Our Changing Forecast Model ECMWF’S Plans for 2008

ECMWF Newsletter No. 114 – Winter 2007/08 EDITORIAL

EDITORIAL Contents Editorial ...... 1 News Our changing forecast model ECMWF’s plans for 2008 ...... 2 New items on the ECMWF web site ...... 3 Changes to the operational forecasting system ...... 3 ECMWF’s forecast model (the IFS model) has been under - Celebration of Tony Hollingsworth’s life ...... 4 going important upgrades in its physical parametrizations Two new Co-operation Agreements ...... 4 during the past two years, and more are on the way. Ensemble Prediction Workshop, 7–9 November 2007 ...... 5 Cycle 31r1, introduced on 12 September 2006, con - A wealth of ocean data makes its appearance on the public web at ECMWF ...... 6 tained a set of revisions to the cloud scheme, namely Signing of the Co-operation Agreement between supersaturation with respect to ice and a more accurate ECMWF and Montenegro ...... 7 discretization and fall speed for ice particles. Another major Book about high performance computing in meteorology . . . 8 change was the replacement of the old and unrealistic ECMWF workshops and scientific meetings in 2008 ...... 8 68 th Council session on 10–11 December 2007 ...... 9 “orographic roughness length” by a modern orographic 11 th Workshop on Meteorological Operational Systems . . . . 10 form drag. New High Performance Computing Facility ...... 13 With Cycle 32r2, introduced on 12 June 2007, further ECMWF Educational Programme 2008 ...... 13 revisions to the orographic drag were made, and the rep - Fifteenth anniversary of EPS ...... 14 resentation of ice particles was further improved. However Meteorology the main change in this cycle was the replacement of the Toward a forecast of aerosols with shortwave radiation scheme by the Rapid Radiative the ECMWF Integrated Forecast System ...... 15 A new partitioning approach for Transfer Model, developed by AER, Inc. This new model ECMWF’s Integrated Forecast System ...... 17 has 14 spectral intervals, and supports an advanced rep - Probability forecasts for water levels at resentation of cloud-radiation interactions called McICA the coast of The Netherlands ...... 23 (Monte-Carlo Independent Column Approximation). This Advances in simulating atmospheric variability with IFS cycle 32r3 ...... 29 provides an elegant solution to the long-lasting problem General of optimizing the cloud superposition between two model Special Project computer allocations for 2008–2010 ...... 39 levels. Further information about the new radiation pack - Member State computer allocations for 2008 ...... 44 age can be found in ECMWF Newsletter No. 112 . Representatives and Contact Points ...... 44 Finally on 6 November 2007 a major change of the TAC Representatives, Computing Representatives convection scheme was introduced (Cycle 32r3). The use and Meteorological Contact Points ...... 45 ECMWF Council and its committees ...... 46 of the “large-scale moisture convergence” was discontin - ECMWF Calendar 2008 ...... 47 ued, and the scheme will now use only the properties of ECMWF publications ...... 47 the convective updraft. This brings the IFS in line with Index of past newsletter articles ...... 48 recent theories of convection, and results in a major Useful names and telephone numbers within ECMWF . . . . 51 change in the model activity as documented in this issue Publication policy of the ECMWF Newsletter (see page 29). In addition, the The ECMWF Newsletter is published quarterly. Its purpose is to soil hydrology was completely revised and geographical make users of ECMWF products, collaborators with ECMWF and the wider meteorological community aware of new devel - variations in soil texture were introduced for the first time opments at ECMWF and the use that can be made of ECMWF in the IFS, resulting in a more realistic representation of products. Most articles are prepared by staff at ECMWF, but the soil moisture reservoirs. The vertical diffusion above articles are also welcome from people working elsewhere, the boundary layer was also improved. especially those from Member States and Co-operating States. The ECMWF Newsletter is not peer-reviewed. Further work is currently addressing the overall consis - Editor: Bob Riddaway tency of the cloud processes and the representation of Typesetting and Graphics: Rob Hine shallow cumulus by a dual mass-flux approach. Any queries about the content or distribution of the ECMWF Such extensive and rapid changes in the model physics Newsletter should be sent to [email protected] have not taken place for many years. Due to these various Contacting ECMWF upgrades the systematic errors of the IFS model have Shinfield Park, Reading, Berkshire RG2 9AX, UK been considerably reduced. This benefits not only the Fax: ...... +44 118 986 9450 medium-range forecasts but also the assimilation of Telephone: National ...... 0118 949 9000 observations. In addition the changes provide an International ...... +44 118 949 9000 enhanced ability to represent the Earth’s climate in long ECMWF Web site ...... http://www.ecmwf.int integrations coupled with ocean, sea-ice and land surface Front cover Storm surge barrier in Zeeland, The Netherlands, which was models. In fact our IFS is getting much closer to being a built after the storm disaster in 1953. true “Earth System Model”. Philippe Bougeault

1 NEWS ECMWF Newsletter No. 114 – Winter 2007/08

ECMWF’s plans for 2008 obtained – such as a probability distribution or EPSgram. Two important new datasets are expected this year. The first one will be the re-forecast dataset providing a perma- DOMINIQUE MARBOUTY nently updated set of re-forecasts for calibration. Such a dataset will allow many new product developments by the THE four-year programme of activities 2008 –2011, adopted Member States. The second one will be the Interim reanalysis by the ECMWF Council at its 68 th session in December, covering the period 1989 to present: it will improve signifi- provides the backbone of our plans for 2008 (the cantly reanalysis quality over the satellite era. An important programme is available at www.ecmwf.int/about/ programmatic ). feature is that the Interim reanalysis will be continued and These plans will build upon our 2007 achievements which updated in delayed mode, i.e. with a few weeks delay. included: Further developing the Centre’s support to the European N Supercomputer procurement. Union has been an important element in ECMWF policy N Aggressive action to ensure the prompt operational use over the recent years and will be continued. Our main of MetOp data. contribution is directed to the GMES initiative, in partic- N Successful forecasts of several major severe weather ular to the development of the atmospheric services comp- events in Europe (e.g. the extreme winds associated with onent for which we are currently coordinating the GEMS storm Kyrill which affected Western Europe in January, project (which deals with global monitoring of atmospheric flooding in the UK in July and the storm-surge in the gases and aerosols and improved air quality forecasts). North Sea in November). Several GEMS developments are expected this year, N Coordination of the preparation of the MACC (Monitor- including air chemistry reanalyses. As mentioned earlier a ing Atmospheric Composition and Climate) proposal for follow-up project (MACC) has been prepared and will be the developments of atmospheric services to support discussed with the commission with a view to it starting in GMES (Global Monitoring for Environment and Security). 2009. Another important theme will be to prepare and In December the Council accepted the Centre’s proposal secure funding for the next generation reanalysis (ERA-75). to select IBM for the provision of our next supercomputer Following discussions with the Council, it was agreed that and decided a further increase of the funding from 2011. ECMWF will review and continue to improve its governance The new machine will be installed in steps throughout and financial management. This year’s focus will be on: 2008, starting with an initial test system in May and the N Developing a long-term solution for the Centre’s first cluster being ready for trial in the autumn. Meanwhile, pensions’ payments. several infrastructure systems – power, cooling, networks N Considering changing accounting and auditing policies and archive – will be enhanced, and our various operational to new international standards. suites will be prepared for the migration. The new super- N Developing a scheme for activity-based costing. computer facility will provide increased performance by a N Preparing a technical risk analysis. factor of five (to 19 teraflops sustained), allowing for the The number of States supporting ECMWF will continue to resolution of the model to be increased from 2009. increase. New States are expressing interest in developing Most of ECMWF’s activity is dedicated to improving our co-operation with ECMWF. Also the process of accepting the forecasting system, with a particular focus on severe weather. amended Convention is making good progress and is These improvements will, as usual, be incorporated in new expected to be finalised by the remaining Member States cycles of the model and implemented throughout the year. this year. It will open the door for new States (or current They will include assimilation of many more satellite data, Co-operating States) to join ECMWF as Members. several important developments of the wave model and a Finally this year we want to dedicate serious effort to the specific effort on improving the ensemble in the tropics. It long-term future of the ECMWF forecasting system. The is worth noting that, as presented in the editorial, we will pilot project on the non-hydrostatic dynamical core will be finalise a cycle of general upgrade of the IFS physics which pursued and we expect to reach conclusions as to its started in 2006. Another effort, in collaboration with Aladin suitability of the IFS for higher resolution beyond 2015. We and Hirlam, will be dedicated towards modularization of will also start a new project concerning the scalability of the code so as to simplify its maintenance. our codes on massively parallel machines, in particular the Effort will be put on the development of verification. data assimilation system. Future supercomputers are likely It will include finalising a revision of the verification applica- to be composed of hundreds of thousands of processors tion and database and implementing new verifications and we need to evaluate how our codes should evolve in (e.g. for Extreme Forecast Index (EFI) and ensemble waves). order to run efficiently on such machines. This work will be First and foremost, however, will be developing ways of crucial in defining the necessary priorities for ECMWF in verifying forecasts of severe weather events. ECMWF will the coming years. lead a verification review to be undertaken in WMO. The plan sets out an ambitious programme of work, but Another specific focus will be on the development of prod- we are confident that the plan will be implemented and ucts, mainly through web services. This will include develop- lead to improved services for Member States and the wider ing clickable charts allowing additional information to be meteorological community.

2 ECMWF Newsletter No. 114 – Winter 2007/08 NEWS

New items on the ECMWF web site

ANDY BRADY days, through months and seasons, to required to move towards petaflops multi-annual timescales. See page 14 for computing. The workshop is planned Release of ECMWF Calendar 2008 a report about the workshop. to be held from 3 to 7 November 2008 The 2008 ECMWF Calendar is the www.ecmwf.int/newsevents/meetings/ (exact dates to be confirmed). official timetable for all major events workshops/2007/ensemble_prediction/ www.ecmwf.int/newsevents/ that are planned to be held at the meeting s/workshops/2008/ Centre, including meetings of Council Workshop on GPS Radio- high_performance_computing_13th/ and its committees, training courses Occultation and workshops. A joint ECMWF-EUMETSAT/GRAS ECMWF Annual Seminar 2008 www.ecmwf.int/newsevents/calendar/ SAF Workshop on GPS Radio- Increasingly parametrization work 2008.html Occultation will be held at the Centre focuses on how processes interact with from 16 to 18 June 2008. The aim of each other and the dynamics, and Presentations from the Workshop the workshop is to consider the whether the feedbacks are represented on Meteorological Operational various assimilation approaches, realistically. This is particularly Systems publicise the assimilation software relevant at very high resolution where The 11 th Workshop on Meteorological developed by the GRAS-SAF part of a process might be resolved. Operational Systems was held from 12 consortium, review the latest The emphasis of the 2008 Seminar, to to 16 November 2007. The objective of assimilation experiments with GPS be held from 1 to 5 September, will be the workshop was to review the state RO measurements, and promote the on the interaction of parametrized of the art of meteorological opera- value of these measurements in the processes with the resolved dynamics. tional systems and to address future context of POST-EPS. Workshop www.ecmwf.int/newsevents/meetings/ trends in the use and interpretation of attendance is by invitation only. annual_seminar/2008/ medium- and extended-range forecast www.ecmwf.int/newsevents/ guidance, operational data manage- meetings/workshops/2008/ Workshop on Atmosphere-Ocean ment systems and meteorological GPS_radio_occultation / Interaction visualisation applications. See page 10 A Workshop on Atmosphere-Ocean for a report about the workshop. Workshop on High Performance Interaction will be held at the Centre www.ecmwf.int/newsevents/meetings/ Computing in Meteorology from 10 to 12 November 2008. The workshops/2007/MOS_11/ Every second year ECMWF hosts a workshop will address the require- workshop on the use of high perform- ments for ocean-atmosphere coupling Presentations from the Ensemble ance computing in meteorology. The from the very short time scales to the Prediction Workshop emphasis of the 13 th Workshop on monthly time range with a focus on The Workshop on Ensemble Prediction High Performance Computing in ocean near-surface processes. was held from 7 to 9 November 2007. Meteorology will be on running Workshop attendance is by invitation This workshop reviewed the most recent meteorological applications at only. advances in ensemble techniques sustained teraflops performance in a www.ecmwf.int/newsevents/ applied to data-assimilation and forecast production environment, and on the meetings/workshops/2008/ systems for predictions ranging from application specific developments ocean_atmosphere_interaction / Changes to the operational forecasting system

DAVID RICHARDSON N New formulation of convective N Assimilation of SBUV (NOAA-17, entrainment and relaxation timescale. NOAA-18) and monitoring of OMI Implementation of cycle Cy32r3 N Reduction in free atmosphere ozone data. A new cycle of the ECMWF forecast vertical diffusion. The main changes to the EPS and analysis system, Cy 32r3, was N New soil hydrology scheme. included in this cycle are: introduced on 6 November 2007. The N New radiosonde temperature and N Initial perturbation amplitude new cycle includes significant changes humidity bias correction. reduced by 30%. to the model physics, including the N Increase in the number of radio N Singular vectors targeted on convection scheme, that increase the occultation data from COSMIC. tropical cyclones are computed with model activity in the tropics. The main N Assimilation of AMSR-E, TMI and the new moist physics package in the changes included in this cycle are: SSMIS window channels (clear sky). tangent-linear and adjoint models (as

3 NEWS ECMWF Newsletter No. 114 – Winter 2007/08

used in the operational in 4D-Var Planned changes provided with EPS output uniformly since Cy 32r3). The monthly extension to VarEPS is up to 32 days ahead, once a week. It More information about the impact now undergoing pre-operational will also introduce a coupled ocean- of the Cy 32r3 changes on atmospheric testing. This will lead to the full atmosphere model for the forecast variability is given in the article integration of the Monthly System and range day 10 to 15 for the 00 UTC starting on page 29. VarEPS and will enable users to be based forecast on a daily basis. Celebration of Tony Hollingsworth’s life

DOMINIQUE MARBOUTY ECMWF: David Burridge, Lennart the widespread admiration and affec- Bengtsson and Adrian Simmons. All tion for Tony will at least be a small TONY HOLLINGSWORTH , who died these focussed on Tony’s achieve- comfort to them. on 29 July 2007, was employed at ments and the qualities that made him ECMWF from March 1975 until his such a delightful friend and colleague. death. A full account of his key role in It was clear that Tony had made a the development of the Centre and his significant contribution to the success involvement in various other inter- of ECMWF, and that he would be national activities can be found in the sorely missed. article by David Burridge in ECMWF During the proceedings there was Newsletter No. 113 . some cello music played by Tony’s Due to Tony’s long association with daughter, Déirdre. The final ECMWF, his wife Breda asked whether presentation was from Tony’s son, the Centre would host a celebration of Cormac. He described the many ways Tony’s life. I considered it an honour in which Tony had been a loving for the Centre to be involved in such an father and husband. event. Also I was delighted that there At the end of the formal part of the would be an opportunity for Tony’s proceeding refreshments were served. many friends and colleagues to cele- I am sure that Tony would have brate the outstanding contribution he approved of the convivial atmosphere made to many aspects of meteorology. and the bringing together of people The event was held on 3 November from different aspects of his life. with over 120 people attending. As Nothing will take away the feeling well as a presentation from me, there of loss for Breda, Déirdre and Cormac, were presentations from three other but I hope that the celebration of Cormac Hollingsworth speaking at his father’s people who had worked with Tony at Tony’s life and the demonstration of memorial service held at ECMWF. Two new Co-operation Agreements

MANFRED KLÖPPEL N Has exactly the same access to by Council. ECMWF products as a Member State To date Co-operation Agreements TWO NEW Co-operation Agreements (e.g. dissemination, MARS and have been concluded with the Czech have been signed. On 1 November software). Republic, Croatia, Estonia, Hungary, 2007 an agreement was signed N Has access to servers (ECGATE) Iceland, Lithuania, Montenegro, between ECWMF and Slovakia which but does not have a supercomputer Morocco, Romania, Serbia, Slovakia came into force on 1 January 2008. allocation. and Slovenia. Also an agreement was signed with The needs of the Co-operating In 2005 the Council decided to Montenegro on 5 November with effect States are overseen by the Advisory amend ECMWF’s Convention that from the day of signature. Montenegro Committee of Co-operating States. restricts membership to the current 18 and Slovakia are ECMWF’s 11 th and 12 th This committee provides the ECMWF Member States. Once the amended Co-operating States. Council with opinions and recommen- Convention has been ratified by all Where a Co-operation Agreement dations on the programme of activities Member States, “new” countries will has been established, the Co-operating of the Centre, as well as giving consid- be eligible to apply for full Member State: eration to any matter submitted to it State status.

4 ECMWF Newsletter No. 114 – Winter 2007/08 NEWS

Ensemble Prediction Workshop, 7 –9 November 2007

ROBERTO BUIZZA, RENATE HAGEDORN, FRANCO MOLTENI, TIM PALMER On the merit of multi-analysis/ multi-model approaches to ensemble A THREE-DAY workshop on designed to further increase the value prediction, there was a clear consensus Ensemble Prediction was held at of probabilistic ensemble prediction. that the multi-model approach to ECMWF from 7 to 9 November 2007. In the second part of the workshop, ensemble prediction is an extremely Detailed information about the discussions were held in three pragmatic and useful mechanism for workshop can be found at: working groups that focused on key sampling model uncertainty. However, www.ecmwf.int/newsevents/meetings/ areas of research and development. it was recognized that this approach workshops/2007/ensemble_prediction should not be used to avoid the About sixty participants from around Representation of initial condition difficult problems of improving the the world met to discuss progress in: and model uncertainty representation of model uncertainty or N The representation of uncertainties The first working group focused on reducing model errors and biases. in initial conditions (e.g. ensemble issues related to the representation of The working group spent some time data assimilation, ensemble transform initial conditions and model uncer- discussing the trade off between Kalman filter, bred vectors, singular tainty. During the discussion, it was ensemble size and ensemble member vectors). pointed out that it is difficult to resolution, and concluded that this N Model formulation (e.g. multi- separate the contribution to forecast issue needs to be revisited with the model ensembles, perturbed error due to initial condition uncer- goal of potentially increasing the parameter ensembles, stochastic tainty from that due to model uncer- resolution of the ensemble members. parametrization). tainty. Although a clear separation N Validation and calibration methods. cannot be achieved, it was recognized Methodologies for downscaling and N Use of ensemble forecasts in hydro- that model uncertainty makes a calibration of ensemble forecasts logical prediction. significant contribution to initial The second working group discussed In the first part of the workshop, condition uncertainty. Possible ways the relative merits of methodologies twenty speakers from Europe, to diagnose the contribution to for the downscaling and calibration of Canada, Japan, Korea and the USA forecast error associated with model ensemble forecasts on different time reviewed the current status of error have been suggested – these are scales. As far as calibration is operational ensemble forecasting based on a perfect model assumption concerned, results from various systems. They also discussed research and on initializing forecasts using the approaches indicated a consistent plans and future developments Ensemble Kalman Filter. benefit in terms of skill scores for

5 NEWS ECMWF Newsletter No. 114 – Winter 2007/08

weather parameters. There was Verification and application of the EPS should be given the same general consensus about the useful- Ensemble Prediction Systems level of attention as the deterministic ness of generating re-forecast data for The third working group focused its forecast system. medium-range, monthly and seasonal discussions on the verification and It was recommended that research ensemble forecasts. It was pointed out application of Ensemble Prediction on how to best combine high- that unresolved issues on the choice of Systems. The topics discussed resolution deterministic forecast and optimal methodologies, re-forecast included the most appropriate choice EPS should be continued at ECMWF. versus ensemble size, and the actual of verification measures, statistical It was, however, also stated that a increase of forecast skill could only be significance of results, accounting for user/application-specific combination addressed by carrying out research on the uncertainty of verification data, may be superior to a generic combina- extensive sets of re-forecasts. avoidance of false skill, verification of tion. In such cases, the production of Downscaling issues were mostly rare/extreme events, and communi- an optimally combined PDF would fall discussed in the context of seasonal cation of probabilistic forecasts. into the responsibility of Member forecasts, an area where the advantages The working group made some States or individual end-users. (and disadvantages) of dynamical recommendations concerning the versus statistical methods are still to be design and testing of forecast systems, Key message properly evaluated. The working group stating that, based on predictability The workshop closed with a plenary agreed that, so far, evidence to justify the theory, medium-range and later-range discussion during which the key additional complexity and computing forecasts should be issued in conclusions of the three working requirements of dynamical downscaling probabilistic form. Therefore, the groups were analyzed by all is limited. However, results presented prime aim of ECMWF should be to participants. The key message that can at the workshop on a reduction of false- provide the tools to predict a reliable be drawn from the plenary meeting is alarm rates in downscaled simulations and sharp probability density function that many groups are working to with a regional climate model warrant (PDF) of the atmospheric state rather improve the current ensemble further investigation. The degradation than just a single deterministic high- prediction systems, and that work is of results arising from the use of resolution forecast. The PDF should progressing towards the development boundary conditions at 12-hour versus be based on all available information, of a complete probabilistic approach 6-hour frequency in dynamical down- i.e. the EPS and the high-resolution to the data assimilation and scaling was also pointed out. deterministic forecast. Consequently, forecasting problem.

