Scheduled Downtime
On Friday 21 April 2023 @ 5pm MT, this website will be down for maintenance and expected to return online the morning of 24 April 2023 at the latest

Unclosed moisture budget in MYNN scheme


New member
While running some idealized simulations with WRF version 4.1, I noticed severe problems in the moisture budget when using the MYNN scheme.
For a systematic test, I used the single-column model test case (em_scm_xy, no code modifications) with the attached input sounding and various PBL scheme settings (see attached namelist files).
Then I compared the accumulated moisture sources minus sinks (integral of QFX - RAINNC, only liquid precipitation occurred) to the change in total water content of the atmosphere, delta Qt:
The resulting moisture budget terms are displayed here:
View attachment waterbudget_qbud_wet_scm.pdf

For the LES (high-resolution without PBL scheme), MYJ and YSU schemes the two curves (red and violet) are reasonably close to each other, whereas for the two MYNN scheme variants they are not.
With the MYNN2 scheme, the atmosphere keeps losing more water (large negative delta Qt) than expected from the sum of the surface moisture flux and precipitation (QFX-RAINNC).
With the MYNN3 scheme instead, the atmosphere does not lose enough water given the large amount of precipitation.

Playing around with the namelist options bl_mynn_edmf, icloud_bl, scalar_pblmix, and bl_mynn_mixscalars did not change the curves significantly.
Dry air mass is conserved in all simulations, as I checked.

What is the problem here? I thought water mass should be conserved in WRF. But for the MYNN scheme, the budget is really bad!

Thanks for helping!



  • namelist.MYNN3.input
    4.8 KB · Views: 48
  • namelist.MYNN2.input
    4.8 KB · Views: 57
  • namelist.MYJ.input
    4.9 KB · Views: 48
  • namelist.YSU.input
    4.8 KB · Views: 52
  • namelist.LES.input
    6.4 KB · Views: 56
  • input_sounding.txt
    36.4 KB · Views: 56
Hi Matthias,
I would like to apologize for the delay in response. We have tried to reach out to the developer for MYNN, but have not received a response yet. We will let you know as soon as we do.
Are you using a microphysics scheme without an ice specie?
If so, could you please retry your experiment with a more complicated scheme?

I ask these questions because I'm guessing that this is some "strange" configuration that we don't test. We have never seen this drastic of a moisture budget problem in idealized of real cases, even going out 5+ day forecasts.

I may have a fix for simple microphysics schemes if this really is the problem.

Thanks in advance,
Thanks for the replies. As can be seen in the namelist file I use mp_physics=2, which has ice processes included.
Hi Matthias,

I saw your email, but I've been slammed with deadlines/conferences/workshops/proposals/paper reviews/etc. I don't mean to ignore this problem. In fact, I hope to attack it soon. Maybe we can iterate directly through email and just post a fix to this forum when we find one. I have a few ideas about where this problem may be coming from. It may take some coordinated testing or perhaps you can share your specific model setup(?).

I run into the same issue. I am using WRF-v4.3.3.

I am using WRF-v4.3.3. I want to verify the domain-wise moisture budget with IWV, ACLHF, RAINNC, and RAINC. I output in 1-min interval snapshots, and average every 60 output so that I end up having hourly-average data of all variables. I compute IWV at time t using the following formula
IWV(t) = SUM{ [MU(t)+MUB(t)] / g * [QVAPOR(t) + QRAIN(t)] * DNW }

and the change rate of IWV is
dIWVdt(t) = [ IWV(t) - IWV(t-1) ] / dt

where dt = 1hr. I then compute the precipitation rate with
PRECIP(t) = [ TTL_ACC(t) - TTL_ACC(t-dt) ] / dt

For moisture flux related to latent heating, I compute as
ACL_QFX(t) = ACLHF(t) / L_q
QFX(t) = [ ACL_QFX(t) - ACL_QFX(t-dt) ] / dt

where L_q = 2.5e6 J/kg is the heat of vaporization defined in WRF share/module_model_constants.F
Since I am using sf_sfclay_physics=1 (Revised MM5 Monin-Obukhov scheme), L_q should be a constant, as I understood from WRF/phys/module_sf_sfclayrev.F at release-v4.3.3 · wrf-model/WRF The verification is computed as the residue
RES(t) = dIWVdt(t) - [ QFX(t) - PRECIP(t) ]All data are first computed in 2D/3D. Once RES is obtained, I average REShorizontally. Additionally, I apply a moving average 25-hr on RES. The following figure shows (a) Integrated water vapor (IWV), (b) precipitation, (c) Q_flx (surface moisture flux), and (d) residue = dIWV/dt - (Q_flx - P). It shows that MYNN2.5 budget does not close. This is the same conclusion if I define

IWV(t) = SUM{ [MU(t)+MUB(t)] / g * [QVAPOR(t) + QRAIN(t) + QICE(t) + QSNOW(t) + QCLOUD(t)] * DNW }

to include cloud, ice and such.


I also attach the namelist.input for MYNN2.5 here.


  • namelist.input
    5.9 KB · Views: 0
Last edited:

MYNN has significant updates in WRFV4.6. Can you run this newest version and check the moisture budge again? Please keep me updated about the results. Thanks.
@Ming Chen

I use the same procedure and check WRFV4.6. The updated figure is as follows.

MYNN's budget imalance is less than 0.1 mm/day, and become the similar magnitude as YSU and MYJ when approaching equilibrium. It is a very good improvement.

Also notice that YSU is not the same by the end of day 3 between versions.


  • namelist.input
    5.9 KB · Views: 0
Last edited:
Here is the same figure but 0-9 days using WRFV4.6


  • 1718670630792.png
    531.7 KB · Views: 2
Last edited: