Home

User Guide

1. 3 The products e These parameters are computed at 6 hourly intervals from 6 to 120 hours and at 12 hourly intervals from 120 to 240 hours based on 12 UTC data The 2 metre temperature and dew point and the 10 metre wind are computed from the values at the lowest model level approx 30 metres above ground and at the surface taking into account a pre scribed state of the surface albedo roughness etc Analysis fields for 00 06 12 and 18 UTC including additional fields such as model orography land sea mask percentage of vegetation etc are also available However it should be borne in mind that surface parameters and cloud and radiation parameters are not analysed in the present system The analysis and forecast output is archived into MARS the ECMWF archiving system of meteor ological data cf Meteorological Bulletin M1 9 2 Dissemination products A subset of parameters is available to ECMWF Member States through the operational dissemina tion system table 5 cf Meteorological Bulletin M 3 1 2 for a description of the system Upper air parameters are available in spectral form or in grid points Table 5 ECMWF dissemination products Upper air parameters on pressure levels and Upper air parameters on pressure levels and model levels model levels Mean sea level pressure Mean sea level pressure to day 7 from day 71 2to day 10 2 metre temperature 2 metre temperature 2 metre dew point 2 metre d
2. Persson A 1991 Kalman filtering a new approach to adaptive statistical interpretation of numer ical meteorological forecasts Lecture presented at the WMO Training Workshop on the Interpreta tion of NWP Products Wageningen The Netherlands 29 July 9 August 1991 WMO TD No 421 Phillips N A 1990 Dispersion processes in large scale weather prediction WMO No 700 WMO 1992 Methods of interpreting numerical weather prediction output for aeronautical meteor ology Technical note No 195 Ensemble predictlon Molteni F R Buizza T N Palmer and T Petroliagis 1994 The ECMWF Ensemble Prediction System methodology and validation Submitted to Q J Roy Meteor Soc Mureau R F Molteni and T N Palmer 1993 Ensemble prediction using dynamically condi tioned perturbations Q J R Meteorol Soc 119 299 323 Ocean wave modelling Komen G J L Cavaleri M Donelan K Hasselmann S Hasselmann and P A E M Janssen 1994 Dynamics and modelling of ocean waves Cambridge University Press User Guide to ECMWF Products 2 1
3. Introduction While good operational use can be made of the forecast weather parameters in the ECMWF model their primary function lies in their impact on the overall atmospheric flow A ten day integration makes it absolutely necessary to include effects with relatively long time scale even as subtle as the evaporation by vegetation in order to handle the flow pattern more accurately The different time scales and feed back mechanisms between the various processes makes the computations extremely complex and expensive The mechanisms for these processes are mainly related to small scale disturbances in space and time smaller than the scales explicitly resolved by the model from convective clouds down to molecular processes The effect that these subgrid scale processes have on the larger scales can be computed only by parametrization i c formulating indirectly their overall effect in terms of known grid scale variables The Planetary Boundary Layer The treatment of the planetary boundary layer PBL the lowest part of the troposphere plays a fundamental role for the whole atmosphere earth system It is through the surface exchanges of momentum heat and moisture that the atmosphere feels that it moves over a rough land surface or a wet smooth sea In the ECMWF model the lowest six levels are at around 30 140 340 600 900 and 1200 m above the model surface Even with this fairly high resolution the vertical gradients of temperatur
4. The representation of the orography used in the ECMWF model was changed in April 1995 Since 1983 an envelop orography had been used whereby the grid square mean height was enhanced by an amount proportional to the standard deviation of the sub grid scale orographic heights This had a positive impact on the forecast dynamics but was detrimental in particular with respect to the physical processes The new representation uses the mean orography and four additional fields describing the height orientation anisotropy and slope of the sub grid orography This allows a more realistic representa User Guide to ECMWF Products 2 1 2 The ECMWF global atmospheric model 773 2 The ECMWF global atmospheric model CS tion of the mountain drag novel and important part of the scheme is that depending on dynami cal criteria it can block the low level flow rather than make the air going over the orography The new orography together with the new physical parametrization produces more realistic precip itation in mountainous regions whereas in the old version the largest rainfall tended to fall on the tops of mountain ranges it now falls on the slopes The land sea mask The model surface is logically divided into sea and land points A grid point is defined as a land point if more than 50 of the actual surface of the grid is land calculated from a 1 6 x 1 6 refer ence field With a T213 resolution islands like Corsica Cre
5. associated with the computation of the perturbations and the use of a non perfect model The oper ational problems are related to the clustering technique and to the coarse Gaussian grid in the T63 model e Limitations related to the model The resolution of the model used to compute the ensemble is currently T63 L19 The T63 abil ity to maintain the mean level of eddy kinetic energy is deficient This implies that the EPS cannot create realistic small scale cyclones and partly as consequence of this a reluctance beyond D 5 to create blockings and strong cut off lows This will affect the forecasted flow pattern probabilities making their interpretation more difficult For example since blockings are generally underestimated beyond D 5 those occasions when a minority of the members still forecast a block become highly significant The T63 resolution should also be taken into account when working with EPS weather param eters The coarse grid smooths out many orographic features and give biased values of rain and temperature in particular in mountainous and coastal regions e Limitations related to the perturbations ue i AAN User Guide to ECMWF Products 2 1 5 The Ensemble Prediction System 3 3 User Guide to ECMWF Products 2 1 5 The Ensemble Prediction System 20 The initial perturbations are currently computed at T42 resolution which is adequate to catch most of the uncertainties in the initial state It has
6. precipitation cloud visibility temperature provided that climatological data for the location exists Some techniques also partly compensate for the model s systematic errors There are two traditional statistical techniques the Perfect Prog Method PPM and the Model Output Statistics MOS technique Perfect Prog Method PPM In the PPM approach a statistical relationship is established between observed values of the pre dictand 2m temperature rainfall etc and one or more observed or analyzed values of any rele vant predictor from the free atmosphere like 850 hPa temperature 700 hPa wind 500 hPa vorticity etc Taking then forecast values of these parameters as predictors the most likely value of the pre dictand is produced with the underlying assumption that the forecast is correct Model Output Statistics MOS Instead of observed or analyzed values MOS relate forecast parameters to the parameters to be predicted The MOS technique enables the use of Direct Model Output DMO values as predictors If the model has a tendency to under or over forecast any predictor this will be compensated for by the MOS technique gd User Guide to ECMWF Products 2 1 41 3 ATA E 4 0703 7773 3 rra dei e 42 Advantages and disadvantages of MOS and PPM While the PPM can be developed using observed data for very long periods and can also be applied to different models the MOS technique is al
7. some countries started operational NWP at 500 hPa three days ahead using this simple barotropic model Objective analysis In the early years of NWP initial conditions for the simulations were obtained from manually ana lyzed meteorological charts laboriously interpolated to pre defined grid points Though the need of an automatic data handling and analysis system was realized already in the late 40 s it was not until the mid 50 s that the current concept of fitting observations to a prognostic first guess field was sug gested and successfully tried The impetus came from both mathematicians and meteorologists the mathematicians were inspired by newly derived estimation theories the meteorologists by their experience of analysing synoptic weather charts over data sparse areas by looking at the likely evo lution from previous analyses Baroclinic models Though the barotropic mode produced useful forecasts at 500 hPa up to three days ahead its disad vantages were its restriction to the middle atmosphere and inability to create baroclinic develop ments The 50 s saw intense efforts in several countries to explore the baroclinic nature of the atmosphere Several quasi geostrophic models were designed where the computations were in principle made for the geopotential height fields the winds and temperatures being derived as spa tial derivatives This allowed simple modelling of heating from below friction and orography The increased comp
8. 3 eo 4 The operationa medium range forecast Sunday 15 May 1994 12z ECMWF Forecast t 48 VT Tuesday 17 May 1994 12z Sunday 15 May 1994 12z ECMWF Forecast t 120 VT Friday 20 May 1994 12z A ARS BS TGS TED ES Es NEZ e G y Figure 16 Propagation of differences over a large part of the northern Hemisphere 500 hPa geopotential difference between consecutive forecasts contour every 2 dam dotted lines negative If a back tracking of differences points to a source region over an area with a dense network of reli able sounding stations and good density of aircraft reports the new forecast may be regarded as an improvement compared to the previous one If the source region is in areas with few or less reliable data the forecast should be regarded with some doubt and yesterday s forecast might provide a bet ter solution 52 User Guide to ECMWF Products 2 1 User Guide to ECMWF Products 2 1 a ec 4 The operational medium range forecast R Summary how to approach the different parts of the medium range The following is intended as a guide to what the forecaster at present can rely on in the forecast from ECMWF beyond day 3 and how the different forecast ranges can be approached e Early medium range 72 to 120 hours Long wave patterns are fairly well predicted but cyclones or frontal systems cannot always be followed up from the initial analysis and their predicted positions and i
9. 31 level vertical resolution Horizontal resolution The ECMWF model uses two different horizontal numerical representa tions a spectral representation based on 213 wave numbers for the representation of upper air fields and the computation of the horizontal derivatives a grid point representation used for computing non linear adiabatic terms and the dia batic physical parametrization The spectral technique where the wave like atmospheric features are represented by trigonometri cal functions is more efficient than the grid point technique Though the grid can describe features with half a wavelength down to 0 6 degrees due to numerical reasons the waves would have to be twice as long to be properly advected by a finite difference scheme In this sense the grid point rep User Guide to ECMWF Products 2 1 3 2 The ECMWF global atmospheric model q 3 e resentation is coarser than the spectral representation which has an effective resolution of about 0 8 degrees It is convenient to integrate spectral models by carrying out much of the computation on a so called Gaussian grid which is regular in longitude and almost regular in latitude Due to the convergence of the longitudes toward the poles the east west distance between the grid points decreases pole ward At 60 latitude it is half that of the equator in the polar region north of 80 latitude it is less than 1 20 compared to the equator This very d
10. 5 The Ensemble Prediction System 307 Many end users have also found that they can draw larger benefits from forecasts expressed in prob abilistic terms either in numbers or in words Though the general public is said to favour forecasts expressed in categorical terms experience has shown that information provided in probabilistic terms is also quickly understood The basic principle which makes forecasts expressed in probabilistic terms superior to categorical relates to the fact that a decision maker who runs the risk of losing L due to bad weather is con cerned only when the forecasted risk exceeds c L if the cost of protecting against bad weather is c Verification of the EPS The verification of the Ensemble Prediction System is different in nature from the verification of a categorical forecast The ensemble system is designed to provide a statistical estimation of the error of the unperturbed control forecast If the estimation is reliable the differences between the per turbed forecasts and the control should be representative of the control forecast error Results with the current system indicate that the spread of the ensemble is on average too small i e the deviation of the perturbed forecasts away from the control tends to be too small compared with the control error This can be seen from the number of cases when the verifying analysis is outside the ensemble around 25 in the early medium range over Europe down to 15 at
11. ECMWF global atmospheric model ec The data assimilation Data avallabllity In many areas of the globe the density of available observations is far below what is needed to sup port the analysis with the required accuracy In those areas the assimilation relies mainly on satel lite based observations cloud cleared radiance data and temperature and humidity data from the NOAA satellites and cloud motion wind data from geostationary satellites between 50 N and 50 S Typical coverages of non satellite observations for one analysis cycle are shown in figure 5 150 W 120W 90 W eow sw O 30E 60 E 90 E 120E 150 ak ao PILOT and wind profiler observations Figure 5 Typical coverage of conventional observations 1 July 1994 12UTC User Guide to ECMWF Products 2 1 23 be IZ SionpoJd AMW 0 SPIND 19SN SUONBAJOSQO URIK 3 S o suolealesgo Aong pasoow pue Bung S009 8 D SUONBAJesqO JIHS PUE JONAS 3 lepow spreydsouje 190918 JMNIA 9YL 2 3 3 3 2 The ECMWF global atmospheric model Data quality In addition to the problem of data availability serious difficulties arise from the highly varying quality of the data itself from one station to another and from one time period to another ECMWF therefore undertakes regular monitoring of all observations used in the data assimilation with emphasis on those from the global radiosonde network Statistics on the differences be
12. Humidity Available on pressure levels 1000 850 700 500 and 200 hPa Forecast ranges from T 12 10 T4240 every 12 hours Derived products Products derived from the individual ensemble members include fields from the clusters table 8 and probabilities of occurrence of selected weather events table 9 The probabilities are estimated Table 8 Products from clusters Geopotential height at 1000 and 500 hPa Temperature at 850 and 500 hPa Forecast ranges T 72 to T 168 every 12 hours The fields provided are the mean and standard deviation of the relevant parameters across the ensemble members belonging to the cluster Available for five clustering areas Europe and four areas as shown fig 18 User Guide to ECMWF Products 2 1 59 eS 5 The Ensemble Prediction System TE 3 C78 Table 9 Probabilities of occurrence of weather cvents Temperature at 850 hPa warm anomaly of at least 4K warm anomaly of at least 8K cold anomaly of at least 4K cold anomaly of al least 8K Precipitation at least mm in 24 hours at least 5mm in 24 hours at least 10 mm in 24 hours at least 20 mm in 24 hours Wind Speed at 10 metres at least 10 m s at least 15 m s Forecast ranges T 24 to T 240 every 24 hours Average Temperature at 850 hPa warm anomaly of at least 2K cold anomaly of at least 2K Mean Precipitation Rate at least 3 mm day at least 5 mm day at most i mm day No rain de
