https://ams.confex.com/ams/27WAF23NWP/webprogram/Handout/Paper273676/GFS4cycle_2015_NWP_Chicago_15A.pdf
you might have not seen it before
GFS 06Z and 18Z cycle are less skillful than the 00Z and 12Z cycles. • The difference in the NH is the largest
• 06Z data count is always less (10~30%) than other cycles. • Even though the 18Z-cycle forecast score is worse than that of the 00Z cycle, 18Z conventional data count is not less than 00Z data count!
18Z has the same amount of Aircraft data as the 00Z does, but has worse forecast skill scores. • 12Z has less Aircraft data than 00Z, but has similar forecast skill scores to 00Z. • Difference in Aircraft data cannot explain forecast skill differences between 00Z/12Z and 06Z/18Z cycles.
There is no difference in satellite data count among the four cycles. • 18Z and 00Z have the same amount of conventional data, but 18Z has worse forecast skills than 00Z. On the other hand, 12Z has less conventional data than 00Z, but 12Z and 00Z have similar forecast skills. • 06Z and 18Z cycles have 10 to 20 times less land rawinsonde observations than the 00Z and 12Z cycles. Is the lack of RAOBS responsible for the lower forecast scores of the 06Z and 18Z cycles? More investigation is required to understand its impact
The forecast skills of GFS 06Z and 18Z cycles are not as good as that of the 00Z and 12Z cycles.
(based on the rest of the paper it indicates that these factors take effect later on in the forecast period)