Presentation is loading. Please wait.

Presentation is loading. Please wait.

The fundamentals of Seasonal Forecasting

Similar presentations


Presentation on theme: "The fundamentals of Seasonal Forecasting"— Presentation transcript:

1 The fundamentals of Seasonal Forecasting
RA VI Clips Workshop The fundamentals of Seasonal Forecasting J.P. Céron – Météo-France

2 Some Vocabulary Long Range Forecasts and Climate Forecasts
Forecast Range, Forecast period and Lead time. Lead Time Forecast Period Forecast issue time Forecast Range

3 May June July Aug Sept Octo Nov
LT - 1 month Seasonal Forecast 1 May June July Aug Sept Octo Nov Forecast issue time Typical scheme for a coupled forecast (like ECMWF’s one) Coupled Forecast : Range of 6 months

4 LT - 2 month Seasonal Forecast 2 May June July Aug Sept Octo Nov Forecast issue time Coupled Forecast : Range of 6 months

5 LT - 3 month Seasonal Forecast 3 May June July Aug Sept Octo Nov Forecast issue time Coupled Forecast : Range of 6 months

6 The Scientific bases The evolution of the atmosphere is partly driven by the evolution of external forcing conditions (SST and continental surfaces). The evolution of external forcings is often slow and predictable. It gives a slow memory to the atmosphere ; the evolution of the latter becoming partly predictable. The successive instantaneous states of the atmosphere have a limited predictability while the mean states of the atmosphere have a greater predictability. The mean circulation in tropical regions is strongly inflenced by the large scale organised convection.

7 Limitation of numerical forecast : Daily forecast
Daily Scores over Northern Hemisphere (Geopotential Height at 500 hPa). The diagram represent the anomaly coefficient correlation for the daily values of the geopotential at 500 hPa vs the range of the forecast. The used GCM is the Arpege-Climat model but these behaviour is really general. The quality of daily numerical forecasts decreases very rapidly from 1 to 10 days; Beyond this limit there is no more informations.

8 Limitation of numerical forecast : Daily forecast
Daily Scores over Northern Hemisphere + Persistence Scores In red, one can see the persistence scores … Worst than the previous one …. (Fortunatelly for numerical modeling!!!).

9 Limitation of numerical forecast : Daily forecast
Daily Scores over Northern Hemisphere + Perfect model Scores Again in red, the Perfect Prog Scores … quite improved ; note that the perfect model score is obtained taking one member of an ensemble forecast as the perfect forecast an comparing the other members to this one. The predictability limit is around 15 days as expected. Finally, one can a look on the remaining place for improvment (comparing blue and red curves).

10 Limitation of numerical forecast : Monthly forecast
Daily Scores over Northern Hemisphere + Monthly running mean Scores The month integration of the signal improve the predictability and the predictability limit increases (in time) from 15 to more than 20 days. Of course in Meteorology the time integration of the signal corresponds to an equivalent space integration (namely signal on a larger space scale than the synoptic one).

11 Limitation of numerical forecast : Seasonal forecast
Daily Scores over Northern Hemisphere + seasonal running mean Scores Again, the time integration increase the predictability. Consequently, it’s the mean state which is predictable and not successive instantaeous state of the atmosphere.

12 Limitation of numerical forecast : Seasonal forecast
Daily Scores over Northern Hemisphere + Ensemble forecast, seasonal running mean and SST forecast If, additionnaly to the Atmosphere we bring information on the Ocean Surface evolution, we improve again the predictability of the mean state of the atmosphere in relationship with the driving of the mean state of the atmosphere by external forcings (notably but not only SST and particularly in Tropical regions).

13  realistic, one can be quite confident in this forecast
The Predictability « a Thunderstorm will be observed next Sunday over the Toulouse « Météopole » between 15h and 16h »  Irrealistic, the confidence that one can have in this forecast is very low « a rainy system will cross the Toulouse region Sunday afternoon  »  realistic, one can be quite confident in this forecast Note that the main difference in these two forecasts is in both space and time scales.

