geocat-rs

Rust-accelerated drop-in replacements for GeoCAT-comp computational functions.

GeoCAT-comp is NCAR's Python replacement for NCL. geocat-rs accelerates the compute-heavy functions using Rust + rayon parallelism, while leaving the xarray/scipy wrapper functions alone (they're already fast).

Performance

Real-world benchmarks on CESM-like 3D atmospheric fields. Every result verified identical to geocat.comp.

The #1 workflow: interp_hybrid_to_pressure

The function CESM researchers use most — converting hybrid-sigma vertical coordinates to standard pressure levels (replacing NCL's vinth2p).

Tested with 14 drop-in compatibility tests: standard/full CESM grids, custom levels, multiple timesteps, mountain pressure extremes, 4 random seeds. All pass.

Meteorology functions on 3D fields (12.5M elements)

WGS84 gradient functions (250K grid points)

Overall across all accelerated functions

Accuracy

All functions verified against geocat.comp to machine precision:

pytest tests/ -v
python tests/bench_interp_hybrid.py
python tests/bench_workflow.py

Installation

pip install geocat-rs

Requires metrust for interp_hybrid_to_pressure (uses its Rust interpolation engine).

Usage

interp_hybrid_to_pressure (drop-in replacement)

# Before
from geocat.comp import interp_hybrid_to_pressure

# After (same API, 7.5x faster)
from geocat_rs.interp_hybrid import interp_hybrid_to_pressure

# Usage is identical
output = interp_hybrid_to_pressure(
    data, ps, hyam, hybm,
    new_levels=new_levels,
    lev_dim='lev',
)

Meteorology (array functions via Rust + rayon)

from geocat_rs._geocat_rs import meteorology as met

# Scalar
td = met.dewtemp_scalar(300.0, 50.0)  # T(K), RH(%) -> Td(K)
hi = met.heat_index_scalar(95.0, 50.0, False)  # T(F), RH(%) -> HI(F)
rh = met.relhum_scalar(300.0, 0.01, 101325.0)  # T(K), w(kg/kg), P(Pa) -> RH(%)

# Array (rayon-parallel)
import numpy as np
td_array = met.dewtemp_array(temp_k, rh_pct)
rh_array = met.relhum_array(temp_k, mixr, pressure)
svp_array = met.saturation_vapor_pressure_array(temp_f)

WGS84 gradient (66-70x faster)

from geocat_rs._geocat_rs import gradient as grad

radius = grad.rad_lat_wgs84_array(lat_grid.ravel())  # meters
arc = grad.arc_lat_wgs84_array(lat_grid.ravel())      # meters from equator

What's accelerated

What's NOT accelerated (and why)

License

Apache-2.0 (same as GeoCAT-comp)