The simulation is based on an hourly data file (*.MET). When only monthly values are available, PVsyst generates hourly values according to a model in a stochastic (random) process.
For irradiances, PVsyst uses the "Collares-Pereira" model, which generates a series of daily irradiations, and then hourly irradiances within each day, using fully stochastic process (Markov matrices). This model tries to reproduce irradiance time series, with a statistical behaviour analogous to measured values in several sites in the world. The result will be a synthetic hourly year, which is completely different from one execution to another one.
During simulation of a grid-connected system, the differences on the yearly yield may be usually up to 0.5 to sometimes 1%. The effect will be higher when simulating systems with non-linear beghaviour like stand-alone.
For temperatures, as the temperatures are mainly governed by atmospheric circulations rather than irradiation, we don't have any model relying the daily temperature to the irradiations. Nevertheless the temperature within a given day is rather well correlated to the irradiance. It takes a rather sinusoidal shape, with an amplitude about proportional to the daily irradiation, and a phase-shift of about 3 hours. In absence of a day-to-day model, the passage to the next day is randomly modified in order to obtain the required monthly average and a reasonable amplitude.
Therefore, the model waits for the "meteo" monthly average temperature (24h)" (not the Daytime average !), and irradiance data.
For constructing a Synthetic Meteo file, please use "Tools" / "Synthetic Hourly Data Generation".
When choosing a site within a project, PVsyst will automatically create the corresponding Synthetic file if there is no corresponding file in the \Meteo\ database elaborated by Meteonorm.
NB: The Collares-Pereira model has been established in the 1980's, when the irradiance was not so well understood as nowadays. To my knowing nobody has published about this model since these early researches. The version 5.xx of PVsyst uses this original version of the model.
However the Meteonorm team has significanntly improved the model, and implemented it in the new Meteonorm program. The version 6 of PVsyst makes use of this new version of the synthetic generation model.