Best practice in term of spatial resolution

[balisto]

Hello,

I am using WRF with data from Era5 ECMWF that have a spatial resolution of 0.25deg in longitude and latitude. My objective is to get a spatial resolution of 1kmx1km so approximately 0.01 degree in longitude and latitude.
The final area that I want to obtain is quiet small (~30x30km).

I run WRF with DX=DY=0.25degree in the namelist.wps file and then applied several nested domains inside each others with a parent grid ratio of 3 in order to obtained my final resolution of 0.01degree.

My questions are : Should I use a resolution of 0.25degree in the file namelist.wps because it is my input data resolution or can I directly use my final resolution of 0.01degree (and then not use any nested domains)? Is there a difference between the two methods ?

If I am interested in an area of 30x30km but the input data have a resoltution ok 27x27km, how much bigger should be the input area in order to have accurate result? Is it better to consider a bigger area around ?

Thank you very much for your help.
Kind regards

Bastien

 

[kwerner]

Hi Bastien,
Here are some resources that may be helpful to you:

Best Practices WPS Namelist Page: https://www2.mmm.ucar.edu/wrf/users/namelist_best_prac_wps.html
This page will describe the common WPS namelist variables and describe recommended practices for each.

Nesting Presentation given at the WRF Tutorial: https://www2.mmm.ucar.edu/wrf/users/tutorial/202001/werner_nesting.pdf
This presentation will discuss how to set up nests appropriately.

To specifically answer your questions, no you do not need to use 0.25 degrees in the namelist. As that is the input data, you can think of that data and global domain as the boundaries for your first domain (kind of like a nest to the global data). For this initial domain, you'll want to have a resolution ratio of no more than about 7:1. You cannot just immediately jump down to 1km for the first domain because that is too large of a ratio and the boundaries will be very noisy. Within the namelist, we recommend ratios of 5:1 or 3:1, so it would make sense to have a set-up of:

dx = 9000, 3000, 1000
grid_ratio = 1, 3

or

dx = 5000, 1000
grid_ratio = 1, 5

 

[balisto]

Thank you kwerner for your answer. It is much more clear to me now.

 

[ladyvader]

Hi Bastien,
Here are some resources that may be helpful to you:

Best Practices WPS Namelist Page: https://www2.mmm.ucar.edu/wrf/users/namelist_best_prac_wps.html
This page will describe the common WPS namelist variables and describe recommended practices for each.

Nesting Presentation given at the WRF Tutorial: https://www2.mmm.ucar.edu/wrf/users/tutorial/202001/werner_nesting.pdf
This presentation will discuss how to set up nests appropriately.

To specifically answer your questions, no you do not need to use 0.25 degrees in the namelist. As that is the input data, you can think of that data and global domain as the boundaries for your first domain (kind of like a nest to the global data). For this initial domain, you'll want to have a resolution ratio of no more than about 7:1. You cannot just immediately jump down to 1km for the first domain because that is too large of a ratio and the boundaries will be very noisy. Within the namelist, we recommend ratios of 5:1 or 3:1, so it would make sense to have a set-up of:

dx = 9000, 3000, 1000
grid_ratio = 1, 3

or

dx = 5000, 1000
grid_ratio = 1, 5

Hi kwerner,

I did not want to open another topic, since this is quite similar to the question I have. Well, it's actually 2 questions...

1) I understand the answer of starting the domain in 9km to reach the finest of 1km. However, I found a paper (https://rmets.onlinelibrary.wiley.com/doi/full/10.1002/joc.6686) where they mention that starting directly with 10km (in the case that they deal with), causes distortion in the large-scale dynamics and even a 30km domain is necessary as the parent domain (all this with ERA5). This is the only paper I found and it is a bit recent, I don't know what valid it will have. I did not find much more information about it.

I also found a case study (https://nar.ucar.edu/2019/ral/monkeypox-study) where the parent domain starts at 12km and to help to prevent the model from “drifting” from the observed state, spectral nudging is used at the top of the atmosphere (12 levels higher).

I am interested in a simulation of ~ 15 years, using ERA5 focused on a 1km nested domain. What would be the best way to approach my simulation?

2) I also understand the relationship of the domains 1:3 and 1:5, however I have never seen them mix. For example, start at 15km, in the second domain use 3km and in the final domain 1km. is there any reason not to do the simulations this way?

I'm sorry if the questions I have are very obvious, or if I should have opened another topic for them.

kind regards

 

[kwerner]

@ladyvader,

1) Does the paper you refer to mention the resolution of input data from ERA5? I'm not as familiar with that data, so I'm not sure if there are different options for resolution. It seems silly to need a parent with 30km grid-spacing, if the input data are a finer resolution than that (e.g., 0.25 degree). You do not have to start with 9km. You can start larger if you want. You could test out a few options and run some quick/small tests to see what you think will give you the most realistic results.

As for spectral nudging, it may be an option you will want to consider. Take a look at this presentation on nudging options. It discusses several different nudging options, but if you scroll through a bit, you'll find information on spectral nudging, in particular.

2) Yes, you can have different resolution ratios between domains. It would be okay to have a 5:1 and then a 3:1, if that helps your case.

 

[ladyvader]

@ladyvader,

1) Does the paper you refer to mention the resolution of input data from ERA5? I'm not as familiar with that data, so I'm not sure if there are different options for resolution. It seems silly to need a parent with 30km grid-spacing, if the input data are a finer resolution than that (e.g., 0.25 degree). You do not have to start with 9km. You can start larger if you want. You could test out a few options and run some quick/small tests to see what you think will give you the most realistic results.

As for spectral nudging, it may be an option you will want to consider. Take a look at this presentation on nudging options. It discusses several different nudging options, but if you scroll through a bit, you'll find information on spectral nudging, in particular.

2) Yes, you can have different resolution ratios between domains. It would be okay to have a 5:1 and then a 3:1, if that helps your case.

Hi,

yes, they use the data ERA5 (0.25°, hourly). It seemed strange to me, that's why I came to ask here better to get out of the doubt. Anyway, I'm going to try as you said at the beginning of the topic with 9km, 3km and 1km, and I'll add the spectral nudging only on the largest domain.

thank you very much Kelly!!