Monitoring a Sentinel Species from Satellites: Detecting Emiliania huxleyi in 25 Years of AVHRR Imagery - Satellite-based Applications on Climate Change - page 271

18.2.3 Detecting E. huxleyi Blooms

E . huxleyi blooms were classified in AVHRR daily imagery by applying an R rs
threshold tuned to the well-validated SeaWiFS maps. Whenever two AVHRR
sensors were flying, daily composites were created incorporating both sensors.
A threshold of 1.0 was established for R rs so the spatial extent of E . huxleyi from
AVHRR was similar to maps derived from SeaWiFS for blooms in the Barents Sea
(Smyth et al. 2004 ) and North Atlantic Ocean south of Iceland (Raitsos et al. 2006 ),
where in situ samples had confirmed E . huxleyi presence at high numbers. For each
AVHRR scene, pixels of R rs >

1.0
were assigned 0. An R rs threshold of 1.5 was applied for pixels with solar zenith angles
between 50 o and 70 o after observing several false-positive results correlated to high
solar zenith angles. Weekly composite AVHRR E . huxleyi maps were created by
summing these binary data over 8 days for each pixel, such that pixel values could
range from 0 ( E . huxleyi absent) to 8 ( E . huxleyi always present), then dividing by the
number of valid observations in the weekly composite to estimate the percentage of
time a pixel possessed E . huxleyi during the period. Fractions were averaged over the
entire region for a final value ranging between 0 (no E . huxleyi in the region) and 1
(a bloom covered the whole region for the 8-day period). Only pixels with a minimum
of 25% valid data, or 2 days out of eight, were included in the final area average.
Repeating our analyses with more stringent cloud and aerosol screening criteria
yielded nearly identical patterns, albeit fewer data, suggesting that this E . huxleyi
classification method is not biased by nearby clouds.
In order to minimize false-positives in both AVHRR and SeaWiFS data, we
excluded imagery collected during winter months (i.e., October-March in the
Northern Hemisphere, vice versa in the Southern) when materials such as diatom
frustules are resuspended by high winds and spectrally mimic E . huxleyi blooms
(Broerse et al. 2003 ). We applied a bathymetric threshold (depth

1.0 were assigned a value of 1; pixels of R rs <

150 m) between
45ºS and 45ºN to avoid the incorrect classification of shallow carbonate shelves as
coccolithophore blooms.
The number of times E . huxleyi blooms were identified in a pixel of both AVHRR
and SeaWiFS (

>

1 week) was summed as was the number of times either AVHRR or
SeaWiFS detected a bloom when the other did not. From this map, four open ocean
and coastal regions were selected where E . huxleyi was frequently present and the
matchups of blooms derived from both sensors were most consistent as well as being
locations where blooms had been sampled during cruises (Fig. 18.2 ): the Bering Sea
west of Alaska, the North Atlantic south of Iceland, the North and Norwegian Seas,
and the Patagonian Shelf east of Argentina. Over 216 coincident weeks, the correla-
tion between AVHRR and SeaWiFS E . huxleyi detection was significant: r ¼ 0.46
(Bering Sea), r ¼ 0.63 (south of Iceland), r ¼ 0.39 (North and Norwegian Seas), and

r ¼ 0.30 (Patagonian Shelf). Although the Black Sea had an abundance of blooms

consistently identified by both sensors, it was excluded because the focus of this study
was on open ocean systems.
The areal extent of blooms in each region was estimated by averaging bloom
pixels. After integrating over each year, the maximum bloom area was normalized