Annual Report 2017

www.ecmwf.int

Foreword 2

2017 At a glance 4

Advancing weather science 10

Delivering global predictions 16

Sustaining high-performance computing 22

Supporting ECMWF 26

Serving Member and Co-operating States 30

How we work 36

Member States as of January 2018

Austria Belgium Croatia Denmark Finland France Germany Greece Iceland Ireland Italy Luxembourg The Netherlands Norway Portugal Serbia Slovenia Spain Sweden Switzerland Turkey United Kingdom

ecmwf.int ECMWF Annual Report 2017 1

AR-2018-ARTWORK.indd 1 08/06/2018 16:01 A year among years

Florence Rabier Director-General

There is a lot to say about 2017, and the following pages allow us to remember or discover some of the key moments. Possibly the most striking aspect that comes to mind is the way our past, present and future collided in that speci c year.

2017 was a rework of anniversaries. Trust is earned, and 2017 has allowed It marked 20 years of 4D-Var at ECMWF, us to pay our dues and serve our Member and of seasonal forecasts, and 25 years of States and other users with some noticeable ensemble prediction and wave forecasting. progress towards our Strategy to 2025. It was also the year when ECMWF’s Council Some good examples include the work approved the development of our future undertaken to assess the best ensemble data centre in Bologna, Italy, and voted con gurations to make a 5 km ensemble a budget allowing a substantial increase affordable; the more ef cient radiation in our computing capability. The end of the scheme, completely recoded to be more year saw the Government of the UK, our exible and delivering ef ciency gains up host nation, formally offering to relocate to 30-35%; the release of the new seasonal ECMWF’s headquarters to accommodate system, SEAS5, notably improving our requirements. El Niño prediction.

A common denominator between all those Our partnership with the European Union events is that they illustrate the relationship is also thriving and delivering results that based on trust which connects ECMWF cut across the whole of ECMWF. The new and its Member States. The science behind climatology derived from aerosol data ensemble prediction or 4D-Var was not born developed by the EU-funded Copernicus in one day. It took years and years before Atmosphere Monitoring Service that we it even looked like something which could operate on behalf of the EU is contributing be adequately demonstrated. Developing to improving our forecasts. The latest a state-of-the-art data centre is also version of our reanalysis, ERA5, produced a substantial investment in the future. by our Copernicus Climate Change Service These are strong statements of support now provides a useful benchmark and from the Member States. reference for identifying interannual

2 ECMWF Annual Report 2017 ecmwf.int

AR-2018-ARTWORK.indd 2 08/06/2018 16:01 Day Probabilistic skill 7.5 ECMWF’s duty to Member 7 States is to continue

6.5 improving medium-range forecast skill. Results for 6 the northern hemisphere

5.5 extratropics show that the skill of the ensemble 5 forecast in predicting 24-hour 4.5 precipitation totals continues

4 to increase. The computation of skill is based on the 3.5 continuous ranked probability

3 skill score (CRPSS). The chart shows 12-month running 2.5 average values of the forecast 2 range at which the CRPSS 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 drops below 0.1.

variations in predictability; last but not Andy joined us one year into the least, this strong partnership with the EU implementation of the Strategy to 2025, is reinforced by the work they have selected with sizeable challenges to address … and us to deliver on re prediction and human an impressive scienti c team to support him. carbon emissions. Science in numerical weather prediction Our work with the WMO continued to could not function any more without ourish in 2017, with Nguyen Thanh Tung the support of computing sciences and from Vietnam nishing his fellowship in technology, so Council’s approval of our next September after 12 months of mutually HPC budget at its December session was bene cial exchanges. This year the a great outcome. Six months after approving WMO designated ECMWF as a World the building of our new state-of-the-art Meteorological Centre and endorsed data centre in Bologna, Council con rmed us as the Lead Centre for Wave its con dence in our teams by endorsing Forecast Veri cation. our request for a substantial increase in computing capability. As we continue to ECMWF is above all a scienti c organisation, push our science towards the goals of so the arrival of Dr Andy Brown as Director our Strategy to 2025, there could not of Research is also a milestone of 2017. have been a better way to end 2017. © Stephen Shepherd photography © Stephen Shepherd

ecmwf.int ECMWF Annual Report 2017 3

AR-2018-ARTWORK.indd 3 08/06/2018 16:01 2017 At a glance

Windstorm Doris January On 23 February, Storm Doris hit Ireland and Great Britain, causing disruptions CAMS model upgrade and at least one fatality. The risk of Changes to the aerosol model of the extreme gusts was predicted well Copernicus Atmosphere Monitoring 6 –7 days in advance, but there were Service (CAMS) global forecasting still uncertainties in the details in the system operated by ECMWF on behalf forecast one day before. of the EU resulted in more accurate forecasts. The upgrade introduced two new satellite datasets: aerosol March observations from the Polar Multi- Earlier dissemination Sensor Aerosol Product (PMAp) Open data events of ensemble forecasts provided by EUMETSAT and ozone ECMWF marked International On 7 March, ECMWF started observations from the Ozone Mapping Open Data Day with a week of events disseminating medium-range Pro ler Suite (OMPS) provided by that included an open data hackathon probabilistic forecasts to its users 40 NOAA through EUMETSAT’s third-party and several workshops. The events minutes earlier than before. The move data programme. These will help to aimed to raise awareness of freely was made possible by an upgrade of de ne the initial state on which available data from ECMWF and ECMWF’s supercomputers in 2016. forecasts are based. the EU Copernicus programme Forecasts are now available as early and of the socio-economic impacts as 0800 UTC. This can be bene cial of numerical weather prediction to users who need to make weather- February (NWP) data. based decisions early in the day. Fire danger reanalysis goes online Copernicus re danger reanalysis products from the Global ECMWF Fire Forecast (GEFF) system became available online, providing historical records of global re danger conditions from 1980 to today. The products were developed at ECMWF over the last three years as part of the EU-funded Copernicus Emergency Management Service. They complement other Copernicus products related to re, such as the biomass-burning emissions made available by CAMS.

4 ECMWF Annual Report 2017 ecmwf.int

AR-2018-ARTWORK.indd 4 08/06/2018 16:01 © Maria-Helena Ramos

Predicting weather impacts Support for WMO Scientists, developers and emergency May Year of Polar Prediction managers from across Europe came ECMWF launches eLearning As the World Meteorological together at ECMWF to drive forward Organization (WMO) launched its To meet the training needs of its the three-year ANYWHERE project, Year of Polar Prediction (YOPP), Member and Co-operating States, funded under the EU Horizon 2020 ECMWF began work to generate ECMWF launched the rst of its Programme. The project aims to an extended two-year dataset publicly available eLearning modules, develop forecasts of weather impacts to support the programme. which give learners exibility in terms in Europe, such as damage from of when, where and how they want to windstorms, oods and heatwaves. learn. The new modules are enabling ECMWF’s key role in the project OPP D the Centre to shorten face-to-face F Y AT is to provide weather forecasts A training events and reduce costs W S in support of other partners’ M for Member States. E C modelling applications. T

Ten years of atmospheric E

T C C

E E M composition forecasts M

April S W W

A A

ECMWF’s rst forecast of atmospheric A

T T T T T

F F

Y Y

A A A A

O D D P Support for Peru ood response composition in May 2007 paved the P ECMWF forecast data were rapidly way for the Copernicus Atmosphere made available to Peru’s national Monitoring Service (CAMS), which meteorological and hydrological held its second General Assembly service (SENAMHI) as heavy rainfall in Warsaw, Poland, in May this year. and ooding swept across the country during the early months of 2017. Forecasts, including our new point- rainfall product, helped SENAMHI forecasters to issue warnings of heavy rainfall and highlight events that might hinder rescue operations.

ecmwf.int ECMWF Annual Report 2017 5

AR-2018-ARTWORK.indd 5 08/06/2018 16:01 2017 At a glance

in Europe and further a eld. In hands- on sessions, they investigated what the weather would have been had the strong El Niño of 2015/16 been different.

Bologna to host data centre At the 90th session of the ECMWF Council, Member States approved the proposal by the Italian Government and the Emilia-Romagna Region to host ECMWF’s new data centre in Bologna, Italy, by 2019.

