Explainer: Weather models, and what they mean in forecasting

We’ve all heard it from meteorologists on the evening news: “Here is what the models say.” Or perhaps, “The models are predicting....”

What models, you might think? What are they talking about? Is it the kind of model pictured above? Or when I was very young, I thought about becoming an architect and building models like this:

Well, maybe not that good, which is probably why I became a meteorologist. But no, we are referring to a totally different thing: weather models. These are basically computer programs that can help predict what the weather will be in the future, be it tomorrow, three days out or even next season. Modern meteorology and forecasting is to a large extent based on what is called numerical weather prediction, which depends greatly on these models. NOAA has a fantastic wrap-up about the history of NWP.

How do weather models work?

First, understand that a computer knows nothing about weather or climate. Without any data or knowledge of the basic science of weather, the most advanced computer will produce nothing. So the computer needs to learn some things first, chiefly the fundamentals of science, the mathematical laws that underpin the workings of our universe. These laws, such as Newton’s laws of motion or the core equations of chemistry, govern everything from how far a football can be thrown (even the best NFL passer can’t put a football into orbit) to how rapidly a pot of water will boil.

Armed with these laws, a weather model (or its operator, really) is ready to do some work. One of the first attempts at numerical prediction was performed during World War I by British scientist Lewis Fry Richardson, who spent more than a month calculating the pressure change in a tiny part of the air over a six-hour period. Not a very effective way to tell the weather, but still revolutionary for the time. Richardson realized the limits of his modeling efforts and dreamed of a day when thousands of mathematicians would work in “forecast factories,” solving the fundamental equations that govern the weather around the world.

That time has come about, sort of: The “forecast factory” is the computer. With computers, we put initial observations into a number of fundamental equations, solve these equations for hours or days and we have a forecast of the future weather. Easy right? Not really, because we can’t measure everything to every cubic inch. The scale of the earth environment is just too vast and today’s smartest supercomputers couldn’t handle all the data even if we could measure every inch of the earth.

But we are able to measure and predict on finer and finer scales. Look at the difference and detail in measuring the temperature of the Atlantic and Gulf Stream on a scale of tens of miles versus just a few miles:

And the same goes with air temperature and wind and pressure and cloud cover and precipitation. Data from every source, from thermometers near the ground to Doppler radars measuring precipitation and wind to new weather satellites (that’s why the new NPP satellite is so important), are fed into supercomputers with the fundamental equations that govern the science. The resulting “model” is a three-dimensional simulation of the future weather, oceans and climate. Here is one simulation for the winds about 5 miles above the United States this Saturday:

General Forecast System (GFS) from NOAA's National Centers for Environmental Prediction

And this is a simulation of what the pressure and precipitation will be at the surface on that same day:

GFS output of surface pressure, 6 hour precipitation and surface winds

Use your imagination, and you can see the wave above us in the top image. The blue shading shows where the highest winds and the jet stream are and on the lower image is a simulation of weather at the surface, with the “L” or low-pressure area being a possible storm off the coast.

What does this model “say”? It says that Saturday will be wet and cold, but that is not the end of the story. After all, we are meteorologists, not just model readers. There are many, many more tools we look at before making a forecast, from just watching the trends of storms to using the latest Doppler radars and hour-by-hour observations. Knowing the local quirks and vagaries of the Mid-Atlantic weather (snow more likely in the mountains, rain more likely farther south and east) also helps with an accurate forecast, as does climatology and of course our education to become meteorologists.

The future of weather simulations (I like that word better than “models”) is really in ensemble forecasting, which I wrote about in a previous post about “forecasting Spaghetti-style.” We know our data is not perfect. We make approximations about some of the physics because we can’t measure every drop of water. But if a variety of simulations converge toward a common solution, we can feel more confident about the event, whether it is a forecast 24 hours from now or a projection of climate change 100 years from now. Here’s what the latest ensemble for Saturday looks like, thanks to Penn State's great weather page:

The fun for all of us is of course the “big” weather days ahead as winter comes. Here at ABC7, we have our own very hyper-local weather simulation called a Futurecast with clouds, temperatures and precipitation. Here it is out to Friday morning. A cold morning for sure, but I think 35 degrees might be a bit low for D.C.:

Keep an eye on our unique simulation (OK, “model”), which we plan to link to in our forecasts. We hope this is a useful tool for you in seeing the hour-by-hour weather where you are or where you're planning to go. As all of us meteorologists who always loved the weather and grew up trying to predict it and communicate it to you... happy weather-and-Futurecast watching.