The new ECMWF Strategy to 2030 foresees a pushing of the boundaries of weather science to achieve global ensemble forecasts at kilometre-scale resolution. In this interview, ECMWF Director of Science Andy Brown explains how research, forecasting and computing must work together to achieve the goals set out in the Strategy.
What is the purpose of ECMWF’s Strategy?
The societal needs for better weather forecasts and climatology are huge, but the science challenges are very big and significant as well. That is why we need a vision on a ten-year timescale in order to prioritise the science we’re doing. In terms of the evolving ECMWF products and services for our users and ways of distributing data like the European Weather Cloud, we also need to share an agreed long-term vision about where we’re trying to get to.
The Strategy covers ten years but a new one is drawn up every five years. Why?
Very deliberately, the Strategy doesn’t change every year. We’re setting a multi-year direction, which we hope is robust. Yet, at the same time, to think we have a perfect crystal ball for ten years ahead would be a bit arrogant. So, every five years it’s good to take stock, and very often large parts of the previous Strategy will still be valid. That’s also the case this time, but there are also exciting new directions where the science, technology or user needs have moved on.
“Operationally we are talking about three- to four-kilometre resolution, in research mode we’ll be experimenting with higher still. That’s challenging both computationally and scientifically.”
What are the main innovations in the new Strategy?
Beyond the continuity regarding Earth system science, seamless modelling, and ensembles, I think there’s a real pushing of the boundaries, both scientific and technical. We are moving towards global kilometre-scale modelling. Operationally we are talking about three- to four-kilometre resolution, in research mode we’ll be experimenting with higher still. That’s challenging both computationally and scientifically.
The strategy also contains a lot on the use of observations. New observations are coming, like EUMETSAT MTG (Meteosat Third Generation), EPS–SG (EUMETSAT Polar System–Second Generation) and further Copernicus Sentinel satellites. We also have to learn how to get even more value out of satellite data. An exciting opportunity is the application of machine learning in our field, where we will seek to combine the best of what data-driven approaches can provide with the huge strengths and physical understanding encapsulated in our existing forecasting systems. We are also developing the tools for sharing the vast amounts of data that we use and produce, through the European Weather Cloud and moves to open data.
“We’ll need to move to non-hydrostatic forecasting models. We’ll have to do some of the physics differently as we start resolving convection.”
What are the main targets for research?
As I mentioned, there will be moves towards higher-resolution ensembles and we will work to overcome the scientific and technological challenges in order to get full benefit from those directions. We’ll need to move to non-hydrostatic forecasting models. We’ll have to do some of the physics differently as we start resolving convection. And we will continue to increase the range of processes modelled where additional Earth system complexity can be demonstrated to lead to improved forecast outputs.
To get even more value from observations, we need to continue to develop data assimilation and take a more consistent approach across different Earth system components. Also, we are launching a new, big initiative called ‘from all-sky to all-surface’. We did a ten-year research programme to be able to use microwave data in cloudy conditions. This all-sky initiative delivered a much more extensive use of microwave data and demonstrated improvements in the quality of the forecast. But we’re still less able to use satellite data over land, snow and sea ice. We’re envisaging a similar multi-year effort to extract more value out of those observations.
“For extreme temperature anomalies and hydrological impacts, the Strategy calls for skilful predictions up to three weeks ahead. It’s a huge scientific challenge.”
What are the main targets for forecasts?
The current Strategy is continuing to focus on user-relevant improvements in forecasts. There is thus a continued focus on ensemble forecasting, near-surface weather measures, and extremes. We’re continuing to push for more accurate forecasts of high-impact weather into the second week to give better lead-time warning of severe weather. For extreme temperature anomalies and hydrological impacts, the Strategy calls for skilful predictions up to three weeks ahead. It’s a huge scientific challenge.
