Uncertainty propagation in SST averaging tools

Nick Rayner (NR) and me (Chris Merchant, CM)

Many SST CCI products will be output at 0.05 deg spatial resolution, daily day-and-night values per sensor ("L3U" data in GHRSST terminology). For various applications there will be a need to average to coarser resolutions. This telecon was about defining a standard approach to this for the project, which may be embodied in a software tool. The key difficulty is propagating the new uncertainty information in SST CCI products appropriately.

NR agreed to further consideration of:
Whether to use simple, uncertainty-weighted, or windsorised means for averaging.
How to define exactly the effective number and location of "synoptically correlated areas" when averaging beyond synoptical space-time scales.

CM agreed to develop an look up table to estimate coverage uncertainty (the uncertainty from not having observations for every cell when averaging up to coarser scales), on a 5 deg basis, for developing regional averages.

NR and CM agreed on options for outputting and describing the components of uncertainty.

Algorithm Selection and Comparison exercise -- time line

As followers of this blog are aware, there is an algorithm selection exercise that is an opportunity for external participation, as well as involving algorithms developed by the project team. This exercise will be done via use of a common "Round Robin Data Package".

We had hoped to have distributed this RRDP at the start of July. Some technical difficulties have led to a delay. We estimate that the RRDP covering year 2010 will be available at the start of September. The other target years for this dataset are 2005-2009 inclusive and 1991-1993. These will follow during September.