Architecture Overview

This overview reflects the current project layout and workflow graph.

Packages

  • Core workflows

    • i7aof.convert — CMIP conversions (thetao/so → ct/sa; ct/sa → TF; back-conversion of annual ct/sa/tf → thetao/so).

    • i7aof.remap — vertical interpolation to z_extrap and horizontal remapping to the ISMIP grid.

    • i7aof.extrap — horizontal and vertical extrapolation using external Fortran executables; includes post-extrap conservative resampling (z_extrap → z).

    • i7aof.biascorr — classic bias-correction of CMIP ct/sa toward observational climatology.

    • i7aof.time — monthly → annual averaging utilities and a CMIP driver for annual means.

  • Shared data and helpers

    • i7aof.grid — ISMIP grid specification and helpers.

    • i7aof.vert — vertical interpolation and conservative resampling utilities.

    • i7aof.io — NetCDF writing and CF time normalization; i7aof.io_zarr — Zarr append/finalize helpers.

    • i7aof.coords — coordinate helpers: attach grid coords, strip _FillValue on non-data, time-bounds handling.

    • i7aof.topo — topography datasets (download/preprocess/remap to ISMIP).

    • i7aof.imbie — IMBIE basin masks (download/generate on demand).

    • i7aof.config — configuration loader/merger (default.cfg + model/climatology/user configs).

    • i7aof.cmip and i7aof.clim — packaged model and climatology configs consumed by i7aof.config.

    • i7aof.download — simple HTTP download helpers with progress.

Data flow (high level)

End-to-end CMIP path (per model; most steps run per scenario):

  1. Split (optional) — cut very large CMIP monthly files into manageable chunks (split).

  2. Convert — thetao/so → ct/sa on native grid (convert).

  3. Remap — vertical to z_extrap, then horizontal to ISMIP grid (remap).

  4. Extrapolate — fill shelves/cavities/ice/bathymetry using Fortran executables; then conservatively resample to z (extrap).

  5. Bias-correct (classic) — run once per model + future scenario with a reference climatology; applies to historical and that future scenario (biascorr).

  6. Thermal forcing — compute TF from bias-corrected CT/SA (convert.ct_sa_to_tf).

  7. Annual means — compute weighted annual means for CT/SA/TF as needed (time).

  8. Optional back-conversion — convert annual ct/sa/tf back to thetao/so (convert.ct_sa_to_thetao_so).

Climatology-only path (no time dimension):

  1. Remap → 2) Extrapolate → 3) TF → 4) Optional back-conversion. The TF and back-convert CLIs have dedicated “clim” entry points.

I/O patterns and reliability:

  • Zarr-first for chunked operations (append by time), then finalize to a single NetCDF with atomic rename; internal ready markers enable idempotent reruns.

  • NetCDF writes use consistent CF time encoding (numeric days since 1850, declared calendar), per-variable _FillValue and optional compression, and unlimited time when present.

See package pages for developer details and the User Guide for end-to-end examples.