"Real time" forecasting, integration and dissemination

[JimCircadian]

This issue is being extended / adapted to capture a whole set of developments relating to this activity

This issue will capture ongoing tasks following implementation to assess the predictive capability and additional developments that arise using forecast data over reanalysis data.

• Ensure dataset differences don't result in land mask artifacts in SIC estimates
• Validate forecasts from the forecast data against ERA5 produced SIC predictions
  • Finish developments relating to solar variables from forecast feeds

[JimCircadian]

Validate forecasts from the HRES data against ERA5 produced SIC predictions

Comments relating to this task. The below discussion point from Tom provides a nice foundation to multi workflow testing in the pipeline, so quite useful

So we get the two forecast datasets in that case. We also kind of have two ground truths - ERA5 SIC and RT SIC. So we could

1. measure the change in performance relative to ERA5 SIC when we switch to RT (not a huge fan of that),
2. measure the difference between the two forecast datasets to quantify the perturbation from using RT data,
3. compare the RT-initialised performance on RT data with the ERA5-initialised performance on ERA5 data.

From 2, we hope the perturbation is not too large. From 3, we hope the performance is very similar.

If both those are true we know we are pretty safe running the model with RT data.

But an additional control variable in this experiment is that we can train the model on RT data (as you're saying). So there's actually a third potential forecast dataset to analyse: RT-trained-and-initialised.

So it would be interesting to repeat 2) and 3) but with the third forecast dataset.

[JimCircadian]

Ensure dataset differences don't result in land mask artifacts in SIC estimates

One comment to factor into this:

And on the normalisation/climatology stuff, I would expect the normalisation parameters to be pretty similar between ERA5 and RT data, so that shouldn't be a problem. I am slightly concerned about the climatology though, because slight differences in the land mask or something might lead to weird artefacts. It would be worth plotting and comparing the anomaly variables between ERA5 and RT.