13. NE 3 wi D D Mee dE 2 GC Ge gg LL sil Li Zi EL Gt SE Y gz th dn 91 vi Ob St erm ge ee PEZ e z vz OF amp ZE EE o ei Qi Oz lg IEE LE HE OSE AC EEE de ic H H BE zz H GE MS BL HER OL La ware I TT tir nonsense 66 Fel Ota e fe j Eo E Eo E Ex E User Guide to ECMWF Products 2 1 16 En 8 3 F 2 The ECMWF global atmospheric model allows longer time steps where a limitation is the need to approximate curved trajectories by straight lines or great circles on the globe during each time step Although it would be possible in theory to carry out predictions in a Lagrangian framework by fol lowing a set of marked fluid parcels in practice shear and stretching deformations tend to concen trate parcels inhomogeneously so that it is difficult to maintain uniform resolution over the forecast region A semi Lagrangian scheme is used to overcome this difficulty In this version the grid points are stationary At each time step the scheme computes a backward trajectory from every grid point The point reached defines where the air parce was at the begin ning of the time step The value of the variable in that point is then carried forward to the grid point applying the various physical processes Tests have shown that a semi Lagrangian timestep can be around four times longer than the Eule rian without significant difference in accuracy Parametrization of physical processes
14. Var there is no direct horizontal correlation each vertical observation point is treated separately The horizontal coupling comes later during the sub sequent optimum interpolation In 3D Var surrounding observation points are simultaneously taken into account the adjustment along one vertical is not unrelated to adjacent adjustments For conventional observations 3D Var operates generally as optimum interpolation its advantage being e amore accurate mathematical treatment of the first guess error characteristics e better capability to handle observations of quite different types for example satellite data can be used directly better analysis of planetary waves which will be analysed as a whole and not pieced together from analyses in numerous boxes amore general mass wind balance constraint on analysis increments possible e the first guess errors should not be assumed to be barotropic i e the vertical correlation or influence of an upper air observation is primarily vertical Instead it should be able to have sloping influences This will make it possible to use single level data to correct the position of a slightly wrongly placed front or a trough in a vertically consistent way In order to make the first operational implementation a bit simpler the quality control will be made using the optimal interpolation scheme described earlier 3D Var also has the advantage to facilitate the extension into 4D Var where time en
15. ageostrophic deviation from the streamline isohypses Since the ECMWF model treats the wind and mass fields very realistically jet winds are predicted with useful synoptic skill up to at least five days In some cases the forcing from a jet streak can cause rapid developments In addition the subtropical jet which derives its speed from air transported at high levels from the equatorial latitudes in the Hadley circulation can hardly be traced from the 500 hPa gradients It is particularly important for the weather over the Mediter ranean especially when it interacts with a southerly branch of the Polar Front Jet figure 7 Stratospheric charts are plotted more rarely and then very much as the traditional 500 hPa charts with isohypses occasionally also isotherms or isotachs Potential vorticity might be a useful parameter in increasing understanding of the processes involved SAN U DI Figure 7 250 hPa height and wind ECMWF analysis 16 December 1993 12UTC User Guide to ECMWF Products 2 1 37 8 383 a 3 e Geographical coverage The charts plotted to be used in medium range forecasting are often too limited in geographical coverage A European Atlantic coverage preferably also including the easternmost part of North America is necessary to be able to cope with forecast problems in the short range up to 3 days For forecasts up to 5 days the coverage should be extended to the whole of the North Ame
16. been shown however that crucial analysis errors can have quite small dimensions and these relatively coarse perturbations may have dif ficulties to describe them In addition since only those perturbations which are likely to amplify most within 48 hours are used cases when an initial error start to amplify after 48 hours are not taken into account On other occasions an initially rapidly growing error may weaken after 48 hours and have little impact on the medium range but still be included among the perturbations e Limitations related to the size of the ensembles The current ensembles comprise 32 members The ideal size of an ensemble is difficult to assess as it depends on factors which are not yet well understood such as the representative ness of the ensemble members and of the unperturbed forecast It also depends on the users expectations conceming the reliability of the probabilistic estimates However indications are that much larger ensembles are required in order to sample properly the range of possible solu tions Clustering When the ensemble forecast has been computed similar ensemble members are grouped together into clusters The mean and standard deviation of each of these clusters are computed as well as the mean and standard deviation of the overall ensemble Instead of having to inspect 32 or even more different individual forecasts the forecaster can just look at a limited number at most six in the current
17. day 10 against about 6 as expected in principle with a 32 member ensemble The verification of the ensemble forecast is also done on a case by case basis from the synoptic point of view This is usually based on the cluster fields The results confirm the tendency of the ensemble to be relatively too close to the control however the spread usually gives a consistent indication of the uncertainty of the forecast This can be particularly useful in changing synoptic sit uations as illustrated by the example above The ensemble mean is verified as a categorical forecast with RMS and ACC objective scores However it must remembered that its actual spectral resolution is lower than T63 due to the averag ing process and therefore the straight comparison of these scores with scores from higher resolu tion fields e g from the T63 or T213 models is misleading From the synoptic point of view initial monitoring shows that the ensemble mean is quite skilful in predicting the large scale atmos pheric pattern The last component of the EPS validation is the verification of the derived probability products An example of reliability diagram is shown in figure 20 for cold temperature anomalies Results for temperature generally exhibit a high degree of reliability even in the later part of the forecast range Results for precipitation and for surface wind speed show that the system is generally over confi 3 34 ra 307 5 The Ensem
18. e compute temperature changes in the atmosphere due to release of latent heat or cooling in con nection with evaporation In the general convection scheme sub grid vertical fluxes of mass heat water vapour and momen tum are computed at each model level with the help of a simple cloud model interacting with its environment The scheme is applied to penetrative convection shallow convection and mid level convection They are mutually exclusive so only when the scheme fails to create cloud of one type does it try the next Deep convection predominantly occurs in disturbed situations with a deep layer of conditional instability and large scale moisture convergence The downdraught mass flux is assumed propor tional to the updraught mass flux Shallow convection predominantly occurs in undisturbed flow in the absence of large scale conver gent flow For example trade wind cumuli under a subsidence inversion convection occurring in the ridge region of tropical easterly waves or day time convection over land The moisture supply is from surface evaporation It does not normally produce precipitation Mid level convection describes convective cells which originate at levels above the boundary layer A well known example is Altocumulus castellanus floccus Less clearly visible but frequent are rain bands connected to extratropical cyclones Clouds Although in general the forecast cloud cover gives useful guidance the main purpose of the cl
19. either fall out as precipitation be carried higher up in the cloud or be advected horizontally out of the cloud In this latter case and only then the scheme will use this part of the condensate to create stratiform and cirrus clouds altocumulus castellanus anvils and other remains from convective processes Stratocumulus clouds are linked to the boundary layer moisture flux produced by the vertical diffu sion scheme Stratiform clouds e g low level stratus and medium level nimbostratus types are determined by the rate at which the saturation specific humidity decreases due to upward vertical motion and radi ative cooling Evaporation processes are accounted for in several ways large scale and cumulus induced subsid ence and radiative heating evaporation at the cloud sides due to turbulent processes and turbulent motion at the cloud tops Precipitation processes do not only take into account the local water ice content but also different precipitation enhancement processes The effect of evaporation of falling precipitation is also taken included The hydrological cycle e Precipitation two mechanisms to generate precipitation are included in the ECMWF model for convective and for stratiform frontal or dynamical precipitation convective precipitation the condensation leading to precipitation is assumed to be liquid water or snow No ice crystals or liquid water are assumed to be stored in clouds or floating in the air as cl
20. fields of geopotential Temperature observa tions from TEMPs are used only to check the reported geopotential heights Surface observations of 2m temperature and dew point are used for the humidity analysis In addition surface pressure data from PAOB received from the Burcau of Meteorology Mel bourne are used They are so called pseudo observations covering the Southern Hemisphere derived by experienced meteorologists on the basis of all available information including satellite imagery Pre processing Before being passed to the analysis the observations are decoded and go through a quality control procedure The code format is checked and the bulletins that do not follow a WMO code form syntax errors are displayed on a screen for manual correction by the metcorological assistant on duty when possible This is the only manual intervention in real time in the ECMWF analysis and forecast system No attempts are made to perform corrections based on meteorological considera tions The quality control checks that the reported values are realistic It includes hydrostatic check of TEMP data check of the displacement of ships and drifting buoys etc It conforms to WMO rec ommendations User Guide to ECMWF Products 2 1 3 3 3 M ec 2 The ECMWF global atmospheric model a a a a User Guide to ECMWF Products 2 1 26 The data assimilation and quallty control Sever
21. forecasts of a synoptic feature is not necessarily due to analysis problems of that particular system It is quite often linked with inconsistent treatment of a synoptic feature fur ther upstream If this upstream system appears to be correctly analyzed and forecast increase confi dence can be attributed to the development downstream The synoptic phenomenon of downstream development is familiar to forecasters figure 17 a strong cyclogenesis over the western Atlantic is often followed one or two days later by an ampli fied ridge over Iceland and one or two days later by a low deepening over the North Sea This chain of events is often part of a fundamental dynamic process energy dispersion whereby weather sys tems interact over long distances Usually energy does not propagate with the same speed as the weather system but faster at the group velocity A typical atmospheric group velocity is 25 30 User Guide to ECMWF Products 2 1 4 The operational medium range forecast IZ SIONPOJA AMWI3 01 OPjNH 19SN SUSO sory ui odomg YIL IIA Bose PUBIPUNOJMON OY Woy SIDUINYUIL WY SUB IOI Aen Jod 9pni3uo jo sao1dop eleydsiwey ujeujnog pue u ujequoN Aep ewes eu uo pea 01 ep pue Aep je sjse2e10j TSW JO Sejdurex3 S SIDRA EZ veel DK Li Kepung LA OP2 1 15090103 MIO 721 9661 dy z Arpan ee 221 vest ud y A Aepung LA pZ 1 1696104 SANOS 221 v661 dv 91 Apps 1889910 eue uinjpeu euojjesedo eu
22. from satellites Depending on the complexity of the approach we distinguish between 1 3 and 4 dimensional var iational analysis 1 dimensional variational analysis 1D Var From the orbiting satellites observed radiances from the atmosphere are relayed to stations on the earth where they are translated into temperatures or geopotential thicknesses These SATEMs are sent out on the GTS and have been used at ECMWF for the global analysis in the stratosphere and over the oceans although only over the Southern Hemisphere since Spring 1991 The computation of SATEMs is not trivial or reliable and a significant part of the errors emanate during this process A better approach would be to use the radiances directly The model tempera ture and humidity are interpolated to the locations of the satellite measurements The corresponding first guess radiances are computed and compared with the measured radiances If they roughly agree in principle no major modification is made to the vertical structure of the analysis if they dis agree the 1D Var makes an adjustments of the vertical first guess temperature and humidity pro file so as to produce a radiance that fits the observation better User Guide to ECMWF Products 2 1 2 The ECMWF global atmospheric mode 773 73 E 7 3 pra 2 The ECMWF global atmospheric model Three dimensional variational analysis 3D Var In the adjustment of the vertical profile by 1D
23. range forecast The forecaster and the medium range The r le of the forecaster in the interpretation of the medium range output from Numerical Weather Prediction systems NWP is not as well established as in the short range where a number of tradi tional and modern techniques are available The short range forecaster combines NWP information with later and or additional observations automatic stations radar satellite images etc to make the best possible forecast for every particular need assuming that new observations will affect the forecast in an almost linear way This assumption is clearly not valid beyond the first 72 hours of the forecast Another difficulty is that although the medium range NWP output looks realistic even in the small est scales the actual skill is concentrated in larger scales which are not immediately apparent on conventional charts Associated to that is the problem of the numerical forecast changing from one day to the other with the risk of conveying inconsistent information to the customer Weken 20 Apri 1934 122 ECMWF Forecast 6192 VT Tase 23 Apr 1994 122 Thursday 21 Apri 1994 122 ECMWF Forest 169 VT Thursday 20 Apri 1998 122 NEA Y A d IN Figure 11 Two ECMWF forecasts valid at the same time Both cannot be correct in the end the last one was right but only caught the broad scale The realistic looking details did not verify Therefore it is essential for the forecaster in the m
24. setting to see the main forecast alternatives identified by the EPS and the clustering algo rithm The clustering technique thus provides a convenient way to assess the EPS output from a synoptic point of view However the advantage of having a reduced number of charts to inspect is balanced by some loss of information All the members which are used to create a cluster are similar but not identical and differ in certain aspects which are smoothed out in the clustering process The problem is that what is considered to be important varies with weather situation location and forecast practices Five sets of clusters are computed one for the entire European area and four for the smaller areas indicated in figure 18 The European clustering naturally emphasizes the most dominant synoptic features blockings cut off lows and zonal flow regimes covering the whole or a major part of the area The clusters over smaller geographical areas should generally be more detailed On the other hand with smaller areas there is an increased risk of grouping together forecasts which happen to look similar in a limited region but are quite different on a larger scale The present clustering is performed simultaneously on the D 5 D 6 and D 7 forecasts This means a Cluster will contain not only forecasts that are similar at a specific forecast range but also have followed similar synoptic evolution in the range from day five to day 7 A cluster will there f
