Skip to content

2024‐08‐15

Jump to bottom
Michael Levy edited this page Aug 15, 2024 · 3 revisions

Aug 15, 2024

Agenda:

  1. Updates
  2. Regridding discussion

Notes

  • Updates:

    • Key metrics notebooks are starting to fill in! We have merged in both Isla's nmse_PSL notebook and Gunter and Kate's LIWG_SMB_diagnostic notebook!
    • As noted in our last meeting, we are hoping to get key metrics in for an initial CESM Project Meeting discussion on August 27th, and we'll have a second discussion on the CESM workflow on September 10th (dates subject to change).
    • We are compiling a list of variables that need regridding-- please add any variables to this list that you'd like to have regridded.
    • We shared an example of running timeseries generation and diagnostic notebooks through CUPiD during the CESM tutorial.

  • Discussion: Regridding

    • Three options for regridding: ADF, x4c, xESMF

      • Pros and cons for each
      • What do we want to use? x4c might be easiest to implement

    • x4c requires previously-generated weight files (but we're mostly concerned with standard CESM grids)

    • Goal: get from native unstructured to any structured grid. CUPiD could take raw ne30 and get it on a 1x1; leave it up to notebooks for further remapping (e.g. mapping to obs grid for comparison)

      • Counterpoint: CISM runs on very high-res (but Cartesian), so maybe doesn't need to be regridded

      • uxarray has been great for plotting CLM fields, but ILAMB needs regular grid; by default land could just remap ILAMB fields

      • Comparing against observation is still a big sticking point! uxarray is good for model vs model, but not model vs obs

      • Trying to pin down use-case, and there are a couple

        1. Need to write some new re-gridded data to disk as part of post-processing (e.g. for ILAMB)
        2. Want to be able to regrid on the fly in notebooks, for model vs obs comparisons

      • For derived variables, we'll let the time series generation handle the derivation but regridding could certainly apply to those new files

    • Regridded data wouldn't need to be archived!

    • cross-component diagnostics: both CAM and CLM need to be on native grid, or regridded to same grid for some diagnostics. Probably makes the most sense to do comparisons on native grid and regrid result!

    • Need to ensure that once data has been regridded, we don't perform operation again (check to see if regridded output is available before calling the tool)

    • Question: is vertical remapping also a concern?

      • Yes, but horizontal grid is static while vertical grids tend to vary in time... so it's better to address it separately
      • ESMPy does handle 3D remapping! Does't support vertical-only (currently just full 3D or horizontal-only), but could be updated if we wanted
      • ESMF_RegridWeightGen supports multiple grid formats (not just mesh files): SCRIP, UGrid and CFtile files. Can mix and match (mesh file for source, something else for destination)

    • ESMF has NASA proposal to support ESMPy better; if funded, would add dask support and could address other CUPiD concerns

  • Next steps:

    1. Move x4c to NCAR github org
    2. Add CUPiD script akin to timeseries.py that regrids fields as specified in config file
    3. Point land folks to regridded CLM output
Clone this wiki locally