Range checking

[agstephens]

• dachar version: *
• Python version: *
• Operating System: *

Description

When we come to developing a range check then we can learn a lot from:

https://github.com/cp4cds/cmip6_range_check

Martin gave the following description...

defining thresholds is quite problematic. At the moment I'm looking at snow depth. There are 12 models with a maximum in the range 3-5m, 2 with a maximum close 10m, one near 20m (but with an odd distribution), 6 between 35 and 40m, and 5 in the range 1000 to 3100m. My guess is that the 12 in the 3-5 m range are correct, but I can't rule out the two near 10m.

My approach is to extract a set of percentiles (0.1, 0.5, 1., 5., 10., ..... 90, 95, 99, 99.5, 99.9) as well as max and min. Having the percentiles makes it possible to diagnose some problems without going back to the data files. I'm hoping to achieve a degree of automation by checking whether the percentiles are ordered in the sense min(0.5th pct) > max (0.1th pct) etc for the percentiles I've extracted, taking the min and max across the multi-model ensemble.  If variables pass this test, and the multi-model ensemble has a reasonable size, I think it will be reasonable to define a range based automatically.