At each instant, the actual solar irradiance has a specific spectral distribution that most generally differs substantially from the standardized reference conditions under which photovoltaic (PV) cells are rated. To help reduce the dimensionality of the calculations needed to quantify the spectral effects on PV cells, various spectral indices have been introduced in the literature to simply characterize the incident spectral distribution with a single number. In previous studies, diverging conclusions have been reached about whether or not a single number can effectively represent a specific spectral distribution. This disagreement needs to be resolved before the usage of spectral indices can be considered best practice. To that effect, this study proposes a novel modeling approach. Here, three spectral indices (Average Photon Energy, APE; Blue Fraction, BF; and Useful Fraction, UF) are critically evaluated and compared using the SMARTS radiative transfer model and a large database of atmospheric inputs that are representative of the vast diversity of possible conditions worldwide, thus avoiding the rarity and limitations of costly experimental measurements. A total of 28,260 pairs of direct normal and global horizontal spectra are obtained, from which the three indices are calculated over six different wavebands that characterize different PV technologies. The key variables that have a significant impact on the indices are found to be: air mass, aerosol optical depth, Ångström exponent, and precipitable water. Using a variety of tests to evaluate the behavior of the three indices relatively to the reference direct and global spectra, results show that UF is prone to anomalous variations under a variety of circumstances. APE and BF behave similarly and reliably, but APE is found more responsive to even small changes in atmospheric conditions, and is thus preferable. A critical finding is that none of the indices can be considered bijective because compensations between red-shifting and blue-shifting atmospheric processes often result in different spectral distributions having the same index value.
Read full abstract