July Model upgrade improves Symposium on dynamical forecast quality June meteorology and NWP ECMWF’s Integrated Forecasting Leading WMO of cials and experts in User meeting focuses on storms System (IFS) was upgraded to IFS NWP came together at ECMWF for a Cycle 43r3. The changes included More than 100 forecasters and scientists symposium to mark the departure of improvements in the modelling of attended the Using ECMWF’s Forecasts Erland Källén from the Centre after convection, a new radiation scheme (UEF) meeting, where the theme this year eight years as Director of Research. and aerosol climatology, and better use was ‘Storms’. It focused on areas such Delegates heard about new results in of dropsonde and other observations. as the processing of model outputs to dynamic meteorology, operational NWP The upgrade brought a range of support the forecasting of severe storms and climate science as well as about improvements in forecast skill and in and associated weather phenomena; the Centre’s past, present and future. diagnostics involving tools or studies that tropical cyclone forecasts in particular. highlight strengths and weaknesses of OpenIFS user workshop ECMWF’s Integrated Forecasting System New Research Director The fourth OpenIFS user workshop, (IFS) in predicting storms; and the impact takes up post held in Trieste, Italy, was devoted to of storms on sectoral applications. Dr Andrew Brown, previously Director ‘Atmospheric Variability: seasonal of Science at the UK Met Of ce, predictability and teleconnections’. became ECMWF’s Director of Research. It attracted scientists from institutes

6 ECMWF Annual Report 2017 ecmwf.int

AR-2018-ARTWORK.indd 6 08/06/2018 16:01 © Mike Watson Images/moodboard/Thinkstock © Mike Watson

ECMWF Fellowships renewed ECMWF designated World The three original ECMWF Fellows, Meteorological Centre September Tim Palmer, Tilmann Gneiting and ECMWF was designated as a World Royal Society Meteorological Centre (WMC) by the Rupert Klein, began second terms Research Fellowship of three years. WMO. WMCs have to meet a range of standards on data, products, ECMWF scientist Peter Düben was awarded a prestigious Royal Society Powerful climate monitoring documentation and training. University Research Fellowship to dataset from Copernicus They provide WMO Members with a range of forecast products. continue his research into uncertainty The Copernicus Climate Change and the computational ef ciency of Service (C3S), operated by ECMWF NWP. The ve-year fellowship is Heatwave in Europe on behalf of the EU, released the intended for outstanding early-career rst part of a powerful global climate The unusually high temperatures in scientists who have the potential monitoring dataset, known as the ERA5 early August, across much of central to become leaders in their eld. climate reanalysis. This initial release and southern Europe, were well of ERA5, covering 2010 to 2016, made forecast by ECMWF. Potential Annual Seminar on a vast amount of climate data freely heatwave conditions were evident ensemble prediction and openly available. Data include around four weeks in advance and air temperature, pressure and wind short-range forecasts correctly More than 100 people from across the at different altitudes, and surface predicted the severity of the event. world joined ECMWF scientists and 28 parameters such as rainfall and sea ice. speakers at ECMWF’s Annual Seminar on ‘Ensemble prediction: past, present and future’ to discuss how to further August improve the accuracy and reliability of ensemble-based forecasts. Co-operation agreement with INPE, Brazil ECMWF and Brazil’s National Institute for Space Research (INPE) signed a formal co-operation agreement on 1 August. This came after many years of working together, especially in the development of the Metview meteorological workstation application.

ecmwf.int ECMWF Annual Report 2017 7

AR-2018-ARTWORK.indd 7 08/06/2018 16:01 2017 At a glance

Test phase for new interpolation package ECMWF’s new interpolation package, MIR, was made available to ECMWF users for testing. MIR is a library of routines for the interpolation and regridding of meteorological elds. It illustrates how NWP research and software/hardware development must go hand in hand. Its exible design will allow it to be adapted to future developments.

November Anniversaries and celebrations 2017 provided ample opportunity to reect on scienti c achievements and progress. The year marked 25 years of ECMWF’s production of ensemble forecasts, 25 years of wave forecasting at ECMWF, 20 years since ECMWF implemented a pioneering four- dimensional variational data assimilation technique (4D-Var), CO Human Emissions 2 and 20 years of producing (CHE) project October seasonal forecasts. ERA5 reanalysis simulates October saw the start of a major Great Storm of 1987 project led by ECMWF to explore the Towards global modelling development of a European system at resolutions under 10 km The power of the new ERA5 climate to monitor human-related carbon The ECMWF Strategy aims for global reanalysis was demonstrated with dioxide (CO ) emissions across the a detailed, hourly simulation of the 2 ensemble forecasts at a resolution world. Such capacity is vital to support of 5 km by 2025 and this requires evolution of this notoriously dif cult- Europe’s leading role in worldwide to-capture storm over Western Europe, new ways of representing small-scale action to address climate change. processes, such as convection, including the sting jet that made The project brings together a it so destructive. in forecasting models. A workshop consortium of 22 European partners brought together 50 scientists to foster and will last for over three years. international collaboration and discuss how best to make progress on the modelling of physical processes in the grey zone of horizontal resolutions ranging from 1 to 10 km grid-spacing. © arquiplay77/iStock/Thinkstock

8 ECMWF Annual Report 2017 ecmwf.int

AR-2018-ARTWORK.indd 8 08/06/2018 16:01 New EFAS and GloFAS ocean. SEAS5 brings better forecast ood forecasting products skill for periodic warming/cooling events New products became operational in the tropical Paci c (El Niño/La Niña) in the European Flood Awareness and other improvements. The forecast System (EFAS, part of the Copernicus data is also available through C3S Emergency Management Service): as part of the multi-model seasonal now-casting ash ood forecasting forecast service that combines data based on radar images, and a Rapid from SEAS5, the UK Met Of ce Risk Assessment tool. The Global and Météo-France. Flood Awareness System (GloFAS) also launched new seasonal forecast products. Developed jointly with the December University of Reading, GloFAS seasonal ECMWF to contribute provides global hydrological forecasts to EU re project out to four months. ECMWF became the computational SEAS5 brings better centre for the Copernicus Emergency El Niño forecasts Management Service – Fire. ECMWF will provide re danger calculations A new seasonal prediction system from high-resolution and ensemble (SEAS5), which gives an indication of Symposium for Adrian Simmons forecasts up to 15 days ahead on average, large-scale weather conditions a daily basis. The development of A symposium celebrated the lifelong out to 13 months ahead, was launched a long-range forecast is also planned work of ECMWF scientist Adrian by ECMWF. The new system includes and a pre-operational suite will be Simmons, who has made major an upgraded ocean model and a model made available. contributions to progress in weather which allows sea-ice cover to respond forecasting, climate science and to changes in the atmosphere and atmospheric composition.

2017 extends exceptionally warm period C3S datasets show temperatures for 2017 were higher than the 1981–2010 climatological average over most regions of the world. They were most above average over parts of the Arctic. 2017 was close to the warmest year on record despite cooling La Niña conditions both early and late in the year over the equatorial eastern Paci c Ocean. The warmest months of 2016 © Copernicus Climate Change Service, ECMWF occurred in the declining phase of a strong El Niño, which also inuenced temperatures in 2015, making 2017 the warmest year on record that was not inuenced by an El Niño. Temperature difference between 2017 and 1981-2010

