Before evaluating the overall model performance, the diurnal
behavior of the GHI predictions for each scheme was investigated
for cloud-free cases, as determined both from the measurements
and the model. In this case the ±5 min averages of GHI measurements
were used to improve the comparisons at large SZAs. As seen
in Fig. 2, for Thessaloniki, there is a diurnal dependence of the ratio
model to measurements in three of the SW radiation schemes. The
magnitude of this dependence varies among models, and because it
is symmetric with respect to local noon it can be associated with
the variation of SZA. Generally, all models overestimate the GHI
throughout the day. Dudhia's scheme shows the smallest SZA
dependence, of the order of 5e10%, while the GFDL scheme shows
the largest diurnal pattern with overestimation up to 50% at SZAs
around 80. The other two schemes show similar behavior with
largest overestimation of 20e25%. Apparently, there is also a seasonal
dependence of the model-to-measurements ratio, with April
and July showing smaller and less SZA-dependent deviations
compared to October and January. These seasonal differences could
be due to aerosols which are more abundant during the warm
period, leading to reductions in the measured GHI, while the model
estimates are unaffected. The large diurnal variations cannot be
attributed to measurement errors since the angular response error
of this type of pyranometers is substantially lower (nominally ±2%