222
SYLVIE POULIQUEN
The advantages of such system are:
¾ Optimisation of the resources (network, CPU, Memory, etc) among
the contributors,
¾ Data stay where they are generated preventing non compatible
duplicates among the network
¾ Built on internationally agreed standards that guaranty its efficiency
in the long term and its adaptability because it will benefit from
international shared developments.
The disadvantages are:
¾ The system is not easy to set up because it needs a lot of
international coordination, especially for metadata.
¾ Even more work for small contributors because it requires important
computer expertise
¾ It can be unreliable if some data providers cannot guaranty data
service on the long term. To be reliable such a system must rely on
sustained data centres.
5.2
Quality control procedures
These procedures have to be adapted to the allowed delay of the delivery.
In real-time, most of these QC are made automatically and only outliers are
rejected. In delayed mode, more scientific expertise is applied to the data and
error estimation can be provided with the data.
Data quality control is a fundamental component of any ocean data
assimilation system because accepting erroneous data can cause incorrect
forecast, but rejecting extreme data can also lead to erroneous forecast by
missing important events or anomalous features.
The challenge of quality control is to check the input data against a preestablished "ground truth". But who really knows this truth when we know
that the ocean varies in time and space, but also that no instrument gives an
exact value of any parameter but only an estimation of the "truth" within
some error bars.
For operational oceanography, other problems must be solved. First, the
forecast requires quality-controlled data within one day. This means that
only automated or semi-automated quality control procedures can be
applied. Second, most of the data are processed by different actors, but used
all together by operational models: this implies a clear documentation of the
quality control procedures, an homogenisation of the quality flags, a
reliability of different actors in applying these rules. Third, for re-analysis
SYLVIE POULIQUEN
The advantages of such system are:
¾ Optimisation of the resources (network, CPU, Memory, etc) among
the contributors,
¾ Data stay where they are generated preventing non compatible
duplicates among the network
¾ Built on internationally agreed standards that guaranty its efficiency
in the long term and its adaptability because it will benefit from
international shared developments.
The disadvantages are:
¾ The system is not easy to set up because it needs a lot of
international coordination, especially for metadata.
¾ Even more work for small contributors because it requires important
computer expertise
¾ It can be unreliable if some data providers cannot guaranty data
service on the long term. To be reliable such a system must rely on
sustained data centres.
5.2
Quality control procedures
These procedures have to be adapted to the allowed delay of the delivery.
In real-time, most of these QC are made automatically and only outliers are
rejected. In delayed mode, more scientific expertise is applied to the data and
error estimation can be provided with the data.
Data quality control is a fundamental component of any ocean data
assimilation system because accepting erroneous data can cause incorrect
forecast, but rejecting extreme data can also lead to erroneous forecast by
missing important events or anomalous features.
The challenge of quality control is to check the input data against a preestablished "ground truth". But who really knows this truth when we know
that the ocean varies in time and space, but also that no instrument gives an
exact value of any parameter but only an estimation of the "truth" within
some error bars.
For operational oceanography, other problems must be solved. First, the
forecast requires quality-controlled data within one day. This means that
only automated or semi-automated quality control procedures can be
applied. Second, most of the data are processed by different actors, but used
all together by operational models: this implies a clear documentation of the
quality control procedures, an homogenisation of the quality flags, a
reliability of different actors in applying these rules. Third, for re-analysis