THE FUTURE OF WEATHER AND CLIMATE MODELLING - WWW.HPC-ESCAPE.EU
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
A collective approach to weather and climate prediction The first-of-its-kind collaborative project ESCAPE is helping to facilitate high-resolution weather forecasting. By changing numerical algorithms and employing new programming models, significant improvements to both weather and climate predictions will be possible and affordable, leading to reliable and timely forecasted warnings While it is possible to reduce the established. It represents the first has helped define and initiate ESCAPE. human impact on the global climate, it collaborative project between European ‘The team of weather prediction centres, is impossible to control the weather – weather prediction centres, academia, high- academic institutions and computer irrespective of whether it is a result of global performance computing (HPC) centres and companies make sure that ESCAPE warming or naturally occurring. However, computer hardware vendors. developments are fit-for-purpose for both as technology develops and becomes regional and global applications, and that more and more innovative, both weather The project would not be possible without the latest developments in numerical and climate predictions are increasingly the excellent international collaboration techniques, programming environments accurate, providing reliable forecasts and that underpins this work. ESCAPE joins up and hardware capabilities are brought timely warnings. operational meteorological services from together,’ says Bauer. Switzerland, France, Germany, Belgium and Despite these advances, it is critically Denmark, a UK university, high-performance KEEPING COSTS DOWN WHILST important that more precise forecasts are computing centres in Ireland and Poland MAINTAINING PERFORMANCE developed as soon as possible, not least and three industrial partners providing The chief aim of the ESCAPE collaboration because of the aforementioned long-term, computing hardware. These organisations is to reformulate the key model components human-induced climate change that is are all working hard to investigate new that drive the significant increases in cost. taking place all around us. For, while the formulations of forecast model components The proposed reformulation considers weather has always generated natural to achieve better performances in scientific elements, such as different disasters around the world, the frequency the future. Importantly, the European equations, mathematical solvers and and severity of hazards related to the Centre for Medium-Range Weather algorithms, and computational elements, weather and climate is on the rise. Forecasts (ECMWF) – an independent such as programming models, software intergovernmental organisation supported libraries and more efficient processors. One While the need for improvements to weather by 34 states – is strongly involved in the crucial consideration is the need to ensure and climate predictions is evident, it is not a joint research effort. that by reformulating these elements the straightforward process. Weather forecasts scientific performance is not deteriorated. are necessary throughout each day, where ESCAPE, coordinated by Dr Peter Bauer, they are continuously updated to improve benefits from evolutions in extreme-scale However, taking such a significant step their reliability, and they therefore must be computing capabilities that have taken in defining the fundamental algorithmic generated within strict time constraints. On place over the last decade. Bauer notes that building blocks of weather predictions the other hand, climate predictions are more ECMWF has a long history of HPC facility is no mean feat. Indeed, one of the first focused on the long term, but both models management and optimising computer deliverables of the project is the definition of need to produce somewhere between 100 codes on a range of architectures for those building blocks. These follow from the and 1000 forecast days every real day. The achieving the best possible efficiency given hierarchy and sequence of logical tasks in problem is that the accuracy of weather and the available resources: ‘Their foundation the forecast model that reflects how groups climate prediction is strongly affected by the as a single European unit also helped of physical processes are represented in the computational affordability and, as models centralise HPC capabilities to allow access simulation model and which mathematical become increasingly complex, they become to larger computing power than available algorithms are needed by the chosen extremely expensive. Thus, a fundamentally in any individual Member State. This has numerical methods. As the latter are different approach to computing and data enabled cutting-edge scientific research also part of the redefinition step, several management is required. to be performed and translated into options for numerical methods and thus operational services.’ mathematical methods are investigated at THE BENEFITS OF PARTNERING the same time. WITH WORLD-CLASS FACILITIES Ultimately, Europe’s experience with With that in mind, the Energy-efficient advanced and explorative projects in As the project evolves over time, the multi- SCalable Algorithms for weather Prediction numerical weather prediction, with HPC, disciplinary team will be able to determine at Exascale (ESCAPE) project has been code management and code sharing which building blocks are important in
The team of weather prediction centres, academic Project Insights institutions and computer companies make sure that FUNDING ESCAPE developments are fit-for-purpose for both European Union Horizon 2020 regional and global applications PARTNERS Danish Meteorological Institute (DMI) • German Meteorological Service (DWD) • Royal Meteorological Institute of Belgium (IRM) • Météo-France • Meteo Swiss • Institute of Bioorganic Chemistry, Polish Academy of Sciences • Loughborough University, Leicestershire • National University of Ireland, Galway • Bull SAS • NVIDIA Corporation • Optalysys Ltd CONTACT Peter Bauer Project Coordinator T: +44 118 949 9080 E: peter.bauer@ecmwf.int W: www.hpc-escape.eu PROJECT COORDINATOR BIO terms of their relevance to future models, opportunities and provides computer Dr Peter Bauer joined ECWMF in with consideration given to the fact that the training with access to the latest hardware January 2000 and is the Deputy Director level of importance might change. Indeed, architectures, as well as practical experience of the Research Department. ESCAPE includes steps down the line to with very large science codes in some of the re-select and re-evaluate the components largest HPC facilities in the world. These as the understanding of them develops. opportunities are very attractive for young Importantly, the computational cost of researchers and present them with career the building blocks can be ascertained