In this Strategy, while medium-range forecasting remains our bread and butter, and while we do seasonal forecasts as well as climate monitoring, we particularly emphasise the extended range, up to 46 days ahead. Doing better in the extended range is also an exacting test of whether we’re doing the right thing at the medium range. Hence, we’re interested in the extended range as an end in itself because the users are interested in it, but the way we’ll make the extended range better will also have knock-on benefits for the medium range.
What is ECMWF’s strategic interest in the Copernicus Services?
The Copernicus Services are very valuable to society: the Copernicus Atmosphere Monitoring Service provides air quality forecasts and reanalyses; and the Copernicus Climate Change Service provides weather reanalysis over many decades and the multi-model seasonal forecast. We want to build on that in future Copernicus agreements. We are also looking to expand into monitoring and verification support for CO2 emissions. And we operate the global and European flooding services, GloFAS and EFAS, and the fire danger forecast service as part of the Copernicus Emergency Management Service.
There’s huge value to us scientifically in the synergies between the Copernicus work and the work we do for numerical weather prediction in the medium and extended range. The reanalysis of past weather, for example, is fundamental to our efforts to make the weather forecast better. Over the period of this Strategy, more and more of the processes that are of interest to Copernicus, for example in atmospheric composition, become relevant for numerical weather prediction, too.
“There are likely to be more hybrid architectures in the future, CPUs and GPUs for example. There’s an opportunity for us to exploit these new technologies.”
What developments in computing are important for the next ten years?
One of the reasons why our forecasts have got better over many years is bigger computers to let us run higher-resolution models and use more complex algorithms. That is why we continue to need more computing capacity. There are likely to be more hybrid architectures in the future, CPUs and GPUs for example. There’s an opportunity for us to exploit these new technologies, but there’s also a challenge with adapting our codes to take advantage of them.
We are very actively working to be ready for a future procurement to use these hybrid architectures. We also try and develop our codes to be more flexible. As well as adapting to the known changes, we’ve got a longer-term strand of work trying to make ourselves resilient to as yet unknown future evolutions in computing.
“We’ll be thinking over the coming years of how to take advantage of the opportunities offered by remote meetings.”
What are the main changes in education and communication?
It’s clearly part of our role to educate and communicate, so we have a very active programme of seminars, and workshops and courses. And we will continue to expand that. I think there are interesting opportunities. The recent COVID crisis has forced us all to work remotely, but through many of our events being entirely virtual, we have seen an increased reach.
We’ll be thinking over the coming years of how to take advantage of the opportunities offered by remote meetings when we can come back together. I think a hybrid approach is called for, with a central hub and people on site while making sure that a wider audience is fully engaged. This will also help to improve environmental sustainability by decreasing our carbon footprint.
What about partnerships and collaborations?
We have huge numbers of partnerships across everything we do, both at the user end and on the science and technology side. We can’t have all the skills and expertise ourselves, so we work with partners in many universities and research institutes around Europe and the world. Many of our Member and Co-operating States have strong and active research units. We work, for example, very closely with some of the limited-area model consortia, with whom we share code.
We have the Fellows programme, which gives us access to some of the top expertise worldwide. And we work very closely with EUMETSAT, ESA (European Space Agency) and the WMO (World Meteorological Organization). For example, we’re reliant on the exchange of data that’s coordinated through the WMO, and we’re strongly represented in the science side of the WMO.
“People are at the heart of what we do. I think the Strategy rightly elevates people to one of its three headings.”
What are the Strategy’s key messages about organisation and people?
We know that our forecasts have improved for three reasons: new and better observations, bigger computers, and scientists developing new techniques and new code. It’s not all about hardware and infrastructure. People are at the heart of what we do. I think the Strategy rightly elevates people to one of its three headings.
How do we attract, recruit and train a diverse workforce while overcoming the challenges of multiple sites that we’re moving to? I think an absolute commitment to our people, finding efficient ways of working that are flexible, in a diverse organisation across multiple sites, is crucial to our success.