25. since 1979 an overview 9 The end use of the ECMWF medium range weather forecasts 10 2 The ECMWF global atmospheric model a 11 The model formulation MEM TC 11 The basic equations EE 11 The resolutions in time and Space a 11 The numerical formulation PM tacos 13 Parametrization of physical processes RS 17 Introduction ee 17 The Planetary Boundary Layer pr 17 A Eet 18 The orography PH ardikia 8 The lind ses EAE 19 Sea surface ll eh 19 GA me MH PERI ER 19 Eegenen TE 19 The soil representation EM nnan 19 Gravity wave drap EE 20 Kata ege 20 CONVECHON sssssssssopiisusssssonunsooidiiaiSanisniondonaiosenbsoosianidk saed b rio kassera bisa aieia 21 Clones A A A NN 21 The hydrological ES 22 The data ssimil tion il 23 Data availability EEN 23 Data quality aia a A ia 25 The analysis and forecast parameters a 25 Pre processing EAEN 25 The data assimilation and quality control 26 Preparation of the data for the analysis a 26 ATZA EEE 26 The assumed observation errors Op an 27 The assumed First Guess FG errors CA 27 Objective optimum interpolation analysis 27 22 December 1995 1 QQ Meteorological Bulletin M3 2 m The normal mode initialisation sens
26. time step the Lagrangian scheme 1 When the spectral technique was introduced with lower resolution models in the early 80 s the purpose was also to improve the accuracy in computing advective terms which was crucial for correct phase speed estima tion User Guide to ECMWF Products 2 1 C uU E E 2 to CG a 9 9 E C a 2 o E o ul e siuiod ees ejealpui sjop sesjewesep ui siyBiey edoing 1640 pub verssnes jepow ey SIDRA BA H ti Ob G 9b wb SL EI oz cr Betis 6 6 6 Bue 69 L Sos O 03 LE 6 vi EL BL Ob ME OF E 9 6 soc Ire ade v e GI E sian A e fe uic LL oil a fos and elI yl K L BLU ad DEE DES L zu eof 2 Dur N beum E E User Guide to ECMWF Products 2 1 14 2 The ECMWF global atmospheric modal 15 1u09 edong 1840 pub ueissnes jepow ey patera pr rr ra D Sn e D 3 eier e zz a 3 MA vt ST ERR e e 9 e 4 Se o e see ee Y eS OE bY db Lo Sy So R SEE E LEE AC ET rees G se GOT AA OZ Gi BL Li OL ZE OZ Z ze oz gy p Se n es ss a BAN a E be inan neetttiosstssugog E o nz RRES 999955 do User Guide to ECMWF Products 2 1 Eo Bue Boa dE Ez E Exa ES E E Eon E e EB Gen El 2 The ECMWF global atmospheric model E uoo edoinz 4810 pub ueissnec jopou ey y eei NETTE iD E V rm TE PORTES CI E EG md A H Eak AAL
27. unlikely e to produce local probabilistic forecasts of weather parameters Like the ECMWF operational deterministic model cach of the individual EPS members pro duces forecast values of near surface parameters such as wind temperature cloud and precipi tation for every location The statistical distribution of these parameters can be used to produce local probabilistic weather forecast as with the high resolution model the users are advised to apply a post processing correction scheme to account for systematic biases In the deterministic model occasional day to day forecast inconsistencies act to complicate the interpretation of the latest forecast This problem should be significantly reduced with the EPS In such periods with large inconsistencies in the deterministic model the EPS should consistently dis play the different solutions among its members They might change in number and thus in likeli hood from one forecast to the next one but the transitions should be smooth A general thumb rule would be that 2 3 to 3 4 of the forecast alternatives should persist from one EPS run to the other The probabilities of a certain cluster solution should at most change by 15 20 i e around five ensemble members Limitations of the current EPS Although the present system already provides useful guidance in all the respects mentioned above it has limitations caused by current scientific and operational problems The scientific problems are
28. 1 3 7773 8 se OEEEMEXAM H EAL LLIIIZP V xcrr s t XXX L U L reten References and further literature 1307 References and further literature ECMWF documentation and publications A quarterly ECMWF Newsletter is distributed to national weather services in the Member States and users of the GTS products worldwide It deals with topics in meteorology and the operational activities at the Centre and provides short descriptions of operational changes to the analysis and forecasting system The newsletter also deals with computing topics Comprehensive documentation of the analysis and forecasting system the archiving and dissemina tion is given in the Meteorological Bulletins The Computer Bulletins provide the guidance to the Centre s computing facilities Scientific and technical aspects of the Centre s work are discussed in informal ECMWF Technical Memoranda A limited distribution within the ECMWF Member States applies to these three types of documentation Individual copies are available from the Cen tre s library on request Proceedings from the Centre s annual seminar and workshops are distributed widely to the national weather services and scientific institutions of the meteorological community ECMWF publishes reviewed papers of results in its own series of Technical Reports available in the libraries of most national weather services and scientific institutions A document
29. 91 Four dimensional variational data assimilation using the adjoint of a multilevel primitive equation model Q J Roy Meteor Soc 117 1225 1254 Forecast model Beljaars A C M and A A M Holtslag 1991 Flux Parametrization over land surfaces for atmos pheric models J Appl Meteor 30 327 341 Hortal M and AJ Simmons 1991 Use of reduced Gaussian grids in spectral models Mon Wea Rev 119 1057 1074 Miller M J T N Palmer and R Swinbank 1989 Orographic gravity wave drag its parametriza tion and influence in general circulation and numerical weather prediction models Meteor Atmos Phys 40 84 109 Miller M J A C M Beljaars and T N Palmer 1992 The sensitivity of the ECMWF model to the Parametrization of evaporation from the tropical oceans J Climate 5 418 434 Miller M J 1993 The modelling of hydrological processes in the atmosphere In Modelling Oce anic Climate Interactions Ed J Willebrand and D L T Anderson NATO ASI Series I Vol 11 Springer Verlag 1 33 Morcrette J J 1991 Radiation and cloud radiative properties in the ECMWF operational weather forecast model J Geophys Res 96D 9121 9132 Reed R A Hollingsworth W A Heckley and F Delsol 1988 An evaluation of the performance of the ECMWF operational system in analyzing and forecasting tropical easterly wave disturbances over Africa and the tropical Atlantic Mon Wea Rev 116 824 865 Simmons AJ D M Burridge M
30. AM model was provided by the advent of remote sensing tech niques for measurements of the ocean surface by means of microwave instruments altimeter scat terometer and synthetic aperture radar SAR Satellite observations may be used to validate and initialise wave models Assimilation of altimeter data was introduced in the global version of the wave model in August 1993 Buoy data are not assimilated instead they serve as an independent check of the quality of modelled wave height The products Two versions of the WAM model are running at ECMWF be global model has a regular latitude longitude grid with a resolution of 1 5 degrees The advection time step is 30 minutes while the source term integration time step is 15 minutes The wave spectrum has 25 frequency bins and 12 directions e the Baltic and Mediterranean model has a resolution of 0 25 degrees shallow water effects are included and the advection and source time step is 10 minutes The wave spectrum has 25 fre quency bins and 24 directions The 2D spectra are the so called wave variance spectra which follow from the energy spectrum by division by a factor Pwater x E All integral parameters can be obtained from the 2D spectrum The current product range is indicated in table 10 and an example of forecast is shown in figure 21 Analysed wave spectra and analysed and forecasted integral parameters are archived Figure 21 shows wave forecast verification over the northern H
31. Centre for Medium Range Weather Forecasts In May 1985 the spectral truncation was extended to wave number 106 the number of levels was increased to 19 in 1986 Finally in September 1991 a much higher resolution spectral model was put into operations whereby the spectral truncation was extended to wave number 213 and the number of levels was increased to 31 The shortest half wavelength resolved is 90 km The model uses a computational grid with a resolution of about 60 km The global grid contains 4 154 868 points in all three dimensions At each of these grid points the meteorological variables are re cal culated every 15 minutes out to ten days ahead The total number of computations amounts to about 20 000 000 000 000 With the current Cray C90 this takes approximately 2 hours The end use of the ECMWF medium range weather forecasts The ECMWF forecasts are used for a wide range of social and economic activities The following list is not exhaustive agriculture harvesting protection frost warnings energy generation oil refining consumption construction heavy engineering general building transport and marine ship routeing off shore activities road ice breaking fishing environmental pollution forest fires water distribution sea level estimations public tourism and leisure health marketing emergencies protection and safety 10 User Guide to ECMWF Products 2 1 b 2 The ECMWF glo
32. Jarraud C Girard and W Wergen 1988 The ECMWF medium range prediction models Development of the numerical formulations and the impact of the increased resolution Meteor Atmos Phys 40 28 60 Tiedtke M 1989 A comprehensive mass flux scheme for cumulus parametrization in large scale models Mon Wea Rev 117 1779 1800 User Guide to ECMWF Products 2 1 References and further literature 3 8 78 7 U8 23 3 References and further literature Use of products Cattani D 1994 Application d un filtre de Kalman pour adapter les temp ratures 2 metres four nies par le mod le ECMWF aux stations m t orologiques de la Suisse Arbeitsbericht der Schwei zerischen Meteorologischen Anstalts Z rich Glahn H R Murphy A H Wilson L J Jensensius J S 1991 Lectures presented at the WMO Training Workshop on the Interpretation of NWP Products in terms of Local Weather Phenomena and their Verification WMO Programme on Short and Medium Range Weather Prediction Research PSMP Report Series No 34 WMO TE no 421 Miller M J 1993 The analysis and prediction of tropical cyclones by the ECMWF global fore casting system progress problems and prospects In Tropical Cyclone Disasters ed J Lighthill Z Zhemin G Holland and K Emanuel Peking Univ Press Bejing 220 231 Murphy A and Katz R W eds 1985 Probability statistics and decision making in atmospheric sciences Westview Press
33. The reports from the two groups were completed in August 1971 and at the conference of ministers in the same year it was decided to create the European Centre for Medium Range Weather Fore gm casts The ambition laid out in the plans was to produce five day forecasts with the same accuracy as subjective two day forecasts in the 50 s ee The ECMWF convention was signed in October 1973 ECMWF was established by eighteen Euro pean States and cooperation agreements have been concluded with Iceland Hungary WMO EUMETSAT and ACMAD The objectives of the Centre were laid down as follows m To develop dynamic models of the atmosphere with a view to preparing medium range weather forecasts by means of numerical methods To prepare on a regular basis the data necessary for the production of medium range weather gm forecasts To carry out scientific and technical research directed towards the improvement of these fore casts 7 ID collect and store appropriate meteorological data To make available to the meteorological offices of the Member States in the most appropriate form the results of the studies and research provided for in the first and third objectives above e and the data referred to in the second and fourth objectives H DD make available a sufficient proportion of its computing capacity to the meteorological offices of the Member States for their research priority being given to the field of numerical for
34. ainous areas many relatively dry valley stations are compared to precipita tion forecasts over higher model orography Observation errors the only quality control applied to the observations prior to their use in ver ification is a gross error check Nevertheless the verification of direct model output values against observations gives a good guid ance as to the regional and seasonal behaviour of weather parameter errors do acu 3 e Random errors These are measured using the standard deviation which is a measure of de biased random forecast errors Typical values over Europe at day 3 60 72 hour forecasts are for precipitation 1 5 mm 6h in winter and 3 0 mm 6h in summer total cloud cover 3 3 5 octas It is fairly constant throughout the year Clouds associated with synoptic frontal systems are generally well predicted 2 metre temperature around 3 degrees Also fairly constant throughout the year e 2 metre humidity between 1 g kg in winter and 2 g kg in summer e 10 metre wind around 3 m s in force and 40 to 50 degrees in direction e Precipitation practically no bias during the cold half of the year In summer especially over mountainous regions the model overestimates precipitation during the day and to a lesser extent underestimates during the night Total cloud cover on average the model has no bias during the night a negative bias of about 0 5 octa during the day since the introducti
35. al and practical means to make any quantitative predictions Bjerknes around 1919 initiated the qualitative approach that has been known as the Bergen School model of air masses fronts and cyclones This conceptual model made it possible for the forecasters to analyse the current atmospheric situation even from a few observations It allowed him to extrapolate the motion of important atmospheric flow patterns and to interpret the weather connected to this flow qualitatively taking the effects of radiation convection and turbulence into account Despite its considerable practical value as a forecasting tool the Bergen school model also had seri ous limitations It could only be used for one or two day forecasts and more importantly could not give any guidance in predicting rainfall amounts or the onset of a blocking After the end of the Second World War the renewed exchange of meteorological observations and the development of a hemispheric network of upper air stations made daily analyses of the general circulation possible At the same time the first electronic computer was constructed and made com plex mathematical calculations feasible The time was ripe for a new onslaught on the problem of computing the weather 1 The last decade of deeper understanding of NWP has shown that Richardson s computations were actually correct the westward drift of the flow pattern in the analytical forecast was due to the barotropic assumption The unrea
36. al quality checks are performed within the analysis itself The observation is compared with the 3 6 or 9 hour forecast first guess field interpolated to the time of the observation SYNOP which report more than once during the 6 hour period or appear in duplicate form are reduced in number the observation closest to analysis time is chosen High density aircraft data and drifting buoy data are thinned when necessary All observations except super observations are compared with surrounding observations and once again with the first guess field There is a specific check for multi level data A flag is set to serve as a quality indicator in the analysis scheme The flag can have one of four values 0 observation is accepted as being correct 1 observation is doubtful but will be used 2 observation is very doubtful and will not be used 3 observation will not be used An observation which has been rejected in the early stages cannot be considered in later checks Observations accepted in the first test can in a later test have their flag changed to the better or to the worse and possibly be rejected Preparation of the data for the analysis In contrast to some other analysis schemes grid points are not analyzed individually in the ECMWF system but are grouped together in boxes of variable horizontal size depending on the data density The boxes reach from the surface to 100 hPa and from 300 hPa to the model s top leve
