LMWG support for Earth System Prediction

[wwieder]

The LWMG would like to support new application of the land model in emerging model configurations. This includes:

• High resolution configuration that are used for numerical weather prediction (NWP), SIMA, EarthWorks, DYAMOND experiments, etc;
• Initialized predictions, like those used by SMYLE & crated by the ESPWG); and
• Impact studies, which may rely on fidelity of land model output.

Following conversations at the ESPAT workshop, this seems like a growth area for NCAR and one that the LWMG would like to support.

Broadly, this discussion is to support simulations which requires creating high resolution surface datasets and/or "non-standard" initial condition. This is Discussion Board is intended to track these efforts as we develop expertise and tools in this area. See also Project Board #31

---

This thread first came up when several of us have been having isolated conversations, so I wanted to start a general discussion here to organize what the LMWG needs to do to help support DYAMOND (3.75km atmosphere & land) and SIMA (high resolution over CONUS for recent NW heat dome and Texas freeze events).

I'm less clear on a timeline for this, although it seems like it's more on the order of weeks than months. For now I just wanted to start a conversation and put this on everyone's radar.

Generally it seems like we need to provide:

1. Surface datasets:

• Present day surface dataset for the (0.25 degree?) land only run used in (2) for generating land initial conditions with ERA5.
• 3.75 km for the DYAMOND run (that may already be available from EarthWorks or @fischer-ncar?)
• Variable resolution for the SIMA runs focusing on CONUS.

2. Land Initial Conditions:

• For now we've discussed generating this from CTSM5.1sp simulations forced with high resolution ERA5 reanalysis.
• This simulation could likely start with initial conditions from a 1 degree CTSM5.1sp-GSWP3 forced run?
• @briandobbins can tell us what start date(s) are needed for the experiments.

@briandobbins does this seem accurate?
@dlawrenncar suggested that showing we can provide these data may ultimately take some pressure off of CGD / CESM.

If so, I'd suggest the following:

• @ekluzek and @slevis-lmwg may be able to help with surface dataset needs? These don't need to be from CTSM5.2, but do the new mksrfdata_esmf tools make things easier?
• @swensosc, can you create datm data from ERA5 files that @adamrher may already have on disk?
• @olyson, after these are done, can you run a year or two of spinup with these ERA5 datm files (at the ERA5 resolution).

[ekluzek]

I'm thinking that the CTSM5.2 branch should be used for making surface datasets. I think you'll run into a ton of problems trying to creating the mapping datasets as well as run mksurfdata_map for these ultra high resolution cases. Yes, @slevis-lmwg and I can help with that if needed.

[fischer-ncar]

I've been able to run I2000Clm50SpRs at 3.75km on derecho with 224 nodes and under subscribing with 14,336 tasks. But I haven't tried running any F compsets. Also, this is with CTSM5.2 and a memory fix from Mariana.

[adamrher]

@fischer-ncar do you have a core hour p simulated year for the i case?

[edit - also, when you undersubscribe, does it tie up the entire node? I'm inferring that your using 14,336/224 = 64 processors per node.]

[fischer-ncar]

@adamrher I've only been running for 3 timesteps. So I don't have any useful core hour p simulated year.

Yes, I'm tying up the entire node.