Abstract
This paper describes algorithm of cloud detection and atmospheric correction for deriving Sea surface temperature (SST) using FY-2 Infrared Channel. It also briefly introduce SST operational processing system.

Introduction
China first launched the FY-2 Geostationary Meteorological Satellite on 10 June 1997. The Satellite carries a multi-channel scan radiometer, with a visible channel, a thermal infrared channel, and a water vapor channel. The infrared window channel (10.5mm-12.5mm) can be used to derive SST. SST is important operational product in the ground application processing system of FY-2 and useful for climate research and oceanographic application. At present, SST products derived from FY-2 are generated every at 0.5° latitude by longitude box covering the area from 50° N to 50° S and 55° E to 155° E.

SST Algorithm for FY-2
The key problems of SST derived from satellite data are recognition of cloud- contaminated data and correction of atmospheric Attenuation, because the radiation reaching the sounder aboard satellite has been contaminated by clouds and attenuated by atmosphere.

Cloud detection
The cloud detection Are based on the premises that: sea surface brightness temperature is higher cloud top, sea surface reflectance is lower than cloud top, and SST is slowly varying in space and time etc. These methods and discrimination equations for FY-2 are summarized as follows:

Histogram method
Histogram is constructed in 0.5° latitude by 0.5° longitude interval for IR (10.5 - 12.5mm) of FY-2, the following methods are used for histogram:

Sea / land discrimination
Sea / land discrimination is Preprocessing step before cloud detection.

The sea/land tags table is used to discriminate between sea and land. When 20% pixels in the histogram are located at land, the grid point of 0.5° by 0.5° is labeled as land.

Visible (VIS) and Infrared (IR) gross cloud test

R <10% (for each pixel)         (1)
Where, R is the reflectance of FY-2 visible channel. This test is only for daytime, When the reflectance of a pixel in the histogram is great than 10%, the grid point of 0.5° by 0.5° is discriminated as cloud.
Tb <270° k (for each pixel)         (2)
Where, Tb is the brightness temperature of IR. This test is based on the fact sea water frizzed at 271° k. Over open ocean when Tb is less than 270° k, we can safely assume cloud is present. When Tb of a pixel in the histogram is less than 270° k, the grid point of 0.5° by 0.5° is discriminated as cloud. .

IR uniformity test and Frequency test

Tbmax -Tbmin <Tth        (3)
Where, Tbmax and Tbmin are maximum and minimum brightness temperature of IR in the histogram, respectively, and Tth is threshold value, Tth = 2.°k. If the difference between Tbmax and Tbmin is more than 2°k, the grid point of 0.5° by 0.5° is discriminated as cloud. After uniformity test passing. Then frequency test is performed.
N >80% (N=(f 1+f 2+f 3)/F)         (4)
Where, f1, f2, f3 are the Frequencies for three class temperatures (tb1, tb2, tb3) of IR on the warm side of the histogram, respectively, and F is total frequencies of IR of the histogram. IF N is more than 80% for passing test, Tb₀, the mean brightness temperature of cloud-free of the histogram, can be calculated:
Tb₀=(tb1*f1+tb2*f2+tb3*f3)/(f1+f2+f3)         (5)
Following steps are necessary for SST to perform the quality control of cloud detection:

SST Climatology test

|SST-SST_clim|<6°C        (6)
Where, SST is the retrieved SST, and SST_clim is SST climatological temperature for the nearest 1° latitude by longitude intersection. If the difference is more than 6° C, the SST is rejected.

SST Comparison test everyday 8 times

SST_max - SST_t < 1.5° C (t=1,2...8)         (7)
Where, SST_max is maximum SST in everyday 8 times, t represents the number of time in 8 times, SSTt is the SST of t time. If the range of the differences is more than 1.5° C, the SST of t. time is rejected.

SST comparison test in the five days

|SST_max - SST_t| <3.5° C (t=1,2...5)         (8)
Where, SST_max is maximum SST in the five days, t represents the number of day in five days, SST_t is the SST of t day. If the range of the differences is more than 3.5° C, the SST is rejected.

Atmospheric Attenuation correction
The scan radiometer on the FY-2 satellite has only one infrared window channel, SST are derived from FY-2 using a statistical estimation method. Lead to the linear approximation for the radiative transfer equation, it may be expressed as matrix equation:
Y = K^T X         (9)
Where Y is the matrix of observation brightness temperature differences (relative preliminary state); K is the weighted function matrix, and X is the estimated matrix. On the basis of the statistical estimative principle, estimative solutions of the equation have been got, as
ÙX = S_xK (K^TS_xK+SY) - ¹ Y         (10)
Where Sx is the covariance matrix of a priory information; SY is the measurement error matrix. Then SST can be computed from the estimative values.