pyepwmorph

A Python package for morphing EnergyPlus Weather (EPW) files with climate model data. Supports CMIP6 projections from Google Cloud and custom CSV-based model data for both future and historical scenarios.

Overview

pyepwmorph enables building performance analysts and researchers to create morphed weather files by applying scientifically validated procedures (Belcher et al. 2005, Jentsch et al. 2013) to existing EPW files. The package can morph EPWs using:

• CMIP6 data fetched automatically from Google Cloud (Pangeo)
• Custom CSV data from any climate model, including historical reconstructions
• CH2025 station scenarios for Switzerland, shipped with the package and indexed by global warming level

Installation

With uv (recommended)

uv add pyepwmorph

Development setup

git clone https://github.com/justinfmccarty/pyepwmorph.git
cd pyepwmorph
uv sync --extra dev

With pip

pip install pyepwmorph

Quick start

CMIP6 workflow (future projections)

import pyepwmorph.tools.workflow as workflow

results = workflow.morphing_workflow(
    project_name="MyBuilding_Future",
    epw_file="weather.epw",
    user_variables=["Temperature", "Humidity", "Clouds and Radiation"],
    user_pathways=["Middle of the Road"],   # ssp245
    percentiles=[50],
    target_years=[2050],
    output_directory="output/",
)

Custom CSV workflow (historical or any model)

import pyepwmorph.tools.workflow as workflow

custom_data = {
    "reference": {
        "tas": "data/reference_tas.csv",
        "tasmax": "data/reference_tasmax.csv",
        "tasmin": "data/reference_tasmin.csv",
    },
    "target_1990s": {
        "tas": "data/target_tas.csv",
        "tasmax": "data/target_tasmax.csv",
        "tasmin": "data/target_tasmin.csv",
    },
}

results = workflow.morphing_workflow(
    project_name="Historical_Morph",
    epw_file="weather.epw",
    user_variables=["Temperature"],
    user_pathways=["target_1990s"],
    percentiles=[50],
    target_years=[1990],
    output_directory="output/",
    data_source="custom",
    custom_data=custom_data,
    reference_scenario="reference",
)

Custom CSVs should have a date column (parseable by pandas) and a column named after the CMIP6 variable (e.g. tas, tasmax). Rows should be monthly.

The reference and target scenarios must cover different years, the same way the CMIP6 historical and sspXXX experiments do. The two series are concatenated before the baseline and target periods are sliced out, so overlapping years get averaged together and weaken the climate signal. Keep baseline_range inside the years the reference scenario covers.

Switzerland (CH2025 warming levels)

For an EPW inside Switzerland, data_source="ch2025" morphs from MeteoSwiss CH2025 station scenarios. There is no target year: each pathway is a global warming level relative to 1991-2020, and the run is offline.

results = workflow.morphing_workflow(
    project_name="Zurich_GWL2",
    epw_file="zurich.epw",
    user_variables=["Temperature", "Humidity", "Wind", "Radiation", "Dew Point"],
    user_pathways=["GWL 2.0"],
    percentiles=[50],
    output_directory="output/",
    data_source="ch2025",
)
morphed = results["gwl2.0"]["50"]

The site is matched to the nearest station that has the requested variables (or, with ch2025_full_coverage=True, the nearest station that has every CH2025 variable), with an elevation penalty so a much higher station is not chosen just because it is close on the map. Supported variables are temperature, humidity, dew point, wind, and radiation (global, diffuse, and direct). Pressure and cloud cover are not in CH2025; requesting them raises an error. Sky cover is left at the EPW's baseline values when radiation is morphed. Horizontal infrared follows the morphed temperature and humidity (see "Variable dependencies"), but not cloud changes.

The signal is the change from the 1991-2020 reference climate to the chosen warming level. Each model chain's change is computed first, over the chains that reach that warming level, and the percentiles are taken of those changes. The EPW's period is read from the Period of Record in its header. If that is not exactly 1991-2020, a UserWarning recommends a TMY built from 1991-2020 data. If the header states no period, a second warning says the period could not be detected, and the mismatch warning still fires if the data rows' years fall outside 1991-2020. The signal is applied to the file as it stands either way. tools.io.write_period_of_record adds the statement to files built without one. The same caveats are listed in MorphConfig.ch2025_notes and the matched station in MorphConfig.ch2025_station.

CH2025 data: MeteoSwiss & ETH Zurich (2025), Climate CH2025 - Daily Datasets, CC-BY 4.0, https://doi.org/10.18751/climate/scenarios/ch2025/data/1.0/

Measured data (actual meteorological year)

Moved to weather-file-builder (2.1 and later), together with ISO 15927-4 typical years from station data: weather_file_builder.amy.station_table_to_epw and weather_file_builder.station_tmy.station_table_to_tmy_epw. pyepwmorph.tools.amy still works in 3.4 but warns on import and will be removed in 4.0.

