Difference between total precipitation over Greenland from CLM variable (RAIN+SNOW) and CAM (PRECC+PRECL)

[Ziqi Yin]

Hi!

I am working with Adam Herrington on analyzing the output of a fully-coupled CESM2 simulation with an interactive Greenland Ice Sheet (CISM activated). The simulation uses the variable-resolution grid ARCTIC, which features ¼˚ regional refinement over the Arctic. There is a former simulation with the same settings except for using the conventional f09 grid. Currently I am comparing the integrated total precipitation over the Greenland Ice Sheet (GrIS) in these two simulations.

I tried using both the precipitation variables from the land component (RAIN+SNOW) and also from the atmosphere model outputs (PRECC+PRECL). For all these four individual variables, my calculation is the same: GrIS integrated precipitation = sum( precipitation fields * Greenland mask * land fraction * glacier fraction). Then I added either the RAIN and SNOW time series, or the PRECC and PRECL time series together to get the total precipitation. The results are shown in the figure below. The upper subfigure represents results for the f09 grid simulation and the bottom one is for the ARCTIC grid simulation. Since PRECC is really small, the red PRECL line is behind the purple (PRECC+PRECL) line.

I thought, theoretically, the two sums, (RAIN+SNOW) and (PRECC+PRECL), should be the same (of acceptable difference). It is true for the ARCTIC simulation, in which the gray line is behind the purple line. However, for the f09 simulation, the difference of the purple and gray lines is quite large. So far we haven't found a reason for it.

For the f09 simulation, the history files are in /glade/campaign/collections/cmip/CMIP6/timeseries-cmip6/b.e21.B1850G.f09_g17_gl4.CMIP6-1pctCO2to4x-withism.001/ and monthly outputs are used. The Greenland mask is calculated by setting where 'GLACIER_REGION' = 2 in /glade/work/aherring/grids/uniform-res/f09/clm_surfdata_5_0/surfdata_0.9x1.25_hist_78pfts_CMIP5_simyr1850_c200226.nc. to 1 and other grids to 0. Land fraction is achieved from the variable 'landfrac' in any of the clm2.h0 output file. Glacier fraction is index 3 of the variable 'PCT_LANDUNIT' in the clm2.h0 output files (since CISM is activated, the glacier fraction is changing).

So may I ask what could possibly result in this difference? Thanks a lot!

Best,
Ziqi

 

[aherring]

I asked Ziqi to post this to the forums to get a bigger audience. Of note is the the f09 runs are from an older code base (something close to, if not the official CESM2 release tag -- paper is here), whereas the Arctic run is from more recent clm/cam tags. Was there a bug fix in that time period related to this? Is this a FV vs. SE dycore issue, in which only the latter preserves the relationship between ATM and LND models?

 

[oleson]

There was a change made related to these fields:

Old:
- RAIN: "atmospheric rain" - NOT downscaled
- Rainf: "atmospheric rain" - NOT downscaled
- SNOW: "atmospheric snow" - NOT downscaled
- RAIN_REPARTITIONED: "atmospheric rain, after rain/snow repartitioning based on temperature" - DOWNSCALED
- SNOW_REPARTITIONED: "atmospheric snow, after rain/snow repartitioning based on temperature" - DOWNSCALED

New:
- RAIN: "atmospheric rain, after rain/snow repartitioning based on temperature" - DOWNSCALED
- Rainf: "atmospheric rain, after rain/snow repartitioning based on temperature" - DOWNSCALED
- SNOW: "atmospheric snow, after rain/snow repartitioning based on temperature" - DOWNSCALED
- RAIN_FROM_ATM: "atmospheric rain received from atmosphere (pre-repartitioning)" - NOT downscaled
- SNOW_FROM_ATM: "atmospheric snow received from atmosphere (pre-repartitioning)" - NOT downscaled

However, that was some time ago (2017).
I don't know of any differences related to FV versus SE. If SE works, then I would think FV should work.
How do grid averages of RAIN+SNOW and PRECL+PRECC in the FV simulation compare? Is it just the subgrid quantities that differ?
I can look into this further but I will ping @sacks here in case anything occurs to him offhand.

 

[Ziqi Yin]

Thanks Keith!

I think the grid averages of RAIN+SNOW and PRECC+PRECL also differ, not only the subgrid quantities over glaciated regions, since in the above calculation and plot, I used the same mask for both of them.

I tested this by plotting the difference of the global fields of (PRECC + PRECL - RAIN -SNOW) of one random month of the f09 B1850G simulation in the figure below. The difference over Greenland is large enough, while at other places it is three orders of magnitude smaller. I integrated this difference over Greenland Ice Sheet and it is about 180 Gt/yr, like the above time series figure shows (difference between the purple and grey lines).

However, I also checked that this large difference between RAIN+SNOW and PRECC+PRECL over Greenland in this B1850G simulation using f09 grid doesn't exist in F-cases using f09 grid.

 

[sacks]

It took me a while to come up with an explanation for this, but I finally remembered this:

Option for rain-to-snow to immediately run off in some regions by billsacks · Pull Request #586 · ESCOMP/CTSM

Description of changes Up until now: When repartition_rain_snow is .true. (which is the default for CLM5), rain that falls when the near-surface temperature is cold is converted to snow. This repar...

github.com

In some CMIP6 runs (probably including this one), we turned on an option over Greenland that caused any rain-to-snow conversion to runoff immediately rather than becoming snow. This is explained in the comments in Option for rain-to-snow to immediately run off in some regions by billsacks · Pull Request #586 · ESCOMP/CTSM, including the rationale for not bringing this to the master branch.

You could confirm that this is what's going on by adding QRUNOFF_RAIN_TO_SNOW_CONVERSION to your RAIN+SNOW sum. (I think you should also find that RAIN_FROM_ATM+SNOW_FROM_ATM = PRECC+PRECL = RAIN+SNOW+QRUNOFF_RAIN_TO_SNOW_CONVERSION.)

 

[Ziqi Yin]

Hi Bill, thanks! I tested that, for this f09 B1850G simulation, RAIN_FROM_ATM+SNOW_FROM_ATM = PRECC+PRECL = RAIN+SNOW+QRUNOFF_RAIN_TO_SNOW_CONVERSION.