A wealth of ocean data makes its appearance on the public web at ECMWF

ALBERTO TROCCOLI, more people. This is the first time that and in the original model grid (more FRANCISCO J. DOBLAS-REYES such an extensive dataset with homo- information about the ocean analyses geneous NetCDF Climate and Forecast is in the article in ECMWF Newsletter IN ECMWF Newsletter No. 113 (page (CF) Metadata conventions for either No. 113 , pages 8 –16). At present these 4) there was a news item about atmospheric or ocean variables has data are available for the ECMWF ENSEMBLES public data disseminat- been created and made available to model but within the framework of ion in which it was reported that the public. ENSEMBLES, results from other ocean data would be available in the The ENSEMBLES ocean data are models should be available soon. near future. Rather quicker than from two sources. The main difficulty in making the expected a wealth of ocean data has N Data assimilation analyses data available, once they have been now been made available on the covering the 48-year period produced, is to ensure that the OPeNDAP server installed at ECMWF 1959 –2006 – currently the longest NetCDF files destined for the OPeN- under the EU project ENSEMBLES period for publicly available ocean DAP server follow well defined (ensembles.ecmwf.int/thredds/catalog.html ). analyses. standards. Extensive work has been In fact, limited amounts of ocean data N Retrospective forecasts (or hind- undertaken for the atmospheric data were made publicly available on the casts ) – from three experiments, two also, as discussed in the news item ECMWF OPeNDAP server under the covering the period 1991 –2001 and mentioned above. However, new FP 5 EU project ENACT ( www.ecmwf.int/ one covering the period 1960 – 2006. conventions have been introduced for research/EU_projects/ENACT/ ), but a The analyses are available both in a the ocean data in order to make them greatly expanded data set is now avail- regular grid to allow an easier inter- CF-compliant. This process, together able which may trigger the interest of comparison between different models with the actual preparation of the CF-

6 ECMWF Newsletter No. 114 – Winter 2007/08 NEWS

Examples of data retrieval An example of how data can be retrieved using free software such as nco (nco.sourceforge.net ) is given here. In this example the nco command (ncks) is requesting the temperature (the three-dimensional thetao field) along the equator (-d latitude,0) for the upper 1000 m (-d level), for the 1959 –2006 period (-d time for the 576 months) and for the five ensemble members (-d ensemble): ncks -a -h -v thetao,time_bnd -d latitude,0. -d level,0.,1000. -d time,0,575,1 -d ensemble,0,4,1 -O -o fileout.nc -p http://ensembles.ecmwf.int/thredds/dodsC/ocean/ecmwf/eg thetao By applying this command to programming/visual languages such as IDL one can analyse these data in a relatively easy way. For instance, the figures illustrated here were produced using a simple code which is available on request or at: www.ecmwf.int/research/EU_projects/ENSEMBLES/data/data_dissemination.html. In both cases illustrated here, the results change only marginally when fields are taken on the regular grid rather than on the original grid. 2 0

100 10 1 –5 –4 5 3 –3 200 –2 0 1 0.5 – .1 .5 1 –0– 01 0 ..1 .50. – 00 D 300 1 –

–1 e p t h ( T m 400 e m –2 ) p e r a 500 t u 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 0 30 60 90 120 150 180 210 240 270 300 330 360 r e Year Longitude (E) a n

Timeo series of the temperature anomaly of the ocean analysis on the The equatorial section of the analysed temperature anomaly for m

regula ar grid averaged over the top 300 m and over the Niño3.4 region. the ensemble mean for the month of December 1997 (close to the l The by lue envelope represents the full spread of the five members. The peak of the 1997/98 El Niño event). ( ° abseC nce of spread, as in the present period, indicates that observa - ) tions are numerous and provide a strong constraint to the analyses. compliant NetCDF, is one that has physically retrieved (the targeted even at different locations. required intensive work, involving part) and this operation can be carried More sophisticated data analyses others at ECMWF and elsewhere. out without visiting directly the web than those illustrated here are The main advantage of the page where the server and the data certainly possible and we would like to OPeNDAP server ( www.opendap.org ) is reside (the remote part). As a encourage all those interested in using the targeted remote data access. This consequence, computer codes can be historical oceanic data as well as long means that only portions of data (e.g. written that access targeted data from hindcasts to start exploiting our equatorial cross-sections) need to be more than one OPeNDAP servers, targeted remote data access facility. Signing of the Co-operation Agreement between ECMWF and Montenegro

MANFRED KLÖPPEL Predrag Nenezic stated: the Hydrometeorological Institute of “The European Centre for Medium- Montenegro will benefit to a great ON 5 November 2007, Predrag Range Weather Forecasts is the world extent from their close co-operation in Nenezic, Minister of Tourism and leader in its area of numerical weather meteorology.” Environment of Montenegro, and prediction and we are proud of having Dominique Marbouty said: Dominique Marbouty, Director of concluded this co-operation agreement “ECMWF’s worldwide leadership in ECMWF, signed a co-operation with this organisation. Its products will the field of numerical weather agreement in Podgorica, Montenegro. greatly assist the Hydrometeorological prediction is based on close collabo- Luka Mitrovic, Director of the Institute of Montenegro in fulfilling its ration with the meteorological Hydrometeorological Institute of mission, in particular with regard to community. Governments recognise Montenegro, attended the signing the protection of life and property. I the necessity of improving the quality ceremony. am confident that both ECMWF and and accuracy of numerical weather

7 NEWS ECMWF Newsletter No. 114 – Winter 2007/08

prediction, allowing them above all supercomputer system will be vital for advance warning of severe weather improving the overall quality of our events, such as storms, heat waves forecasts and for our warning services, and floods. I am looking forward to in particular giving the public and the close collaboration with the responsible authorities early advice on Hydrometeorological Institute of the likelihood of extreme weather Montenegro, which will now have events. Our meteorological and access to our products, especially technical staff will benefit immensely medium-range and seasonal weather from increasing their contacts with forecasts.” their colleagues at ECMWF. We will Luka Mitrovic said: be using ECMWF’s products to extend Dominique Marbouty, Director of ECMWF, and Predrag Nenezic, Minister of Tourism “This co-operation agreement is a both the range and the validity of our and Environment of Montenegro, signing a significant milestone for the forecasts to the benefit of the people co-operation agreement in Podgorica, Hydrometeorological Institute of of Montenegro. We very much Montenegro on 5 November 2007. Montenegro. The data from ECMWF’s welcome this agreement.” Book about high performance ECMWF workshops computing in meteorology and scientific meetings in 2008 GEORGE MOZDZYNSKI

EVERY second year ECMWF hosts a BOB RIDDAWAY workshop on the use of high performance computing in Workshop on “Use of GPS radio- meteorology. These bring together occultation” (16 to 18 June 2008) manufacturers, computer scientists, A joint ECMWF-EUMETSAT/GRAS researchers and operational users to SAF workshop will be held to consider share their experiences and to learn the assimilation of GPS radio- about the latest developments. occultation (GPS RO) measurements. The last workshop was held from 30 The aim of the workshop is to: October to 3 November 2006. Its N Outline the strengths and weak- context was the sustained performance nesses of the various assimilation at ECMWF now being 4 Teraflops approaches. averaged over the main IFS applica- N Publicise the assimilation software tions. This represents an 80,000 fold developed by the GRAS-SAF increase over the last 25 years. Will consortium. computer vendors be able to continue N Review assimilation experiments this kind of increase in the next 25 The papers from the workshop with GPS RO measurements using years? Will power consumption present the state of the art in the use various assimilation strategies. ultimately constrain performance by of parallel processors in the fields of N Promote the value of these measure- that time? Will the applications at meteorology, climatology and ments in the context of EUMETSAT’s ECMWF be able to run efficiently on oceanography. They have now been Post-EPS. computers that have just ten times the published by World Scientific under Workshop attendance is by invita- number of processes that we have the title “ Use of High Performance tion only. Further information is today? These are just some of the Computing in Meteorology – available from: questions addressed by the workshop. Proceedings of the Twelfth ECMWF www.ecmwf.int/newsevents/meetings/ Consideration was also given to Linux Workshop ”. The book is aimed at workshops/2008/GPS_radio_occultation clusters and parallel algorithms. The graduate students, academics and presentations from the workshop can researchers specializing in high ECMWF 2008 Annual Seminar on be found at: performance computing. Further “Parametrized physical processes www.ecmwf.int/newsevents/ information about the book can be and their dynamical interactions” meetings/workshops/2006/ found at: (1 to 5 September 2008) high_performance_computing-12th/ www.worldscibooks.com/ Increasingly models are being devel- presentations compsci/6638 oped and used at higher resolutions. However, many physical processes

8 ECMWF Newsletter No. 114 – Winter 2007/08 NEWS

remain unresolved and need to be A registration form and further the power of supercomputers in the parametrized. Even the highest information is available from: field of meteorology. resolution limited area models still www.ecmwf.int/newsevents/meetings/ Included will be a discussion of the need a parametrized representation of annual_seminar/2008 challenges of creating a computer shallow convection, turbulence, centre infrastructure for High microphysics, radiation and land Thirteenth Workshop on Performance Computing. surface processes. Schemes for deep “High performance computing Attendance at the workshop is by convection and subgrid orography will in meteorology” invitation and will be limited to still be needed in the foreseeable (3 to 7 November 2008) around 100 people. If you are future for global NWP and climate The emphasis of this workshop will be interested in attending, please go to: models. on achieving sustained teraflops www.ecmwf.int/newsevents/ The seminar will give an overview performance in a production environ- meetings/workshops/2008/ of the relevant issues. Emphasis will ment, and on developing a vision for high_performance_computing_13th be on the interaction of parametrized getting towards Petaflops computing. processes with the resolved dynamics. The aim is to provide a venue where: Workshop on Increasingly, parametrization work N Users from ECMWF’s Member “Ocean-atmosphere interaction” focuses on how processes interact with States and around the world can (10 to 12 November 2008) each other and the dynamics, and report on their experience and The workshop will address the require- whether the feedbacks are represented achievements in the field of high ments for ocean-atmosphere coupling realistically. This is particularly performance computing during the from the very short time scales to the relevant at very high resolution where last two years; plans for the future and monthly time range with a focus on part of the process (e.g. convection) requirements for computing power ocean near-surface processes. The might be resolved. Interactions and will also be presented. question to be discussed will include: feedbacks act at all scales from N Vendors of supercomputers will N Which processes need to be taken planetary and synoptic scales to meso- have the opportunity to talk to into account? scales (e.g. cloud/radiation interaction managers and end users of meteoro- N What is the impact of air-sea with the general circulation, boundary logical computer centres about their processes on atmospheric phenomena? layer turbulence with cyclones, land current and future products. N Which models are the most surface processes with precipitation at N Meteorological scientists can appropriate at different timescales, the meso-scale and the interaction of present their achievements in the and what are the implications of convection with tropical variability). development of parallel computing seamless forecasting? Consideration will also be given to the techniques and algorithms, and can Workshop attendance is by invita- role of cloud resolving models in exchange ideas on the use of tion only. Further information about studying interactions and developing supercomputers in future research. this workshop will be available at: parametrization schemes, and on the N Computer scientists can give an www.ecmwf.int/newsevents/ need for stochastic components in update on their efforts in providing meetings/workshops/2008/ schemes. tools which will help users to exploit ocean_atmosphere_interaction 68 th Council session on 10 –11 December 2007

MANFRED KLÖPPEL UK Ltd to replace the HPCF at the would increase by 3.0% over the Centre from 2009 onwards. The budget for the previous year. CHAIRED by its President, Dr Adérito Council also agreed unanimously to N Pension Scheme . The Council Vicente Serrão from Portugal, the increase the HPCF budget by £1million adopted the investment policy as ECMWF Council held its 68 th session from 2011 on top of the additional proposed by the Centre for the funds in Reading on 10 –11 December 2007. £3.5million from 2009 onwards in the dedicated pension account of Besides several decisions on already agreed in December 2006. the Funded Pension Scheme. financial and staff matters (e.g. N Four-year Programme of Activities . N Governance and financial adoption of Reports from the Co- The Council unanimously adopted the management . The Council noted ordinating Committee on updated “Four-Year Programme of ECMWF’s vision for the development Remuneration), the main results of Activities” for the period 2008 –2011 of the management practices at the the 68 th session were as follows. (for further information see Centre: N High Performance Computing www.ecmwf.int/ about/programmatic/ ). “ECMWF is committed to an Facility (HPCF) . The Council unani- N Budget 2008 . Council unanimously ongoing process of reviewing and mously authorised the Director to agreed that Member States’ improving its management policies to enter into a Service Contract with IBM contributions to the budget for 2008 ensure that it:

9 NEWS ECMWF Newsletter No. 114 – Winter 2007/08

N Delivers best value for Member States. N Monitors and further improves its high level of efficiency. N Is well positioned to address future technological and scientific challenges in order to maintain its worldwide acknowledged leadership position in Numerical Weather Prediction.” The Council agreed that the following points should be given priority: (a) development of a long-term solution for the pension issue, (b) consideration of the introduction of International Public Sector Accounting Standards (IPSAS) and standards of the International Organisation of Supreme Audit Dr Tim Palmer (Head of Probabilistic Forecasting and Diagnostic Division at ECMWF) giving Institutions (INTOSAI), (c) consider- a lecture to Council on “ Ensemble Prediction System – 15 years on ”. ation of the introduction of activity- based costing, (d) conduct a technical N Products . The Council agreed to (a) a second term of office. risk analysis and manage the make a subset of ECMWF wave N Election of President and Vice- identified risks, (e) review of the products available to WMO Members President . Dr Adérito Vicente Serrão strategic risk analysis and the in support of extreme sea state from Portugal and Mr Wolfgang management of the identified risks, forecasting, and (b) add several Kusch from Germany were re-elected and (f) review of the financial strategy. EUROSIP (EUROpean Seasonal-to- as President and Vice-President of the The Council agreed to the Centre’s Inter-annual Prediction) multi-model Council, respectively, for a second road map to address these issues. ensemble products to the Catalogue of term of office of one year. N Conditions for new Member States ECMWF real-time products. Dr Tim Palmer (Head of joining ECMWF . The Council agreed N Appointment of new Auditor . The Probabilistic Forecasting and on the calculation method of the Single Council appointed Mr José Manuel Diagnostic Division at ECMWF) gave Additional Contribution for new Adán Carmona as auditor and a lecture to Council on “ Ensemble Member States, in particular for former Ms Mercedes Rodriguez Tarrida as Prediction System – 15 years on ” Co-operating States. New membership deputy auditor (both from Spain) for showing the impressive development to ECMWF is possible when the the financial years 2008 –2011. of the EPS techniques at the Centre amended Convention is ratified by all N Scientific Advisory Committee . from the beginnings in the early 1990s current Member States (to date 11 Council appointed Dr Bouttier, Dr until today and outlining the future Member States have ratified). Rodriguez-Camino and Dr Adrian for potential of EPS. 11 th Workshop on Meteorological Operational Systems

HORST BÖTTGER, ALFRED HOFSTADLER, BAUDOUIN RAOULT, DAVID RICHARDSON, STEPHAN SIEMEN

THE 11 th Workshop on Meteorological N Meteorological visualisation month and for the season require Operational Systems was held at applications (Session 3) model bias corrections before they can ECMWF on 12 to 16 November 2007. The workshop proved to be very be passed to the users. Several The objective of the workshop was to popular with over 80 participants presentations addressed the use of review the state of the art of meteoro- from ECMWF Member States and Co- forecasts in weather risk management, logical operational systems and to operating States, and from other parts mainly using probabilistic forecast address future trends in: of Europe and beyond. information. Also consideration was N The use and interpretation of The presentations given in Session 1 given to applications based on medium- and extended-range forecast addressed in particular the post- predictions of severe weather events. guidance (Session 1) processing of direct model output to The weather alarm and warning N Operational data management meet user requirements. It was noted system in Europe is now much systems (Session 2) that extended-range forecasts out to a developed and there is an increasing

10 ECMWF Newsletter No. 114 – Winter 2007/08 NEWS

demand for early warnings of severe severe event. A number of suggestions forecasts for specific locations. While weather, mainly in the range of 3 to 5 were made for additional EFI care must be taken using some of days. parameters to allow alerts to be these tools for severe weather events, In Session 2, several presentations provided for a wider range of there are procedures that are suitable discussed the emerging new distrib- situations, including snowfall and for such situations. It is important to uted information systems, such as the minimum temperature. There was calibrate longer-range (monthly and WIS (WMO Information System) and also interest in the ongoing work to seasonal) forecasts with the help of the system created to handle the develop products to complement the re-forecast datasets to account for TIGGE (THORPEX Interactive Grand EFI. Probabilities of exceeding model error. Recent research has Global Ensemble) data exchange and thresholds relative to the climate shown that there is also potential data service. Providing the users with distribution and warnings expressed benefit from using re-forecast datasets efficient interfaces for data mining in terms of return periods were to calibrate medium-range model and retrieval and ensuring interoper- considered to be potentially valuable output. The planned ECMWF ability were seen as the primary tasks additional information. re-forecasts from the unified monthly/ when creating these systems. The operational ECMWF tropical VarEPS system will allow users to Several operational data manipula- cyclone tracking was considered to be calibrate medium-range forecasts to tion and visualization systems were very useful, and development of complement the current statistical presented in Session 3. Also there products showing tropical cyclone downscaling. were demonstrations of some systems genesis during the forecast was The verification of severe weather during the exhibition. encouraged. It would be worthwhile events was acknowledged to be an Three working groups were estab- investigating corresponding products important but difficult topic. It was lished to discuss issues relevant to the for extra-tropical cyclones. noted that this is an area of active session topics. In general the direct model output research, but a substantial amount of is not provided to the end users. work remains to be done. Working group on severe and Rather, statistical post-processing high-impact weather events procedures are applied. A variety of Working group on interoperability The working group discussed the methods, including Model Output Traditionally Meteorological Services requirements for forecasting severe Statistics (MOS), Kalman filtering and have been very successful at and high-impact weather events, “Perfect Prognosis” are used, often establishing interoperability within focusing particularly on early warnings incorporating locally available their community. Strong governance for the medium-range. observations. Such procedures can through WMO allowed the definition The Extreme Forecast Index (EFI) add significant value to the products of well defined and agreed standards was noted to be widely used as an delivered to end users and are and methods. Increasingly there has alert to forecasters of a potential considered essential to provide skilful been the need to participate in inter-

11 NEWS ECMWF Newsletter No. 114 – Winter 2007/08

disciplinary activities, mainly driven any semantics. Such semantics are can be a challenge for network band- by commercial, scientific and defined in data schemas that still need width. humanitarian interests, but also new to be defined and agreed upon. For In addition there were many legal requirements like the European these reasons, it would be best for use questions raised which would need to INSPIRE directive. Consequently it of XML to be limited to the exchange be addressed before larger web has been necessary to review and of small data items with other applications could be developed: adjust these standards in a way that communities. It was agreed that N How scalable will the applications will allow the Meteorological Services within the meteorological community, be? to interoperate with a much wider data exchange should continue to be N How to deal with demand, and community in the near future. driven by efficiency. therefore availability, which can be The Internet was seen as a There was general agreement that weather dependent? successful example of the benefit of OGC (Open Geospatial Consortium)/ N Are there any development tools interoperability: a small set of simple, ISO standards will play an important (debugging and profiling)? stable, non-proprietary and accepted role in the near future. Both the Euro- N How can a web application be standards that connect the inform- pean INSPIRE directive and Euro- operationally maintained? ation providers to a large, partly control, for example, have decided to N What are the licences and legal unknown user community, leading to define OGC compliant data represent- issues? unforeseen usage of the published ations based on GML (Geography N How can the security be guaranteed? information. Markup Language). Meteorological Further it was questioned whether a Similarly, it was recommended that Services are already in the process of full featured web-based meteorolog- the meteorological community should building expertise on OGC but more ical application should be the aim of use a limited number of standards collaboration is needed. Better coord- developments. The desktop based that impose minimal constraints on ination of the activities could lead to a workstation has progressed well over the users. This would open up a large common reference implementation of the last decade and gives good services variety of information from many OGC compliant web services for to forecasters. different providers to known and meteorological data, which could be Instead of heavy rich internet unknown user groups who might also proposed to WMO for standardisation. applications, the possibility of using find new ways to make use of our data smaller, lighter web applications was sets. Working group on visualisation discussed. These would be tailored to To continue the analogy, the and web applications the need of the users. Accepted Internet standards cover data formats The working group discussed the standards could be used to exchange (HTML), transfer protocols (HTTP) possibility of porting a meteorological information from different disciplines. and requesting mechanisms (URL). workstation from a desktop applica- Standards already considered by many Again, the meteorological community tion to a web application, a so-called are the web services defined by the must define similar standards: rich internet application. The idea was OGC. See also the conclusion of the N Data formats : GRIB, BUFR, CF- fuelled by many talks at the workshop working group on interoperability. NetCDF, (GML_BUFR), GeoTIFF, which indicated an increasing trend of A further workshop to exchange KML, GIF, PNG, JPG. providing web-based products to fore- experience on OGC/GML, covering N Transfer protocols : (s)ftp, http(s), casters and clients. Also recent both data management and graphics, DVB, OGC Web Services, SMS, developments in web technologies would be beneficial to facilitate such VOICE, RSS. have meant that many web applica- cooperation. N Requesting mechanisms : WMO file tions have appeared and had much naming convention, OGC Query success, increasing the expectations of Further information language. web users. The findings of the working groups The use of XML based data formats It became clear that the time is not were presented and discussed in a was discussed. Because of the support yet right for a web-based workstation. final plenary session which concluded by industry and the wealth of available The reasons mentioned were: an informative and successful tools these formats are very popular. N Calculations performed using workshop. Summaries of the working On the downside, the representation substantial local resources are still groups along with the workshop in XML can be bloated and produce required in meteorology. programme and presentations can be very large files, and processing such N User interface toolkits are not found at: files can be CPU intensive. mature enough – fast changing and www.ecmwf.int/newsevents/meetings/ Furthermore, the XML format is just not yet future safe. workshops/2007/MOS_11/index.html an agreed syntax and does not specify N Download of (initial) application

