Texas Emergency Management ONLINE 2016 Vol. 63 No. 1

What You Need to Know About the Use of Numerical Weather Prediction Models

Introduction of Numerical Weather Prediction and its history
By Troy Kimmell, Jr.

Born in the 1950s with the advent of computer technology—crude compared to today's standards—forecasters began to use computer models to predict the state of the atmosphere into the future. In short, a computer-generated weather forecast. The advancement of Numerical Weather Prediction (NWP) models since then is nothing short of amazing. Models are now able to make reasonably accurate forecasts a week in advance, an unthinkable occurrence just a decade ago. Computer models are now even used in making long-term seasonal and climate forecasts. Although used heavily by today's forecasters, these computer models are not perfect. Each model has its own biases, weaknesses and strengths. Meteorologists and weather forecasters must know these model traits inside and out before using them to make forecasts.

Weather Prediction Model
Weather Prediction Model

There are five global-scale NWP models used on a daily basis by meteorologists here in the United States:

  • National Weather Service Global Forecast System (GFS)
  • European Center for Medium-Range Weather Forecasts (ECMWF)
  • Global Ensemble Forecast System (GEFS)
  • United Kingdom Meteorological Agency (UKMET)
  • U.S. Navy Operational Global Atmospheric Prediction System (NOGAPS)

There are other higher resolution mesoscale models, including the North American Mesoscale (NAM) and High-Resolution Rapid Refresh (HRRR) models as well as specialized models, such as specific hurricane models. These models produce many types of forecasts ranging from surface pressure and precipitation forecasts to upper air jet stream and vorticity (atmospheric spin) forecasts. Many of the global scale models also have ensembles where atmospheric conditions are slightly tweaked and changed each time before the model is processed by the computer. For example, the European Center Model (ECMWF) has over 50 different ensemble members (solutions) while the GFS model has over 30 of these ensemble members. Meteorologists look at these ensemble data sets to see if the individual ensemble members are similar or different. Similar forecasts would indicate higher confidence in the forecast model. Large differences in the ensembles would indicate a high level of uncertainty in the forecast output. In addition, model output statistics (MOS) are created for specific forecast locations.

Many smart phone apps are heavily reliant on MOS data for a given location, such as city or zip code, so it is important to remember that the MOS data will be inaccurate if the model is not performing well! These models and their associated output are prepared once every six hours for global models and as frequently as hourly in the shorter range high resolution mesoscale models. There are tremendous amounts of NWP data streaming into smart phone apps every hour of every day. It’s important to understand how these data are generated to be able to use them wisely.

The fine print
With the widespread use of the Internet, just about everyone has access to NWP forecasts. Even with models that meteorological agencies consider proprietary in nature, like the ECMWF, basic data are freely available. Most of us have seen a local meteorologist show temperature and precipitation trends for their viewing area over the next 24 to 48 hours and even out to seven to ten days. These data are from models.

In addition, most everyone has access to specialized model solutions for hurricane development, strength and forward track. In short, all of this publicly available model information has made anyone with a smart phone an armchair meteorologist, but without having the necessary knowledge of the biases, strengths and weaknesses of the respective models being used. One of my Texas A&M University meteorology professors, a number of years ago, told the class that perhaps one of the most important things for meteorologists to remember when using NWP data is when to discard the model output! When you realize that the data is leading you down the wrong path, don't use it!

If you’re looking for proof that computer models are not perfect, check out the National Weather Service Subjective List of Model Performance Characteristics or the twice-daily issued NWS/Weather Prediction Center Prognostic Meteorological Discussion, where real-time model performance is judged by NWS meteorologists.

Tropical Cyclone “Spaghetti” Plots
A spaghetti plot is a map that displays the output of many different models on the same graphic so that the user can compare the solutions. Most commonly, we see spaghetti plots showing the indicating forecast center positions of high and low pressure areas, including the center of tropical storms and hurricanes. In the graphic on the left, from tropical cyclone Danny in August of 2015, the model solutions, as it regards future movement of the center of the system, are fairly consistent in their forecast of the future path of the system. Consistency would bring higher confidence in the final forecast track. On the other hand, the spaghetti plot on the right, from tropical cyclone Lisa in the eastern Atlantic Ocean basin in 2010, displays a great deal of disagreement. As a result, there is less confidence in the final forecast track. When faced with the divergent forecasts in this second example, it is important for the meteorologist to know each model's characteristics in detail before making a specific track forecast.

Tropical Storm Danny
Tropical Cyclone Lisa

If you use these hurricane track forecasts, please be sure to read the tropical cyclone discussions being prepared by the meteorologists at the NWS/National Hurricane Center. They outline the respective model strengths and weaknesses and how their finalized public forecast track was made. Another reminder—although not related to the primary discussion here—when looking at the forecast track of any tropical cyclone, remember that a tropical cyclone is never a point, it may cover hundreds if not thousands of square miles.

In Closing
Numerical weather prediction is a tool, and, as part of the meteorologist's tool box, it is an important for forecasting the state of the atmosphere into the future. It is a specialized tool, however, and one that requires meteorological experience and understanding about the various models strengths, weaknesses, limitations and biases. These biases are each different based upon the model type, the time of year as well as the type of weather pattern. While these models can be used by non-meteorologists, it is essential that everyone understand that these atmospheric models are not perfect. A model user must also take the time to read NWS weather discussion products to get a better understanding of how the respective models have been performing and are expected to perform in the near future.

For additional information:


  • NOAA/National Weather Service
  • UCAR/University Corporation for Atmospheric Research
  • Florida State University
  • Bob Rose, Chief Meteorologist, Lower Colorado River Authority, Austin, Texas
  • Dr. Kevin Kloesel, Director, Oklahoma Climatological Survey, University of Oklahoma, Norman, Oklahoma

Share |