I asked this question a few days ago. Ansley kindly looked at the original data and solved the mystery.
I was wondering what this time axis is: It starts from "01-JAN-0001 00:00:00" and it doesn't include any CALENDAR attribute. I paste the output from "ncdump -h" below.
I thought the default calendar is Gregorian or similar, but the timeseries doesn't include any leap year. It covers from 1990-01-01 to 1997-12-31 but there is no Februrary 29 in the listing of list t[gt=eta] .
Without a CALENDAR attribute, Ferret assumes the Proleptic Gregorian calendar. So, there should be leap years.
The trick is that the original daily data increases the timestep from 1 day to 2 days between the February 28 and March 1 of the leap year!
Instead of this trickery, the creator of the dataset should have set the calendar attribute to "NOLEAP" and used a uniform timestep of 1 day throughout. . . .
(This is a climatological simulation, a numerical simulation under a repeated annual cycle, so the original calendar is likely to have been the no-leap calendar.)
Ryo