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

Increasing model simulation rate

This post was from a previous version of the WRF&MPAS-A Support Forum. New replies have been disabled and if you have follow up questions related to this post, then please start a new thread from the forum home page.


New member
Good day Mcurry
Please how can i make my simulation to run faster, for example, I want to make the temporal resolution to be a little coarser. currently, with the namelist.atmoshpere file I attached, I set my config_dt to 60, the simulation runs at every 1min but outputting files at every 1hr.
But I want the simulation to run at every 1hr and output at every 1hr, I set the config_dt to 3600, when it reaches the second run, It stop and exit with code Exit_statu = 139

I have tried different values of config_dt, like 1800(30min), 1200(20min), 600(10mn), it all stops on the second time step.
I understand the rule of thumb that config_dt should be 6sec times the minimum spatial resolution( in my case that is 10km) but that is too slow for me, it was going to spend more 5days to run just 10days of simulation, I can't envisage how long it will take to run 40years of simulations at such a run that is why I want to make or increase the config_dt.

please any help will be appreciated.

The 'config_dt' value specified in the namelist.atmosphere file is the step size used in the model's time integration scheme, and it cannot be chosen to be arbitrarily large due to numerical stability constraints. When config_dt is chosen to be too large, the model integration will become unstable, as you've seen when choosing a step size of 1800 s. The time integration scheme used in MPAS-Atmosphere is described in Klemp et al. (MWR, 2007), but if you're new to numerical atmospheric modeling, the Wikipedia article on Runge-Kutta methods might provide a better introduction.

So even though we typically save the simulated state of the atmosphere to output files every three or six hours, we still need to use a model time step of around 60 seconds (in the case of a mesh with a minimum grid spacing of 10 km).

The primary means of increasing the model simulation rate is through the use of more processors. For example, we at NCAR have run simulations on a 15-3 km mesh using 6144 processors in order to finish five-day simulations in a few wall-clock hours. Producing 40 years of simulation on a mesh with a minimum grid spacing of 10 km will certainly be computationally demanding. There are a more details on estimating the computational cost of simulation in this forum post.
Thank you so much, Mgduda.
For the 10days simulations I tested, I used 12cores each in 13nodes and it took more than 24hrs to simulate 2days and 59hr( at every minutes, 6s x 10) before it gets stopped because of cluster maintenance.
This is so helpful and understandable, thanks once again. I will read through all the links to know more and have more understanding about atmospheric modeling, as you have guessed right, I am new to atmospheric modeling, in fact this is my first simulation ever.

Thanks so much.