37. and the Anomaly Correlation Coefficient ACC RMSE measures the difference between the forecast field and the verifying one In most cases the smaller the value the better However RMSE will yield lower apparently better values for smoothed fields or for a forecast model with decreasing activity User Guide to ECMWF Products 2 1 3 8 8 3 The products User Guide to ECMWF Products 2 1 e ACC is the correlation between the forecast and analysed anomalies It mainly measures the skill of the forecast positions and thus implicitly of the phase speed The magnitude of the ACC depends strongly on the size of the area on which it is computed For Europe it has been observed empirically that forecasts with less than about 0 60 correlation have little if any prac tical skill ACC has a tendency to score higher in anomalous situations for example in blocked situations Both RMSE and ACC are computed for specific verification times In the medium range this can give a too critical impression when the exact timing is not crucial for the practical usefulness For example a deep low forecasted to move rapidly over Europe between 120 and 144 hours may cause poor RMSE and ACC if the timing is wrong by 12 hours Verification of weather parameters Verification procedure The weather parameters precipitation cloud cover 2m temperature and humidity and 10m wind are verified against surface observations on a dail
38. and tracing the evolution of meteorolog ical disturbances and air masses It may also help us to understand dynamical processes especially in the upper troposphere and lower stratosphere though exactly how still remains to be explored A large part of the conservation often seen in PV maps can however also be seen in absolute AV Trajectories Two or three dimensional trajectories can be computed from the ECMWF analyses and forecasts Trajectories have mostly been used to trace pollution especially radio active fall out but they can also be an interesting air mass tracer which results can be compared with other thermal and dynam ical tracers mentioned above It will also help the forecaster to grasp the differences between streamlines and trajectories important for medium range interpretation Dynamical statistical interpretation The effect of the large scale flow on the local weather can only partly be described by the model because local conditions such as topography and surface characteristics have a major effect on pre cipitation cloud wind and temperature 3 The products User Guide to ECMWF Products 2 1 3 3 The products Figure 9 A 3 dimensional trajectory Analyses from 12UTC 24 June to 12 UTC 29 June 1994 cross plotted every 6 hours Trajectory starting at 500 hPa A statistical interpretation or dynamic climatology can be produced for any particular weather parameter predictand e g
39. ation of the analysis and forecast model can be found in the ECMWF Research Many als Data assimilation scientific documentation Meteorological Bulletin 1 5 1 Forecast model adiabatic part Meteorological Bulletin 1 6 3 Forecast model physical parametrisation Meteorological Bulletin 1 6 2 User Guide references Historical background Bushby F H 1986 A history of numerical weather prediction Collection of papers presented at the WMO IUGG NWP symposium Tokyo Platzman G W 1967 A retrospective view of Richardson s book on weather prediction Bull Am Met Soc 48 pp 514 552 Platzman G W 1979 The ENIAC computations of 1950 gateway to numerical weather predic tion Bull Am Met Soc 60 302 312 Analysis system Daley R 1991 Atmospheric Data Analysis Cambridge University Press User Guide to ECMWF Products 2 1 69 2 EE 7 3 8 3 3 7 e Eyre J R G A Kelly A P McNally E Andersson and A Persson 1993 Assimilation of TOVS radiances information through one dimensional variational analysis Q J Roy Meteor Soc 119 1427 1463 Lorenc A C 1981 A global three dimensional multivariate statistical interpretation scheme Mon Wea Rev 109 701 721 Machenhauer B 1977 On the dynamics of gravity oscillations in a shallow water model with application to normal mode initialization Beitr Phys Atmos 50 259 271 Th paut J N and P Courtier 19
40. bal atmospheric model ec rm e 2 The ECMWF global atmospheric model m The model formulation m The model characteristics can be summarized by six basic physical equations the resolution in time and space and the way the numerical computations are carried out E The basic equations Of the six equations governing the ECMWF primitive equation atmospheric model two are diag m nostic and tell us about the static relation between different parameters The GAS LAW gives the relation between pressure density and temperature The HYDROSTATIC EQUATION shows the relationship between the density of the air and m the change of pressure with height The other four equations are prognostic and describe the changes with time of the horizontal wind m components temperature and water vapour content of an air parcel and of the surface pressure The EQUATION OF CONTINUITY ensures that the mass is conserved and makes it possible to determine the vertical velocity and change in the surface pressure m The EQUATION OF MOTION describes how changes in the wind velocity are caused by the pressure gradient and the Coriolis force and what the effects of friction are near the carth s sur face m e The THERMODYNAMIC EQUATION expresses how a change in an air parcel temperature is brought about by adiabatic cooling or warming due to vertical displacements latent heat release radiation from the sun and the earth s surface and frictional or turbulent pr
41. between post processing times such as evaporation User Guide to ECMWF Products 2 1 an oy e Suggested post processing Conservative parameters To gain an understanding of the thermodynamics of a flow certain thermal and dynamical parame ters with conservative properties can be useful Thermal parameters like potential temperature and wet bulb potential temperature can be used as air mass tracers but there are also some interesting conservative dynamical parameters The simplest is the absolute vorticity AV applied at the gen erally non divergent conditions at 500 and 150 hPa AV C f constant This simple parameter can to a high degree explain the changes of the large scale flow the deepen ing of a vortex moving equatorward the amplification of a high moving poleward A further extension is the barotropic potential vorticity PV PV f D constant where D is the height of an air volume The baroclinic potential vorticity where the height is defined as the distance between two isentropes 0 and 0 88 at pressure p and p p PV f 80 6p constant has during the last years received increased attention as a tool to diagnose dynamical processes in the atmosphere Dynamical weather disturbances that have sharp gradients such as jets and fronts are associated with large anomalies in the potential vorticity PV s conservative properties make it useful for identifying
42. ble Prediction System dent in predicting the event occurrence which is consistent with lack of spread This should be accounted for when using the products Sample climate d 0 40 10 0 200 30 0 400 500 600 700 800 orecast probability W I 19504 cases 4639 occurrences Brier score 0 123 Skill score based on sample climate 0 32 90 0 100 0 Figure 20 Reliability diagram for the forecast probability of cold temperature anomaly events degt erature anomaly at 850 hPa less than 4K over Europe March May 1995 forecast range ours The histogram in the top left corner shows the distribution in percent of the predicted probabilities of occurrences also indicated by the numbers next to the reliability curve Future developments The Ensemble Prediction System is already contributing usefully to medium range forecasting In particular its skill in differentiating between predictable and less predictable situations prevents over interpretations of realistic looking forecast fields from the T213 model During the coming years the EPS will be further refined Initial perturbations will be computed to better cover possible analysis errors both in space and amplitude The T63 model will be replaced with a higher resolution T106 and thus enhance the ability to forecast realistic synoptic features The size of the ensemble will be increased All this should improve the value of the EPS in the medium range for the prediction of uncertaint
43. can be retrieved both from model and pressure levels Orography When choosing a grid point to represent a station or a certain area the user should keep in mind that the geographically closest grid point might not always be the most suitable Vertically interpolated values can for some parameters distort or smooth vital information Due to the difference between model orography and actual orography at many grid points the direct model output of 2m temperature is representative of an altitude significantly different from the real one correction is therefore necessary simply using the Standard Atmosphere Temporal resolution Forecast fields are produced every 6 hours out to day 5 then every 12 hour out to day 10 Interpola tion to 3 hours to coincide with the synoptic hours can be performed when required Care is needed with parameters which display a strong diurnal variation In such cases more refined meth ods than linear interpolation ought to be used Precipitation forecasts can present some problem in that they are time integrated values for the last Six hours or the last 12 hours beyond day 5 and do not exactly fit in synoptically with the pressure pattern which refers to the end of the accumulation period It is of course possible to plot 12 hour accumulated values centred around the forecast time Beyond day 5 24 hour accumulated precipi tation could be plotted The same applies to the other parameters which are accumulated
44. ch grid point and used by the model to estimate the roughness and the evaporation Gravity wave drag When stably stratified air flow crosses a mountain ridge gravity waves are excited into the flow Depending on the static stability and vertical wind shear these gravity waves can propagate verti cally until they have sufficiently large amplitude to break This process has a certain impact on the large scale flow it makes it slightly less zonal and contributes to the formation of blocking highs and cut off lows In 1986 this previously unrepresented physical process was incorporated into the model as the gravity wave drag GWD scheme It represents the momentum transport due to sub grid gravity waves Some of the wave drag occurs in the stratosphere and there is also significant low level drag The dependence on static stability and wind speed implies a maximum in winter and a minimum in summer Radiation In view of the importance of cloud radiation interaction in both long and short term processes ECMWF has placed high emphasis on the treatment of the absorption and scattering by clouds of solar and terrestrial radiation About 1 5 of the overall computational time is devoted to the radia tion scheme as much as for the dynamics The radiation spectrum is divided into eight frequency bands two in the short wave spectrum direct and diffuse radiation from the sun and six in the long wave spectrum from the earth and within the atmosphe
45. cy is unavoidable with a realistic NWP model at every stage in the forecast period synoptic features at all resolved scales should develop User Guide to ECMWF Products 2 1 49 3 3 3 org a 3 ra e 50 with the same frequency even though many will not verify Drifts in the model climate indicate undesirable deficiencies in the simulation of the atmospheric processes A reduction of the day to day inconsistency is therefore not an indication per se that the model has improved such a reduction can be achieved by changes in the model which make it less realistic e g developing fewer and fewer cyclones or blockings in the later stages of the forecast period At present the ECMWF model has a practically stable model climate over the Northern Hemisphere whereas in the Southern Hemisphere a minor but significant increase in eddy kinetic energy takes place figure 15 The model s ability to create features of all scales throughout the simulation will unavoidably com plicate the interpretation of the forecasts However the task of removing unpredictable synoptic features cannot be accomplished during the forecast computation it must be left to the post processing preferably through a mixture of automatic means and of subjective interpretation by forecasters Day to day inconsistency is not only a problem for the operational forecaster in his relation with the NWP output it is also important with re
46. d an intermediate medium range category for which it was necessary to consider the details of both the initial state and the external forcing From the experience gathered with short range and climatological simulations there was in the late 60 s enough know how in both these fields to attack the medium range forecast problem defined as the interval from three to ten days ahead The scientific and technical problems were still formida ble and only few countries had enough expertise to tackle them This made medium range fore casting an ideal candidate for multi national co operation When a PE model began operating in the USA in 1966 there were moves in Europe to build up a similar system to provide weather forecasts for up to 10 days ahead The creation of ECMWF In October 1967 the Council of Ministers of the European Communities adopted a resolution to implement a programme to promote joint scientific and technical research A proposal for a Euro User Guide to ECMWF Products 2 1 1 The European Centre for Medium Range Weather Forecasts ec pean Meteorological Computer Centre for Research and Operations occupied the first place on a m list of meteorological projects submitted by an expert group in April 1969 The proposal was accepted and other European nations were invited to participate In April 1970 an expanded expert group initiated two study groups to look into the economic and scientific motivations for the project
47. dium range forecast Day to day forecast differences Consistency as an indication of skill During the 1980 s many medium range forecasters began to use the day to day agreement consist ency between successive forecasts to give an indication of forecast quality Several objective and subjective investigations have however failed to confirm any useful relation between day to day consistency and skill The typical correlation in the medium range is disturbingly low only 0 1 0 3 mainly because successive bad forecasts quite often are also consistent Consistency correlates highly 0 7 0 8 with the skill of yesterday s forecast This can unfortu nately not be of much use since it only means that day to day agreement mostly occurs when the one day old forecast is better than normal But a good D 6 is generally of the same quality as a typ ical D 5 anomaly correlation of around 75 at 500 hPa over Europe so there is actually nothing for the forecaster to gain So using day to day consistency as a measure of skill actually means making the confidence dependent on a one day old forecast Experience has shown that in cases of inconsistency the fore caster may be reluctant to issue forecasts beyond D 4 Still verifications show that 75 80 of all D 5 and D 6 are better than the D 6 and D 7 from the day before On the other extreme in cases of consistency the forecaster may be over confident and tempted to over interpret the forecast syn op
