Did I noticed that only the first initial step of forecast (GFS initialized 00Z-06Z-12Z-18Z) presents values of 2m dewpoint very underestimated (6-8°C), especially over sea

I use NMM core with Ferrier MP and BMJ Cu scheme.
The problem does not seem due to the 2m temperature. It may be due to incorrect initialization of the model, someone had already noticed this?