Looking for explanation about cycle behavior of adaptive time-step

[LluisFB]

Dear all,

This is my first time that I use adaptive time-step in WRF.

I am trying to run over Ecuador (South America) down to 2 km. See domain below:

I am forcing with ERA5. I have insane CFL problems to start the simulation so, I recall about adaptive time-step. So I thought to perform 1 simulation day with a lot of dynamical restrictions to get a first restart and then start a proper simulation from there. So in my namelist (I attached the full version) I have:

&domains
 smooth_option                       =     1,
 use_adaptive_time_step              = .true.,
 step_to_output_time                 = .true.
 target_cfl                          = 1.0, 1.0, 1.0,
 max_step_increase_pct               = 5, 51, 51,
 max_time_step                       = 20,
 min_time_step                       = 1,
 smooth_cg_topo                      = .false.
/


&dynamics
 hybrid_opt                          =      2,
 w_damping                           =      1,
 diff_opt                            =      2,      2,     2,
 km_opt                              =      4,      4,     4,
 diff_6th_opt                        =      0,      1,     0,
 diff_6th_factor                     =   0.12,   0.12,  0.12,
 diff_6th_slopeopt                   =      1,
 diff_6th_thresh                     =    0.1,    0.1,    0.1,
 base_temp                           =   290.,
 damp_opt                            =      3,
 zdamp                               =  4000.,  4000.,  4000.,
 dampcoef                            =    0.2,    0.2,    0.2,
 khdif                               =      0,      0,      0,
 kvdif                               =      0,      0,      0,
 epssm                               =    0.9,    0.9,
 non_hydrostatic                     = .true., .true., .true.,
 moist_adv_opt                       =      1,      1,      1,
 scalar_adv_opt                      =      1,      1,      1,
 gwd_opt                             =      1,      0,      0,
/

So far I managed to simulate, but then I realize that in the rsl.error.0000 appears de actual time step being used during the simulation as for example:

(...)
Timing for main (dt= 15.53): time 2018-11-24_01:37:53 on domain   1:   13.43241 elapsed seconds
Timing for main (dt=  8.15): time 2018-11-24_01:38:01 on domain   2:    3.40214 elapsed seconds
Timing for main (dt=  8.15): time 2018-11-24_01:38:09 on domain   2:    2.84357 elapsed seconds
Timing for main (dt= 16.31): time 2018-11-24_01:38:09 on domain   1:   16.08749 elapsed seconds
Timing for main (dt=  8.56): time 2018-11-24_01:38:17 on domain   2:    3.40707 elapsed seconds
Timing for main (dt=  8.56): time 2018-11-24_01:38:26 on domain   2:    3.69092 elapsed seconds
Timing for main (dt= 17.13): time 2018-11-24_01:38:26 on domain   1:   16.56759 elapsed seconds
Timing for main (dt=  8.99): time 2018-11-24_01:38:35 on domain   2:    3.73254 elapsed seconds

I realized that the time step fluctuates. Curious about that (I would thought that as times goes by it will become somewhat stable) I got the values for the first12 hours of simulation and I plot them as you can see below:

Fig2. Left panel time-step used to integrate each domain (d01: red, d02: green) during the first 12 hours. Right panel, zoom of the first half hour. Gray lines indicate calls to the radiative scheme (radt = 3,3),whereas blue lines indicate output written in file (history_interval = 180, 15,)

I checked namelist.output and does not seem to be any other parameter in the namelist that could explain this cyclic behavior of around 5 minutes.

Just to understand the behavior of the model, does anyone have any explanation for this?

Many thanks in advance,

Lluís

 

[kwerner]

Hi,
I'm attaching the WRF Technical Note. Please see section 3.3.3, which describes the details of the way the model handles the adaptive-time-stepping option.