wrf2wps

Drive a fine-resolution WRF domain from a coarse WRF run — in pure Python, with no nest-ratio constraint.

wrf2wps converts a coarse WRF-ARW history (wrfout) into WPS intermediate files, which metgrid.exe → real.exe then turn into wrfinput / wrfbdy / wrffdda for a finer domain.

It is an alternative to ndown that carries no nest-ratio constraint. Because metgrid is a free horizontal interpolation (not a parent/child nest), the inner grid dimensions are unconstrained — you are not bound by WRF's (e-1) % parent_grid_ratio == 0 rule, and you can keep an existing, byte-frozen inner domain while changing the coarse driver.

coarse WRF run ──► wrfout (eta levels, Mercator)
                      │   wrf2wps
                      ▼
                 WPS intermediate (regular lat/lon, pressure levels)
                      │   metgrid.exe ─ real.exe
                      ▼
                 wrfinput / wrfbdy / wrffdda  ──►  fine WRF run

• No nest-ratio constraint. Nesting and ndown both enforce integer grid ratios and aligned dimensions. Going through WPS intermediate sidesteps that entirely — useful when your fine domain dimensions can't be divided by a clean ratio, or when the coarse:fine jump is too large for a single nest.
• No Fortran, no wrf-python. Diagnostics (temperature from θ, geopotential height, sea-level pressure, wind destagger, η→pressure interpolation) are computed with numpy. Runs on minimal/old cluster Pythons where building wrf-python is painful.
• Battle-tested format core. The intermediate read/write is vendored from the author's cmip6-to-wrfinterm and has driven multi-decade regional downscaling runs.

pip install wrf2wps        # or: pip install -e .  (from a clone)

Requires Python ≥ 3.8 and numpy, scipy, pandas, xarray, netCDF4.

cp examples/config.example.ini config.ini
# edit wrf_root / out_root / the time window
wrf2wps config.ini

Then point metgrid at the output by setting, in namelist.wps:

&metgrid
 fg_name = 'WRFITM',
/

prepare your inner-domain geo_em, run metgrid.exe and real.exe as usual.

A single [w2w] section (see examples/config.example.ini):

deltlat, deltlon, and frq must be positive, and etl_end_ts must not precede etl_strt_ts. If a multi-time wrfout is opened and its Times metadata does not contain the requested timestamp, wrf2wps stops instead of silently using the first record.

Two are bundled:

• wrf-arw (default) — native Noah 4-layer soil (ST000010/ST010040/ST040100/ST100200 + moisture).
• wrf-arw-2soil — a 10–200 cm aggregate (ST000010/ST010200), for a metgrid/real path tuned to a 2-layer soil source (e.g. some CMIP6 setups).

Each row maps an internally computed field (calc) to a WPS field name (aim_v) with its units, type (3d/2d), and desc. Copy one and edit to emit a different field set.

• Mercator / rectilinear sources only. The horizontal regrid is a fast separable 1-D interpolation, which is exact when wrfout rows are constant latitude and columns constant longitude. Lambert / polar-stereographic sources need a true 2-D regrid and wind rotation to earth-relative — not yet implemented; a warning is logged if a non-rectilinear grid is detected. (Tracked for v0.2.)
• Noah 4-layer soil is assumed (TSLB/SMOIS with 4 levels).
• Winds are written IS_WIND_EARTH_REL = 1, valid for Mercator (unrotated); revisit when 2-D projections land.

pip install -e ".[test]"
pytest

Tests cover the format round-trip and an end-to-end conversion on a synthetic in-memory wrfout (no large data files committed).

Author: Zhenning Li (zhenningli91@gmail.com). The WPS-intermediate format core derives from cmip6-to-wrfinterm. If wrf2wps supports your research, please cite it (see CITATION.cff).

Licensed under the MIT License.