• Blender, R., , U. Luksch, , K. Fraedrich, , and C. C. Raible, 2003: Predictability study of the observed and simulated European climate using linear regression. Quart. J. Roy. Meteor. Soc., 129, 22992313.

    • Search Google Scholar
    • Export Citation
  • Buizza, R., , and A. Hollingsworth, 2002: Storm prediction over Europe using the ECMWF Ensemble Prediction System. Meteor. Appl., 9, 289305.

    • Search Google Scholar
    • Export Citation
  • Cassou, C., , C. Deser, , L. Terray, , J. W. Hurrell, , and M. Drévillon, 2004: Summer sea surface temperature conditions in the North Atlantic and their impact upon the atmospheric circulation in early winter. J. Climate, 17, 33493363.

    • Search Google Scholar
    • Export Citation
  • Compo, G. P., , and P. D. Sardeshmukh, 2004: Storm track predictability on seasonal and decadal scales. J. Climate, 17, 37013720.

  • Conil, S., , H. Douville, , and S. Tyteca, 2009: Contribution of realistic soil moisture initial conditions to boreal summer climate predictability. Climate Dyn., 32, 7593.

    • Search Google Scholar
    • Export Citation
  • Czaja, A., , and C. Frankignoul, 2002: Observed impact of Atlantic SST anomalies on the North Atlantic oscillation. J. Climate, 15, 606623.

    • Search Google Scholar
    • Export Citation
  • Della-Marta, P. M., , M. A. Liniger, , C. Appenzeller, , D. N. Bresch, , P. Köllner-Heck, , and V. Muccione, 2010: Improved estimates of the European winter windstorm climate and the risk of reinsurance loss using climate model data. J. Appl. Meteor. Climatol., 49, 20922120.

    • Search Google Scholar
    • Export Citation
  • DelSole, T., , and J. Shukla, 2010: Model fidelity versus skill in seasonal forecasting. J. Climate, 23, 47944806.

  • Donat, M. G., , G. C. Leckebusch, , J. G. Pinto, , and U. Ulbrich, 2010: Examination of wind storms over Central Europe with respect to circulation weather types and NAO phases. Int. J. Climatol., 30, 12891300, doi:10.1002/joc.1982.

    • Search Google Scholar
    • Export Citation
  • Douville, H., 2009: Stratospheric polar vortex influence on Northern Hemisphere winter climate variability. Geophys. Res. Lett., 36, L18703, doi:10.1029/2009GL039334.

    • Search Google Scholar
    • Export Citation
  • Fischer, E. M., , S. I. Seneviratne, , P. L. Vidale, , D. Lüthi, , and C. Schär, 2007: Soil moisture–atmosphere interactions during the 2003 European summer heat wave. J. Climate, 20, 50815099.

    • Search Google Scholar
    • Export Citation
  • Fletcher, C. G., , S. C. Hardiman, , P. J. Kushner, , and J. Cohen, 2009: The dynamical response to snow cover perturbations in a large ensemble of atmospheric GCM integrations. J. Climate, 22, 12081222.

    • Search Google Scholar
    • Export Citation
  • García-Morales, M. B., , and L. Dubus, 2007: Forecasting precipitation for hydroelectric power management: How to exploit GCM’s seasonal ensemble forecasts. Int. J. Climatol., 27, 16911705.

    • Search Google Scholar
    • Export Citation
  • Goddard, L., , S. J. Mason, , S. E. Zebiak, , C. F. Ropelewski, , R. Basher, , and M. A. Cane, 2001: Current approaches to seasonal-to-interannual climate predictions. Int. J. Climatol., 21, 11111152.

    • Search Google Scholar
    • Export Citation
  • Hagedorn, R., , F. J. Doblas-Reyes, , and T. N. Palmer, 2005: The rationale behind the success of multi-model ensembles in seasonal forecasting—I. Basic concept. Tellus, 57A, 219233.

    • Search Google Scholar
    • Export Citation
  • Harrison, M., 2005: The development of seasonal and inter-annual climate forecasting. Climatic Change, 70, 201220.

  • Ineson, S., , and A. A. Scaife, 2009: The role of the stratosphere in the European climate response to El Niño. Nat. Geosci., 2, 3236.

    • Search Google Scholar
    • Export Citation
  • Jin, E. K., and Coauthors, 2008: Current status of ENSO prediction skill in coupled ocean–atmosphere models. Climate Dyn., 31, 647664, doi:10.1007/s00382-008-0397-3.

    • Search Google Scholar
    • Export Citation
  • Johansson, A., 2007: Prediction skill of the NAO and PNA from daily to seasonal time scales. J. Climate, 20, 19571975.

  • Jones, P. W., 1999: First- and second-order conservative remapping schemes for grids in spherical coordinates. Mon. Wea. Rev., 127, 22042210.

    • Search Google Scholar
    • Export Citation
  • Kharin, V. V., , and F. W. Zwiers, 2003: On the ROC score of probability forecasts. J. Climate, 16, 41454150.

  • Klawa, M., , and U. Ulbrich, 2003: A model for the estimation of storm losses and the identification of severe winter storms in Germany. Nat. Hazards Earth Syst. Sci., 3, 725732.

    • Search Google Scholar
    • Export Citation
  • Knippertz, P., , U. Ulbrich, , F. Marques, , and J. Corte-Real, 2003: Decadal changes in the link between El Niño and springtime North Atlantic oscillation and European–North African rainfall. Int. J. Climatol., 23, 12931311.

    • Search Google Scholar
    • Export Citation
  • Kushnir, Y., , W. A. Robinson, , P. Chang, , and A. W. Robertson, 2006: The physical basis for predicting Atlantic sector seasonal-to-interannual climate variability. J. Climate, 19, 59495970.

    • Search Google Scholar
    • Export Citation
  • Leckebusch, G. C., , U. Ulbrich, , L. Fröhlich, , and J. G. Pinto, 2007: Property loss potentials for European midlatitude storms in a changing climate. Geophys. Res. Lett., 34, L05703, doi:10.1029/2006GL027663.

    • Search Google Scholar
    • Export Citation
  • Leckebusch, G. C., , M. Donat, , U. Ulbrich, , and J. G. Pinto, 2008a: Midlatitude cyclones and storms in an ensemble of European AOGCMs under ACC. CLIVAR Exchanges, No. 13, International CLIVAR Project Office, Southampton, United Kingdom, 3–5.

    • Search Google Scholar
    • Export Citation
  • Leckebusch, G. C., , D. Renggli, , and U. Ulbrich, 2008b: Development and application of an objective storm severity measure for the northeast Atlantic region. Meteor. Z., 17, 575587.

    • Search Google Scholar
    • Export Citation
  • Lee, J.-Y., and Coauthors, 2010: How are seasonal prediction skills related to models’ performance on mean state and annual cycle? Climate Dyn., 35, 267283.

    • Search Google Scholar
    • Export Citation
  • Luksch, U., , C. C. Raible, , R. Blender, , and K. Fraedrich, 2005: Decadal cyclone variability in the North Atlantic. Meteor. Z., 14, 747753.

    • Search Google Scholar
    • Export Citation
  • Mason, S. J., , and N. E. Graham, 1999: Conditional probabilities, relative operating characteristics, and relative operating levels. Wea. Forecasting, 14, 713725.

    • Search Google Scholar
    • Export Citation
  • Mason, S. J., , and N. E. Graham, 2002: Areas beneath the relative operating characteristics (ROC) and relative operating levels (ROL) curves: Statistical significance and interpretation. Quart. J. Roy. Meteor. Soc., 128, 21452166.

