ECMWF User Meeting 14-16 June 2006 The use of ECMWF ensemble and lagged deterministic forecasts for 3-30 day outlooks in Sweden 1.Monthly instead of seasonal.

Slides:



Advertisements
Similar presentations
How Well Forecast Were the 2004 and 2005 Atlantic and U. S
Advertisements

Climate Prediction Applications Science Workshop
UNITED NATIONS Shipment Details Report – January 2006.
Sales Forecasting using Dynamic Bayesian Networks Steve Djajasaputra SNN Nijmegen The Netherlands.
February 7, 2002 A brief review of Linear Algebra Linear Programming Models Handouts: Lecture Notes.
6 - 1 Copyright © 2002 by Harcourt, Inc All rights reserved. CHAPTER 6 Risk and Return: The Basics Basic return concepts Basic risk concepts Stand-alone.
Rossana Dragani Using and evaluating PROMOTE services at ECMWF PROMOTE User Meeting Nice, 16 March 2009.
THOR Annual Meeting - Bergen 9-11 November /25 On the impact of initial conditions relative to external forcing on the skill of decadal predictions:
Norwegian Meteorological Institute met.no LAMEPS – Limited area ensemble forecasting in Norway, using targeted EPS Marit Helene Jensen, Inger-Lise Frogner,
Federal Department of Home Affairs FDHA Federal Office of Meteorology and Climatology MeteoSwiss Extended range forecasts at MeteoSwiss: User experience.
Page 1© Crown copyright 2004 Seasonal forecasting activities at the Met Office Long-range Forecasting Group, Hadley Centre Presenter: Richard Graham ECMWF.
Norwegian Meteorological Institute met.no TEPS/LAMEPS at met.no Marit Helene Jensen, Inger-Lise Frogner, Hilde Haakenstad and Ole Vignes.
HB 1 Forecast Products Users'Meeting, June 2005 Users meeting Summary Performance of the Forecasting System (1) Main (deterministic) model -Outstanding.
User experience with extended range forecasts -- climatic aspects of ECMWF products Christof Appenzeller Wolfgang Müller Heike Kunz Mark Liniger ERA-40.
The Wave Model ECMWF, Reading, UK.
ECMWF Slide 1Met Op training course – Reading, March 2004 Forecast verification: probabilistic aspects Anna Ghelli, ECMWF.
Slide 1ECMWF forecast User Meeting -- Reading, June 2006 Verification of weather parameters Anna Ghelli, ECMWF.
User Meeting 15 June 2005 Monthly Forecasting Frederic Vitart ECMWF, Reading, UK.
Slide 1ECMWF forecast products users meeting – Reading, June 2005 Verification of weather parameters Anna Ghelli, ECMWF.
The COSMO-LEPS system at ECMWF
Seasonal forecasts Laura Ferranti and the Seasonal Forecast Section User meeting June 2005.
Slide 1 Forecast Products User Meeting June 2006 Slide 1 ECMWF medium-range forecasts and products David Richardson Met Ops.
Use of Medium and Extended Range Forecasts in Slovenia Jure Cedilnik ARSO [EARS – Environmental Agency of Slovenia, Met. service]
Page 1 © Crown copyright 2005 ECMWF User Meeting, June 2006 Developments in the Use of Short and Medium-Range Ensembles at the Met Office Ken Mylne.
Severe Weather Forecasts
ECMWF Training Course 2005 slide 1 Forecast sensitivity to Observation Carla Cardinali.
Slide 1The Wave Model - Last part Validation of wind & wave analysis using satellite & buoy. Altimeters onboard ERS-1/2, ENVISAT and Jason Quality is monitored.
1 NEXTOR Monitoring and Modeling NAS Performance at the Daily Level Mark Hansen Performance Metrics TIM May 2002.
Sub-seasonal to seasonal prediction David Anderson.
Lecture 2 ANALYSIS OF VARIANCE: AN INTRODUCTION
CS1512 Foundations of Computing Science 2 Week 3 (CSD week 32) Probability © J R W Hunter, 2006, K van Deemter 2007.
1 Contact details Colin Gray Room S16 (occasionally) address: Telephone: (27) 2233 Dont hesitate to get in touch.
The role of the stratosphere in extended- range forecasting Thomas Jung Alfred Wegener Institute Helmholtz Centre for Polar and Marine Research Germany.
1 NCAS SMA presentation 14/15 September 2004 The August 2002 European floods: atmospheric teleconnections and mechanisms Mike Blackburn (1), Brian Hoskins.
Robin Hogan Ewan OConnor Damian Wilson Malcolm Brooks Evaluation statistics of cloud fraction and water content.
NCAS Conference December 2007, Park Inn Hotel, York The Indian monsoon and climate change Andrew Turner, Julia Slingo.
Climate Prediction Division Japan Meteorological Agency Developments for Climate Services at Japan Meteorological Agency 1.
Chapter 7 Sampling and Sampling Distributions
1 Understanding Multiyear Estimates from the American Community Survey.
Chapter 13 – Weather Analysis and Forecasting
Richmond House, Liverpool (1) 26 th January 2004.
Module 4. Forecasting MGS3100.
Elementary Statistics
(This presentation may be used for instructional purposes)
Page 1 NAE 4DVAR Oct 2006 © Crown copyright 2006 Mark Naylor Data Assimilation, NWP NAE 4D-Var – Testing and Issues EWGLAM/SRNWP meeting Zurich 9 th -12.
EU Market Situation for Eggs and Poultry Management Committee 21 June 2012.
1 Slides revised The overwhelming majority of samples of n from a population of N can stand-in for the population.
指導教授:李錫堤 教授 學生:邱奕勛 報告日期:
Discrete Event (time) Simulation Kenneth.
Oil & Gas Final Sample Analysis April 27, Background Information TXU ED provided a list of ESI IDs with SIC codes indicating Oil & Gas (8,583)
VOORBLAD.
Equations of Lines Equations of Lines
Risk and Return Learning Module.
Meteorological Training Course, 20 March /25 Using Combined Prediction Systems (CPS) for wind energy applications European Centre for Medium-Range.
Slide 1 October 2011 Verification for polar regions  Scores computed for polewards of 65°  NB proposed for CBS is polewards of 60°  Verification at.
Model and Relationships 6 M 1 M M M M M M M M M M M M M M M M
Institut für Physik der Atmosphäre Institut für Physik der Atmosphäre Object-Oriented Best Member Selection in a Regional Ensemble Forecasting System Christian.
Deutscher Wetterdienst Zentrale Vorhersage DWD Workshop "Use and Verification of LEPS products", Geneve, May, Verification of LEPS products.
LRF Training, Belgrade 13 th - 16 th November 2013 © ECMWF Sources of predictability and error in ECMWF long range forecasts Tim Stockdale European Centre.
ECMWF long range forecast systems
Willem A. Landman & Francois Engelbrecht.  Nowcasting: A description of current weather parameters and 0 to 2 hours’ description of forecast weather.
Predictability and Chaos EPS and Probability Forecasting.
A Regression Model for Ensemble Forecasts David Unger Climate Prediction Center.
Eidgenössisches Departement des Innern EDI Bundesamt für Meteorologie und Klimatologie MeteoSchweiz Statistical Characteristics of High- Resolution COSMO.
NOAA’s Seasonal Hurricane Forecasts: Climate factors influencing the 2006 season and a look ahead for Eric Blake / Richard Pasch / Chris Landsea(NHC)
Short-Range Ensemble Prediction System at INM García-Moya, J.A., Santos, C., Escribà, P.A., Santos, D., Callado, A., Simarro, J. (NWPD, INM, SPAIN) 2nd.
Figures from “The ECMWF Ensemble Prediction System”
Ryan Kang, Wee Leng Tan, Thea Turkington, Raizan Rahmat
Presentation transcript:

