Hi Paul,
thanks for your help. Especially the
handbook link proved helpful to get some better understanding.
But upfront an apology: the quoted very low system performance of 7850kWh/a was not based on TMY from PVGIS, but from
EnergyPlus
(too many files on my disk right now). The TMY from PVGIS (in it's 2007 - 2016 incarnation) results in a system output of 9849kWh. This narrows the gap to merra, but it's still ~16%; the PVGIS output sounds more reasonable, talking to local providers of PV systems here.
So, let's dig a level deeper. First, the handbook (p. 33, chapter 2.4.3) educated me on the fundamental equation (... now you know from which level of understanding I'm coming from :-( ):
GHI = DNI*cos(SZA) + DHI
For merra, Siren's makeweatherfiles help file states:
DNI is computed using formulae derived from the NREL Centre for Renewable Energy Resources DISC DNI Model. The handbook gives some background on p.33ff. with the caveat '
The model is not a rigorous physical algorithm, because the coefficients for computing clear-sky transmittance values were derived from empirical regression analyses of measured DNI and GHI data from Atlanta, Georgia'
The handbook also explains in 5.4.15 for PVGIS (albeit for an old version and not the current v5.0) that '
Monthly averages of daily sums of global and diffuse irradiation measured or calculated for 566 ground meteorological stations distributed over the region.'
Based on this, I investigated GHI - DHI = Direct_Irradiation, which is then the direct contribution to the radiation. Plotting Direct_Irradiation for SARAH vs. Merra (with hourly data), it becomes obvious that in many instances where SARAH has zero or very small values, Merra has substantial contributions (up to 600 W/m
2) of direct radiation. To quantify this, let's bucket Direct_Irradiation(SARAH) into buckets of 10 W/m
2 and then for each bucket calculate
Gs = sum(GHI(SARAH))
Gm = sum(GHI(merra))
Delta = (Gm - Gs)/Gs
giving the normalization advantage to SARAH as being the more trustworthy model. With this, we get
Hence, it appears that DISC overestimates direct radiation for cases where there is not much of it. Conversely, DHI is estimated similarly for SARAH and Merra (559kWh/a and 511kWa/a respectively) with an R
2 = 0.68 for a scatter on an hourly basis, which sounds reasonable to me.
Some more background, in case somebody might want to follow up (sounds like a diploma thesis to me :-) ):
- coordinates used are in Germany, 51.78N, 6.10E
- PVGIS doesn't provide files with GHI, DNI and DHI nicely listed. But one can cheat by configuring a PV system with panel slope = 0deg. and download without radiation components. This results in G_i = GHI; then one can download the same with radiation components to get D_i = DHI. The validity of this approach was verified by comparing resulting GHI, DNI and DHI with TMY data (incarnation 2007 - 2016). Parameters correlated perfectly for Jan - Apr, but not for May - Dec, which resulted in a bug report to PVGIS.
- the handbook also states that PVGIS also accounts for sky obstruction (shadowing) by local terrain features (hills or mountains) calculated from the DEM. This should not be relevant here, as diverging data points occur throughout the day and the horizon is <= 4 deg. around here.
I'm willing to elaborate further if anybody wishes. But for now, I stay with 'my' SARAH data sets and move on. Thx. again for your help!
Kind regards, Stefan