    • Search Google Scholar
    • Export Citation
  • Mathieu, P.-P., , R. T. Sutton, , B. Dong, , and M. Collins, 2004: Predictability of winter climate over the North Atlantic European region during ENSO events. J. Climate, 17, 19531974.

    • Search Google Scholar
    • Export Citation
  • Müller, W. A., , C. Appenzeller, , and C. Schär, 2005: Probabilistic seasonal prediction of the winter North Atlantic Oscillation and its impact on near surface temperature. Climate Dyn., 24, 213226.

    • Search Google Scholar
    • Export Citation
  • Munich Re, 2007: Zwischen Hoch und Tief—Wetterrisiken in Mitteleuropa. Munich Re, Order Number 302–05481, 57 pp.

  • Munich Re, NATCATSERVICE cited 2010: Significant natural disasters since 1980. [Available online at http://www.munichre.com/en/reinsurance/business/non-life/georisks/natcatservice/significant_natural_catastrophes.aspx.]

    • Search Google Scholar
    • Export Citation
  • Murnane, R. J., 2004: Climate research and reinsurance. Bull. Amer. Meteor. Soc., 85, 697707.

  • Murnane, R. J., , M. Crowe, , A. Eustis, , S. Howard, , J. Koepsell, , R. Leffler, , and R. Livezey, 2002: The weather risk management industry’s climate forecast and data needs. Bull. Amer. Meteor. Soc., 83, 11931198.

    • Search Google Scholar
    • Export Citation
  • Orsolini, Y. J., , and N. G. Kvamstø, 2009: Role of Eurasian snow cover in wintertime circulation: Decadal simulations forced with satellite observations. J. Geophys. Res., 114, D19108, doi:10.1029/2009JD012253.

    • Search Google Scholar
    • Export Citation
  • Palmer, T. N., and Coauthors, 2004: Development of a European Multimodel Ensemble System for Seasonal-to-Interannual Prediction (DEMETER). Bull. Amer. Meteor. Soc., 85, 853872.

    • Search Google Scholar
    • Export Citation
  • Palutikof, J. P., , T. Holt, , and T. J. Osborn, 2002: Seasonal forecasting of strong winds over Europe. Extended Abstracts, 16th Conf. on Probability and Statistics in the Atmospheric Sciences, Orlando, FL, Amer. Meteor. Soc., J3.12. [Available online at http://ams.confex.com/ams/pdfpapers/30343.pdf.]

    • Search Google Scholar
    • Export Citation
  • Pinto, J. G., , E. L. Fröhlich, , G. C. Leckebusch, , and U. Ulbrich, 2007: Changing European storm loss potentials under modified climate conditions according to ensemble simulations of the ECHAM5/MPI-OM1 GCM. Nat. Hazards Earth Syst. Sci., 7, 165175.

    • Search Google Scholar
    • Export Citation
  • Pinto, J. G., , S. Zacharias, , A. H. Fink, , G. C. Leckebusch, , and U. Ulbrich, 2009: Factors contributing to the development of extreme North Atlantic cyclones and their relationship with the NAO. Climate Dyn., 32, 711737, doi:10.1007/s00382-008-0396-4.

    • Search Google Scholar
    • Export Citation
  • Pinto, J. G., , M. Reyers, , and U. Ulbrich, 2011: The variable link between PNA and NAO in observations and in multi-century CGCM simulations. Climate Dyn., 36, 337354, doi:10.1007/s00382-010-0770-x.

    • Search Google Scholar
    • Export Citation
  • Qian, B. D., , and M. A. Saunders, 2003: Seasonal predictability of wintertime storminess over the North Atlantic. Geophys. Res. Lett., 30, 1698, doi:10.1029/2003GL017401.

    • Search Google Scholar
    • Export Citation
  • Rodwell, M. J., , and C. K. Folland, 2002: Atlantic air–sea interaction and seasonal predictability. Quart. J. Roy. Meteor. Soc., 128, 14131443.

    • Search Google Scholar
    • Export Citation
  • Rodwell, M. J., , and F. J. Doblas-Reyes, 2006: Medium-range, monthly, and seasonal prediction for Europe and the use of forecast information. J. Climate, 19, 60256046.

    • Search Google Scholar
    • Export Citation
  • Saunders, M. A., , and B. D. Qian, 2002: Seasonal predictability of the winter NAO from North Atlantic sea surface temperatures. Geophys. Res. Lett., 29, 2049, doi:10.1029/2002GL014952.

    • Search Google Scholar
    • Export Citation
  • Schwierz, C., , C. Appenzeller, , H. C. Davies, , M. A. Liniger, , W. Müller, , T. F. Stocker, , and M. Yoshimori, 2006: Challenges posed by and approaches to the study of seasonal-to-decadal climate variability. Climatic Change, 79, 3163.

    • Search Google Scholar
    • Export Citation
  • Shongwe, M. E., , C. A. T. Ferro, , C. A. S. Coelho, , and G. J. Van Oldenborgh, 2007: Predictability of cold spring seasons in Europe. Mon. Wea. Rev., 135, 41854201.

    • Search Google Scholar
    • Export Citation
  • Toth, Z., , O. Talagrand, , G. Candille, , and Y. Zhu, 2003: Probability and ensemble forecasts. Forecast Verification: A Practitioner’s Guide in Atmospheric Science, I. T. Jolliffe and D. B. Stephenson, Eds., John Wiley & Sons, 137–163.

    • Search Google Scholar
    • Export Citation
  • Troccoli, A., 2010: Seasonal climate forecasting. Meteor. Appl., 17, 251268, doi:10.1002/met.184.

  • Troccoli, A., , M. Harrison, , D. L. T. Anderson, , and S. J. Mason, Eds., 2008: Seasonal Climate: Forecasting and Managing Risk. NATO Science Series, Springer Academic Publishers, 467 pp.

    • Search Google Scholar
    • Export Citation
  • Ulbrich, U., , G. C. Leckebusch, , and J. G. Pinto, 2009: Extra-tropical cyclones in the present and future climate: A review. Theor. Appl. Climatol., 96, 117131.

    • Search Google Scholar
    • Export Citation
  • Uppala, S. M., and Coauthors, 2005: The ERA-40 Re-Analysis. Quart. J. Roy. Meteor. Soc., 131, 29613012.

  • Van Oldenborgh, G. J., 2005: Comments on “Predictability of winter climate over the North Atlantic European region during ENSO events.” J. Climate, 18, 27702772.

    • Search Google Scholar
    • Export Citation
  • Vitart, F., 2006: Seasonal forecasting of tropical storm frequency using a multi-model ensemble. Quart. J. Roy. Meteor. Soc., 132, 647666.

    • Search Google Scholar
    • Export Citation
  • Vitart, F., and Coauthors, 2007: Dynamically-based seasonal forecasts of Atlantic tropical storm activity issued in June by EUROSIP. Geophys. Res. Lett., 34, L16815, doi:10.1029/2007GL030740.

    • Search Google Scholar
    • Export Citation
  • Wang, B., and Coauthors, 2009: Advance and prospectus of seasonal prediction: Assessment of the APCC/CliPAS 14-model ensemble retrospective seasonal prediction (1980–2004). Climate Dyn., 33, 93117, doi:10.1007/s00382-008-0460-0.

    • Search Google Scholar
    • Export Citation
  • Wang, W., , B. T. Anderson, , R. K. Kaufmann, , and R. B. Myneni, 2004: The relation between the North Atlantic Oscillation and SSTs in the North Atlantic basin. J. Climate, 17, 47524759.

