Future weather files explained: how morphing turns today's weather into a future climate

What is a future weather file?

Building simulation needs hourly weather for a specific place, but climate models work on coarse grids, and their projections are usually supplied as changes in monthly averages. A future weather file bridges the two. It takes an observed hourly year, usually a typical year in EPW format, and adjusts it so that each month's statistics match a projection for a future period, while keeping the day-to-day and hour-to-hour pattern of real weather.

Belcher, Hacker and Powell described this approach in 2005 and called it morphing. It is one of several ways of downscaling climate projections, and it underpins widely used data sets: CIBSE has published future weather years for the UK made by morphing since 2009, and its 2025 release morphs a 1994–2023 baseline to the UKCP18 projections using the revised algorithms of Eames et al.

How does morphing work?

Each operation acts on one month of hourly data, using a change factor: the projected change in that month's average relative to the baseline period. Belcher et al. defined three operations:

• Shift: add the projected change to every hour. The mean moves and the spread stays the same; it is used, for example, for atmospheric pressure.
• Stretch: multiply every hour by the fractional change. It is used for solar radiation, so that night-time values stay at zero and the shape of each day is kept.
• Shift and stretch: move the mean and scale the variation around it. It is used for dry-bulb temperature, with the stretch set by the projected change in the daily temperature range (daily maximum minus daily minimum).

Belcher (2005) vs Eames et al. (2024): what changed?

In an open-access paper in Building Services Engineering Research and Technology, Eames, Xie, Mylona, Shilston and Hacker identified two weaknesses in the original algorithms. For temperature, the shift and stretch keeps the projected change in the mean and in the average daily range, but not the projected changes in the average daily maximum and daily minimum individually. And unbounded operations can push variables past physical limits: cloud cover cannot go below clear sky or above full cover, and morphed solar radiation could produce peaks inconsistent with the sky.

Their revised algorithms, a weighted stretch for variables with one change factor and a bounded temperature weighted stretch for dry-bulb temperature, add conditions on the maximum and minimum as well as the mean. They keep clear and fully overcast hours unchanged while morphing partly cloudy ones, and they preserve all three temperature change factors. In the authors' building simulations, the revised method gave lower heating energy on average, and lower cooling energy and peak loads, for the same projection, which would lead to different design decisions.

The algorithms were first developed at Arup for CIBSE TM49 and used in the 2016 CIBSE weather years and in Arup's WeatherShift tool; the 2024 paper sets out their mathematical proof.

Where do the climate inputs come from?

Change factors come from climate model projections, for example from CMIP6, the sixth phase of the Coupled Model Intercomparison Project, whose models underpin the IPCC's Sixth Assessment Report. They are run under Shared Socioeconomic Pathway (SSP) scenarios. National projections, such as the UK's UKCP18, are also used.

Raw global model output is coarse, around 1°, so data services downscale and bias-correct it. The World Bank's Climate Change Knowledge Portal, for example, regrids up to 30 CMIP6 models to a common 1° grid, then bias-corrects and downscales them to 0.25° against ERA5 with a quantile-mapping method (BCSD). It publishes the multi-model median with 10th and 90th percentiles for SSP1-1.9, SSP1-2.6, SSP2-4.5, SSP3-7.0 and SSP5-8.5 up to 2100, and notes that the downscaled data inherit fine-scale spatial patterns from the reference period rather than from the models themselves.

What can a future weather file tell you, and what can it not?

• It can show how the same design performs as the climate shifts, for example lower heating demand, more cooling or more overheating hours, and passive strategies that gain or lose hours.
• It keeps the baseline's weather sequences. Warm spells become warmer versions of the baseline's warm spells, not longer or more frequent ones. Eames et al. also note that the transformation raises each day's maximum by the same amount, whereas the hottest temperatures are not expected to rise in line with the average daily maximum.
• It inherits the baseline's purpose. A morphed typical year is still a typical year, so overheating checks need a hot or extreme baseline year as well.
• It carries the uncertainty of the scenario and the models. Scenarios have no likelihood attached, so test a design against more than one, and look at the model range as well as the median.

How C4B makes future weather files

In projects on paid plans, C4B morphs the site-specific weather file to any year from 2025 to 2100 under five SSP scenarios, using the World Bank Climate Change Knowledge Portal's CMIP6 multi-model median for temperature, humidity and solar radiation and, mainly, the Eames et al. (2024) method. When a search completes it prepares four scenarios for the building's default mid-life year, 27 years ahead, and overlays present and future results on temperature, UTCI outdoor comfort, the wind rose, cloud cover, PV output and the energy balance.

Downloading future EPW files is available on the Technical and Enterprise plans.