2. The service (2013)

contact: support-sales(-at-)soda-is.com


A Typical Meteorological Year is one year of meteorological parameter(s), which is representative of a given situation (median  Percentile 50, pessimistic e.g. Percentile 75 or optimistic e.g. Percentile 10).

No missing data in a TMY: gap-filling method is required.

Why to use a TMY instead of using the long term dataset of meteorological parameters?

  • “Widely-used” simulation software in solar energy (e.g. PVSYST, System Advisor Model) require one year of meteorological data <=> at the same time, industrials generally use TMYs for their (pre)feasibility studies of solar energy project
  • It can be useful for comparing two different locations (even if the temporal coverage of the data are different)
  • For “not-so-long LT”, TMY’s analysis increases the combinatorial possibilities (monthly data blocks)10 years LT => ~ 62 billions (12^10) possible years of data
  • Gain of time when you have many sites to proceed


Fig. 1: from a long term time series (at least 10 years) to a one representative year: the TMY

1. State of the art

The starting point of the service is the NREL (National Renewable Energy Laboratory) method, published in Hall et al. (1978) and Kalogirou (2003).

Here are the specifications of this TMY method:

  • Long term of hourly values
  • data-blocks of monthly meteorological datasets
  • TMY based on the linear weighted combination of the Filkenstein-Schafer distance, where weights are chosen depending on the application.

The model defines that the selected month is the one which minimizes the following expression:

5 dFS(GHI) + 5 dFS(BNI) + 4 dFS(T) + 1 dFS(WS) + 4 dFS(DWT)

where GHI: Global Horizontal Irradiation, BNI: Beam Normal Irradiation, T: Temperature, WS: Wind Speed, DWT: relative humidity. In the rest of this page, this expression, which is a function of different meteorological parameters will be named "driver".

NB: For the percentiles outside P50, the aggregated value of the driver over the temporal blocksize is used.
NB2: Please note that the distance is based on the histogram. This implies to select a bin width to generate the histogram and thus the CDF, and different bin width will provide different histograms and thus different TMYs.

Fig. 2: top left: example of histogram. Top right: CDF (Cumulative Distribution Function) derived from the histogram. Bottom left: two histograms. Bottom right: the Filkenstein-Schafer distance is defined as the space between the two CDFs derived from the two histograms (dFS), represented in pink

2. Method and Innovations

2.1 Innovations:

The main innovation of the new generation method is to take into account the characteristics of the conversion system under concern for the selection of the representative months (or other block size):

  • Tilt of PV panels
  • Type of solar Tracking
  • Security issues...

This resulted in the creation of a "driver", as introduced in the previous section: the driver is a one-dimensional composite time series from the multidimensional meteorological long-term dataset more related (or more linearly correlated) to the energy production of the solar energy conversion system of interest (PV, CPV, CSP, etc...).

Fig. 3: Influence of the types of system or plane orientations on the radiation received by the plane

Other improvements are:

Fig. 4: Example of the cover page of the report provided together with the TMYs.

To give an example of driver, let's take the example of a one-axis sun tracking: the effective part of the BNI for this system is the cosine of the incident angle. It exists power output thresholds due to:

  • Min irradiance (min start-up temperature)
  • Max irradiance (defocusing)
  • Max wind speed (tracking security)

In that situation, the driver would be derived from the cosine of the incident angle, where the values would be equal to 0 (or set to lower values) when the threshold in temperature, irradiance and wind speed are reached.

2.2 Service description:

The service is so far (Sept. 2013) an offline service, and the routines are available in Matlab. It is envisaged to develop a Web service on the new SoDa website by the end of the ENDORSE project in Dec. 2013, maybe with the use of "tokens".

Inputs: the routines are now fully operational, and are able to take into account the following inputs (non-exhaustive list):

  • HelioClim3-version4 or HelioClim3v4-MC (McClear) (radiation values)
  • Ground station measurements (radiation and other meteo param.) provided by the customer or by another national weather station
  • MERRA reanalyzes, NASA (other meteo values)
  • HelioClim3 values calibrated using either:
    • ground station measurements,
    • or the solar Atlas PACA (service S1 ENDORSE)
  • ...

Method: Whatever the input data selected as input to the service, a quality check is applied on the data (following the requirements of one of the outcomes of the research also carried out in ENDORSE -see the poster), and a calibration is operated if required. Then the data are completed in order to avoid gaps in the provided TMY.

Then the TMY is generated taking into account the requirements of the user concerning:

  • The site coordinates,
  • The type of system and the necessary characteristics to define the driver,
  • The time step of the TMY,
  • The temporal block size,
  • The percentile(s),
  • The output format (compatible with SAM or PVsyst),
  • The type of radiation data to process (HelioClim only, ground station measurements...)

Outputs are the TMY(s) and associated report.

Fig. 5: The ENDORSE "TMY generation" service

Note on the data completion tool: MINES ParisTech has developed a statistical tool to complete the missing values of the set of input data before the TMY generation. The principle is to fill the gaps with values taken in the temporal neighborhood of the missing period. A distance is computed between, on the one hand, the radiation values before and after the temporal gap with, on the other hand, the values in the previous and following days. The selected period of data corresponds to the minimum for this distance.  

3. Results and validation

3.1 Results:

The test site is Carpentras, and the ground station measurements belong to the Baseline Surface Radiation Network (BSRN).

3.2 Validation

Fig. 6: Illustration of the risk of inconstency between one year of yield computed from the 10 years of the meteorological values and from a TMY

Please find below the validation procedure and the results performed by MINES ParisTech:

Fig. 7: validation procedure

Fig. 8: Results 1: with the Filkenstein-Schafer distance, the TMY based on the adequate driver returns better results than the ones obtained with TMY3 and BNI-only driver

Fig. 9: Results 2: with the P50 distance, the TMY based on the adequate driver returns better results than the ones obtained with TMY3 and BNI-only driver

Fig. 10: Results 3: The TMY P10 based on the adequate driver returns similar results than with BNI-only driver.

Transvalor has also performed its own validation of the method. The yield has been simulated over 9 years of data, for a CPV system and a wind limit of 7.5 m/s, and computed an average yield over the year. The obtained yield was compared to the one obtain with:

  • a driver which does not represent the system under concern (driver GHI): the bias was 4.47%
  • the adequate driver, taking into account the maximum wind speed threshold: bias: 0.73%

As a conclusion, the validation processes have agreed and demonstrate the enhancement of this TMY generation method compared to the state of the art, with respect to the proctution of the solar system.

4. Price

The price strategy is now mature, and mostly depends on the need of a prior calibration of the long term HelioClim-3 dataset with:

  • the in-situ measurements provided by the Customer
  • the data retrieved from a national station network such as Meteofrance
  • the PACA Atlas
  • ....

Supplied results: TMYs P50, P75 or P90 and a complete report. Once the payment is received, the TMYs and associated reports are sent by email.

Please, contact us to receive a quotation!

5. Leadtime

The TMY and report are generated in approximately half an hour, if no preliminary calibration is required by the user.


  • Market:
    • Beforehand of the ENDORSE project, a few TMYs have already been sold => a market was already existing.
    • ENDORSE confirms this fact: 32 TMYs have been sold so far in 2013. We plan approx. 50 TMY requests for next year.
  • The service is already fully operational and sustainable on a medium and long term basis.
  • A Web service is envisaged for the new SoDa website. If we opt for this development, it will be available by the end of the ENDORSE project (Dec. 2013)

This ENDORSE service is a success, since Transvalor has enhanced the service provided to the Customer with highly qualified routines, pertinent information, and relevant validation processes.