Add RRFS-SD reader #154

bbakernoaa · 2023-12-14T20:29:57Z

create a separate reader for RRFS-SD. This is based on _rrfs_cmaq_mm but removes many of the functions as rrfs-sd is obviously much simpler.

bbakernoaa · 2023-12-14T20:31:52Z

this addresses #152

zmoon

Just few initial comments. I'll work on the pressure calculation once I have extracted some sample data.

monetio/models/_rrfs_sd_mm.py

zmoon · 2023-12-14T23:40:43Z

monetio/models/_rrfs_sd_mm.py

+            if "ug/kg" in dset[i].attrs["units"]:
+                # ug/kg -> ug/m3 using dry air density
+                dset[i] = dset[i] * dset["pres_pa_mid"] / dset["temperature_k"] / 287.05535
+                dset[i].attrs["units"] = r"$\mu g m^{-3}$"


If we really want a pretty version here could use unicode, e.g. μg/m³ or μg m⁻³. But maybe better to use just ASCII ug/m3 or ug m-3.

Let’s stick with ascii. It may cause issues down the pipeline if we save the raw files out. It might not be recognized as CF convention otherwise

Then I think ug m-3 best for CF (though they would want kg m-3 for canonical).

zmoon · 2023-12-14T23:45:44Z

monetio/models/_rrfs_sd_mm.py

+
+
+
+    # convert "ug/kg to ug/m3"


Do we want this to be optional like Jordan's?

I dont see why

monetio/models/_rrfs_sd_mm.py

Co-authored-by: Zachary Moon <[email protected]>

monetio/models/_rrfs_sd_mm.py

bbakernoaa · 2023-12-15T00:42:57Z

monetio/models/_rrfs_sd_mm.py

+    phalf(k) = a(k) + surfpres * b(k)
+
+    Mid layer pressures are calculated by:
+    pfull(k) = (phalf(k+1)-phalf(k))/log(phalf(k+1)/phalf(k))


not sure where this formula came from

This is the formula used in the post_fv3 https://github.com/NOAA-EMC/fv3atm/blob/a82381c0b751a15e5343de5078ef836b2c444c89/io/post_fv3.F90#L4266-L4276

@zmoon still not sure where this formula came from

@rschwant do you know? I think that you originally did this

Raffaele sent it to me to describe the vertical structure in the model when I was helping with the boundary conditions. I'll forward you all the email. This is how he described it to me.

Interface pressure levels are computed using the hybrid interface formula:
p(k) = a(k) + ps * b(k)

where ps is the (actual/reference) surface pressure. These pressure levels correspond to phalf in your output dyn*.nc files, while pfull are computed as:
pfull(k) = (phalf(k+1)-phalf(k))/log(phalf(k+1)/phalf(k))

If there is an official typical way of calculating this we can use that instead too. We haven't really tested this much since we have not used the aircraft evaluation in MELDODIES MONET much in the RRFS-CMAQ model.

I'll compare the two methods' results.

There are differences but they are all less than 0.2 hPa in the column I tested.

Differences in Pa:

this is how Raffaele calculated it for UFS-Aerosols in the exglobal_aero_init function https://github.com/NOAA-EMC/global-workflow/blob/develop/ush/merge_fv3_aerosol_tile.py#L99-L101

I think that we should use that

That is calculation of delta p, similar to dpres in the dataset but not the same:

zmoon · 2023-12-15T23:08:00Z

monetio/models/_rrfs_sd_mm.py

+    p = dset.dp_pa.copy().load()  # Have to load into memory here so can assign levels.
+    srfpres = dset.surfpres_pa.copy().load()
+    for k in range(len(dset.z)):
+        if surf_only:
+           pres_2 = 0.0 + srfpres * 0.9978736
+           pres_1 = 0.0 + srfpres * 1.0
+        else:
+           pres_2 = dset.ak[k + 1] + srfpres * dset.bk[k + 1]
+           pres_1 = dset.ak[k] + srfpres * dset.bk[k]
+        p[:, k, :, :] = (pres_2 - pres_1) / np.log(pres_2 / pres_1)


Maybe something like this

Suggested change

p = dset.dp_pa.copy().load() # Have to load into memory here so can assign levels.

srfpres = dset.surfpres_pa.copy().load()

for k in range(len(dset.z)):

if surf_only:

pres_2 = 0.0 + srfpres * 0.9978736

pres_1 = 0.0 + srfpres * 1.0

else:

pres_2 = dset.ak[k + 1] + srfpres * dset.bk[k + 1]

pres_1 = dset.ak[k] + srfpres * dset.bk[k]

p[:, k, :, :] = (pres_2 - pres_1) / np.log(pres_2 / pres_1)

surfpres = dset.surfpres_pa

a = dset.ak

b = dset.bk

if surf_only:

phalf_kp1 = 0.0 + surfpres * 0.9978736

phalf_k = 0.0 + surfpres * 1.0

else:

phalf_kp1 = a.shift(z=-1) + surfpres * b.shift(z=-1)

phalf_k = a + surfpres * b

p = (phalf_kp1 - phalf_k) / np.log(phalf_kp1 / phalf_k)

going for consistency with the docstring variable names

Since ak and bk are dataset attrs, not variables

phalf_kp1 = a[1:] + surfpres * b[1:]

but may need to make them DataArrays for the dims to match up properly and such, maybe create

a_k = xr.DataArray(ds.ak[:-1], dims="z") a_kp1 = xr.DataArray(ds.ak[1:], dims="z") b_k = xr.DataArray(ds.bk[:-1], dims="z") b_kp1 = xr.DataArray(ds.bk[1:], dims="z")

zmoon · 2023-12-19T17:23:30Z

@bbakernoaa I made an extraction (forecast hour 6 of the run you shared, 5 levels closest to surface, selected variables, etc.), now available at https://csl.noaa.gov/groups/csl4/modeldata/melodies-monet/data/example_model_data/rrfssd_example/rrfs-sd_dynf006.nc (104M) for testing.