14 The Predictability The predictability depends on :
The scale of the forecasted phenomenum (Thunderstorm, Easterly Wave, Blocking situation, ENSO, …) The Range of the forecast (NowCasting, Short , Medium , Seasonnal , Climatic)

15 Predictability Space Scales Local  10-100 km Regional  100-1000 km
Synoptic  km Supra-synoptic > km seasonal Forecasting - supra-synoptic scales So, it’s important to notice that, because of the strong relationship in Meteotology between space and time scale , the considered space scale are the large scales. So, consequently, Seasonal forecasting is dealing with the large space and time scales that is to say those of the mean state of the atmosphere which correspond to the scale of the general circulation.

16 Predictability Actors and Associated Scales
Just looking to some of the components of the climate system which bring predictability at the seasonal to interanual scales.

17 Predictability The different views of the Predictability
Through the observations Through the models

18 The evolution of external forcing conditions
Evolution of Sea Surface temperature (SST) Interannual variability (like ENSO) Decadal variability (like PDO) Evolution of continental surface conditions Influence of continental surface conditions (snow, albedo, ..), Intraseasonal variability (notably soil moisture), Mutual influences Decadal/ENSO ENSO/Intraseasonal Intraseasonal/Synoptic Some important points related to external forcings to be considered in seasonal forecasting. At present time one uses mainly the SST information, but in the next, notably with the improvment of the observation of the climate system (notably with reference to the next statellite generation), the evolution of continental conditions should be used in a more efficient and realistic way.

19 The ENSO The planetary influence of El Niño (left) and La Niña (right)
One of the major interannual external forcing change is the ENSO phenomenum. As explain in the ENSO presentation, it has a planetary influence through deep modifications of the General Atmospheric and Oceanic Circulations. The two extremes of this phenomenum are, on an oceanic point of view, La Niña and El Niño events. Be cautious with these figures, they give a tendancy to get specific features in El Niño or La Niña events but it’s not always the reality (e.g. South Africa and the 97 events ; the forecast was dry while the observations was FLOODS!!!). So, the right way to interpretate these maps is that the probability of the described patterns are larger than usually but the events are never certain.

20 The ENSO Through the observations in Winter
A quite simple (but efficient) way to look at the Nino/Nina influence. The 3 classes are defined using the tercile rainfall distribution. Roughly speaking, in Niña years the « Below Normal » classe is more frequent while it’s the « Above Normal » classe in Nina years. One can remark first the concerned regions change depending of the Nina vs Niño years and second e.g. for the Atlantic sector of Europe that the behaviour in Niña years is more homogeneous compared to the Nino’s one and consequently, we should be more confident in the forecast « Below Normal » in Nina years vs forecast « Above Normal » in Niño years.

21 The ENSO Through the observations in Summer
Again than noticeable signals which again are not symetric between El Niño and La Niña And are not necessarily over the same regions than in Wintertime.

22 The ENSO Through the observations in Winter
Same comments but with less signal in the winter rainfall compared to the temperature.

23 The ENSO Through the observations in Summer
Again but may be more signal for summer rainfall (western part for El Niño years and Northern part for La Niña years).

24 The fundamentals of seasonal Forecasting
The climatic variability The forecasting models Statistical models SST forced Atmospheric General Circulation Models Ocean/Atmosphere Coupled General Circulation Models The verifications Verification of the forecasts Verification of the usefulness of the forecats The chaos Link with the climatic variability Link with the ensemble forecast

25 The fundamentals of Seasonal Forecasting
The climatic variability : slow variation in the Atmosphere NAO PNA mode PDO QBO or TBO and Barnston and Livezey 1987, Mon. Wea. Rev., 115, ) East Atlantic (EA) , East Atlantic Jet (EA-Jet) , East Atlantic/Western Russia , Scandinavia (SCAND) , Polar/Eurasia Asian Summer , West Pacific (WP) , East Pacific (EP) , North Pacific (NP) , Tropical/Northern Hemisphere (TNH) , Pacific Transition (PT)

26 The « North Atlantic Oscillation»
The NAO looks like a dipole betwwen the subtropical regions (near the Azore Anticyclone) and the northern part of the Atlantic (near the Iceland region). The NAO Index is a measurement of the pressure meridionnal grandient over the eastern part of the Atlantic Ocean (normalized SLP in Lisbonne -Portugal- minus normalzed SLP in Stykkolmur-Iceland-). The NAO is mostly active in winter (DJFM) and the time serie of the NAO index shows a clear multidecadal variability (which is quite common for the Atlantic). It seems that the NAO is partly related to the internal dynamic of the atmopshere (so not predictable at the seasonal time scales) but (and fortunatelly for seasonal forecasting) partly related to SST forcings in the Atlantic. Lastly, the NAO and the Artic Oscilation are more or less the same phenomenun as the figures on the right show.