ecmwf.int ECMWF Annual Report 2017 9

AR-2018-ARTWORK.indd 9 08/06/2018 16:01 Advancing weather science

ECMWF carries out world-leading SEAS5 research into all aspects of A seamless forecast numerical weather prediction. In November 2017, ECMWF implemented system makes The advances made are a new seasonal forecasting system, SEAS5. Seasonal forecasts provide predictions up forecasts more incorporated into operations to to 13 months ahead. They describe how consistent across improve the Centre’s forecasts for the atmospheric, ocean and land surface different time ranges. the bene t of users. Collaboration conditions over particular areas and periods of time are likely to be different from the with space agencies, national long-term average. SEAS5 includes several meteorological services and the upgrades in the ocean model, atmospheric research community worldwide is resolution and land surface initialisation. a key part of this research effort. The changes have led to better predictions in the tropics and the Arctic compared Developing new science takes time. In 2017, to SEAS5’s predecessor, S4. ECMWF scientists brought several projects In developing SEAS5, care was taken to to a successful conclusion, such as the move towards a seamless approach to development of a new seasonal forecasting forecasting across timescales, in line with system, SEAS5; the production of a new ECMWF’s Strategy to 2025. This means that Improved forecast skill coupled satellite-era reanalysis; changes to SEAS5 is con gured almost identically to Anomaly correlation for the warm-rain microphysics parametrization extended-range ensemble forecasts, which ensemble mean 2-metre to improve the representation of precipitation have lead times of 15 to 45 days. A seamless temperature predictions for near land–water boundaries; and a range of forecast system makes forecasts more December–January–February improvements in the use of atmospheric in situ consistent across different time ranges. from 1 November for S4 (left) observations, including the ability to account It also makes it easier to identify and x and SEAS5 (right). for radiosonde drift. Examples of substantial problems, to the bene t of forecast quality Improvements can be seen progress in other areas include research into at all timescales. in the tropics and the Arctic. how coupling the atmosphere with the ocean Anomaly correlation is a improves tropical cyclone predictions; tests measure of skill. A value of of different ensemble forecast con gurations; 1 corresponds to a perfect and the evaluation of a new satellite-based deterministic forecast, sea-ice thickness product. while 0 means no skill.

-1 -0.9 -0.8 -0.7 -0.6 -0.4 -0.2 0.2 0.4 0.6 0.7 0.8 0.9 1 -1 -0.9 -0.8 -0.7 -0.6 -0.4 -0.2 0.2 0.4 0.6 0.7 0.8 0.9 1

10 ECMWF Annual Report 2017 ecmwf.int

AR-2018-ARTWORK.indd 10 08/06/2018 16:01

-1 -0.9 -0.8 -0.7 -0.6 -0.4 -0.2 0.2 0.4 0.6 0.7 0.8 0.9 1 TC Neoguri, uncoupled TC Neoguri, coupled

100 100 20°N 20°N

120°E 140°E 120°E 140°E

100 200300 400500 600800 880 880900 920940 960970 980990 1000 1100 Net surface heat flux (W/m2 ) Central pressure (hPa) Forecast Best track

Typhoon Neoguri is an example of a TC Track and intensity Ocean coupling in which a shallow warm layer in the ocean forecasts was quickly depleted, causing Neoguri to Tropical cyclones (TCs) are one of the Five-day high-resolution develop less deep central pressure than deadliest weather phenomena, but some track and intensity forecasts might otherwise be expected. ECMWF’s of their features can be tricky to predict. (squares) starting on 5 July coupled model is able to represent 2014 together with ‘best Research at ECMWF has shown that these interactions. track’ estimates (triangles) interactions between the ocean and the and the predicted net surface atmosphere can have a decisive in uence on heat ux (shading) for the evolution of TC intensity. Heat exchanges Typhoon Neoguri using the between these two Earth system components uncoupled model (left) and need to be taken into account to avoid the coupled model (right). overpredicting TC intensity in some cases.

The main energy source for TCs is heat transport from the ocean. Not coupling the atmosphere and the ocean in the forecast means that the ocean acts as an undiminished source of energy for the atmosphere during the forecast period. This allows TC intensity in the forecast Satellite image to increase unrealistically if the warm Typhoon Neoguri on top layer of the ocean is shallow. 4 July 2014. © EUMETSAT

ecmwf.int ECMWF Annual Report 2017 11

AR-2018-ARTWORK.indd 11 08/06/2018 16:01 Advancing weather science

Mean Sea Level Pressure in Tropical Cyclone Centre (hPa) solid =HRES; dot =Ens Mean

1020 1010 Hurricane Irma 1000 intensity forecasts 990 980 Forecasts from 5 September 970 of central pressure for 960 950 Hurricane Irma using the (hPa) 940 current 18 km operational 930 con guration (top) and a 5 km 920 910 ensemble (bottom). The 900 higher-resolution forecasts 890 Sep 2017 Forecast valid date match best-estimate values much better than the 1020 operational con guration. 1010 1000 990 980 970 960 950 (hPa) 940 930 920 910 900 890 Sep 2017 Forecast valid date

Ensemble forecast Ensemble mean Best estimate

Experiments at ECMWF have con rmed Ensemble conguration that a 5 km ensemble brings clear bene ts compared to today’s 18 km ensemble. It An ensemble forecast is a set of forecasts would, for example, have produced much 18g5 km that represent the range of possible future better forecasts of the intensity of Hurricane Experiments have con rmed weather evolutions given the observed Irma, which struck several Caribbean islands that a 5 km ensemble brings current state of the Earth system. A large and Florida in September 2017. Severe clear bene ts compared to ensemble size and high spatial resolution oods in Livorno, Italy, in September 2017 today’s 18 km ensemble. are generally advantageous, but they are would also have been better predicted. also computationally expensive. Meanwhile, it is not clear that the current As ECMWF moves towards the goal of 5 km operational ensemble size of 50 should be global ensemble forecasts set in its Strategy reduced to make global 5 km forecasts to 2025, there is a need to investigate affordable. Experiments show that signi cant the relationship between ensemble size, reductions in ensemble size are liable to spatial resolution, forecast quality reduce probabilistic skill in the medium range. and computational affordability. Conversely, further increases in skill appear possible if ensemble size is increased beyond the current operational con guration.

12 ECMWF Annual Report 2017 ecmwf.int

AR-2018-ARTWORK.indd 12 08/06/2018 16:01 0.05 Forecast skill for 30 km, 50 members different ensemble con gurations 18 km, 20 members Difference in CRPS (a measure of forecast 0.03 18 km, 50 members skill) relative to a 50-member 30 km, 200 members 30 km resolution ensemble, for three different combinations of horizontal resolution and 0.01 ensemble size. Lower values

CRPS mean higher relative skill. The black line shows the current operational con guration. The blue line -0.01 shows that, for some lead times at least, a larger ensemble size can improve medium-range forecast quality even at a lower spatial -0.03 resolution. The forecasts 0 3 6 9 12 15 are for 850 hPa temperature veri ed against analyses, Lead time (days) in the northern extratropics June–July–August 2016. the representation of Earth system Thin sea ice interactions in the Integrated Forecasting System (IFS). Sea-ice thickness is important for medium- range to seasonal predictions because thin First evaluation results show encouraging ice evolves much more quickly than thick similarities between observations and ice. Small differences in ice thickness at the ECMWF’s OCEAN5 analysis although beginning of summer can also make a large there are also some regional discrepancies. difference to the timing of its complete The next step would be the development disappearance during the melt season. of sea-ice data assimilation schemes which can fully exploit the sea-ice In 2017 ECMWF routinely obtained and thickness information provided by satellite processed a product (SMOS-SIT) based on observations to arrive at an improved satellite observations for sea-ice thicknesses sea-ice analysis. This work will provide of up to about 1 metre from the University important building blocks in efforts to of Hamburg. This type of innovative product improve predictions in the polar regions. should in the future help ECMWF to improve

1 1 Sea-ice products 0.9 0.9 Daily-mean elds for 0.8 0.8 16 April 2017 showing 0.7 0.7 sea-ice concentration from

0.6 0.6 the OSTIA product (left) and sea-ice thickness from the 0.5 0.5 (m) SMOS-SIT product (right). 0.4 0.4 The thickness plot provides a 0.3 0.3 lot more detail on the amount 0.2 0.2 of sea ice in some areas than

0.1 0.1 the sea-ice concentration plot alone. The area around the 0 0 North Pole is greyed out because of a lack of satellite observations in this region.

ecmwf.int ECMWF Annual Report 2017 13

AR-2018-ARTWORK.indd 13 08/06/2018 16:01 Advancing weather science

IASI temperature IASI humidity Differences in background departures Better Worse The plots show differences 10 between CERA-SAT and 200 an uncoupled control Better Worse experiment. Values smaller 30 than one indicate an 400 improved t of 24-hour forecasts to temperature and 100 humidity observations from 600 the IASI satellite instrument.