    • Search Google Scholar
    • Export Citation
  • Weigel, A. P., , M. A. Liniger, , and C. Appenzeller, 2007: The discrete Brier and ranked probability skill scores. Mon. Wea. Rev., 135, 118124.

    • Search Google Scholar
    • Export Citation
  • Weigel, A. P., , D. Baggenstos, , M. A. Liniger, , F. Vitart, , and C. Appenzeller, 2008: Probabilistic verification of monthly temperature forecasts. Mon. Wea. Rev., 136, 51625182.

    • Search Google Scholar
    • Export Citation
  • Weisheimer, A., and Coauthors, 2009: ENSEMBLES: A new multi-model ensemble for seasonal-to-annual predictions—Skill and progress beyond DEMETER in forecasting tropical Pacific SST. Geophys. Res. Lett., 36, L21711, doi:10.1029/2009GL040896.

    • Search Google Scholar
    • Export Citation
  • Wilks, D. S., 2006: Statistical Methods in the Atmospheric Sciences: An Introduction. International Geophysics Series, Vol. 91, Elsevier, 475 pp.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Climatological track density (tracks per DJF per latitude) of (a) ERA-40 reanalysis and (b)–(l) DEMETER and ENSEMBLES models in the North Atlantic/European region for the period covered by the respective dataset (cf. Table 1). Isoline interval is 0.2 starting at 1.2.

  • View in gallery

    Monthly distribution of windstorm frequency as represented by (left) the absolute mean monthly number of windstorms per month and (right) the same, but during DJF, JFMA, and November–April (NDJFMA), for ERA-40 (black), DEMETER (solid), and ENSEMBLES models (dashed) with standard deviations for the periods covered by the respective models (cf. Table 1) and over the region of interest (dashed boxes in Fig. 1).

  • View in gallery

    ROC curves for windstorm frequency in the lower tercile (blue) and upper tercile (red) during DJF for the 1980–2001 period for different single and multimodel ensembles, decreasingly sorted according to the ROCSS of the upper-tercile forecast. The curves show the hit rate (vertical) and false alarm rate (horizontal) for different warning threshold probabilities.

  • View in gallery

    Predictive skill (RPSS) of windstorm frequency during (a),(c) DJF (i.e., lead time of 1–4 months) and (b),(d) JFMA (2–6 months) for the (a),(b)1980–2001 and (c),(d) 1960–2001 periods in the single-model ensembles of (left) the respective multimodel ensemble (DEM and ENS) and the multimodel ensemble combining all 11 models (GRAND), (middle) ENSEMBLES, and (right) DEMETER. The box (whiskers) indicates the 50% (90%) confidence interval based on a bootstrap method.

  • View in gallery

    Fraction of times the skill in the 1980–2001 period outperforms the skill of 10 000 random samples of the entire 1960–2001 period with the same number of years for model ensembles available for the entire period (solid edged/shaded bars for ensembles with significant skill p < 0.05/p < 0.20).

  • View in gallery

    The 21-yr running correlation between PNA and NAO indices in ERA-40 (black), DEMETER (solid lines), and ENSEMBLES single-model ensembles (dashed lines). The x axis denotes the central year of the running correlation. Significant correlation coefficients (p < 0.05) are marked.

  • View in gallery

    Proportions of winters (DJF) with high (dark gray), medium (white), and low (light gray) windstorm frequency in a subsample of winters with skill (left) above and (right) below the median for the 1980–2001 period based on hindcasts of (a) the DEMETER multimodel ensemble and observed storm frequency, (b) the ENSEMBLES multimodel ensemble and observed storm frequency, (c) the DEMETER multimodel ensemble and predicted storm frequency, and (d) the ENSEMBLES multimodel ensemble and predicted storm frequency.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 49 49 9
PDF Downloads 34 34 13

The Skill of Seasonal Ensemble Prediction Systems to Forecast Wintertime Windstorm Frequency over the North Atlantic and Europe

View More View Less
  • 1 Institute of Meteorology, Freie Universität Berlin, Berlin, Germany
  • | 2 Munich Re, Munich, Germany
© Get Permissions
Full access

Abstract

The science of seasonal predictions has advanced considerably in the last decade. Today, operational predictions are generated by several institutions, especially for variables such as (sea) surface temperatures and precipitation. In contrast, few studies have been conducted on the seasonal predictability of extreme meteorological events such as European windstorms in winter. In this study, the predictive skill of extratropical wintertime windstorms in the North Atlantic/European region is explored in sets of seasonal hindcast ensembles from the Development of a European Multimodel Ensemble System for Seasonal-to-Interannual Prediction (DEMETER) and the ENSEMBLE-based predictions of climate changes and their impacts (ENSEMBLES) projects.

The observed temporal and spatial climatological distributions of these windstorms are reasonably well reproduced in the hindcast data. Using hindcasts starting on 1 November, significant predictive skill is found for the December–February windstorm frequency in the period 1980–2001, but also for the January–April storm frequency. Specifically, the model suite run at Météo France shows consistently high skill.

Some aspects of the variability of skill are discussed. Predictive skill in the 1980–2001 period is usually higher than for the 1960–2001 period. Furthermore, the level of skill turns out to be related to the storm frequency of a given winter. Generally, winters with high storm frequency are better predicted than winters with medium storm frequency. Physical mechanisms potentially leading to such a variability of skill are discussed.

Current affiliation: Swiss Reinsurance Company, Zurich, Switzerland.

Current affiliation: School of Geography, Earth and Environmental Sciences, University of Birmingham, Birmingham, United Kingdom.

Current affiliation: Leibniz-Institut für Meereswissenschaften an der Universität Kiel (IFM-GEOMAR), Kiel, Germany.

Corresponding author address: Dominik Renggli, Institute of Meteorology, Freie Universität Berlin, Carl-Heinrich-Becker-Weg 6-10, D-12165 Berlin. E-mail: dominik.renggli@met.fu-berlin.de

Abstract

The science of seasonal predictions has advanced considerably in the last decade. Today, operational predictions are generated by several institutions, especially for variables such as (sea) surface temperatures and precipitation. In contrast, few studies have been conducted on the seasonal predictability of extreme meteorological events such as European windstorms in winter. In this study, the predictive skill of extratropical wintertime windstorms in the North Atlantic/European region is explored in sets of seasonal hindcast ensembles from the Development of a European Multimodel Ensemble System for Seasonal-to-Interannual Prediction (DEMETER) and the ENSEMBLE-based predictions of climate changes and their impacts (ENSEMBLES) projects.

The observed temporal and spatial climatological distributions of these windstorms are reasonably well reproduced in the hindcast data. Using hindcasts starting on 1 November, significant predictive skill is found for the December–February windstorm frequency in the period 1980–2001, but also for the January–April storm frequency. Specifically, the model suite run at Météo France shows consistently high skill.

Some aspects of the variability of skill are discussed. Predictive skill in the 1980–2001 period is usually higher than for the 1960–2001 period. Furthermore, the level of skill turns out to be related to the storm frequency of a given winter. Generally, winters with high storm frequency are better predicted than winters with medium storm frequency. Physical mechanisms potentially leading to such a variability of skill are discussed.

Current affiliation: Swiss Reinsurance Company, Zurich, Switzerland.

Current affiliation: School of Geography, Earth and Environmental Sciences, University of Birmingham, Birmingham, United Kingdom.

Current affiliation: Leibniz-Institut für Meereswissenschaften an der Universität Kiel (IFM-GEOMAR), Kiel, Germany.

