Huge cold bias on snow covered and very complex terrain with Noah LSM and ACM2 PBL

[emaiorana]

Hi, I'm running some simulations using WRF V4.0.3 on Northern Italy. We use ECMWF data initialization and the domain resolution is 4km (in reality we run a 2.2km horizontal resolution domain all over Italy for operational use, we made those 4km simulations only for testing the impact of pbl and lsm physics on certain conditions). We use Corine Landuse and Aster Gdem v2 orography.
When using ACM2 PBL with Noah LSM togheter, we find a HUGE cold bias during night time on 2m temperatures on Alps. It seems to be related to snow depth init from ECMWF, which appears too coarse and high.





Above are shown 2 simulations, one with snow from ecmwf init, the other with snowh, snowc and snow setted to 0 in wrfinput file from real. The huge bias disappears, while the good performances on the Po Valley with nighttime temperature inversion are not touched.






If we run a third simulation using Noah-MP LSM with ACM2, the snow cover on Alps is much better tratted respect to the simulation with Noah LSM and Snow from Ecmwf in wrfinput... but on the other hand, we lose the good performance on temperature inversion in the Po Valley. We also observed better performances on maximum 2m temperatures on Po Valley when using Noah LSM and ACM2.

The run with ACM2 PBL and Noah LSM appears very unstable too, for sure because of the snow tratment on Alps (often the run without snow in wrf input end successfully while the run with ecmwf snow depth init data fails).

Has anyone encountered this kind of problem? Is it possible to treat albedo better on complex orography using Noah LSM? Thank you!




PS: That is the snow depth at t +0 from ecmwf init data. Those values are unrealistically high and spreaded across valleys where there is not snow at all. This influences the 2m temperatures on a large number of grid points (the effect is reported also in the 2.2km horizontal resolution domain)

 

[kwerner]

Hi,
Can you attach the namelist.input file you're using for these runs?

Thanks!

 

[emaiorana]

Hi, here is the namelist.input

 

[kwerner]

Thank you for sending that. I have a couple of suggestions that you could consider:

1) It may be that your first model level (eta_levels) is too close to the ground. You have to start with 1.0000, but then your next one is 0.9980, which is very close. You could keep running with 51 levels, but as a test, you may want to consider shifting the first layer up to something closer to 0.9950.

2) If there aren't better results with 1), you could try using the YSU PBL scheme instead of ACM2 to see if that makes any difference.

Unfortunately surface schemes are extremely complicated, and there isn't always a straight-forward solution, as there are so many different types of terrain/climates, and the interaction between schemes can vary so drastically. Using the Noah-MP scheme (vs. Noah) is likely going to fix your initial problem if there is snow on the ground (and I assume there is in that location), so it makes sense that you would see improvement with the cold bias. I would recommend just experimenting with various physics combinations (e.g., 2) above, and/or perhaps switching the sfclay scheme). If nothing seems to improve both the fields you have mentioned, you unfortunately may have to choose which is the most important to your application. Let us know if you find something that helps, though, as this could benefit someone in the future, with a similar simulation.

Kelly

 

[emaiorana]

Hi, we found that the major problem is caused by unreal amounts of snow on ECMWF init data on Alps, even at low altitudide (Aosta, for example). For our real time runs we decided to "cut" snow amounts under a certain altitude, that solved must of the cold bias on alps. Thank you!

 

[kwerner]

Hi,
I am glad to hear that you were able to solve the problem. Thank you for updating the post!

 

[Xun_Wang]

Hi, we found that the major problem is caused by unreal amounts of snow on ECMWF init data on Alps, even at low altitudide (Aosta, for example). For our real time runs we decided to "cut" snow amounts under a certain altitude, that solved must of the cold bias on alps. Thank you!

Hi,

I am running WRF using ERA5 from ECMWF and have the same problem: unreal snow depth causes day time cold bias especially in winter. I want to ask how do you "cut" the snow amounts? Do you set the snow amounts to zero under certain altitude? Or do you set it to a certain number?

Thanks and cheers!
Xun

 

[emaiorana]

Hi, we found that the major problem is caused by unreal amounts of snow on ECMWF init data on Alps, even at low altitudide (Aosta, for example). For our real time runs we decided to "cut" snow amounts under a certain altitude, that solved must of the cold bias on alps. Thank you!

Hi,

I am running WRF using ERA5 from ECMWF and have the same problem: unreal snow depth causes day time cold bias especially in winter. I want to ask how do you "cut" the snow amounts? Do you set the snow amounts to zero under certain altitude? Or do you set it to a certain number?

Thanks and cheers!
Xun

Sorry for the late reply! I "solved" it by setting (using nco) the SNOW* fields from wrfinput_d01 to 0 under certain conditions like hgt<800m. Anyway, this is not the best way to resolve the problem, but for sure the easiest, we're still investigating on that