Peak pressure (hPa) Peak pressure (hPa) The comparison spans the 300 full year from September 800 2015 up to and including August 2016. Bars indicate 95% con dence intervals. 1000 1000

0.995 1.005 0.995 1.005 Normalised standard deviation

Tropics Extratropics

observations are assimilated in a consistent Twenty years of 4D-Var manner. The production of CERA-SAT as part of the EU-funded ERA-CLIM2 project CERA-SAT is a 2017 marked 20 years of ECMWF’s use of is thus a milestone on the road towards milestone on the the 4D-Var data assimilation technique to a strategic goal: the implementation of initialise forecasts. The method is particularly road towards a an Earth system approach in all parts well suited for the assimilation of satellite strategic goal: of ECMWF’s forecasting system. data and has brought major improvements the implementation in forecast skill. In 2017, it was successfully Experiments show that in the tropics the extended to a coupled system in the of an Earth system differences between observations and production of a satellite-era research approach in all parts 24-hour forecasts are generally smaller reanalysis called CERA-SAT. in CERA-SAT than in uncoupled control of the forecasting experiments, indicating an improved system. CERA-SAT reconstructs the state of the reanalysis. Work is under way to improve atmosphere, the ocean, sea ice, ocean waves results in the extratropics, where CERA-SAT and the land surface between 2008 and 2016. generally performs less well. It is coupled in that atmospheric, ocean, sea-ice, ocean-wave and land-surface

14 ECMWF Annual Report 2017 ecmwf.int

AR-2018-ARTWORK.indd 14 08/06/2018 16:01 Ensemble forecasts can provide indices of the risk of severe events and probabilities of the occurrence of weather events.

seasonal timescale, but also to provide Twenty-ve years of estimates of the initial state of the Earth ensemble forecasting system. These ensemble-based forecasts and analyses contain more complete at ECMWF information than single, deterministic 2017 was the 25th anniversary of the year forecasts. For example, they can provide when the Centre started to issue operational indices of the risk of severe events; ensemble forecasts. The availability of such probabilities of the occurrence of weather forecasts marked a paradigm shift in events; the range of possible values at weather prediction: for the rst time, speci c locations; alternative weather forecasters and users were provided with scenarios; and weekly-mean anomalies. reliable and accurate estimates of the range of possible future scenarios, and not just Thanks to model upgrades, improvements with a single realisation of the future. in the data assimilation system, the use of more observations, and ensemble Today ensembles are used not just con guration changes, the performance to provide forecasts for the short- and of ensemble forecasts has increased medium-range, the monthly and the substantially during the past 25 years.

0.9

0.8

0.7

0.6

0.5 CRPSS 0.4

0.3

0.2

0.1

0 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017

T+240 hours T+168 hours T+120 hours T+72 hours T+24 hours

Ensemble forecast skill evolution Time evolution, from 1 January 1995 to 2017, of the Continuous Ranked Probability Skill Score (CRPSS) of ensemble forecasts of 500 hPa geopotential height over the northern hemisphere, for lead times of 24 hours, 72 hours, 120 hours, 168 hours and 240 hours. Forecasts are veri ed against operational analyses. A CRPSS of 1 is the best value possible. The more or less regular pattern of peaks and troughs in each line stems from differences in predictability related to the seasons: winter weather tends to be more predictable than summer weather.

ecmwf.int ECMWF Annual Report 2017 15

AR-2018-ARTWORK.indd 15 08/06/2018 16:01 Delivering global predictions

Hurricane Ophelia In October Hurricane Ophelia transported a mixture of smoke, dust and sea salt aerosol across Europe leading to red skies. © Inner_Vision/iStock/Thinkstock

To keep improving forecasts, The Centre also extended its monitoring ECMWF systematically system to cover ocean observations and continues to work closely with the World evaluates the performance Meteorological Organization (WMO) on of its operational forecasts, forecast veri cation and developing the engages with users and applies global observing system.

the latest scienti c advances. Veri cation results show that the skill Across Europe, the potentially of the Centre’s medium-range forecasts, devastating impact of extreme as measured by a set of headline scores, weather such as heatwaves and has continued to improve. To help assess progress towards the goals set out in the windstorms was felt keenly in Centre’s Strategy to 2025, two additional 2017, underlining the importance headline scores have been agreed for the of improving forecasts to help ensemble forecast. protect life and property. Timeliness is also important: on 7 March, ECMWF started disseminating medium- In 2017 ECMWF implemented a substantial range probabilistic forecasts to its users upgrade of its Integrated Forecasting 40 minutes earlier than before. This earlier System (IFS), bringing signi cant availability means users in ECMWF’s improvements in forecast skill. Cycle 43r3 Member and Co-operating States can now included changes in the model; in the way use the most recent 0000 UTC forecast in observations are used; in software key applications instead of having to rely infrastructure; and in the assimilation on the previous day’s 1200 UTC forecast. procedure used to generate the initial conditions for forecasts.

16 ECMWF Annual Report 2017 ecmwf.int

AR-2018-ARTWORK.indd 16 08/06/2018 16:01 The science in Cycle 43r1 contributed to a better representation of Earth system processes.

a) Zonal wind bias (previous) b) Zonal wind bias (new) Wind bias Bias in the day-5 forecast of 925 hPa zonal wind in the Indian monsoon region in summer (June–August) for (a) the previous IFS cycle, 43r1, and (b) IFS Cycle 43r3 with the new aerosol climatology. Saturated colours indicate areas where the signal is signi cant at the 95% con dence level. -33 -15 -9 -3 3 9 15 39 -27 -15 -9 -3 3 9 15 33 (0.1 m/s) (0.1 m/s)

Forecasting system upgrade New aerosol climatology IFS Cycle 43r1 became operational on A new climatology derived from data 11 July, enabling the use of more observations provided by the Copernicus Atmosphere and improving their assimilation. Changes in Monitoring Service (CAMS) has led to an data assimilation and in the way dropsonde improved representation of the Indian observations are handled improved the summer monsoon. A reduction in the accuracy of the initial conditions on which absorption of shortwave radiation over forecasts are based, especially for tropical Arabia leads to less solar heating and hence cyclones. Model changes included a new a reduction in the strength of the Arabian radiation scheme, improvements in the heat low. This in turn reduces the bias in modelling of convection, and a new aerosol westerly wind into India by around 25%, climatology, all contributing to a better which halves the previous overestimate representation of Earth system processes. of rainfall over the west coast of India. These advances represent essential steps towards the Centre’s goals and have led to signi cant improvements in forecast skill in medium-range and monthly forecasts.

Looking ahead, the implementation of future model upgrades will be streamlined as a result of software infrastructure updates made in this model cycle. Notably, these updates will facilitate further work on single precision, which is expected to make an important contribution to the Centre’s Scalability Programme.

ecmwf.int ECMWF Annual Report 2017 17

AR-2018-ARTWORK.indd 17 08/06/2018 16:01 Delivering global predictions

7 Headline score The chart shows the lead

6.5 time at which the anomaly correlation for 500 hPa geopotential in the northern

6 hemisphere extratropics

drops below 80% in ya

D ECMWF’s high-resolution

5.5 forecast (HRES) and in those of other global forecasting CMC ERA5 centres. ERA-Interim JMA ERA-Interim 5 and ERA5 are shown UKMO NCEP for reference. ECMWF DWD

4.5 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

ensemble forecast skill for precipitation Forecast quality continues to be at a CRPSS skill score level of at least 10% out to day 7. Model upgrades in November 2016 and July 2017 have enabled ECMWF to remain The 12-month running mean value of the the leading centre in terms of overall ensemble forecast skill for temperature medium-range forecast skill. The ERA5 at 850 hPa in the northern hemisphere reanalysis produced by the EU-funded extratropics is now consistently at a Copernicus Climate Change Service now CRPSS skill score level of at least 25% provides a useful benchmark and reference out to 9 days, which is an increase of for the identi cation of interannual variations more than a day over the last decade. in predictability, replacing ERA-Interim in this regard. Tropical cyclone track errors were the smallest ever in 2017 for the high-resolution The resolution upgrade from 32 to 18 km for and ensemble forecasts. Improvements are the ensemble forecast in 2016 contributed expected in forecasting tropical cyclone to a noticeable increase in skill. Veri cation intensity in future model cycles. scores for precipitation forecasts reached their best-ever levels in 2017, and several Member and Co-operating States reported positive feedback on this improved quality.