12 ECMWF Newsletter No. 114 – Winter 2007/08 NEWS

New High Performance Computing Facility

MANFRED KLÖPPEL

At its 68 th session on 10-11 December 2007, the ECMWF Council authorised the Director to enter into a service contract with IBM United Kingdom Ltd to replace the High Performance Com- puting Facility (HPCF) at the Centre from 2009 onwards. The contract between ECMWF and IBM UK Ltd was signed on 20 December 2007. IBM will provide a POWER6 system for the period 2009 –2010, to be replaced with a future POWER system for the period 2011 to mid-2013. The Ms C Isaac and Mr A Sharples of IBM signing the contract with the Director of ECMWF, new computer system will enable Dominique Marbouty, for the next High Perfomance Computing Facility at ECMWF. ECMWF to further improve its global under climate change. ECMWF will warnings will be crucial to save lives, NWP models, in particular by contribute to mitigating the effects of enabling people to be evacuated from increasing the horizontal resolution of climate change by providing early endangered areas (e.g. in the event of the models and by making the best use warnings of severe weather. This is the storm surges in the North Sea, such of satellite data. seen as a key contribution to help as in November 2007), or to take Severe weather is predicted to society adapt to the dangers and precautionary action to avoid major become more frequent and more threats associated with global warming. threats to goods during severe storms, intense in some parts of the world The increased time gained by such such as ‘Kyrill’ in January 2007. ECMWF Educational Programme 2008

ELS KOOIJ-CONNALLY The course is aimed at both current and potential users of the Centre’s facilities. ECMWF has an extensive education N Use and interpretation of ECMWF and training programme to assist products . This discusses the ECMWF Member States and Co-operating products in operational weather fore- States in the training of scientists in casting available to the Member States. numerical weather forecasting, and in It is mainly aimed at forecasters or making use of the ECMWF computer people with forecasting experience. and archive facilities. N Numerical Weather Prediction . A booklet has been prepared that This covers various aspects of research describes the content of the courses in NWP at ECMWF, including data and methods of applications. The assimilation and use of satellite data, training courses consist of modules numerical methods and adiabatic that can be attended separately. Since formulation of models, parametrization the courses do not vary much from of diabatic processes, and predict- one year to another someone may ability, diagnostics and extended-range decide to attend different modules in forecasting. separate years. The courses fall into In addition there is a one-week The booklet describing the educa- three broad categories. seminar consisting of a series of tional programme can be obtained by N Use of computing facilities . This lectures dedicated to one specific contacting Els -Kooij.Connally@ecmwf. int. provides an introduction to ECMWF’s topic. In 2008 the subject is Alternatively information can be computing and archive facilities. The “Parametrized physical processes and accessed via the ECMWF website by topics covered include SMS/XCDP, their dynamical interaction ”. More going to: GRIB API, MARS, MAGICS, METVIEW details about the seminar are given in www.ecmwf.int/newsevents/ and use of supercomputing resources. the news item on page 8. training/2008

13 NEWS ECMWF Newsletter No. 114 – Winter 2007/08

Fifteenth anniversary of EPS

TIM PALMER, ROBERTO BUIZZA 9 EPS 8 Deterministic forecast DECEMBER 2007 saw the fifteenth EPS control anniversary of the ECMWF Ensemble 7 Prediction System (EPS). In fact the EPS was run in real time from 6 November 1992, but EPS products were disseminated from December 5 onwards. In those days, the EPS ran L e just three days a week; full daily a 4 d t running began in May 1994. Since i m

September 2006, the EPS has been e 3 ( d extended to 15 days with a variable a y s resolution (VAREPS) system: T399 up ) 2 to forecast day 10 and T255 after- 1995 1997 1999 2001 2003 2005 2007 wards, with initial uncertainties Lead time (days) at which the Ranked Probability Skill Score of 850 hPa temperature for simulated using singular vectors and a the northern hemisphere extra-tropics has dropped to a value of 0.3 for the EPS (blue), the stochastic representation of model high-resolution deterministic forecast (black) and the control forecast of the EPS (green). The threshold of 0.3 has been selected so that the lead time of the EPS in 2006 corresponds uncertainties. A summary of progress to the lead time at which the high-resolution deterministic forecast loses useful skill in terms and developments over the last 15 of anomaly correlation. The Ranked Probability Skill Score is computed using a climatological years is described in ECMWF Tech. distribution as reference forecast – a value of 1 corresponds to a perfect deterministic fore - cast and a value of 0 to no skill compared to climatalogy. The lead times are computed from Memo. No. 540 . low-pass filtered monthly scores. Although the first operational ensemble forecast system was implemented at the Met Office in the After the implementation of the Although in the early days it was 1980s (for monthly forecast customers) , medium-range ensemble, use of not clear whether the EPS would have the operational implementation of the ensemble prediction has become a permanent place in the operational EPS at ECMWF saw a real explosion of commonplace on all timescales; schedule, one can safely say that the activity in operational probabilistic seasonal and short-range ensemble EPS is here to stay! We look forward forecasting worldwide. This activity forecast systems were developed later to the next 15 years of developments continues to the present day and global in the 1990s, and climate-change in ensemble-based operational ensemble forecasts have become part of ensembles and ensembles of analyses probabilistic forecasting. the operational suites of ten meteoro- were (and continue to be) developed logical centres: ECMWF, BMRC, CMA, in the current decade. Meteorological centres producing CPTEC, FNMOC, JMA, KMA, Met The EPS has become a standard global ensemble forecasts Office, MSC and NCEP. Other centres tool in the forecast office, and is used N ECMWF (e.g. Météo-France, Spanish Instituto routinely by a number of commercial N Australian Bureau of Meteorology Nacional de Meteorología, Norwegian and humanitarian organisations for (BMRC) Meteorological Institute and COSMO whom a quantifiable estimate of N Chinese Meteorological Consortium incorporating the German, forecast uncertainty is paramount. Administration (CMA) Greek, Italian, Polish, Romanian and This year saw reports of very valuable N Brazilian Centro de Previsão de Swiss National Meteorological Services) probabilistic flood predictions issued Tempo e Estudos Climáticos (CPTEC) are running or testing short-range using hydrological models driven by N United States Fleet Numeric Meteo- regional ensemble systems. the EPS (e.g. the flooding of two rivers rological Operational Center (FNMOC) In December 2007, eight of the ten in Bangladesh in the summer of N Japanese Meteorological organizations (ECMWF, BMRC, CMA, 2007). Administration (JMA) JMA, KMA, Met Office, MSC and The beginning of 2008 will see the N Korean Meteorological Administration NCEP) are delivering medium-range range of VAREPS extended to 32 days (KMA) global forecasts and one centre once-a-week, when the 15-day N Met Office (Météo-France) is deliverying short- VAREPS and the coupled ocean- N Meteorological Service of Canada range regional forecasts to the TIGGE atmosphere monthly ensemble (MSC) (THORPEX Interactive Grand Global systems are joined into a unified N National Centres for Environmental Ensemble) archive at ECMWF. Also ensemble system ranging from day 0 Prediction (NCEP) CPTEC has started sending test data. to day 32.

14 ECMWF Newsletter No. 114 – Winter 2007/08 METEOROLOGY

Toward a forecast of aerosols with the ECMWF Integrated Forecast System

At present, the model prognostic aerosols are not JEAN-JACQUES MORCRETTE, LUKE JONES, interactive with the radiation scheme and are therefore JOHANNES KAISER, ANGELA BENEDETTI, OLIVIER BOUCHER passive tracers. However, their optical thickness is eval - uated at different wavelengths as diagnostic quantities that A PART of the EU-funded GEMS project (which can be compared to surface measurements such as those concerns global environmental monitoring using satel - taken by AERONET (Aerosol Robotic Network) or lite and in situ observations), a prognostic representation derived from satellite measurements like those of MODIS of aerosols is being developed in the ECMWF Integrated (Moderate Resolution Imaging Spectroradiometer). Forecast System (IFS) in both its analysis and forecast The analysis module has been developed and is modules. In this short note, consideration is given to the currently running for GEMS’s official reference study forward modelling, outside of the analysis. period, 2003 –2004. The analysis ingests MODIS aerosol optical depth observations and uses the total aerosol The forecast model mixing ratio as the control variable. Results so far are An experimental version of the forecast model now encouraging and show the capability of the analysis to accounts for five tropospheric aerosol types (i.e. sea-salt, draw to the observations. desert dust, organic matter, black carbon and a sulphate- For the results presented here, there is no assimila - related variable, SO 2). Both the sea-salt and dust are each tion of any data related to aerosols. The model is run represented by three bins, whose limits are chosen as to from a given starting date in a series of 12-hour forecasts have roughly 10%, 20% and 70% of the mass of each starting every 12 hours from the ECMWF operational aerosol type in the various bins. The package of physical analysis. The model aerosols are free-wheeling. They start parametrizations dedicated to aerosol processes is stan - from null concentrations of aerosols on the starting dard. Sources of sea-salt and desert dust are interactive with date. Then the various aerosols spin up for about 8 –12 surface and near-surface variables of the model. The days (the time their contents establish themselves), with surface flux of sea-salt aerosols is parametrized from the the aerosols produced from their surface emission 10-m wind at the free ocean surface. For the production fluxes, and going through the physical processes (dry of desert dust, the source depends on the 10-m wind, soil deposition, sedimentation, wet deposition). After 12 moisture, the UV-visible component of the surface albedo, hours of the forecast the aerosols are stored and passed and the fraction of cover by vegetation when the surface as initial conditions for the next forecast starting 12 is snow-free. Other aerosol sources are taken from hours later. This is in essence not very different from monthly-mean climatologies or inventories (Global Fire what is done within a transport model, except that the Emission Database, Speciated Particulate Emission Wizard, aerosol processes are fully consistent with the dynam - Emission Database for Global Atmospheric Research) ics and all physical parametrizations. until more temporally-resolved data based on satellite observations are provided as part of the GEMS project. Some results All aerosols undergo sedimentation, and dry and wet Since 1 June 2007 this experimental version of the fore - deposition (this last one by large-scale and convective cast model, which includes prognostic aerosols and precipitation). For organic matter and black carbon, two runs at T159L91, produces near-real time aerosol fore - components, hygrophobic and hygrophilic, are consid - casts that are available from the ECMWF web site at: ered. SO 2 is considered as one variable with no explicit www.ecmwf.int/products/forecasts/d/inspect/catalog/ chemistry included. Recent developments in the model research/gems/aer/ dynamics and package of physical parametrizations Figure 1 illustrates the potential of the future aerosol allow the aerosols to be advected, and the vertical diffu - forecasts. For an ascending orbit over Africa (Figure 1(a)), sion and the mass-flux convection schemes to account a classification of cloud and aerosol produced by the explicitly for tracers such as aerosols. The wet and dry CALIPSO Science Team (Figure 1(c)) is compared with deposition schemes are standard, whereas the sedi - the corresponding model cloud and aerosol (Figure mentation of aerosols follows closely the scheme recently 1(d)). Even for this relatively low horizontal resolution introduced for the sedimentation of ice particles. (T159), the ECMWF model generally produces the cloud and aerosol in the proper location both horizontally AFFILIATIONS and vertically. Over the same orbit, the total aerosol Jean-Jacques Morcrette , Luke Jones , optical depth at 550 nm produced by the model is Johannes Kaiser , Angela Benedetti : ECMWF, Reading, UK compared to the equivalent optical depth retrieved from Olivier Boucher : Met Office, Exeter, UK MODIS observations over ocean and dark land surfaces

15 METEOROLOGY ECMWF Newsletter No. 114 – Winter 2007/08

a Satellite track c Cloud / aerosol classification from CALIPSO Clouds Aerosols Stratospheric clouds Total attenuation 12:42:14 12:47:14 12:52:14 12:57:14 13:02:14 100

200

300 400 500 P r e

b Aerosol optical depth at 550 nm s 700 s u

MODIS Model r 50N 0.97 1e 000 (

h –24.4 –6.3 11.8 29.9Latitude 47.8 P

0.9 a 40N ) 19.2 15.1 11.3 7.0Longitude 1.4 0.83 0.75 d Model aerosol and cloud 30N 100 0.68 20N 0.61 0.54 200 10N 0.46 300 0.39 400 EQ 0.32 500 P

0.25 r e

10S s 700 s

0.17 u r 0.1 1e 000 (

20S h

P 20°E 18°E 16°E 14°E 12°E 10°E 8°E 6°E 4°E 2°E

0 a 10W 0 10E 20E 30E10W 0 10E 20E 30E ) 20°S 10°S 0° 10°N 20°N 30°N 40°N O

Fip gure 1 (a) The orbit of the A-train of satellites on 16 July 2007 between 1242 and 1302 UTC. (b) The aerosol optical depth at 550 nm t i c

dea rived from MODIS-Aqua observations (left) and produced by the ECMWF forecast model (right). (c) The cloud/aerosol classification l ded rived from CALIPSO measurements along the orbit shown in panel (a). (d) The cross-section along the same orbit as used for (c) show - e ingp the aerosol (yellow) and cloud (blue) quantities produced by the ECMWF forecast model. The MODIS and CALIPSO data were t doh wnloaded from the NASA Giovanni server. a t 5 5 0 a Blida n 0.9 m 0.8 0.7 0.6 0.5 0.4 0.3

O 0.2 p t

i 0.1 c a l

d 0 e

p 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 1 t

h July 2007 August a

t b Nes Ziona

5 1 5 0 0.9 AERONET n Total m 0.8 Sulphate Figure 2 The observed optical depth at 550 Dust nm from AERONET stations at (a) Blida and 0.7 Organic matter (b) Nes Ziona for July 2007 compared to the Sea salt values from the model (total optical depth 0.6 Black carbon 0.5 plus contributions from various aerosols). Dust aerosols dominates the optical depth at 0.4 Blida, whereas sulphate aerosol dominates 0.3 in Nes Ziona. Data for Blida and Nes Ziona were obtained from the AERONET web site. Thanks 0.2 go to B. Holben, A. Karnieli and M. Boughedaoui 0.1 for their efforts in establishing and maintain - 0 ing the Blida and Nes Ziona AERONET sites. 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 1 July 2007 August

16 ECMWF Newsletter No. 114 – Winter 2007/08 METEOROLOGY

in the absence of extended cloud cover (Figure 1(b)). physical parametrizations, but also on the representa - For desert dust, the agreement is usually good reflect - tiveness of the sources. With the exact sources of aerosols ing the high quality of the initial conditions and of the (in particular, those of anthropogenic origin) not avail - atmospheric motions in the subsequent 12-hour forecast. able in real time, the aerosol analysis, through the Over Central Africa, the sources of sulphate, and organic assimilation of aerosol-related observations, will provide and black carbon aerosols linked to biomass burning are initial conditions representative of the true aerosol well represented in the inventories, and the agreement burden in the atmosphere. The development of a on the optical depth of the plume moving towards the successful aerosol analysis is therefore fundamental to South Atlantic is also good. the quality of the subsequent aerosol forecast. Here As a further example, Figure 2 compares the model the forecast model including prognostic aerosols was aerosol optical depth at 550 nm over Blida (Algeria) and shown to provide a reasonable basis for this analysis. Nes Ziona (Israel) to that derived from AERONET surface observations at those stations for July 2007. FURTHER READING These results show that the model describes reason - Morcrette , J.-J ., P. Bechtold , A. Beljaars , A. Benedetti , ably well the temporal variability of the aerosols. The A. Bonet , F. Doblas-Reyes , J. Hague , M. Hamrud , J. Haseler , model also captures the dominant aerosol type, dust in J.W. Kaiser , M. Leutbecher , G. Mozdzynski , M. Razinger , Blida and sulphate in Nes Ziona. D. Salmond , S. Serrar , M. Suttie , A. Tompkins , A. Untch & The quality of the results depends not only on the A. Weisheimer , 2007: Recent advances in radiation transfer dynamics of the model and the adequacy of the aerosol parametrizations. ECMWF Tech. Memo. No. 539 . A new partitioning approach for ECMWF’s Integrated Forecast System

day forecast model at T799L91 resolution, and a fifteen- GEORGE MOZDZYNSKI day ensemble prediction system (EPS). The EPS comprises a control forecast at T399L62 resolution and a further 50 SINCE the mid-1990s the Integrated Forecast System forecasts at T399L62 using perturbed initial states. These (IFS) at ECMWF has used a two-dimensional scheme for run out to ten days, when the resolution is truncated to partitioning grid-point space to MPI tasks where MPI T255L62 before completing the runs out to fifteen days. is a protocol for the data exchange and synchronisation The key parallelisation scheme for IFS was developed between the executing tasks of a parallel job. While in the mid-1990s ( Barros et al. , 1995; Isaksen & Hamrud , this scheme has served ECMWF well there has still been 1996). This scheme consists of a series of data transpo - some areas of concern, namely: sitions between grid-point space, Fourier space and N Communication overheads for IFS reduced grids at spectral space as shown in Figure 1. With this approach, the poles to support the Semi-Lagrangian scheme. the complete data required is redistributed at these N The halo requirements needed to support the inter - stages of a time step so that the computations between polation of fields between model and radiation grids. two consecutive transpositions can be performed with - The halo is an area of a task’s memory reserved for out any inter-process communication. data that is owned by neighbouring tasks. The focus of this article is the distribution of data in These issues have been addressed by the imple - grid-point space where the bulk of the IFS computa - mentation of a new partitioning scheme called tions take place for the physics and dynamics aspects of EQ_REGIONS which is characterised by an increasing the model. This two-dimensional distribution of grid- number of partitions in bands from the poles to the point space to 256 MPI tasks is shown in Figure 2(a) equator. The number of bands and the number of parti - using an Aitoff projection. Here each task is associated tions in each particular band are derived so as to provide with a single partition. This projection allows us to easily partitions of equal area and small ‘diameter’. The see the main features of this distribution for the whole EQ_REGIONS algorithm used in IFS is based on the globe, with 16 north-south bands (or sets) and 16 east- work of Paul Leopardi, School of Mathematics, west bands, where 16 is the square root of 256. For 512 University of New South Wales, Sydney, Australia. tasks there is no exact square root, so the nearest factors of 512 are used, namely, 32 north-south bands and 16 east- IFS parallelisation west bands, as shown in Figure 2(b). The reason for The Integrated Forecast System (IFS) at ECMWF consists having more north-south bands than east-west bands in of a large suite of software used primarily to produce a this case is down to performance, as IFS applications daily ten-day forecast. Key components of the IFS include run marginally faster this way. Finally, Figure 2(c) shows the processing of observations, a 4D-Var analysis, a ten- a case for 1024 MPI tasks.

17 METEOROLOGY ECMWF Newsletter No. 114 – Winter 2007/08

Grid Point Computations made a small difference on performance and some - times had a negative effect. What was required was a new B approach for the partitioning of grid-point space. z P1 P2 There are clearly many approaches to partitioning a A λ P3 P4 sphere, a good example being that of Lemaire & Weill T r a g (2000) which uses constant area quadrangles as shown μ n s → p l

o in Figure 3. However, many of these approaches share e s s e o little in common with the IFS 2D partitioning; as a g p s P → n z V 1

a result they would require a major effort (many person l r W

T P 2 P3 years) to incorporate any of them into the IFS. λ P4 Inverse μ Fourier Fourier EQ_REGIONS Transform Transform The EQ_REGIONS partitioning scheme ( Leopardi , 2006) z V P1 P W 2 P3 appeared attractive from the beginning, as can be seen

m T l

P r in Figure 4 for a sphere with 33 partitions. This practi -

4 a n →

μ s cal algorithm came with a MatLab package titled m p o e W s s

e Recursive Zonal Equal Area Sphere Partitioning Toolbox o p

m=0 m=3 m=4 m=1 m=2 m=5 m=6 l → s http://eqsp.sourceforge.net/

n which can be found at . As a m a r z V T comparison IFS 2D partitioning would need 11 bands P1 P3 m P2 P4 2D partitioning Inverse μ a 256 MPI tasks Legendre W Legendre Transform Transform m=0 m=3 m=4 m=1 m=2 m=5 m=6 z P V P1 3 T r m P P4 a 2 n s

→ m p s o e s s

n e o W m p s →

n NN a r m=0 m=3 m=4 m=1 m=2 m=5 m=6 s T P2 z P1 P3 P4

m n b 512 MPI tasks Spectral Computations

Figure 1 IFS model time step showing data transpositions. The Semi-Lagrangian scheme in the IFS requires data from neighbouring tasks, the width of this halo being a function of the maximum possible wind speed and the time step. It is clear that the ideal shape for two- dimensional partitions (hereafter referred to as 2D partitions) to achieve the smallest halo area is square- like. However, the 2D partitions shown in Figure 2 are more rectangular than square-like with a width to height c 1024 MPI tasks ratio of 2:1 or 4:1, and further they have a nasty ‘cake’ feature for those partitions at the poles. These polar partitions present a problem as they not only require a large halo area, but also that this halo area requires communication with all partitions converging on the relevant pole. Reducing the number of north-south bands and increasing the number east-west bands would make partitions more square-like at the equator, however this would only make matters worse at the poles with the increased communication. Some attempts to resolve the polar halo problems were made by using a differ - ent partitioning strategy for the first and last partitions Figure 2 2D partitioning of grid-point space to (a) 256 MPI tasks, (i.e. a polar cap). However, these developments only (b) 512 MPI tasks and (c) 1024 MPI tasks for the T799 model.

