SATELLITE MEASUREMENTS
171
algorithms were also affected, a reworking of the semi-physical model by
including the stratospheric aerosols in the radiation model led to algorithms
which cope well with the volcanic problem (Merchant et al. 1999). This
approach should also be robust in situations such as those where dust from
the Sahara is lifted into the troposphere over the Atlantic.
The atmospheric correction algorithms produce maps of SST at fine
resolution (about 1.1 km) for each overpass. However, the atmospheric
correction methods cannot retrieve SST when cloud wholly or partly
obstructs the field of view. Therefore at this stage cloud must be detected
using a variety of tests (e.g., Saunders and Kriebel, 1988), so that only
cloud-free pixels are retained for oceanographic applications, such as
assimilation into models. The most difficult cloud contamination to identify
is that by sub-pixel size clouds, thin cirrus or sea fog where only small
deviations of temperature occur.
Failure to detect cloud leads to
underestimation of the SST and can produce cool biases of order 0.5 K.
Thus confidence in the cloud detection procedure is just as important as
atmospheric correction for achieving accurate SST. Where uncertainty
remains in cloud detection, this should be flagged in the error estimate fields
attached to SST products. Cloud detection is generally more successful
during daytime, when visible and near-IR image data can be used, than at
night.
The SSTs measured in individual overpasses are incorporated into global
composite datasets by averaging all individual pixel contributions to each
larger cell over a period of a few days. The larger cells are defined by
longitude and latitude on a grid with spacing typically 1/2 or 1/6 degree
(about 50km or 16 km at the equator). The multi-channel sea surface
temperature (MCSST) (Walton et al., 1998) was the standard global
composite product derived from AVHRR, until superseded by the Pathfinder
SST (Vasquez et al., 1998). This is a re-processing of the archived pixellevel AVHRR data with algorithms incorporating the best knowledge of
sensor calibration drift and making full use of the available drifting buoy
dataset (Kilpatrick et al., 2001). It also makes more use of night-time data
than previous analyses, and aims for long term consistency. Reynolds and
Smith (1994) developed an OI-SST archive that is an optimal interpolation
of both in situ and satellite data, and should therefore provide more
climatological continuity with pre-satellite SST records before 1980. The
global composite product from ATSR is the ASST (Murray, 1995) which is
being re-processed using the more robust atmospheric algorithms (Merchant
et al., 1999).
3.3.3
Microwave radiometers on satellites
Microwave radiometers detect the brightness temperature of microwave
radiation which, like the infrared, depends on the temperature of the emitting
171
algorithms were also affected, a reworking of the semi-physical model by
including the stratospheric aerosols in the radiation model led to algorithms
which cope well with the volcanic problem (Merchant et al. 1999). This
approach should also be robust in situations such as those where dust from
the Sahara is lifted into the troposphere over the Atlantic.
The atmospheric correction algorithms produce maps of SST at fine
resolution (about 1.1 km) for each overpass. However, the atmospheric
correction methods cannot retrieve SST when cloud wholly or partly
obstructs the field of view. Therefore at this stage cloud must be detected
using a variety of tests (e.g., Saunders and Kriebel, 1988), so that only
cloud-free pixels are retained for oceanographic applications, such as
assimilation into models. The most difficult cloud contamination to identify
is that by sub-pixel size clouds, thin cirrus or sea fog where only small
deviations of temperature occur.
Failure to detect cloud leads to
underestimation of the SST and can produce cool biases of order 0.5 K.
Thus confidence in the cloud detection procedure is just as important as
atmospheric correction for achieving accurate SST. Where uncertainty
remains in cloud detection, this should be flagged in the error estimate fields
attached to SST products. Cloud detection is generally more successful
during daytime, when visible and near-IR image data can be used, than at
night.
The SSTs measured in individual overpasses are incorporated into global
composite datasets by averaging all individual pixel contributions to each
larger cell over a period of a few days. The larger cells are defined by
longitude and latitude on a grid with spacing typically 1/2 or 1/6 degree
(about 50km or 16 km at the equator). The multi-channel sea surface
temperature (MCSST) (Walton et al., 1998) was the standard global
composite product derived from AVHRR, until superseded by the Pathfinder
SST (Vasquez et al., 1998). This is a re-processing of the archived pixellevel AVHRR data with algorithms incorporating the best knowledge of
sensor calibration drift and making full use of the available drifting buoy
dataset (Kilpatrick et al., 2001). It also makes more use of night-time data
than previous analyses, and aims for long term consistency. Reynolds and
Smith (1994) developed an OI-SST archive that is an optimal interpolation
of both in situ and satellite data, and should therefore provide more
climatological continuity with pre-satellite SST records before 1980. The
global composite product from ATSR is the ASST (Murray, 1995) which is
being re-processed using the more robust atmospheric algorithms (Merchant
et al., 1999).
3.3.3
Microwave radiometers on satellites
Microwave radiometers detect the brightness temperature of microwave
radiation which, like the infrared, depends on the temperature of the emitting