climaclass

Classical climate classification from monthly climatologies — pure Python, no Earth Engine required.

climaclass turns a 12-month climatology of temperature and precipitation into a climate class under three classical schemes:

It grew out of the Climate_Classifications Earth-Engine notebooks, but reimplements the algorithms as dependency-free Python so they run anywhere — a laptop, a CI job, or inside a hydrological model pipeline — without a Google Earth Engine account.

Install

pip install climaclass                 # core: zero dependencies
pip install "climaclass[symfluence]"   # + SYMFLUENCE attribute-processor plugin

Use

from climaclass import MonthlyClimate, classify

reykjavik = MonthlyClimate(
    temp=[0.1, 0.4, 0.9, 3.3, 6.8, 9.4, 10.9, 10.5, 7.7, 4.5, 1.7, 0.4],   # °C, Jan..Dec
    precip=[76, 72, 82, 58, 44, 50, 52, 62, 67, 86, 73, 79],               # mm,  Jan..Dec
    latitude=64.1,
)

results = classify(reykjavik)
print(results["koppen"].code)        # 'Cfc'  (subpolar oceanic)
print(results["holdridge"].name)     # e.g. 'Boreal moist forest'
print(results["thornthwaite"].code)  # moisture/thermal province code
print(results["koppen"].details)     # MAT, MAP, Pthreshold, ... for QA

Run a single scheme directly:

from climaclass import koppen
koppen.classify(reykjavik).zone      # 16  (Beck et al. legend)

Input

Everything keys off one small record, MonthlyClimate:

• temp — 12 monthly mean temperatures (°C), January → December (required)
• precip — 12 monthly total precipitation (mm), January → December (required)
• latitude — optional; refines Thornthwaite PET day-length and fixes hemisphere
• tmin / tmax — optional, reserved for future scheme variants

SYMFLUENCE plugin

The optional symfluence extra ships an attribute processor that emits climate classes as per-HRU catchment attributes (climate.koppen_code, climate.holdridge_zone, climate.thornthwaite_code, …), registered via the symfluence.attribute_processors entry point and discovered automatically. Use them as a regionalization / PUB grouping variable.

Two data sources, in order of preference:

1. Remapped forcing store (data/model_ready/forcings/*_remapped_*.nc) — the meteorology already areal-weighted to each HRU. This is the preferred path: true HRU resolution, consistent with what the model runs on, no ~1 km ceiling. Temperature/precip are auto-detected (CF standard names) and a 12-month climatology is built per HRU; per-HRU latitude drives Thornthwaite.
2. WorldClim fallback — zonal statistics over the monthly rasters when no forcing store is present.

With a DEM present, per-HRU mean elevation enables the Holdridge altitudinal refinement (climate.holdridge_altitudinal_belt, …_latitudinal_region, …_is_altitudinal). No Earth Engine anywhere.

The integration is fully decoupled: importing climaclass never imports SYMFLUENCE, and the engines (classify_forcing_store, classify_catchment, record_to_attributes) are testable without it.

Maps

The viz extra renders per-HRU classifications onto catchment geometry, with the official Köppen-Geiger colours:

from climaclass.viz import plot_classifications
plot_classifications(catchment_shp, attributes, out_path="climate_maps.png")

pip install "climaclass[symfluence,viz]"

Develop

pip install -e ".[dev]"
pytest -q
ruff check src/

License

Apache License 2.0 — see LICENSE and NOTICE. The classification algorithms were prototyped in the author's Climate_Classifications Earth-Engine notebooks and relicensed by the copyright holder.