If you aren't interested in the text, at least look at the picture (below).
1. The Climate Change Initiative (CCI) should contribute a long-term scientific legacy of climate-quality satellite observations with uncertainty
estimates. CCI climate data records (CDRs) should be as precious to Earth System Science in 100 years' time as the fundamental meteorological and oceanographic observations made in the 19th century are to us now. Practically,
this implies a transition from the present Phase 1 activities (user
consultation, algorithm selection, system definition and prototyping, off-line
production) to routine CDR production and problem-solving. Routine production will support provision
of climate services, the scientific/societal/policy demand for which will
likely be long-term.
2. ESA have wisely required scientists and system engineers
to work together in the prototyping and system specification in Phase 1. This
close co-operation must continue through Phase 2 and thereafter: transition is
not merely a software implementation task, and should include embedded science teams for problem-solving in the CDRs.
3. The
maturity and functionality of prototype software from Phase 1 will vary
between ECVs. For SST and several other variables with functional prototypes, Phase 2 should focus on
commissioning existing prototypes in routine reliable infrastucture and using this for initial operations (including at least one
improvement cycle in Phase 2 I'd suggest – see figure above). This may involve code revision/refactoring, although complete rewriting/re-engineering of code from DPMs should not be necessary (and would likely be scientifically counter-productive).
4. During Phase 2 (in parallel with routine production), we should demonstrate the process of going through at least one cycle of feedback and improvement of the CDR. A concept for this is presented in the figure above. Including this aspect
in Phase 2 will prove the approach for the later phases of ‘continuous
development and operations’ that should continue the work of the CCI long term.
5. Therefore, ‘routine CDR generation’ must be viewed broadly,
as in the figure, to include
·
routine, reliable provision of climate-quality
observations via robust implementation of excellent scientific methods; and
·
the agile scientific work-flow that delivers
upgraded CDRs in response to new requirements and problems.
6. Continuous development is essential in parallel to
routine generation because
·
we continually learn how better to reduce
uncertainties and improve stability of CDRs
·
experiences of users will always uncover new
problems in these huge, complex data sets
·
inputs (level0 and/or level1) will periodically
be reprocessed and improved by agencies
·
new sensors will require seamless integration
into the CDR, while preserving stability and other quality aspects
·
new user requirements will emerge, demanding
new, value-added or better quality CDRs.
7. Routine generation must include continuous
availability of a full, consistently reprocessed, continuously updated CDR. In the figure, this is the
purpose of having overlapping periods between CDR version. The new version running in parallel would be designated "pre-operational" in a meteorological agency. Pre-operational provision gives users time to move across to the next version
CDR.
8. The improvement cycle (loop from problem identification
to reprocessing) requires agile implementation of scientific innovations (e.g.,
new algorithms) within a robust and traceable system environment. The full
effect on the CDR is apparent only on reprocessing the full archive. The system
must therefore be capable of maintaining routine delivery of ‘new’ data while
simultaneously undertaking rapid (preferably multiple) reprocessing as part of
this improvement cycle.
9. Distinctions between near-real time operations and
routine CDR generation are real and should be recognised. One distinction is the
need for the CDR improvement cycle as part of the operations, discussed above.
Another is that stability and consistency take on a greater importance for a
CDR. Lastly, short-time delivery with a lag of days is generally
acceptable. This time delay can, and should, be exploited to maximize the
climate-quality aspects of the CDRs, beyond what is possible with near-real
time provision. (For example, processing for consistency across a constellation, or stability in time can take advantage of the short time lag in delayed-mode
delivery.)