18 ECMWF Newsletter No. 114 – Winter 2007/08 METEOROLOGY

Figure 3 Partitioning the sphere with constant area quadrangles Figure 4 Partitioning of a sphere into 33 regions using the (taken from Lemaire &Weill , 2000). EQ_REGIONS MatLab package. north-south and 3 bands east-west for this number of sioned array) used to store the number of partitions in partitions. The reason why EQ_REGIONS was attractive each band. is due to the similarity between the IFS 2D bands and The results for this balanced EQ_REGIONS imple - what EQ_REGIONS calls collars – the partitioning of a mentation can be seen in Figure 6(a) for 256 partitions, sphere resulting in two polar caps plus a number of Figure 6(b) for 512 partitions and Figure 6(c) for 1024 collars with increasing partition counts as we approach partitions. The characteristic features of this parti - the equator. tioning approach are square-like partitions for most of The description of the algorithm and mathematical the globe and polar caps, together with a significant proof are described in great detail in Leopardi (2006). improvement in the convergence at the poles. This algorithm results in partitions of equal area and From a code point of view the differences between small ‘diameter’. However, this would not be sufficient the old 2D partitioning and the new EQ_REGIONS for an IFS implementation, as the density of grid-points partitioning are relatively simple as shown in Box A. on the globe varies with the latitude, the greatest density being at the poles and the least density at the equator. Radiation Grid This imbalance is 13% for a T799 model with 512 parti - Radiation computations in the IFS (as in other models) tions when using the EQ_REGIONS algorithm to are very expensive and to reduce their relative cost we provide the bounds information (start/end latitude, both reduce the frequency of such computations (i.e. start/end longitude) for each partition as shown in every hour for a T799 model) and run these computa - Figure 5. tions on a coarser grid than the model grid. This coarse The solution to this imbalance issue was to use the grid was initially a sampled model grid ( Dent & EQ_REGIONS algorithm to only provide the band Mozdzynski , 1996) but more recently a reduced grid information, i.e. the number of north-south bands and (linear) was used which gave a small improvement in the number of partitions per band. Then the IFS parti - meteorological skill. As an example, for a T799 model tioning code would use this information in a similar way grid (843,490 grid points) the corresponding radia - to that used for 2D partitioning, resulting in an equal tion grid would be T399 (213,988 grid points), with an number of grid-points per partition. With this approach approximate ratio of 4:1. Of course to use this reduced there was only one new data structure (an single-dimen - radiation grid we must interpolate data from the model

1850

1800

1750

1700

1650

1600

N 1550 u m

b 1500 Figure 5 Number of grid-points per parti - e

r tion for T799 resolution using EQ_REGIONS o

f 1450 ‘area’ partitioning for 512 tasks. g r

i 1 65 129 193 257 321 385 449 513 d -

p Partition o i n t s

19 METEOROLOGY ECMWF Newsletter No. 114 – Winter 2007/08

EQ_REGIONS partitioning Box A a 256 MPI tasks Coding differences between 2D and EQ_REGIONS partitioning

For the 2D partitioning there were loops such as: DO JB=1,NPRGPEW DO JA=1,NPRGPNS ... ENDDO ENDDO where NPRGPEW and NPRGPNS are the number of east-west and north-south bands (or sets). b 512 MPI tasks For EQ_REGIONS partitioning loops were simply transformed into: DO JA=1,N_REGIONS_NS DO JB=1,N_REGIONS(JA) ... ENDDO ENDDO where N_REGIONS_NS is the number of north-south EQ_ REGIONS bands. N_REGIONS(:) is an array containing the number of partitions for each band. In total some 100 IFS routines were modified with such transformations. It should be noted that the c 1024 MPI tasks above loop transformation supports both 2D and EQ_REGIONS partitioning, i.e. to use 2D parti - tioning, a simple namelist variable would be set LEQ_REGIONS=F , which would result in the follow - ing initialisation: N_REGIONS_NS=NPRGPNS N_REGIONS(:)=NPRGPEW

routines needed for supporting the halos already existed for the Semi-Lagrangian scheme and could easily be Figure 6 EQ_REGIONS partitioning of grid-point space to (a) 256 MPI tasks, (b) 512 MPI tasks and (c) 1024 MPI tasks for the T799 model. used for the radiation interpolation. However, there was a downside to this second grid required for the radiation computations and after approach. The halos required for the radiation inter - such computations interpolate data back to the model polations were much larger than one would expect, grid from the radiation grid, in both cases using 12 particularly for large numbers of tasks. It was only after point bi-dimensional interpolations. the geographic position of some partitions that required Two approaches are possible for such interpolation. large halos were plotted that the problem was understood. The first approach would be to send each field to a sepa - Figures 7 and 8 show the model and radiation partitions rate task using a global gather operation, perform the (using 2D partitioning) for task 201 (Africa partition) and interpolation and then return the data by using a global task 11 (Polar partition) of 512 for a T799 model and scatter operation. This approach has the disadvantage T399 radiation grid. One would have expected parti - that global operations are relatively expensive on today’s tions for two different grids to be at the same geographic parallel computers, particularly when the number of position, as they use the same partitioning code. The fields (variables × levels) to interpolate is less than the reason this is not the case stems from the definition of number of tasks being used. The second approach, the these two grids, one is clearly not a projection of the other. one used in IFS, was to use a halo for the data required Differences between these two grids are due to the from neighbouring tasks and perform the interpolations requirements for equal spacing of grid points in each grid locally. This approach was assumed to be better as all (with the exception of polar latitudes). In addition there tasks are used for the interpolation and only local is a need for the number of points on the same line of communication is used. The second approach also had latitude to be divisible by factors 2, 3, and 5 – a require - an advantage from a code maintenance viewpoint as the ment for the Fourier transforms (see Figure 1).

20 ECMWF Newsletter No. 114 – Winter 2007/08 METEOROLOGY

It is now interesting to see the effect of using EQ_REGIONS parti - tioning and how it addresses the above problems. To make a fair comparison, tasks 220 and 4 were chosen that by comparison had a relatively large halo but also were located in comparable geographic positions (Africa and Polar) to tasks 201 and 11 for 2D partitioning. This can be seen in Figures 9 and 10 respectively. The conclusion of this compari - son is that EQ_REGIONS partitioning requires smaller halos than 2D partitioning for the interpolation when using the same number of Figure 7 2D partitioning – model (blue) and radiation (red) partitions for task 201 (Africa tasks. This translates into less data partition) of 512 tasks for a T799 model/T399 radiation grid. being communicated and therefore improved performance. To take this further Figure 11 shows graphically the halo area (total halo grid points including a partition’s own grid points) for all tasks for the radiation grid interpolation. The top two lines show the halo area required for in - terpolating from the model to the ra - diation grid and the bottom two lines for interpolating from the ra - diation to the model grid. In both cases it can clearly be seen that EQ_REGIONS partitioning results in smaller halos when compared with 2D partitioning. 4D-Var

Besides the model and radiation Figure 8 2D partitioning – model (blue) and radiation (red) partitions for task 11 (Polar grids presented earlier, there is partition) of 512 tasks for a T799 model/T399 radiation grid. another grid used in 4D-Var analyses – it is part of the J b wavelet scheme (Fisher , 2003; Fisher , 2004) within the minimisation steps of a 4D-Var cycle. The grid used in this scheme is a ‘full grid’, where all latitudes contain the same number of points. For the J b wavelet scheme many such grids of increasing resolution are used (the wavelet scales) from some minimum truncation (default being T21) up to the resolution of the minimisation step. As an example, for a T255 minimisation step, wavelet grids are used at T255, T213, T159, T127, T95, T63, T42, T30, and T21. This can present a problem when scaling to large numbers of tasks as the lowest trunction T21 only has 2048 grid Figure 9 EQ_REGIONS partitioning – model (blue) and radiation (red) partitions for task points for a full grid (32 latitudes 220 (Africa partition) of 512 tasks for a T799 model/T399 radiation grid.

21 METEOROLOGY ECMWF Newsletter No. 114 – Winter 2007/08

each with 64 points). Of course the performance of the J b wavelet code is dominated by the higher wavelet scales so one should not pay too much attention to the low resolu - tions. Unfortunately, when testing EQ_REGIONS partitioning it was exactly the lowest resolution that presented a problem; the problem being a constraint of IFS partitioning code that required that a line of lati - tude could not be split more than once between bands (or A-sets). For 2D partitioning this was not found to be a problem. After some investigation and discussion with the author of the J b Figure 10 EQ_REGIONS partitioning – model (blue) and radiation (red) partitions for task 4 wavelet scheme a solution was found. (Polar partition) of 512 tasks for a T799 model/T399 radiation grid. The solution was simply to use reduced grids (preferably linear While the overall performance advantage for using EQ_REGIONS may grids) instead of full grids. It was only be a few percent, one should bear in mind that IFS applications have decided to implement this uncon - had many years of code optimisation and finding even a few percent is ditionally as it was realised that the increasingly rare. ‘Rich pickings’ are often found in new code, but less so overall performance of a 4D-Var with code that has been around a long time. cycle improved by about 2% inde - By inspecting some low-level timers in the IFS, it can be seen that some pendent of whether 2D partitioning areas of code are now running faster due to a combination of reduced halo or EQ_REGIONS partitioning was sizes used for the Semi-Lagrangian scheme and radiation grid interpola - used. The reason for this perform - tion, and the associated reduction in memory use (always good for ance improvement was obvious, performance). However, it can also be seen that there is an increase in the about 30% fewer grid points were communications used for the transposition of data between grid-point used compared with full grids in the space and Fourier space (see Figure 1). This increased communication is above scheme. The wavelet scales due to the fact that the distribution of latitudes in Fourier space is a very used were preset in some large data close match to the north-south bands when using 2D partitioning. The trans - files, so it was not practical in this position to Fourier space from grid-point space requires complete latitudes implementation to change these. Further, some of these scales were 5000 not linear grids (e.g. T213 is a quad - 4500 ratic grid). In the near future it is planned to reset the wavelet scales so 4000 that all have corresponding linear grids. This is expected to further 3500 improve the overall performance of 3000 4D-Var by an additional 1%. 2500 Performance H

a 2000

In Table 1 there is a comparison of l o the performance of a T799 forecast a r

e 1500 model and 4D-Var analysis when a using 2D and EQ_REGIONS parti - 1000 tioning. The advantage in using 2D T799 to T399 2D T399 to T799 500 EQ_REGIONS was measured at EQ_REGIONS T799 to T399 EQ_REGIONS T799 to T399 3.9% for the forecast model and 2.7% for the 4D-Var analysis. For 0 4D-Var the 2% improvement gained 0 33 85 97 129 161 193 225 257 289321 353 385 417 449 481 513 Task number by using reduced grids for the J b wavelet grids discussed in the previ - Figure 11 Comparing EQ_REGIONS and 2D partitioning requirement for Halo Area (total ous section is not included in the halo grid points including a partition’s own grid points) for radiation grid interpolation for latter 2.7%. a T799 model/T399 radiation grid.

22 ECMWF Newsletter No. 114 – Winter 2007/08 METEOROLOGY

2D EQ_REGIONS 2D/ FURTHER READING Tasks × Application partitioning partitioning EQ_REGIONS Barros , S., D. Dent , L. Isaksen , G. Robinson , G. Mozdzynski Threads (seconds) (seconds) speed-up & F. Wollenweber , 1995: The IFS model: a parallel production weather code. Parallel Computing , 21 , 1621 –1638. Model 512 × 2 3648 3512 1.039 Dent , D. & G. Mozdzynski , 1996: ECMWF operational fore - 4D-Var 96 × 8 3563 3468 1.027 casting on a distributed memory platform: Forecast model. In Proc. of the Seventh ECMWF Workshop on the Use of Parallel Processors in Meteorology , G.-R. Hoffman and N. Kreitz editors, Table 1 Comparing the performance of 2D partitioning and EQ_REGIONS partitioning for a T799L91 IFS model and 4D-Var analysis. World Scientific, 144 –154. Fisher , M., 2003: Background error covariance modeling. as the first dimension, with model levels being distrib - In Proc. of ECMWF Seminar on Recent Developments in Data uted as the second dimension. For systems with many Assimilation for Atmosphere and Ocean , September 2003, CPUs per node (such as ECMWF’s IBM p5-575 clusters) http://www.ecmwf.int/publications/library/ecpublications/.pdf/seminar/ a relatively large part of this communication can be 2003/sem2003_fisher.pdf . performed within each node (when using 2D parti - Fisher , M., 2004: Generalized frames on the sphere, with tioning). We know that intra-node communication via application to the background error covariance modeling. memory is faster than inter-node communication via the In Proc. of ECMWF Seminar on Recent Developments in Federation switch. Therefore this seems a reasonable Numerical Methods for Atmospheric and Ocean Modelling , explanation why 2D partitioning is better for this partic - September 2004, http://www.ecmwf.int/publications/library/ecpublica - ular communication phase. This scenario naturally tions/-pdf/seminar/ 2004/sem2004.fisher.pdf . changes when truly large numbers of tasks (thousands) Isaksen , L. & M. Hamrud , 1996: ECMWF operational fore - or thinner nodes are used. In both these cases, most of casting on a distributed memory platform: Analysis. the data in such transpositions (if not all transposi - In Proc. of the Seventh ECMWF Workshop on the Use of Parallel tions) will need to be communicated via the systems Processors in Meteorology , G.-R. Hoffman and N. Kreitz editors, switch, with further gains to EQ_REGIONS over 2D World Scientific, 22 –35. partitioning. Time will tell. Lemaire , C. & J. Weill , 2000: Partitioning the sphere with EQ_REGIONS partitioning was introduced in IFS constant area quadrangles . 12 th Canadian Conference on cycle Cy31r2 (November 2006) as the default parti - Computational Geometry, 16 –19 August 2000, tioning scheme for grid-point space. This new citeseer.ist.psu.edu/lemaireOOpartitioning.html . partitioning approach together with ongoing develop - Leopardi , P., 2006: A partition of the unit sphere into ments should allow IFS to continue to scale well on regions of equal area and small diameter. todays and future high performance parallel computers. Electronic Transactions on Numerical Analysis , 25 , 309 –327. Probability forecasts for water levels at the coast of The Netherlands

Storm Surge Warning Service (SVSD) of Rijkswaterstaat, HANS DE VRIES the Public Works and Water Management Authority. KNMI, DE BILT, THE NETHERLANDS When there is a severe storm surge, the SVSD takes the final responsibility for the water level forecasts that ACCURATE storm surge forecasts are of paramount are issued and warns other appropriate authorities. importance for a country which, like The Netherlands, The standard tool available for numerical opera - has an important part of its economic activity and popu - tional storm surge forecasts is the two-dimensional lation concentration close to the coast in a part of the shallow-water model WAQUA/DCSM98. This model country which is near or even below sea level. The covers all of the NW European Continental Shelf, which primary usage is for coastal defence and the operation includes the North Sea, with a resolution of 8 km. of the storm surge barriers in the Oosterschelde and the Figure 1 shows the area covered by the model and the Nieuwe Waterweg, the entrance to the harbour of water depths. The atmospheric forcing for day-to-day Rotterdam. But the forecasts are also used for assisting forecasts for up to 48 hours ahead is provided by the shipping traffic in and out of the Rotterdam harbour 10-metre wind and mean sea level pressure from KNMI’s and by inland water management boards to assess their operational HIRLAM limited area model. This system ability to sluice water into the sea at low tide in case of provides sea level forecasts four times per day. A steady- heavy rainfall. state Kalman filter is used to assimilate water level Forecasts are issued by KNMI, the Royal Netherlands observations from tide gauges on the east coast of Meteorological Institute, in close cooperation with the Britain and the Dutch coast.

23 METEOROLOGY ECMWF Newsletter No. 114 – Winter 2007/08

2000 62°N

600

400

58°N 300

200 s e r t e M 150 54°N

100

50°N 20 10 Figure 1 Depth map of the WAQUA/DCSM98 storm surge model. 12°W 6°W 0° 6°E 12°E

Recently, there has been growing attention to the SVSD to indicate the severity of the surge. The proba - increased possibility of and the vulnerability to flood - bilities never span the full 100% range. One reason is ing due to the effects on sea level and storm climate that the derivation of the probabilities explicitly allows caused by global warming. This has prompted the exten - for a small fraction of the forecasts to lie outside the sion of the forecast range, together with the need for ensemble. Also ensemble members might be so close information on the uncertainty of the forecasts. together, especially at the edge of the ensemble, that ECMWF’s Ensemble Prediction System (EPS) is an obvi - parts of the probability density function collapse. For ous candidate to provide input for such forecasts. A forecasts less than 48 hours ahead, lack of spread in the comparatively simple model which calculates surges ensemble makes more of the probability fall outside the using a statistical technique had already been coupled ensemble and generally not more than the 25% to 75% to the EPS to produce forecast plumes. But now, the interval can be determined. increased capacity of computers allows for a water level The forecasts are being used, for example, for ensemble based on the numerical two-dimensional medium-range planning of the personnel required in storm surge model, which is generally more accurate case of a severe storm. But there is also a keen interest and also more flexible in its output. The results of this from inland water management boards. They use the ensemble are being used to produce probability fore - information on predicted low-tide levels for medium- casts for water levels in selected locations. range planning of the regulation of inland water levels when heavy rainfall is forecast. The forecasting system In November 2007 a storm occurred which reached WAQUA/DCSM98 with input from the EPS has been the alarm level for the whole of the Dutch coast, except running in real-time experimental mode since February for Vlissingen in the south where it fell just a few centime - 2007. With data from the winter 2003/04, hindcasts tres short. In Hoek van Holland the water reached the have been run to establish the calibrations which are highest level since the disaster in 1953. It was the first needed to generate probability forecasts from the time in more than 30 years that warnings and alarms had ensembles. Results for eight locations along the coast to be issued for the entire Dutch coast. The storm surge of The Netherlands are presented on the Internet. barriers in the Oosterschelde and the Maeslantkering Figure 2 gives an example of a forecast. and Hartelkering near Rotterdam had to be closed, the Probabilities derived from the ensembles are repre - latter two for the first time in the event of a storm surge. sented by bars with varying thickness and colours to The high levels were successfully captured as far as five indicate the range. Astronomical extremes are given as days ahead. Figure 2 shows the forecast for Hoek van black crosses and observations, if already available, as Holland three days ahead. Such forecasts are used to red circles. Also shown are the levels defined by the decide whether to close the barriers near Rotterdam.

24 ECMWF Newsletter No. 114 – Winter 2007/08 METEOROLOGY

3 AP GP WP 2 VP IP

W 0

a 97% t

e 93% r 90% l e -1 75% v

e 50% l

( 25% m 10% ) -2 7% 3% -3 0 24 48 72 96 120 144 168 192 216 240 6 Nov 7 Nov 8 Nov 9 Nov 10 Nov 11 Nov 12 Nov 13 Nov 14 Nov 15 Nov 16 Nov 00:00 00:00 00:00 00:00 00:00 00:00 00:00 00:00 00:00 00:00 00:00 Forecast time (hours)

Figure 2 Example of a probability forecast for Hoek van Holland (00 UTC on 6 November 2007). Astronomical extremes are given as black crosses and observations as red circles. The horizontal lines indicate the levels defined by the SVSD to indicate the severity of the surge: information (IP), pre-warning (VP), warning (WP), storm surge (GP) and alarm level (AP).

