Feature/convert stat #344
Feature/convert stat #344
Conversation
Codecov Report
@@ Coverage Diff @@
## develop #344 +/- ##
==========================================
Coverage ? 95.99%
==========================================
Files ? 102
Lines ? 4119
Branches ? 0
==========================================
Hits ? 3954
Misses ? 165
Partials ? 0
Continue to review full report at Codecov.
|
|
@clouden90 , we need a few pr's to go in before we can think about merging that work. @travissluka and I will address #339 (#210 is in review) since this is going to break a few of our change of variables ... I think. Your work should be a little easier to fit in after #210 and #339 to then address and hopefully close #340 and #309, possibly enhancing whatever we'll come up with in #339. |
@guillaumevernieres , sure, it sounds a perfect plan to me. I do see #210 before reopening this draft pull request. This draft PR is mainly for tracking purpose as well as exchanging ideas. I will also take a look #210 and other dependencies to see how I am going to fit in my work. |
|
@travissluka , I accidentally removed you from the reviewer's list. Just added you back to keep you in the loop. No need actions now. |
clouden90
force-pushed
the
feature/convert_stat
branch
2 times, most recently
from
f64e03a to
7b34a13
Compare
clouden90
force-pushed
the
feature/convert_stat
branch
from
7b34a13 to
d4d7cea
Compare
clouden90
force-pushed
the
feature/convert_stat
branch
2 times, most recently
from
da54b13 to
8884397
Compare
clouden90
force-pushed
the
feature/convert_stat
branch
from
3cdabb4 to
4710d04
Compare
| ! set hocn for src & des grid | ||
| self%hocn_src = hocn%val | ||
| hocn2%val = des%h | ||
| self%hocn_des = hocn2%val |
There was a problem hiding this comment.
Why do you need hocn2?
There was a problem hiding this comment.
hocn2 is target grid cell thickness. We keep it here in case in the future if we have regional mom6 model setup and one would like to use the converted results to hotstart the model directly.
| ! set lon lat for src & des | ||
| if (field_src%name == "hocn" .and. field_des%name == "hocn" ) then | ||
| return | ||
| end if |
There was a problem hiding this comment.
Sorry, I lost the logic. Why do we return at that point?
There was a problem hiding this comment.
hocn has been set in soca_convertstate_setup so skip here.
|
|
||
| end do | ||
|
|
||
| end subroutine soca_hinterp |
There was a problem hiding this comment.
Note to self: Lost focus here, need to look in more detail at hinterp.
There was a problem hiding this comment.
@clouden90 , I went through most of the code and left a few comments. Please have a look.
Things that needs to be done before you spend more time on that work:
- Contact GFDL, or dig into MOM6, it seems to me that this feature already exists.
- We need two
convert_statetests: fine -> coarse and coarse -> fine - Since we have the exact solution of the 2 tests, we also need to compare the interpolated state to the truth.
That's a lot of work so thanks!
|
@clouden90 , sorry for the travis-ci failures over the last few days. It should work now, please merge develop into your branch and push. If you are working in containers, update your image. It was updated a few days ago. |
guillaumevernieres
added
the
waiting for another PR
| @@ -1,5 +1,5 @@ | |||
| inputVariables: | |||
| variables: &soca_vars [tocn, socn, hocn, cicen, uocn, vocn] | |||
| variables: &soca_vars [ssh, tocn, socn, uocn, vocn, hocn, cicen] | |||
| inputresolution: | |||
There was a problem hiding this comment.
All good besides minor comments. I revived issue #348 which I was suppose to address for your work and I didn't ... But we'll need to revisit in the future since only checking the size of the domain isn't good enough to figure out if 2 geometries are the same ... Subject of an other PR, "soon".
Description
SOCA State Converter
What bug does it fix, or what feature does it add?
This application performs state conversion both horizontally and vertically. A few changes has been made based on the previous comments
.
The major change is now the intermediate grid step is removed. The mom_diag_remap module is used and the re-gridding is done in soca_gridgen step.
Now the application will do conversion based on the following steps:
(0) Pre-processing: run soca_gridgen to convert grid coordinate to z* grid for both source & target grid;
(1) Call soca_convertstate_setup to read pre-computed z* grid and original grid from soca_gridspec nc file for both source & target grid;
(2) Call soca_convertstate_change_resol to convert selected state variables (except target grid cell thickness, which has been defined in step (1)) by following steps:
a. Vertical remapping from source grid to source_z* grid for selected variable;
b. Horizontal remapping from source_z* grid to target_z* grid for selected variable;
c. Vertical remapping from target_z* grid to target grid for selected variable.
Several Issues:
The horiz_interp_and_extrap_tracer() is the subroutine used for horizontal interpolation in MOM6 . The main problem we cannot directly use horiz_interp_and_extrap_tracer() is that it only supports tracer variables (e.g. T, S), but we need conversion of vector variables in our application. Also I found that fill_miss_2d may not work properly in the case that grid cell is surrounded by land mask.
The example of how to use the state converter can be found in test case “test_soca_convertstate” and associated yaml file “convertstate.yml”. This case shows conversion from global domain (5deg) to coarser global grid (generated from test_soca_gridgen_small). I have done several tests to validate the accuracy/performance of the SOCA state converter and the results are well-documented in a separated report.
Issue(s) addressed
closes #309
Testing
How were these changes tested? Updated test_soca_convertstate (Global -> Coarse global grid)
What compilers / HPCs was it tested with? gcc 7.4, Intel 18, Intel 19
Are the changes covered by ctests? Yes
Dependencies
#341 , #339 and possibly #340, or #340 could be addressed here.
meshgrid and soca_fill_miss_2d will be removed after PR: Change meshgrid and fill_miss_2d to public #11
#348 (future work)