Pressure surface and total pressure (PSFC, P+PB) oscillations related to domain size

[MLandreau]

Hi everyone,

I noticed some strange pressure oscillations in my test simulations. Is it a known issue or is there a namelist option I chose wrong ?

Description of my simulation test :
- 1 domain
- DX=DY=9.477 km, NX=NY=121, NZ=52, DT=60s, DT_acoustic=20s
- PBL=MYJ, land_surface=NoahLSM, LW=RW=RRTMG,
- forced by ERA5, 1hourly
- More details in the namelist linked

Description of the analysis
- I first plotted the pressure over time interpolated to a given height (200m, 1500m, 10000m, 17000m) at a location in the middle of my domain where I output a time series. The figures 01_PXXXX shows this evolution for the time series output (blue) and the 10min averaged computed (black). One can see some oscillations with a period of about 1h30min
- I then computed a 1st order derivative with the 10-min time-averaged surface pressure (PSFC). I used several points on a cross-coast line that covers approximately 1/4th of the domain. The first derivative is plotted in 02_dPdt. The higher frequency oscillation (about 1h30) becomes more evident in the derivative plot. Note for all location, the oscillation is in phase
- I plotted the time autocorrelation of the dPdt signal to better measure the oscillation period (03_autocorrelation_dPdt)

Sensitivity
- I ran 4 more simulations :
-- blue : reference (domain length = 1.1 E6 m)
-- orange : NX=200 (domain length = 1.9 E6 m)
-- green : DX=5km (domain length = 0.6 E6 m)
-- red : DT=20s, DT_acoustic=5s (domain length = 1.1 E6 m)
-- purple : DX=3.159km, NX=361, DT=20s, DT_acoustic=20/3s (domain length = 1.1 E6 m)
- The autocorrelation plot is attached in figure 04_comparison_autocorrelation_dPdt. It shows that the frequency is modified for orange and green curves but is constant for the three other simulations where the domain length remains the same as the reference.
- I plotted the fft of the autocorrelation for these 5 simulations (05_fft_autocorr_dPdt) and then normalized the frequency with the domain length and the sound velocity (05_fft_normalized_pulsation). The position of the main peak collapses for all curves.

Conclusion
- The oscillation seems to be related to a sound wave oscillation through the whole domain. The normalized pulsation is on the order of 10^0 which means that the period is near the time needed for a sound wave to travel through the domain.

 

[MLandreau]

Additional note :
- I also tested the sensitivity to PBL+SL model, radiation timestep, LSM model, cumulus model but they didn't have any influence

Thank you,

Mathieu

 

[kwerner]

Hi Mathieu, I forwarded your question to our physics/dynamics expert and this is their reply.

I think these are very small and you really have to look closely at the pressure to see them. I doubt they appear in any other fields, or affect the meteorology. I think it is right these these are domain-wide sound waves. Due to the low frequency we don't damp them much in our acoustic damping. In other words, it is not harming the simulation and no changes need to be made.