Small negative surface precipitation using Morrison 2-moment

[anewman]

Hi,

We have a simulation using WRFV3.2.1 (this is fixed due to other model infrastructure) using the Morrison 2-moment scheme. In the first few hours of the simulation we see very small negative surface precipitation values, ranging from -1e-30 to as negative as -1e-20.

Right now we're treating them as roundoff and are setting them to zero for our application. However, we're wondering if there is a larger issue with our simulation configuration, or if this behavior is known with the Morrison scheme or other microphysics schemes?

A few other relevant details. The surface precipitation variables are not impacted by advection routines, thus moist_advect_opt and mp_zero_out have no impact. Our configuration uses a grid spacing of 50 km with roughly a 4 minute (240 s) model time step. We've also tested using a 60 second time step with the same results.

Thanks,
Andy

 

[Ming Chen]

Andy,
We are not aware of the negative precipitation when using Morrison scheme. Are you looking at rainnc or rainc+rainnc? Would you please send me your namelist.input to take a look? It is also helpful if you can tell me the forcing data you used.
By the way, I suppose the negative value disappeared after longer integration time? Please let me know if I am wrong.

 

[anewman]

Hi Ming,

We see the negative values in rainnc and snownc scattered across the domain and they do eventually all become positive. However, we have not checked the internal time step rainncv and snowncv variables to see if the Morrison scheme is consistently generating small negative values throughout the simulation. See attached for our WRF namelist.

We are using the ERA-I reanalysis to generate initial and boundary conditions across a pan-Arctic domain.

We're on Cheyenne if you want to look at our specific configuration there.

Thanks,
Andy

 

[Ming Chen]

Andy,
I looked at the code module_mp_morr_two_moment.F, in which you can find the piece of code below:

3639 ! GET PRECIPITATION AND SNOWFALL ACCUMULATION DURING THE TIME STEP
3640 ! FACTOR OF 1000 CONVERTS FROM M TO MM, BUT DIVISION BY DENSITY
3641 ! OF LIQUID WATER CANCELS THIS FACTOR OF 1000
3642 
3643         PRECRT = PRECRT+(FALOUTR(KTS)+FALOUTC(KTS)+FALOUTS(KTS)+FALOUTI(KTS)+FALOUTG(KTS))  &
3644                      *DT/NSTEP
3645         SNOWRT = SNOWRT+(FALOUTS(KTS)+FALOUTI(KTS)+FALOUTG(KTS))*DT/NSTEP
3646 ! hm added 7/13/13

I believe this is where the negative values first appear and then pass on to rainnc and snownc. I believe that rainncv and snowncv at internal time step should also be negative. Unfortunately I don't have time to further debug which FALOUT is the culprit. I will contact the developer and get back to you if I know for sure the problem.

 

[anewman]

Hi Ming,

Thanks for checking the Morrison code, and thanks for contacting the developer. I had also dug around a little, but was also unsure of which component term would cause the issue and the best way to fix it. Thinking a little more, if I had to guess I would think rain and/or graupel fallout given the higher fall speeds of those species. I'll try to dig more also and let you know if I figure anything out.

One other question, do you know of this behavior in other microphysics schemes? I can certainly check the code of each scheme, but just curious if this has come up elsewhere?

Cheers,
Andy

 

[Ming Chen]

Andy,

I am not aware of the similar behavior in other microphysics. We usually look at 12-36hr forecast of rainfall and verify against StageIV. So it is possible that we might miss some information at the right beginning of the simulation.

Ming