less critical. Cloud screening algorithms focus on applications for which the
absolute accuracy of the SST value is paramount and rely on differences in
emissivity, reflectivity, temperature and spatial structure between the ocean surface and clouds (Donlon 2010; Cayula and Cornillon 1996).
Degradation of the accuracy of SST can occur when the atmosphere contains
high concentrations of aerosols, such as Saharan or volcanic dust, which generally
introduce a negative bias error (retrieved SST values are too cold) of up to a few
degrees Celsius. For example, the eruption of Mt Pinatubo in June 1991 drastically
affected the ability to monitor temperature globally (Reynolds 1993). Hence, the
affected time periods must either be excluded from any data analysis, or a specific
set of coefficients estimated for the atmospheric correction algorithm corresponding to the particular conditions must be utilized. This has been mitigated to a
certain degree by upgrading algorithms to include a dust index scheme (Merchant
et al. 2006).
Uncertainties in satellite-derived SST can occur as a result of the above mentioned skin effect and diurnal heating, the effects of clouds and aerosols, and
imperfections in the atmospheric correction algorithm (Minnett and Barton 2010).
The standard approach to determine the uncertainties in SST is to compare them
with coincident measurements from independent sources. Several approaches to
validating satellite-derived SST that use different instruments are reviewed in
Minnett (2010). The optimal method to determine the error characteristics of the
satellite SST retrievals is with a ship-based radiometer, such as the MarineAtmospheric Emitted Radiance Interferometer (M-AERI; Minnett et al. 2001),
which can mimic the satellite measurement most closely. For AVHRR, the
coefficients used in the atmospheric correction algorithms are derived by robust
regression between the AVHRR-derived temperatures and in situ measurements
from drifting buoys (Kilpatrick et al. 2001). The measured SSTs are bulk temperature readings that are related to the skin SST through the mean temperature
differences in the matchups between satellite and in situ data. Similarly for MODIS, SST retrievals are of skin temperature, and the mean biases that result from
comparison with the buoy bulk temperature measured at a depth of a meter or
more are a manifestation of the thermal skin effect (Donlon et al. 2002). A
summary of the accuracies of SST derived from various satellite sensors determined by comparison with independent measurements is provided in Minnett
(2010). For example, uncertainties in SST from cloud-free comparisons between
Pathfinder AVHRR retrievals from 1985 to 1998 and drifting buoys provide a
mean bias of 0.02 °C (SD = 0.53 °C). Uncertainties in SST retrievals from all
MODIS 11 lm retrievals compared to bulk buoy temperatures provide a mean bias
of -0.16 °C (SD = 0.55 °C). These are derived from [600,000 matchups with
drifting buoys measuring the bulk temperature. However, when compared with the
radiometric SST from M-AERI, the Pathfinder AVHRR retrievals provide a mean
bias of 0.14 °C (SD = 0.36 °C), and the MODIS retrievals a mean bias of 0.02 °C
(SD = 0.55 °C). This implies differences still exist between derived and in situ
temperatures. In summary, all other things being equal, MODIS data will be cooler
on average by *0.16 °C than AVHRR data since MODIS is calibrated with
316
S. J. Weeks et al.
absolute accuracy of the SST value is paramount and rely on differences in
emissivity, reflectivity, temperature and spatial structure between the ocean surface and clouds (Donlon 2010; Cayula and Cornillon 1996).
Degradation of the accuracy of SST can occur when the atmosphere contains
high concentrations of aerosols, such as Saharan or volcanic dust, which generally
introduce a negative bias error (retrieved SST values are too cold) of up to a few
degrees Celsius. For example, the eruption of Mt Pinatubo in June 1991 drastically
affected the ability to monitor temperature globally (Reynolds 1993). Hence, the
affected time periods must either be excluded from any data analysis, or a specific
set of coefficients estimated for the atmospheric correction algorithm corresponding to the particular conditions must be utilized. This has been mitigated to a
certain degree by upgrading algorithms to include a dust index scheme (Merchant
et al. 2006).
Uncertainties in satellite-derived SST can occur as a result of the above mentioned skin effect and diurnal heating, the effects of clouds and aerosols, and
imperfections in the atmospheric correction algorithm (Minnett and Barton 2010).
The standard approach to determine the uncertainties in SST is to compare them
with coincident measurements from independent sources. Several approaches to
validating satellite-derived SST that use different instruments are reviewed in
Minnett (2010). The optimal method to determine the error characteristics of the
satellite SST retrievals is with a ship-based radiometer, such as the MarineAtmospheric Emitted Radiance Interferometer (M-AERI; Minnett et al. 2001),
which can mimic the satellite measurement most closely. For AVHRR, the
coefficients used in the atmospheric correction algorithms are derived by robust
regression between the AVHRR-derived temperatures and in situ measurements
from drifting buoys (Kilpatrick et al. 2001). The measured SSTs are bulk temperature readings that are related to the skin SST through the mean temperature
differences in the matchups between satellite and in situ data. Similarly for MODIS, SST retrievals are of skin temperature, and the mean biases that result from
comparison with the buoy bulk temperature measured at a depth of a meter or
more are a manifestation of the thermal skin effect (Donlon et al. 2002). A
summary of the accuracies of SST derived from various satellite sensors determined by comparison with independent measurements is provided in Minnett
(2010). For example, uncertainties in SST from cloud-free comparisons between
Pathfinder AVHRR retrievals from 1985 to 1998 and drifting buoys provide a
mean bias of 0.02 °C (SD = 0.53 °C). Uncertainties in SST retrievals from all
MODIS 11 lm retrievals compared to bulk buoy temperatures provide a mean bias
of -0.16 °C (SD = 0.55 °C). These are derived from [600,000 matchups with
drifting buoys measuring the bulk temperature. However, when compared with the
radiometric SST from M-AERI, the Pathfinder AVHRR retrievals provide a mean
bias of 0.14 °C (SD = 0.36 °C), and the MODIS retrievals a mean bias of 0.02 °C
(SD = 0.55 °C). This implies differences still exist between derived and in situ
temperatures. In summary, all other things being equal, MODIS data will be cooler
on average by *0.16 °C than AVHRR data since MODIS is calibrated with
316
S. J. Weeks et al.