27 The « North Atlantic Oscillation»
Temperature in Winter There is a lot of very famous and interesting works on NAO and its relationship with weather and climate over Europe – Notably but not only from Hurell (see Web page : ) For these figures , it’s the correlaiton between the NAO index and temperature and rainfall in winter over the whole globe by H Cullen and M Visbek from the Lamon Doherty Earth Observatory – NOAA. Rainfall in Winter

28 The Pacific Decadal Oscillation
The PDO concern all the Pacific basin. On left the warm phase, on right the cold phase. Obviously the time serie shows the multidecadal variability.

29 The Pacific Decadal Oscillation
As a matter of evidence, the PDO is abble to modulate the ENSO events.

30 The climatic variability
Some atmospheric patterns which could be in relationship with European climate. For instance, the correlaiton between the EA Index and the winter temperature Anomalies can reach 0,7 which is quite interesting for seasonal forecasting purposes.

31 The climatic variability

32 The fundamentals of Seasonal Forecasting
The climate variability The annual rainfall over Sahelian regions is influenced by the ENSO variability (eg see ) which is superimposed on a multidecadal variability (notably in connection with the Atlantic SST variability). It’s a good example of the mixing of the different variability and consequently of the complexity of the seasonal forecasting.

33 The fundamentals of Seasonal Forecasting
The climatic variability Composite map of global SST drawn using dry years minus wet years for Sahelian regions. One can note that the Pacific signal is really noticeable (please, note also this way to proceed In order to adress the predictability at seasonal scales complementary to the first Composite maps presented at the beginning).

34 The fundamentals of seasonal Forecasting
The climatic variability Correlation between the Sahelian rainfall and the global SST. The correlation is very sample dependant in connection with the oceanic multidecadal variability. Particularly on can see the more or less correlation with the Pacific SST ; the value being more on the recent period. Please note the correlation over Indian Ocean which are not, at present time, used in sahelian statistical models but which could be introduced at a later time.

35 The fundamentals of seasonal Forecasting
The climatic variability Futur improvement of the knowledge of the Atlantic behaviour and consequently of it’s influence. Better description of the initial state in surface and subsurface conducting to a better coupled forecast. In this point of view the recent Pirata buoy network is really very usefull and should become the equivalent of the well knownTAO buoy network (over Equatorial Pacific). Atlantic “El Nino” – Pirata buoy network

36 The fundamentals of seasonal Forecasting
The climatic variability Some uncertainty : the role of the deep oceanic circulation and it’s relationship with the long time variability. Any way, one can see that it still exists some prespective in the futur about long range forecast at the interannual scales. JAS Observed Sahel Rainfall Vs JAS Observed THC index r = 0.45

37 The fundamentals of seasonal Forecasting
Forecasting models Statistical models SST forced Atmospheric Global Circulation Models Océan/Atmosphère Coupled General Circulation Models There is different way to forecast! All are interesting and have force and weakness

38 The fundamentals of seasonal Forecasting
The Statistical models East African Rainfall vs Nino3 Index Thank’s to Simon Mason

39 The fundamentals of seasonal Forecasting
The Statistical models East African Rainfall vs Nino3 Index The category transformation and the building of a contengency table.

40 The fundamentals of seasonal Forecasting
The Statistical models East African Rainfall vs Nino3 Index … which allow to evaluate the conditionnal probability. If there Is some predictability, we should get some preferential distributions Of the rainfall categories vs the SST categories. But, we have not yet informations that can be used for forecasting!

41 The fundamentals of seasonal Forecasting
The Statistical models First tool dedicated to categorical forecast : Discriminant Analysis. Here 2 predictors (giving informations from the SSTs).

42 The fundamentals of seasonal Forecasting
The Statistical models The most popular model is the linear one : Strong assumptions – gaussian pdf of the predictors for each categories of the predictand (here the rainfall) and within dispersions matrix the same gain For each rainfall categories. These assumptions allow to simplify the decisional model and lead to a linear model (equation of the red line). One can adjust some threshold which allow to formulate The decisional model (see figure) but interestingly taking into account in a quite easy Way economical consideration of the user of te forecast such as cost and lost evaluations. Aditionnal interesting parameters can be performe like conditionnal probabilities (Hit, Miss Correct Rejection, False Alarm) or the probability to be in a category when the value of the used predictor is observed.