48. e wind moisture etc in the PBL cannot be described very accurately let alone the turbulent transports of momentum heat and moisture For the estimation of these parameters the model uses the larger scale variables such as wind temperature and specific humidity with the assumption that the trans ports are proportional to the vertical gradients At the earth s surface the turbulent transports of momentum heat and moisture are computed as a function of air surface differences and surface characteristics User Guide to ECMWF Products 2 1 e 18 Eulerian dynamics 15 b Fourier transforms 3 Semi Lagrangian dynamics 22 Legendre transforms 10 Physics 50 Figure 4 The computational time devoted for the different parts in the forecast computation The surface The model orography determines the relief of the lower model boundary but other characteristics such as vegetation snow cover sea ice distribution albedo etc strongly influence the surface proc esses in the model To have a realistic simulation in the PBL the knowledge of such factors as the heat and moisture content in the surface soil is essential The degree to which convective clouds are created in the model depends on the stored heat and moisture in the soil Over land areas snowdepth soil temper ature and wetness are forecast variables calculated by a model of the soil with 4 layers at depths 7 21 72 and 189 cm The orography
49. e 28 um MEDEL DEG UG EAO 28 1 dimensional variational analysis 1D Var urnes 28 Three dimensional variational analysis 3D Var ees 29 Four dimensional variational analysis 4D Var eere 29 STEE 31 P The operational edin 31 l Direct model GOR 32 P Dissemination AAA 33 Products on the DTE 34 D ta Services ii iaa 34 Verification of the upper air fOrecasts eese eese rores tates statutas 34 Verification of weather parameters en 35 Verification procedure A 35 em Error sources in the verification a 35 Random errors See 36 LIKE 36 m Presentation of medium range forecasts a 36 ST ica 36 USTE eea 38 em AA sers saisies 38 Spatial and temporal resolutions in the retrieval sn 39 Horizontal resolutions sss esse 39 m A T iones tr A 39 lur V E HD Ge 39 Temporal resolution AEE 39 l Suggested A du ea AAE 40 j Conservative parameters a 40 Trajectories A aaa ak gai 40 Dynamical statistical interpretation senc 40 Perfect Prog Method PPM sss sss ennaa nnan onnenn nen 41 m Model Output Statistics MOS en 41 Advantages and disadvantages of MOS and PPM es 42 Adaptive techniques ere 42 j 4 The operational medium range forecast 45 m The forecaster and the medium range ea 45 l j 2 22 December 1995 P L 77 3 2 0 Meteorological Bulletin M3 2 Scale pr
50. ecasting The allocation of the proportions would be determined by Council To assist in implementing the programmes of the World Meteorological Organization To assist in advanced training for the scientific staff of the meteorological offices of the Mem ber States in the field of numerical weather forecasting EO The first operational forecast was produced on 1 August 1979 pm The ECMWF forecasting system since 1979 an overview The ECMWF forecasting system consists of two components a general circulation model and a m data assimilation system Every day ECMWF makes a forecast to ten days ahead and distributes it from its computer system to the systems of the national meteorological services of its Member States via a dedicated telecommunication network En The first ECMWF numerical model was a grid point model with 15 levels in the vertical and a hor izontal resolution of 1 875 degrees of latitude and longitude corresponding to a grid length of 200 m km on a great circle In April 1983 the grid point model was replaced by a T63 spectral model i e P a spectral representation in the horizontal with a triangular truncation at wave number 63 The spectral technique was found to be more accurate than the grid point model for the same computa tional cost The number of levels in the vertical was increased to 16 E User Guide to ECMWF Products 2 1 77 3 718 LS NN cec 1 The European
51. edictability and forecast range uses 45 The large scale flow analysis inferred weather forecast 46 The hemispheric flow pattern analysis inferred weather forecast 47 Day to day forecast differences nn 49 Consistency as an indication of skill atia 49 Why does inconsistency occur eerte 49 The model climate and its relation to inconsistency a 49 Investigating day to day differences sem 50 Summary how to approach the different parts of the medium range 54 5 The Ensemble Prediction System 55 The limits of predictability and the concept of ensemble prediction 55 The usefulness of the EPS oir iia nia 55 Limitations of the current ERE a 56 Le pp A nes TRE 57 The TEE 59 Direct OP aca 59 Derived B Gua Ua ERE 59 Deterministic use of EPS nina ia 60 Use of ensemble mean a 62 Probabilistic use Of EPS a 62 Verification OF DAO EAE 63 Future Le aT 1 a 64 6 The ocean wave forecast aa 65 A EES Ee 65 Th POC rie 65 References and further literature etti dE TUS 69 ECMWF documentation and publications 69 User Guide ueu 69 Historical background ai 69 Analysis SEM RM 69 Forecast IN OE iaa cin 70 Use Of Drouet reegt rieden Edge 71 22 December 1995 3 3 a o 3 Meteorological Bulletin M3 2 2 0 En
52. edium range to focus on the scales which are nor mally predictable at this time range The interpretation can be enhanced further by concentrating on the atmospheric developments which are relevant in particular the behaviour of the planetary waves and the phenomenon of downstream development Handling of day to day forecast differ ences also benefits from the use of these concepts Scale predictability and forecast range The targer the atmospheric system the more predictable it is a polar front wave is predictable some days in advance a squall line of thunderstorms only 12 18 hours a local shower just 30 60 minutes User Gulde to ECMWF Products 2 1 7 4 3 pra 3 3 3 3 T3 g 3 e Forecasters have always well before the advent of NWP been aware of this relation between atmospheric scale and possible extension and detail of the forecast A general rule in weather fore casting is to associate any forecast event with as large as possible a scale and disregard scales that are normally not predictable at the forecast range under consideration Aviation forecasters making 2 12 hour cloud turbulence and visibility forecasts concentrate on turbulent radiative and convective processes short range forecasters looking 12 36 hours ahead are more concerned with the position and intensity of fronts trough lines and baroclinic develop ments In the same way medium range forecasters dealing typically with range
53. ellite data and will also better account for the forecast model equations The assumed observation errors o In the ECMWF analysis system all observations of the same type are ascribed typical error vari ances So computed from long series of observations compared with analysis Except for TEMPS there is generally no distinction between different platforms all are assumed to be of equal quality The assumed typical wind errors are for SYNOP SHIP 3 4 m s DRIBU 5 6 m s TEMP and PILOT are assumed to have a typical error of 2 3 m s in the lower troposphere compared to 3 m s for AIREP and SATOB In the upper troposphere TEMP and PILOT are assumed to have errors around 3 m s AIREP 4 m s and SATOB 6 m s The assumed typical height errors are for SYNOP 7 m 1 hPa SHIP and DRIBU 14 m 2 hPa and PAOB 32 m 4 hPa For TEMP there are three quality classes defined according to the monitoring statistics The typical 200 hPa geopotential error in the first group is 13 m in the second 20 m and in the third 26 m Certain vertical error correlations are used to spread out the influence of e g an observation of an AIREP in the vertical For example a wind report at 250 hPa is assumed to correlate 0 5 with the wind at 200 and 300 hPa only 0 1 with 500 hPa and 100 150 hPa In the same way a height obser vation at 500 hPa is assumed to correlate around 0 5 with the height at 700 and 350 hPa The assumed First Guess FG errors o The as
54. emisphere for the period October November 1995 User Guide to ECMWF Products 2 1 6 The ocean wave forecast s681 1equie2e S uo peseq ISE9810 G uonoeupp pue jybiey eABA UBau jo 15879104 LS SIDA 3 001 M 09 A4 08 MAO Ak MOEk EE ees E E E Ee S EAE E en ee e pee E eia SS E n es etea TANKY SS A Mui RR TM ED co gor V AAA ack SSS SE o DN MS Ze E ert AA enera ee err ES SERRE LII Pe TE SE 22 TER SSS PERS ETE Gea EE Gea SS User Guide to ECMWF Products 2 1 3 7 4 8 P A oro 6 The ocean wave forecast e Standard deviation of wave height forecast error metres October November 1995 13 13 11 1 09 DA Oa 94 os DA 63 x 9 1 2 3 4 6 7 D 0 10 He Day Figure 21 Wave forecast verification against analysis northern Hemisphere The corresponding value for a persistence forecast is about 1 2 m from day 3 onwards Table 10 Wave forecast products 2D spectra Significant wave height Mean wave direction Peak period of 1D spectra mean wave period Global model 1 5 x 1 5 latitude longitude T 0 to T 120 every 6 hours T 132 to T 240 every 12 hours Baltic and Mediterranean model 0 25 x 0 25 latitude longitude T 0 to T 120 every 6 hours Code form FM92 Ext GRIB with local extension User Guide to ECMWF Products 2 1 67 e uc rg mc nr os 68 6 The ocean wave forecast User Guide to ECMWF Products 2
55. ense east west resolution not matched by a similar resolution along the north south direction takes more computer time than is necessary for accuracy To save computer time a reduction of the Gaussian grid was introduced in 1991 The reduced Gaussian grid is defined as a stepwise reduction of the number of grid points along a latitude line to keep the east west separation almost constant Between 20N and 20S the grid is identical to a regu lar Gaussian longitude latitude grid By doing this the computational time has been reduced by about 25 to 30 The numerical formulation The choice of a numerical scheme is the result of a compromise between the need to preserve numerical accuracy and the need to save computer time and speed up the forecast When the T213L31 model was implemented in 1991 the traditional Eulerian scheme would have been too expensive and a semi Lagrangian scheme was introduced to reduce the computer cost The basic difference between an Eulerian and a Lagrangian representation can be seen from the equation in a one dimensional space 40 20 90 0 dt di dr which in an Eulerian way expresses that the local changes in Q are due to the advection of Q by the wind U 00 _ _ 2Q dr dy or in a Lagrangian way that Q is conserved for a given parcel dQ dt Whereas most Eulerian schemes require small time steps to avoid numerical instability the quan tity Q must not be advected by more than one grid length per
56. erature as air mass indicators 10 m wind arrows give extra force to the MSL charts especially if coloured according to the 850 hPa temperature 3 The products User Guide to ECMWF Products 2 1 3 18 3 The products e The 850 hPa temperature is the main parameter for tracing air mass and significant fronts The forecasts of temperature and moisture reflect the atmospheric conditions just above the friction layer The wind is very close to the pressure gradient wind at the surface The 700 hPa chart is a suitable field to study temperature advection and movements of convec tive systems In particular the 700 hPa relative humidity is highly correlated with the synoptic patterns as fronts or cloud systems It is recommended for use as forecast guidance in prefer ence to the vertical motion which can be adversely affected by small scale gravity waves The 500 hPa chart derives its usefulness from its long use as a representative picture of the tropospheric average flow Especially for assessments beyond D 5 the 500 hPa forecast can provide valuable indications of large scale changes Long experience has been collected in interpreting the typical gradients and configurations It is often useful to note the positions of troughs and ridges in the contour and temperature thickness patterns 200 250 or 300 hPa level charts have their operational importance in the representation of the jet streams with their strong
57. ernative less likely forecasts can be based on the other meteorolog ical patterns indicated ed the egile two cases this case occurs in about 10 of situations The spread is eegen large aM there are several different patterns equally supported within the ensemble theoptimal deterministic forecast is still based on the T213 model the confidence in that forecast is lower than average only the very large scale influ ences are predictable one or several alternative less likely forecasts can be based on the other ensemble patterns this isa difficult c case occurring in about 20 of situations The general rule is to con sider both forecasts main EPS and T213 as equally likely alternatives However the emphasis can be put on the EPS or the T213 according to the following broad guidelines considering that the current EPS is run at T63 resolution e onthe EPS when it has consistently updated its solutions over the last days or the most supported pattern forecasts blocking s on T213 in other cases SMALL ENSEMBLE SPREAD i e all ensemble members forecast similar meteorological patterns 15 to 30 of cases depending on the season The single EPS pattern i i ment wi 13 for Then e the optimal deterministic forecast is based on the T213 model the confidence in that forecast is higher than average there arc no alternative forecasts hen single EPS is noL i zrcemen with the T21 I User Gu
58. es even blocking anticyclones The model seems to be able to identify changes in the atmospheric activity but not their geographical location Synoptic scale is almost totally unreliable No skill in weather parameters The general approach is to identify movements and changes in the whole or part of the hemispheric flow pattern User Guide to ECMWF Products 2 1 3 7773 773 3 5 The Ensemble Prediction System 5 The Ensemble Prediction System The limits of predictability and the concept of ensemble prediction Due to improvements in numerics in the realism of the model and in the definition of the initial state the skill of numerical models should continue to improve during the coming years But because of unavoidable model approximations and limitations in the observational coverage and perhaps more fundamental because of the turbulent nature of the atmosphere there will always be a limit to how long ahead and how detailed a deterministic categorical forecast can be produced with some skill It will probably never be possible to forecast thunderstorm cells 24 hours in advance individual cyclones more than a week in advance and large scale flow patterns more than a couple of weeks in advance The Ensemble Prediction System EPS is intended to extend the range of usefulness of the numer ical products by exploring the concept of probabilistic forecast The main idea behind the EPS is to take into consideration the uncer
59. essive movements of the planetary waves often coincide over large sections of the hemisphere A change from stationary to non stationary conditions often starts with one long wave but within a couple of days has spread over a large section of the hemisphere It is not uncommon to see adjacent Rossby waves moving in the same direction westward or eastward Ver ifications show that the ECMWF model with some skill and consistency forecast periods of retro User Guide to ECMWF Products 2 1 47 TRENES SS See Lee e S SE IX SN VS SENN SN RSS SOONG ROS SSRN S CS NS 150 E 4 The operational medium range forecast 21 4 1994 12z 500 HEIGHT SO ICAA ea eae Doo 8 as 9 8 9 8 Y 8 8 g a E g amp 8 8 8 31VQ SISATVNY E E ES is E EE D E 30E 60E 90E 120E 0 sow 60W 30w 20 1 180 W 180 f the flow pattern over a part of the hemisp Retrogression of the large scale flow as seen from trough ridge diagran average between 35N and 60N igure 14 F here even beyond day 6 ual waves often are in error For Europe a general progressive move gression or progression o though the positions of ivid ind ment of the largest scale indicates increasing maritime influence whereas retrogressive movement indicates increasing continental influence User Guide to ECMWF Products 2 1 9g 03 83 3 773 774 4 The operational me
60. ew point 10 metre wind 10 metre wind total precipitation total precipitation total cloud cover total cloud cover to day 7 from day 71 2to day 10 Additional weather parameters large scale precipitation convective precipitation low cloud cover medium cloud cover high cloud cover snowfall snowdepth throughout the forecast range User Guide to ECMWF Products 2 1 3 e Products on the GTS 3 Tho products A limited quantity of ECMWF analysis and forecast products is disseminated via the GTS The product range is summarized in table 6 Table 6 ECMWF products on the GTS Northern and southern hemisphere MSL pressure 850 hPa temperature 500 hPa geopotential Validity 12 UTC analysis 24 48 72 96 120 144 hour forecasts Tropics 30N 305 850 hPa winds 200 hPa winds Validity 12 UTC analysis 24 48 72 hour forecasts Code form FM47 V GRID 5 x5 resolution FM92 Ext GRIB 2 5 x 2 5 resolution Data services ECMWF operates a comprehensive data service from its archives In particular it maintains an archive of level II a atmospheric data in support of projects associated with the WMO World Climate Research Programme Verification of the upper air forecasts Several types of objective verification scores are computed after every forecast run for a number of areas and parameters and stored in a historical data base The most commonly used are the Root Mean Square Error RMSE
