In a realistic run, DA will wait for observations to arrive which is usually about 40 minutes after the top of the hour. This gives enough time for the previous forecast to complete a one-hour forecast.
In the sandbox run, DA does not wait for observations so it starts quickly once a cycle is activated. At this time, the previous forecast may not complete a one-hour forecast yet and because we use links instead of copy, this may cause the next fcst task to start on an unfished 1hr fcst file.
A simple solution to this for now is to set STARTTIME_FCST="00:40:00".
An ideal solution may be that the DA task depends on the fcst_f01 task for the previous cycle.
But this will generally NOT be a problem if we run realistic data assimilation cycles.
In a realistic run, DA will wait for observations to arrive which is usually about 40 minutes after the top of the hour. This gives enough time for the previous forecast to complete a one-hour forecast.
In the sandbox run, DA does not wait for observations so it starts quickly once a cycle is activated. At this time, the previous forecast may not complete a one-hour forecast yet and because we use links instead of copy, this may cause the next fcst task to start on an unfished 1hr fcst file.
A simple solution to this for now is to set STARTTIME_FCST="00:40:00".
An ideal solution may be that the DA task depends on the fcst_f01 task for the previous cycle.
But this will generally NOT be a problem if we run realistic data assimilation cycles.