For weather forecasts to be useful, they must be delivered in a timely manner.
On 7 March, ECMWF started disseminating medium-range probabilistic forecasts to its users 40 minutes earlier than before.
For example, national meteorological services and other clients now receive a 15-day ensemble forecast based on initial conditions at 0000 UTC as early as 0800 UTC.
The change is hugely significant for users in ECMWF’s Member and Co-operating States: it can mean being able to use the most recent 0000 UTC forecast in key applications instead of having to rely on the previous day’s 1200 UTC forecast.
“A number of our customers have important decisions to make early in the morning or at the end of the day to organise their activities in the short and medium term,” says Jean-Marie Carrière, the Director of Meteorological Services at Météo-France.
“These decisions depend on ensemble data or on forecast products derived from the ensemble. This is particularly true for customers in sectors such as energy, agriculture, food distribution, road maintenance, transport, and the safety of people and goods,” he explains.
The change in the dissemination schedule also applies to ECMWF’s provision of ensemble boundary conditions for limited-area models run by national meteorological services.
The Hungarian Meteorological Service (OMSZ) has tested the impact of such boundary conditions and has been using them since November 2016. “It is very welcome that ensemble boundary conditions are going to be available earlier in ECMWF’s new schedule,” says Mihály Szűcs, a meteorologist at OMSZ.
Computational challenge
Ensemble forecasts account for the uncertainties inherent in weather prediction by producing a set of possible outcomes.
At ECMWF they comprise 51 individual forecasts, or ensemble members, which together describe the range of possible weather scenarios and their likelihood of occurrence.
Producing such a large number of forecasts requires substantial supercomputer resources.
Before the forecast can begin, it is necessary to process tens of millions of weather observations from all over the world. These are combined with model data to build a representation of the current state of the atmosphere and related Earth system components. This procedure takes several hours.
For a 15-day ensemble forecast based on initial conditions at 0000 UTC, production of the 51 ensemble members and post-processing only begins at just after 0500 UTC. It used to take about three and a half hours.
Cutting that time by 40 minutes was a major undertaking.
“This substantial reduction has been facilitated by last year’s upgrade from ECMWF’s two Cray XC30 supercomputer clusters to two Cray XC40s,” says Umberto Modigliani, the Head of User Support at ECMWF.
“However, even with this upgrade we had to squeeze out extra minutes by making the production process as parallel and efficient as possible,” he adds.
ECMWF’s new ten-year Strategy puts a strong emphasis on ensemble prediction.
“The place of ensemble forecasts in the Strategy and their importance to users means that efforts to make them available as early as possible will continue,” Umberto says.
Image: BrianAJackson/iStock/Thinkstock