Paul Spinks (SCL), Norman Fomferra (BC, also Ralf Quast and Martin Boettcher for part), Nick Rayner (Met Office), Gary Corlett (Leicester, for part) and me (Chris Merchant, Edinburgh)
The purpose of the telecon was to identify, from among the possible tools suggested within the project, which are essential to be implemented as formal tools (by our system engineering team), which as informal tools (by team members who immediately need to do a task) and which can exist as capabilities (e.g., step by step instructions). In some cases, our immediate aim will be an informal tool or capability, but we will specify a formal tool as a desirable part of the full system that will follow after the present Phase.
Of the 11 possible tools reviewed, 2 were identified as definite requirements within the project, and a further 1 requires further thought. Most of the remainder will be IDL implementations, not formally tools.
The definite requirements are (i) a RegionalAveraging tool, that will support development of global, hemispheric and other time series, and (ii) a Regridding tool, that will generate SST products of lower spatio-temporal resolution than the SST CCI L2P, L3C and L4 products. The RegionalAveraging tool has been fully specified by NR with some inputs from CM and is ready for implementation by NF.
Actions:
NR to send the one remaining auxiliary file to NF.
CM to send NF links to selected ARC netcdf files for tool development and testing (agreed 1 year of ATSR including some overlap period).
Tuesday 20 September 2011
Uncertainty estimates for OSTIA
Matt Martin (Met Office) and me (Chris Merchant, Edinburgh) by telephone
MM updated me on the work in his OSTIA group on improving background error covariance estimates for the analysis of SST. (An "analysis" takes spatially incomplete satellite and/or in situ data, and constructs a spatially complete field of SST by sophisticated extrapolations accounting for various components of uncertainty.) The work for this (Wp209) for SST CCI is just starting.
I explained that delayed algorithm selection exercise is now imminent, but that the details of uncertainty characterisation in the SSTs we will produce depend on the outcome of that algorithm selection: different algorithms inevitably generate different uncertainties.
However, the decomposition of the uncertainties that will be pursued is clear, and is specified in the SST CCI Product Specification: we'll provide uncorrelated uncertainty (mainly arising from sensor noise), synoptically correlated uncertainty (mainly arising from algorithm effects), and correlated uncertainty (mainly arising from satellite calibration and forward modelling bias). There is a matchup dataset from the ARC project that has these uncertainties for the case of ATSRs, and from this I can give MM sufficient general information for his group to proceed. The magnitude of uncorrelated uncertainty is instrument dependent, varies if different channel combinations are used, and for a 0.05 deg cell, varies greatly according to how many full resolution pixels contribute to the cell. Synoptically correlated uncertainty requires a length and timescale for the analysis system; evaluation of these scales is not planned for IR retrievals within SST CCI, so generic synoptic scales will need to be assumed.
For passive microwave SST, we will only know how uncertainties map into these categories from the result of a future SST CCI work package, which is presently delayed due to data unavailability.
Action: by end of this year (2011), CM to provide MM with typical values and ranges of uncertainty components for ATSR and AVHRR (including different algorithm types), and, if possible, a map of how the synoptically correlated uncertainty varies geographically.
MM updated me on the work in his OSTIA group on improving background error covariance estimates for the analysis of SST. (An "analysis" takes spatially incomplete satellite and/or in situ data, and constructs a spatially complete field of SST by sophisticated extrapolations accounting for various components of uncertainty.) The work for this (Wp209) for SST CCI is just starting.
I explained that delayed algorithm selection exercise is now imminent, but that the details of uncertainty characterisation in the SSTs we will produce depend on the outcome of that algorithm selection: different algorithms inevitably generate different uncertainties.
However, the decomposition of the uncertainties that will be pursued is clear, and is specified in the SST CCI Product Specification: we'll provide uncorrelated uncertainty (mainly arising from sensor noise), synoptically correlated uncertainty (mainly arising from algorithm effects), and correlated uncertainty (mainly arising from satellite calibration and forward modelling bias). There is a matchup dataset from the ARC project that has these uncertainties for the case of ATSRs, and from this I can give MM sufficient general information for his group to proceed. The magnitude of uncorrelated uncertainty is instrument dependent, varies if different channel combinations are used, and for a 0.05 deg cell, varies greatly according to how many full resolution pixels contribute to the cell. Synoptically correlated uncertainty requires a length and timescale for the analysis system; evaluation of these scales is not planned for IR retrievals within SST CCI, so generic synoptic scales will need to be assumed.
For passive microwave SST, we will only know how uncertainties map into these categories from the result of a future SST CCI work package, which is presently delayed due to data unavailability.
Action: by end of this year (2011), CM to provide MM with typical values and ranges of uncertainty components for ATSR and AVHRR (including different algorithm types), and, if possible, a map of how the synoptically correlated uncertainty varies geographically.
Subscribe to:
Posts (Atom)