Cold pools and forecast philosophy All - Home - RTMA/URMA Discussion Group

Some common sense: Complex terrain areas such as Creede and Westcliffe suffer from complications due to terrain in that models or analysis systems are likely to miss or misrepresent the effects of large terrain "variation" on dynamics (winds) and thermodynamics (temperature) plus moisture, which are all inter-coupled; and, data measurements are much harder and more unreliable.

Note that there are always "subgrid-scale" physical/dynamical features, which are missed in numerical models (either explicitly or through parameterization), both in horizontal and vertical dimensions. Highly inhomogeneous, anisotropic, and non-stationary flows (in 3-D) complicate the stagnation of flow in valleys and/or diversion over/around crests, which are closely linked to "cold pools".

UNLESS these subgrid-scale features are systematically represented in modeling (simulated or parameterized), obs. data pre-processing (bias correction and quality control), data assimilation, downscaling, or whatever, the wind/temperature bias issues won't be solved systematically.

It is not whether we have too little or much data and from what kind of sources we get data from (e.g., METAR or mesonet), but how we sort out garbage from a mixed set of good and bad data.

That's the key rationale behind our CaRDS request (sorry if you're tired of me talking about this):

https://docs.google.com/document/d/1sX8JZ0-0vNsSVLTzuC_7k-crVKSeArsimXiOu00utM4/edit

We're planning to write requirements based on this, which are scientific and general enough to address some key problems in mesoscale analysis with focus on terrain variation effects, especially on flow around complex terrain. Then, NBM’s blending will improve through more trustworthy URMA. 

yj kim

NWS/AFS11 Chief

Hey Paul-

Your CWA (and ABQ) are some of the places where the URMA struggles the most, in my opinion. You can see in the first image below that the southern Rockies harbor some of the biggest deviations between the NDFD nowcast and URMA for MinT (left side). Angel Fire, NM might be the ultimate, not surprisingly, but the Creede and Westcliffe areas show up well. There are also some of the larger corrections from the background (right side) in that area, although they are much smaller than the NDFDNow differences from URMA. Keep in mind that this is data for the month of December, so it probably included both URMA v2.6 and v2.7 information. I expect that once February data rolls in the left ide biases will improve substantially.

I have found it surprising that these relatively wide basins with multiple corroborating observations that URMA sticks so close to the background. URMA already is sensitive to terrain complexity such that it weights the obs more heavily in complex terrain, which has improved its performance significantly in Western Region since v2.7 came online in December. I wonder if valleys like the San Luis (or even the one Westcliffe is in) are sufficiently wide that they might as well be Kansas to the URMA. 

In your ppt you also talk about how this should relate to forecasting. One other thing that is surprising to me is that that at KALS, the NBM is actually not too bad compared to NDFD in forecast MAE (second plot below). I assume that your office uses MatchObsAll and therefore you are bias correcting to an Obs database that directly reflects the observed temps in Alamosa.  Despite being severely handicapped by the URMA deviations from the KALS obs, the NBM is still competitive forecasting the mimimum temperatures there! In the bias plots you can see that NBM has relatively constant warm bias, while NDFD starts out unbiased but is just as bad as NBM beyond day 3.  Switching the URMA and NBM might now degrade the forecast as much as you expect, at least during mid-winter cold-pool season. As an aside, apparently MOS is just really good at KALS. 

-Matt

Resurrecting an old thread here. URMA/RTMA developrs: do you know if Matt's hypothesis is correct, i.e. whether valleys can be wide enough that the RTMA system introduced in 2.7 is not identifying it as a mountain valley? I can't remember what the system is called but the idea was if a pixel is in a mountain valley an ob can pull the analysis farther from the background, thus allowing RTMA to better analyze cold pools.  

I ask because I noticed the Uinta Basin of Utah in January/February we had multiple days where the URMA analysis was likely not as cold as reality, and this happens to be a wide basin where it stands to reason that it could be errantly treated as "Kansas" instead of a mountain valley prone to cold pools.  I tried to grab some example images for this post but the Blend Viewer page seems to have already purged the Analysis Review from that time frame.