An Algorithm for Oceanic Front Detection in Chlorophyll and SST Satellite Imagery

1
An Algorithm for Oceanic Front Detection in
Chlorophyll and SST Satellite Imagery
Igor M. Belkin,
University of Rhode Island, and John E.
OReilly, NMFS/NOAA
ibelkin_at_gso.uri.edu and jay.oreilly_at_noaa.gov
1. Abstract. An algorithm is described for
oceanic front detection in chlorophyll (Chl) and
sea surface temperature (SST) satellite imagery
(Belkin and OReilly, 2008). The algorithm is
based on gradient approach. The main novelty is a
shape-preserving, scale-sensitive, contextual
median filter applied selectively and iteratively
until convergence. This filter has been developed
specifically for Chl since these fields have
spatial patterns such as chlorophyll enhancement
at thermohaline fronts and small- and meso-scale
chlorophyll blooms that are not present in SST
fields. Linear Chl enhancements and point-wise
blooms are modeled as ridges and peaks
respectively, whereas conventional fronts in Chl
and SST fields are modeled as steps or ramps.
Examples are presented of the algorithm
performance over a broad range of spatial and
temporal scales, using modeled (synthetic) images
as well as Chl and SST imagery. Satellite data
from several thermal and color sensors (AVHRR,
SeaWiFS and MODIS/Terra and Aqua) were processed
with the new algorithm to generate climatology of
SST and Chl fronts off the U.S. Northeast,
encompassing the Mid-Atlantic Bight, Georges Bank
and Gulf of Maine (Belkin et al., 2008).
3. INTELLIGENT MEDIAN FILTER REMOVES SPIKES
(IMPULSE NOISE) PRESERVES FEATURES STEP-LIKE
FRONTS CHL ENCHANCEMENT at FRONTS CHL PEAKS AND
LOCALIZED BLOOMS ITERATES UNTIL
CONVERGENCE DETECTS OSCILLATIONS
Contextual median filtering of ridges. Thin
(1-pixel wide) and wide (3-pixel wide) spiral
ridges are shown before MF (left), after standard
MF (center), and after contextual MF(right).
Insets are enlarged in the next figure below.
2. Transversal structure of Chlorophyll and SST
fronts
Figure 2. Spatially-averaged Chl concentration
as a function of frontal swath width for the ramp
and peak models (see below).
Figure 1. Schematic of the Shelf-Slope Front
(SSF).
Contextual median filtering of ridges. Thin
(1-pixel wide) and wide (3-pixel wide) spiral
ridges (top panel) are processed with standard MF
(left) and contextual MF (right). Standard MF
removes thin ridge and blunts the crest of wide
ridge (bottom left panel), whereas contextual MF
leaves both ridges intact (bottom right panel).
Median filtering of the model Gulf Stream and its
rings. The model Gulf Streams edges are frayed
with horizontal spread and swap spikes that are
smoothed by standard MF3 (right column).
Contextual MF3in5 leaves the Gulf Stream fronts
edge intact. Isolated vertical spikes within
rings are removed by both standard MF3 and
contextual MF3in5.
Figure 3. Chl distribution in the NW Atlantic,
September 2002. Chl concentration over the shelf
is relatively uniform and significantly higher
than offshore. This distribution is described by
the ramp model (Figure 4).
Figure 4. Ramp model of Chl distribution across
SSF.
Examples of the algorithm performance on synoptic
satellite images of SST (top row 3 May 2001)
and chlorophyll (bottom row 14 October 2000).
Left column, original images. Right column,
gradient magnitude.
Contextual median filtering of peaks and spikes.
Contextual MF3in5 removes 1-point spikes but
leaves intact sharp 3- and 5-point peaks.
Figure 5. Chl distribution in the NW Atlantic,
April 2001. Chl concentration peaks at SSF. This
type of Chl distribution is described by the peak
model (Figure 6).
REFERENCES Belkin, I.M. and J.E. OReilly (2008).
An algorithm for front detection in chlorophyll
and sea surface temperature satellite imagery.
Journal of Marine Systems. Belkin, I.M., J.E.
OReilly, K.J.W. Hyde, and T. Ducas (2008).
Satellite climatology of chlorophyll and sea
surface temperature fronts in the Northeast U.S.
Large Marine Ecosystem. In preparation.
ACKNOWLEDGEMENTS We are grateful to NOAA for
funding this project under the Research to
Operations program and through a contract to the
University of Rhode Island.
Figure 6. Peak model of Chl distribution across
SSF.