Operational forecasts are not comparable

[JimCircadian]

Not sure I had a ticket for this, but fundamentally the ERA5 and operational forecast data seem to produce vastly different, unacceptable (in the latter case) forecasts. This really shouldn't be the case to the level at which it happens, so tracking the issue herein...

• [x] Fail back to most recent full days from HRES only, until we can validate that there is some ability to use partial days (which is a dataset availability issue based on our usage agreement!)
• [x] Investigate why strd/ssrd accumulated flux fix is not working to make forecast data equivalent to the means in the reanalysis data. This was, if I remember correctly, down to the expression of the units within the file in discrete chunks that needed an equivalent calculation to produce the accumulated values over the lead-up period.
• [ ] Determine the significance of the differences between SIC observed and SIC forecasted in this time period
• [ ] Determine the reason for the coastal SST differences

[JimCircadian]

A really very likely culprit is, surprise surprise, the solar radiation parameters. I did do a rudimentary fix for these at some point but seem to have knackered it in the collection of the source data somehow. Comparison between training (reanalysis/observed SIC) data and operational forecast data highlights this well:

There is also the issue of the SIC forecast provided from the ECMWF model, which is the operational version of the SEAS @tom-andersson (I've CC'd you as you might find this interesting 😄 ) used in assessing the previous IceNet, and the SST differences. I'll add all these to evaluate these difference in the operationally forecasted data into the tasklist for this issue, and for at the moment we can continue running through ERA5 for predictions alongside operational forecasted data.