eeweather-epa is a frozen, unmodified snapshot of opendsm/eeweather master at commit 5dd1959 (2026-10-13), published so that epathermostat can depend on a released build of eeweather's redesigned API. Only packaging metadata differs (see NOTICE). It installs the same eeweather import package, so do not install it alongside eeweather. For anything else, use and report issues against upstream eeweather.

EEweather: Weather for energy-efficiency modeling

EEweather answers "what was the weather here?" — it matches locations to weather stations, fetches observed and typical-year data, and returns analysis-ready frames with warnings and provenance.

Documentation lives in docs/ (publishing to opendsm.energy).

Usage

from datetime import datetime, timezone

import eeweather

location = eeweather.WeatherLocation(34.2, -118.4)   # or .from_place("zcta", "91104")
df, warnings = location.load_data(
    datetime(2024, 1, 1, tzinfo=timezone.utc),
    datetime(2024, 12, 31, tzinfo=timezone.utc),
)
location.provenance     # which station served each variable, and how far away

saved = location.to_json()  # coordinates, sources, and the stations resolved at load
location = eeweather.WeatherLocation.from_json(saved)  # later loads (e.g. a reporting
                                                       # period) reuse the same stations

station = eeweather.WeatherStation("USW00023152")     # or by registry id
station = eeweather.WeatherStation.from_usaf("722880")  # or by historical ids
df, warnings = station.load_data(
    datetime(2024, 1, 1, tzinfo=timezone.utc),
    datetime(2024, 12, 31, tzinfo=timezone.utc),
    variables=("temperature", "relative_humidity"),
)
typical, _ = station.load_data(
    datetime(2024, 1, 1, tzinfo=timezone.utc),
    datetime(2024, 12, 31, tzinfo=timezone.utc),
    source="tmy3",                                    # typical-year data by explicit pin
)

Installation

$ pip install eeweather

Features

• One loading verb over pluggable sources: GHCNh observations (NOAA), NASA POWER gridded meteorological and solar data, TMY3 (NREL) and CZ2010 (CEC) typical years; every load returns an aligned UTC frame plus data-quality warnings and per-source provenance
• A packaged station registry with opaque ids and identifier translation (USAF, WBAN, ICAO), precomputed climate-zone assignments (IECC, Building America, California), ZCTA place codes, and per-source observation inventory
• Station matching: ranked candidates for any point with zone, quality, distance, and availability filters (location.candidates()), plus data-sufficiency selection
• Extension protocols for custom data: station-keyed feeds (e.g. a BigQuery table keyed by any translatable id system) and location-keyed gridded sources
• A shared sqlite weather cache (eeweather.cache.set_path/clear)

Contributing

Dev installation:

$ python -m venv .venv
$ source .venv/bin/activate
$ pip install -e .[dev]

Run tests:

$ pytest

Run tests on multiple python versions:

$ tox

Or with Docker:

$ docker compose run --rm test

Registry updates

The packaged registry (GHCNh station list, observation inventory, quality ratings, identifier aliases) keeps itself current: when the live data is more than six months old, loading data starts a background update into the platform user data directory, and those copies take precedence over the wheel's in new processes. Updates download the ready-made files a scheduled workflow publishes to the repository's rolling release (CDN-served, so any number of clients costs NOAA nothing and no client rebuilds locally) and fall back to rebuilding directly from the live NOAA files when that channel is unreachable or stale, so they keep working even if the repository goes dormant. Concurrent workers on a machine coordinate through an atomic claim file (one attempt per day, however many race), and the update thread waits a minute before touching the network, so short-lived pipeline workers exit without generating traffic. Set EEWEATHER_AUTO_UPDATE=0 to suppress the traffic entirely, or manage it directly with:

$ python -m eeweather.registry.update           # refresh on demand
$ python -m eeweather.registry.update --clear   # revert to packaged data

A scheduled workflow runs the same refresh monthly and opens a pull request so the wheel's snapshot stays current; maintainers can rebuild all packaged data with python -m eeweather.build.

Notice Regarding CZ2010 Data

There may be conditions placed on their international commercial use. They can be used within the U.S. or for non-commercial international activities without restriction. The non-U.S. data cannot be redistributed for commercial purposes. Re-distribution of these data by others must provide this same notification.

See further explanation here.