Running WRF with DFI cannot find start time in wrffdda_d01

[tritch]

Hi all, I'm trying to test running DFI with some short range forecasts. I've set the DFI to opt 3 and to run backward 20 min and forward 20 min.
In this scenario real.exe succeeds but wrf.exe fails when the filter moves forward. It fails to find the start time in the wrffdda_d01 file.

In an alternate scenario of choosing DFI opt 1 (and altering nothing else), real.exe fails because d02 "time on domain" moves to an absurd date of 2035-09-06. (rsl.error.000.DFI1.txt).

I've not found much information online regarding proper DFI setup, so I may be missing something simple, but I'd appreciate any insight as to what might be going wrong.
Thanks

 

[kwerner]

Hi,
I don't think you can turn on FDDA during the DFI (especially when it involves backward model integration), and you probably shouldn't. Once you produce a filtered initial condition file (you can stop after running DFI), you can start the model as if it is from a directly-interpolated file from real, and then you can use FDDA. But there may be a direct conflict between the two: when you run DFI, you change the initial condition file (with noise removed), but when you nudge the model, you are forcing the model with un-filtered analysis. DFI is really only useful for short-range runs, which nudging is typically not needed for.

 

[tritch]

Thank you, turning off the FDDA options allowed the model to progress. I was aware of the differences in purpose between obs nudging and DFI, but didn't anticipate they'd actually conflict within the model.