Corresponding author address: Dominik Renggli, Institute of Meteorology, Freie Universität Berlin, Carl-Heinrich-Becker-Weg 6-10, D-12165 Berlin. E-mail: dominik.renggli@met.fu-berlin.de

1. Introduction

Predictions on time scales of months to seasons (beyond medium-range weather forecasts) are of major scientific and economic interest. Skillful longer-term predictions of frequency and intensity of extreme meteorological events with enormous loss potential could help to improve foresightful risk management. Windstorms in the North Atlantic/European region are of particular interest as they cause about two-thirds of overall and insured loss due to natural hazards in this region (Munich Re 2007). For example, the windstorm “Lothar” (December 1999) ranks among the 10 most severe natural hazards worldwide with respect to insured losses in the years 1980–2009 (Munich Re 2010). Consequently, there is a need for skillful forecasts of such events with different lead times (Murnane et al. 2002; Murnane 2004). Whereas short-term forecasts of extremes have been considerably improved in recent decades (Buizza and Hollingsworth 2002), little is known about the longer-term predictability of windstorms on seasonal time scales.

Research on seasonal predictions has greatly advanced in the last two decades (Troccoli et al. 2008; Schwierz et al. 2006; Harrison 2005; Goddard et al. 2001). Seasonal predictability has been extensively studied for many variables, such as surface temperature, precipitation, mean sea level pressure, and sea surface temperatures (SST), including indices of the El Niño–Southern Oscillation (ENSO), and tropical cyclones (e.g., Weisheimer et al. 2009; Jin et al. 2008; Wang et al. 2009; Vitart et al. 2007; Shongwe et al. 2007; Rodwell and Doblas-Reyes 2006; Vitart 2006; Palmer et al. 2004). Today, seasonal predictions for most of these variables are operationally issued by several institutions such as the International Research Institute of Climate and Society, the UK Met Office, and the European Centre for Medium-Range Weather Forecasts (ECMWF).

Seasonal predictability is generally assumed to be much lower in the midlatitudes than in the tropics and to stem from the oceans, especially the North Atlantic (e.g., Kushnir et al. 2006; Czaja and Frankignoul 2002; Rodwell and Folland 2002), or from sea ice variability. Other possible sources include variations of land surface properties such as the extent of continental snow cover (Fletcher et al. 2009; Orsolini and Kvamstø 2009) or soil moisture (Conil et al. 2009; Fischer et al. 2007), stratospheric variability (e.g., Douville 2009), or the remote influence of tropical variability (e.g., ENSO; Ineson and Scaife 2009). However, the amount of skill in the North Atlantic region arising from ENSO variability is still strongly under debate (Van Oldenborgh 2005; Mathieu et al. 2004).

Studies of predictive skill for the extratropics focusing on variables other than surface temperature and precipitation are rare. Müller et al. (2005) and Johansson (2007) demonstrated that there is significant skill in dynamical predictions for the mean winter (December–February) North Atlantic oscillation (NAO) in a multimodel ensemble of seasonal hindcasts with a lead time of 1–4 months. Our analysis of seasonal predictability of windstorms was motivated by the general relation between the NAO and windstorm climate over the North Atlantic (Donat et al. 2010; Pinto et al. 2009). It must be noted, however, that this relation is not as simple as a linear rise of storm number with the value of the NAO index. A mean negative or neutral phase of NAO in a given winter does not necessarily mean that there are no severe windstorm events; in fact, about 30%–40% of gale days in Europe occur during negative or neutral NAO phases (Donat et al. 2010).

A number of empirical studies explored the potential of seasonal wind climate predictability (Qian and Saunders 2003; Palutikof et al. 2002). For example, Qian and Saunders (2003) found correlation coefficients of about 0.6 between the mean number of gale days in winter and the extent of July–August snow cover in the Northern Hemisphere during the preceding summer. Using dynamical climate model runs with prescribed SST anomalies, Compo and Sardeshmukh (2004) found few prospects for the potential predictability of storm tracks (2–7-day bandpass-filtered variance of 500-hPa vertical velocity) on seasonal and decadal time scales for the North Atlantic region. In a study on the estimation of return periods of windstorms, no significant predictive skill was found for a monthly averaged measure of intensity of windstorms [Extreme Wind Index (EWI)] in seasonal forecast data of the ECMWF except in the first month (Della-Marta et al. 2010). In contrast to these studies, a different, event-based definition of wintertime windstorm climate is used in this study, and predictive skill is analyzed not only in one model but also in ensembles of several dynamical seasonal prediction models.

In Europe, seasonal predictability was investigated particularly within two European Union (EU)-funded projects, namely, the Development of a European Multimodel Ensemble System for Seasonal-to-Interannual Prediction (DEMETER; Palmer et al. 2004) and, more recently, the ENSEMBLE-based predictions of climate changes and their impacts (ENSEMBLES; Weisheimer et al. 2009). Both projects used coupled ocean–atmosphere climate models to generate retrospective seasonal forecasts (so-called hindcasts) for about the last 40 years. In this study, data from both projects are used to analyze the predictive skill of wintertime windstorm frequency in the North Atlantic and European region (45°–70°N, 45°W–20°E) within hindcasts of different (multi) model ensembles, periods, and target months, and additionally in different versions of the same models. Furthermore, the dependence of the variability of predictive skill on the windstorm frequency in a given winter is analyzed. Thus, a systematic skill assessment in dynamical seasonal predictions is carried out.

The definition and identification of winter windstorms in coupled climate model and reanalysis data is briefly described in section 2. Descriptions of the model and reanalysis data used for verification are given in section 3. The probabilistic verification measures and windstorm frequency classes are defined in section 4. The representation of windstorms in the hindcast data, predictive skill, and aspects of the variability of skill are shown in section 5, and discussion of the results and conclusions are given in section 6.

2. Definition and identification of wintertime windstorm events

So far, studies on the seasonal predictability of wind climate in the North Atlantic/European region have mostly used approaches involving gridpoint-wise measures, such as the number of days with wind speeds above a certain (absolute or relative) threshold (Qian and Saunders 2003; Palutikof et al. 2002) or measures spatially aggregated over a predefined domain (Della-Marta et al. 2010). In contrast, this study is based on the identification of individual windstorm events as entities extending over both space and time. This windstorm identification approach was used by Leckebusch et al. (2008b) to study properties of wintertime windstorms in climate change scenarios. In this study, individual windstorm events (in contrast to the individual model or reanalysis grid points) are used to define seasonally averaged properties of the windstorm climate and then analyze the hindcasts accordingly.

As in Leckebusch et al. (2008b), a winter windstorm is defined as a spatially coherent area (with a minimum extent of about 150 000 km2) characterized by extreme surface (10 m) wind speeds that exceed a local threshold and can be tracked for at least 18 h. Windstorms are thus identified with an event-based approach, using the local 98th percentile of surface wind speed as threshold. Areas of adjacent grid boxes with wind speeds exceeding the percentile threshold are tracked in consecutive time steps, resulting in individual events defined in time and space. The local 98th percentile of a given dataset was used for two reasons: first, the influence of differences in the local (model-specific) wind climatologies is less important (e.g., systematic model biases in wind speeds over the continents or different boundary layer parameterizations). Additionally, it was shown that exceedances of the 98th percentile correspond well with local storm damage climatologies (Leckebusch et al. 2007; Pinto et al. 2007; Klawa and Ulbrich 2003). Therefore, this event definition focuses on the impacts of severe windstorms.