43 The fundamentals of seasonal Forecasting
The Statistical models One can suppress the assumption on the within matrix and in that case The discriminant model becomes a quadratic one. Take care with such kind of model, one can get very strange results infered By the quadratic form of the discriminant surface and generally speaking, The more sophisticated the model is, the less robust are the results.

44 The fundamentals of seasonal Forecasting
The Statistical models No more comments on Multiple Regression (very popular model)! Take care with the number of predictors with respect to the sample size (in seasonal forecasting we have generally not so much data – once a year). Be sure to test the quality of the model on an independant file (learning Fiel vs Test file or paliative methods). Don’t forget operational considerations (namely availability of the predictors to preform the forecast)

45 The fundamentals of seasonal Forecasting
The Statistical models

46 The fundamentals of seasonal Forecasting
Numerical models All considerations integrated in the recent numerical models generally through the set of parametrisation and coupling technics. The truncation of the models is generally from T42 up to T106 Forced AGCM used just the surface informaiton from the Ocean while the coupled models used the whole information (from the surface and the sub-surface that is to say taking into account All the vertical dynamical and thermodynamical structure of the ocean).

47 The fundamentals of seasonal Forecasting
The numerical models Take care that the different models have different performances depending of the considered regions and years. So, it’s particularly important to get the evaluation of the model over a long time period in order to take these evaluation in the interpretation of the forecats of the model (particularly the confidence in the forecast). Obviously, these performance should be used in multimodel approach (eg weighting each model in mixing the different model simulations ; this in relationship with its score; the greater the score the larger the weight). For Africa, note that there is a lack of performance over the ZCIT; it seems that it’s a general behaviour of GCM (notably related to the convection scheme). Additionnally, note the behaviour over Australian regions wher one have some large signal with the ENSO; that is to say that we should not confuse the predictability and the performance of the model. Then, one can notice that evaluaiton are generally provided over the all period but that The performance generally depends on the forcing (namely the stronger is the forcing the better is the model) so it’s should be better to get the evaluation by large type of forcing events (e.g. El Niño, La Niña and Neutral conditions).

48 The fundamentals of seasonal Forecasting
Coupled vs Forced models On right pannel, one can clearly see that for a forecast lead more than 4 month, the predictability of the SST in NIno3 box is really improved when we use the oceanic subsurface data (here TAO data). The predictability stay at a reasonnable value until 4 month using the surface data ; this being related to the forecast lead used in forced SST AGCM.

49 Coupled vs Forced models
Generally speaking, it’s seems that coupled version compared to their forced version Are better for tropical regions and slighty less for mid-latitude regions.

50 The fundamentals of seasonal Forecasting
Forecast Verifications Verification in « real time » (following up of the bias, pointing out and monitoring of the errors, …), Verification in hindcast forecast mode , Verification of the predictability of forecasting events, Verification of the forecast value in a user’s point of view, Verification of the use and impact of the forecast, « Deterministic » vs « Probabilistic » Verifications Comparison with climatology and persistence (often use as references by users), … Problem of relevant and reliable dataset for verification purpose.

51 Score/Skill and Value 2 complementary point of view :
The scientific point of view : Quality of the forecast = Scores Interest of the use of the forecast = Skills The user point of view : Usefulness of the forecast = Skills (using current forecast strategy of the users – e.g. Climatology) Value of the forecast = Economonical evaluation of the use of the forecast (Cost/Lost approach) Schematically one try to answer to different point of view using these different Approaches : Scores : Are the forecasts comparable to the « observations »? Are the forecasts good? Skills : Are the forecasts better than a reference forecast strategy? Value : Is using these forecasts one can save some money?

52 Score/Skill and Value Score point of view :

53 Score/Skill and Value Event Event n11 n12 n21 n22 c Cost/Lost approach
2 categories : e.g. dryer / wetter and ratio Cost/Lost e.g. = 0.5 C1=Averaged cost using climatological forecast C2 =Averaged cost using perfect model forecast C3= Averaged cost using real model Event obs non obs forecast c non forcast L Event obs non obs forecast n11 n12 non forecast n21 n22 Just an example for the economical evaluation of the forecast. Generally used as a theoterical approach.

