ECMWF and GFS are two leading global weather models. This means that when you look at the weather forecast for your favorite spot and activity, you most likely see data from one of these two models. However, there is a logical question: what is the difference between the models, and which one is more accurate? Are there any alternatives? In this article, we will answer it.
Before talking about the differences, it is important to look at the similarities to understand where we are coming from.
First and most importantly, ECMWF and GFS are the two most common global weather models. This means that they provide weather forecasts for the entire world due to their Spatial resolution or grids with a series of points where the weather is predicted, which cover the entire globe.
The same models are also used as a basis for making forecasts by various regional models for specific areas of the earth, e.g. France and its neighboring countries, the Mediterranean Sea, separately the USA and North America, and others.
The second important similarity is that both models are professional governmental projects, not private or amateurish. This is understandable: to use global models you need incredibly powerful supercomputers (and knowledge), which private companies and individual meteorologists do not have. Plus, it is very expensive to run a weather model.
This also explains the popularity of these two models — being the largest government projects, they have logically become the most widespread.
Third, both models produce weather forecasts for all seven main weather elements, namely: general weather conditions (sunny, overcast, and so on), air temperature, precipitation, wind, atmospheric pressure at sea level, relative humidity, and visibility. But each of the models has dozens more advanced elements for the atmosphere, ocean, and other environments. We will name the specific elements later because this is more of a difference between the models.
The fourth thing the two models have in common is that they both produce medium-range weather forecasts. This is a type of forecast depending on their duration which equals a period from 72 hours (3 days) to 240 hours (10 days). ECMWF and GFS give a maximum of 10 days medium-range forecast. But no more than that.
This is the end of the main similarities of the models, although you can find other less important ones if you want.
To summarize: ECMWF and GFS are the two most popular governmental global weather models with maximum 10 day medium-range forecasts for seven major weather elements, including the three main ones: temperature, precipitation, and wind.
ECMWF weather model forecast for Burghfield Sailing Club, Reading, United Kingdom, in the Windy.app for iOS. Marco Zuppone / Unsplash
ECMWF and GFS have a lot in common, but there are even more differences. Let’s look at the main ones in the same way.
ECMWF (European Center for Medium-Range Weather Forecasts) is the main European global weather service from an independent intergovernmental organization supported by 34 nations of Europe, so it is the main European weather service with headquarters in Shinfield, Reading, Berkshire, South-East England, United Kingdom but also in Bologna, Emilia-Romagna, Italy, where its data center is located now, and Bonn, North Rhine-Westphalia, Germany, from where they run their Copernicus Programme operations, including the Copernicus Climate Change Service.
In other words, when you look at the weather forecast on one of the European services, say Ilmeteo.it, you probably see a forecast from ECMWF which is also not hard to find out by looking at the data source above the forecast table. For that reason, it’s also often called the "Euro weather model", or just "Euro". Basically, the Americans do it when they contrast it with the GFS.
The official website of the organization is Ecmwf.int.
GFS (Global Forecast System) is the main US global weather model from the National Centers for Environmental Prediction (NCEP) of the National Weather Service (NWS) of the National Oceanic and Atmospheric Administration (NOAA), the main US weather service, with headquarters in Washington, D.C.
In other words, when you check the weather on one of the US services say Weather.gov, you see a GFS forecast. Similarly, this model is often referred to as the "American" model by the Americans themselves, as well as people from other parts of the world.
The ECMWF model's spatial resolution is 14 km (8.6 mi) while the GFS model has 27 km (16.7 mi), which is why the latter also has another popular name, GFS27. Does it mean that the bigger number means the more accurate forecast? Not at all. More: it means exactly the opposite.
Spatial resolution is the distance between two points of the weather model grid. Hence the smaller the number, the better the resolution, and hence the quality of the forecast. It is like the pixels on your computer monitor, smartphone, or photo camera.
From this, we can make a preliminary conclusion that the ECMWF model is more accurate and we will talk more about that right away.
Forecast step is the period of the weather forecast within one day or how many hours ahead you can see the forecast. With ECMWF, it is 1 hour, while GFS has 3 hours. That is you can see the forecast of the first model with 1-hour frequency: for 6 am, 7 am, 8 am, and so on, and for the second model — with only 3 hour frequency: for 6 am, 9 am, 12 am, and so on.
The forecast for each hour is also more convenient. ECMWF also wins here.
Weather models produce weather forecasts from one to several times a day. This is called "update frequency" or "expected update". Accordingly, the more often it is updated, the fresher the forecast in your weather application or on your favorite website.
The ECMWF model is updated twice a day or every 12 hours. This means, for example, that the 15 o’clock forecast you see was made early in the day and will be updated only once more during the day. The GFS has a better figure — 4 times a day or every 6 hours.
Does that mean now that GFS is better? Again, no. Because the model's spatial resolution is much more important than its refresh rate. In other words, as a source of weather forecasts, it’s better to have a more accurate model which makes one accurate forecast per day, than four less accurate forecasts from a less accurate model in general.
