Typical Meteorological Year (TMY): What It Is and How It Is Created

What is a Typical Meteorological Year?

A Typical Meteorological Year (TMY) is one year of hourly weather data, typically including air temperature, humidity, solar radiation, wind and pressure, that is meant to represent the typical conditions of a location. Engineers use it when a simulation needs a single, manageable year instead of decades of records.

The key point is how it is built. A TMY is an assembly of real historical months drawn from different years. It is not the average of the hourly values, not an observed calendar year, not a forecast and not an extreme year.

How is a TMY created from historical weather data?

The procedure starts with a multi-year hourly record for one location, from stations, satellites or reanalysis. The data is checked for gaps and quality problems. For each of the twelve calendar months, every available year is a candidate: all Januaries, all Februaries and so on.

One candidate is then chosen per calendar month, using statistics that compare the candidate with the long-term record of the same month. The twelve winners are joined end to end into one year. The sequence of months in the TMY follows the calendar, while the source years behind them differ, for example a January from one year followed by a February from another.

Details differ between methods. NOAA documents its own TMY distribution and procedures, the TMY3 data sets are described in the NREL user manual, and the PVGIS generator documents selected months and provenance for its own output. Variable weights, smoothing at month joins and persistence checks are method-specific, so a TMY should always be read together with its documentation.

Statistical selection of representative months

Most TMY methods rate candidate months with the Finkelstein-Schafer (FS) statistic. For one variable, FS compares the cumulative distribution of the candidate month with the cumulative distribution of that calendar month across all years. A small value means the candidate resembles the long-term distribution.

Several variables are combined, typically as a weighted sum of their FS values, so the choice reflects more than one quantity. Because the weights express the purpose of the dataset, a solar-oriented TMY and a building-energy TMY can pick different months for the same site. Some procedures then apply further checks, such as screening for persistence of weather conditions, before the final month is chosen.

The result is a typical month in a distributional sense. It is not guaranteed to be the mean month, and it is not designed to contain the most severe month on record. Huld et al. (2018) assess one specific method for building energy use; that study concerns that method and is not a validation of any other TMY product, including Climetry.

Why does an hourly TMY contain 8,760 hours?

The number follows from the calendar: 365 days multiplied by 24 hours gives 8,760 hourly values. An hourly TMY is therefore laid out as a non-leap year. Leap days are not part of the standard layout.

Do not assume that every product reaches 8,760 values in the same way. Source months may come from leap years, and files can differ in time zone (local standard time or UTC), in hour numbering and in whether timestamps are real dates or labels. Check the time convention in the metadata of the export before aligning a TMY with load profiles, tariffs or measurements.

Hourly resolution matters because many engineering questions concern duration and coincidence: how many hours exceed a threshold, or whether high humidity and low temperature occur together. Monthly means cannot answer these questions.

TMY vs historical weather data vs extreme years

The three types of data answer different questions. Picking the wrong one is a common source of misleading results.

Applications in engineering, solar energy and corrosion screening

Engineering and building simulation. Energy models for heating, cooling and ventilation commonly run on a TMY, because one year keeps simulation time low while covering all seasons.

Solar energy. Yield estimates for photovoltaic systems use a representative year of irradiance and temperature to compare designs and sites. A typical year gives an expected annual order of magnitude, but it does not show year-to-year variability or low-resource periods.

Corrosion screening. Hourly temperature, humidity and dew point can be turned into indicators, such as hours above a relative humidity threshold or hours with a small temperature-dew point spread. These indicators help to compare locations. They are not observed condensation, not a measure of corrosion damage and not a life estimate. Material response, surface conditions and contaminants such as chlorides need separate assessment.

How Climetry generates and provides representative-year data

Climetry Environmental Year 20 (CEY20) is a representative year built from a 2006-2025 source period. It is derived from ERA5-Land data and delivered for a 0.25-degree land grid, so the values describe a grid-cell background climate and not conditions at an exact installation. This source period and grid belong to CEY20 only and should not be assumed for other products.

In the implementation, each calendar month of each source year is a candidate. A candidate is scored with FS statistics against the pooled values of that month across all years. The statistics cover temperature, relative humidity, dew point and three derived indicators: relative humidity of at least 80 percent, a time-of-wetness indicator and a dew point spread of at most 2 K. These variables carry different weights. The candidate with the lowest weighted score is selected, with the earlier year winning a tie. RH90 hours are reported as an output indicator and are not a default selection weight. The method is inspired by ISO 15927-4 principles. We make no claim about formal certification here.

The selected months are joined into a delivered year of 8,760 hours on a synthetic non-leap calendar. The hourly CSV labels month, day and hour (hour_0based, positions 0 to 23) as synthetic positions in this representative year. They should be read as positions, not as observation timestamps, and they do not declare a UTC or local-time instant, so no time zone should be assumed. The export also carries the source period, version identifiers and limitations. In this implementation the selected months are joined directly, without smoothing or persistence adjustment at the month joins, so continuity across a join is not assumed.

Where an export is provided, the attached metadata is the authority for that product's source period, grid and calendar. Contains modified Copernicus Climate Change Service information; this is not an official Copernicus product.

Limitations

• A TMY is not an observed year, not a forecast and not a prediction of future weather or climate change.
• It is not an extreme year, a P90 case or a service-life prediction. A typical year can miss prolonged or extreme conditions that matter for design.
• Selected months do not preserve the multiyear chronology, and nothing guarantees continuity at the joins between months.
• Different methods, weights and source datasets give different TMYs for the same site.
• Humidity and dew point indicators do not prove condensation, corrosion damage or service life.
• Grid-cell data is background climate and not a measurement at your exact site.

References

• NOAA National Centers for Environmental Information: Typical Meteorological Year
• European Commission JRC: PVGIS typical meteorological year (TMY) generator
• Wilcox, S. and Marion, W. (2008). Users Manual for TMY3 Data Sets. NREL/TP-581-43156
• Huld, T. et al. (2018). Assembling Typical Meteorological Year Data Sets for Building Energy Performance Using Reanalysis and Satellite-Based Data. Atmosphere 9(2), 53

The summaries of these sources are paraphrased and do not replace them. The description of the Climetry implementation is based on its own repository code.