The identification scheme is applied to both reanalysis and model data. The windstorm frequency for each month is defined as the number of identified storms occurring in the considered region (45°–70°N, 45°W–20°E) in the respective months and extracted from the reanalysis data and the model ensembles.

3. Data

The seasonal forecast model data from DEMETER and ENSEMBLES used in this study consist of ensembles of 6-month hindcasts from 11 different coupled atmosphere–ocean climate models with nine ensemble members each. These nine ensemble members were generated by slightly altered initial conditions for the coupled model runs [cf. Palmer et al. (2004) and Weisheimer et al. (2009) and references therein for further description]. Concentrating on winter windstorms, hindcasts starting on 1 November and running to the end of April were used in this study. Within the DEMETER project, three models are available for the 1959–2001 period, and four additional models cover at least the 1980–2001 period (Table 1, upper rows). From ENSEMBLES, four models delivered the required data for the period 1960–2001 (Table 1, lower rows). Generally, 6-hourly surface (10 m) wind speed data were used, except for the ECMWF model (labeled ECMF in the following; see Table 1 for acronyms and expansions of the other models). A numerical problem in this model led to an unrealistic intensification of extreme surface wind speeds (see http://www.ecmwf.int/products/forecasts/seasonal/documentation/system3/knownissues.html). Therefore, ECMWF recommends using wind speed on the 925-hPa level instead. However, pressure level wind data are only available in 12-hourly resolution. Hence, the windstorm identification for ECMF had to be performed with a lower temporal resolution compared to the other models. Therefore, the original threshold of minimum lifetime of 18 h (i.e., four time steps for 6-hourly data) was set to 12 h for ECMF (i.e., two time steps for 12-hourly data). This weaker constraint leads to a systematic positive bias in the number of identified windstorms (cf. Table 2 and Fig. 2), but this bias is successfully accounted for by the simple correction scheme applied in this study (cf. Table 3).

Table 1.

Acronyms, institutions, periods covered, and atmospheric and ocean components of the seasonal prediction models used for the study. See also Palmer et al. (2004) for further details on the DEMETER models (upper rows) and Weisheimer et al. (2009) for the ENSEMBLES models (lower rows).

Table 1.
Table 2.

Monthy distribution of windstorm frequency expressed as the mean number of identified windstorms per month and during DJF and JFMA, respectively, in (top) ERA-40; and deviations from ERA-40 of (middle rows) the seven DEMETER models (in %) and (bottom rows) the three ENSEMBLES models (in %) for the periods covered by the respective models (cf. Table 1). Significant deviations (p < 10%, two-sided rank sum test) are denoted in boldface. The systematic overestimation in ECMF is due to a weaker threshold concerning the duration of windstorm events related to the different temporal resolution of the data (see section 3 for details).

Table 2.
Table 3.

As in Table 2, but scaled for every model with its multiyear monthly mean frequency including all months (November to April)—that is, corrected for systematic errors.

Table 3.

Windstorm events identified on the basis of surface wind speed from the 40-yr ECMWF Re-Analysis (ERA-40) dataset (Uppala et al. 2005) are used for validation. Reanalysis data, DEMETER data, and model data from ECMF and LFPW were available on a common regular 2.5° grid. INGV and IFMK were provided on the original T63 resolution and were interpolated to the common regular 2.5° grid using a first-order conservative remapping scheme (Jones 1999).

4. Methods

a. Measures of predictive skill

In this study, the predictive skill of the windstorm frequency during December–February (DJF) and January–April (JFMA) is analyzed in hindcasts starting on 1 November (i.e., with lead times of 1–4 and 2–6 months, respectively) in the 11 single-model ensembles, in the multimodel ensembles of DEMETER (labeled DEM in the following) and ENSEMBLES (ENS), and in the combined DEMETER–ENSEMBLES multimodel ensemble (GRAND). Owing to the availability of the DEMETER and ENSEMBLES data, two different periods are verified: 1960–2001 and 1980–2001.

Both the relative operating characteristic (ROC) and the ranked probability skill score (RPSS) are used for forecast verification. The ROC measures the ability of a set of forecasts to discriminate between occurrence and nonoccurrence of specific events, namely the occurrence of a certain observed windstorm frequency class in the case of this study. These classes are defined by tercile thresholds (“high,” “medium,” and “low” windstorm frequency). The predicted probabilities are defined as the fraction of ensemble members below the first tercile (low frequency), between the first and the second tercile (medium frequency), and above the second tercile (high frequency). The tercile thresholds are defined for each model separately. The observed probabilities are defined as 1 for the occurred class and 0 for the two other classes. Calculation of the ROC requires dichotomous events and is therefore performed for each windstorm frequency class separately. Assuming that a set of n forecast–observations pairs is divided into n1 ocurrences of the event and n2 nonoccurrences, the ROC score is defined as
e1
where Pi (Pj) is the forecast probability issued prior to the occurrence (nonoccurrence) of a given frequency class, and I is set to 1 if the condition in parentheses holds, and 0 otherwise (Shongwe et al. 2007; Mason and Graham 2002); A is equivalent to the area under the ROC curve (i.e., the line spanned by the hit rate and false-alarm rate for different warning thresholds). The ROC skill score (ROCSS), defined as
e2
(Mason and Graham 1999; Wilks 2006), is 1 for a perfect forecast and below 0 for a forecast worse than a random forecast. The critical levels of statistical significant ROCSS are estimated from 10 000 random forecasts. For p < 0.05 (p < 0.10) they are 0.44 (0.35) for the 1980–2001 period (in consistency with Shongwe et al. 2007) and 0.26 (0.20) for the 1960–2001 period.
In contrast to the ROCSS, the ranked probability skill score measures the quality of a multiple-class forecast with respect to a reference forecast. Generally, the ranked probability score (RPS) of a given forecast–observation pair (RPSfc) is defined as
e3
Here Pk and Ok denote the predicted and observed cumulative probabilities for the kth class, and K is the number of classes (Toth et al. 2003; K = 3 in this study). The reference forecast is a simple climatological forecast with a constant probability of ⅓ for each of the three frequency classes. Similar to the RPSfc, the RPS of the climatological forecast (RPSclm) is defined as the sum of the squared differences between the cumulative climatological probabilities and the cumulative observed probabilities. The RPSS is defined as
e4
The horizontal bars denote the average of scores for a given period. Thus, a perfect forecast would have an RPSS of 1. A positive (negative) RPSS means that the model forecast is better (worse) than a climatological forecast; that is, it contains additional (no additional) information compared to a forecast based only on climatological knowledge. To account for the ensemble size, the debiased version of the RPSS is used (Weigel et al. 2007).

The statistical significance of the RPSS is calculated by a bootstrap method (Weigel et al. 2008). The hindcasts are resampled in such a way that a certain year could occur several times, whereas another is not included at all. The skill score is recalculated for this modified sample. The resampling is repeated 10 000 times. If the 5th percentile of the resulting distribution is greater than 0, the respective skill score is assumed to be significantly positive.

b. Relation between predictive skill and windstorm frequency

An interesting question is whether all winters are equally well predicted, or whether there is a systematic relation to the windstorm frequency of a given winter. From a scientific point of view, such a relation may allow processes contributing to predictive skill to be identified and evaluated. From a user perspective, it would be desirable to know whether, for example, winters with high storm frequency are better predicted than average winters. Since the ROCSS is evaluated for different windstorm frequency classes, it will be possible to draw some conclusions from this evaluation. In a complementary approach, the relation between skill and windstorm frequency is also analyzed for the RPSS. To this end the storm frequency class for every winter in a given period is calculated from ERA-40 as described in section 4a. Additionally, it is possible to calculate a measure of skill for every winter in the same period for a given forecast ensemble as
e5
Note that Eq. (5) is very similar to Eq. (4) except that the quotient is not evaluated for the average but rather for every winter t of a given period. Based on RPSSt two subsamples are computed, the first containing winters with good skill (RPSSt above the median) and the second, winters with low skill (RPSSt below the median). The proportion of each frequency class is then calculated in both subsamples. A disproportional representation of any frequency class in the subsamples is indicative of dependence of skill on storm frequency. For example, the extreme case with all medium-frequency winters in the well-predicted sample and none in the poorly predicted sample would indicate that medium-frequency winters are generally better predicted than high- and low-frequency winters.