61. fined as less than 0 1 mm over the period Forecast ranges five day period from day six to day ten two day period from day six to day seven three day period from day eight to day ten from the ratio between the number of elements from the ensemble indicating the event occurrence and the total number of elements 33 including the unperturbed control forecast A last category of derived products is available namely the fields from thc ensemble mean and ensemble standard deviation Deterministic use of EPS The following guidelines arc geared towards the combined usc of the EPS and T213 model between D 4 and D 7 depending on the spread of the ensemble and on the agreement with the T213 fore cast The aim is to decide on a most probable deterministic forecast define a level of confidence in this forecast User Guide to ECMWF Products 2 1 CT EE d 5 The Ensemble Prediction System 20 If the most probable forecast still has a low probability less than 50 it may be possible to decide upon one or sometimes two likely alternatives LARGE ENSEMBLE SPREAD i e the ensemble members forecast different meteorological patterns 70 to 85 of cases depending on the season ast about 50 of the optimal deterministic forecast is based on the T213 model the confidence in that deterministic forecast is about average details in the T213 forecast should not be over interpreted e one or several alt
62. hindsight look quite realistic He modifies the 06 and 12 UTC and these will in turn have feedback on his current analysis for 18 UTC User Guide to ECMWF Products 2 1 29 TT 4 r FT TZ 30 2 The ECMWF global atmospheric model User Guide to ECMWF Products 2 1 ZI 3 EE 3 3 ZI 3 Sie TE 73 73 o3 er a4 43 3 3 3 The products cC 3 The products The operational schedule ECMWF produces global analyses for the four main synoptic hours 00 06 12 and 18 UTC and global 10 day forecasts based on 12 UTC analysis The operational schedule with the approximate running times of the analysis and forecast is shown in figure 6 As a forecasting centre with the emphasis on the medium range ECMWF operates with long data collection times varying between 15 hours for the 00 UTC analysis and 8 hours for the 12 UTC final analysis This schedule ensures the most comprehensive global data coverage including the Southern Hemisphere surface data and global satellite sounding data An additional analysis is run for 00 UTC with a cut off time of around 3 hours followed by a global 3 day forecast to pro vide some Member States with boundary conditions to their limited area models Data observation time 1501 2100 2101 0300 0301 0900 0901 1500 Approximate time of data cut off gt 0200 1530 1730 2000 ANALYSIS ANALYSIS VT 0600 VT 1200 ANALYSIS ANALYSIS VT 1800 VT 0000 Initialisation In
63. ide to ECMWF Products 2 1 61 J 3 3 a e 5 The Ensemble Prediction System very difficult situation which seldom occurs 1 or 2 of cases the emphasis can be on the EPS or on the T213 according to the previous remarks for the cases with large spread the confidence should be lower than average Use of ensemble mean A better way of using the EPS than just assessing the expected skill is to consider the mean of the whole ensemble figure 19 A high degree of synoptic details can remain even at D 7 especially iN EO SSA Y Figure 19 Analysis for 17 April 1995 top left and day 6 forecasts for the same He T213 above and ensemble mean left when the sprcad is small However thc resulting fields can be very smooth and sometimes mean ingless when very different forecast patterns are averaged together Probabilistic use of EPS The optimal way of using the products from the EPS is however to consider possible alternatives as shown by the clusters or the forecasted frequencies Weather forecasting has in reality always been probabilistic the forecaster has in his wording indicated uncertainties and even suggested alternative developments Tests have shown that forecasters even without EPS do formulate realis tic and uscful probabilities in quantified terms both for short and medium range 62 User Guide to ECMWF Products 2 1 d User Guide to ECMWF Products 2 1
64. ies and alternative scenarios in the synoptic flow and lead to reliable probability forecasts 64 User Guide to ECMWF Products 2 1 E d EE 7 3 3 7 3 3 6 The ocean wave forecast 6 The ocean wave forecast Background Interest in ocean wave forecasting started during the second world war when it was realised that information on the sea state could be of vital importance The first operational predictions were based on the use of empirical wind sea and swell laws An important advance was the introduction of the concept of a wave spectrum in the mid 1950 s followed by a dynamical equation describing the evolution of the wave spectrum This equation has become known as the energy balance equa tion It describes the rate of change of the wave spectrum due to advection wind input dissipation due to white capping and non linear wave wave interactions The wave spectrum gives the distribu tion of wave energy over frequency and direction and gives a complete specification of the sca state The wave model that is used for ocean wave forecasting at ECMWF is the WAM model developed during the 1980 s The WAM model is the first model that solves the complete energy balance equa tion including the computationally expensive non linear interactions A global version of the model became operational at ECMWF in 1992 followed after a few months by a Mediterranean imple mentation A big stimulus for developing the W
65. istribution between bare land and vegetation based on geographical data at each grid point The soil is represented by a four layer model described earlier The soil hydrology is simulated in the same four layers A skin or brightness temperature characteristic of a layer with no heat capacity at the soil atmosphere interface represents the balance between radiative fluxes sensible and latent heat fluxes and a flux into the ground Temperature in the four soil layers changes due to the combined effect of ground heat flux and snow melting where appropriate User Guide to ECMWF Products 2 1 19 8 X33 3 T 8 20 Snow cover the thermal properties of snow covered ground depend only on the depth of the snow cover actually the amount of water in the snow No considerations are made for the age of the snow i e the model will treat old dark snow as if it was white Soil moisture the soil moisture is divided into skin and soil reservoirs The skin reservoir which mainly is moisture on vegetation evolves under the action of its own evaporation and its ability to collect dew and intercept precipitation The soil reservoir takes into account con tributions from precipitation and snow melt as well as losses due to deep penetration gravity evaporation over bare ground and root uptake by vegetation i e to the transpiration e Further climatological fields the vegetation ratio is specified in ea
66. itialisation Initialisation Initialisation FORECAST VT 1200 FORECAST FORECAST VT 0000 VT 0600 TO 10 DAYS Figure 6 The ECMWF operational schedule all times shown in UTC User Guide to ECMWF Products 2 1 FORECAST 31 3 3 3 Gees 3 CC 3 The products 32 Direct model output The model variables for the computation of the forecasts are temperature wind and specific humid ity These primary parameters are converted into other atmospheric parameters Tables 3 and 4 sum marize the main output of the forecast model Table 3 Upper air parameters Geopotential height not on model levels Temperature Vorticity and Divergence Wind U and V components Vertical Velocity Specific Humidity Upper air parameters are produced on the original model levels and on standard pressure levels 1000 925 850 700 500 400 300 250 200 150 100 70 50 30 and 10 hPa Table 4 Surface and single level parameters Mean sea level pressure 10 metre wind 2 metre temperature 2 metre dew point Maximum and minimum 2m temperature since previous post processing Large scale and convective precipitation Snowfall Surface temperature and soil wetness Snowdepth Total cloud cover Low medium high and convective cloud cover Surface fluxes surface stress surface roughness albedo Solar and thermal radiation User Guide to ECMWF Products 2 1 773 77 8 7 d
67. l The analysis for all the grid points in each box is carried out using the same data from the box and surrounding boxes assuring consistency of the analysis between the grid points The analysis is evaluated on the model levels not on standard pressure levels Data from TEMPs are normally only taken from 15 standard levels If any are missing the nearest significant level will be used The heights to the significant levels have been computed during the hydrostatic check If there is a significant difference between a reported standard level and the re computed one the former will be rejected and the latter used The analysis Though the need of an automatic data handling and analysis was realized already in the late 40 s it was not until the mid 50 s that the current concept of fitting observations to a prognostic first guess or preliminary field was suggested and successfully tried Though this method was refined during the 60 s and early 70 s the numerical forecast component was given greatest importance From the outset ECMWF invested considerable resources in devel oping an assimilation system and a sophisticated analysis system combined with the most advanced techniques and ideas WE 2 The ECMWF global atmospheric model e The assimilation and analysis system has been further refined during the 80 s and will during the 90 s be further advanced by the so called variational analysis which will make better use of sat
68. lexity introduced new problems and it was not until the 60 s that the baroclinic quasi geostrophic models began to show operational utility and then only for one or two days longer forecasts were still made by barotropic models By that time work was already under way with the introduction or rather re introduction of the primitive equation model Richardson s old invention though now refined and enlarged The primitive equation PE model In the PE model the changes in wind and geopotential fields are not restricted by any geostrophic linkage but are allowed to interact freely Physical parametrizations such as convection which are User Guide to ECMWF Products 2 1 3 c pra 3 pra 3 d Cg a a a M i t n 1 The European Centre for Medium Range Weather Forecasts 207 difficult to handle in the quasi geostrophic model are realistically incorporated so that the tropical regions essential for forecasts over Europe beyond two or three days can be included The first global PE model began operating in 1966 with 300 km grid and six layer vertical resolu tion During the 70 s several other PE models were implemented either hemispheric or global or as Limited Area Models which ran on a higher resolution over a smaller area and took boundary val ues from a larger hemispheric or global model The general connection between inc
69. listic pressure changes were due to spurious gravity waves associated to the absence of an initialisa tion process User Guide to ECMWF Products 2 1 3 3 Ee e 1 The European Centre for Medium Range Weather Forecasts ee o The first barotropic NWP model In 1946 the physicist and mathematician John von Neumann suggested that numerical forecasting start from the simplest of all models the barotropic equations of atmospheric motion where the evolution of the flow is determined by the effect of the conservation of absolute vorticity of an air parcel This was taken as too simplistic an approach by the meteorologists who regarded it neces sary to take account of the baroclinic nature of the atmospheric flow In 1948 J Charney and A Eli assen derived quasi geostrophic equations describing the large scale evolution of a baroclinic atmosphere where sound and gravity waves were mathematically eliminated Since this approach was too technically demanding for the computers of the time the first numerical meteorological forecast run in 1950 had to use the simple barotropic model as originally Suggested by von Neu mann The first experiments were surprisingly successful the general mid tropospheric flow pattern was forecast 2 3 days in advance with greater skill than previous subjective methods Even evolutions which looked baroclinic often tumed out to be mainly barotropic in nature From the mid 50 s with improved computers
70. lution can be found to accommodate the different requirements of all various users The trend should be that more and more Member States access the individual ensemble members and produce a clustering as Suited to their needs as possible ECMWF will continue to provide some compromise solution 5 The Ensemble Prediction System Ges a e aa e a ea e EO 58 User Guide to ECMWF Products 2 1 EE 3 7 8 778 5 The Ensemble Prediction System eo The product range Products from the Ensemble Prediction System include direct output and derived products The standard dissemination system is used for most products a limited set of graphical products is sent to Member States by telefax to assist with the initial implementation of the system No products from the EPS are available on the GTS The EPS output is archived into MARS The control forecast is available in the same way as the T213 forecast but at T63 model and pressure levels and N48 Gaussian Grid surface parameters A wide selection of parameters from the ensemble members and derived fields is available cf Meteorological Bulletin M1 9 2 Direct output The list of fields available from the control forecast and the individual ensemble members is given in table 7 upper air fields Table 7 Fields from individual ensemble members Geopotential height Temperature Vorticity and Divergence Wind U and V components Vertical Velocity Specific and Relative
71. m rep 00096 Meteorological Bulletin M3 2 User Guide ECMWF Products L p MA European Centre for Medium Range Weather Forecasts Wai Europ isches Zentrum f r mittelfristige Wettervorhersage Centre europ en pour les pr visions m t orologiques moyen terme t H Meteorological Bulletin M3 2 j r User Guide ECMWF Products ew PVR European Centre for Medium Range Weather Forecasts Lui Europ isches Zentrum f r mittelfristige Wettervorhersage Centre europ en pour les pr visions m t orologiques moyen terme TO 1 3 Meteorological Bulletin M3 2 CS User Guide to ECMWE Products Designed edited and printed by ECMWF O Copyright 1995 ECMWF User Guide to ECMWF Products 2 1 22 December 1995 E Ez E E E T 203 8 3 8 3 3 4 3 CT 2 0 Meteorological Bulletin M3 2 ec 1 The European Centre for Medium Range Weather Forecasts 5 Historical background to Numerical Weather Prediction NWP ue 5 From von Helmholtz to Richardson a 5 The first barotropic NWP model a 6 E MES 6 Baroclinic AAA AEA 6 The primitive equation PE modela 6 The history AZUR AA LEE 8 Towards medium range and global models 8 The creation of ECMWE ssssssssocsssscssesssssssessceceesacsocesecssecerensesenessessaseasees 8 The ECMWF forecasting system
72. ntensities might be wrong Weather parameters have skill as daily means and as means over the period after removal of possible systematic errors by statistical or other means The general approach is to trace the source of significant day to day forecast differ ences to establish if the main weather systems originate in areas where analysis uncertainties are large lack of reliable observations or might have a large impact on the flow evolution The upstream flow should be examined to see if any new develop ment is forecasted which might affect the downstream evolution e Middle medium range 120 to 168 hours Forecast of the long wave pattern is fairly reliable though the creation of blocks can be underestimated The predicted positions and intensities of cyclones or frontal sys tems are doubtful and spurious features may appear Weather parameters have skill as means over the whole period The general approach is to identify the long waves and changes in the intensity and movements especially possible retrogressions Tracking of forecast differences is still possible but only in rare cases is it possible to trace the source region The use of the forecasts from the last days as a Poor Man s ensemble forecast might indicate possible alternative unless any of the previous forecasts is suspect Late medium range 168 to 240 hours There is some skill in the forecast of the hemispheric flow pattern but little in the long wav
