232
P.J. Minnett
(NWP) where the sea surface is the bottom boundary condition for an atmospheric
model (e.g. Chapter 15 by Beggs, this volume). A temperature measurement that is
relatively inaccurate is therefore given less weight than one that has smaller uncertainties. A root-mean-square (rms) error of 0.3 K about a zero mean bias is the
target accuracy for WMO observational requirements for global NWP applications
(Eyre et al., 2009). A more relaxed requirement of 0.5 K is given as the “breakthrough” level which if achieved would result in a significant improvement in the
targeted application. An rms uncertainty of 1 K is defined as the “threshold” accuracy, meaning that SSTs with greater uncertainties would not be of use in NWP
applications.
14.4 Validation Techniques
The standard approach to establish the uncertainties in satellite-derived SSTs
is to compare them with coincident measurements from independent sources.
This is called “validation” as it leads to a verification of the in-flight calibration and the performance of the algorithms used to derive SST from the
“top-of-atmosphere” brightness temperature measurements. The objectives of the
validation exercise are to reveal the residual effects of instrumental artifacts in
the raw measurements that have been imperfectly corrected, and of uncompensated effects of the intervening atmosphere. If patterns are identifiable in the
uncertainties that reveal a systematic component to the sources of uncertainties,
these may lead to improved correction algorithms. Included in such analyses is
the quest for dependences on other relevant parameters that influence the satellite measurements, such as the water-vapor content of the atmosphere. If no
clear patterns or dependences are found, the properties of the random or nonsystematic uncertainties provide guidance on the averaging, spatial or temporal,
that may be required to reach a specific level of accuracy required for a particular
application.
Ideally, the reference measurements should be free of error and be taken at the
same time and place as the satellite measurements, and have the same temporal and
spatial sampling characteristics. But such measurements do not exist, so we have to
endeavor to ensure that they are at least more accurate than the satellite retrievals,
and they should share as many of the same characteristics as possible. This means
that radiometric validation measurements are preferred over a sub-surface thermometric measurement, so that “like is compared with like” and additional sources of
uncertainty caused by the near surface temperature gradients (Fig. 14.1) can be rendered negligible. Since the reference measurements are imperfect, having their own
uncertainties, and the method of comparison introduces additional uncertainties, an
error budget has to be constructed that takes into account the contributions from all
sources.
There are several approaches to validating satellite-derived surface temperatures
that use different instruments. Some are mounted on aircraft, others on ships or
buoys.
P.J. Minnett
(NWP) where the sea surface is the bottom boundary condition for an atmospheric
model (e.g. Chapter 15 by Beggs, this volume). A temperature measurement that is
relatively inaccurate is therefore given less weight than one that has smaller uncertainties. A root-mean-square (rms) error of 0.3 K about a zero mean bias is the
target accuracy for WMO observational requirements for global NWP applications
(Eyre et al., 2009). A more relaxed requirement of 0.5 K is given as the “breakthrough” level which if achieved would result in a significant improvement in the
targeted application. An rms uncertainty of 1 K is defined as the “threshold” accuracy, meaning that SSTs with greater uncertainties would not be of use in NWP
applications.
14.4 Validation Techniques
The standard approach to establish the uncertainties in satellite-derived SSTs
is to compare them with coincident measurements from independent sources.
This is called “validation” as it leads to a verification of the in-flight calibration and the performance of the algorithms used to derive SST from the
“top-of-atmosphere” brightness temperature measurements. The objectives of the
validation exercise are to reveal the residual effects of instrumental artifacts in
the raw measurements that have been imperfectly corrected, and of uncompensated effects of the intervening atmosphere. If patterns are identifiable in the
uncertainties that reveal a systematic component to the sources of uncertainties,
these may lead to improved correction algorithms. Included in such analyses is
the quest for dependences on other relevant parameters that influence the satellite measurements, such as the water-vapor content of the atmosphere. If no
clear patterns or dependences are found, the properties of the random or nonsystematic uncertainties provide guidance on the averaging, spatial or temporal,
that may be required to reach a specific level of accuracy required for a particular
application.
Ideally, the reference measurements should be free of error and be taken at the
same time and place as the satellite measurements, and have the same temporal and
spatial sampling characteristics. But such measurements do not exist, so we have to
endeavor to ensure that they are at least more accurate than the satellite retrievals,
and they should share as many of the same characteristics as possible. This means
that radiometric validation measurements are preferred over a sub-surface thermometric measurement, so that “like is compared with like” and additional sources of
uncertainty caused by the near surface temperature gradients (Fig. 14.1) can be rendered negligible. Since the reference measurements are imperfect, having their own
uncertainties, and the method of comparison introduces additional uncertainties, an
error budget has to be constructed that takes into account the contributions from all
sources.
There are several approaches to validating satellite-derived surface temperatures
that use different instruments. Some are mounted on aircraft, others on ships or
buoys.