ECMWF User Meeting June 2006 The use of ECMWF ensemble and lagged deterministic forecasts for 3-30 day outlooks in Sweden 1.Monthly instead of seasonal forecasting 2.The used of lagged forecasts (as a complement to the EPS) 3.Problems with weighting together different forecast systems For details see : forecast_products_user/Presentations2006/index.htm

ECMWF User Meeting June The seasonal forecasts Not used, partly because the forecasts seem to repeat themselves

ECMWF User Meeting June 2006 Warmer than normal The last four years ECMWF summer forecasts (issued in April) Warmer than normal Warmer than normal Warmer than normal Colder than normal Not warmer than normal Colder than normal Colder than normal

ECMWF User Meeting June The monthly forecast Used and found skilful, but tendencies of jumpiness in transitional periods

ECMWF User Meeting June April 27 April 2-3 week forecast of negative anomalies changed into 1-2 week forecast of positive Climate value Climate value

ECMWF User Meeting June 2006 Max +5 to 10 Max +10 to 15 Jumpiness experienced at a specific location 850 hPa temperature plume for Norrköping, southern Sweden clim

ECMWF User Meeting June 2006 Forecast for week June 2006Forecast for week June 2006 From 1June From 8June A very recent example of jumpiness (2m temp anom)

ECMWF User Meeting June The 21 day forecast (+9 days) Using the last three days 21 d forecasts enables us to inder the trends beyond day 10, even beyond day 15 For details see : forecast_products_user/Presentations2006/index.htm

ECMWF User Meeting June d20 d 30 d normal temperature statistics Last days ECMWF fc and EPS Last days Control +21 d forecasts Main method since summer 2003 ECMWF monthly forecast

