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

Drastic change of result per timestep

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.

Saszalez

New member
Hi everyone. I did a few runs to extract the precipitation from a point for a month, and I found the following situation. If I don't modify anything except the adaptive time step (I have 3 nested domains, and I adjust the time step from 3dx to 5dx), the accumulated precipitation goes from 147mm to 448mm. I have repeated it several times, and the result is correct. I understand that there may be some variation when running faster, but so much is very strange to me. On the other hand, with the wind, for example, there is no variation. Any ideas?
 
Hi,
Such a dramatic change in precipitation is unexpected. Would you please clarify what variables are you looking at? How long did the model run before you look at the precipitation? What physics options did you use for this case? Which domain of precipitation are you looking at?

If the file is not too large, can you upload the wrfout files at the time you check but with time step of 3dx and 5dx, respectively?

It will also be helpful to know which version of WRF did you run and what data you used to drive the case. Thanks.
 
Hello.

I will give you the necessary information to handle the case.

- The WRF version is 4.1.
- I am using 3 nested domains.
- I am extracting the data from domain 3.
- The input data are those of the NCEP/NCAR Global Reanalysis Products database, 1948-continuing (ds0.90), and, more specifically, pgb.f00 and grb2d.
- The precipitation is obtained from the sum of the variables rainnc, rainc, and rainsh. I do it with NCL.
- Although I extracted the data from a point to a CSV, I also did it for other points in the domain, and the result, despite being a different amount of precipitation, has differences depending on the time step used.
- I did two tests, a short and a long one. The short runs from 6/15/20 to 8/1/20. The long one runs from 1/1/2018 to 4/1/21.
- In the short simulation, the difference is much more noticeable than in the long simulation. Rain is most evident in mid-August.
- In the following link, you will find the wrfout of the long and short simulation for time_step at 3dx and 5dx. You also have the NCL script, the namelist.input, and the CSV files with the precipitation data for each run and the same point. The link expires in a week; if you need more time, let me know. http://gofile.me/6AXXD/8KmNFAEI2
- The physics I used (I did tests with other physical parameters in mp_physics and cu_physics, and there is also the same problem):
Code:
&physics
mp_physics = 8,8,8,
cu_physics = 1,1,0,
ra_lw_physics = 1,1,1,
ra_sw_physics = 1,1,1,
bl_pbl_physics = 1,1,1,
sf_sfclay_physics = 1,1,1,
sf_surface_physics = 2,2,2,
bldt = 0,0,0,
cudt = 0,0,0,
radt = 10,10,10,
isfflx = 1,
icloud = 1,
ifsnow = 1,
surface_input_source = 1,
num_soil_layers = 4,
num_land_cat = 21,
sf_urban_physics = 0,0,0,
sf_ocean_physics = 1,
topo_wind = 1,1,1,
 
Saszalez said:

Hi, your domains are super-small and for such coarse resolution and step_to_output_time = .false., your point in time for output step is probably way different between two runs with different time steps so that you get very different amount of precipitation into bucket... IMHO.

So you might be able to fix that with step_to_output_time = .true.

If that is not reason then just try to set up more proper domain sizes and see if you still get different results. Any way you should create domains of at least 100x100 points and set more vertical levels (at least 35).
Ivan
 
Hello. I know that the domains, in this case, are small. I usually use at least 80x80 and upwards. But in this case, and due to the computational capacity that I have, it is necessary to make them smaller.

I will try to put more vertical levels. The point in time does not vary much despite having step_to_output_time = .false., As it is only a few seconds. That shouldn't make such a big difference in precipitation. I can still do the tests with that variable turned on.
 
Top