TMY data generation

Information from the multi-year time series can be summarized into a Typical Meteorological Year (TMY), which reflects the most frequent weather conditions of a particular site. This dataset is designed for being used in simulation software that is able to compute the electricity yield of a solar energy system. Only complete years are used for the construction of TMY.

The TMY P50 is constructed by selection of most representative months from the available series (the most typical January, February, March, etc) which are finally concatenated into one artificial and representative single year.

For TMY Pxx (P90, P75, etc.), to increase the likelihood of finding better candidate months, the original data is split into smaller chunks of 15 days, and many candidate months are generated by mutual combinations of these smaller data chunks.

The selection of representative months follows two criteria:

• Minimum difference between statistical characteristics (annual average, monthly averages) of TMY and time series. This criterion is given about 80% weighting.
• Maximum similarity of monthly Cumulative Distribution Functions (CDF) of TMY and time series, so that the occurrence of typical hourly values is well represented for each month. This criterion is given about 20% weighting.

The importance of each parameter when selecting the typical months is weighted according to the type of solar energy application under consideration. This way, direct (DNI), global (GHI) and diffuse (DIF) irradiance and also Air Temperature at 2 metres (TEMP) may have different weights on the TMY construction.

Other meteorological parameters can be also included, but typically these are additional parameters with lower accuracy and less relevance in the analysis, thus they don’t have weight in deciding about the choice of the representative month.

In general practice, various datasets can be derived from the time series:

• The TMY P50 data set represents, for each month, the average climate conditions and the most representative cumulative distribution function, therefore extreme situations (e.g. extremely cloudy weather) are not represented in this dataset.
• The TMY P90 data set represents for each month the climate conditions, which after summarization of irradiation values for the whole year, result in the value close to P90 derived by statistical analysis of uncertainties and interannual variability. Thus TMY for P90 represents generally a conservative estimate, i.e. a year with the annual irradiation which is close to the lowest identified within the time series.
• The TMY P75, P99 or in general any Pxx can be constructed as well.

The method for calculation TMY P90 (or similarly other scenario as P75, P99 or in general Pxx) data set is based on the TMY P50 method described previously, which is modified in the way how a candidate month is selected.

The search for the set of twelve candidates is repeated in iteration until the condition of minimization of difference between the annual P90 value and the annual average of the new TMY is reached (instead of minimization of the differences in monthly means and CDFs, as applied in the P50 case).

Once the selection converges to the minimum difference, the TMY is created by concatenation of selected months. Note that P90 annual values are calculated from the combined uncertainty, which besides the model generic performance, takes into account the inter-annual variability due to local weather conditions.

Some popular simulation software are able to generate synthetic hourly time series from long-term monthly averages. Mathematical models for generating synthetic time series have been tuned for few specific sites in temperate climate. The performance of these models is not validated in other climatic zones such as rain tropics or arid deserts. This is a widely used approach. However, it generates a distorted representation of actual time series.

Typical Meteorological Year data are closest to real time series as they include 12 fragments of real data that describe a climate more realistically based on the given selection criteria. Energy Yield simulations using TMY data represent better accuracy compared to energy yield simulations done using synthetic time series data.

It is important to note that the data reduction in TMY is not possible without loss of information contained in the original multiyear time series. As a result of generating TMY and mathematical rounding, long-term monthly and annual averages calculated from TMY data files may not fit accurately to the statistical information calculated from the multiyear time series.

Therefore time series data are considered as the most accurate reference suitable for the statistical analysis of solar resource and meteorological parameters of the site. Only time series data can be used for the statistical analysis of solar climate.