UV Index from Global Solar Irradiance: Performance according to cloudiness and air mass

Ultraviolet (UV) irradiance from the Sun has potential negative health impacts. Erythemal UV irradiance (UV-E) is obtained by weighting UV radiation in the 250-400 nm range with the average skin erythemal response function. UV-E is proportional to the UV Index (UVI), a tool used worldwide to inform the public about this environmental hazard. Since this magnitude is not currently measured at most meteorological sites, satellite based estimates are often used to generate UV-E information over broad areas. However, for a climatological characterization of variability and typical doses of UV-E, long time series with low biases are required. An alternative approach is to estimate UV-E from readily available information with which UV-E is highly correlated. This work builds upon previous research, which evaluated a simple model (Power Model or PM) to estimate UV-E from Global Horizontal Irradiance (GHI), relative air mass, and total atmospheric ozone concentration at the 10-minute level. In that general analysis, the model showed an uncertainty below 12% (as measured by its relative Root Mean Squared Deviation or rRMSD) when compared with ground UV-E measurements. Here, we present a more in-depth assessment of this model, using high-quality data from four mid-latitude temperate sites. The deviations of the model estimates from UVI ground measurements are analyzed under different cloudiness conditions and across a broad range of air masses. Our results confirm that PM performs best under clear skies and low air masses, precisely the conditions associated to the higher incident solar irradiances and potentially dangerous UV levels. Under these conditions, the PM has an average accuracy (rRMSD across sites) of less than 6% of the average of the measurements and outperforms. These results confirm that the model, particularly in its unbiased version, is a reliable and practical tool for the climatological analysis of erythemal UV irradiance.