The statistical significance is estimated by a bootstrap method where the winters of a given period are randomly sampled into two subsamples and the proportion for every frequency class is calculated. This is repeated 10 000 times. In the case of 22 winters (i.e., the 1980–2001 period) and three storm frequency classes, it was found that a given storm frequency class is significantly underrepresented (overrepresented, p < 0.05) if it attains at most 9% (at least 55%).

Focusing on the applicability of the seasonal forecast systems, a measure of confidence is introduced by a slight modification of the aforementioned approach. Instead of the windstorm frequency class being calculated on the basis of ERA-40 (observed frequency class), the storm frequency is calculated as the mean storm frequency of the forecast ensemble (predicted frequency class). A relation between the level of skill and the predicted storm frequency would help a potential user to estimate the confidence in a forecast for a given winter before it is actually verified.

5. Results

a. Representation of windstorms in reanalysis and seasonal hindcast data

In a first step, the representation of windstorms according to the definition used in this study is evaluated in the reanalysis and seasonal hindcast data. The average spatial distributions of the identified windstorms (in terms of track density) in the reanalysis and model data during DJF are shown in Fig. 1. In the reanalysis data, the highest values are found over the North Atlantic with a maximum between Iceland and the British Isles (Fig. 1a). The tracks of the windstorms split into two preferred pathways over the eastern North Atlantic: one leads northeastward into the Norwegian Sea, and the other leads eastward into the North and Baltic Seas. The British Isles and areas adjacent to the North Sea are very exposed regions. A small local maximum is discernible over the northwestern Mediterranean. The track density of cores of cyclones in the North Atlantic region shows a very similar pattern but is shifted to the northwest (Leckebusch et al. 2008a). This is consistent with the assumption that the highest wind speeds in an extratropical low pressure system most frequently occur to the south and east of its center.

Fig. 1.
Fig. 1.

Climatological track density (tracks per DJF per latitude) of (a) ERA-40 reanalysis and (b)–(l) DEMETER and ENSEMBLES models in the North Atlantic/European region for the period covered by the respective dataset (cf. Table 1). Isoline interval is 0.2 starting at 1.2.

Citation: Monthly Weather Review 139, 9; 10.1175/2011MWR3518.1

Because of the large ensemble size of nine members per model, the track density patterns are smoother in the seasonal forecast model data than in the reanalysis (Figs. 1b–l). The models are generally able to capture the main features of the observed track density. For example, all models consistently show the maximum track density over the North Atlantic. However, most models overestimate the track density and reproduce the center of the storm track shifted to the southwest compared with the reanalysis. This is most prominent in the models using Action de Recherche Petite Echelle Grande Echelle (ARPEGE; CNRM, CRFC, and LFPW; cf. Table 1), which show the maximum track density between 45° and 54°N (Figs. 1c,f,i), and less obvious for models using ECHAM (SMPI, SCNR, INGV, IFMK; Figs. 1b,e,h,k). Furthermore, models using the Integrated Forecast System (IFS) of ECMWF as their atmospheric component underestimate the track density especially over Scandinavia and central Europe (SCWF, LODY; ECMF to a smaller degree; Figs. 1d,g,j). The splitting into two preferred pathways revealed in the reanalysis data is discernible in several of the climate models (SMPI, SCWF, SCNR, LODY, ECWF, and UKMO).

The intraseasonal distribution of the identified windstorms is expressed as the monthly number of windstorms (i.e., the monthly windstorm frequency) affecting the North Atlantic and European region (dashed boxes in Fig. 1). In ERA-40, the monthly distribution of windstorm frequency peaks in January (about seven events on average) with a subsequent decrease to below one event in April (Fig. 2; Table 2). In the region of interest, 27.7 storms are identified on average from November to April (i.e., about 4.6 storms per month), with considerable interannual variability (Fig. 2, black whiskers). An average number of 19 windstorms is counted during DJF (Table 2).

Fig. 2.
Fig. 2.

Monthly distribution of windstorm frequency as represented by (left) the absolute mean monthly number of windstorms per month and (right) the same, but during DJF, JFMA, and November–April (NDJFMA), for ERA-40 (black), DEMETER (solid), and ENSEMBLES models (dashed) with standard deviations for the periods covered by the respective models (cf. Table 1) and over the region of interest (dashed boxes in Fig. 1).

Citation: Monthly Weather Review 139, 9; 10.1175/2011MWR3518.1

In general, the seasonal forecast models reproduce the monthly distribution well (e.g., the peak in January; Fig. 2). However, significant biases exist for certain models and months (Table 2). The biases of storm frequency during DJF range from −9.8% (SCWF) to +30.1% (UKMO; Table 2) and are model specific. For example, the two models with IFS (SCWF and LODY) underestimate the absolute windstorm frequency, which is consistent with the aforementioned underestimation of track density (Fig. 1). These biases are accounted for by scaling the monthly storm frequency with the multiyear monthly mean frequency including all months (November–April; e.g., 4.6 storms per month for ERA-40) and for each model separately. For the seasonal forecast models, the scaling factor is calculated as the average over all ensemble members. Thus, in the following, the storm frequency is expressed as an anomaly with respect to the models’ climatologies.

The bias-corrected windstorm frequencies of the models during DJF and JFMA (the focus of the remainder of the study) are not significantly different from ERA-40 (Table 3). Hence, the climatology of North Atlantic/European windstorms is reproduced sufficiently well in the hindcast data to conduct studies of the skill of windstorm frequency predictions on seasonal time scales.

b. Predictive skill

One major outcome of the study is that there is significant predictive skill, both in terms of ROCSS and RPSS, particularly in the multimodel ensembles. The skill reveals a dependence on the period considered. Furthermore, the level of the ROCSS indicates a dependence on the observed windstorm frequency.

Concentrating on winters with an anomalous high or low number of windstorms, ROCSSs have been computed for windstorm frequency in the lower tercile (low windstorm frequency) and upper tercile (high windstorm frequency), thereby avoiding known negative biases for the middle tercile class (Kharin and Zwiers 2003). Figure 3 shows the corresponding ROC curves for windstorm frequency during DJF in the 1980–2001 period. For most of the ensembles, the upper-tercile ROC curve deviates from the diagonal, indicating successful discrimination of the high windstorm frequency winters. For example, LFPW exhibits an almost perfect ROC curve (Fig. 3a). In contrast, the ROC curve of the lower-tercile category is above the diagonal for fewer ensembles (LFPW, GRAND, CRFC, DEM, SCWF, CNRM), and close or even below otherwise. As expected from Fig. 3, all models attain positive ROCSSs for the upper tercile, the majority significantly (p < 0.05; Table 4). Although generally lower, the skill for the upper tercile remains positive in the longer period 1960–2001, except in CNRM (Table 4). In contrast, the ROCSSs of the lower tercile are only significant in two ensembles (DEM, LFPW) in the 1980–2001 period (Table 4). Generally, the scores of the upper-tercile category are higher than of the lower-tercile category.

