Thursday, December 16, 2010


Over the months, I get lots of questions about the various models local forecasters use...and considering that the weather is relatively benign right now....perhaps this is a good time to talk about them.

Carpenters have hammers and power saws, plumbers have wrenches, and meteorologists have computer models. Now these are not physical models of course, but rather simulation models.
The equations that describe the atmosphere are programmed on digital computers--that is the model. We start with a 3D description of the atmosphere based on observations, known at the initialization, and then the model can integrate the equations forward in time to provide the future state of the atmosphere in 3D. The equations are generally solved on a 3D array of grid points and the distance between them is generally termed resolution. A 12-km resolution model has the horizontal grid points separated by 12 km. These models are some of the most complicated constructs of our species...encompassing hundreds of thousands to millions of lines of code...and represent physical processes from the microscopic to the planetary. The most powerful computers of the planet are used to run the models. Be impressed, very impressed.

Now at a typical NWS weather service office, the forecasters have access to a dozen or more models. Models with different resolution, different physical descriptions, different ways of solving the equations, different approaches to initialization and run at different modeling centers. So a meteorologist is like a carpenter with a dozen or more hammers, each different, each better in some situations than others.

The National Weather Service runs two major models: The Global Forecast System Model (GFS), which runs over the entire globe at a resolution of roughly 35km, and NAM model (also called NMM-WRF) which runs at 12-km resolution over North America and adjacent oceanic areas. The best global model is NOT American--a terrible and unnecessary embarrassment for our nation--- and is run by the European Center for Medium Range Weather Forecasting. The UK Met Office has an excellent global model as well. The Navy runs a global model (NOGAPS) that is second tier, but useful, and has a very good regional model called COAMPS.

The research community mainly uses a model called WRF (Weather Research and Forecasting Model) developed at the National Center for Atmospheric Research. This is a state-of-the-art high resolution weather prediction system with all the bells and whistles. At the UW we run WRF operationally at 36, 12, 4, and 1.3 km resolution and our colleagues at the National Weather Service use it heavily. This is the model output you see most frequently on this blog

One of the most important tasks of an operational forecaster is to look at this palette of model forecasts and make a decision of what they think will be the most probable evolution of the atmosphere. Which models have been good lately? Which has done best in the past for similar situations? And the forecast can use the variability of the model forecasts to reach conclusions of the possible future evolutions and which are most likely to occur.

And there is MUCH, MUCH more. The model output can often be improved by statistical post-processing. For example, if a model is typically too cool at certain hours, you could alter the output with the average error. And I haven't even talked about the newest prediction technology....ensemble which we run the model tens or hundreds of times with slightly different initializations and physics to get probabilities of what will happen in the future.

This is a highly sophisticated technology--and I am highly optimistic for future improvements. And forecast HAS gotten much better over the past decades.

So next time you are thinking about making fun of meteorologists or laugh at the classic jokes (you guys get paid for forecasting 50%!, meteorologists have a pair of dice in the backroom, I wish I could get paid for being wrong all the time, etc), bite your lip!


Michael DeMarco said...

Cliff, your discussion of modeling and the personalities of the various models is appreciated. My question is, given the wide diversity of our regional geography and topography, when does the reliance on models transition to the inclusion of current meteorological data? When do you use the "old fashioned" readings. Being a longtime resident of the area when does your "gut" get your attention? Does this ever happen anymore? Thanks.

Avalanche said...

I just wanna give a shout out tonight to the TCWB MM5, aka ole reliable :P

Now more serious, I dunno, maybe its used, but it looks like UW doesn't incorporate the EURO in any computations. Is there some type of copyright or licensing trademark on it?

I get the sense too that the NWS and in the discussions, its like, "we're going to trust what our models show, however the EURO shows this." Always as a double check, confirmation thing, lol.

WX said...

Are the 1.3km wrf results plotted and available? If not, are the wrfout_d02_yyyy-mm-dd_hh:00:00 output files accessible to plot from?


for what I'd like to be able to use them for instead of running wrf myself.

WX said...

Are the 1.3km wrf results plotted and available? If not, are the wrfout_d02_yyyy-mm-dd_hh:00:00 output files accessible to plot from?


for what I'd like to be able to use them for instead of running wrf myself.

WX said...

Ah while we are talking about models, have you or anyone you know substituted real time snow cover data for the monthly average that wrf uses as standard?

How about using higher than 30s resolution topography. I noted the other day that in the 30s data, Mt. Baker is less than 7000ft high, compared to its actual 10,800.


Robert Okrie said...

Curious about how much faster model "runs" are accomplished now compared to the past vs number of data points processed, etc. Also, who physically runs these? Public, private or cloud services (an ultimate pun)...?

Connie said...

Meteorologists do it with models.

Joel Levin said...

Are the models self-impoving based on verification of their predictions?

jj said...

Alright, how about this: You guys get paid to forecast weather from a nice warm office, whereas Jim Foreman has to be right in the Middle of it!

Yes, I know, the obvious comeback is "but we don't get the nice Jacket."

Kenna Wickman said...


What do the models say about Monday night - in terms of viewing the total lunar eclipse?


catman306 said...

Back in the day, my Fortran instructor said she worked for NASA on code for Apollo: Half a million lines of Fortran.

Someone else can compare the complexity.

Scherer Family said...

Hey Cliff, are we going to be able to see the confluence of the total lunar eclipse that is happening on the winter solstice for the first time in 400+ years this upcoming Tuesday?

Mainly interested in Western WA… or should we make plans to head over the mountains?

Can’t get much darker outside than that!

Anonymous said...

Changing the subject. . . why is it that sometimes the frost on my car is the whorled, patterned Jack Frost type, but other times (most typically) the frost is just a thin solid layer with no designs?

Craig said...

Just wanted to tell you Cliff, I just relived the 2 weeks of snow from two years ago on your blog. This is how bad I need a shift in weather patterns from "normal" December weather to Snomageddon 2010.

Anonymous said...

Good info. Thanks. I think areas over 9000 feet out west will see close to 15 feet of snow by Monday. To find out more on my thoughts you can check out my website...

j.Konovsky said...

Cliff--I use Sailflow and it has a couple of other models for wind speeds. They are CMC and WRAMS, plus NAM and GFS and something they call default. The predictions are always wildly divergent and in my simple view of things, I always wonder which one is the most reliable. Any suggestions?

kosmo said...

Seems the models got last nights wind storm wrong. If I understand, it was to be a traditional pass driven wind event, with pressure looking to balance between eastern/western WA, where the lowlands are not impacted. So, power out in Federal Way makes for a forecast miss as I remember the forecast zone to be east county areas, N. Bend, Enumclaw. What happened?