Climate scenarios

Morphing variables

• Temperature -- dry bulb temperature (shift + stretch)
• Humidity -- relative humidity, stretched in specific humidity space
• Pressure -- atmospheric pressure (shift)
• Wind -- wind speed (stretch)
• Radiation -- global/diffuse/direct radiation, without changing sky cover (all sources; the only radiation option for CH2025)
• Clouds and Radiation -- global/diffuse/direct radiation and sky cover (CMIP6 and custom data). Morphed sky cover keeps two decimals (EnergyPlus reads it as a real number), so small monthly changes are not lost to rounding
• Dew Point -- recalculated from morphed temperature and humidity

Variable dependencies

Some variables cannot be morphed on their own:

• Humidity requires Temperature and Pressure, except with CH2025, where relative humidity is stretched directly
• Dew Point requires Temperature, Humidity, and Pressure (Temperature and Humidity only with CH2025)

Dependencies are added automatically and are written to the output file. Asking for Dew Point alone therefore returns an EPW with morphed pressure, temperature, relative humidity, and dew point, which keeps the file internally consistent. MorphConfig.resolved_variables shows exactly what will be written, in the order it is computed.

Some fields are recomputed from the morphed ones without climate data of their own (MorphConfig.recomputed_variables):

• Dew point whenever Temperature or Humidity is morphed. After a temperature-only morph it comes from the morphed dry-bulb and the file's own relative humidity, so relative humidity is held as it was.
• Horizontal infrared whenever Temperature, Humidity, Dew Point or Clouds and Radiation is morphed. The file's own value is scaled by the change in eps_sky * sigma * T^4 from the Clark and Allen sky model (EnergyPlus's default), computed from dry-bulb, dew point and opaque sky cover. Missing values (9999) stay missing. With CH2025, which has no cloud data, it follows temperature and humidity only.

Header blocks

A morphed file's header is brought in line with its new climate:

• GROUND TEMPERATURES shift with the change in monthly mean dry-bulb. The annual-mean change passes through in full; the seasonal part is damped and lagged at each depth the way the source header's own ground temperatures are against the source air. A file without ground temperatures gets them at 0.5, 2 and 4 m from the morphed air (Kusuda and Achenbach).
• DESIGN CONDITIONS are replaced by DESIGN CONDITIONS,0, because the source's ASHRAE values describe the baseline climate. Size equipment from the morphed hours, not from the header.

Both are noted in COMMENTS 2. CMIP6 morphs also name the grid cell each model's data came from and its distance from the site.

Caching

Climate model data is cached locally after the first download, and the cache is consulted before the remote catalogue is opened so a hit costs no network traffic.

import pyepwmorph.models.access as access

stats = access.get_cmip6_cache_stats()
access.clear_cmip6_cache()

Cache entries live in the per-user cache directory (~/Library/Caches/pyepwmorph on macOS, ~/.cache/pyepwmorph on Linux, %LOCALAPPDATA%\pyepwmorph on Windows) and are keyed by location, pathway, variable, model sources, and time slices. Two environment variables override the defaults:

Available climate models

from pyepwmorph.tools.utilities import available_models

models = available_models()

Development

git clone https://github.com/justinfmccarty/pyepwmorph.git
cd pyepwmorph
uv sync --extra dev

# Run tests (fully offline)
uv run pytest

# Run with coverage
uv run pytest --cov=pyepwmorph

# Lint
uv run ruff check pyepwmorph tests gui

Releases

./release.sh [patch|minor|major]

The script refuses to run on a dirty tree, off main, with failing lint or tests, or without a matching CHANGELOG.md section. It bumps the version, tags, and pushes; creating the GitHub Release then triggers the PyPI publish workflow.

gh release create v3.0.0 \
  --title "v3.0.0" \
  --notes "See CHANGELOG.md."

Changes

See CHANGELOG.md for the full history. The most recent release corrects several morphing calculations, so morphed humidity, dew point, wind speed, cloud cover, and direct/diffuse radiation all differ from files produced by earlier versions.

Requirements

• Python >= 3.9
• pandas >= 2.2
• Internet connection (for CMIP6 data download; the custom CSV and CH2025 workflows run offline)

License

MIT License. See LICENSE.

Citation

McCarty, J. (2026). pyepwmorph: A Python package for climate-informed
EPW file morphing. Version 3.0.0.
https://github.com/justinfmccarty/pyepwmorph