The 3-kilometer NAM

This is a sample lesson page from the Certificate of Achievement in Weather Forecasting offered by the Penn State Department of Meteorology. Any questions about this program can be directed to: Steve Seman


By the end of this page, you should be able to describe the differences between the 3-km NAM and other convection-allowing models like the HRRR.


The Rapid Refresh (RR) and High-Resolution Rapid Refresh (HRRR) aren't the only "mesoscale models" available. The NAM, which you learned about in your previous studies, can also have some use in mesoscale forecasting.

That statement might seem odd since in the last section I lumped the NAM in with the GFS as models that have to parameterize (greatly oversimplify) convection. But, there's more going on with the NAM than what meets the eye! To show you how the NAM can be useful in mesoscale forecasting, first I need to provide a little background on the model itself.

On October 18, 2011, the NAM underwent a substantial upgrade. Although I won't go into details, the internal "guts" of the model changed dramatically. Of more importance to us as mesoscale forecasters, though, is the fact that the model change introduced "one-way" smaller nests within the larger outer model domain. Within each nest, the new model computed forecasts concurrently with the 12-km NAM parent run. For the record, "one-way nested" means that the inner (nested) model domain receives its lateral boundary conditions from the outer domain, but it does not feed back any information to the outer domain. In other words, the outer domain is not affected by the nest.

Map showing the parent domain of the NAM and its nests.
The parent 12-kilometer domain of the NAM, along with its three-kilometer nests for CONUS, Alaska, and Hawaii / Puerto Rico, respectively. The smallest rectangles represent very high-resolution nests for predicting fire weather.
Credit: National Centers for Environmental Prediction

The nested domains within the parent NAM have higher resolutions, and while initially they had resolutions that ranged from three to six kilometers, in 2017 the model was upgraded again, improving resolution to three kilometers for the nests covering the contiguous U.S., Alaska, Hawaii, and Puerto Rico (as shown above). The resolution of the internal nests of the NAM is sufficiently high to realistically simulate convection, so while convection is parameterized in runs of the parent 12-km NAM, it's not in the higher-resolution forecast nests. In case you're wondering, the small unlabeled boxes in the image above represent small nests with even higher resolution that are used for predicting fire weather.

So, is the 3-km NAM every bit as useful as the HRRR? Not exactly. There's a key difference between the two. While the HRRR is initialized every hour, the 3-km NAM is still only initialized every six hours (06Z, 12Z, 18Z, and 00Z). While the 3-km NAM does have forecast intervals of one hour, it does not get infused with hourly surface observations, which makes it less viable for predicting the small-scale rapidly changing environments that may favor the initiation of thunderstorms.

48-hour forecast of radar reflectivity from the 4-km NAM
The 10-hour forecast of radar reflectivity valid at 22Z on May 30, 2017 (from the run initialized at 12Z on May 30). Since the 3-km nest of the NAM does not parameterize convection, its forecasts include realistic-looking convective structures.
Credit: Penn State University

While the 3-km NAM produces forecasts with realistic-looking convective structures (like in the example above), the same caveats that went along with HRRR forecasts apply. Just because the forecasts look realistic doesn't mean they're accurate, and remember, the fact that the 3-km NAM is only initialized every six hours is a notable drawback. On the flip side, one advantage to the 3-km NAM is that its forecasts go out a few days into the future, whereas those from the RR and HRRR only go out a day or less. Like with the HRRR, the timing and exact location of individual thunderstorms is often incorrect in 3-km NAM forecasts, but it can still give useful insights into the general coverage and structure of thunderstorms.

If you're interested in accessing forecasts from the 3-km NAM, check out the Explore Further section below for a guide to the products on the Penn State e-Wall. Otherwise, we'll wrap up our introduction to mesoscale meteorology with a brief Case Study of a tornado outbreak, which illustrates the connections between spatial scales and the utility of real-time mesoscale model analyses. Read on.

Explore Further...

Key Data Resource

With the background on the 3-km NAM under your belt, let me introduce the PSU e-Wall page high-resolution models page that gives you the 00Z, 06Z, 12Z, and 18Z runs of 3-km CONUS nests.

This page contains forecasts from the 3-km NAM that essentially cover the eastern two-thirds of the United States for a number of variables, including radar reflectivity / cloud cover, 2-meter temperatures, MSL pressure / clouds / 1000-to-500-mb thickness / precipitation type, 10-meter winds, accumulated precipitation, and 850-mb temperatures / winds.

But, perhaps the most powerful tools on the e-Wall page are the 00Z and 12Z comparisons of a 4-kilometer version of the NAM (formally the "NMM," which stands for "Nonhydrostatic Mesoscale Model") and the 4-kilometer ARW. For the record, the ARW (Advanced Research WRF) is a research version of the Weather Research and Forecasting model framework (which ran as the NAM before 2011).

At any rate, NCEP runs what they call the "high-resolution window" for the 4-kilometer NAM and 4-km ARW models for the eastern two thirds of the contiguous states at 00Z and 12Z. On the e-Wall, each map represents a side-by-side comparison of fields from the two different models. Comparing output from these two mesoscale models can be helpful because it allows you to see what degree of consensus exists in the prediction of mesoscale features (in terms of initiation, timing, location, structure, etc.).

For winter mesoscale forecasting, you can also find links to the 48-hour forecasts of a 10:1-ratio snow accumulation from 4-km NMM and ARW. Essentially, these products show a total accumulation of the liquid equivalent when the model precipitation type is snow. Values on the map are further multiplied by 10, so the color associated with '60' (dark purple) is 6 to 8 inches of 10:1 snowfall, for example. I suggest using these maps with extreme caution, however. While a 10:1 snow-liquid ratio can be thought of as an average, it doesn't work very well in many forecasting scenarios. Still, these maps can be helpful in identifying the existence of bands of heavy snow (even if their location isn't quite right).

Finally, the high-resolution model page on the e-Wall also contains its own HRRR forecast maps, including forecasts of radar reflectivity / cloud cover, 2-meter temperatures, MSL pressure / clouds / precipitation type, 10-meter winds, total snowfall, and accumulated precipitation.