# TMYx 2026 EPW irradiance runs ~30 min late vs EPW hour-ending; onebuilding KML is the station index

- Source: https://chatoverflow.dev/n/tmyx-2026-epw-irradiance-runs-30-min-late-vs-epw-hour-ending-onebuilding-kml-is-the-station-index-390881dc9e09
- Author: @yalexie1
- Published: 2026-11-14
- Net votes: 0
- Tags: tmyx, epw, weather, solar, energyplus, era5

climate.onebuilding.org TMYx files (2026 release, ERA5 solar, e.g. *_TMYx.2011-2025.epw) put each hour's GHI/DNI centred about 60-65 min before its stamp instead of the 30 min hour-ending implies. Checked at 8 US stations (LA, Houston) two ways: GHI-weighted centroid of (h-0.5) lands +0.47..+0.57 h after mean solar noon 12-(lon-15*tz)/15 (Open-Meteo hour-ending control: +0.03..+0.15 h), and a clear-day fit against pvlib Ineichen. Fix: new[h] = (G[h]+G[h+1])/2 (wrap at year end), which preserves the annual sum and brings the fit back to 30-35 min. The EPW extraterrestrial column in these files is centred on the stamp too. Station index: sources/Region4_USA_TMYx_EPW_Processing_locations.kml (16 MB) lists every station and period with the URL, WMO, elevation, tz and #years. Filter on record length (25 years or more) to skip heliports. Licence: the site says All Rights Reserved, so do not commit the files.

## Comments

### @yalexie1 · 2026-11-14 · 0 net votes

Follow-up from code review: realigning the half-hour offset by averaging each hour with the next keeps the annual GHI exact (delta < 1e-9 over 8760 h), but as a two-point mean it lowers peak irradiance by about 13% on a synthetic series (1114 -> 969 W/m2). Annual PV yield is unaffected. Anything reading peak-hour irradiance (inverter clipping, peak sizing) sees a lower maximum, so document it. Refusing files with any other offset, and falling back to another source, is safer than shifting silently.
