Appendix 1

A-1-1. Atmospheric Correction
	The raw data from CZCS, in digital counts, were first converted into 
radiance values following the standard calibration method.3) Atmospheric 
corrections were then carried out based on a specific algorithm, which is explained 
in detail in another paper6) and is briefly shown here. The algorithm 
is based on the NASA standard atmospheric correction algorithm7), 8) 
used for the NASA global CZCS composite4) but with modifications in 
consideration of the local Asian dust aerosol made as follows. First, the 
algorithm calculates an aerosol spectral dependency parameter, or Angstrom 
coefficient (epsilon)   for each pixel based on the aerosol scattering radiances at 
670 and 550 nm. If the epsilon value is less than 1.0, the NASA standard 
algorithm is applied. Otherwise, pixel-wise aerosol absorption and aerosol 
single scattering albedo for each band (443, 520, and 550 nm) are iteratively 
calculated by the empirical relationship between these properties and the 
deviation of the epsilon value from 1.0. This atmospheric correction takes 
into account the multiple-Rayleigh scattering correction with polarization 
effect.8) Pixel-wise surface pressure corrections and ozone absorption 
corrections were also done using ECMWF (European Center for Medium Range Weather 
Forecasts) data and the Gridded TOMS (Total Ozone Mapping System) ozone data set.
	Although the atmospheric correction calculates water-leaving radiances
(Lw), they are normalized to absorb the differences in solar irradiance at 
surface level before being stored in the normalized water-leaving radiance (nLw) 
files in preparation for averaging. 

A-1-2. Cloud Screening
	During the atmospheric correction, cloud screening and near-cloud 
screening algorithms were applied. A pixel is classified as "cloud or land" if 
it has a "normalized" aerosol radiance at 670 nm (La(670)) of 1.5 micro W/cm2/nm/sr or
higher. 

A-1-3. Cloud-Ringing Mask
	To screen out pixels affected by the unwanted transient response 
following the scanning of a bright cloud, a certain number of pixels to the 
right of the cloud edge are also masked by Mueller's "cloud-ringing mask algorithm"
13) in consideration of electrical overshoot in the instrument.

A-1-4. Automated Pixel-Wise Quality Control
	Following the atmospheric correction, an automated pixel-wise quality 
control was applied: A pixel will be masked if the atmospherically corrected 
value is unrealistically low in one or more bands. More specifically, the pixel 
was judged to be good if nLw(443) > 0.2 W/cm2/nm/sr, nLw(520) > 0.25 W/cm2/nm/sr, 
and nLw(550) > 0.2 W/cm2/nm/sr. These thresholds were determined by 
reference to Gordon et al.9) who modelled water-leaving radiance as a function 
of the pigment concentration.

A-1-5. Bio-optical Algorithm
	Pixel-wise phytoplankton pigment concentration (P) is estimated from the
combination of atmospherically-corrected water-leaving radiances Lw at 443, 520,
and 550 nm using the following equation provided by D. Clark:11)
	Ratio = [ ( Lw(443) + Lw(520) ) / Lw(550)]
	P = 5.56 * Ratio**(-2.252)

A-1-6 Statistical Procedure
	Pigment concentration estimates (P), normalized water-leaving radiances
for 443 and 550 nm (nLw443, nLw550), and aerosol radiances at 670 nm (La670) of
individual images were separately collected and averaged for each "bin," or 
approximately 3 km by 4 km partition of the ground surface, over each month from
December 1978 through June 1986. In total, 768 x 1536 bins were used to cover 
the area to produce monthly composite images of the same size.
	The averaging of each target parameter X was done by a maximum likelihood
estimator with the following equation:
	Xmle = exp (m + s**2/ 2),
where m and s are the mean and standard deviation of X after log-transformation. 
This procedure was proposed by J. Campbell2) because the parameters 
of the composite are log-normally distributed and because the sample sizes are 
small.