266
H.M. Beggs
analyses range in global mean RMS error from 0.4 to 0.6 ◦ C (Table 15.4) when compared with independent observations from Argo floats or drifting buoys. In order to
meet the WMO goals for SST product accuracy for various operational systems of
0.1–0.3 ◦ C, it is essential that calibration and validation of satellite TIR over the
ocean be improved. In particular, calibration of all satellite SST data needs to be
improved over the Southern and Arctic Oceans (Reynolds et al., 2010; Donlon et al.,
2009a). More in-situ SST observations at high latitudes are required to drive these
improvements both for calibration and validation.
One important issue limiting satellite SST accuracy is that drifting and moored
buoys are generally used to calibrate SST, with their associated representativeness
and instrument errors and limited precision. Sea surface temperature measurements from buoys are recorded at a precision of 0.1 ◦ C on the WMO Global
Telecommunications System (GTS) with an instrument error of around 0.3 ◦ C for
point measurements at 0.5–2 m depth, whereas TIR sensors on satellites measure
the skin SST over an ocean area of between 1 km × 1 km and 6 km × 6 km depending on sensor. A possible solution is to use physical retrieval methods such as those
applied to AATSR IR SST (Merchant and Le Borgne, 2004) and more recently to
AVHRR IR SST (Bogdanoff et al., 2009) to calibrate IR SST rather than regression
against in-situ SST. In-situ measurements of skin SST can then be used to validate
the satellite skin SST produced using this method. Since ship-borne radiometers are
capable of routinely measuring skin SST to <0.1 ◦ C (Donlon et al., 2008), a global
sustained array of these sensors on ships will be necessary to validate satellite IR
SST to the goal accuracy of 0.1 ◦ C.
There is a clear need for global coverage (ideally 6 satellites in orbit) of at least
hourly skin SST from geostationary satellites (Donlon et al., 2009a), and the calibration of SST from IR sensors on GOES and MTSAT-1R geostationary satellites
in particular needs to be improved to approach the accuracy of AVHRR sensors
(Beggs et al., 2009b). Physical retrieval calibration methods should be investigated
for application to IR sensors on geostationary satellites.
Not only should the calibration of all single view satellite IR SST sensors be
improved, but also methods for determining the single pixel SST bias and standard
deviation. Operational systems require SST inputs with reliable error estimates per
pixel determined by the producer, thereby removing the necessity for user-designed
bias-correction schemes depending on either in-situ SST or other satellite IR SST
data such as from AATSR. Logically, the IR SST data producers should be more
expert in the error characteristics of their data stream than the operational users.
Were operational systems to use consistent error estimates for each input SST data
stream, then this would significantly aid SST inter-comparison efforts such as the
GHRSST Multi-Product Ensemble and Intercomparison Project. 13
The processing of the level 2 IR SST should also be improved, with a systematic
review of cloud screening and ice masking methods being undertaken to potentially
improve the quality of satellite IR SST data sets (Donlon et al., 2009a). Following
13 http://www.ghrsst.org/Todays-global-SST.html
H.M. Beggs
analyses range in global mean RMS error from 0.4 to 0.6 ◦ C (Table 15.4) when compared with independent observations from Argo floats or drifting buoys. In order to
meet the WMO goals for SST product accuracy for various operational systems of
0.1–0.3 ◦ C, it is essential that calibration and validation of satellite TIR over the
ocean be improved. In particular, calibration of all satellite SST data needs to be
improved over the Southern and Arctic Oceans (Reynolds et al., 2010; Donlon et al.,
2009a). More in-situ SST observations at high latitudes are required to drive these
improvements both for calibration and validation.
One important issue limiting satellite SST accuracy is that drifting and moored
buoys are generally used to calibrate SST, with their associated representativeness
and instrument errors and limited precision. Sea surface temperature measurements from buoys are recorded at a precision of 0.1 ◦ C on the WMO Global
Telecommunications System (GTS) with an instrument error of around 0.3 ◦ C for
point measurements at 0.5–2 m depth, whereas TIR sensors on satellites measure
the skin SST over an ocean area of between 1 km × 1 km and 6 km × 6 km depending on sensor. A possible solution is to use physical retrieval methods such as those
applied to AATSR IR SST (Merchant and Le Borgne, 2004) and more recently to
AVHRR IR SST (Bogdanoff et al., 2009) to calibrate IR SST rather than regression
against in-situ SST. In-situ measurements of skin SST can then be used to validate
the satellite skin SST produced using this method. Since ship-borne radiometers are
capable of routinely measuring skin SST to <0.1 ◦ C (Donlon et al., 2008), a global
sustained array of these sensors on ships will be necessary to validate satellite IR
SST to the goal accuracy of 0.1 ◦ C.
There is a clear need for global coverage (ideally 6 satellites in orbit) of at least
hourly skin SST from geostationary satellites (Donlon et al., 2009a), and the calibration of SST from IR sensors on GOES and MTSAT-1R geostationary satellites
in particular needs to be improved to approach the accuracy of AVHRR sensors
(Beggs et al., 2009b). Physical retrieval calibration methods should be investigated
for application to IR sensors on geostationary satellites.
Not only should the calibration of all single view satellite IR SST sensors be
improved, but also methods for determining the single pixel SST bias and standard
deviation. Operational systems require SST inputs with reliable error estimates per
pixel determined by the producer, thereby removing the necessity for user-designed
bias-correction schemes depending on either in-situ SST or other satellite IR SST
data such as from AATSR. Logically, the IR SST data producers should be more
expert in the error characteristics of their data stream than the operational users.
Were operational systems to use consistent error estimates for each input SST data
stream, then this would significantly aid SST inter-comparison efforts such as the
GHRSST Multi-Product Ensemble and Intercomparison Project. 13
The processing of the level 2 IR SST should also be improved, with a systematic
review of cloud screening and ice masking methods being undertaken to potentially
improve the quality of satellite IR SST data sets (Donlon et al., 2009a). Following
13 http://www.ghrsst.org/Todays-global-SST.html