54 Score/Skill and Value Value point of view :
Please not the theoretical approach and the gain with the multi-model approach. The 0.43 event correpond with the upper tercile in the gaussian assumption.

55 The fundamentals of seasonal Forecasting
Verifications WMO Normes (parameters, scores, zones) The WMO has already proposed through notably the CBS a Standardised Verification System (SVS).

56 Verification in Hindcast mode
« retrospective » forecasts European research Projects PROVOST (CEPMMT, UKMO, EDF, LODYC, MPI, IMGA, DMI, U. Alcala) : Perfect Océan forecast, 4 different models, 15 years x 4 seasons x 9 membres Résults : clear in Tropics, some skill on North hemisphere in winter, but local uncertainties. Interest of the Multi-model approach. ELMASIFA (SNM Maroc, Algérie Tunisie) POTENTIALS (DMI, CINECA, LMD, MPI) DEMETER (CEPMMT, UKMO, LODYC, CERFACS, MPI, ADGB, IMGA, DMI, JRC, U. Liverpool, INM) : Forecasted Océan using coupled models, 6 different models, 40 years x 4 seasons x 6 month x 9 membres Résults : Provost revisited, improvment in Tropical regions and degradation in mid-latitude, extension of the range of the forecast. Please, note the crucial aspect of the retrospective forecasts in order to get , in a forecasting mode, the climatology of the behaviour of the model. You must use this information in building the forecast (see slide 74 up to 77) And without this information, you will not be abble to provide the forecast of an expected parameter (e.g. aridity index, hydrological budget, ….) even if the model is abble to compute it. So needs must be known by the modeling groups before the hindcasts forecasts in order to be sure that the relevant parameters will be properly performed and archived.

57 The fundamentals of seasonal Forecasting
Chaos and ensemble forecast Uncertainty Sources : Differences between analysis and real initial state Assimilation system Imperfection Lack of observations Model Errors (both Oceanic and/or atmospheric) Natural variability of the climate system Interpretation of the forecast Uncertainty sources conducting to the ensemble forecast. The butterfly effect! Note that the interpretation and the communication of the forecast are not really included in the butterfly effect …. Nevertheless … (e.g. Presao in Burkina-Faso, e.g. Daily Mirror page, …)!

58 The fundamentals of seasonal Forecasting
Chaos and ensemble forecast Model errors (Océanic and/or atmospheric) An illustration of the models error of the Arpege/OPA coupled models (from CERFACS) in the Nino3 box. One can remarks both the initial state errors and the different behaviour of the forecast. The figure shows also the continuous process and the update of the forecast for the Nino3 box for the 1997 Nino event.

59 Butterfly effect (JFM 2003)
Arpege-Climat – ensemble mean – 9 members Here we used the Operationnal ECMWF dissemination to get the atmospheric and SST initial conditions. Please focuses for instance on the northern coast of south America or the western part of the Pacific or ….

60 Butterfly effect (JFM 2003)
Same model, same atmospheric initial conditions, same SST …. And different forecast see on focused aera). The atmepheric and SST initial condiitons were get here through the Mars Archive System and this give some slight differences (typically in the range of 0.1/0.2 °C for temperatures) due to truncation and …

61 The fundamentals of seasonal Forecasting
Chaos and ensemble forecast To sample the initial state uncertainty  analysis disturbances Corresponding to the most unstable modes Compatible with analysis errors Methods : Singular vectors, breeding … To sample the modelisation uncertainty  model disturbances Using several models Using stochastic physics Modifying some physical parameters To sample all the possible solutions for the Ocean Atmosphere system The base of the ensemble forecast, some solutions.

62 The fundamentals of seasonal Forecasting
Chaos and climatic variability HOT & DRY COLD & WET One can interpretate the different and major states of the climatic system like attractors in the chaos theory. (see chaos presentaiton for details). But even on the same side of the attractor, one can have very different solutions!

63 The fundamentals of seasonal Forecasting
Chaos and Climate variability Close intial conditions can lead to very different solutions (e.g. red and purple curves) Be sure that the imagination of nature has no limits! And we must sample all these possible solutions!