Calibration of deterministic forecasts shown by the dashed (average) and red lines (average The transformation of the ‘raw’ ensemble forecasts ± standard deviation). For wind speeds below 8 m s–1 the into probability forecasts and an evaluation of their averages lie close to the main diagonal, but the excess quality involved several steps. above this is on average 14% less in the ECMWF model The first step was the validation of deterministic than in HIRLAM. The green line in Figure 4 summarises water level forecasts with input from the deterministic this. Similar plots for wind directions show no system - high-resolution ECMWF atmospheric model. This was atic differences. done for the period September 2003 to April 2004, a A rerun of the storm surge model shows that apply - recent winter which contained a reasonable amount of ing a wind correction of +14% on the excess of 8 m s–1 storm surges. As one of the validation tools, the corre - indeed removes the systematic under-prediction of the lation between forecast tidal extremes and observations results. This is shown in Figure 3(b). For the control run is used. Figure 3(a) shows the correlation for IJmuiden. of the EPS the same correction is found. This is essentially a scatter plot which has been trans - formed into a two-dimensional histogram. The contour Calibration of the ensemble lines indicate the density of the points, starting from To assess the quality of the sea level ensemble and 10% below the highest level and go logarithmically derive probabilities, forecast ensembles have been down to single points. The deviation from the main diag - regenerated for the winter 2003/04, both with and onal, as seen in Figure 3(a), indicates a systematic without the correction of the wind speed described under-prediction of surges. earlier. The ensembles are studied by means of Rank As storm surge forecasts with input from HIRLAM do Histograms, also known as Talagrand Diagrams. To not suffer from such systematic under-prediction, a construct these, a ranking from low to high is established comparison was made between the 10-metre wind fields for the forecasts for each of the tidal extremes for each (i.e. the main input to the storm surge model) from of the sea level ensembles. This gives unequally spaced HIRLAM and the ECMWF model. This comparison is forecast intervals for each of the tides, which are unique shown in Figure 4 for a collection of wind fields for the for every ensemble forecast and for every tidal extreme. same forecast time from stormy periods. As in Figure 3, For an ideal forecasting system the realization of the this gives the densities of the two-dimensional histogram sea level is equally likely to fall in either of these inter - of the wind speeds over the North Sea from HIRLAM vals, including two for everything below or above the and the ECMWF model interpolated to the grid points ensemble. The Talagrand Diagram is a histogram of of the storm surge model. It shows an under-predic - these realizations. Its shape characterises the forecast - tion for higher wind speeds. To quantify this, averages ing system. In the ideal case where all intervals are and standard deviations of the densities have been populated equally, the histogram is flat. An ensemble calculated perpendicular to the main diagonal; these are system with too little variation between the members

25 F o r e c a s t h i g h t i d e

M( ETEOROLOGY ECMWF Newsletter No. 114 – Winter 2007/08 m )

a Original 10-metre wind the scatter between the different intervals and hence 2.5 gives a clearer characterisation of the system. Figure 5 gives an example of a Talagrand Diagram 2 and the corresponding CRH. The diagrams have been produced from all high tide forecasts from 60 to 96 1.5 hours ahead for Hoek van Holland. The red curves are without the correction of the 10-metre wind and the black curves include this correction. 1 The under-population of the lower and over-popu - lation of the higher forecast intervals indicate that the 0.5 ensemble considerably underestimates the surges. Although the correction of the 10-metre wind works the 0 right way and brings down the fraction of the forecasts

F where the observation is higher than the ensemble, o r e -0.5 from 12% to 8% in this case, it is not enough by far. c a

s Moreover, the under-population of the lower forecast t

h intervals does not change significantly. An explanation i g -1 h for that can be found in the observation that the lower t i d surges that populate these bins are caused by lower e

( -1.5 m winds for which the correction is smaller. ) -1.5 -1 -0.5 0 0.5 11.522.5 These features are typical for the whole range of the Observed high tide (m) Dutch coast and all forecast ranges from 48 hours ahead. b Corrected 10-metre wind For ranges up to 48 hours the diagrams are charac - 2.5 terised by too little variation in the ensemble members. This, however, is a well-known feature of the EPS. In 2 order to overcome the under-prediction, various corre - lations between ensemble parameters and observations have been investigated. The most promising is the rela - 1.5 tion between the standard deviation of the ensemble and the difference of the ensemble mean from the obser - 1 vation as illustrated in Figure 6; this figure shows an increase of the average difference with ensemble width. 0.5 30

-0.5 ) 1

-1 – 20 s m ( d e

-1.5 e p -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 s d n

Observed high tide (m) i w F

Figure 3 Validation of deterministic forecasts from T+48 to T+96 W 10 for high tides for IJmuiden with (a) original 10-metre wind and (b) M C corrected 10-metre wind. E leads to realizations which are often at the edges or outside of the ensemble, and hence gives a u-shaped histogram. However, too much variation gives under- population of the edges and hence a bell shape. A skew 0 diagram indicates under- or over-prediction. The 0 10 20 30 Talagrand Diagram can be summed to get a so-called HIRLAM wind speed (ms–1) Cumulative Rank Histogram (CRH). This gives the rela - Figure 4 Correlation between HIRLAM and ECMWF winds for tion between ensemble probabilities and observed stormy periods. Averages and standard deviations of the densities frequencies. An additional advantage of the CRH over calculated perpendicular to the main diagonal are shown by the the Talagrand Diagram is that the former smooths out dashed (average) and full red lines (average ± standard deviation).

26 ECMWF Newsletter No. 114 – Winter 2007/08 METEOROLOGY

a Talagrand Diagram b CRH Diagram 100 14 Without wind correction Without wind correction With wind correction With wind correction

12 80

10 60 8

6 40 O O b b s 4 s e e r r v v e e 20 d d f f r r e 2 e q q u u e e n n c c

y 0 y 0 ( ( % % 0 10 20 30 40 50 0 20 40 60 80 100 ) ) Bin number EPS probability (%)

Figure 5 (a) Talagrand Diagram and (b) CRH diagram for all high tide forecasts from T+60 to T+96 for Hoek van Holland without correc - tions to the ensemble. The red lines are without the 10-metre wind correction, the black lines include the correction.

For positive surges there is an increasingly negative zero by shifting the ensembles according to their stan - difference between the ensemble mean and the obser - dard deviation. The Talagrand Diagrams, regenerated vations and for negative surges this difference gets more with this correction, exhibit a very much improved and more positive. This trend points to a bifurcation in shape and the systematic under-prediction has been the probability density function for the larger surges, removed, as shown in Figure 7. positive or negative. A number of ensemble members does catch this surge, but others do not and the standard Skill of the forecasts deviation of the ensemble as a whole increases. The corrected CRHs can be used to calculate calibrated From this relation a simple correction has been probabilities from the ensemble forecasts. To validate constructed to bring the average difference back to these, the Brier Skill Scores for the exceedance of differ - ent levels have been determined. The sample 0.75 climatology has been used as reference. As an example, Figure 8 gives the Brier Skill Scores for the exceedance 0.5 of a range of surge levels for 36 hours starting from day ) m

( 2 and day 5 for IJmuiden for low (black curves) and high

n 0.25

o tides (red) and both with the corrected (solid) and the i t

a uncorrected (dashed) CRH diagrams. v r e

s 0 For surges between -0.5 m and +0.8 m there is skill, b

o even after day 5, when the corrected CRH is used. s u

n Outside this surge range there were only a few cases

i -0.25

m during the period, which makes the results statistically n a e -0.5 insignificant and the Brier Scores vary wildly. m For the results shown in Figure 8 the 10-metre wind e l b correction was applied, but without the wind correction m -0.75 e

s the results are similar. This means that, although the n E Positive surges correction improves the deterministic forecast, it can -1.0 Negative surges be omitted for the generation of the probability fore - casts. Therefore, it is not applied in the real-time -1.25 00.10.20.30.4 ensemble system. Ensemble standard deviation (m) Because the 10-metre wind correction can be omit - ted for the generation of the probability forecasts, Figure 6 Ensemble mean minus observation versus ensemble stan - dard deviation. Red crosses are from positive surges, blue crosses recalibration becomes easier. The ensembles do not from negative surges. have to be rerun once a new 10-metre wind correction

27 METEOROLOGY ECMWF Newsletter No. 114 – Winter 2007/08

a Talagrand Diagram b CRH Diagram 100 14 Without wind correction Without wind correction With wind correction With wind correction

12 80

10 60 8

6 40 O O b b s s

e 4 e r r v v e e 20 d d f f r r e 2 e q q u u e e n n c c y y

( 0 ( 0 % 0 10 20 30 40 50 % 0 20 40 60 80 100 ) ) Bin number EPS probability (%)

Figure 7 (a) Talagrand Diagram and (b) CRH diagram for all high tide forecasts from T+60 to T+96 for Hoek van Holland with correc - tions to the ensemble. The red lines are without the 10-metre wind correction, the black lines include the correction.

a T+48 to T+84 b T+120 to T+156

High tides Calibrated 60 60 Uncalibrated Low tides Calibrated Uncalibrated 40 40

20 20 B B r r i i e e r r S S k k i i l l l l S S c c o o r r 0 0 e e ( ( % % ) )

-100 -50 0 50 100150 -100 -50 0 50 100150 Surge (cm) Surge (cm)

Figure 8 Brier Skill Scores for different forecast intervals for IJmuiden: (a) T+48 to T+84 and (b) T+120 to T+156. Black curves are for low tides, red curves for high tides. The solid curves are for the calibrated, dashed curves for uncalibrated probabilities.

has been established, but a new calibration for the changes in the underlying forecasting models. A major ensemble can simply be derived from the forecasts that change of the deterministic atmospheric model and have already been generated. EPS occurred in February 2006 when the resolution of both was upgraded. The calibrations of both the deter - Recalibration ministic water level forecasts and the water level A drawback of a calibrated forecasting system is that the ensemble were re-established from the data for Sept- calibration should be maintained, especially after ember 2006 to March 2007. A fortunate advantage for

28 ECMWF Newsletter No. 114 – Winter 2007/08 METEOROLOGY

this purpose was that this winter, like the one of 2003/04, which are in use by various institutions around the North contained several significant storm surges. Sea or input for the storm surge model from a HIRLAM Corrections for the 10-metre wind from the deter - ensemble which is under development. ministic model have reduced from 14% extra for the excess of 8 m s–1 to 11% for the excess of 10 m s–1. Also the calibrations for the ensembles have changed, but FURTHER READING the under-prediction is still present and can be corrected de Vries , H. , 2007: Real-time water level probability forecasts. for in the same way as described earlier. http://www.knmi.nl/~jwdv/WAQUA/EPS . Gerritsen , H. , H. de Vries & M. Philippart , 1995: Future developments The Dutch Continental Shelf Model. In Quantitative Skill The medium-range probability forecasts have found an Assessment for Coastal Ocean Models , Eds. D. Lynch and important application in providing guidelines for the A. Davies, Vol. 47, Coastal and Estuarine Studies, planning of activities. For short-range forecasts there is American Geophysical Union. also a desire for uncertainty information, which will help Kok , C.J. , 2001: Calibration of EPS derived probabilities. decision making, e.g. in case of a storm surge for the clos - Scientific Reports WR-2001-04 , KNMI, ISBN 90-369-2201-1. ing of storm surge barriers. The EPS unfortunately does Wilks , D.S. , 2006: Statistical Methods in the Atmospheric Sciences . not help on these time scales, but other possibilities Vol. 91, International Geophysics Series, Academic Press, include the use of an ensemble of storm surge models London, second edition, ISBN 13:978-0-12751966-1. Advances in simulating atmospheric variability with IFS cycle 32r3

The relative activity of a model can be defined as the PETER BECHTOLD, MARTIN KÖHLER, THOMAS JUNG, standard deviation of forecast anomalies divided by the MARTIN LEUTBECHER, MARK RODWELL, FREDERIC VITART standard deviation of observed (i.e. analysed) anom - alies from the ERA-40 climatology. Figure 1 shows the Forecasting the state of the atmosphere involves the relative activity as a function of forecast lead time for fore - prediction of a slowly evolving mean equilibrium state casts with model cycle 32r2 (hereafter Cy32r2) that was (also referred to as the climate) and the temporal and operational between 5 June 2007 and 6 November 2007, spatial variations around this state. Predicting the ampli - and cycle 32r3 (Cy32r3) that became operational on 6 tude and phase of the planetary, synoptic and mesoscale November 2007 and which is the topic of this article. perturbations is not an easy task. This is especially so in Relative activity is shown for 850 hPa temperatures in the the tropics where with decreasing latitude the radius of northern and southern hemispheres (poleward of 20°), influence (also called Rossby radius of deformation) over and for 925 hPa temperatures in the tropics. The 850 which a given perturbation impacts on the background hPa temperature variations in mid-latitudes are a good flow extends to infinity. Consequently the perturba - measure of cold and warm advection associated with tions can actually circle the entire tropical belt. synoptic systems, whereas tropical temperatures varia - Issues related to atmospheric variability as predicted tions at the 925 hPa level are representative of the using ECMWF’s Integrated Forecast System (IFS) partic - low-level convergent/divergent circulations. ularly concern the following. Figure 1 indicates that Cy32r2 slightly underestimates N A decrease in amplitude of tropical synoptic and mid-latitude atmospheric activity, but more importantly planetary-scale activity within the first ten days of the underestimates activity in the tropics. This problem was forecast. also present in all previous cycles. In contrast, in Cy32r3 N A slight underestimation of mid-latitude synoptic model activity is close to the climatology within the first activity in the medium range. ten days in both the mid-latitude and tropical regions. N A lack of growth of the ensemble spread in mid-lati - Indeed the dots show that the activity is significantly tudes in the Ensemble Prediction System (EPS) better compared to Cy32r2 at the 95% confidence level. that is compensated by choosing large initial pertur - The higher, but more realistic level of atmospheric activ - bations. ity has been achieved without compromising classical N An underestimation in the total amount of precipi - deterministic forecast scores. tation and the amplitude of the annual cycle over the In the rest of this article we consider the main features tropical continents. introduced with Cy32r3 and their impact on the model Some of these issues are not specific to the IFS, but climate and deterministic forecasts. We also discuss the are shared by many global atmospheric modelling consequences of the model changes for the EPS and the systems. monthly forecasts.

29 METEOROLOGY ECMWF Newsletter No. 114 – Winter 2007/08

Cy32r2 Cy32r3 its own equilibrium. As a first step the convection scheme 1.05 Northern hemisphere 850 hPa temperature has been rendered quasi-linear which facilitates the optimisation of parameters. The entrainment is set 1 proportional to the relative humidity of the environ - 0.95 ment based on the observations showing that convective cloud development is controlled by mid-tropospheric 1.05 Tropics 925 hPa temperature relative humidity. Furthermore the entrainment is scaled 1 with a vertical function decreasing exponentially with height thereby mimicking an ensemble of clouds. The 0.95

R convective adjustment time is no longer a constant vary - e l a 1.05 Southern hemisphere 850 hPa temperature ing with model resolution (720 s at T799, and 3600 s at t i v

e T159). Instead the adjustment time is a quantity equiv -

a 1 c

t alent to the estimated convective turnover time scale with i v i t 0.95 a weak resolution dependency of a factor of two between y 12345678910model resolutions of T799 and T159. Finally, rain evap - Lead time (days) oration below convective clouds is set to occur with environmental relative humidities of less than 0.9 (0.7) Figure 1 Relative activity in Cy32r2 and Cy32r3 as a function of forecast lead time for 850 hPa temperatures in the northern and over water (land). Different values are used over land southern hemispheres, and tropical temperatures at 925 hPa. The and water in order to account for differences in the relative activity is defined as the ratio between the standard devi - horizontal variability of the water vapour field. ation of forecast anomalies and the standard deviation of anomalies Furthermore it has been realised for some time that from an ERA-40 based climatology. The blue dots denote improve - ments that are significant at the 95% confidence level. for both stably stratified boundary layers and outside the boundary layer the model turbulent mixing, which is intended to reduce static and shear instabilities, is too Features of Cy32r3 strong. The consequence is that stratocumulus layers The two main physical processes that lead to the gener - tend to be eroded, and the vertical shear of the wind ation of kinetic energy in the atmosphere are radiative is reduced. However, it turns out that stronger vertical cooling/heating and convective heating, with the differ - diffusion is beneficial for deterministic forecast scores. ential radiation being the actual driving force of the Therefore, as a pragmatic step, it was decided to inter - general circulation. On sufficiently large space and polate the diffusion coefficients between the former time scales the two processes are in quasi-equilibrium Louis et al. (1982) formulation in the surface layer and above the boundary layer. Inside the boundary layer, it the formulation given by the Monin-Obukhov theory is the turbulent diffusion that regulates the exchanges higher in the free troposphere. of momentum and heat between the surface and the Cy32r3 also contains two other major changes. atmosphere, and the dissipation of kinetic energy. N A change in the soil hydrology scheme including a soil Therefore, from a physical point of view, one would texture map and a new set of hydraulic properties for suspect one of these processes as being responsible for unsaturated soils, as well as a sub-grid surface runoff. the lack of model activity. Recently, the radiation scheme N A new bias correction scheme for radiosonde temper - received a substantial upgrade with a new shortwave ature and humidity data as a function of solar code and the introduction of McICA (Monte Carlo elevation and radiosonde type. Independent Column Approximation) ( Morcrette et al. , These changes have a beneficial impact on soil mois - 2007), so that the focus is now on convection and verti - ture and surface runoff. They also lead to a slightly cal diffusion. However, there exists another source of moister analysis in the free atmosphere that more closely kinetic energy dissipation in the model. This is the fits the observations. numerical diffusion that is inherent to the interpolations In addition Cy32r3 includes changes in the assimila - in the semi-Lagrangian advection scheme, but this tion and observing system. These concern an increase in source of dissipation is not affected by the changes the number of radio occultation data from COSMIC and associated with Cy32r3. use of additional sources of satellite data (see Table 1). Before Cy32r3 the formulation of deep convection Changes to the IFS, particularly to the physical and imposed a strong coupling between the convection and numerical aspects, require thorough testing. It is impos - the large-scale dynamics through an entrainment formu - sible to cover all of them here, but some key diagnostics lation based on the large-scale moisture convergence. of the model climate and the quality and activity of the This, together with an iterative procedure, leads to a forecasts will now be discussed. nonlinear feedback between the large-scale flow and convection that is difficult to control. Therefore, it was Model climate decided to remove any imposed large-scale control of Figure 2 compares the precipitation average from the the convection through the ω field (where ω = d p/d t) Global Precipitation Climatology Project (GPCP, a or the moisture convergence, and to let the model find merger of raingauge data and satellite-derived rain -

30 ECMWF Newsletter No. 114 – Winter 2007/08 METEOROLOGY

Instrument Platform precipitation in the West Pacific. The global average AMSR-E (Advanced Microwave Scanning precipitation rate is reduced from 3.0 to 2.9 mm/day Aqua Radiometer – Earth Observing System (which is more in line with observational estimates). Also the tropical troposphere becomes significantly Defense moister between 850 and 600 hPa (not shown). (SSMIS) Special Sensor Microwave Meteorological Imager/Sounder Satellite Program The rainfall changes in Cy32r3 not only reflect a (DMSP) local effect due to the changes in convection, but also reflect changes in the large-scale circulation and the Tropical Rainfall TMI (TRMM Microwave Instrument) transport of moisture from the subtropics to tropical Measuring Mission regions. Finally, take a closer look at the precipitation SBUV (Solar Backscatter Ultraviolet) NOOA-17/NOAA-18 rates averaged over the Amazon basin and parts of the OMI (Ozone Monitoring Instrument) Aura Andes (Figure 3). It is noticeable that compared to a special set of raingauge data ( Betts et al. , 2005) Cy32r3 also improves the amplitude of the annual cycle of rain - Table 1 Additional sources of satellite information used in Cy32r3. fall which has been rather flat in previous IFS cycles. rates) for the December to March period with the values The tropical activity in the long integrations is effec - obtained from long integrations at resolution T159L91 tively evaluated with so called wavenumber-frequency for Cy31r1, Cy32r2, and Cy32r3. diagrams for either Outgoing Longwave Radiation (OLR), N Cy31r1 has been operational between 12 September 850 hPa winds or 200 hPa velocity potential. Theory ( Gill , 2006 and 5 June 2007; it is also the cycle used in the 1980) says that an equatorial local heat source generates, Interim Reanalysis and the current seasonal fore - amongst others, westward moving Rossby waves with casting system known as System 3. This cycle (and phase speeds around 5 m s–1, and eastward moving Kelvin previous cycles) underestimates precipitation over the waves with phase speeds of 15 –20 m s–1. tropical continents, but overestimates precipitation Figure 4 compares the wavenumber-frequency spec - over the Pacific Inter Tropical Convergence Zone tra of the predicted OLR averaged over the latitude (ITCZ) and the southern Indian Ocean. band 10°S –10°N to spectra of the OLR observed by N Cy32r2 with the revised radiation scheme reduces the NOAA satellites. Notice the relative lack of power in the precipitation biases, especially the convective rainfall easterly modes (Kelvin waves) in Cy31r1 and Cy32r2. over the tropical continents, due to stronger cloud- However, Cy32r3 realistically represents both Rossby radiative feedback and a better description of surface and Kelvin mode activity, but now somewhat overesti - shortwave heating. mates the power in the low-frequency low-wavenumber N With Cy32r3 the overall rainfall distribution is further range. The dominant mode in the OLR spectrum within improved everywhere, including a better land sea the positive wavenumber 1 –2 band and frequency range contrast, apart from an overestimation of convective of 20 –60 days (0.05 –0.016 cycles per day in Figure 4) is a GPCP precipitation b Cy31r1 – GPCP 16 10 12 4 10 2 8 0.5 6 -0.5 4 -2 2 -4 1 -10 c Cy32r2 – GPCP d Cy32r3 – GPCP 10 10 4 4 2 2 0.5 0.5 -0.5 -0.5 -2 -2 -4 -4 -10 -10

Figure 2 (a) Average precipitation rate (mm/day) for December, January and February 1990-2005 from GPCP. (b) Difference between the seasonal type integration at resolution T159 with Cy31r1 and GPCP. (c) and (d) As (b) but for Cy32r2 and Cy32r3.

