Dynamic Downscaling of Wind Events - Sugestions?

Forum dedicated to older versions of EMS package (WRFEMS v3.2, v3.1 or older). Support is user-to-user based, so please help others if you can.
Post Reply
Posts: 9
Joined: Fri Nov 09, 2012 3:59 pm

Dynamic Downscaling of Wind Events - Sugestions?

Post by WxJAK » Tue Nov 27, 2012 6:28 pm


I am working to produce as accurate as possible surface (10m) windfield as possible for a number of wind events in the VA / NC area of the eastern US, including baroclinic events such as strong post frontal winds and nor’easters, as well as landfalling tropical cyclones. I am currently tinkering with EMS WRF using the NARR for my boundary conditions and an ARW (LC) three deep nest (21km (200x200), 7km (283x283), 2.333 (409x409)) and 55 vert. layers. I turned on two way nesting and used Yonsei PBL, NOAH sfc with Multi-layer (BEP scheme (Alberto Martilli)) activated for the 2.33km nest, and used the New Thompson graupel MP scheme. For CU param. I hesitantly tried KF at 21km, Grell 3D at 7km, and none at 2.33. First results for Irene’s landfall look decent. I am considering tinkering with an ARW (LC) nest of 15km (200x200) and 3km (471x471) with similar physics, save that I would start off trying Grell 3D, none for my partent, child domain CU param.

I would welcome any thoughts, experiences from similar endeavors, suggestions, etc. Is 55 layers overkill for my ends? I have been trying to keep my outer domain extent well over 1000 km, tending toward 3000 km. Is that reasonable for a dynamic downscaling effort, or would a smaller extent suffice? Are that any reanalysis datasets that would be expected to be superior to the NARR for my purposes (ERA Interim)?

Cheers and my thanks in advance for any advice,

-- Jacob

Posts: 173
Joined: Sun Jun 24, 2012 8:42 pm
Location: Los Angeles, CA, USA

Re: Dynamic Downscaling of Wind Events - Sugestions?

Post by pattim » Fri Dec 28, 2012 5:36 pm

Hi Jacob: I have read that the best solutions to any FD/FV calculation (WRF is FD), if you ignore grid solver stability issues, are when the aspect size of grid cells is roughly uniform. So you would like the number of vertical levels to give you roughly the same horizontal and vertical dimensions (but never so coarse as to ignore important atmospheric regions). I have also heard both opinions: that 2-way nesting is bad and 2-way nesting is good. Warner's "Numerical Weather and Climate Prediction" (2011) states pretty strongly that domain boundaries are a major source of error. Every time you introduce a new nest, you introduce a new set of boundaries. The only known fix seems to be to locate the boundaries far away from the zone of interest. As the "reanalysis fields" which you download get better, this may be less of a concern. The idea is to run a bunch of experiments and see if unphysical fields emerge at the boundaries (e.g., shocks) and propagate into the domain. This is messed-up dynamics from the (unphysical) boundaries.

So I think one-way nesting does the best at fixing this. If I were to use nests again, I would use one-way nesting and "nudge" the larger domains toward the "reanalysis fields" I've downloaded so that the finest nest (the one of interest) gets the best possible boundary conditions at every timestep. (At boundaries of non-nested domains, all WRF gets are the 6 hour reanalysis field data.)

As far as forecasting strictly into the future, I'm not sure how to do that (I've done only re-forecasting). In that case I'm not sure how WRF handles the boundary conditions (BC's). The two ways to handle BC's which I can think of (from my readings) are [1] downscaling BC's from a global climate model (such as ModelE or GEOS5), or [2] doing a coarse Global-WRF simulation with "nudging," and one-way (?), non-nudged nesting to a small region of interest. Again, as soon as you run WRF into the future, you can't "nudge" any more, but at least you have a dynamically-consistent starting field, so the global forecasts should be reasonable to at least a few days to a week or so, and these will feed the BC's of the nested region very well.

Let us know how it goes. Especially, if Robert (or any other experts :-) want to comment - much appreciated!! In particular, I believe nudging will be included in the coming version of EMS, so that will help implement the scheme [2] I mentioned above. If anyone has a good scheme to forecast into the strict future (how to handle BC's and set up EMS) please share!! Also, opinions on my scheme [2] above are appreciated.

Happy Holidays!!

patricia :D

Post Reply