Skip to content

Would a longer preprocessed quickstart bundle be possible? #2

@sudshu

Description

@sudshu

Hi! Quick note — got data-quickstart-v2 (LL 5° and 2.5°) running end-to-end on GPU in a single afternoon (Julia 1.11, CUDA.jl on an NVIDIA A10G). Docs were very clean; clone, instantiate, run actually worked first try.

I'm now experimenting with a surface-only point source from LA (small local patch adding an n_layers knob to gaussian_blob so the IC sits in the lowest model layer only), and I'd like to let it run for a winter month or two so the plume can spread out and reach a quasi-stationary continental footprint.

The path there is non-trivial as far as I can tell: the native ERA5 download is ~1.5 TB/month, and the 2.5° preprocessor reads spectral input rather than a pre-coarsened grid, so requesting 2.5° directly from CDS doesn't shortcut it.

Would you consider publishing a longer preprocessed bundle (e.g. quickstart_ll_dec2021_to_feb2022_v1.tar.gz, even just one full month at 2.5°) under the same release scheme as data-quickstart-v2? It would open up a lot of second-day experiments without putting every new user through the full CDS + preprocess loop.

Happy to help test or stage anything if useful.

Thanks!
— Sudhanshu

(Setup: Julia 1.11.5, AtmosTransport main @ 5ab951c, running ll144x73_advonly with a surface-only LA point source, single A10G.)

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions