contaminants
• Dissolved Nutrients
• Dissolved O2
• Fish Abundance
• Fish Species
• Heat flux
• Ice Distribution
• Ocean color

• Optical properties
• Pathogens
• Phytoplankton species
• Salinity
• Sea Level
• Surface Currents
• Surface Waves
• Temperature
• Zooplankton abundance
• Zooplankton species