It is a misconception that the pre-PSM NSRDB data came from ground stations. The 1961-1990 version contained only 7% observations; the rest of the data came from an empirical model (METSTAT) that used meteorological data as inputs. The 1991-2005 version contained only 1% observations; the rest of the data came from the METSTAT model or the "SUNY Albany" satellite model (a precurser to CPR's SolarAnywhere product). Thus only a miniscule portion of the data in the TMY2 or TMY3 products consists of observations. The situation with Meteonorm is similar. Although some people say that Meteornorm has ~8,000 ground station, in fact only meteorological variables were measured at most of those stations. There are about 1,600 stations with measurements. The rest of the data comes from a satellite model; it is used to fill gaps in the ground site spatial coverage (which tend to be large outside of Europe). You can learn all these things by reading the NSRDB users' manuals or the Meteonorm software and theory documents. But I agree that the values from PSM tend to be high.