Uncaught HIRS outliers in HIRS-IASI matchups in new harmonisation

[gerritholl]

In the January 2019 harmonisation run, there is a small number of HIRS radiances getting through that are bad:

This was not the case in the October 2018 run:

I can't think of anything that's changed in the harmonisation processing for HIRS-IASI matchups (which were always unfiltered), so it must be something else.

It affects 0.03% of the HIRS data, all from a handful of orbits:

In [66]: with printoptions(linewidth=250):
    ...:     print(numpy.unique(ds2["time2"].isel(M=nm2>2e-12).values.astype("M8[h]")))
    ...:
['2013-03-20T06' '2013-03-20T07' '2014-03-26T06' '2014-03-26T07' '2014-03-26T08' '2014-03-26T14' '2014-03-26T15' '2014-04-09T06' '2014-04-09T07']

In the matchup data from October, no data occur at those times:

In [87]: ds1.assign_coords(M=ds1["time2"]).sel(M="2013-03-20T06").dims["M"]
Out[87]: 0

In [88]: ds1.assign_coords(M=ds1["time2"]).sel(M="2013-03-20T07").dims["M"]
Out[88]: 0

In [89]: ds1.assign_coords(M=ds1["time2"]).sel(M="2013-03-20T08").dims["M"]
Out[89]: 0

In [90]: ds1.assign_coords(M=ds1["time2"]).sel(M="2014-03-26T07").dims["M"]
Out[90]: 0

So it looks like there has been a regression...

[jmittaz]

This may have been solved in pull request #386 where extra filtering has been added to typhon