Fig. 3.
Fig. 3.

ROC curves for windstorm frequency in the lower tercile (blue) and upper tercile (red) during DJF for the 1980–2001 period for different single and multimodel ensembles, decreasingly sorted according to the ROCSS of the upper-tercile forecast. The curves show the hit rate (vertical) and false alarm rate (horizontal) for different warning threshold probabilities.

Citation: Monthly Weather Review 139, 9; 10.1175/2011MWR3518.1

Table 4.

ROCSS of windstorm frequency hindcasts during (left) DJF and (right) JFMA in (top) 1980–2001 and (bottom) 1960–2001 in (top three rows) the combined DEMETER–ENSEMBLES multimodel ensemble (GRAND), the ENSEMBLES multimodel ensemble (ENS), and the DEMETER multimodel ensemble (DEM); (middle four rows) the ENSEMBLES single-model ensembles; and (bottom seven rows) the DEMETER single-model ensembles for the events as indicated in the header. Statistically significant scores (p < 0.05/p < 0.10) are denoted in bold/italic.

Table 4.

Compared to DJF, the ROCSS is generally smaller for windstorm frequency during JFMA, as expected from the longer lead times (Table 4). Only INGV has consistently significant scores for both the lower and upper terciles, which is also reflected in significant skill for the GRAND and ENSEMBLES multimodel ensembles.

Especially successful are the the hindcasts of the LFPW and SCWF single-model and the GRAND multimodel ensembles. They attain positive skill scores in all cases shown in Table 4.

The ROCSSs for an additional above-median category agree well with the upper-tercile category (not shown).

The performance of the dynamical prediction models to forecast the probability of three windstorm frequency classes with respect to a climatological reference forecast is quantified in terms of RPSS (Fig. 4). For the period 1980–2001, 7 of the 11 single-model ensembles exhibit positive skill scores for the windstorm frequency during DJF (Fig. 4a). In particular, the models from Météo France (CRFC and LFPW) attain statistically significant skill scores of 0.24 and 0.41, respectively. Additionally, SMPI has a skill of 0.16, but with a significance marginally above 0.05 (p = 0.058 estimated from the distribution of the random sample as described in section 4a). All considered multimodel combinations consistently exhibit significant positive skill, but on a lower level (about 0.15). For the hindcasts of the windstorm frequency during JFMA the skill scores are generally lower (Fig. 4b). Still, LFPW and SCWF have significant skill (0.22 and 0.18, respectively), as does the combined GRAND multimodel ensemble (0.07).

Fig. 4.
Fig. 4.

Predictive skill (RPSS) of windstorm frequency during (a),(c) DJF (i.e., lead time of 1–4 months) and (b),(d) JFMA (2–6 months) for the (a),(b)1980–2001 and (c),(d) 1960–2001 periods in the single-model ensembles of (left) the respective multimodel ensemble (DEM and ENS) and the multimodel ensemble combining all 11 models (GRAND), (middle) ENSEMBLES, and (right) DEMETER. The box (whiskers) indicates the 50% (90%) confidence interval based on a bootstrap method.

Citation: Monthly Weather Review 139, 9; 10.1175/2011MWR3518.1

For the longer period 1960–2001, skill is generally lower than for the shorter period (e.g., compare the different skill scores of the LFPW model in Fig. 4). The three DEMETER single-model as well as the DEMETER multimodel hindcasts of windstorm frequency during DJF show negative skill (Fig. 4c). In contrast, the ENSEMBLES multimodel ensemble and two of its single-model ensembles (ECMF, LFPW) attain positive skill (0.06–0.14), the latter even significantly (Fig. 4c). For the longer period 1960–2001, the skill scores differ less for JFMA than for DJF (Fig. 4d). Interestingly, although none of the ENSEMBLES single-model ensembles has significant skill, both its multimodel ensemble and the GRAND multimodel ensemble have significant RPSS of 0.07. This demonstrates that information in multimodel seasonal forecasts of windstorm frequency with a lead time beyond three months outperforms a climatological forecast. It also illustrates the value of large (multimodel) ensembles in climate predictions (Hagedorn et al. 2005). Because of the large ensemble size, the confidence intervals are smaller for the multimodel ensemble than for the single-model ensembles. Thus, it is possible to conclude that even a relatively low skill score is significantly different from a random sample. This is more difficult for smaller ensembles with the same level of skill. Furthermore, the use of the multimodel is motivated by the fact that it is almost impossible to know which model performs best before the prediction is verified by observations. Note that the significant skill scores of the multimodel ensembles generally remain significant even if the LFPW model is removed from the ensembles. The only exception is the ENSEMBLE multimodel prediction of windstorm frequency during DJF for the 1980–2001 period (Fig. 4a). Note that the results from the ROCSS and RPSS analysis agree very well. Forecast ensembles with high ROCSS tend to have high RPSS.

The single-model ensembles show considerable differences in predictive skill. However, there are no obvious relations between skill and the resolution of the models (Table 1) or their systematic biases (Tables 2 and 3).

In addition to the standard climatological reference forecast, the RPSS was also calculated using three different reference forecasts based on persistence. The first reference forecast equals the observation of the preceding year (“interannual persistence”); the second considers the preceding four years (weighted by the autocorrelation function of the observed windstorm frequency; “multiyear persistence”); and the third equals the observed windstorm frequency during November, the initial month of the dynamical hindcasts (“November persistence”). The RPSS of the dynamical models with respect to the different persistence reference all become positive and in most cases statistically significant (not shown). Thus, the persistence reference forecast is a weaker benchmark for the dynamical predictions than its climatological counterpart [i.e., the RPSclim in Eq. (2) is generally higher for the persistence than for the climatological reference], consistent with Qian and Saunders (2003). Note that these results do not exclude the possibility that an optimal combination of persistence and climatological reference forecasts could yield a stronger benchmark.

The results shown so far are based on windstorm frequency. Additionally, predictive skill was analyzed in terms of the intensity of winter windstorms as measured by the Storm Severity Index (SSI; Leckebusch et al. 2008b) and three storm intensity classes defined according to the accumulated SSI values of the storms in a given winter. No significant skill was found for this particular measure of storm intensity (not shown).

c. Some aspects of the variability of skill

It has been noted that considering only average skill may mask potentially important skill variations (Compo and Sardeshmukh 2004). Therefore, some aspects of the variability of skill are addressed in the following, namely differences in skill between the 1980–2001 and 1960–2001 period, and the relation between skill and windstorm frequency.

1) Predictive skill 1980–2001 versus 1960–2001

Both skill measures tend to score more highly in the 1980–2001 than the 1960–2001 period (cf. Figs. 4a,b; Table 4). One possible reason could simply be the higher quality of the initial conditions due to improved observational data for the later period, especially with the introduction of satellite data, also related to more accurate storm identification in the reanalysis data. The analysis of the impact of better observations would ultimately depend on very specialized hindcast runs (e.g., by withholding new observations or observation types to mimic the state of the observation system in the earlier periods). Such hindcasts do not exist up to now.

