Long run with no crash, but bad results - insights?

[bkbainy]

Hello,

I am running a 5 year simulation for a region in Northeast Brazil. I'm using FNL 0.25deg as inicial and boundary data and ERA5 for SST updates and using spectral nudging for the coarse domain. I've set 2 domains, 9 and 3 km resolution. All the other options in namelist.input are pretty much usual. I took lots of tips from other posts here. I did change SST lines on METGRID.TBL following https://dreambooker.site/2017/12/20/Initializing-the-WRF-model-with-ECMWF-ERA-Interim-reanalysis/ because I was getting the error "grid%tsk unreasonable" while running ./real - and it solved.
I've simulated about 3 months so far, and as I was comparing some data with observation at two points within the finer domain, I got very good results in one of them (Poço Verde) and bad results in the other (Brejo Grande). Any ideas why? I'm not an experienced modeller, so I'm wondering wether I can actually improve my results. My goal is to obtain reliable data for another point (Japaratuba).
I do not believe changing parameterizations would improve significantly my results.
I attached my domains, terrain plot, weather stations over a map, namelists (.wps and .input) and plots for T and TD for the 2 stations.

I appreciate any insight, solutions or ideas.

 

[bkbainy]

New info:

Since my suspicions for my problem are mostly regarding either nudging, SST or topography, I checked on my wrfout files the variables SST and SST_INPUT (which is supposed to be data from wrflowinp).
- When checking ncdump -v SST wrfout_d02_2016-01-06_00\:00\:00, I got reasonable values (around 300 K) and zeroes (which I believe are for the land points)
- When checking ncdump -v SST_INPUT wrfout_d02_2016-01-06_00\:00\:00, all I got were zeroes. Could that be my problem? Or in case of zeroes (absent data) the model would run just as if I didn't have any extra SST information?

 

[kwerner]

Hi,
I honestly am not sure how the SST_INPUT variable works, but I have a case that has reasonable results and SST_INPUT is 0 everywhere, as well, so I don't *think* that's causing a problem. Out of curiosity, does your other case/location have values for that variable in the wrfout* files?

As for the bad results, unfortunately there is no secret ingredient to making the simulation output match the observations. It can sometimes just take a bit of playing around with some settings. These are the things I would try, but I don't know if my suggestions will give you the desired outcome:
1) Using nudging for the entire simulation is typically useful when the size of your domain is about the size of a continent (e.g., a CONUS domain), but your domain is pretty small, and therefore it's probably not necessary to use nudging for the entire simulation, though I don't know that it would hurt to do so. You may want to just try a simulation where you only use it near the beginning - just to see if the results differ.
2) As I mentioned, your domains are pretty small. We recommend using domains with a size of at least 100x100, and yours meet that requirement, but not much over. I would especially increase the size of d02 some. It doesn't need to be huge, but I think it may help to have the boundaries of that domain further away from the station of interest. Just remember that you want d02 to be about 1/3 the distance on all sides from d01, so if you make d02 much larger, you may want to also increase d01 a bit. Right now, you have plenty of room for an increase in d02 size.
3) The only physics modification I would recommend trying is perhaps to use NoahMP for surface.

 

[bkbainy]

Hi,

Thanks for your reply.
I was testing the same case but with ERA5 data only, including SST update, and the SST_INPUT variable in the wrfout* files was still only zerores, but with reasonable values in the SST variable... I haven't checked the output yet.
I'm running a test now, increased D02 from 115 to 169 points each way and with the surface scheme physics you suggested. I excluded nudging from this test.