So, those are the main differences between the two models.
To summarize: ECMWF and GFS are weather models from two parts of the world, Europe and the United States, from two different weather services, the former being a merger and the latter being a single service. ECMWF resolution is twice as much as GFS — 14 km vs. 27 km, but the latter is updated twice as often — every 6 hours vs 12 hours. At the same time you can see ECMWF forecast for every hour — three times as often as GFS.
But that's not all with the differences. There are several additional ones that will be better understood by experts in meteorology and physics. Let’s also name the main ones.
In addition to the basic weather elements, the models provide a forecast for dozens of advanced elements that are not found in the general weather forecasts. For example, GFS, predicts land-soil variables and sea-ice.
ECMWF and GFS use different equations to predict weather phenomena in the atmosphere. This brings us to hydrostatics, a branch of continuum physics that studies fluid equilibrium or the theory of fixed fluid behavior. It also refers to topography.
So, ECMWF is a nonhydrostatic weather model. This means that it uses altitude in the forecast, i.e. more accurately takes into account the influence of the terrain on the weather. More: ECMWF is the only global nonhydrostatic model used in the world for general weather forecasting.
Hence, the GFS is a hydrostatic model. Instead of altitude, it uses atmospheric pressure (which as we know changes with altitude). This means that compared to nonhydrostatic models they perform worse at higher resolution of the model grid, i.e. they take the topography into account worse.
Hydrostatics is also one of the reasons for the frequency of model updates. Nonhydrostatic models like ECMWF require more computation, while hydrostatic models require less.
Most likely you get the weather forecast for free or for a small fee from your favorite service. But that doesn't mean that creating a forecast isn't expensive or doesn't require money at all either. On the contrary: weather models that require supercomputers to use are very expensive. So money is another difference between the models, more understandable to professionals.
Because ECMWF is a more complex model, it is more expensive to compute, and its data is distributed for payment to third-party services — such as the outdoor weather forecast app Windy.app.
GFS is simpler and less expensive, so it’s free. You can even download and upload its data for your weather service, or you can use the model itself — if you had a supercomputer.
Based on cost, GFS is also a more common initial forecast source for third-party services than ECMWF, for which you have to pay.
Let’s summarize the additional differences of the weather models as well: the two leading weather models contain dozens of additional weather elemets each. ECMWF is a nonhydrostatic model, which works better at lower weather grid resolutions, while GFS is a hydrostatic model and is generally worse at predicting the weather. But the latter model is in the public domain, while the former is distributed for money.
GFS weather model forecast for Ellipse, Washington, D.C., United States, in the Windy.app for iOS. Andy Feliciotti / Unsplash
ECMWF vs GFS is an old dispute about which weather model is better. Meteorologists have been arguing about this for more than 30–40 years since the US National Meteorological Center's Global Spectral Model, GFS's precursor, was introduced in August 1980 and the ECMWF debuted in May, 1985.
As confirmation of the models dispute, there are many publications like this one from Forbes by Marshall Shepherd, the President of the American Meteorological Society (AMS), from Feb 14, 2019, in which he calls the opposition of models nothing less than a meteorological war.
So which model is more accurate? As it became clear from comparing the main and additional differences between the models, the ECMWF is considered to be more accurate because of its more powerful supercomputer infrastructure and better resolution, despite the sharper update. This is noted by the experts themselves who make the forecasts in publications such as those cited above as well as by many regular users of the weather forecasts who are not shy with words when describing models.
The advantage of ECMWF over GFS in forecast accuracy is also confirmed by research published by another leading American meteorologist Ryan Maue in which this and other scientists compared the forecast quality of both models for 14 years from 2008 to 2022. The graph below shows that the models go head-to-head, with the European model giving a more accurate forecast all the time. The data are for a 5-day forecast for the Northern Hemisphere between 20 and 80 degrees.
Comparison of ECMWF and GFS accuracy from 2008 to 2022 for a 5-day forecast in the Northern Hemisphere / Ryan Maue
Does this mean that ECMWF is absolutely and always more accurate? No. Because historically, there have been many cases recorded by meteorologists over the same period where GFS predicted individual weather events, particularly from severe and extreme weather, better. For example, GFS predicted the formation of Tropical Storm Dorian long before ECMWF. It was a catastrophic Category 5 Atlantic hurricane under the Saffir-Simpson Hurricane Wind Scale (SSHWS) with the highest wind speed of 185 mph between August 24 and September 10, 2019, with damage of more than USD 5.1 billion.
And that’s very important. Because as another scientist, Pavel Konstantinov, a meteorologist from the opposite part of the world, says, "the future of meteorology is in predicting extreme weather" to prevent the devastating effects of it: floods, wildfires and droughts, and more. In other words, accurate prediction of such phenomena is more important than general weather forecasting.
Read more about this in an article "How Big Data analytics helps in weather forecasting. An example of Windy.app" on the blog.