ECMWF User Meeting June 2006 Consecutive daily 18-day lagged average forecast of 850 hPa temperature forecasts made from eight member averages of 21-day forecasts Already in in mid-April the lagged system provide hints about the temperature in early May 15 April15 May verif

ECMWF User Meeting June The problem of verification For details see : forecast_products_user/Presentations2006/index.htm

ECMWF User Meeting June 2006 Level of useful forecasts Introduction of more ECMWF data

ECMWF User Meeting June 2006 What to do? Two ways to go: 1.Political (cover up, play illusionist tricks or change the norms) 2.Scientific (go to the roots of the problem)

ECMWF User Meeting June 2006 ACC=98% Slope=0.8 I just happen to have some fresh verifications here, depicting the results during the first half of this year... Political trick: Selective sampling

ECMWF User Meeting June 2006 Anomaly correlation of monthly forecast for Stockholm (2 m temperature) More ECMWF input Verif Prog But it didnt look that bad….

ECMWF User Meeting June 2006 Scientific approach: The conventional verification disregarded three factors: 1. Variable range of variation between 2002 and More than one verification method should be used 3.Twelve forecasts per year is a too small sample For details see : forecast_products_user/Presentations2006/index.htm

ECMWF User Meeting June Lower correlation Smaller errors 2002 Higher correlation Larger errors

ECMWF User Meeting June 2006 f-a f-c a-c The RMSE in vector form yields angles as correlation measures β a f c ACC =cosβ

ECMWF User Meeting June 2006 a f-a f-c c Large variability high correlation (small β) but large errors β f a-c β a f Small variability low correlation (high β) and small errors f-a

ECMWF User Meeting June 2006 Introduction of more ECMWF data reduced the errors! Another verification method RMSE MABSE

ECMWF User Meeting June 2006 Verifying two years at a time (Lagged) verification over 24 months compared to over 12 months 12 months 24 months

ECMWF User Meeting June 2006 Mid-2006 Continued progress

ECMWF User Meeting June Swedish concerns about the quality of the centres EPS 1. Forecasters at SMHI and the Air Force do not find much use of the deterministic EPS compared to an elaborate use of the deterministic model 2. The scientists at SMHI and the MISU (Univ. Stockholm) are critical about the perturbations + (recently) the stochastic physics 3. My impression is not that the EPS is bad or has become worse, but has had problems to keep pace with the improvements of the deterministic model

ECMWF User Meeting June The size of the T42 EPS perturbations is very large The picture depicts the status before 1 February Since then the resolution of the deterministic system has increased by 50%, but the EPS perturbations which remain at their 1995 level of T42

ECMWF User Meeting June 2006 Before 2001 there was little quality difference between perturbed and non-perturbed forecasts, amounting beyond D+5 to an ACC difference. Since then it has increased to 10-15% ? Difference in ACC between the unperturbed Control and a randomly selected EPS member

ECMWF User Meeting June Over spreading in during the first hours made it difficult to use the EPS as BC for the HIRLAM 2.In cases of extreme or interesting events the signals often come 1-2 earlier in the T799 lagged system 3.In cases of consistent and skilful T799 performance the EPS keep the forecaster uncertain too long For more details see presentation at the OD Workshop November 2005

ECMWF User Meeting June 2006 The EPS perturbations make the forecasts 1 ½ days worse than Control! 1.5 days Unperturbed Control perturbed members The RMSE of individual EPS members The 2 m temperature forecasts for London winter Lagged EPS mean

ECMWF User Meeting June 2006 The RMSE of individual EPS members The 2 m temperature forecasts for London Feb-April 2006 perturbed members 1 day EPS mean Lagged Unperturbed Control

ECMWF User Meeting June 2006 Figure 2.1: Schematic image of the RMS error of the ensemble members, ensemble mean, and control forecast as a function of lead- time. The asymptotic predictability range is defined as the average difference between two randomly chosen atmospheric states. In a perfect ensemble system the RMS error of the ensemble members is a factor larger than the RMS error of the ensemble mean. Courtesy, L. Bengtsson, MISU climate RMSE(pert member)= (=sqrt2) RMSE (ensemble mean) Perturbed member Ensemble mean Control Tim Palmers Law

ECMWF User Meeting June 2006 Figure 4.2: Comparison of RMS error of the ensemble mean (green), the ensemble members (blue), the control forecast for the EPS (red) as well as the deterministic forecast as a function of lead-time (light blue). This comparison is made globally for the periods DJF (a) as well as JJA (d), and for regions 1 and 2 described in the text for the same time periods (plots b, c, e and f). The RMS errors are averaged over the globe and over the periods DJF and JJA. Courtesy L. Bengtsson, MISU

ECMWF User Meeting June Use of the last T799 runs forming lagged ensembles (work under development)