Headline scores for surface parameters focus on precipitation for both the high-resolution and ensemble forecast. The 12-month running mean value of the

18 ECMWF Annual Report 2017 ecmwf.int

AR-2018-ARTWORK.indd 18 08/06/2018 16:01 New headline scores Monitoring and verication The new scores In April, the ECMWF Technical Advisory The WMO designated ECMWF as a World measure forecast Committee (TAC) Subgroup on Veri cation Meteorological Centre in May, acknowledging Measures recommended the introduction the Centre’s compliance with the standards skill in the medium of two new headline scores for monitoring set out in the Global Data-Processing and and extended ranges ensemble forecast skill in the medium Forecasting System manual, including for and were selected and extended ranges. product veri cation and forecast performance. It also endorsed ECMWF as being most The additional score for the medium range as the Lead Centre for Wave Forecast relevant to users. will be based on the frequency of large Veri cation. ECMWF’s main activity will be 2 m temperature errors as measured by to collect ocean wave forecasts from global the continuous ranked probability score and regional centres and to evaluate them (CRPS) exceeding a given threshold. using a common set of observations. For the extended range, the new headline score will be the skill in predicting weekly ECMWF is also contributing to a WMO means of 2 m temperature anomalies as project to develop new near-real-time measured by the ranked probability skill monitoring for the global observing system, score applied to terciles or higher quantiles. the WIGOS Data Quality Monitoring System Both scores will be computed for the (WDQMS). This will provide much more northern extratropics. The new scores are ef cient feedback to observation providers guided by the needs of our Member and through modernised procedures. Co-operating States and contribute to the overall evaluation of progress towards To support coupled data assimilation ECMWF’s strategic goals. developments, in 2017 ECMWF extended its data monitoring system to cover ocean observations. This will provide vital information on the availability and quality of observations used in the ocean data assimilation system, leading to quick reaction to anomalies in the observations, better understanding of model errors and, ultimately, better Earth system predictions.

8 New headline score in the medium range The plot shows the 12-month running mean percentage of

7 continuous ranked probability score (CRPS) values for 2 m temperature exceeding 5 K at day 5 in the extratropics (poleward of 30° latitude), 6 veri ed against surface synoptic observations.

5 Percentage of large errors

4 2005 2007 2009 2011 2013 2015 2017 Year

ecmwf.int ECMWF Annual Report 2017 19

AR-2018-ARTWORK.indd 19 08/06/2018 16:01 Delivering global predictions

Fire risk Summer heatwave and Extreme heat conditions wildres in southern Europe increase the risk of forest res. The summer of 2017 was dominated by hot temperatures in southern Europe but cold conditions in the north. Especially during June, the Iberian Peninsula was severely affected by wild res, and a devastating re near Coimbra, Portugal, claimed more than 60 lives.

Since 2012 ECMWF has been producing probabilistic re forecasts as part of the EU-funded activities of the Copernicus Emergency Management Service. The Global ECMWF Fire Forecasting (GEFF)

system provides daily global re danger © A.Mateus Photography/iStock/Thinkstock levels to 10 days ahead. The Instituto Português do Mar e da Atmosfera (IPMA) For the week of 12–18 June, there was a gained access to GEFF data at the signi cant heatwave over south-western beginning of the summer season. It reported Europe. The forecast captured the structure that GEFF products warned days in advance of the positive temperature anomaly well of the establishment of extreme danger up to 8–14 days in advance. Even longer conditions in the area and contributed to the forecasts predicted a likelihood for the better planning and the fast response of the anomaly, indicating that the initial or crisis units. More lives might otherwise have boundary conditions of the forecast been lost. favoured a warm period.

Temperature anomaly forecasts The charts show the analysis of two-metre temperature anomalies for 12–18 June 2017 (left) and the forecasts for two-metre temperature anomalies for the same period 5–11 days ahead (middle) and 26-32 days ahead (right).

ANALYSISANALYSIS DayDay 5-11 5-11 DayDay 26-32 26-32

<-10 ˚C -10.. -6 -6.. -3 -3.. -1 -1.. 0 0.. 1 1.. 3 3.. 6 6.. 10 > 10 ˚C <-10 ˚C -10.. -6 -6.. -3 -3.. -1 -1.. 0 0.. 1 1.. 3 3.. 6 6.. 10 > 10 ˚C

20 ECMWF Annual Report 2017 ecmwf.int

AR-2018-ARTWORK.indd 20 08/06/2018 16:01 provided a consistent signal for a dangerous Severe windstorms windstorm, which grew stronger with time. On the night of 28 October a deepening Earlier in the year, in August, a severe cyclone, named Herwart by the Free windstorm hit northern Poland, causing the University of Berlin, moved southeastwards deaths of ve people, signi cant damage across southern Sweden and the southern to trees and power disruptions affecting Baltic Sea. Within the large circulation of this 340,000 households. The risk of severe cyclone, very strong winds developed, most convective hazards in the affected region notably over Germany, the Czech Republic was captured in ECMWF medium-range and Poland. There were at least four fatalities, forecasts by the Extreme Forecast index damage to trees and buildings and disruption and Shift of Tails product for a composite to infrastructure. According to reports in the parameter that combines CAPE and European Severe Weather Database (ESWD), wind shear. However, ECMWF wind gust the Czech Republic was probably worst forecasts failed to predict anything near affected. The short-range Extreme Forecast the observed values. Index (EFI) for wind gusts agreed well with those reports. Even at a lead time of six days, These two cases illustrate very different the ECMWF EFI and Shift of Tails (SOT) skill for the same variable (wind gusts). clearly highlighted a greatly elevated risk of a The meteorological conditions behind severe wind event over a large area. Indeed, the two events were very different, and the strongest signal in the EFI was centred on such differences should be borne in mind the Czech Republic. Throughout the lead-up when interpreting veri cation results. to this event, ECMWF ensemble forecasts

Storm Herwart

0.3 0.3 Ensemble-based forecasts

0 for day 6 of Extreme Forecast 0

0 0 Index (EFI, shaded) and Shift of Tails (SOT, black contours)

0 0 0 for maximum 10-metre wind 0

0.3 0.3 1 gusts during 29 October 2017, 1 highlighting an elevated risk of a severe wind event over a large area.

1 1

1 0 1 0

0 0 0 0

0.3

0.3 0 0 0.5 0.5 0.60.6 0.70.7 0.80.8 0.9 0.9 1 1

ecmwf.int ECMWF Annual Report 2017 21

AR-2018-ARTWORK.indd 21 08/06/2018 16:01 Sustaining high-performance computing

Collaboration between the HPC and Scalability teams aimed to maintain focus on technological advances and sustainable efﬁciency gains. © A. Brookes/ECMWF Copernicus © A. Brookes/ECMWF

In 2017 ECMWF’s high- to ensure that we will get the most suitable performance computing facility technology and best value for money. (HPCF) consisted of two Cray We also started to set up a framework XC40s with 260,000 processor contract to purchase the computer storage cores and more than 900 systems that will support the data centre both in the current location in Reading terabytes of memory. These and in the future location in Bologna. two supercomputers have been used for operational forecasts A research and consultancy company carried out a study in 2017 as part of since June 2016 and are due the Business Case for the investment in to be replaced in 2020, in the HPC2020. The result of the study showed new data centre currently being very clearly the large bene ts that will be developed in Bologna, Italy. realised in our member countries from investing in the new HPC at ECMWF.