64 The fundamentals of seasonal Forecasting
The forecast sytem The description of the initial state of the Ocean/Atmosphere system Atmospheric Data Oceanic Data Data from the continental surface Assimilation data scheme Elaboration of products Direct Methods (Deterministic vs Probablilistic products) Indirect Methods (notably PPM or MOS) Adaptation of the products (notably downscaling) Interpretation of the forecast Transformation of the forecast to the benefit of the user Following-up of the process Update of the forecast User’s Evaluation of the forecast (value, use and impact) Please, note that if short time period are only available for the model dataset, PPM are preferable compare to MOS which must have long time period model simulations in order to calibrate the statistical adaptation. In the former case, one can used some set of Reanalysis (or Analysis if any) assuming that the model behaviour in forecast mode will not be very different than in assimilation mode. Again for downscaling approach, a lot of technics can be used from statistical tools up to numerical tools. However, one have to be cautious and don’t forget the fundemantals of seasonal forecasting, namely it’s the mean state of the atmosphere which has some predictability at this time scale. So, consequently even if one use a LAM model with typically a mesh of 10 km, the downscaling will allow to represent the mean effects of the atmospheric general circulation at finer space scales (eg over a mountainous area) but not to predict finer space scale meteorological phenomenum (like thuderstorms, ….). The user’s evaluation is very important to adress. If the users’ decision models are not sensitive to the meteorological input the gain using the forecast will be very low even if the forecast is very accurate. In the reverse position, the user could gain a lot in unsing the forecast, even if the forecast is not very good. Note that this evaluation requests to have a formalisation of the use of the forecast by the user which is not so easy to get in concrete form!

65 The fundamentals of seasonal Forecasting
The forecasting suite Generally, the atmospheric conditions came from operational model analyses, the Surface Oceanic conditions came from observations (eg TAO network) and assimilation (notably satellite informations) or from oceanic climatology. Finally, the continental surface conditions came for the main part from climatology even if this could change in the next (notably with MSG products). But, the big scientific and oprationnal effort, at present time, is done for the oceanic assimilation. As seen in the Model presentation, all these elements give the boundary and initial conditions to the numerical model which is run using a big computer (here a VPP 5000) the final result being the global forecast over the whole globe.

66 The fundamentals of seasonal Forecasting
Description of the initial state of the Ocean/Atmosphere system Just to see the data coverage and the lack of observations partly conducting to the uncertainty of the initial state.

67 The fundamentals of seasonal Forecasting
Description of the initial state of the Ocean/Atmosphere system

68 The fundamentals of seasonal Forecasting
Description of the initial state of the Ocean/Atmosphere system

69 The fundamentals of seasonal Forecasting
The Oceanic data assimilation On right pannel, for a OACGCM (Arpege-OPA-OASIS), on can see the improvment of the simulation of the 97 Nino event using the assimilation of the Oceanic topography (Topex-Posseidon). The lower pannel shows the Pacific subsurface oceanic structure at the Equator as observed by the TAO network. The middle pannel shows the OAGCM reference simulation gotten using the assimilation of the observed SST. The upper pannel shows the same simulation when, additionnaly, we use the assimilation of the topography Satellite information. As a matter of evidence, the latter simulation is really better despite some discrepancies compared to the TAO observations.

70 The fundamentals of seasonal forecasting
Assimilation of the surface wind One combine in situ satellite measurements and modeling simulations in order to adress the oceanic repsonse to a wind forcing in the Equatorial Pacific. The surface wind is on one hand derived from SSMI observations and, on the other hand extracted from ECMWF operationnal analysis. Top pannel : stress wind differences between satellite measurements and ECMWF analysis. Middle pannel : The oceanic model give a better estimation of the La Nina event. The greater cooling obtained using the SSMI dataset is closer to the AVHRR measurment and buoy observations. Bottom pannel : the greater difference is observed in the Equatorial domain and under the surface despite the greater differences in the wind stress are out off the Equatorial waveguide (beyond 15°N and S). The difference (looking to the model heat budget) demonstrate that they are mainly due to vertical advection problems and overelevation of the thermocline.

71 The fundamentals of seasonal Forecasting
Description of the initial state of the Ocean/Atmosphere system As already noticed, there is a lack of observation for continental conditions. Even if the new GCM used refine Soil/Vegetation Scheme the data used are generally provided by climatologies instead of observations. This can be quite different for albedo conditions for instance in relationship with the vegetation growing. The soil moisture or snow cover are also some important parameters to be highlighted.

