This is a complex problem indeed.
In the years 1990 we had passed a full-day seminar on this question with all involved people in Switzerland (including the Meteonorm team), and did not find any satisfactory solution. With our mountains, meteo measurements in Switzerland are especially concerned with this question of course !
With any ground-measured data, the horizon loss of irradiance is already included in the meteo data.
This is very difficult to use because:
- the horizon line is very sensitive to the exact location: the horizon at the exact sensor position may be very different of the horizon seen by the PV system.
- if applied to monthly values, the synthetic generation doesn't "know" the horizon, and will generate values with beam component below the horizon. Which will be incorrect in the simulation of course.
Therefore in any case (except on-site meteo measurements), the horizon data should ideally be horizon-free and the horizon effect should be computed within PVsyst, using the effective horizon line of the site.
There is no simple mean for avoiding the problem.
With monthly values, we could generate a full year of hourly values, evaluate the loss (on beam) due to the pre-specified horizon of the site, and then add these losses to the monthly original values and re-generate a synthetic hourly file. This file will have better "free horizon" values on which we can apply the real horizon of the PV system.
With hourly values, we should reconstruct the supposed beam component below (and behind) the horizon. We have now models for doing this with some confidence (probabilistic continuity of the weather).
We have tested these methods, and they can give acceptable results. But sorry, I did not yet implement this special case in PVsyst.