Providing ECMWF with new supercomputers One of the key aspects of 2017 in this is a long process which started when our eld was the necessary and effective close Council approved a substantial additional collaboration between HPC and Scalability investment for the next generation of teams. The aim was to maintain focus supercomputers at its session in December on both technological advances and 2017. The project, known as HPC2020, sustainable ef ciency gains. The impact kicked off with a Request for Information to of projects funded by the European Union vendors as part of the business case aiming supporting future and emerging technologies

22 ECMWF Annual Report 2017 ecmwf.int

AR-2018-ARTWORK.indd 22 08/06/2018 16:01 and E-Infrastructures, together with core In addition to signi cant work on information activities, has been noticeable. Already the security policies, work was carried out to rst generation of projects like ESCAPE, develop and deploy a new identity and 25% NextGenIO and ESiWACE have produced access management system due to be signi cant results, including: completed before the move to the new of ECMWF’s data centre in Bologna. supercomputing resources • the first head-to-head comparison of are allocated to the Centre’s present and future Integrated Forecasting Member States. System (IFS) model options on different Special Projects processor types ECMWF allocates 25% of supercomputing • the assessment of full single-precision resources to the Centre’s Member States. model runs, and Of these, up to 10% are reserved for Special • the development of a realistic workload Projects: experiments or investigations of a simulator for benchmarking. scienti c or technical nature, undertaken by one or more Member States, likely to be of The second generation of these projects interest to the general scienti c community. such as ESCAPE-2, MAESTRO and EPiGRAM-HS will continue this legacy In 2017, 8,250,000,000 System Billing Units in the coming years. (SBUs) and 31,800,000 gigabytes of data storage were made available for this activity. Towards the end of 2017, ECMWF started to develop a proposal for ExtremeEarth, During the year, 45 projects started in a preparatory action for a Flagship previous years continued, while 25 new programme with strong involvement from Special Projects from 10 different countries the weather and climate community as were added. These new projects included well as ECMWF Member and Co-operating investigations into the predictability of States. It will also include expertise from sea-ice conditions at very small scales; the food, water, energy, health, geophysics the inuence of microphysical processes and nancial risk management sectors. on the dynamics of weather systems; The outcome will be known in late 2018. and perturbation strategies for the development of convection-permitting On a different note, the year also saw an ensemble forecasting. unusual number of ransomware attacks, infecting thousands of systems around the world. Thanks to the effort of our teams, ECMWF Microsoft Windows systems stayed safe. Despite this success, information security remained a key focus of the Centre.

ecmwf.int ECMWF Annual Report 2017 23

AR-2018-ARTWORK.indd 23 08/06/2018 16:01 Sustaining high-performance computing

(positional image)

Single precision-30 -28 -24 -20 -16Double -12 precision -8 -4 0 4 8 12 16Analysis -30 -28 -24 -20 -16 -12 -8 -4 0 4 8 12 16 °C °C°C

Within the latest IFS cycle (43r3), single Single and double Single precision to save precision simulations with the forecast precision simulations HPC computing resources model achieved almost equivalent results Surface temperature in degree for both short- and long-term simulations Celsius for ve-day forecasts Almost all weather and climate models in research mode. This includes simulations for 8 January 2017 0000 UTC. use double precision on high-performance at the resolution of operational forecasts This date is during the computing hardware. In this method 64 bits (9 km and 18 km). European cold wave that are used to represent each real number and caused very low temperatures the numbers are represented with at least However, there are still some hurdles before in Eastern and Central Europe. 15 decimal digits precision. In single single precision can be implemented. It was Results are shown for single precision, 32 bits are used and at least found that mass conservation error was precision and double precision 6 decimal digits are available. signi cantly larger in single precision simulations at 9 km (TCo1279) simulations and differences in temperature resolution (left and middle) In 2017 ECMWF started experimenting using at 100 hPa as well as in surface temperature and the analysis as a single precision to improve computational were still measurable. reference (right). Differences ef ciency to make the most of current between single and double and future high-performance computing Once this technique is stable, the use precision are very small. hardware as part of the Scalability of single precision will potentially allow Programme. A single precision version signi cant savings of computing cost in of the IFS has been developed and during the near future and an increased throughput the year work to re ne it continued. on supercomputers.

24 ECMWF Annual Report 2017 ecmwf.int

AR-2018-ARTWORK.indd 24 08/06/2018 16:01 The RMDCN serves to ensure the secure and timely delivery of ECMWF forecasts to Member States and the exchange of meteorological observations.

This suite is owned by RHMS and monitored ecFlow suite for Serbia by ECMWF. It was developed under the Framework for Member State time-critical In 2017 the Republic Hydrometeorological applications. Service of Serbia (RHMS) set up a suite using ecFlow to run a forecast model using their share of resources on the ECMWF HPCF. Regional Meteorological Data The forecast model is based on the B-grid Communication Network version of the Nonhydrostatic Multiscale The Regional Meteorological Data Model. Initial data and lateral boundary Communication Network (RMDCN) provides conditions are taken from ECMWF high- a computer network infrastructure for resolution forecasts and disseminated to the the meteorological community in World ECMWF HPC. The suite runs twice per day. Meteorological Organization (WMO) Region VI and beyond. It was set up in 2000 and The NMM-B regional model has been provides any-to-any connectivity between operational at RHMS since January 2013. more than 50 sites. Among other things, Due to the lack of its own computer the RMDCN serves to ensure the secure resources, RHMS could not increase the and timely delivery of ECMWF forecasts model resolution and/or ensemble runs. to its Member States and the exchange Other operational applications of weather of meteorological observations between and hydrological forecasts are linked to the connected sites. Following an upgrade of NMM-B model, so it needs to be executed the connections of many sites, the network in a timely fashion and on a regular basis. is now nearly complete.

RMDCN global coverage By the end of 2017, 55 sites were connected to the RMDCN network. The shaded countries indicate ECMWF Member and Co-operating States.

ecmwf.int ECMWF Annual Report 2017 25

AR-2018-ARTWORK.indd 25 08/06/2018 16:01 Supporting ECMWF

A winning team Every year, ECMWF staff take part in the Green Park triathlon to help raise money for Sport Relief.

© gmp von Gerkan, Marg & Partner Supporting ECMWF is in Bologna. After discussions and votes, Bologna site about ensuring that the of ce representatives of ECMWF’s Member States Artist’s impression of the site were satis ed with the high-level agreement proposed in Bologna, Italy, accommodation, conferencing proposed by Italy and approved Bologna as for ECMWF’s new data centre. activities and the high- the host city for ECMWF’s new data centre.

performance computing facility The Italian proposal had been evaluated as are all t for purpose, and that part of an international competition and was staff are given the means to judged to best meet ECMWF’s requirements. be as good as they can be. The building is to be delivered to ECMWF by 2019 and will host the Centre’s new supercomputers, whilst the Centre’s 2017 was an exceptionally busy period in headquarters are to remain in the UK. this area, with our Council approving the Italian proposal to host the new data centre in The BOND (Bologna: Our New Data centre) Bologna and the UK Government committing programme will remain a high priority for to providing new of ce and conferencing ECMWF in the coming years. accommodation by 2023/2024. The data centre, however, was not the only relocation project discussed in 2017. Of ce A new home for accommodation was also on the agenda supercomputers and in December, the UK Government made an of cial proposal to have the University On 22 June 2017, ECMWF’s Council of Reading as the future long-term location approved the proposal made by the Italian for ECMWF’s new of ce accommodation. Government and the Emilia-Romagna Region to host ECMWF’s new data centre

26 ECMWF Annual Report 2017 ecmwf.int

AR-2018-ARTWORK.indd 26 08/06/2018 16:01

After an extensive process of evaluation by a panel of international experts and representatives of ECMWF’s Member States, Council accepted the proposal made by the Italian Government to host the future data centre in the ‘Tecnopolo di Bologna’, in the Emilia-Romagna Region of Italy.

Signing agreement 22 June 2017 - ECMWF Director-General Florence Rabier and Col. Silvio Cau, the Head of the Italian national meteorological service, sign the high-level agreement on the data centre.

ecmwf.int ECMWF Annual Report 2017 27

AR-2018-ARTWORK.indd 27 08/06/2018 16:01

Supporting ECMWF The Graduate Training Programme is an important aspect of ECMWF’s strong scientiﬁc cooperation with Member and Co-operating States. © Florian Rathgeber

Weather discussion Weekly weather discussion in ECMWF’s Weather Room.