72 The fundamentals of seasonal forecasting
Elaboration of products Direct Methods (deterministic vs Probabilistic products) Indirect Methods (Statistical adaptations notably) Adaptation of the products (downscaling)

73 The fundamentals of seasonal forecasting
Elaboration of products Deterministic vs Probabilistic product Is the product usefull? Is the product well adapted to the user? If no, can we do something? Can the quality, the value and the use of the product be checked and verified? Require a Forecaster/User discussion through Multidisciplinary groups (Experts in forecasting, communication, users’s needs)

74 Elaboration of numerical products
Direct Methods (deterministic and probabilistic products) formulation as Indices or Anomalies Model Forecast : raw information not usefull ! F is the raw forecast

75 Elaboration of numerical products
Direct Methods (deterministic and probabilistic products) formulation as Indices or Anomalies Debiased model forecast : better formulation Better formulation taking into account the bias of the forecasting suite

76 Elaboration of numerical products
Direct Methods (deterministic and probabilistic products) formulation as Indices or Anomalies Normalized model forecast : Model forecats compared to its own climatology The right formulation taking into account the climatology of the forecasting model. Generally, the spread of the forecast is less than the spread of the atmosphere ; so it’s very important do divide by the climatological standard deviation of the forecast (that is to say on a long period of reference – 15 years for Provost and 40 years for Demeter). Note that if we want to perform a multimodel forecast it’s the only formulation which allow to mixe different forecasts from different models.

77 Elaboration of numerical products
Direct Methods (deterministic and probabilistic products) formulation as Indices or Anomalies Anomalies : Adaptation to « local » observation properties Indices : Model forecats compared to its own climatology Generally the user would like to use quantitative value of the parameter instead of adimentionnal value (Indice). So, one have to transform the previous indice in quantitative forecast using an observed climatology. Take care with the climatology, they can be quite different depending of the sources, the way to mixe the data, the used periods, the used models (for Reanalyses), ….

78 Downscaling Problem G l o b a l
Seasonal predictability and associated scales  adaptation to the user G l o b a l L o c a l One has to don’t forget that the predictability is infered by the time average and that we are dealin,g with the mean state of the atmosphere. So, if we want to have chance to succeed in downscaling we must verify first That at the finner scale, it still exists a mean effect, even at the local scale, which is large scale forced (e.g. orographical effect in tropical islands). If this mean effect doesn’t exist, we will get just the picture of the tool we will use! statistical methods : Observations, Downscaling models numerical methods : Numerical models using GCM simulations as boundary conditions (single column, LAM, …)

79 Dowscaling Problem Seasonal
Seasonal predictability and associated scales  adaptation to the user Seasonal intra seasonal Some works had been already achieved in tropical regions for the Intraseasonal variabilit and predictability. So, some products are already possible giving some hight or low probability periods of rainfall within the season. However, some improvments must be achieved to get more accurate intraseasonal forecasts. Take care to don’t confuse monthly forecasts and seasonal forecasts, the Scientific Principles are not exactly the sames and for instance, for monthly forecasts, the surface assimilation parameters is crucial (notably in comparison with seasonal forecasts over Continental regions), the spin-up of the model must be adressed, …. statistical methods : Observations, Downscaling models numerical methods : Numerical models using GCM simulations as boundary conditions (single column, LAM, …)

80 The fundamentals of seasonal forecasting
Interpretation of the forecast transformation to the benefit of the user Translation in terms of actions, risks, scenario, … and associated probabilities Following_up of the process Update of the forecast – continuous process Evaluation on a user point of view – Processs Experience Feedbacks How looking at the forecast one can interpretate it in terms of action or decision for the user? Note the necesssity of Multidisplinary groups in this context. This is a real and big challenge!

81 Highlights of seasonal forecasting
Basically Probabilistic forecast, Forecast of the Mean State and not of the Weather, Not usefull elsewhere neither for everything, Confidence in the forecast depending of the year and the parameter, Evaluation of both aspects quality (scientific) and usefulness (economical value, use), Usefull in a decision making context and in meteo sensitive activities (in an economical sense), Since a few years better knowledge of the limits of the seaonnal predictability and its potential uses, Operational forecast systems aiming to provide targeted products, Improvements : ERA40, coupled models, donwscalling, intraseasonal forecasts, Observation system, …


Download ppt "The fundamentals of Seasonal Forecasting"

Similar presentations


Ads by Google