We can set it up like I did here, which also uses data stored in that location.

bbakernoaa · 2023-12-19T19:23:15Z

@bbakernoaa I made an extraction (forecast hour 6 of the run you shared, 5 levels closest to surface, selected variables, etc.), now available at https://csl.noaa.gov/groups/csl4/modeldata/melodies-monet/data/example_model_data/rrfssd_example/rrfs-sd_dynf006.nc (104M) for testing.

We can set it up like I did here, which also uses data stored in that location.

maybe we can add some netcdf tricks to compress that even further using integers instead of floats and the add_offset and scale_factor netcdf attributes

zmoon · 2023-12-19T20:05:52Z

maybe we can add some netcdf tricks to compress that even further using integers instead of floats and the add_offset and scale_factor netcdf attributes

Perhaps, but I think it is better to keep format closer to the original. And I used NCO lossy compression, which already decreases the size a lot due to quantization, I don't know if additionally doing the int packing transform would make it any more compressible.

zmoon · 2024-01-04T23:18:06Z

monetio/models/_rrfs_sd_mm.py

+        Geoptential height with attributes.
+    """
+    sfc = f.surfalt_m.load()
+    dz = f.dz_m.load() * -1.0


This (loading all dz) is what Maggie's run is failing on, specifically:

Unable to allocate 129. GiB for an array with shape (37, 65, 2961, 4881) and data type float32

Like the pressure calc we should be able to write this in a Dask-friendly way.

Also like pressure calc, for the surf-only case there is a short version.

zmoon · 2024-01-04T23:23:33Z

monetio/models/_rrfs_sd_mm.py

+    # These are negative in RRFS-CMAQ, but you resorted and are adding from the surface,
+    # so make them positive.
+    dz[:, 0, :, :] = dz[:, 0, :, :] + sfc  # Add the surface altitude to the first model level only
+    z = dz.rolling(z=len(f.z), min_periods=1).sum()


I would think .cumsum() could be used.

add rrfs-sd reader

57c466b

bbakernoaa requested a review from zmoon December 14, 2023 20:29

bbakernoaa added the enhancement New feature or request label Dec 14, 2023

bbakernoaa linked an issue Dec 14, 2023 that may be closed by this pull request

RRFS-SD Reader #152

Open

bbakernoaa added 3 commits December 14, 2023 16:23

fix pressure name

d3d2484

Refactor code to improve performance and readability

3530ea1

Fix can_do function docstring

570d0c9

zmoon reviewed Dec 14, 2023

View reviewed changes

monetio/models/_rrfs_sd_mm.py Outdated Show resolved Hide resolved

bbakernoaa and others added 6 commits December 14, 2023 19:05

Update monetio/models/_rrfs_sd_mm.py

a9c3b16

Co-authored-by: Zachary Moon <[email protected]>

Update monetio/models/_rrfs_sd_mm.py

969816a

Co-authored-by: Zachary Moon <[email protected]>

Update monetio/models/_rrfs_sd_mm.py

c9a2f11

Co-authored-by: Zachary Moon <[email protected]>

Update monetio/models/_rrfs_sd_mm.py

af0f173

Co-authored-by: Zachary Moon <[email protected]>

Update monetio/models/_rrfs_sd_mm.py

fce56be

Co-authored-by: Zachary Moon <[email protected]>

include all changes

5cb2c83

bbakernoaa commented Dec 15, 2023

View reviewed changes

monetio/models/_rrfs_sd_mm.py Show resolved Hide resolved

bbakernoaa commented Dec 15, 2023

View reviewed changes

zmoon reviewed Dec 15, 2023

View reviewed changes

zmoon reviewed Jan 4, 2024

View reviewed changes

zmoon changed the title ~~add rrfs-sd reader~~ Add RRFS-SD reader Jan 30, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add RRFS-SD reader #154

Add RRFS-SD reader #154

bbakernoaa commented Dec 14, 2023

bbakernoaa commented Dec 14, 2023

zmoon left a comment

zmoon Dec 14, 2023

bbakernoaa Dec 15, 2023

zmoon Dec 15, 2023

zmoon Dec 14, 2023

bbakernoaa Dec 15, 2023

bbakernoaa Dec 15, 2023

bbakernoaa Dec 15, 2023

bbakernoaa Dec 19, 2023

bbakernoaa Dec 19, 2023

rschwant Dec 19, 2023

zmoon Dec 19, 2023

zmoon Dec 19, 2023

bbakernoaa Dec 19, 2023

bbakernoaa Dec 19, 2023

zmoon Dec 19, 2023

zmoon Dec 15, 2023

zmoon Dec 19, 2023 •

edited

Loading

zmoon commented Dec 19, 2023 •

edited

Loading

bbakernoaa commented Dec 19, 2023

zmoon commented Dec 19, 2023 •

edited

Loading

zmoon Jan 4, 2024

zmoon Jan 4, 2024

zmoon Jan 4, 2024

Add RRFS-SD reader #154

Are you sure you want to change the base?

Add RRFS-SD reader #154

Conversation

bbakernoaa commented Dec 14, 2023

bbakernoaa commented Dec 14, 2023

zmoon left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

zmoon Dec 19, 2023 • edited Loading

Choose a reason for hiding this comment

zmoon commented Dec 19, 2023 • edited Loading

bbakernoaa commented Dec 19, 2023

zmoon commented Dec 19, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

zmoon Dec 19, 2023 •

edited

Loading

zmoon commented Dec 19, 2023 •

edited

Loading

zmoon commented Dec 19, 2023 •

edited

Loading