This week we collected updates on model development.

FV3

Dan gave the update.

  • 3D Var, 3D EnVar are working
    • NICAS covariance
    • Minimization
    • Increments
    • Low Resolution, Low number of ensemble members
  • Working on getting 4D Var running
    • Tangent Linear/Adjoint (TL/AD) is coded
    • Testing is next
  • Will create a repository with tests soon
    • 3D Var
  • Next step is to get 4D Var increment working
    • Issue with halos, MPI communication

MPAS

BJ gave the update via the following slides.


LFRic

Steve and Marek (UKMO) gave the update.

  • Have been focusing on the interface tests
    • Not all working yet
    • Localization is failing
  • Switching focus to HofX, 3D Var and 3D EnVar
    • HofX working with radiosonde if not running from restart files
    • Barrier is with interpolation function
  • What is status of interpolation?
    • Yannick: Getting close, within days (as opposed to weeks) of getting it fixed
  • How are others dealing with interpolation?
    • MPAS, FV3 have a way to bypass interpolation
    • UKMO should be able to do the same
      • This discussion will be taken offline from this meeting
  • Have other got the localization tests working?
    • FV3 - not yet
    • Probably okay for UKMO to proceed, for now, without this test passing

SOCA

No update since Guillaume was not in attendance.

WRF

Ming gave the update.

  • Just getting started
  • Can run MPAS
    • Using MPAS to figure out flow
  • Building a repository for WRF
  • Working on H(x)
  • Marek: Do you have TL/AD working?
    • Ming: Yes, will take discussion offline from this meeting

NRL

Sergey gave the update.

  • Just getting started
  • Looking at MPAS and QG for guidance
  • Considering two models
    • Semi-Lagrangian
    • Spectral Element
  • The goal is to get a prototype of one of these models working this year

NCEP

Stylianos gave the update.

I am working on integrating:

  1. a model agnostic approach for 2D wave height fields based on grib2 files of prediction and
  2. the WAVEWATCH III® (it's wave model) for having the analysis at the model -4D- state.

Currently, I am focused on the second and, eventually, the two parts will be merged into one "model."

Discussion

  • Yannick mentioned the need for another meeting to discuss the NICAS interface. This interface is supposed to be solidifying, but it keeps changing and thrashing those of us that are using it. The meeting would be held to discuss the roadmap/schedule for when the NICAS interface will be stable.
  • Blocking points
    • Interpolation (background linearization) function

    • Something to do with CRTM, need to meet with Benjamin

      Not sure if I captured this correctly

  • Singularity: should we run inside our outside the container?
    • LFRic, FV3 are running outside Singularity
      • Don't have admin privileges, therefore cannot install Singularity
    • MPAS
      • Running small jobs on desktop inside Singularity
    • Chris Harrop is working on getting UFO to run on Cheyenne
      • Outside container (CISL is not supporting Singularity at this point)
      • Will send a Cheyenne build script to the group once he gets the flow working
      • Has successfully got this flow working outside container on Theia
    • The tricky part when running outside the container is getting all the libraries properly installed
      • Boost and Eigen3 seem to be the more troublesome libraries
  • Containers (everyone has the same, known environment) are good for:
    • Training
    • Running in the cloud
    • Running tests
  • Running container on a production (HPC) machine is problematic
    • Have to convince their IT group to provide Singularity
      • Cannot build a local copy in your home directory
    • Licensing (Intel compilers, for example) can be problematic when running inside a container
    • It would be advantageous for us to find an HPC machine where we could use the container
      • Could run performance tests comparing inside and outside the container
        • Get a handle on the overhead presented by the container
    • Because of these issues, we need to make sure that our systems will run both inside and outside the container
  • We should run tests inside and outside of the container to check overhead
    • JEDI will soon be hiring a new software engineer to focus on testing
  • There is a need for automating the initialization of ensemble forecasts
    • This is a manual process that is too time consuming
    • A solution for this is being discussed, but no solid plans yet

  • No labels