Mobt 496 enforce forecast consistency

[brhooper]

Acceptance test data for metoppv/improver#1900

Percentile test data copied from #5
Probability test data copied from old enforce_consistent_probabilities acceptance tests

[gavinevans]

Thanks @brhooper 👍

I've plotted the difference between the known good outputs and the input forecast below for the two tests, which seem reasonable.

For percentiles (for sites that have a difference), the known good output is larger than the input for some sites for wind gust, which seems reasonable.

For probabilities, the hybrid total cloud probabilities have been reduced within the known good output, which I think makes sense as the total cloud is acting as a cap.

[gavinevans]

Just as a query, should we consider reducing the number of the percentiles in the percentile forecast tests? Just wondering, if that might help as they're relatively large in terms of file size right now.

[brhooper]

I've reduced the percentile files down to 3 percentiles (rather than the original 13). I've plotted the new difference between the forecast and kgo below. As expected, the kgo is greater than the forecast, where changes have been made.

[brhooper]

An acceptance test has been added which uses realization data. The difference between the kgo and input forecast is shown below. As expected, the kgo is always smaller than the forecast, as it has been enforced to be 90% of the reference forecast (or less).