James Barry

Sounds good, I look forward to the results. I would assume that the DIRINT or DIRINDEX model is the best bet, unless you would like to delve into Engerer...

As quick extra note: according to the comparison paper by Gueymard and Ruiz-Arias [1] quoted above, the best performing model is in fact "Engerer 2" [3], since it is developed specifically for one minute data. However the next best model is the DIRINDEX [4] model, which is basically DIRINT with an additional predictor based on applying a clear sky model. It would be great it PVsyst could consider using some of the more modern decomposition models, especially considering the use of high-frequency irradiance data. [3] Engerer, N. A. (2015). Minute resolution estimates of the diffuse fraction of global irradiance for southeastern Australia. Solar Energy, 116, 215–237. https://doi.org/https://doi.org/10.1016/j.solener.2015.04.012 [4] Perez, R., Ineichen, P., Moore, K., Kmiecik, M., Chain, C., George, R., & Vignola, F. (2002). A new operational model for satellite-derived irradiances: description and validation. Solar Energy, 73(5), 307–317. https://doi.org/https://doi.org/10.1016/S0038-092X(02)00122-6

There is already a huge amount of literature on this topic. As a start I would recommend [1]. As is usually the case with models, the parameters have been tuned for certain locations, so it makes sense to test them at many locations. In general one can say that Erbs usually overestimates the diffuse component under clear sky conditions, and this bias is severe (of the order of 25% using 10 minute data from a location in Western Europe). You won't see that statistic when looking at all the data (all sky), but if you specifically look at clear sky conditions, you will see that basically Erbs overestimates the atmospheric turbidity / aerosol optical depth, leading to a higher diffuse component. Put another way, the sky is clearer than what Erbs models it to be, there is less scattering due to aerosols and more direct light. This error leads to underestimation of the direct component and underestimation of shading losses in PVsyst It would definitely make sense to include DIRINT [2] in PVsyst, since that model uses more predictors, not just the clearness index. Specifically, DIRINT has a time component that takes into account variability, i.e. the change in clearness index. That helps then to distinguish between different weather conditions, and thus model them differently. All of that being said, the best is to import diffuse irradiance, which is however not always possible if it is not measured on site. [1] Gueymard, C. A., & Ruiz-Arias, J. A. (2016). Extensive worldwide validation and climate sensitivity analysis of direct irradiance predictions from 1-min global irradiance. Solar Energy, 128, 1–30. https://doi.org/https://doi.org/10.1016/j.solener.2015.10.010 [2] Perez, R. R., Ineichen, P., Maxwell, E. L., Seals, R. D., & Zelenka, A. (1992). Dynamic global-to-direct irradiance conversion model. ASHRAE Transactions, 354–369.

I recently imported high-resolution (1 minute) weather data (GHI, DHI, ambient temperature) from a BSRN station for the year 2020, which was a leap year. This was to estimate sub-hourly clipping losses. The CSV file has 527040 lines of data, which corresponds to 366 days of one minute data, starting at 2020-01-01 00:01 and ending at 2021-01-01 00:00, with end-of-interval convention. However, the CSV import gave the error "Error while reading date 31/12/20 at 23h02 Limitation of the custom file import at 12 months." The resulting MET file ends at 23:00 on the last day of the year, so in the end all is good and it won't make much difference to the simulation. However I am not sure why this error is thrown, since the data is correct and the "limitation" should not be there.

As Michele said, if the diffuse component was not also imported from satellite, it would have to be estimated using the Erbs model, which is highly inaccurate for high resolution data. It was developed using hourly, daily and monthly data. In general Erbs often overestimates the diffuse irradiance under clear skies, thus underestimating the direct component and underestimating shading losses. However I am not sure whether this explains your discrepancy. In general, satellite data does not have such high granularity, at least the raw data such as cloud properties may only be in 15 minute intervals. So it is important to know the limitations of the data. Some satellite data is also instantaneous and not really averaged over an hourly window for instance.