73. ocesses dif e fusion The CONTINUITY EQUATION FOR MOISTURE assumes that the moisture content of an air parcel is constant except for losses due to precipitation and condensation or gains by evapora tion from clouds and rain or from the oceans and continents As in the quasi geostrophic models the hydrostatic assumption makes vertically propagating sound waves impossible Geostrophy itself which would also have excluded horizontal sound waves can m not be assumed as it would have deleted all gravity waves which play a vital role as messengers or mediators in the adjustment process between the wind and geopotential fields m The resolutions in time and space Temporal resolution The computational time step has to be chosen with care in order to avoid numerical instabilities and ensure enough accuracy The present system uses a time step of 15 pn minutes E Vertical resolution The atmosphere is divided into 31 layers The resolution measured in geo metric height is highest in the planetary boundary layer where the levels follow the earth s User Guide to ECMWF Products 2 1 8g a 3 7 e 2 The ECMWF global atmospheric model 12 surface lowest in the stratosphere where they coincide with the pressure surfaces In between a smooth transition is applied Table 2 Pressure of model levels when the surface pressure is 1015 hPa 1 2 3 4 5 6 7 8 9 10 Pressure hPa Figure 2 The
74. on of a new cloud scheme in April 1995 the previ ous scheme had a tendency to strongly underestimate the cloud cover 2m humidity after substantial changes to the surface and soil parametrization scheme in August 1993 the model s boundary layer humidity fits observations well removing a previous tendency to dry out the lowest layers 2m temperature night time too low by about 2 on average in all seasons more in winter than in summer 2m temperature day time too low in winter and spring In summer the forecast has had little bias since the introduction of the cloud scheme in April 1995 the previous scheme produced a large positive bias over continental areas 10m wind the wind force is generally underestimated by 1 to 2 m s during the day There is almost no bias at night This is consistent with the fact that small scales in local wind systems cannot be resolved in the model Presentation of medium range forecasts Charts Together with the mean sea level surface pressure the parameters from the free atmosphere have for long been the forecasters most important tools for synoptic weather forecasting The 500 hPa geopotential field the 850 hPa temperatures and the 250 hPa velocity field etc provide him with data that are essential for the conceptual models of air mass and long waves MSL pressure or 1000 hPa charts can favourably be plotted with fields superimposed such as 850 hPa temperature or moist potential temp
75. ore represent a logical and consistent development It also has the advantage of yielding the same number of clusters at every forecast range 3 H A i i d H E H iens ps Figure 18 Clustering areas The clustering is currently based on the RMS differences between the 500 hPa geopotential fields of the various ensemble members This favours the geopotential height and puts less emphasis on the shape of the flow The clustering may therefore occasionally produce two or more clusters which look very similar The differences may then lay with the general level of geopotential heights indi cating colder or warmer air masses or with the strength of gradients indicating differences in the phase speed of cyclones and fronts steered by the main flow The choice of 500 hPa has the advantage of highlighting the broad scale features which are the most easily predicted beyond D 5 Other parameters like the MSL pressure or 850 hPa temperature could also be used for the clustering Tests have shown that they yield generally similar results but from time to time important differences may appear for example when a uniform zonal flow pre dicted by most of the members may contain small lows and frontal systems with different intensi ties and phase speeds Other approaches to clustering are possible and have been tried An approach based on an objective flow pattem analysis has given promising results However it is clear that no general so
76. oud Scheme is to provide input to the radiation computations There have been documented cases when errors in the radiation fluxes have triggered bad forecasts Until 1995 the ECMWF model did not contain any explicit clouds only interpretations from other fields like relative humidity precipitation vertical motion and vertical temperature gradients The clouds could only affect the forecast through their effect on the radiation A new scheme was introduced in April 1995 with clouds as prognostic parameters defined through the cloud fraction and the content of cloud liquid water and cloud ice The cloud scheme is unique in treating the main cloud related processes in a consistent way by forecasting both cloud fraction and cloud water ice content with their own prognostic equations In the new scheme the cloud proc esses are strongly coupled to other parametrized processes Clouds are generated by large scale ascent cumulus convection boundary layer turbulence and radiative cooling They are dissipated through evaporation due to large scale descent cumulus User Guide to ECMWF Products 2 1 21 3 3 8 ra e induced subsidence radiative heating and turbulence at both cloud tops and sides as through pre cipitation processes Convective clouds are computed in parallel with the convective scheme see above Moisture is carried upward and condenses into liquid water ice or condensate This condensate will
77. oud particles The rain snow partly moistens the atmospheric environment by detrainment the rest falls out of the cloud as precipita tion stratiform precipitation precipitation both as water and ice crystals snow are con sidered depending on the temperature of the layer where condensation takes place No condensation is stored as cloud drops or ice crystals The transition from ice to water is supposed to occur when the precipitation passes through layers with temper atures warmer than 2 C Freezing water is not considered in the model except as snow Evaporation it is assumed that precipitation falling from a saturated layer saturates the first layer underneath before reaching the next layer below This may substantially reduce the pre cipitation on the ground Evaporation of the precipitation is not assumed to take place within the cloud only between the cloud base and the ground The layer below a precipitating non convective cloud is always almost saturated Melting melting of falling snow occurs in a thin layer of a few hundreds of metres below the freezing level It is assumed the snow can melt in each layer whenever the temperature exceeds 2 C The melting is limited not only by the snow amount but also by keeping the induced cool ing of the layer such that the temperature of the layer after melting is not less than 2 C User Gulde to ECMWF Products 2 1 2 The ECMWF global atmospheric model 77 2 2 The
78. previous day s D 6 D 7 and D 8 Smoothing in space is accomplished e g by spectral truncation at T10 It is also very simple and has the advantage of highlighting the model s skill in forecasting the phase speeds of the large scale flow pattern 46 User Guide to ECMWF Products 2 1 KO 4 The operational medium range forecast ec ECMWF Analysis VT Thursday 21 April 1994 122 Thursday 21 Ape 1994 122 ECMWF Forecasl lo 72 VT Bunday 24 Apel 1994 122 Figure 13 Example of fields filtered at T10 geopotential at 500 hPa The forecast evolution becomes clearer with non predictable systems smoothed out Doing this the forecaster will realize that there is more consistency from one day s forecast to the next than meets the eye To make best use of the smoothed fields the forecaster has to use his experience of the relation between large scale flow regimes and surface weather Large scale pressure systems have a different synoptic behaviour from the conventional frontal disturbances they can be stationary for long times and even move westward against the prevailing westerly flow retrogression A retrogression in the large scale will have significant consequences for the weather type Since it occurs at low speed it may be disguised by smaller scale systems moving eastward It will appear much more clearly on a sequence of smoothed forecast plots The hemispheric flow pattern analysis inferred weather forecast Retrogressive or progr
79. re The upward and downward diffused radiation is computed for each of the eight spectral bands The parameters influencing the emission and absorption are pressure mois ture cloud cover and cloud water content Assumed parameters are the ground albedo modified according to the snow cover the solar constant and the concentration of CO Oy and aerosols The radiation scheme is designed to take the cloud radiation interactions into account in considera ble detail It allows partial cloud cover in any layer of the model For cloudy grid points computa tions are made both for clear and overcast conditions and the total amount weighted together according to the forecast cloud amount Provision is made to have the radiative effects of various types of acrosols oceanic desert strat ospheric and background taken into account The carbon dioxide has a constant mass mixing ratio over the whole globe corresponding to a volume concentration of 345 ppmv The ozone distribution depends on height latitude longitude and season based on analyses from the 1970 s User Guide to ECMWF Products 2 1 2 The ECMWE global atmospheric model EE 3 3 2 The ECMWF global atmospheric model Convection The convection schemes in the model fulfils five objectives e create a cloud amount to be used by the radiation scheme compute the precipitation compute the vertical transport of moisture compute the vertical momentum fluxes
80. reased skill and model development better use of an increasing amount of observations and improved analysis systems is seen in the verification history fig 1 The present day 3 forecasts of ECMWF are of the same quality as the 36 hour forecasts made sub jectively in the mid 50 s Skill percent Figure 1 The S1 skill score for the operational 36 hour NMC forecast from 1955 92 The S1 score is a measure of normalized error in horizontal pressure gradients The normalization used here is 2 x 70 S1 yielding O for a worthless and 100 for a perfect forecast User Guide to ECMWF Products 2 1 7 ud 77 3 77 3 8 3 8 gece F 4 773 Table 1 Summary of the development in large scale NWP 1950 1990 Simple topog raphy land sea moisture Convection cloud radia tion friction 50 100 km diffusion 20 30 levels 1 The European Centre for Medium Range Weather Forecasts 1000 amp 500 hPa height and thick ness Most atmospheric parameters incl 2m T 10m wind clouds rain snow showers The history of ECMWF Towards medium range and global models In 1955 J von Neumann had outlined an overall strategy in atmospheric modelling and prediction Between short range predictions of motions determined mainly by the initial state of the atmos phere and climate simulations with longer term predictions that are largely independent of the ini tial state he identifie
81. rican continent and easternmost Pacific For forecasts beyond 5 days the whole hemisphere should be considered 140 W 160 W 160 160 E 140 E Figure 8 Suitable areas for different lead times Meteograms Plotting the forecast parameters as time series meteograms is a convenient way to present and familiarize oneself with the weather related forecast output The quality can be enhanced by apply ing statistical corrections to the parameters to be plotted A specific problem with meteograms is that they convey a too confident impression for the non professional user 3 The preducts User Guide to ECMWF Products 2 1 3 rra g 3 The products Spatial and temporal resolutions in the retrieval From the standard parameters in the ECMWF forecast various derived parameters can be com puted ranging from simply corrected 2 m temperature to complex turbulence parameters The user should be aware of the spatial and temporal resolutions in the retrieval and of the role of orography Horizontal resolution The ECMWF forecast products can be retrieved at a wide range of resolutions from a coarse 5 x 5 to the original reduced Gaussian grid of the model For near surface parameters the distinction between land and sea points may be crucial for example for 2 m temperature precipitation or 10 m wind and the use of the model s own reduced Gaussian grid is highly recommended Vertical resolution The data
82. s of three to seven days or even more concentrate on the movements of the large scale patterns They should disregard smaller scales which in the later part of the medium range may for example mean polar front waves or cut off lows The skill of the ECMWF model for various scales can be broadly measured by comparing the energy of the forecast error with the energy of the analysed field in spectral space throughout the forecast range figure 12 It shows that the forecast of planetary scales is skilful up to and including Wave number a 388 3 85888558 ea o o 4 10 Forecast range days 4 The operational medium rango forecast y Figure 12 Wave number for which the energy of the forecast error at a Gk forecast range exceeds the energy of the initialized analysis deduced from spectra of gaon fields of Z500 for December 1993 The forecast for higher wave numbers smaller scales has little skill in practice day 10 while the day 7 forecast represents properly scales of around T10 Because this large scale is more skilfully forecast than the synoptic scale it also suffers less from day to day inconsistency The large scale flow analysis inferred weather forecast The long waves can be identified by smoothing the forecast fields which can be done both in time and space Smoothing in time for example by plotting the mean of D 5 D 6 and D 7 is techni cally very simple The mean field can be compared with a mean of the
83. sation process the model suppresses these gravity waves During this process the mass and wind fields become adjusted in such a way that no further undesired gravity waves will appear For forecasts beyond 1 2 days initialisation is really not necessary since the model itself would have smoothed out most of the gravity waves during the forecast process Initialisation is still per formed for the requirements of the data assimilation Unless initialized spurious waves in the First Guess especially in the surface pressure would have lead to the rejection of good observations and acceptance of bad ones and to a wrong estimation of the necessary increments to be caused by the selected observations Varlational analysis The optimum interpolation method was designed and further developed mainly to cater for obser vations made at the same time and reporting quantities directly related to model variables But in today s operational meteorological environment data vary considerably both in type and quality not to mention that they more often than not are made at asynoptic hours To some extent it has been possible to extend the OI to take into account this diversity in the data flow but several other short comings in the OI have made a new approach necessary the variational analysis in which better use can be made of many new types of measurements which have a complicated indirect relation ship with the analyzed quantities e g radiance data
