182
IAN ROBINSON
GHRSST-PP has specified a core SST product which consists of the same
SST data as received from the original processing agency (e.g. one of the
data products listed in Table 4) with the addition of:
i. a confidence value associated with the probability of the data being
corrupted by such things as cloud cover (IR only), high winds (m/w
only), diurnal warming and proximity to land;
ii. single sensor error statistics (SSES) defining the bias and standard
deviation applicable to that particular sensor depending on the
confidence flag; and
iii. sufficient information to be able to calculate the parameterized values
of the thermal skin deviation and diurnal warming at the precise time
of the satellite sample (allowing conversion to SST fnd ).
Note that for this basic product there is no intention of resampling or
regridding the data specified in Table 5, which are typically Level 2 image
datasets (presented in the sensor co-ordinates). Hence it is referred to as the
GHRSST-PP L2P product (level 2 pre-processed). Neither is there any
merging of data from different sources. It is assumed that the model
assimilation scheme can handle the native grids of the inputs, and any
differences in bias between them, better than any pre-processing or blending
which risks introducing unnecessary additional errors. However, the
addition of (iii) allows the assimilation system to make the conversion to
whatever definition of SST is appropriate for the model, while the provision
of (i) and (ii) allows the influence of data assimilated into a model to be
weighted according to its quality. Note that the SSES in (ii) should be
calculated on the basis of matching against independent validation
measurements of temperature, after making the appropriate correction to
adjust between skin, subskin or foundation SST so that comparison is made
between like quantities. Thus the bias and standard deviation should no
longer contain an element caused by inappropriately comparing unlike
quantities (although they may contain a contribution from errors in the
method used to perform the conversions).
There are four main product types proposed by the GHRSST-PP
processing model:
x The L2P product described in the previous two paragraphs, intended
as the primary source of data for assimilation into numerical ocean
forecasting models, and therefore required in near-real time.
x A match-up data base (MDB) that pairs spatially and temporally
coincident values of SST independently measured in situ and by
satellite.
x Diagnostic data sets (DDS) will be produced for a number of small
regions chosen to represent different ocean and atmosphere conditions
around the world.
x An analysed SST product at level 4 that is a blend of all the available
SST data, each converted to SST fnd and then used in an optimal
IAN ROBINSON
GHRSST-PP has specified a core SST product which consists of the same
SST data as received from the original processing agency (e.g. one of the
data products listed in Table 4) with the addition of:
i. a confidence value associated with the probability of the data being
corrupted by such things as cloud cover (IR only), high winds (m/w
only), diurnal warming and proximity to land;
ii. single sensor error statistics (SSES) defining the bias and standard
deviation applicable to that particular sensor depending on the
confidence flag; and
iii. sufficient information to be able to calculate the parameterized values
of the thermal skin deviation and diurnal warming at the precise time
of the satellite sample (allowing conversion to SST fnd ).
Note that for this basic product there is no intention of resampling or
regridding the data specified in Table 5, which are typically Level 2 image
datasets (presented in the sensor co-ordinates). Hence it is referred to as the
GHRSST-PP L2P product (level 2 pre-processed). Neither is there any
merging of data from different sources. It is assumed that the model
assimilation scheme can handle the native grids of the inputs, and any
differences in bias between them, better than any pre-processing or blending
which risks introducing unnecessary additional errors. However, the
addition of (iii) allows the assimilation system to make the conversion to
whatever definition of SST is appropriate for the model, while the provision
of (i) and (ii) allows the influence of data assimilated into a model to be
weighted according to its quality. Note that the SSES in (ii) should be
calculated on the basis of matching against independent validation
measurements of temperature, after making the appropriate correction to
adjust between skin, subskin or foundation SST so that comparison is made
between like quantities. Thus the bias and standard deviation should no
longer contain an element caused by inappropriately comparing unlike
quantities (although they may contain a contribution from errors in the
method used to perform the conversions).
There are four main product types proposed by the GHRSST-PP
processing model:
x The L2P product described in the previous two paragraphs, intended
as the primary source of data for assimilation into numerical ocean
forecasting models, and therefore required in near-real time.
x A match-up data base (MDB) that pairs spatially and temporally
coincident values of SST independently measured in situ and by
satellite.
x Diagnostic data sets (DDS) will be produced for a number of small
regions chosen to represent different ocean and atmosphere conditions
around the world.
x An analysed SST product at level 4 that is a blend of all the available
SST data, each converted to SST fnd and then used in an optimal