Results differ with different restart times

[chunchih]

Hi,

I ran a real-case simulation with WRF V4.2.1. In my original simulation, I restarted my run at 16:00:00 UTC (4 hours into my simulation) because I wanted my simulation to produce output files at a different time interval.

When I tried another simulation with the same number of processors, however, this time without the restart, I got slightly different results after 16:00:00 UTC than my original simulation that restarted at 16 UTC. Theoretically, the simulation for the same case run with the same number of processors should produce the same results regardless of the restart time or usage.

I read in some archived posts that some previous WRF users have encountered the same issue. Is there an explanation for this?

Attached is my namelist.input. I used nwp_diagnostic = 1 and from one of the posts I read (restarts not producing identical output), this might be the problem?

Thanks,
David

 

[kwerner]

Hi David,

In some of the earlier versions of WRF, for a restart to give identical results, it was necessary to start the restart at a boundary time. I *think* this has now been fixed in the code. As a test, can you give version 4.5.1 a try and see if you're able to get identical results?

 

[chunchih]

Hi kwerner,

The results only differ quantitatively (very slightly) with different restart times, so I don't think I need to rerun my simulation with version 4.5.1.

David

 

[Ming Chen]

Some slight differences in wrfout files from continuous and restart runs could possibly be cause dby rounding errors. To get identical results, you have to recompile WRF in debug mode, i.e., ./configure -D. However, wrf.exe compoiled in debug mode runs very very slow and we won't recommend doing so.