31 METEOROLOGY ECMWF Newsletter No. 114 – Winter 2007/08

10 a signature of the Madden-Julian Oscillation (MJO). The 9 Cy32r3 Cy32r2 increased wave activity in Cy32r3 also has a beneficial 8 Cy31r1 impact on temperatures and winds in the stratosphere 7 and mesosphere as Rossby and Kelvin waves propagate 6 vertically through the atmosphere. However, the increased wave activity (and associated momentum 5 transfer during wave breaking events) is not sufficient 4 to produce the observed Quasi-Biannual Oscillations 3 (QBO) in the modelled stratospheric zonal winds in the 2 tropics. 1 Deterministic forecasts M 0 o

n Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Cy32r3 has been tested in the full T799 (25 km) reso - t h 2000 2001 l

y lution analysis/forecast cycle for ten months from m

e January to October 2007. Figure 5 summarizes the fore -

Fia gure 3 Annual cycle of monthly mean precipitation rates (mm/day) n cast performance of all available 00 and 12 UTC forecasts fop r Amazonia obtained from a one-year integration of a four-member r e

T1c 59 ensemble with Cy31r1 (black), Cy32r2 (red) and Cy32r3 (blue). using the anomaly correlations (i.e. correlation of fore - i p

Thi e raingauge observations are denoted by the green dots and the t cast anomalies with analysis anomalies) for the 1000 a stt andard deviations of monthly mean precipitation rates for the i o and 500 hPa geopotentials in the northern and south - obn servations and Cy32r3 are denoted by green diamonds and blue r

a ern hemispheres at resolution T63 (300 km) based on

crt osses, respectively. e

( an ERA-40 climate. In the southern hemisphere Cy32r3 m

m improves over Cy32r2 at all lead times. As indicated by / d a y ) a NOAA satellite b Cy31r1 0.5 0.5

0.4 0.4 ) ) y y a a d d r r e e

p 0.3 p 0.3 s s e e l l c c y y c c ( ( y y

c 0.2 c 0.2 n n e e u u q q e e r r F 0.1 F 0.1

0 0 -20 -10 0 10 20 -20 -10 0 10 20 Westward Modes Wavenumber Eastward Modes Westward Modes Wavenumber Eastward Modes c Cy32r2 d Cy32r3 0.5 0.5

0.4 0.4 ) ) y y a a d d r r e e p p 0.3 0.3 s s e e l l c c y y c c ( ( y y

c 0.2 c 0.2 n n e e u u q q e e r r F 0.1 F 0.1

0 0 -20 -10 0 10 20 -20 -10 0 10 20 Westward Modes Wavenumber Eastward Modes Westward Modes Wavenumber Eastward Modes

Figure 4 Power in the frequency (cycles per day) and wavenumber space for the Outgoing Longwave Radiation (OLR) in the 10°S –10°N tropical region from (a) NOAA satellite observations. (b), (c), (d) The corresponding results to (a) but for 15 winter runs (1990 –2005) with Cy31r1, Cy32r2 and Cy32r3.

32 ECMWF Newsletter No. 114 – Winter 2007/08 METEOROLOGY

a 1000 hPa geopotential b 500 hPa geopotential 1.0 1.0 Northern hemisphere Northern hemisphere 0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5 A A n n o o m 0.4 Cy32r2 m 0.4 a a

l Cy32r3 l y y c c

o 0.3 o 0.3 r r

r 12345678910 r 12345678910 e e l l a Lead time (days) a Lead time (days) t t i i o 1.0 o 1.0 n Southern hemisphere n Southern hemisphere 0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5 A A n n o o m 0.4 m 0.4 a a l l y y c c

o 0.3 o 0.3 r r

r 12345678910 r 12345678910 e e l l a Lead time (days) a Lead time (days) t t i i o o Fign ure 5 Average anomaly correlations for Cy32r2 and Cy32r3 as a functn ion of forecast lead time for the northern hemisphere (20°N –90°N, top panel) and southern hemisphere (20°S –90°S, bottom panel) for (a) 1000 hPa geopotential and (b) 500 hPa geopotential for January to October 2007. The blue dots denote improvements that are significant at the 95% confidence level. the dots in Figure 5 these improvements are statisti - tion bias actually corresponds to an underestima - cally significant. In the northern hemisphere the tion of the wind turning (veering) with height, a improvements are smaller and only statistically signifi - problem that is present in many models. cant during the first five days. In spite of higher model The near-surface improvements are a combined activity the deterministic scores are neutral or even effect of the revised shallow convection, the reduction improved. In addition Figure 1 gives an indication of a in turbulent diffusion, and the better description of slight over-activity in Cy32r3 in the northern hemi - the soil moisture. sphere after day 5 which seems to compromise somewhat Since August 2005 the IFS has produced forecast- the scores in the medium range. generated satellite images. These correspond to A comparison with surface synoptic data over Europe brightness temperatures (BTs) computed from the is given in Figure 6 for precipitation, two-metre temper - RTTOV (Radiative Transfer model for TOVS, ATOVS ature, cloud cover and ten-metre wind direction. and Other Vertical sounders) in the 10.8 µm infrared N The precipitation between the two cycles shows a band and the 6.2 µm water vapour band that can be small but significant improvement in the short-range directly compared to observed BTs from Meteosat 8 precipitation forecasts (Figure 6(a)). and Meteosat 9. As the infrared BTs have values that are N The two-metre temperature warm bias over Europe very close to the actual cloud top temperatures, the during summer is reduced in Cy32r3, adding a cool - synthetic images are an excellent tool for indirectly ing of 0.3 K near the surface (Figure 6(b)). This verifying the model clouds (cloud cover and conden - cooling is associated with a 1 –2% increase in cloud sate content) and the synoptic and convective activity cover, particularly due to shallow clouds (Figure 6(c)). associated with cloud systems. N The bias in ten-metre wind direction is also reduced Figure 7(a) shows the infrared satellite image from by a few degrees, and now remains nearly constant Meteosat 9 for 1 July 2007 for large parts of Europe and with forecast lead time (Figure 6(d)). The wind direc - North Africa, with the corresponding six-hour forecast

33 METEOROLOGY ECMWF Newsletter No. 114 – Winter 2007/08

images with Cy32r2 and Cy32r3 given in Figures 7(b) As a quantitative verification, time series from 15 and 7(c). Generally the model gives a good represen - June to 14 July 2007 of spatial correlations and mean tation of the mid-latitude synoptic cloud systems, but has errors between daily 6 –24 hours forecast and observed more difficulty with tropical convection. It seems, infrared BTs are plotted in Figure 8. The errors exhibit however, that Cy32r3 reproduces more realistic tropi - a diurnal cycle with the largest positive biases and lowest cal convection associated with the African Easterly correlations occurring during night. This indicates that Waves between the equator and 5°N (two waves are the forecasts have difficulty in producing long-lived apparent in Figure 7). convective systems that extend into the night. With

a Precipitation b Two-metre temperature 0.7 1

0.6 0.5 0.5

0.4 0

0.3 B i a C s o ( r K r 0.2 e –0) .5 l a t i o 0.1 Cy32r2 Cy32r2 n Cy32r3 Cy32r3 0 –1 12345678910 0 20 40 60 80 100 120 Lead time (days) Lead time (hours) c Cloud cover d Ten-metre wind direction 5 20

0 15

–5 10 B i a B s i ( a % –10 s 5 ) ( d

Cy32r2 e Cy32r2 g r

Cy32r3 e Cy32r3 e s

–15 ) 0 0 20 40 60 80 100 120 0 20 40 60 80 100 120 Lead time (hours) Lead time (hours)

Figure 6 Evaluation of the high-resolution deterministic forecasts with Cy32r2 and Cy32r3 against surface synoptic observations for Europe as a function of forecast lead time: (a) correlation of precipitation, (b) bias of two-metre temperature, (c) bias of cloud cover and (d) bias of ten-metre wind direction. Data is for summer 2007.

a Metosat 9 b Cy32r2 c Cy32r3

45°N 45°N

25°N 25°N

5°N 5°N

0° 30°E 0° 30°E 0° 30°E Figure 7 (a) Observed infrared 10.8 µm brightness temperatures (BTs) from Meteosat 9 for 06 UTC on 1 July 2007. (b), (c) The corresponding RTTOV generated BTs from six-hour forecasts with Cy32r2 and Cy32r3.

34 ECMWF Newsletter No. 114 – Winter 2007/08 METEOROLOGY

a 1 0.9 0.8 0.7 0.6 0.5 0.4

C 0.3 o

r 0.2 r e

l 0.1 Cy32r2 Cy32r3 a t

i 0 o

n 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 1 234567891011121314 June 2007 July 2007 b 6

0 M e a

n -2 Cy32r2 Cy32r3 e r r

o 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 1 234567891011121314 r June 2007 July 2007 ( K ) Figure 8 Time series from 15 June to 14 July 2007 of (a) correlation and (b) mean error of forecast synthetic 10.8 µm BTs obtained with Cy32r2 and Cy32r3 against Meteosat 9 observations. Data points correspond to daily forecasts with lead times 6, 12, 18 and 24 hours that are area-averaged over Central Africa (20°S –20°N, 20°W –30°E); missing data are blanked.

Cy32r2 the average spatial correlation between observed error at all forecast ranges. First, a reduction of the and forecast BTs is about 0.5. Cy32r3 slightly improves initial perturbation amplitude by 20% was tested in a this correlation, and in particular improves the warm sample of 13 summer and 13 winter cases. Results indi - bias in the forecast which becomes smaller than 1 K. The cated that a further reduction of the initial perturbation improvement in Cy32r3 can be linked to the more amplitude was required and it was decided to reduce intense continental convection and larger upper-level the initial perturbation amplitude by 30% in Cy32r3. convective detrainment, producing more and colder The faster perturbation growth due to the convection anvil clouds, and to an improved analysis using the new and diffusion changes, together with the reduced initial radiosonde bias correction scheme. Applying the same perturbation amplitude, results in an improved overall type of verification for Europe (not shown), the two agreement between ensemble spread and ensemble cycles produce nearly identical results, but correlations mean RMS error in Cy32r3 (Figure 9(b)). The EPS no between observed and forecast BTs are much higher longer suffers from an over-dispersion in the early fore - with values around 0.8. cast ranges. The EPS evaluation for Cy32r3 is based on nearly seven weeks of daily ensemble forecasts from The Ensemble Prediction System June, August and September 2007. The improved match The characteristics of perturbation growth in the EPS between ensemble spread and ensemble mean RMS were significantly changed due to the convection and error are statistically significant (see confidence intervals vertical diffusion changes in Cy32r3. In a statistically in Figure 9). consistent ensemble, the RMS error of the ensemble The Ranked Probability Skill Score (RPSS) evaluates mean (computed over a sufficiently large sample) the mean squared error of the predicted probabilities. should match the ensemble standard deviation ( Palmer In terms of this score the probabilistic prediction of et al. , 2006). For Cy32r2, the ensemble spread and the 850 hPa temperature is significantly better in Cy32r3 than ensemble mean RMS error agree well from about day its predecessor in both hemispheres and at all forecast 5 onwards (Figure 9(a)). However, the good agree - ranges (Figure 10(b)). For 500 hPa geopotential, the ment from day 5 can only be achieved by using a large impact of Cy32r3 is neutral in the northern hemisphere initial perturbation amplitude which results in an over- and positive in the southern hemisphere (Figure 10(a)). dispersive ensemble during the earlier forecast ranges. Note, that there is no simple relationship between the Initial experimentation with the new convection and probabilistic skill and the deterministic skill of the diffusion scheme indicated that the physics changes ensemble mean. For instance, while the impact of Cy32r3 results in a significant increase in the ensemble standard on the RPSS of 500 hPa geopotential in the northern deviation. This offered the opportunity to reduce the hemisphere is neutral, Cy32r3 has a positive impact on initial perturbation amplitude in order to improve the the RMS error of the ensemble mean (again with more agreement between spread and ensemble mean RMS than 99% statistical significance). By the same token, the

35 METEOROLOGY ECMWF Newsletter No. 114 – Winter 2007/08

a Cy32r2 b Cy32r3 700 Northern hemisphere 700 Northern hemisphere ) )

2 600 600 – – s s 2 2 2

m 500 500 ( d a

e 400 400 r p s

d 300 300 n a E

S 200 200 M R R 100 Spread M 100 Spread S RMSE E RMSE a 0 n 0 d

042 6 8 10 12 14 s 042 6 8 10 12 14 p

Lead time (days) r Lead time (days) e a

1200 d 1200 ( Southern hemisphere m Southern hemisphere 1000 1000 ) ) – – s s 2 2 2 2 800 800

600 600

400 400 R R

M 200 Spread M 200 Spread S S E RMSE E RMSE a a n 0 n 0 d d

s 042 6 8 10 12 14 s 042 6 8 10 12 14 p p

r Lead time (days) r Lead time (days) e e a a d d ( ( m Figure 9 Ensemble standard deviation (spread) and ensemble mean Rm MS error (RMSE) as a function of forecast lead time for 500 hPa geopotential for the northern hemisphere (20°N –90°N, top panel) and southern hemisphere (20°S –90°S, bottom panel) for (a) Cy32r2 and (b) Cy32r3. Statistics are based on 69 cases during June to September 2007. The vertical bars are confidence intervals based on boot - strapping the dates in the sample of cases. If the spread falls within the bars, the ensemble is not significantly over- or under-dispersive (the probability of the spread being above or below the bar by chance is 1%). impact on the RMS error of the ensemble mean is To assess the skill of the monthly forecasting system neutral for 850 hPa temperature in the northern hemi - to predict an MJO event, 32-day coupled ocean-atmos - sphere while a statistically significant positive impact phere integrations using a five-member ensemble have was noted for the RPSS. been performed for each day between 15 December In addition to the changes already mentioned, the 1992 and 31 January 1993 (46 ensemble integrations); tangent-linear and adjoint model in the singular vector this period corresponds to the Intense Observing Period computation targeted on tropical cyclones use the new (IOP) of TOGA-COARE. The MJO is diagnosed in these moist physics package in Cy32r3. The new package has integrations using a method based on the technique of been used operationally in 4D-Var since Cy32r2. Prelim- Wheeler & Hendon (2004). Combined Empirical Ortho- inary experimentation over 27 summer cases with the new gonal Functions (EOFs) of OLR, 200 hPa velocity moist physics package in the EPS singular vector compu - potential and 850 zonal wind averaged over 10°N –10°S tations indicates an improved reliability of forecasts of are calculated using ECMWF operational data between tropical cyclone strike probability and neutral impact 2002 and 2004. The EOF analysis is performed on the on ensemble spread and skill scores in the extra-tropics. anomalies relative to the seasonally evolving climatology. The first two EOFs, which represent 18% and 17% of the Monthly forecasts and the Madden-Julian variance respectively, describe variations associated with Oscillation the MJO. Observations and forecasts are then projected The Madden-Julian Oscillation (MJO) is the dominant onto these two EOFs. Different cycles from Cy28r3 mode of intraseasonal predictability in the tropical onward (operational implementation 29 April 2004) atmosphere. Therefore, it is important that the monthly have been run for the control period in order to assess forecasting system accurately predicts its onset and the impact of changes in the physics on the skill of the evolution. monthly forecast system to predict an MJO event.

36 ECMWF Newsletter No. 114 – Winter 2007/08 METEOROLOGY

a 500 hPa geopotential b 850 hPa temperature 1 1 0.9 Northern hemisphere 0.9 Northern hemisphere 0.8 0.8 0.7 0.7 0.6 0.6 0.5 0.5 R R P P

S 0.4 S 0.4 S S 0.3 0.3 0.2 0.2 Cy32r2 0.1 Cy32r3 0.1 0 0 042 6 8 10 12 14 042 6 8 10 12 14 Lead time (days) Lead time (days) 1 1 0.9 Southern hemisphere 0.9 Southern hemisphere 0.8 0.8 0.7 0.7 0.6 0.6 0.5 0.5 R R P P

S 0.4 S 0.4 S S 0.3 0.3 0.2 0.2 0.1 0.1 0 0 042 6 8 10 12 14 042 6 8 10 12 14 Lead time (days) Lead time (days)

Figure 10 Ranked Probability Skill Score (RPSS) for Cy32r2 and Cy32r3 as a function of forecast lead time for the northern hemisphere (20°N –90°N, top panel) and southern hemisphere (20°S –90°S, bottom panel) for (a) 500 hPa geopotential and (b) 850 hPa temperature. The RPSS is determined for ten climatologically equally likely categories. Cy32r3 is significantly better (worse) than Cy32r2 at the 99% level if the red curve lies below (above) the vertical bars.

Figure 11 shows the amplitude of Principal Compo- are similar for velocity potential at 200 hPa and zonal nent 1 (PC1) averaged over the 46 integrations and the wind at 850 hPa), and with Cy32r3 the OLR anomalies five members of the ensemble as a function of the fore - by day 15 are as intense as in ERA-40. However, the cast time. Results are similar for Principal Component propagation of the MJO with Cy32r3 is slower than in 2 (PC2). Before Cy32r3, all versions of the IFS shared the analysis. In particular, the model has some difficulty the same problem, namely a rapid drop in the amplitude propagating the convection from the Indian Ocean to of PC1 and PC2; the drop in the amplitude of the MJO the Pacific, and instead tends to maintain the convec - can be seen in Figure 11. For instance, with Cy28r3 the tion over the Indian Ocean for too long. This deficiency MJO loses 25% of its amplitude after only five days of may be related to an overestimation of convective precip - integrations, about 33% by day 10, and about 50% by itation over the maritime continent. day 20. This means that the impact of the MJO on the extra-tropics was likely to be strongly underestimated in Concluding remarks those cycles. On the other hand Figure 11 shows that Apart from overall satisfactory forecast scores, the most Cy32r3 is the first IFS cycle able to sustain the amplitude positive outcome of implementing Cy32r3 is that it is over the whole period of the integration. This is an possible to reasonably represent the atmospheric vari - important result since it implies that, for the first time, ability on a wide variety of time and space scales using the monthly forecasting system may now be able to a conventional set of physical parametrizations, without adequately represent the impact of the MJO on the necessarily having to specifically use an explicit global extra-tropics. representation of convection (as has been discussed in Figure 12 illustrates the progress made in the repre - recent forums). This might come as a surprise, but sentation of the MJO with the IFS. In Cy28r3 the OLR considering that the modes discussed here are also eigen - anomalies are very weak by day 15. Each new cycle modes of the (dry) atmosphere ( Wedi & Smolarkiewicz , increases the amplitude of the OLR anomalies (results 2007), it seems reasonable to assume that it should be

37 METEOROLOGY ECMWF Newsletter No. 114 – Winter 2007/08

1.6

1.4

1.2

0.8 A

m 0.6

p Cy32r3 Figure 11 Amplitude of the first combined l i

t EOF of Outgoing Longwave Radiation (OLR), u Cy32r2 d 0.4 200 hPa velocity potential and 850 hPa zonal e Cy31r1 wind as a function of forecast lead time from 0.2 Cy29r1 Cy28r3 to Cy32r3. The amplitude constitutes Cy28r3 an average over each ensemble member and the 46 cases. 0 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 Day ERA-40 Cy28r3 Cy29r1 Cy31r1 Cy32r2 Cy32r3 29/12

05/01

12/01

20/01 D a y s 28/01

04/04

12/02 090E18090W0090E18090W0090E18090W0090E18090W0090E18090W0090E18090W0 Longitude Longitude Longitude Longitude Longitude Longitude

Figure 12 Hovmöller diagrams of the averaged Outgoing Longwave Radiation (OLR) between 10°S and 10°N from 29 December 1992 to 15 February 1993 as analysed by ERA-40 and obtained from forecasts with Cy28r3 to Cy32r3 using concatenated daily forecasts with a 15 day lead time. Red shading denotes warm OLR anomalies (negative phase of MJO), and blue shading cold anomalies (convectively active phase of MJO). possible to develop a set of physical parametrizations that medium-range forecasts (in particular the humidity) in in some optimal way supports these modes. Further- the tropics is expected from this development. more, the results confirm that a good representation of variability comes with a good representation of the FURTHER READING mean state. Betts , A.K. , J.H. Ball , P. Viterbo , A. Dai & J.A. Marengo , If anything, the level of model activity in Cy32r3 is 2005: Hydrometeorology of the Amazon in ERA-40. slightly overestimated. No attempt was made here to J. Hydrometeorol. , 6, 764 –774. Also ERA-40 Project Report 22 . break this down into convection and diffusion compo - Gill , A.E. , 1980: Some simple solutions for heat-induced nents, but it seems to be related to some overestimation tropical circulations. Q. J. R. Meteorol. Soc. , 106 , 447 –462. of mid-tropospheric moisture (in contrast to a too dry Morcrette , J.-J. , H.W. Barker , M.J. Iacono , G. Mozdzynski , mid-troposphere in previous cycles), and possibly to an R. Pincus , D. Salmond & S. Serrar , 2007. A new radiation increase in wind shear near the trade wind inversion. package McRAD. ECMWF Newsletter No. 112 , 22 –32. However, the IFS is still unable to properly represent the Palmer , T. , R. Buizza , R. Hagedorn , A. Lawrence , stratospheric wind variability associated with the QBO, M. Leutbecher & L. Smith , 2006: Ensemble prediction: but current developments indicate that this can be A pedagogical perspective. ECMWF Newsletter No. 106 , 10 –17. achieved with a scheme that parametrizes the non- Wedi , N.P. & P.K. Smolarkiewicz , 2007. A reduced model of orographic gravity wave drag. Finally, there is the Madden-Julian oscillation. Int. J. Numer. Meth. Fluids , development of an upgrade of the simplified linear set published online DOI:10.1002/fld.1612. of physical parametrizations for the assimilation that Wheeler , M.C. & H.H. Hendon , 2004: An all-season real- closely matches the results of the non-linear schemes in time multivariate MJO index: Development of an index for Cy32r3. Some further impact on the analysis and the monitoring and prediction. Mon. Wea. Rev. , 132 , 1917 –1932.

