at SZA 60 and ±6% at SZA 80). It should be noted that this error
applies mainly to the direct irradiance component; therefore for
large SZAs the angular response error in GHI would be even smaller.
The large deviations of the ratios at large SZAs may have been
caused by obstacles close to the horizon, other local effects, and by
uncertainties in the measured GHI which at these SZAs has very
small values. Finally, the noise in the ratio can be attributed to effects
from varying aerosols, thin clouds that were not captured by
the cloud screening algorithm, variations in water vapor which are
not taken into account, and by uncertainties in the model
estimates.
The patterns shown in Fig. 2 are typical for all stations as far as it
concerns the general shape, although there are differences with
respect to absolute deviations. This is an additional indication for
the effects of aerosols which differ in quantity among stations, and
will be discussed later in more detail.
Since the diurnal patterns occur at all stations, it is reasonable to
assume that they are not caused by the measurements, but likely by
the radiation schemes. Therefore, the comparisons of the hourly
data and the relevant statistical analyses are restricted to data
corresponding to SZA of less than 70, as a compromise between
the model errors and the actual GHI levels.