Acceptance test batch 8 #89
Conversation
data/nowcast-optical-flow-from-winds/kgo_15min.ncPrecision level differences, ignorable: Y-direction max fractional change: 1.1917499165292611e-07 data/nowcast-optical-flow/basic/kgo.nc & data/nowcast-optical-flow/remasked/kgo.nc
e.g. Inputs: deriv_xy 0.000394888301671017 0.008808541332823024 float64 data/orographic_enhancement/basic/kgo_hi_res.ncThe original data has the bounds defined in the reverse direction to the direction of the y coordinate. The new kgo gives: This is a result of use of the becomes The change described here SciTools/iris#4466 which allows bounds to be in reverse order and puts in associated checks is probably responsible for the bounds pairs now being reversed appropriately. This change in the bounds actually changes the results around the boundary of the domain, e.g. Some of these changes are substantial (e.g. factor of 2, or infinite when changing to/from 0). I've not attempted to pin down why but I presume some coordinate matching has changed due to the differing bounds. The bounds should be defined as they are in the new result, meaning I believe it was previously wrong. I have tested that this is the source of the differences by modifying the master version of the code as below and running the test with the old environment. With the bounds reordered the output from the old code and environment is identical to the result with the environment upgrade branch of improver and the new environment. The difference we see between the old KGO and the new KGO are plotted below.
data/orographic_enhancement/boundary_height/kgo_hi_res.ncI'm assuming the issues are all the same as above. |
Phase change levelsdata/phase-change-level/without_id_attr/sleet_rain_unfilled_kgo.ncOnly calendar changes The restdata/phase-change-level/with_id_attr/hail_rain_kgo.nc Values change at the call to The returned type is also escalated from float32 to float64 which it was not previously, but this is inserted into a float32 array and so reset upon return from the We can see the structured differences that result from neighbour selection differences, e.g.
|
|
All of the remaining differences are just changes to the calendars. |
bayliffe
changed the base branch from
master
to
environment_upgrade_feature_branch
mspelman07
left a comment
mspelman07
left a comment
There was a problem hiding this comment.
I'm happy that you've addressed and explained all the changes present in these tests
Related to https://github.com/metoppv/mo-blue-team/issues/942
This PR updates the acceptance test data for batch 8. Changes are shown below.
Data comparison