Exploring the structure of the oceanic environment:

A classification approach

Edward GregrEdward GregrKarin BodtkerKarin BodtkerAndrew TritesAndrew Trites

Marine Mammal Research UnitMarine Mammal Research UnitFisheries CentreFisheries CentreUniversity of British ColumbiaUniversity of British Columbia

October 2004October 2004

Why classify oceanic structure?

• related to biological spatial distributions

• temporal changes (e.g. regime shifts)

• Steller sea lion in an ecosystem context

Oceanic structure classified Dodimead et al. 1963

Extending the classification approach

• biological perspective

• quantitative and repeatable

• adaptable– consider temporal variability

(seasons, years, regimes)

– different spatial scales (zooplankton vs. fish vs. sea lions)

A quantitative approache.g. classifying landscapes

High densityResidentialIndustrialRoadsWaterPastureForestWetlandGrass

Data for oceanic classification

Wind stress

Surface current speed

SSH

SSS

SST

1Yi Chao, Jet Propulsion Lab, California Institute of Technology

1 degree ROMS output1, interpolated to equal area grid.

Seasonal averages,1966-1975 and 1980-1989.

Classification methodH - means clustering algorithm1

Sea surface salinity

Sea surface temperature

oC

31 32 33 34 350.0

-0.1

-0.2

-0.3

-0.4

-0.5

-0.6

-0.7

-0.8

+

+

+

+

+

Identify initial clusters

Assign pixels to ‘nearest’ cluster based on maximum likelihood

Iterate until stable

1Hartigan, J. A. 1975. Clustering Algorithms. John Wiley & Sons, New York.

Results: summer, 1966-1975

130°140°150°160°170°180°170°

130°

140°

150°160°

30°

50°

40°

60°

Results: correspond to domains

Summer, 1966-1975

Results: seasonal variability

Results: regime variability

Pre - winterPost - winter

130°140°150°160°170°180°170°

130°

140°

150°160°

30°

50°

40°

60°

- Alaska gyre: evidence of stronger flow post - 1976

- Transitional domain: boundary shift

Results: map comparisonsPre-76 Post-76

• Seasons more similar between regimes than consecutive seasons within each regime

Winter

Spring

Summer

Fall

• Consistency between some seasons differs before and after regime shift

Results: biological relevance

1.38 0.70

1.03

0.560.41

Chl-a, mg/L1

Summer, 1997-2003

1Andrew Thomas, School of Marine Sciences, University of Maine

Summary

• quantitative and adaptable approach

• regions correspond to classic domains

• temporal differences mapped and quantified

• regions have biological relevance

Thanks very much ...

Funding:NOAA, the North Pacific Marine Science Foundation, and the North Pacific Universities Marine Mammal Research Consortium.

Data:Yi Chao, Jet Propulsion Lab, California; Mike Foreman, Institute of Ocean Sciences, British Columbia; Al Hermann, PMEL, Washington; Wieslaw Maslowski, Naval Postgraduate School, California; Andy Thomas, University of Maine, Maine.

Intellectual:Ian Perry, Mike Foreman, Stephen Ban, the MMRU lab, and the attendees of numerous earlier presentations of this work.

Map comparisons

Higher score, more similarSeasons more similar between regimes than

consecutive seasons within each regime.

Summer, 1980 - 1989 Fall, 1980 - 1989

KIA = 0.39AMI = 2.2

Spring, 1966 - 1975 Spring, 1980 - 1989

KIA = 0.49AMI = 2.4

Classification algorithmSelecting the number of clusters to keep

Keep 6 or 8 clusters

Biomes and provinces of Longhurst 1998

• variability within not evident

• boundaries may shift

Oceanic structure classified