38 ECMWF Newsletter No. 114 – Winter 2007/08 GENERAL

Special Project computer allocations for 2008 –2010 The allocations for 2008 have been approved. The figures for 2009 and 2010 indicate what has been requested.

2008 2009 2010 Member Institution Project title State HPCF Data HPCF Data HPCF Data units storage units storage units storage

Univ. Vienna Homogenization of the global radio- 1 5,000 500 X X X X (Haimberger) sonde temperature and wind dataset

Univ. Vienna Convective fluxes diagnosed from 2 1,000 100 X X X X (Hantel) gridscale ECMWF analyses Climate monitoring by advanced Univ. Graz 3 spaceborne sounding and 25,000 400 45,000 450 50,000 500 (Kirchengast) atmospheric modelling Univ. of Natural Austria Resources and 4 Applied Life Modelling of Tracer Transport (MoTT) 30,000 100 30,000 100 30,000 100 Sciences, Vienna (Kromp-Kolb)

Univ. Vienna MESOCLIM – 5 50 10 100 10 100 10 (Steinacker) Mesoscale Alpine Climatology 4D OMEGA FORM – 4 dimensional Univ. Vienna 6 objective mesogamma analysis of 100 10 0 0 0 0 (Steinacker) foehn in the Rhine Valley during MAP Denmark 7 DMI (Amstrup) Data impact studies in HIRLAM 204,000 4,890 900,000 6,000 X X Investigation of coupling the ALADIN CNRM/GMAP, and AROME models to boundary 8 Météo-France 30,000 800 30,000 800 30,000 800 conditions from ECMWF and ERA (Fischer) model data France CERFACS Seasonal to interannual predictability 9 10,000 150 10,000 150 10,000 150 (Rogel) of a coupled ocean-atmosphere model

CERFACS Variational data assimilation with 10 136,000 2,000 150,000 2,000 150,000 2,000 (Weaver) the OPA OGCM

MPI, Hamburg Numerical experimentation with a 11 105,000 700 300,000 800 300,000 900 (Bengtsson) coupled ocean/atmosphere model MPI, Hamburg Regional downscaling of 12 (Bengtsson, ERA-40 data and validation of 157,000 3,200 600,000 4,600 680,000 5,400 Hagemann) the hydrological cycle MPI, Hamburg Global atmospheric chemistry 13 290,000 6,720 500,000 9,000 500,000 10,000 (Budich) modelling Investigation of systematic tendency Freie Univ. Berlin changes and their influence on the 14 10,000 1,000 20,000 1,500 20,000 2,000 (Cubasch, Kirchner) general circulation simulated with climate models Germany ISET Evaluation of the global potential of 15 100 20 100 20 100 20 (Czisch) energy towers Influence of non-hydrostatic gravity DLR 16 waves on the stratospheric flow 136,000 80 150,000 80 150,000 80 (Doernbrack) field above Scandinavia 17 Univ. Munich (Egger) Landsurface-atmosphere interaction 150 10 150 10 150 10

18 Univ. Cologne (Elbern) GEMS: work package WP_RAQ_2 96,000 750 100,000 750 100,000 750 DLR & MPI Impact of anthropogenic emissions 19 Chemistry, Mainz on tropospheric chemistry with a 173,000 2,110 400,000 4,000 400,000 4,000 (Eyring, Steil) special focus on ship emissions

39 GENERAL ECMWF Newsletter No. 114 – Winter 2007/08

2008 2009 2010 Member Institution Project title State HPCF Data HPCF Data HPCF Data units storage units storage units storage

MPI, Hamburg Climate impact of specific economic 20 100,000 2,220 220,000 7,000 150,000 5,000 (Feichter) sectors

21 DLR (Gierens) Ice-supersaturation and cirrus clouds 170,000 100 200,000 100 200,000 100

22 DLR (Hoinka) Climatology of the global tropopause 500 10 500 10 500 10 MPI, Hamburg 23 Regional ensemble prediction 43,000 2,190 84,000 6,500 92,000 7,500 (Jacob)

Univ. Karlsruhe The impact of tropical cyclones on 24 232,000 450 300,000 450 300,000 450 (Jones) extratropical predictability

MPI, Hamburg Community simulations of the last 25 386,000 2,000 2,000,000 2,000 2,000,000 2,000 (Jungclaus) millennium (COMSIMM) Ensemble modelling for the improve - DLR 26 ment of short range quantitative 96,000 100 100,000 100 100,000 100 (Keil, Craig) precipitation forecasts Univ. Karlsruhe Mesoscale modelling using the DWD 27 10,000 50 10,000 50 10,000 50 (Kottmeier) Lokal-Modell Leibniz-Institut Seasonal to decadal forecasting 28 Univ. Kiel with coupled ocean-atmosphere 764,000 7,900 1,500,000 8,000 X X (Latif) general circulation models Statistical analysis of the onset of IMK-IFU 29 the rainy season in the Volta Basin 10,000 30 10,000 30 10,000 30 (Laux) Germany (West Africa) Remote sensing of water and DLR 30 Ice clouds with Meteosat Second 30,000 20 50,000 20 50,000 20 (Mayer) Generation

Ruhr-University Optimisation of water management 31 2,000 3 2,000 3 2,000 3 Bochum (Pahlow) by using ensemble forecasts

Alfred Wegener Ozone and water vapour transport 32 Institute, Potsdam 200 200 200 200 200 200 with the residual circulation (Rex) Alfred Wegener 33 Institute, Potsdam Sensitivity of HIRHAM 100 10 100 10 100 10 (Rinke) Univ. Koln Interpretation and calculation of 34 100 10 120 15 120 15 (Speth) energy budgets Chemical and dynamical Univ. Bremen 35 influences on decadal ozone change 20 20 20 20 20 20 (Weber) (CANDIDOZ) Univ. Mainz Water vapour in the upper 36 1,000 20 1,000 20 1,000 20 (Wirth) troposphere Real-time assimilation of Univ. Hohenheim observations of key prognostic 37 334,000 2,000 1,000,000 2,000 1,000,000 2,000 (Wulfmeyer, Bauer) variables and the development of aerosol operators (RAPTOR) Met Éireann Community Climate Change Ireland 38 20,000 2,000 30,000 2,000 30,000 3,000 (Mc Grath) Consortium for Ireland (C4I)

CNMCA 39 Limited area ensemble Kalman Filter 392,000 500 1,400,000 500 1,800,000 500 (Bonavita, Torrisi) Italy Evaluation of the performance of ISMAR-CNR 40 the ECMWF meteorological model at 157,000 150 221,500 150 221,500 150 (Cavaleri) high resolution

40 ECMWF Newsletter No. 114 – Winter 2007/08 GENERAL

2008 2009 2010 Member Institution Project title State HPCF Data HPCF Data HPCF Data units storage units storage units storage ARPA-SIM Flow dependent error statistic for 41 (Di Giuseppe, satellite data assimilation in 309,000 150 1,000,000 150 1,000,000 150 Marsigli) regional model (FEAR) INGV, Bologna 42 Middle atmosphere modelling 245,000 2,000 350,000 2,300 40,000 2,700 (Manzini) Osservatorio Forecasting of the optical turbulence Astrofisico di for Astronomy applications with the 43 4,000 30 4,000 30 4,000 30 Arcetri, Firenze MesoNH mesoscale model coupled (Masciadri) with ECMWF products 44 ISAC-CNR (Maurizi) GEMS: BOLCHEM 20,000 120 X X X X ARPA-SMR Emilia Romagna Limited-area ensemble forecasts of 45 180,000 100 620,000 120 640,000 140 & UK Met Office windstorms over Northern Europe Italy (Montani, Mylne) ARPA-SMR Emilia Romagna & Improvements of COSMO 46 148,000 700 420,000 720 440,000 740 MeteoSwiss limited-area ensemble forecasts (Montani, Walser) ARPA-SMR Emilia Romagna & Limited area model targeted 47 Italian Met. Service ensemble prediction system 176,000 280 700,000 290 800,000 300 (Paccagnella, (LAM-TEPS) Montani, Ferri) Univ. Genova High resolution numerical modelling 48 30,000 200 50,000 200 50,000 200 (Parodi) of intense convective rain cells ARPA-SMR Emilia Seasonal prediction for Italian 49 Romagna & UCEA 10 100 10 100 10 100 agriculture (SPIA) (Pavan, Esposito) Patterns of climate change: 50 KNMI (Hazeleger) 170,000 500 400,000 500 400,000 500 coupled modelling activities

51 KNMI (Onvlee) The Hirlam-A project 483,000 6,580 1,250,000 8,500 1,500,000 8,500

52 KNMI (Siebesma) Rain in cumulus 170,000 50 250,000 75 300,000 75 KNMI Climate change studies using 53 185,000 500 225,000 500 X X (Selten) the IFS system Netherlands Multi-annual integrations with KNMI 54 the KNMI regional climate model 187,000 2,360 500,000 2,500 X X (van Meijgaard) RACMO2 Chemical reanalyses and sensitivity KNMI 55 studies with the chemistry-trans - 30,000 100 X X X X (van Velthoven) port model TM4 Global chemistry-transport KNMI 56 modelling of natural reactive 96,000 100 100,000 100 X X (van Weele) greenhouse gases DNMI NORLAMEPS: Limited Area Ensemble 57 190,000 500 500,000 500 500,000 500 (Frogner) Prediction System for Norway

58 Univ. Oslo (Isaksen) Ozone as a climate gas 30,000 5 50,000 5 50,000 5 DNMI GLAMEPS – Grand Limited Area 59 478,000 5,560 900,000 6,000 900,000 6,000 Norway (Iversen) Model Ensemble Prediction System DNMI REGCLIM: optimal forcing 60 232,000 1,000 300,000 1,000 300,000 1,000 (Iversen, Kristiansen) perturbations for the atmosphere

DNMI Optimisation of operational NWP 61 140,000 1,000 X X X X (Tveter) at met.no

41 GENERAL ECMWF Newsletter No. 114 – Winter 2007/08

2008 2009 2010 Member Institution Project title State HPCF Data HPCF Data HPCF Data units storage units storage units storage

Portugal 62 Univ. Lisbon (Soares) HIPOCAS-SPEC 0 10 0 10 0 10

Study of the stably stratified Univ. Illes Balears atmospheric boundary layer through 63 96,000 200 100,000 200 100,000 200 (Cuxart) large-eddy simulations and high resolution mesoscale modelling

Spain Univ. de Analysis of land surface-atmosphere 64 Castilla-La Mancha interactions through mesoscale 246,000 1,000 700,000 800 700,000 800 (Gaertner) simulations

Univ. Basque Mesoscale meteorological 65 136,000 1,000 X X X X Country (Saenz) reanalysis over the Iberian Peninsula

Institute for Atmos- pheric and Climate Switzerland 66 Cloud aerosol interactions 201,000 200 250,000 200 250,000 200 Science, ETH Zurich (Lohmann)

Predictability studies with emphasis ESSC, Univ. Reading on extra-tropical and tropical storm- 67 196,000 200 300,000 200 300,000 200 (Bengtsson) tracks and their dependence on the global observing systems

Using data assimilation in a high- Univ. Reading 68 resolution ocean model to determine 207,000 4,670 700,000 7,000 X X (Haines) the termohaline circulation

Attribution of changes in extreme Univ. Oxford 69 weather risk using large ensembles of 17,000 150 25,000 150 25,000 150 (Hanlon) climate model simulations

Univ. Reading Moist Singular Vectors and African 70 75,000 150 75,000 150 75,000 150 (Hoskins) Easterly Waves

Manchester Determining the relative roles of NO x 71 Metropolitan Univ. and CO 2 emissions from 45,000 700 X X X X (Lee) aviation in climate change United DARC, Univ. Kingdom 72 GlobMODEL 129,000 500 X X X X Reading (Migliorini)

Assimilation of geostationary ozone DARC, Univ. 73 measurements for global ozone 136,000 1,000 X X X X Reading (Migliorini) monitoring

DARC, Univ. Assimilation of retrieved products 74 426,000 2,940 900,000 3,000 900,000 3,000 Reading (O’Neill) from EOS MLS

DARC, Univ. How good are simulated water vapour 75 128,000 500 70,000 250 X X Reading (O’Neill) distributions in the UTLS region?

Keele University Direct numerical simulations of 2-D 76 55,000 100 100,000 100 X X (Shrira) freak waves

Assessment of ECMWF forecasts over BAS, Cambridge 77 the high latitude areas of the Southern 0 1 0 1 0 1 (Turner) Hemisphere

42 ECMWF Newsletter No. 114 – Winter 2007/08 GENERAL

2008 2009 2010 Member Institution Project title State HPCF Data HPCF Data HPCF Data units storage units storage units storage

Dynamical downscaling of ICTP 78 seasonal predictions with a 30,000 500 50,000 500 50,000 500 (Kucharski) regional climate model ICTP Decadal interactions between the ICTP tropical Indo-Pacific Ocean and 79 30,000 300 50,000 300 50,000 300 (Kucharski) extratropical modes of variability in an intermediate coupled model The linkage of climate and air JRC-IES JRC 80 pollution: simulations with the 55,000 140 110,000 160 120,000 180 (Dentener) global 2-way nested model TM5 New Projects The TIGGE data base: Univ. Innsbruck Austria 1 atmospheric predictability and 9,000 30 9,000 30 9,000 30 (Ehrendorfer) Bayesian decision making Data assimilation in high resolu - MUMM Belgium 2 tion hydrodynamic and ecological 100,000 700 200,000 700 200,000 700 (Ponsar) forecasts of the North Sea Univ. Frankfurt Combination of seasonal 3 (Casanova, 500 200 500 200 500 200 forecasts by BMA Ahrens)

DLR 4 Support tool for HALO missions 15,000 80 20,000 80 20,000 80 (Doernbrack) Germany Investigations of storms in Freie Univ. Berlin forecasts, hindcasts and climate 5 (Ulbrich, model simulations on daily to 5,000 1,000 5,000 1,500 5,000 2,000 Leckebusch) seasonal and climatological timescales

Met Éireann Changes in the North Atlantic Ireland 6 20,000 1,000 30,000 1,500 30,000 2,000 (Mc Grath) climate and impacts for Ireland

KNMI Data assimilation over the North 7 65,000 1,000 80,000 10,000 110,000 1,000 (de Vries) Atlantic (DANA)

8 KNMI (Haarsma) Storm tracks in a warmer climate 80,000 500 150,000 500 150,000 500

KNMI Netherlands 9 Participation in GLACE-2 136,000 580 150,000 580 150,000 580 (van den Hurk)

Regional modelling of the KNMI Greenland surface mass balance 10 187,000 1,500 500,000 1,500 500,000 1,500 (van Meijgaard) for key episods in the past and the future Seasonal predictability over the DNMI Norway 11 Artic Region – exploring the role 84,000 1,000 160,000 1,000 215,000 1,000 (Benestad) of boundary conditions

SMHI Sweden 12 GEMS: MATCH 6,000 1 8,000 4 8,000 4 (Robertson)

Coupling a regional climate model to a biogeochemical land-surface JRC-IES JRC 13 model in the study of climate 100,000 100 200,000 100 200,000 100 (Dosio) change impacts on the European ecosystem

Total allocated 10,674,830 87,420 22,926,300 113,753 19,479,300 84,223

43 GENERAL ECMWF Newsletter No. 114 – Winter 2007/08

Member State computer allocations for 2008

Data Storage Data Storage Member State HPCF (kunits) Member State HPCF (kunits) (Gbytes) (Gbytes)

Belgium 4,466 36,657 Austria 4,075 33,451

Denmark 3,775 30,986 Portugal 3,337 27,396

Germany 19,167 157,338 Switzerland 4,747 38,966

Spain 8,067 66,218 Finland 3,453 28,345

France 14,682 120,523 Sweden 4,377 35,934

Greece 3,479 28,560 Turkey 3,853 31,632

Ireland 3,186 26,157 United Kingdom 15,426 126,635

Allocated to Italy 12,280 100,810 10,675 87,420 Special Projects Luxembourg 2,511 20,608 Reserved for 2,725 22,580 Netherlands 5,820 47,777 Special Projects

Norway 3,899 32,007 Total 134,000 1,100,000

Representatives and Contact Points

THERE are a variety of Representatives and Contact ing system, including the high-resolution model, EPS, Points within ECMWF’s Member States and Co-oper - seasonal forecasts and optional projects. They are encour - ating States who liaise with staff at ECMWF. The role of aged to provide feedback concerning the performance these Representatives and Contact Points is given below. of the forecasting system to the Centre. In addition they Note that: may refer to the Head of Meteorological Operations N The purpose of the Technical Advisory Committee Section or the Meteorological Analyst at ECMWF if they (TAC) is covered on page 46 in the item about wish to discuss any aspect of the daily model output. “ECMWF Council and its committees”. Security Representatives N A list of TAC Representatives, Computing Represent- atives and Meteorological Contact Points is given in Security Representatives represent their organisation in the table on the next page. matters relating to computer and network security, and receive information about ECMWF’s security arrange - Computing Representatives ments. They liaise with the Security Officer at ECMWF. Computing Representatives co-ordinate the registration Meetings of the Security Representatives are held at of users of ECMWF computing services, and represent ECMWF annually. their organisation in matters relating to the use of ECMWF computing facilities. They play a very important role in Telecommunication Contacts improving the information flow and facilitating various Telecommunication Contacts deal with day-to-day administrative transactions between ECMWF and coun - matters concerning the Regional Meteorological Data tries that have access to ECMWF’s computing services. Communication Network (RMDCN). They liaise with They liaise with the Head of Computer Division and User the Head of the Networking and Computer Security Support at ECMWF. Meetings of the Computing Section and Computer Operators at ECMWF. Representatives are held at ECMWF every 12 to 18 months. Catalogue Contact Points Meteorological Contact Points Catalogue Contact Points are the primary contact for Meteorological Contact Points receive information about external organisations wishing to receive real-time the meteorological aspects of the operational forecast - ECMWF data via one of the ECMWF Member States.

