ECMWF's HRES Weather Computer Model
There is a weather forecast model out there called HRES, but almost nobody refers to it that way.
The HRES model is what is commonly referred to as 'ECMWF', or simply 'The European' or 'The Euro'.
HRES is incorrectly called 'ECMWF' which is the abbreviation for the organization that developed and runs the model,
The European Centre for Medium-Range Weather Forecasts. Given that everyone knows and refers to the model as ECMWF,
I'll continue to do so here.
Established in 1975, ECMWF is an intergovernmental organization supported by most European nations and is based in
Reading, UK. Unlike US models, ECMWF's models are not free. It's actually very expensive data to obtain, which in part is why it's
the leading provider of weather model data. Fortunately there are various sites that you can obtain the data from at a reasonable
price like F5Weather.
What ECMWF Does
The ECMWF organization runs several models. Their Integrated Forecast System or IFS runs their atmospheric HRES deterministic model,
and the Ensemble Prediction System, which is basically 51 iterations of the HRES run at lower resolution and with slightly different
initial conditions to get a range of possibilities in the forecast, giving a sense of uncertainty. ECMWF also provides seasonal outlooks out to 12 months,
air quality forecasts, ocean circulation monitoring and more. Their 'weeklies' are forecasts predicting forecast trends on a weekly
basis for the next 30 days.
The atmospheric HRES forecasts are done 4 times a day at 00z, 06z, 12z & 18z. The 06/18z forecasts only forecast out to 96 hours,
but the 00z and 12z forecasts go out to 240 hours or 10 days.
ECMWF is widely regarded as the world's best medium-range global weather
forecast model. Although it is consistently dependable, there is no better example of it's potential than with
In October of 2012 a storm developed in the Western Caribbean and tracked north. Almost all computer models had this storm staying
off the east coast of the United States and eventually turning out to sea, making little to no affect on the US coastline. However,
there was one model, the ECMWF, which had the storm do a complete 180 and track west right into the US Mid-Atlantic states nearly
8 days in advance of landfall. 5 days later the US GFS model started to converge on the ECMWF solution.
It was about 4 days before landfall that the National Hurricane Center modified their forecast to agree with
the ECMWF model's prediction. It took that long due to lack of support from other models. Long story short -- Hurricane Sandy did exactly
what ECMWF said it would. It was the 4th costliest storm in U.S. history at over $68 billion and killing 233 people.
Due to the superior forecast ECMWF had on Hurricane Sandy, post-storm testing was done to try to understand why ECMWF outperformed
all the other models. It was theorized that ECMWF's intensive satellite assimilation had something to do with it. They ran several
tests by removing satellite data from the initialization data and re-running the model. It wasn't until they removed NASA's polar orbiting
satellite data when the model started to perform poorly. So this proved that at least one aspect of ECMWF's data assimilation was contributing
to the superior forecast performance. You can read a bit more about that study here.
So why is ECMWF so good? There are many reasons. One of which is the superior way of integration and handling of satellite data and
other real observations into their models. This is called 4D-Var assimilation. This optimizes the way the model utilizes observational
data into the model for initial conditions. The US's GFS uses 3D-Var and is a major component to it being behind ECMWF & MetOffice
(British UKMET/Unified Model) in weather modeling. ECMWF processes about 90 satellite data products for its daily assimilation.
Another way to have a superior weather model is to increase the model's resolution, which currently is about 9km grid spacing. For a
global model, 9km is quite impressive. The US GFS is at 13km while the #2 leading model UKMET or the British Unified Model is at 17km.
The Canadian GEM and Australian ACCESS models are at 25km! So ECMWF's 9km is a significant improvement. This not only helps to
resolve things like storm development but also resolves influences from terrain or bodies of water. That's horizontal grid spacing.
There is also vertical grid spacing where the model uses data from air planes and balloon launches to initialize its forecasts
vertically through the atmosphere. This data is important for cloud formation, precipitation and other variables. The resolution
vertically from ECMWF is 137 vertical levels. The US GFS & new GFS-FV3 is only 64 vertical levels.
Here's a representation of the forecast accuracy of ECMWF compared to other models. It shows how ECMWF was consistently
superior during this period, which is December of 2018 over North America. You can use
this link to create your own time series graphs
comparing different models performance.
Resources & Further Reading