84. semble PrediChON iii 71 Ocean wave modelling a 71 4 22 December 1995 2 3 1 The European Centre for Medium Range Weather Forecasts 1 The European Centre for Medium Range Weather Forecasts Historical background to Numerical Weather Prediction NWP From von Helmholtz to Richardson In 1888 the hydrodynamist H von Helmholtz made it possible to formulate mathematically the fun damental laws of atmospheric motion with the aid of the complete set of hydrodynamic and ther modynamic equations At the turn of the century another hydrodynamist V Bjerknes suggested that the weather could be quantitatively predicted by applying these physical laws to a carefully ana lysed initial atmospheric state Bjerknes vision inspired a young mathematician and meteorologist L E Richardson to endeavour to compute weather forecasts along these purely mathematical lines Expressed in today s terms and very much ahead of his time Richardson defined a primitive equa tion model with a 300 kilometre grid and five layers in the vertical in which he set out to integrate the basic atmospheric equations The result published in 1922 was a disappointment Not only did the flow pattern forecast appear to be a total failure it would need 64 000 individuals to perform the manual computations just to keep pace with the weather itself The ways of weather forecasting took other directions during the 20 s and 30 s Lacking the theoretic
85. spect to his customers End users are very sensitive to inconsistent forecasts and the forecaster may risk losing his customer s confidence if he too often changes his forecast drastically from one day to the next But just because the NWP model has to obey a stable climate with frequent changes in the forecasts there is no reason why the forecaster should do the same He does not have to phrase his medium range forecast with the same degree of detail and confidence as his short range forecast He must introduce a climatological drift by con sidering only those scales that are normally predictable at a certain time range Investigating day to day differences Instead of regarding day to day forecast differences as a nuisance the forecaster can actively make use of them It is often useful to compute the difference between today s and yesterday s ECMWF forecasts and trace the differences back to their sources This can be done for any level in the tropo sphere 1000 500 and 300 200 hPa levels yielding the most significant information Up to D 3 the forecast differences evolve linearly and might already have generated the first gen eration of downstream differences with an opposite sign to the initial difference This difference will in tum create a new one downstream and by D S a chain or train of positive and negative dif ferences will have formed over an area covering almost half the hemisphere figure 16 Inconsistencies in the
86. sumed uncertainty of an observation o is combined with the assumed uncertainty of the FG oj resulting in an estimate of the total uncertainty in the analysis For the following FG 6 hours later this uncertainty is increased by about 5096 a value roughly representative of a typical error growth over six hours In the present analysis system neither the analysis error nor the FG error are flow dependent They depend to a large extent on the typical data quality and coverage in the area which in some areas may lead to occasional misfits Objective optimum Interpolation analysis If Z is the extrapolated geopotential or FG and Zp a single new observation and and are the assumed errors then the analysis Z takes the value Z Z 2 2 02 02 02 which means that when the observations are unreliable da is large and Z almost takes the value Ze Then there is little impact on the FG On the other hand when the observations are assumed fairly accurate c is small and Z almost takes the value Z Then the observations will have a substantial impact on the analysis User Guide to ECMWF Products 2 1 3 7 3 77 8 rra EI 8 We e 28 The normal mode initialisation Though the wind and mass fields are balanced within the analysis boxes on the larger scale between the boxes there appear minor imbalances which set up fictitious undesired gravity waves In the normal mode initiali
87. tainties in the description of the initial state exploring what r le they may play in the development of the forecast To simulate the effects of possible analysis errors the basic analysis is slightly modified perturbed in sensitive areas identified by the system depend ing on the current flow pattern Perturbations are computed for these areas and those which would amplify most during the first 48 hours are selected In the current configuration of the system the perturbations are then combined to form 16 different perturbation patterns By just reversing the signs another 16 mirrored patterns are created ending up with 32 slightly different initial analyses No perturbations are chosen from the Southern Hemisphere If the 32 different forecasts for a certain location are quite similar up to a certain range the forecast is likely to be very accurate for that location within that range Divergent forecast solutions beyond that range will provide information on possible alternative developments In the present system only the effects of the uncertainties in the initial state are explored Experi ments have shown that analysis uncertainties are a major factor behind poor forecasts up to around D 6 Model errors reflected in the analysis through the first guess are also covered by the initial perturbations The usefulness of the EPS The EPS can be used on three different levels which complement each other e asa measure of predictability in
88. te and Cyprus are represented by two grid points Mallorca by only one The Faroe Islands Rhodos Gotland and several Danish Islands are not represented by any land point Sea surface temperature The sea surface temperature SST is based on analyses received daily from NMC Washington It is based on ship buoy and satellite observations It is generally of high quality but in small waters like the Baltic where rapid changes in SST can take place during the cold season the real SST can sometimes differ by as much as 5 from the analysis Also in areas where there are few data such as Hudson Bay there can be large differences between real and assumed SST The SST over ice free water is kept constant during the forecast Albedo A background yearly climate field is used according to which the albedo is set to 0 55 over sea ice and 0 07 over open water Over land the albedo has a minimum value of 0 07 and cannot exceed 0 80 The model is able to alter it during the run according to changes in snow cover Ice The sea ice distribution corresponds to the region of SST values below 1 8 C The temperature at the surface of the ice is variable according to a simple energy balance heat budget scheme The dis tribution of sea and sea ice points is kept constant during the forecast no freezing of the water or melting of the ice is allowed The soll representatlon The parametrization of the surface processes over land takes into account the d
89. ters as the fourth dimension Four dimensional variational analysis 4D Var The 4D Var technique actually meets some very old demands from forecasters raised already when the first NWP analysis schemes were introduced in the 50 s e the analysis should not only rely on the latest available information but make use of informa tion also 12 24 hours back OI does this implicitly through the first guess but 4D Var will make this in an optimal way This means that it should be possible to reconsider observations that initially were weighted down but which might appear realistic in light of later data e Uhe errors in the first guess should be dependent of the flow pattern For example larger obser vation weights should be applied near a rapidly moving or deepening cyclone where the fore cast uncertainty is rather large The four dimensional variational analysis is able to cope with these problems since the influence of an observation in space and time is controlled by the model dynamics and is therefore more realis tic The implicit first guess errors are carried by the dynamic flow patterns calculated by the forecast model 4D Var is to some extent similar to a forecaster s analysis of a chart Working on say 18 UTC he goes back to the previous 12 UTC and even 06 UTC charts and modifies them in the light of the 18 UTC analysis he might see that a trough should have been drawn sharper at 12 UTC a rejected observation at 06 UTC might in
90. the atmosphere Depending on the level of general accuracy required by the end user a categorical forecast can be issued as long as the internal spread of the ensemble has not exceeded a certain threshold The spread considered should preferably be the spread of forecasted weather parameters mea sured at the place of interest The general synoptic spread will also serve as a suitable guidance It is not un natural if the spread of an ensemble forecast does not monotonously increase with lead time If the spread of a given EPS run is larger at D 5 than at D 7 it means that the fore cast five days ahead is more uncertain than later on Even if the general synoptic flow is pre dicted with high confidence and the spread of the overall EPS is small there may be large local User Guide to ECMWF Products 2 1 55 SZ 3 3 uncertainties e g if a small developing low passes north or south of a certain location in the different forecasts pe indicator of possible alternative developments When a categorical forecast is not possible the EPS can provide possible alternatives If 70 of the ensemble members forecast a blocking giving dry weather and weak winds and 30 a zonal flow leading to some rain windy and mild conditions this information will be useful for many customers Other customers will benefit from being told also what will not happen in the same example that gusty northerly winds with frequent showers are most
91. tic details Day to day differences serve however a useful purpose by indicating possible alternative solu tions Poor Man s ensemble forecast The figures quoted above also mean that 20 25 of yester day s D 6 and D 7 forecasts are better than the current D 5 and D 6 and thus may contain useful information about possible alternative Why does inconsistency occur Day to day changes in the NWP forecast inconsistency is an unavoidable consequence of a non perfect forecast system bad observations imperfect analysis system and model deficiencies The factors which cause bad forecasts also cause inconsistencies An erroneous observation which has been accepted by the system may introduce an analysis error and cause a forecast failure But the same is true for factors which cause good forecasts A good observation may improve the analysis and subsequently the forecast However in both cases the forecast will be inconsistent in one case to the worse in the other to the better The work at ECMWF aims at reducing both the bad and the good inconsistency by improving the model and the data assimilation system An improved model will not only provide better fore casts it will also support the data assimilation by providing a better first guess field and thus more efficiently be able to identify bad observations The model climate and Its relation to inconsistency Paradoxically a certain degree of day to day inconsisten
92. tween the observed values and the values given by the ECMWF first guess field 6 hour forecast are regu larly accumulated In principle they can indicate problems either with the data or with the first guess itself i e the performance of the model in the region However large biases mean differ ences and large random differences are usually an indication of data problems Comparisons with neighbouring observing platforms of the same type and collocation statistics from different data types are used to check the results The analysis and forecast parameters The ECMWF forecasts are computed exclusively in five atmospheric variables surface pressure temperature wind u and v components and specific humidity The various weather parameters precipitation cloud cover geopotential etc are derived as by products from the above men tioned four variables The data used for the height and wind analysis come from SYNOP SHIP TEMP PILOT DRIBU aircraft data AIREP AMDAR and ACARS SATOB SATEM and TOVS cloud cleared radiance data The analysis of humidity up to 300 hPa is based only on TEMP and satellite data Though temperature is one of the basic analysis and forecast parameters in the current operational system it is not based on temperature observations from SYNOP SHIP TEMP and AIREP but temperature from ACARS and AMDAR reports are used The temperature analysis is derived through the hydrostatic equation from the analyzed
93. variance throughout the forecast range like PPM or with gradually reduced variance like MOS In contrast to MOS and PPM the adaptive filter applied to statistical adaptation does not need any long historical data base If the model changes in any significant way the filter will notice it and gradually adjust the statistical relationship statistical interpretation for any station can start as early as 2 or 3 weeks after a station has been set up Figure 10 shows an example case from northern Sweden in November December 1988 3 The products User Guide to ECMWF Products 2 1 8 3 The products Temperature Celsius Observed T2m ECMWF forecast Kalman filtered T2m 10 20 30 40 November December 1989 Figure 10 Direct model output mi and pro Kalman filtered 12 UTC t 42h temperature forecast valid at 06 UTC for Lulea November December 1988 The 2 dimensional filter actually makes a cold start on 1 November and after two weeks manages to identify a relation between forecast bias and forecasted temperature When a cold spell sets in mid December the filter manages to make useful corrections sometimes in the order of 10 degrees User Guide to ECMWF Products 2 1 ma 3 y CS ee 2 TZ 3 we ra 3 4 ra rre eei SAP 3 3 The products User Guide to ECMWF Products 2 1 3 Ss 83 7 3 4 The operatlonal medium range forecast 4 The operational medium
94. ways related to a specific model and can only use data from periods when there have been no substantial changes in the model The great advantage with MOS is that it also takes the systematic errors of the NWP model into account This is especially important to deal with probabilistic interpretations Since the PPM statistical relationships are constant through the forecast period the statistically derived values will have the same variance at D 10 as at D 1 while the MOS equations will tend to damp the variance with lead time For those end users who are interested in minimizing the fore cast error MOS is to be preferred Those users who like to get a view of the possible events e g tourist forecast should prefer PPM because of its greater variability Adaptive techniques Adaptive techniques in particular the Kalman filter have gained increased attention over the last years as a way to go round some of the drawbacks of PPM and MOS Kalman filtering has been extensively used since its invention around 1960 However applications within statistical interpre tation of NWP have appeared only lately Adaptive filters share MOS advantage of being able to compensate for model errors while at the same time being able to continue to work despite changes in the model characteristics Like MOS but in contrast to PPM it can work on forecast near surface parameters like 2m temperature The filter equations can provide interpretation with the same
95. y and monthly basis The verification performed at ECMWF is based entirely on direct model output values interpolated to the locations of the observation stations It does not involve any statistical correction method except for temperature to which a correction based on the standard atmosphere lapse rate is applied to account for the differences between the height of the model orography and the real station height stations with too large deviation from the model orography are excluded from any statistics of tem perature and wind Error sources in the verification Apart from the errors in weather parameters due to inaccuracies in the forecasts of the synoptic flow the following effects can affect the verification against synoptic observations representativeness Surface observations include local effects that cannot be represented due to the model resolution and or inappropriate description of the state of the surface Since there are often large differences in the structures of the boundary layers over land and adjacent sea points the model land sea mask information is used In coastal areas the values at the nearest land grid point should therefore be used rather than the interpolated value between surrounding land and sea grid points sampling to avoid unrepresentative biases care is taken to avoid an unbalanced choice of observation stations However this is not always possible e g when averaging precipitation errors over mount

User Guide

Contents

Download Pdf Manuals

Related Search

Related Contents