44 ECMWF Newsletter No. 114 – Winter 2007/08 GENERAL

TAC Representatives, Computing Representatives and Meteorological Contact Points

Member States TAC Representatives Computing Representatives Meteorological Contact Points Belgium Dr D. Gellens Mrs L. Frappez Dr J. Nemeghaire Denmark Mr L. Laursen Mr T. Lorenzen Mr G. Larsen Germany Prof G.-R. Hoffmann Dr E. Krenzien Mr T. Schumann Mr D. Ziakopoulos Mr D. Kapniaris Greece Mr A. Emmanouil Mr M. Manoussakis Mr T. Andreadis Mr P. Fragkouli Spain Mr P. del Rio Mr E. Monreal Mr A. Alcazar France Mr B. Strauss Mrs M. Pithon Mr J. Clochard Ireland Mr J. Logue Mr P. Halton Mr M. Walsh Italy Dr S. Pasquini Dr C. Gambuzza Dr T. La Rocca Luxembourg Mr C. Alesch Mr C. Alesch Mr C. Alesch Netherlands Mr T. Moene Mr H. de Vries Mr J. Diepeveen Norway Mr J. Sunde Ms R. Rudsar Mr P. Evensen Austria Dr G. Kaindl Dr G. Kaindl Dr H. Gmoser Portugal Mrs T. Abrantes Mrs M. da C. Periera Santos Mrs I. Soares Switzerland Dr S. Sandmeier Mr P. Roth Mr R. Mühlebach Finland Mrs K. Soini Mr K. Niemelä Mr P. Nurmi Sweden Mr I. Karro Mr R. Urrutia Mr M. Hellgren Turkey Mr M. Fatih Büyükkasabba ¸si Mr F. Kocaman Mr M. Kayhan United Kingdom Dr A. Dickinson Mr R. Sharp Mr A. Radford Co-operating States Croatia Mr I. Cˇacˇi ć Mr V. Malovi ć Mr Cˇ. Brankovi ć Czech Republic Mr M. Janoušek Mr K. Ostatnický Mr F. Sopko Estonia Mr T. Kaldma Mr T. Kaldma Mrs M. Merilain Mrs T. Paljak Hungary Dr Z. Dunkel Mr I. Ihász Mr I. Ihász Iceland Mr H. Björnsson Mr V. Gislason Mrs S. Karlsdottir Lithuania To be decided Mr P. Jalinskas Mrs. V. Raliene Montenegro To be decided To be decided To be decided Morocco Mr H. Haddouch Mr M. Jidane Mr K. Lahlal Romania Dr I. Pescaru Mr R. Cotariu Mrs T. Cumpanasu Serbia Ms L. Dekic Mr V. Dimitrijevi ć Mr B. Bijelic Slovakia To be decided To be decided To be decided Slovenia Mr J. Jerman Mr P. Hitij Mr B. Gregor ˇci ˇc Observers EUMETSAT Mr M. Rattenborg Dr K. Holmlund WMO Mr M. Jarraud

45 GENERAL ECMWF Newsletter No. 114 – Winter 2007/08

ECMWF Council and its committees

The following provides some information about the Scientific Advisory Committee (SAC) responsibilities of the ECMWF Council and its commit - tees. More detail can be found at: The SAC provides the Council with opinions and recom - http://www.ecmwf.int/about/committees mendations on the draft programme of activities of the Centre drawn up by the Director and on any other matters submitted to it by the Council. The 12 members of the Council SAC are appointed in their personal capacity and are The Council adopts measures to implement the ECMWF selected from among the scientists of the Member States. Convention; the responsibilities include admission of new members, authorising the Director to negotiate and conclude co-operation agreements, and adopting Chair : Prof Gerhard Adrian ( Deutscher Wetterdienst ) the annual budget, the scale of financial contributions of the Member States, the Financial Regulations and the Vice Chair : Prof Martin Ehrendorfer (Univer sity of Reading) ) Staff Regulations, the long-term strategy and the programme of activities of the Centre. Technical Advisory Committee (TAC) The TAC provides the Council with advice on the tech - nical and operational aspects of the Centre including the President : Dr Adérito Vicente Serrão ( Portugal ) communications network, computer system, operational Vice President : Mr Wolfgang Kusch ( Germany ) activities directly affecting Member States, and techni - cal aspects of the four-year programme of activities. Policy Advisory Committee (PAC) The PAC provides the Council with opinions and recom - mendations on any matters concerning ECMWF policy Chair : Dr Alan Dickinson ( United Kingdom ) submitted to it by the Council, especially those arising out of the Four-Year Programme of Activities and the Vice Chair : Mr Bernard Strauss ( France ) Long-term Strategy. Advisory Committee for Data Policy (ACDP) The ACDP provides the Council with opinions and recommendations on matters concerning ECMWF Data Chair : Dr Fritz Neuwirth ( Austria ) Policy and its implementation. Vice Chair : Ms Maria Agren ( Sweden )

Finance Committee (FC) Chair : Ms Lillian Wester-Andersen ( Denmark ) The FC provides the Council with opinions and recom - mendations on all financial matters submitted to the Vice Chair : Mr Colin Cuthbert ( United Kingdom ) Council and shall exercise the financial powers dele - gated to it by the Council. Advisory Committee of Co-operating States (ACCS) The ACCS provides the Council with opinions and recom - mendations on the programme of activities of the Centre, and on any matter submitted to it by the Council. Chair : Ms Monika Köhler ( Austria ) Vice Chair : To be elected

Chair : Mr Ivan Cˇacˇi c´ ( Croatia ) Vice Chair : Dr Ion Sandu ( Romania )

46 ECMWF Newsletter No. 114 – Winter 2007/08 GENERAL

ECMWF Calendar 2008

SRNWP Interoperability workshop Training Course – Jan 14–15 Jun 2–6 (organised by the Met Office, UK) Use and interpretation of ECMWF products Feb 11–Mar 7 Computer User Training Course Jun 9–10 Council ( 69 th Session ) Feb 11–12 Introduction to SMS/XCDP Jun 11–13 Forecast Products – Users’ Meeting Feb 13–14 GRIB API: library and tools Meeting of the Pensions Administrative Jun 12–13 Feb 25–29 Introduction for new users/MARS Committee of the Co-ordinated Organisations ECMWF-EUMETSAT/GRAS SAF Workshop on Mar 3–4 MAGICS Jun 16–18 “Use of GPS Radio-occultation” Mar 5–7 METVIEW Seminar on “Physical Parametrization Meeting of the Council of the European Sep 1–5 Mar 6–7 Processes and their Dynamical Interactions” Meteorological Society (EMS) Trilateral Meeting of the Co-ordinating Training Course – Sep 16–17 Mar 10–14 Committee on Remuneration Use and interpretation of ECMWF products Training Course – Mar 31–May 27 Training Course – Numerical Weather Prediction Sep 22–26 Use of supercomputing resources Numerical methods and adiabatic formulation Mar 31–Apr 8 Oct 6–8 Scientific Advisory Committee ( 37 th Session ) of models Oct 8–10 Technical Advisory Committee ( 39 th Session ) Predictability, diagnostics and extended-range Apr 9–18 forecasting Training Course – Oct 13–17 Use and interpretation of ECMWF products Apr 21–30 Data assimilation and use of satellite data for WMO Members May 12–22 Parametrization of diabatic processes Oct 13–14 Finance Committee ( 81 st Session ) Advisory Committee on Data Policy th Apr 7–8 Oct 14–15 Policy Advisory Committee ( 27 Session ) (9th Session ) Advisory Committee of Co-operating States th Oct 20 Apr 8–9 Finance Committee ( 80 Session ) (14 th Session ) Apr 9–10 Policy Advisory Committee ( 26 th Session ) 13 th Workshop on Nov 3–7 May 6–9 WCRP/WWRP Modelling Summit “High Performance Computing in Meteorology” May 13–14 Security Representatives’ Meeting Nov 10–12 Workshop on “Ocean-atmosphere Interaction” May 14–16 Computing Representatives’ Meeting Dec 2–3 Council ( 70 th Session ) ECMWF publications (see http://www.ecmwf.int/publications/) Technical Memoranda 539 Morcrette , J.-J. , P. Bechtold , A. Beljaars , A. Benedetti , 545 Steinheimer , M. , M. Hantel & P. Bechtold : Convec- A. Bonet , F. Doblas-Reyes , J. Hague , M. Hamrud , tion in Lorenz’s global energy cycle with the J. Haseler , J.W. Kaiser , M. Leutbecher , G. Mozdzynski , ECMWF model. November 2007 M. Razinger , D. Salmond , S. Serrar , M. Suttie , A. 544 Jarlan , L. , G. Balsamo , S. Lafont , A. Beljaars , Tompkins , A. Untch & A. Weisheimer : Recent J.C. Calvet & E. Mougin : Analysis of Leaf Area advances in radiation transfer parametrizations. Index in the ECMWF land surface scheme and October 2007 impact on latent heat and carbon fluxes: 536 Orr , A. : Evaluation of revised parameterizations Applications to West Africa. December 2007 of sub-grid. November 2007 540 Palmer , N. , R. Buizza , M. Leutbecher , R. Hagedorn , 534 Zagar , N. , E. Andersson , M. Fisher & A. Untch : T. Jung , M. Rodwell , F. Vitart , J. Berner , E. Hagel , Influence of the quasi-biennial oscillation on the A. Lawrence , F. Pappenberger , Y.-Y. Park , L. von ECMWF model short-range forecast errors in the Bremen & I. Gilmour : The Ensemble Prediction tropical stratosphere. August 2007 System – Recent and ongoing developments. 519 Leutbecher , M. : On the representation of initial October 2007 uncertainties with multiple sets of singular vectors optimised for different criteria. September 2007

47 GENERAL ECMWF Newsletter No. 114 – Winter 2007/08

Index of past newsletter articles

This is a selection of articles published in the ECMWF Newsletter series during the last five years. Articles are arranged in date order within each subject category. Articles can be accessed on the ECMWF public web site – www.ecmwf.int/publications/newsletter/index.html

No. Date Page No. Date Page NEWS NEWS ECMWF’s plans for 2008 114 Winter 2007/08 2 Applying for resources for a “Special Project” 110 Winter 2006/07 8 New items on the ECMWF web site 114 Winter 2007/08 3 Co-operation Agreement signed with Morocco 110 Winter 2006/07 9 Changes to the operational forecasting system 114 Winter 2007/08 3 Celebration of the career of Clive Temperton 110 Winter 2006/07 10 Celebration of Tony Hollingsworth’s life 114 Winter 2007/08 4 Gerbier-Mumm Award used for a project on the impacts of climate variability on Two new Co-operation Agreements 114 Winter 2007/08 4 malaria in Tanzania 110 Winter 2006/07 11 Ensemble Prediction Workshop, The new IBM Phase 4 HPC facility 109 Autumn 2006 5 7–9 November 2007 114 Winter 2007/08 5 Co-operation Agreement with Estonia 106 Winter 2005/06 8 A wealth of ocean data makes it appearance on the public web at ECMWF 114 Winter 2007/08 6 Long-term co-operation established with ESA 104 Summer 2005 3 Signing of the Co-operation Agreement between Collaboration with the Executive Body of the ECMWF and Montenegro 114 Winter 2007/08 7 Convention on Long-Range Transboundary Air Pollution 103 Spring 2005 24 Book about high performance computing Co-operation Agreement with Lithuania 103 Spring 2005 24 in meteorology 114 Winter 2007/08 8 25 years since the first operational forecast 102 Winter 2004/05 36 ECMWF workshops and scientific meetings 2008 114 Winter 2007/08 8 COMPUTING 68 th Council session on 10 –11 December 2007 114 Winter 2007/08 9 ARCHIVING , D ATA PROVISION AND VISUALISATION 11 th Workshop on New Automated Tape Library for the Meteorological Operational Systems 114 Winter 2007/08 10 Disaster Recovery System 113 Autumn 2007 34 New High Performance Computing Facility 114 Winter 2007/08 13 The next generation of ECMWF’s meteorological Fifteenth anniversary of EPS 114 Winter 2007/08 14 graphics library – Magics ++ 110 Winter 2006/07 36 Dr Anthony (Tony) Hollingsworth (1943 – 2007) 113 Autumn 2007 2 A simple false-colour scheme for the representation of multi-layer clouds 101 Sum/Aut 2004 30 ENSEMBLES public data dissemination 113 Autumn 2007 4 The ECMWF public data server 99 Aut/Win 2003 19 Third EUMETCAL Workshop at ECMWF 113 Autumn 2007 5 COMPUTERS , N ETWORKS , P ROGRAMMING , Replacement of the Automated Tape Library for SYSTEMS FACILITIES AND WEB the Disaster Recovery System 113 Autumn 2007 6 Improving the Regional Meteorological Management changes in the Research Department 112 Summer 2007 3 Data Communications Network (RMDCN) 113 Autumn 2007 36 67 th Council session on 28 –29 June 2007 112 Summer 2007 3 New features of the Phase 4 HPC facility 109 Autumn 2006 32 Forecast Products Users’ Meeting, June 2007 112 Summer 2007 3 Developing and validating Grid Technology for the ECMWF Annual Report for 2006 112 Summer 2007 6 solution of complex meteorological problems 104 Summer 2005 22 Workshop on “Flow-dependent Aspects of Migration of ECFS data from TSM to HPSS Data Assimilation” 112 Summer 2007 7 (“Back-archive”) 103 Spring 2005 22 Access to TIGGE database 112 Summer 2007 7 New ECaccess features 98 Summer 2003 31 German State Secretary visits ECMWF 112 Summer 2007 8 Migration of the high-performance computing service to the new IBM supercomputers 97 Spring 2003 20 IASI radiance data operationally assimilated 112 Summer 2007 8 Moroccan Secretary of State visits ECMWF 111 Spring 2007 3 METEOROLOGY Meteosat-9: The new prime satellite at 0º longitude 111 Spring 2007 3 OBSERVATIONS AND ASSIMILATION Operational assimilation of surface wind data from Monitoring of SSMIS from DMSP-16 at ECMWF 111 Spring 2007 4 the Metop ASCAT scatterometer at ECMWF 113 Autumn 2007 6 Update on ERA-Interim 111 Spring 2007 5 Evaluation of the impact of the space component Workshop on the parametrization of clouds in of the Global Observing System through large-scale models 110 Winter 2006/07 6 Observing System Experiments 113 Autumn 2007 16 David Anderson awarded the Sverdrup Gold Medal 110 Winter 2006/07 8 Data assimilation in the polar regions 112 Summer 2007 10

48 ECMWF Newsletter No. 114 – Winter 2007/08 GENERAL

No. Date Page No. Date Page OBSERVATIONS AND ASSIMILATION FORECAST MODEL Operational assimilation of GPS radio occultation Towards a forecast of aerosols with the measurements at ECMWF 111 Spring 2007 6 ECMWF Integrated Forecast System 114 Winter 2007/08 15 The value of targeted observations 111 Spring 2007 11 A new partitioning approach for ECMWF’s Assimilation of cloud and rain observations Integrated Forecast System 114 Winter 2007/08 17 from space 110 Winter 2006/07 12 Advances in simulating atmospheric variability ERA-Interim: New ECMWF reanalysis products with IFS cycle 32r3 114 Winter 2007/08 29 from 1989 onwards 110 Winter 2006/07 25 A new radiation package: McRad 112 Summer 2007 22 Analysis and forecast impact of Ice supersaturation in humidity observations 109 Autumn 2006 11 ECMWF’s Integrated Forecast System 109 Autumn 2006 26 Surface pressure bias correction in data assimilation 108 Summer 2006 20 Towards a global meso-scale model: The high- A variational approach to satellite bias correction 107 Spring 2006 18 resolution system T799L91 and T399L62 EPS 108 Summer 2006 6 The local and global impact of the recent “Wavelet” J b – A new way to model the statistics of background errors 106 Winter 2005/06 23 change in model aerosol climatology 105 Autumn 2005 17 New observations in the ECMWF assimilation Improved prediction of boundary layer clouds 104 Summer 2005 18 system: satellite limb measurements 105 Autumn 2005 13 Two new cycles of the IFS: 26r3 and 28r1 102 Winter 2004/05 15

CO 2 from space: estimating atmospheric CO 2 Early delivery suite 101 Sum/Aut 2004 21 within the ECMWF data assimilation system 104 Summer 2005 14 Systematic errors in the ECMWF forecasting system 100 Spring 2004 14 Sea ice analyses for the Baltic Sea 103 Spring 2005 6 A major new cycle of the IFS: Cycle 25r4 97 Spring 2003 12

The ADM-Aeolus satellite to measure wind METEOROLOGICAL APPLICATIONS AND STUDIES profiles from space 103 Spring 2005 11 Probability forecasts for water levels in An atlas describing the ERA-40 climate The Netherlands 114 Winter 2007/08 23 during 1979–2001 103 Spring 2005 20 Impact of airborne Doppler lidar observations Planning of adaptive observations during the on ECMWF forecasts 113 Autumn 2007 28 Atlantic THORPEX Regional Campaign 2003 102 Winter 2004/05 16 Ensemble streamflow forecasts over France 111 Spring 2007 21 ERA-40: ECMWF’s 45-year reanalysis of the global atmosphere and surface conditions 1957-2002 101 Sum/Aut 2004 2 Hindcasts of historic storms with the DWD models GME, LMQ and LMK using ERA-40 reanalyses 109 Autumn 2006 16 Assimilation of high-resolution satellite data 97 Spring 2003 6 Hurricane Jim over New Caledonia: a remarkable ENSEMBLE PREDICTION numerical prediction of its genesis and track 109 Autumn 2006 21 The ECMWF Variable Resolution Ensemble Recent developments in extreme weather forecasting 107 Spring 2006 8 Prediction System (VAREPS) 108 Summer 2006 14 Starting-up medium-range forecasting for New Limited area ensemble forecasting in Norway Caledonia in the South-West Pacific Ocean – using targeted EPS 107 Spring 2006 23 a not so boring tropical climate 102 Winter 2004/05 2 Ensemble prediction: A pedagogical perspective 106 Winter 2005/06 10 A snowstorm in North-Western Turkey 12–13 February 2004 – Forecasts, public warnings Comparing and combining deterministic and ensemble and lessons learned 102 Winter 2004/05 15 forecasts: How to predict rainfall occurrence better 106 Winter 2005/06 17 Early medium-range forecasts of tropical cyclones 102 Winter 2004/05 7 EPS skill improvements between 1994 and 2005 104 Summer 2005 10 European Flood Alert System 101 Sum/Aut 2004 30 Ensembles-based predictions of climate change and their impacts (ENSEMBLES Project) 103 Spring 2005 16 Exceptional warm anomalies of summer 2003 99 Aut/Win 2003 2 Operational limited-area ensemble forecasts Record-breaking warm sea surface based on ‘Lokal Modell’ 98 Summer 2003 2 temperatures of the Mediterranean Sea 98 Summer 2003 30 Ensemble forecasts: can they provide useful Model predictions of the floods in the Czech Republic early warnings? 96 Winter 2002/03 10 during August 2002: The forecaster’s perspective 97 Spring 2003 2

ENVIRONMENTAL MONITORING OCEAN AND WAVE MODELLING Progress with the GEMS project 107 Spring 2006 5 Climate variability from the new System 3 ocean reanalysis 113 Autumn 2007 8 A preliminary survey of ERA-40 users developing applications of relevance to GEO Progress in wave forecasts at ECMWF 106 Winter 2005/06 28 (Group on Earth Observations) 104 Summer 2005 5 Ocean analysis at ECMWF: From real-time ocean The GEMS project – making a contribution to the initial conditions to historical ocean analysis 105 Autumn 2005 24 environmental monitoring mission of ECMWF 103 Spring 2005 17 High-precision gravimetry and ECMWF forcing Environmental activities at ECMWF 99 Aut/Win 2003 18 for ocean tide models 105 Autumn 2005 6

49 GENERAL ECMWF Newsletter No. 114 – Winter 2007/08

No. Date Page No. Date Page OCEAN AND WAVE MODELLING OCEAN AND WAVE MODELLING MERSEA – a project to develop ocean and Seasonal Forecast System 3 110 Winter 2006/07 19 marine applications 103 Spring 2005 21 Monthly forecasting 100 Spring 2004 3 Towards freak-wave prediction over the DEMETER: Development of a European global oceans 100 Spring 2004 24 multi-model ensemble system for seasonal to MONTHLY AND SEASONAL FORECASTING interannual prediction 99 Aut/Win 2003 8 Seasonal forecasting of tropical storm frequency 112 Summer 2007 16 The ECMWF seasonal forecasting system 98 Summer 2003 17

New web products for the Did the ECMWF seasonal forecasting model ECMWF Seasonal Forecast System-3 111 Spring 2007 28 outperform a statistical model over the last 15 years? 98 Summer 2003 26

50 ECMWF Newsletter No. 114 – Winter 2007/08 GENERAL

Useful names and telephone numbers within ECMWF Telephone E-mail Telephone number of an individual at the Centre is: The e-mail address of an individual at the Centre is: International: +44 118 949 9 + three digit extension [email protected] UK: (0118) 949 9 + three digit extension e.g. the Director’s address is: [email protected] Internal: 2 + three digit extension For double-barrelled names use a hyphen e.g. the Director’s number is: e.g. [email protected] +44 118 949 9001 (international), (0118) 949 9001 (UK) and 2001 (internal). Internet web site ECMWF’s public web site is: http://www.ecmwf.int Ext Ext Director Meteorological Division Dominique Marbouty 001 Division Head Horst Böttger 060 Deputy Director & Head of Research Department Philippe Bougeault 005 Meteorological Applications Section Head Alfred Hofstadler 400 Head of Operations Department Data and Services Section Head Walter Zwieflhofer 003 Baudouin Raoult 404 Head of Administration Department Graphics Section Head Ute Dahremöller 007 Stephan Siemen 375 Meteorological Operations Section Head Switchboard David Richardson 420 ECMWF switchboard 000 Meteorological Analysts Antonio Garcia-Mendez 424 Advisory Anna Ghelli 425 Internet mail addressed to [email protected] Claude Gibert (web products) 111 Telefax (+44 118 986 9450, marked User Support) Fernando Prates 421 Computer Division Meteorological Operations Room 426 Division Head Data Division Isabella Weger 050 Division Head Computer Operations Section Head Jean-Noël Thépaut 030 Sylvia Baylis 301 Data Assimilation Section Head Networking and Computer Security Section Head Erik Andersson 627 Rémy Giraud 356 Satellite Data Section Head Servers and Desktops Section Head Peter Bauer 080 Richard Fisker 355 Re-Analysis Project (ERA) Head Systems Software Section Head Saki Uppala 366 Neil Storer 353 Probabilistic Forecasting & Diagnostics Division User Support Section Head Division Head Umberto Modigliani 382 Tim Palmer 600 User Support Staff Seasonal Forecasting Section Head Paul Dando 381 Franco Molteni 108 Anne Fouilloux 380 Model Division Dominique Lucas 386 Carsten Maaß 389 Division Head Martin Miller 070 Pam Prior 384 Numerical Aspects Section Head Computer Operations Agathe Untch 704 Call Desk 303 Physical Aspects Section Head Call Desk email: [email protected] Anton Beljaars 035 Console – Shift Leaders 803 Ocean Waves Section Head Console fax number +44 118 949 9840 Peter Janssen 116 Console email: [email protected] GMES Coordinator Fault reporting – Call Desk 303 Registration – Call Desk 303 Adrian Simmons 700 Service queries – Call Desk 303 Education & Training Tape Requests – Tape Librarian 315 Renate Hagedorn 257 ECMWF library & documentation distribution Els Kooij-Connally 751