Quantifying the amplified bias of PV system simulations due to uncertainties in solar radiation estimates

Solar radiation databases used for simulating PV systems are typically selected according to their annual bias in global horizontal irradiance (G(H)) because this bias propagates proportionally to plane-of-array irradiance (G(POA)) and module power (P-DC). However, the bias may get amplified through the simulations due to the impact of deviations in estimated irradiance on parts of the modeling chain depending on irradiance. This study quantifies these effects at 39 European locations by comparing simulations using satellite-based (SARAH) and reanalysis (COSMO-REA6 and ERAS) databases against simulations using station measurements. SARAH showed a stable bias through the simulations producing the best Pp c predictions in Central and South Europe, whereas the bias of reanalyses got substantially amplified because their deviations vary with atmospheric transmissivity due to an incorrect prediction of clouds. However, SARAH worsened at the northern locations covered by the product (55-65 degrees N) underestimating both G(POA) and P-DC. On the contrary, ERAS not only covers latitudes above 65 degrees but it also obtained the least biased P-DC estimations between 55 and 65 degrees N, which supports its use as a complement of satellite-based databases in high latitudes. The most significant amplifications occurred through the transposition model ranging from +/- 1% up to +/- 6%. Their magnitude increased linearly with the inclination angle, and they are related to the incorrect estimation of beam and diffuse irradiance. The bias increased around + 1% in the PV module model because the PV conversion efficiency depends on irradiance directly, and indirectly via module temperature. The amplification of the bias was similar and occasionally greater than the bias in annual G(H), so databases with the smallest bias in G(H) may not always provide the least biased PV simulations.Peer reviewe