The implementation of a new Enterprise Staff and funding Resource Planning (ERP) system, which started in 2016, will simplify and automate ECMWF is of course and above all its staff, many of the administrative processes at who in 2017 came from over 30 countries ECMWF. The design phase was completed and included graduate trainees from in 2017 and the ERP project team is now Hungary, Serbia, Sweden, Ireland and working with external consultants to deploy Croatia, and visiting scientists, including two core functionalities. HR, Finance and Payroll from China. This workforce was enriched by will be the rst areas to go live in 2018, Life at ECMWF the work of our Fellows, whose contributions additional elements will be gradually rolled help advance critical areas of science. Regular internal and external out at a later stage. This will bring ECMWF events provide opportunities a step closer to the vision of a paperless The nancial position of ECMWF remains for staff to network and share and streamlined central information hub. experiences. positive and continues to be inuenced by the ongoing success of ECMWF’s forecasting, which in turn has contributed Graduate Trainees to an increase in the sale of ECMWF data Graduate Trainees from and products; and by the high volatility the national meteorological levels in the sterling/euro exchange rate, services of Croatia, Hungary caused in part by political uncertainty and Serbia. over Brexit.

28 ECMWF Annual Report 2017 ecmwf.int

AR-2018-ARTWORK.indd 28 08/06/2018 16:02 The main areas of expenditure related European investment to remuneration and related items in ECMWF (£23.3 million), computer expenses (£19.9 million), pension schemes The 34 Member and Co-operating states (£10.5 million), electricity (£3.0 million) of ECMWF are the core source of nance and other operating activities (£5.4 million). for the Centre, with contributions totalling Costs associated with externally funded £43.5 million. projects amounted to £35.6 million and net nance costs were £4.2 million. External organisations support both core research and the complementary goals ECMWF’s budget remains on a cash basis of the centre with funding of £42.8 million, and the Financial Statements include a while revenue from sales of data and reconciliation of the results under IPSAS and products provide additional income in cash terms. Under cash accounting, the of over £7.8 million. Centre generated a surplus of £3.1 million in 2017, which Council subsequently approved ECMWF continued to invest in its staff, for use in the Data Centre project. infrastructure and systems to provide the

highest quality products to its Member and Note: all numbers exclude Centre tax. Co-operating States. The main areas of expenditure are summarised below, and include capital investment of £0.9 million, principally for IT and infrastructure.

Funding Costs

8% <1% 4%

23%

35%

45% 62% 46% 10%

24% 3% 14% 5% 20%

Member & Co-operating States’ Personnel costs contributions Pension and post-employment bene ts Externally funded income Electricity expenditure 14% Third-party projects Computer expenditure 24% Copernicus Other operating expenditure 62% Administered on behalf of the EU Externally funded expenditure Sales of forecasts and data Net nance costs Other operating revenue

ecmwf.int ECMWF Annual Report 2017 29

AR-2018-ARTWORK.indd 29 08/06/2018 16:02 Serving Member and Co-operating States

Annual Seminar 2017 The Seminar marked 25 years of ensemble prediction at ECMWF, with a keynote address from Eugenia Kalnay, University of Maryland (below).

ECMWF continued to make progress in serving its Member Making deliverables and Co-operating States and expertise available through enhanced training, 2017 saw an active programme of visits forecast outputs, computer and meetings with our Member and Co-operating States. Highly productive, services and data delivery. these activities provide the opportunity for Workshops and other exchanging information on developments collaborations continued and plans for the future. to bring bene ts. We continued to ensure ef cient access to ECMWF products and computing facilities ECMWF also serves its for our Member and Co-operating States. community through its Developments included locating modules of our in-house dissemination software contribution to the EU ‘Data Mover’ at remote centres to ensure Copernicus Services, ongoing reliability and speed of delivery, which deliver open and free and upgrades to ecgate, which gives users environmental information. access to ECMWF’s computing architecture. A working group was set up to consider cloud computing and big data. These are key issues which cut across several vital functions of the Centre, and could offer Member States wider potential for serving their own end-users.

30 ECMWF Annual Report 2017 ecmwf.int

AR-2018-ARTWORK.indd 30 08/06/2018 16:02 New products have been added regularly to ecCharts, including cloud base height and 2m mean temperature.

The Annual Seminar has always been Progress has been made towards the ECMWF’s agship educational event, operational implementation of products but 2017’s event also carried an emotional from the ECMWF ensemble forecast (ENS) 300 dimension as it was marking 25 years to assist in the key area of severe weather of ensemble prediction at ECMWF. A large prediction. A new map layer went into In 2017 ECMWF number of speakers and participants, both ecCharts to show the most probable courses trained over on-site and remote, took part and added instantaneous precipitation type (rain, 300 participants. their expertise and experience to what was snow, freezing rain, dry etc.) and its a great opportunity to discuss the future probability of occurrence at different lead of ensemble prediction. A range of other times. Most of this work has been funded meetings such as ‘Python frameworks for by the H2020 ANYWHERE project. We have Earth system sciences’ and a joint workshop also developed forecasts of point-rainfall with ESA on ‘Using low frequency passive by post-processing ENS gridbox output microwave measurements in research to take account of sub-grid scale variability and operational applications’ continued and biases. One important application to help ECMWF build on its science through will be within the Global Flood Awareness collaboration with delegates from around System (GloFAS), for ash ood prediction. the world. Other new products have been added regularly to ecCharts, including cloud In 2017 ECMWF courses trained over 300 base height and 2m mean temperature participants. In response to requests from over an interval. Member States, we redesigned our courses to teach more participants with less on-site Delivery of forecast products to Member and staff time. For example, we have and Co-operating States relies on created a three-day blended course on the specialised ECMWF software. Software ’Use and interpretation of ECMWF products’ updates have bene ted from contributions which covers similar material to the 5-day received from Member and Co-operating on-site version. Key to these improvements States. ECMWF has released other updates are new eLearning modules and other to its standard packages for ecCodes and types of online self-study. An initial set of Metview. A key milestone was the beta eLearning modules was published on the release of a new interpolation library website, and Member and Co-operating called Meteorological Interpolation and States can also include them in their own Regridding (MIR), which will facilitate training materials. scalability improvements and other additional features. ECMWF is providing access to a specialised dataset for the WMO Year of Polar Prediction and other data for WMO members. Our WMO and ECMWF Fellowship programmes continued to bring tangible bene ts in line with the ECMWF Strategy.

ecmwf.int ECMWF Annual Report 2017 31

AR-2018-ARTWORK.indd 31 08/06/2018 16:02 Serving Member and Co-operating States

Working together Open data week ECMWF Fellows Twenty teams worked day Between 28 February and 5 March, ECMWF In addition to the WMO Fellowship and night at the open data held a week of events focusing on freely programme, ECMWF also appoints its own hackathon (top left); available data, scheduled to coincide with Fellows to strengthen our relationship with ECMWF Fellows Heini Wernli (top right) and Daniel Jacob International Open Data Day. Three events exceptional individuals working in scienti c (bottom right). looked at key issues surrounding open data: and development areas relevant to the the Workshop on Meteorological Operational strategic goals of ECMWF. Work with Systems (MOS), organised biennially by our two most recently appointed Fellows ECMWF, focused on how to serve large (Heini Wernli, ETH Zurich and Daniel Jacob, amounts of open data and make them Harvard University) is starting to bear fruit. easily accessible; a data hackathon For example, ECMWF is working with where 70 developers, scientists and Heini Wernli on warm conveyor belts data enthusiasts explored uses of freely (WCBs), which are regions of ascending available data from ECMWF; and a air, generally ahead of a cold front. WCBs workshop which discussed how to improve can be precursors for regime changes, the socio-economic impact of numerical such as the development of blocked weather prediction. conditions over Europe, act to enhance medium-range forecast uncertainty and lead to severe precipitation.

Daniel Jacob is working on innovative ways of assessing errors associated with long- range transport in models which could be tested within the ECMWF Integrated Forecasting System (IFS).

32 ECMWF Annual Report 2017 ecmwf.int

AR-2018-ARTWORK.indd 32 08/06/2018 16:02 Sentinel-5P Developed by ESA and launched on 13 October 2017, this satellite will make a vital contribution to the ECMWF- run EU-funded Copernicus Atmosphere Monitoring Service (CAMS) by measuring the composition of the atmosphere. © ESA/ATG medialab © ESA/ATG

Partnerships with space agencies Our work with space agencies goes beyond the provision of data and collaborative development of instruments, extending also to educational activities.