paths that have not been accessible to this using code profiling tools, so the team extent before. can focus on those that help keep costs down. In addition to the computational The fact that ECMWF is an international performances, ESCAPE gives special organisation with mostly European attention to power usage, as large HPC members and cooperating states means facilities require substantial electric power that ESCAPE will benefit the European sources that could become impossible to public by providing the highest forecasting sustain for future systems. Because the quality with affordable computational C : 100 C :0 C 100 - M 72 - N 11 project is focused on solving anticipated cost. However, ultimately, the benefits PANTONE® REFLEX BLUE C PANTONE® 032 C PANTONE®REFLEX BLUE C 100% > 16% M : 72 J :0 N : 11 M : 87 J : 87 N :0 100% > 16% problems in the future, it makes sense to of ESCAPE are further reaching than give consideration to the feasibility of the that, not least because improvements to solutions they proffer. weather forecasting will not only save lives – ESCAPE will help save vast amounts of THE FAR-REACHING money currently associated with natural ASSOCIATED BENEFITS disasters around the world. More reliable As part of the proposed structure for the and earlier weather and climate forecasting project, ESCAPE includes a goal that focuses leads to better preparation by those on dissemination and training that provides potentially affected and, given the expected future recruitment opportunities in the field. effects of climate change in the future, Traditionally, the combination of physical, the weather will become an increasingly mathematical and computer science skills societal concern. is exploited in computational fluid dynamics and engineering, but ESCAPE wants to draw ESCAPE represents a huge step forward scientists skilled in those areas into weather in weather and climate modelling, but and climate prediction. it also helps develop interdisciplinary research on energy-efficient HPC which has The project is designed to present young obvious uses in other fields. The future of researchers with career and training computational modelling looks increasingly opportunities in an area with high societal bright, whatever the weather. relevance. ESCAPE offers employment
Impact Objective • Develop world-class, extreme-scale computing capabilities for European operational numerical weather prediction and future climate models The future of weather and climate modelling Dr Peter Bauer is Coordinator of a project established to develop world-class, extreme-scale computing capabilities for numerical weather prediction and future climate models. Below, he discusses the challenges the project seeks to address, the coordination of complex research initiatives, and upcoming dissemination events the dimension of the models will increase are certainly very satisfied with the feedback by a factor of 100–1000 in the next 10 years we receive on the importance and design because of that. Obviously, the acquisition of the project, and its high relevance for and power supply for 100–1000 times larger the weather and climate community. Also, high-performance computer (HPC) facilities through ESCAPE, the HPC community have will be extremely expensive. ESCAPE aims realised how challenging our particular at identifying and reformulating the key application is and how important therefore model components (called ‘weather and our participation in their co-design efforts is. climate dwarfs’ inspired by the Berkeley computational dwarfs) that drive this cost. What plans do you have for ESCAPE in 2017? Coordinating such a complex and Could you begin by describing some of the multifaceted technological research Next year will see the first full cycle through current challenges associated with weather initiative must present some major the iteration loop – from the definition and and climate predictions? obstacles. How you have dealt with these? extraction of the dwarfs via the software adaptation that employs new programming Both weather and climate predictions need The coordination of ESCAPE is clearly models – towards the testing of components to produce forecasts and products within challenging, just like all projects with many on novel hardware, including an estimation strict time constraints. Weather forecasts partners with multidisciplinary backgrounds. of compute performance. This will be a are issued regularly several times per day We have set up an effective project proof of concept of the main ESCAPE and therefore the models need to run in managing structure, including a science development path. specified time slots every day. Since climate coordinator and project manager, which models produce predictions for long time helps separate the scientific and formal Finally, do you have any dissemination periods – often over decades and longer aspects of ESCAPE. Up to now, no major planned where you will be sharing the – they need to complete these predictions obstacles and problems have surfaced. The results from ESCAPE? within reasonable periods and with partners are very competent and responsive, continuous processing. Either through burst and we have not experienced delays in any We have a News and Events column that (weather) or continuous (climate) mode, work area. Also the European Commission your readers can visit at www.hpc-escape. the required production rates are about (EC) project officer has been very helpful eu for more information. Important events 100–1000 forecast days per real day. This and one can see that the EC is really are the first Dissemination Workshop requirement will not change and becomes interested in the subject. in October 2016 and a range of training extremely challenging for high-resolution events. ECMWF’s Scalability Programme and complex models, even more so if The EC recognises that weather and climate (http://www.ecmwf.int/en/about/what- models are run in ensemble mode. prediction offers a very challenging case we-do/scalability) and ESCAPE will also be for future HPC development, and that prominently represented at the biannual In what ways will the Energy-efficient weather and climate forecast requirements European Centre for Medium-Range Weather SCalable Algorithms for weather Prediction need to be fulfilled, given their significant Forecasts (ECMWF) HPC in Meteorology at Exascale (ESCAPE) project overcome importance for European society. Workshop. To find out more about this, these challenges? your readers can head to www.ecmwf.int/ Have you achieved any results that you are en/learning/workshops-and-seminars/17th- The enhancement of spatial resolution and particularly pleased with? workshop-high-performance-computing- complexity of forecast models makes the meteorology. models physically more realistic and thus It is too early to report on actual results will improve forecast skill. We assume that from the software developments, but we
You can also read