Experiment record

Open-Meteo hourly solar-radiation timestamps: preceding-hour means and PV time-series alignment

Open-Meteo hourly non-instant solar-radiation values timestamped T are preceding-hour means, while the instant variants represent the indicated time. This Experiment verifies the distinction and shows how to align those fields with measured PV time series without applying a blanket one-hour shift.

Summary

An hourly API timestamp does not necessarily mean “the value at that instant.”

For the current Open-Meteo hourly solar-radiation fields examined here, non-instant variables such as shortwave_radiation and global_tilted_irradiance are means of the preceding hour. A value timestamped 14:00 therefore belongs to 13:00–14:00. It does not describe 14:00–15:00, and it is not simply the instantaneous radiation at 14:00.

Open-Meteo also provides separate *_instant radiation variables for the indicated time. This distinction matters when radiation is compared with photovoltaic generation or other sub-daily measurements: treating the non-instant value at T as the following hour can create an apparent time offset even when the API is behaving as documented.

A private residential PV series was used only as a secondary consistency check. Across 104 daylight hourly comparisons, the documented T-ending alignment had the highest tested correlation, but the following-hour alternative was nearly equal. The field series therefore does not independently prove the API semantics. First-party Open-Meteo documentation is the decisive Evidence.

Background

The investigation began with an apparent timing mismatch between hourly Open-Meteo solar radiation and measured PV generation.

The retained hourly weather history contained distinct averaged GHI, averaged GTI, instant GHI, and instant GTI fields. That made the central question more precise than “How many hours should this series be shifted?”:

What time or interval does the timestamp on each radiation variable actually represent?

This question is transferable beyond the private PV system that exposed it. Any user joining Open-Meteo radiation to measured power, energy, irradiance, or chart data must decide whether a timestamp labels a point or an interval boundary.

The same hourly API response can also contain variables with different valid-time semantics. A blanket shift of the whole response can therefore align one variable while misaligning another.

Methods

The Experiment used two Evidence layers.

First, current Open-Meteo first-party documentation was checked for the valid-time definitions of hourly shortwave_radiation, global_tilted_irradiance, and the corresponding instant radiation variants.

Second, retained private PV data was used for a bounded sensitivity check. GTI timestamp T was compared with measured one-hour PV generation for intervals ending at:

T-2h, T-1h, T, T+1h, T+2h

Only daylight rows with GTI of at least 20 W/m² were included. The retained period from 2026-09-17 through 2026-09-26 produced 104 matched hourly comparisons.

Here, shift 0 compares GTI(T) with the PV energy accumulated during the hour ending at T, matching the documented preceding-hour semantics. Shift +1 represents the alternative interpretation that GTI(T) belongs to the following hour beginning at T.

Private raw household measurements and sensitive system details are not published. Only aggregate time-alignment statistics are public.

Experiment Log

Valid-time definition

Open-Meteo documents the non-instant hourly solar-radiation fields examined here as preceding-hour means and provides separate instant radiation variables for the indicated time.

Thus, non-instant GTI at 14:00 maps to the 13:00–14:00 interval.

Bounded PV alignment check

The sanitized correlation results were:

The documented alignment had the highest tested correlation, but 0.769 versus 0.764 is too small a difference to treat this private field series as independent proof of the timestamp definition. PV generation depends on additional factors, and adjacent-hour weather values can be strongly related.

The correct Evidence order is therefore: provider documentation establishes the API semantics; the field data checks whether the local implementation is broadly consistent with that interpretation.

Downstream rule

The retained monitoring workflow now records the rule:

Open-Meteo time T = preceding-hour mean

For non-instant hourly radiation, T should be handled as the interval end. When a point-like comparison at exactly T is needed, an instant radiation field should be used.

Conclusion

For the current Open-Meteo hourly solar-radiation fields examined here, a non-instant radiation value timestamped T represents the mean over the preceding hour.

Therefore:

GTI at 14:00 → average radiation over 13:00–14:00

not:

GTI at 14:00 → radiation for 14:00–15:00

and not necessarily:

GTI at 14:00 → instantaneous radiation at 14:00.

The corresponding *_instant fields provide point-like values for the indicated time.

This conclusion is VERIFIED from Open-Meteo first-party documentation. The private 104-point PV comparison is consistent with that interpretation but is deliberately secondary Evidence because the adjacent following-hour alignment produced a similar correlation.

The reusable implementation rule is to preserve each variable’s documented valid-time semantics. Do not apply a blanket one-hour shift to all hourly Open-Meteo fields.

Evidence summary

The Evidence States for the information obtained in this Experiment are as follows.

See Evidence State for the shared definitions.

Operational implications

For time-series implementations, store interval semantics in addition to the timestamp. For non-instant hourly radiation, interval_end = T and interval_duration = 1h make the meaning explicit.

A chart may place an interval value at its end, midpoint, or across a one-hour bar, but that display choice should not change the underlying interval. For instantaneous PV power comparisons, use an instant radiation field when appropriate. For hourly PV energy, align the preceding-hour radiation mean with the same one-hour interval.

Most importantly, inspect the valid-time definition of each Open-Meteo variable separately rather than shifting an entire hourly response together.

A sanitized aggregate of the alignment check is available at time-alignment-summary.json.

Sources

The primary source for the valid-time conclusion is the current first-party Open-Meteo documentation:

• Open-Meteo Weather Forecast API documentation

The local PV and weather series used for the secondary alignment check are retained privately.

Machine-readable experiment record

A machine-readable canonical record of this Experiment is published as JSON.

AIEL-2026-0012 experiment.json

Related experiments

OPENMETEO-MSM-001 addressed collection-path reliability and recovery of recent missing records. AIEL-2026-0012 addresses valid-time semantics and sub-daily alignment; the records are related, but neither conclusion depends on the other.