However, one should not conclude from this at the same time that GFS as a whole is better at predicting hurricanes. For example, four years earlier, ECMWF was better at predicting Hurricane Joaquin in October 2015.
The information from the Forbes and other sources also makes it clear that with mixed success, with the support of the U.S. government, NOAA listens to feedback from experts and users and tries to improve the general accuracy of the weather model.
The last significant update was made in 2019 via the FV3-GFS (Finite-Volume Cubed-Sphere) version of the model, which US National Weather Service director Dr. Louis W. Uccellini himself compared to "replacing the engine of a car." By the way, it was the 16th update to the model in its history. In particular, the improvements included an increase in the vertical resolution of the model from 55 km to 80 km (from 34 to 50 miles). Here's the NOAA's press release about that.
To summarize this section, ECMWF has consistently been better at predicting overall weather on many parameters over the past 14 years, as evidenced by data. However, it is not always accurate. The GFS has been better at reporting severe and extreme weather many times, but it also has not been a universally accurate source. GFS is trying to get better.
Also to increase your confidence: both models also works with general accuracy of 95–96% for up to 12 hours, 85–95% for three days, and 65–80% for 10 days.
ICON13 weather forecast for Offenbach am Main, Germany, in the Windy.app for iOS. Wikipedia
ECMWF and GFS are two weather models which cover almost all needs of ordinary weather users and experts, but they are not 100% universal due to their peculiarities and the complexity of weather forecasting in general. From this, it’s logical also to ask about alternative models in case you can’t find the information you need in one of them, or they don’t suit you for other reasons.
The main alternative is the ICON13 weather model. This is also a global European model, which name stands for Icosahedral Nonhydrostatic (hence, like the GFS, it uses pressure instead of altitude) from Deutscher Wetterdienst (DWD), the main German Meteorological Service with headquarters in Offenbach am Main, Hesse state, Germany.
The official website for the model is https://www.dwd.de.
Why use ICON13? For example, the 13 in its name of the model indicates a better resolution than even the ECMWF, albeit only by one kilometer (believe us, for meteorology that is a lot). So, in Europe, it is generally considered to be even more accurate than the ECMWF due to the better resolution.
ICON13 forecast step is also a maximum of 1 hour.
The model also provides all the basic weather parameters but also some advanced ones. The most important weather elements of the ICON are air density and virtual potential temperature, horizontal and vertical wind speed, humidity, cloud water, cloud ice, rain, and snow.
However, it has some disadvantages, too: like ECMWF, the model is updated twice a day against four times for GFS.
But the main drawback of the model is that it makes forecasts for 5 days only instead of 10 days in the two leading global models. However, perhaps this is still its advantage because, as was explained above, the shorter the weather forecast, the more accurate it is in general. If you don’t need a 10-day forecast but need better resolution, ICON13 rather than ECMWF might be your choice.
At the end, it is also important to note that ICON13 is one of the two models of German weather service on par with ICON7, the local model for Europe which ECMWF and GFS do not have — another of its features and advantages. This can be handy in cases where this model suits you in world forecasts and you travel to Europe — you can keep it.
Like GFS, ICON13 is free of charge, mostly. You can more information about it at Open Data Server page of the DWD site.
So, we have described in detail the two major global weather models in the world and found one alternative to them. Let’s summarize their main similarities and differences in the table:
Global weather models comparison: ECMWF vs GFS vs ICON13 / Windy.app. To get larger image, simply right-click on it and open it in the new tab
You can find weather forecasts from all three major global models for more than hundreds of thousands of spots for various outdoor activities around the world in the Windy.app application.
To do this, on the app’s Home screen or the Weather Map, find your home, favorite, or nearest spot and go to its page. You can also do it visually by icons on the same map.
Outdoor spots search in the Windy.app for iOS
Visual outdoor spots search by icons on the Weather Map in the Windy.app for iOS
Select one of the models in the slider below the wind rose and get a forecast with a set of basic and advances weather elements.
Weather models on the Spot screen in the Windy.app for IOS
To find out which weather parameters are available from each of the models, go to Weather Profiles via the icon to the right of the model slider. Scroll to the Advanced menu section.
ECMWF weather elements in the Windy.app for iOS
GFS weather elements in the Windy.app for iOS
ICON13 weather elements in the Windy.app for iOS
You can even also create your profile by combining parameters from ECMWF, GFS, ICON13, and other models!
Custom weather profile with ECMWF, GFS and ICON13 weather models in the Windy.app for iOS
Another extremely useful function of the application which resonates very much with the theme of this article is to compare the forecasts of these and other models on the same chart. To do this, click on Compare in the slider with the models. The feature is in Pro version of the app.
Comparing ECMWF and GFS and ICON13 in the Windy.app for iOS
Finally, you can get the forecast for the whole world on the Weather Map, which is logical in the case of global weather models, because local models don’t give it, covering only part of the world.
Weather models on the Weather Map in the Windy.app for iOS
Text: Ivan Kuznetsov
Cover photo: Hannah Wei / Unsplash