170
IAN ROBINSON
T s = aT bi +b(T bi - T bj ) + c
(6)
where a, b and c are coefficients to be determined, provide a good basis for
atmospheric correction of the AVHRR (McClain et al., 1985). During the
day the split-window algorithm uses wavebands at 10.3-11.3 Pm and 11.512.5 Pm, while at night the 3.5-3.9 Pm channel can also be used. This
3.7 Pm channel is corrupted by reflected solar radiation in the daytime. A
number of non-linear variants of this basic form have also been developed
(Barton, 1995). The algorithm is also supposed to accommodate the nonblackness of the sea.
Common to each of these approaches for AVHRR is the requirement for
the coefficients to be determined by a best fit between the satellite
predictions and coincident observations of SST from a number of drifting
buoys. The match between the buoys and satellites has a variance of more
than 0.5 K, applicable only to the regions populated by the buoys (Podesta et
al., 1995). The same algorithms are assumed to apply to parts of the ocean
where there are no buoys, although the validity of this assumption needs to
be quantified. Regional algorithms matched to local data may achieve
greater accuracy.
Although the instantaneous distribution of water vapour and aerosols in
the atmosphere are not known, the radiation transfer physics of the
atmosphere is well understood and can be modelled with some confidence in
fine spectral detail. It is therefore possible to simulate T b for a given
combination of T s , atmospheric profile and viewing angle for the spectral
characteristics and viewing geometry of each channel of a particular sensor.
This offers an alternative strategy for atmospheric correction in which an
artificial dataset of matching T s and T bi , T bj , etc. is created using a wide
variety of typical atmospheric water vapour and temperature profiles. The
coefficients for an equation of form similar to (6) are generated by a
regression fit to the artificial dataset. The resulting algorithm should be
applicable to all atmospheric circumstances similar to those included in the
modelled dataset, leading to an estimate of the skin SST. It is independent
of coincident in situ measurements, although they are needed for validation.
This was the approach adopted for the along track scanning radiometer
(ATSR) flown on the ERS polar orbiting satellites (Edwards et al., 1990).
The ATSR scans conically to observe a forward view at about 60q incidence
angle and a near-nadir view about two minutes later. Using the same three
spectral channels as AVHRR for each of the views, it thus acquires six
measures of brightness temperature. The different path lengths for forward
and nadir views make for a more robust algorithm (Zavody et al., 1995), less
reliant on the spectral dependence of the atmospheric attenuation. The
single-view approach was rendered inoperative when large volumes of
volcanic dust were temporarily injected into the stratosphere by the eruption
of Mt. Pinatubo in 1991 (Reynolds, 1993). Although the first ATSR
“
”
IAN ROBINSON
T s = aT bi +b(T bi - T bj ) + c
(6)
where a, b and c are coefficients to be determined, provide a good basis for
atmospheric correction of the AVHRR (McClain et al., 1985). During the
day the split-window algorithm uses wavebands at 10.3-11.3 Pm and 11.512.5 Pm, while at night the 3.5-3.9 Pm channel can also be used. This
3.7 Pm channel is corrupted by reflected solar radiation in the daytime. A
number of non-linear variants of this basic form have also been developed
(Barton, 1995). The algorithm is also supposed to accommodate the nonblackness of the sea.
Common to each of these approaches for AVHRR is the requirement for
the coefficients to be determined by a best fit between the satellite
predictions and coincident observations of SST from a number of drifting
buoys. The match between the buoys and satellites has a variance of more
than 0.5 K, applicable only to the regions populated by the buoys (Podesta et
al., 1995). The same algorithms are assumed to apply to parts of the ocean
where there are no buoys, although the validity of this assumption needs to
be quantified. Regional algorithms matched to local data may achieve
greater accuracy.
Although the instantaneous distribution of water vapour and aerosols in
the atmosphere are not known, the radiation transfer physics of the
atmosphere is well understood and can be modelled with some confidence in
fine spectral detail. It is therefore possible to simulate T b for a given
combination of T s , atmospheric profile and viewing angle for the spectral
characteristics and viewing geometry of each channel of a particular sensor.
This offers an alternative strategy for atmospheric correction in which an
artificial dataset of matching T s and T bi , T bj , etc. is created using a wide
variety of typical atmospheric water vapour and temperature profiles. The
coefficients for an equation of form similar to (6) are generated by a
regression fit to the artificial dataset. The resulting algorithm should be
applicable to all atmospheric circumstances similar to those included in the
modelled dataset, leading to an estimate of the skin SST. It is independent
of coincident in situ measurements, although they are needed for validation.
This was the approach adopted for the along track scanning radiometer
(ATSR) flown on the ERS polar orbiting satellites (Edwards et al., 1990).
The ATSR scans conically to observe a forward view at about 60q incidence
angle and a near-nadir view about two minutes later. Using the same three
spectral channels as AVHRR for each of the views, it thus acquires six
measures of brightness temperature. The different path lengths for forward
and nadir views make for a more robust algorithm (Zavody et al., 1995), less
reliant on the spectral dependence of the atmospheric attenuation. The
single-view approach was rendered inoperative when large volumes of
volcanic dust were temporarily injected into the stratosphere by the eruption
of Mt. Pinatubo in 1991 (Reynolds, 1993). Although the first ATSR
“
”