In particular, we organised a workshop with the EUMETSAT NWP-Satellite Application Facility in May to assess the state of readiness ahead of the launch of the METEOSAT Third-Generation Geostationary satellite (MTG). MTG will provide unprecedented Satellite observations workshop observations with potentially enormous Breakout session in the ‘Assimilation of impacts on analyses and forecasts, but hyper-spectral geostationary satellite the incorporation of such data represents observations’ workshop organised in signi cant scienti c and technical challenges. partnership with EUMETSAT (May 2017). The workshop looked at the latest progress in instrument calibration and spectral characterisation, advances in radiative transfer modelling and the various assimilation strategies being considered to exploit this unique and exciting source of observations.

Our collaboration with ESA is also growing stronger, with the advent of the Copernicus Programme in particular.

ecmwf.int ECMWF Annual Report 2017 33

AR-2018-ARTWORK.indd 33 08/06/2018 16:02 Serving Member and Co-operating States

60°N Seasonal forecasts The proof-of-concept 60°N graphical products from the 30°N C3S multi-model seasonal

30°N forecast suite include 0°Nforecasts for mean sea- surface temperature 0°N anomaly, based on a 30°S combination of three models 30°S from the UK Met Of ce,

60°SMétéo-France and ECMWF.

60°S

180°E 150°W 120°W 90°W 60°W 30°W 0°E 30°E 60°E 90°E 120°E 150°E 180°E 150°W 120°W 90°W 60°W 30°W 0°E 30°E 60°E 90°E 120°E 150°E <-2.0°C -2.0..-1.0 -1.0..-0.5 -0.5..-0.2 -0.2..0.2 0.2..0.5 0.5..1.0 1.0..2.0 >2.0°C <-2.0°C -2.0..-1.0 -1.0..-0.5 -0.5..-0.2 -0.2..0.2 0.2..0.5 0.5..1.0 1.0..2.0 >2.0°C

to 13 months. SEAS5 products are available Delivering environmental to Member and Co-operating States ahead information of their release through C3S. 2017 marked ten years of daily atmospheric Part of the ERA5 reanalysis was also released. composition forecasts at ECMWF, now run operationally through the Copernicus Agreement was reached on the Atmospheric Monitoring Service (CAMS). implementation of the Copernicus Data and CAMS underwent a number of upgrades Information Access Services (DIAS), which through the year, and the rst CAMS will make Copernicus information available reanalysis was produced covering the to users through a cloud-based architecture. period from 2003 to present. Information The work will be done collaboratively between on the emissions and smoke transport from ECMWF, EUMETSAT and Mercator Ocean. devastating wild res in Chile in January ECMWF is leading a new initiative to explore 2017 was provided by CAMS. the development of a European system to

A signi cant milestone was achieved early in monitor human-related carbon dioxide (CO2) November when the new ECMWF seasonal emissions across the world. Such capacity forecast system (SEAS5) became part of the is vital to support Europe’s leading role in Copernicus Climate Change Service (C3S) worldwide action to address climate change.

multi-model seasonal forecast suite. The CO2 Human Emissions (CHE) project Forecast products are based on input from started on 1 October 2017 and brings three core providers: ECMWF, UK Met Of ce together a consortium of 22 European and Météo-France. The current proof-of- partners. It will last for over 3 years and concept phase includes graphical forecast act as a bridge between the European

products for air and sea-surface Commission and its CO2 Task Forces, temperature, mean sea level pressure and space agencies and related industries,

precipitation; the forecasts are updated the CO2 science community, and the every month and cover time ranges up Copernicus Services.

34 ECMWF Annual Report 2017 ecmwf.int

AR-2018-ARTWORK.indd 34 08/06/2018 16:02 The ERA5 global climate reanalysis is now in continuous production.

GloFAS CAMS upgrade The GloFAS web interface The main objective of the CAMS scienti c gives access to the seasonal work performed at ECMWF is to continually river ow outlook, indicating develop and upgrade the con guration of the maximum probability the IFS used to produce global analyses of higher or lower than and forecasts of atmospheric composition. normal river ow during the

www.globaloods.eu four-month forecast horizon, Improvements in 2017 included the for the global river network. incorporation of new satellite data: aerosol Copernicus Emergency observations from the Polar Multi-Sensor Management Service – Aerosol Product provided by EUMETSAT and ozone observations from the Ozone Early Warning Systems Mapping Pro ler Suite provided by NOAA (CEMS-EWS) through EUMETSAT’s third-party data In 2017 ECMWF launched the rst programme. The aerosol module was global-scale hydrological seasonal forecast upgraded to take into account interactions for the GloFAS element of the Copernicus with gases leading to improvements in Emergency Management Service – aerosol forecasts. Improvements were Early Warning Systems (CEMS-EWS) also seen in the UV forecasts. and contributed to other improvements in existing CEMS-EWS products. GloFAS seasonal was developed jointly with the ERA5 University of Reading. The ERA5 global climate reanalysis, being developed as part of the C3S, is in Collaboration has continued between continuous production. In July, ERA5 data ECMWF and external partners through for the period from 2010 to 2016 were training, capacity building and data exchange. released. During 2017, new ERA5 production This includes partnerships with national and streams were added to cover the 1980s and international organisations to support early 1990s, in addition to production streams warning and humanitarian action. covering the periods from January 1999, from January 2009 and a ‘near-real-time’ Copernicus re danger reanalysis products (NRT) stream initiated from December 2014. from the Global ECMWF Fire Forecast Atlantic hurricanes which occurred in 2017 (GEFF) system were made available for were well captured by ERA5 as well as the download through the ECMWF public exceptionally weak Antarctic ozone hole. dataset web interface. The products were developed at ECMWF over the last three C3S reanalysis products are now considered years as part of CEMS. as important scienti c inputs to the State of the Climate assessment by the WMO and are After a preliminary trial service, ECMWF was contributing to the Global Climate Observing awarded the provision of re forecasts for System (GCOS) Implementation Plan. CEMS in 2017, which will be undertaken in collaboration with Météo-France. This makes ECMWF the computational centre for both CEMS-Fire and CEMS-Flood.

ecmwf.int ECMWF Annual Report 2017 35

AR-2018-ARTWORK.indd 35 08/06/2018 16:02 How we work

Policy Advisory Scientiﬁ c Advisory Committee Committee

Technical COUNCIL Finance Committee Advisory Committee 22 Member States

Advisory Committee Advisory Committee of for Data Policy Co-operating States

Director-General

OFFICE OF THE DIRECTOR-GENERAL (Communications, Internal Audit)

RESEARCH FORECASTS COPERNICUS SERVICES COMPUTING ADMINISTRATION DEPARTMENT DEPARTMENT DEPARTMENT DEPARTMENT DEPARTMENT

EARTH SYSTEM DEVELOPMENT C3S SERVICE DATA CENTRE FINANCE ASSIMILATION OPERATIONS

EVALUATION CAMS SERVICE HUMAN RESOURCES EARTH SYSTEM HPC AND STORAGE MODELLING LEGAL PRODUCTION COPERNICUS CONTRACT MANAGEMENT PLATFORMS AND EARTH SYSTEM SERVICES PROCUREMENT PREDICTABILITY COPERNICUS PRODUCTION COPERNICUS COMMUNICATIONS SERVICE DELIVERY PROPERTY SERVICES INTEGRATED USER SUPPORT FORECAST SYSTEMS COPERNICUS USER SUPPORT

Organisation of ECMWF at June 2018

36 ECMWF Annual Report 2017 ecmwf.int

AR-2018-ARTWORK.indd 36 08/06/2018 16:02 Co-operating States as of January 2018

Bulgaria Czech Republic Estonia Former Yugoslav Republic of Macedonia Hungary Israel Latvia Lithuania Montenegro Morocco Romania

Cover image © Stocktrek Images/Thinkstock Slovak Republic ECMWF Shinfield Road Reading RG2 9AX United Kingdom Tel: +44 118 949 9000 www.ecmwf.int