ECMWF User Meeting June 2006 EPS Mean and Lagged Mean 24 March 00 UTC + 84h

ECMWF User Meeting June 2006 Valid Sun 9 April 12 UTC EPS Mean and lagged ECMWF T799 4 April 00 UTC + 132h

ECMWF User Meeting June The public 6-10 day forecasts Once a week, four out of five forecasts verify

ECMWF User Meeting June day forecast presented on TV 26 January 2006

ECMWF User Meeting June 2006

Epsogram for Stockholm

ECMWF User Meeting June When does it pay to weight together forecasts? For details see : forecast_products_user/Presentations2006/index.htm E 1 < E 2

ECMWF User Meeting June 2006 f-a g-a a Ensemble mean error least for uncorrelated (orthogonal) errors f g β

ECMWF User Meeting June 2006 g f-a g-a a Correlated but equal errors β f

ECMWF User Meeting June 2006 g f-a g-a a Correlated, but non-equal errors Ensemble mean is not the optimal solution β f

ECMWF User Meeting June 2006 g f-a g-a a Correlated, but non-equal errors Weighted ensemble mean minimizes the error β f 90º

ECMWF User Meeting June 2006 a The weighted combination of two rather uncorrelated models (f 1 and g 1 ) can yield better forecast than the combination of two better, but correlated models (f 2 and g 2 ) When averaging orthogonality might compensate for lack of skill g1g1 f1f1 f2f2 g2g2

ECMWF User Meeting June 2006 g f-a g-a a At some stage any weighting will not improve the forecasts β f

ECMWF User Meeting June 2006 E 1 =f-a E 2 =g-a a Breaking point: when the fraction between the errors of the two systems equals the error correlation f g β

ECMWF User Meeting June How should x and y, the weights, be calculated taking the forecast error correlation into account? For details see : forecast_products_user/Presentations2006/index.htm E 1 < E 2

ECMWF User Meeting June 2006 Certain combinations of forecasts will not yield an improved weighted mean E 1 < E 2

ECMWF User Meeting June 2006 x 2 +y 2 b2b2 f-a g-a a f g β x y Pythagoras' Theorem not valid for the triangle a2a2 m β

ECMWF User Meeting June 2006 y2y2 m2m2 b2 β But Pythagoras' Theorem is valid for this right- angled triangle b 2 =y 2 +m 2 β

ECMWF User Meeting June 2006 a2a2 m2m2 β x2x2 …and for this right- angled triangle a 2 =x 2 +m 2 β

ECMWF User Meeting June 2006 y2y2 β m y a 2 -b 2 =x 2 -y 2 β a2a2 b2 x2x2

ECMWF User Meeting June 2006 A B β y2 x2x2 β b a 90º

ECMWF User Meeting June 2006 A B β y2 x2x2 β b a

ECMWF User Meeting June 2006 A B β y2 x2x2 β b a bcosβ acosβ

ECMWF User Meeting June 2006 abcos β b2b2 β (x+y) 2 a 2 The Cosine Theorem: (x+y) 2 = a 2 + b 2 - 2abcosβ

ECMWF User Meeting June 2006 a b x y β The Cosine Theorem: (x+y) 2 = a 2 + b 2 - 2abcosβ combined with a 2 - b 2 = x 2 - y 2 yields m

ECMWF User Meeting June 2006 The Cosine Theorem: (x+y) 2 = a 2 + b 2 - 2abcosβ combined with a 2 - b 2 = x 2 - y 2 yields a 2 -b 2 =x 2 -y 2 =(x+y) 2 -2y 2 -2xy a 2 -b 2 =a 2 +b 2 -2abcos(β) -2y 2 -2xy b 2 -a 2 =y 2 -x 2 =(x+y) 2 -2x 2 -2xy b 2 -a 2 =a 2 +b 2 -2abcos(β) -2x 2 -2xy 2b 2 =2abcos(β)+2y 2 +2xy b 2 -abcos(β)=y(x+y) 2a 2 =2abcos(β)+2x 2 +2xy a 2 -abcos(β)=x(x+y)

ECMWF User Meeting June 2006 E1E1 E2E2 a f g x y m cor(E1,E2) …and by replacing a and b with E 1 and E 2, the errors of the two systems and cos(β) with the correlation between E 1 and E 2 we have:

ECMWF User Meeting June 2006 E22E22 a f g y x E12E12 …which for uncorrelated errors boils down to …or the more familiar

ECMWF User Meeting June 2006 Hypothetical error correlations 50% 100% 0% D+0D+15 T799 vs T399 T799(T399) vs UKMO or an arbitrary eps-member If all three models have the same error magnitude and correlation then the weights are 33.3% But if the errors of T799 and T399 are more correlated than the errors of T799 (T399) versus UKMO the UKMO should be weighted the most Extension to three models??

ECMWF User Meeting June Future challenges Extending the monthly forecasts by including precipitation and provide forecasts separately form week1, week2 and week34 - and much more….