issue_comments: 863955559

Yes, they are generated on a .25x.25 lat lon grid in europe, so these values match (when reading the original files there is no nan, which I think excludes this option)

The test is all q values are the same is not meant for the case where I even find nan, but where I don't see them. I should have included the output I get - see below e.q. for the last test I ran.

It say that both original and read back in are F32 - that's what confuses me. I also expected to see a difference in data type to be responsible, but at first glance here it does not seem to be the case.

Below that output I print a timespan of the original and the second dataset, where the values clearly differ - in the last few digits. I can also include the test, where it even returns nan at some places. The full testing code and data is in the link if you want to see that - or I can post it here. ``` original <xarray.Dataset> Dimensions: (level: 26, time: 1464) Coordinates: longitude float32 10.0 latitude float32 38.0 * level (level) int32 112 113 114 115 116 117 ... 132 133 134 135 136 137 * time (time) datetime64[ns] 2014-09-01 ... 2014-10-31T23:00:00 Data variables: t (time, level) float32 dask.array<chunksize=(720, 26), meta=np.ndarray> q (time, level) float32 dask.array<chunksize=(720, 26), meta=np.ndarray> u (time, level) float32 dask.array<chunksize=(720, 26), meta=np.ndarray> v (time, level) float32 dask.array<chunksize=(720, 26), meta=np.ndarray> Attributes: Conventions: CF-1.6 history: 2021-02-05 01:00:40 GMT by grib_to_netcdf-2.16.0: /opt/ecmw...

read back in <xarray.Dataset> Dimensions: (level: 26, time: 1464) Coordinates: longitude float32 ... latitude float32 ... * level (level) int32 112 113 114 115 116 117 ... 132 133 134 135 136 137 * time (time) datetime64[ns] 2014-09-01 ... 2014-10-31T23:00:00 Data variables: t (time, level) float32 ... q (time, level) float32 ... u (time, level) float32 ... v (time, level) float32 ... Attributes: Conventions: CF-1.6 history: 2021-02-05 01:00:40 GMT by grib_to_netcdf-2.16.0: /opt/ecmw...

test for nan - np.any(np.isnan(array)) original q: False t: False u: False v: False read back in q: False t: False u: False v: False

look at one of the problematic portions: (q.values[timespan] - values for same timespan original and read back in) original [0.01286593 0.01290165 0.01218289 0.01229404 0.01238789 0.0125237 0.01275251 0.01274316 0.01292717 0.01308822 0.01309219 0.01304683 0.01299834 0.01299749 0.01267057 0.01274089 0.01281064 0.01282141 0.01286848 0.01291271 0.01302868 0.01290676 0.01276612 0.01273976 0.01273635 0.01271169 0.01244998 0.01250867 0.01229999 0.01256708 0.01265356 0.01276471 0.01274259 0.01243155 0.01195124 0.01166572 0.01124779 0.01097304 0.01091747 0.01098779 0.01105896 0.01114317 0.01122823 0.01133569 0.01147207 0.01155231 0.01154834 0.01154579 0.01155486 0.01158009 0.0114715 0.01169464 0.01170598 0.01151034 0.01124751 0.01127246 0.01125374 0.01128862 0.01127643 0.0112631 0.01126225 0.01126594 0.01154182 0.01162574 0.01169833 0.01176354 0.01183301 0.01184066 0.01187781 0.01194756 0.01208564 0.01224102 0.01244346 0.01260706 0.01236549 0.01256538 0.0127528 0.01287415 0.01304286 0.01327876 0.01366919 0.01396406 0.0142683 0.01445004 0.01449626 0.01438228 0.01404204 0.01419486 0.01447329 0.01472309 0.01493943 0.01512514 0.01532986 0.01552691 0.01566074 0.01577302 0.01581669 0.015832 0.01564515 0.01568768] read back in [0.01286582 0.01290182 0.01218301 0.01229396 0.01238785 0.01252367 0.01275264 0.01274299 0.01292705 0.01308811 0.01309219 0.01304692 0.0129983 0.01299756 0.01267063 0.01274076 0.01281053 0.01282129 0.01286842 0.01291258 0.01302873 0.01290664 0.012766 0.01273965 0.01273631 0.01271182 0.01244982 0.01250883 0.0122999 0.01256709 0.01265355 0.01276488 0.01274262 0.01243164 0.01195107 0.0116657 0.01124785 0.01097287 0.01091757 0.01098771 0.01105896 0.0111432 0.01122818 0.0113358 0.01147199 0.01155215 0.01154844 0.01154584 0.01155474 0.01157998 0.01147162 0.01169465 0.01170615 0.01151021 0.01124748 0.01127234 0.01125379 0.01128867 0.01127642 0.01126306 0.01126232 0.01126603 0.01154175 0.01162562 0.01169836 0.01176367 0.01183306 0.01184049 0.01187797 0.01194773 0.01208578 0.0122409 0.01244352 0.01260717 0.01236559 0.01256523 0.01275264 0.01287398 0.01304283 0.01327885 0.01366924 0.01396389 0.01426819 0.01445002 0.01449641 0.01438211 0.01404219 0.01419471 0.0144734 0.01472315 0.0149395 0.01512505 0.01532989 0.01552694 0.01566091 0.01577298 0.01581677 0.01583198 0.01564532 0.01568763] timespan: 2014-10-04T08:00:00.000000000 2014-10-08T11:00:00.000000000 Test all q values same: False

```