I'm trying to set up a Forecast Model Run Collection using THREDDS. I have a collection of netCDF files, where the time at which the simulation was run is part of the filename:
In the files, the time variable is an integer, and gives time as number of hours since a given date. This date is the same in all my files. Since the times are given in integer hours, there should be absolutely no uncertainty as to whether two timesteps are the same.
However, when the THREDDS server presents this as an FMRC using the "best" model (for a given timpstep, it always serves data from the most recently performed simulation), it converts the time variable to a double, and somehow manages to present some timesteps which are actually the same as slightly different. For example, all of these are actually the same in my data, yet are presented as three different timesteps:
-65.10000033333334, -65.10000033333333, -65.1
The reason that the numbers are not whole hours is that the server also changes the reference point for the time axis from that specified in my files, to the simulation run time of the first file. This is fair enough, if only the data would be presented correctly.
I find the THREDDS documentation to be a collection of nice examples rather than an exhaustive list of all requirements and all available options, so I haven't been able to figure out how to fix this problem.