To test whether sampling issues explain these findings, a bootstrap test is employed by randomly choosing 22 years (the same number as in the period 1980–2001) for each of the 10 single and multimodel ensembles available for the entire 1960–2001 period and computing the corresponding RPSS. This is repeated 10 000 times. From the resulting distribution, the fraction of random periods outperformed by the regular 1980–2001 period (in terms of RPSS) is computed. If this fraction is above 0.50, the regular 1980–2001 period is interpreted as being better predicted. This is indeed the case for 8 of the 10 model ensembles (Fig. 5). Only two models, IFMK and ECMF, show fractions of about 0.20. The performance of these models is doubtful, however, since none of them shows significant skill. In contrast, the LFPW model ensemble, which provides statistically significant skill scores for both the long and the short period, attains a fraction of above 0.99 (i.e., only 1% of the randomly chosen 22 forecast–observation pairs have skill of at least the value found for the regular 1980–2001 period). Additionally, the other models with significant skill in the 1980–2001 period show fractions of 0.80 or above. Therefore, we argue that sampling effects do not explain the better predictability of the 1980–2001 period.

Fig. 5.
Fig. 5.

Fraction of times the skill in the 1980–2001 period outperforms the skill of 10 000 random samples of the entire 1960–2001 period with the same number of years for model ensembles available for the entire period (solid edged/shaded bars for ensembles with significant skill p < 0.05/p < 0.20).

Citation: Monthly Weather Review 139, 9; 10.1175/2011MWR3518.1

Physical reasons may play a role as well. However, the long-term trend of sea level pressure in the North Atlantic region (reflected by a positive trend of the wintertime NAO in the 1960–2001 period) is not confirmed as a possible explanation. First, the forecast models do not reproduce the trend of the winter mean NAO. Second, the long-term trend accounts for only about 10% of the total interannual variability. One possibility could be the temporally varying relation between the Pacific–North American pattern (PNA) and the NAO. It was suggested that there are phases of strong and weak relations, both in data from coupled climate models and reanalyses (Pinto et al. 2011; Luksch et al. 2005). During phases of strong relations, predictive skill from the ENSO region could be propagating into the North Atlantic, potentially leading to higher skill on the seasonal time scale.

To evaluate the PNA–NAO link in the seasonal prediction models, PNA and NAO indices were computed in both hindcast and reanalysis data. The indices are defined as the leading principal components in the North Pacific (20°–87.5°N, 140°E–30°W) and North Atlantic (20°–87.5°N, 110°W–60°E) regions, respectively, based on the pooled monthly anomalies of geopotential height on the 500-hPa level from November to March. Then, the 21-yr running correlations between PNA and NAO were computed for both the reanalysis and hindcast data. The observed link between PNA and NAO is weak in the 1970s but relatively strong in the 1980s (Fig. 6), consistent with Pinto et al. (2011). The same figure reveals that the seasonal prediction models generally overestimate the strength of the PNA–NAO link. The two exceptions are CNRM and CRFC. However, the latter is only available for 22 years and therefore is represented by only two points in Fig. 6. Specifically, the models do not reproduce the decadal-scale variability found in the observations. Thus, the link produced in the models fits the observed link in the 1980s and 1990s, but not in the 1970s. Compared to observations, the model climate in the North Atlantic might be too strongly influenced by the Pacific region in the earlier period.

Fig. 6.
Fig. 6.

The 21-yr running correlation between PNA and NAO indices in ERA-40 (black), DEMETER (solid lines), and ENSEMBLES single-model ensembles (dashed lines). The x axis denotes the central year of the running correlation. Significant correlation coefficients (p < 0.05) are marked.

Citation: Monthly Weather Review 139, 9; 10.1175/2011MWR3518.1

2) Relation between predictive skill and windstorm frequency

The results of ROCSS analysis indicate differences between the skill for high and low windstorm frequency winters. In the following it is tested whether such dependence is also found for the RPSS by analyzing the fractioning of the three windstorm frequency classes in well and poorly predicted years (as described in section 4b).

In the subsample of well-predicted winters (skill above median) based on the DEMETER multimodel ensemble during the 1980–2001 period, the high, medium, and low storm frequency classes attain proportions of 55%, 9%, and 36%, respectively (Fig. 7a). Thus, the high-frequency class is significantly overrepresented in well-predicted years. In contrast, the subsample of poorly predicted winters (skill below median) consists of 18% high-frequency, 55% medium-frequency, and 27% low-frequency winters. Hence, the medium-frequency class is significantly underrepresented in well-predicted winters but significantly overrepresented in poorly predicted winters (cf. section 4b). This tendency is also found in the ENSEMBLES multimodel (Fig. 7b). High- (medium-) frequency winters are more (less) frequent in the well-predicted than in the poorly predicted subsamples. In contrast to the DEMETER multimodel ensemble, low-frequency winters are less frequent in well-predicted years (27%) than in poorly predicted winters (36%). However, the proportions based on the ENSEMBLES multimodel skill are not statistically significant.

Fig. 7.
Fig. 7.

Proportions of winters (DJF) with high (dark gray), medium (white), and low (light gray) windstorm frequency in a subsample of winters with skill (left) above and (right) below the median for the 1980–2001 period based on hindcasts of (a) the DEMETER multimodel ensemble and observed storm frequency, (b) the ENSEMBLES multimodel ensemble and observed storm frequency, (c) the DEMETER multimodel ensemble and predicted storm frequency, and (d) the ENSEMBLES multimodel ensemble and predicted storm frequency.

Citation: Monthly Weather Review 139, 9; 10.1175/2011MWR3518.1

Thus, there is a tendency toward higher (lower) RPSSs in winters with high (medium) observed windstorm frequency, in consistency with high ROCSSs for the upper-tercile category (cf. Table 4). However, this result is derived from information that would not be available at the time the forecast is issued (i.e., the observed storm frequency). Therefore, a question immediately arising is whether a similar relation of skill is manifested also with respect to the windstorm frequency predicted by the multimodel ensemble. In the following, the storm frequency class for a given winter is defined as the ensemble average of the windstorm frequency during DJF of the respective multimodel ensemble (i.e., the predicted storm frequency). Thus, only information available at the time the forecast was issued is used.

For the DEMETER multimodel ensemble, the proportions of the three predicted frequency classes in the well-predicted subsample (Fig. 7c, left) are the same as for the observed frequency classes (Fig. 7a, left). However, the fractioning in the subsample of poorly predicted winters is more pronounced. The high-, medium-, and low-frequency classes attain proportions of 9%, 64%, and 27%, respectively (Fig. 7c, right). Thus, the fractioning of both the high- and medium-frequency classes is statistically different from a random distribution for both the well-predicted and poorly predicted years. A tendency to more pronounced fractioning is also visible in the hindcasts of the ENSEMBLES multimodel ensemble. The well-predicted subsample consists of 36% high-frequency and 45% low-frequency but only 18% medium-frequency winters. In contrast, the poorly predicted years consist of 27% high- and 18% low-frequency winters, but the medium-frequency winters are significantly overrepresented with a proportion of 55%. Hence, the tendency toward higher (lower) predictive skill of high- (medium-) frequency winters is also discernible in the predictions themselves.

These tendencies are also manifested in the multimodel ensembles for the period 1960–2001, but not significantly (not shown). The single-model ensembles generally agree but exhibit more variability.

6. Discussion and conclusions

The seasonal forecast systems of the DEMETER (Palmer et al. 2004) and ENSEMBLES (Weisheimer et al. 2009) projects were investigated in terms of their ability to produce reliable forecasts of windstorm climate over the North Atlantic and Europe. Previous studies (Johansson 2007; Müller et al. 2005) demonstrated small (at a level of about 0.15 in terms of RPSS) but statistically significant skill in the DEMETER seasonal prediction ensembles in forecasting the mean winter NAO when started on 1 November. The statistically significant skill of windstorm frequency of the same magnitude (about 0.15 for the multimodel ensembles) is comparatively high and even suggests an economic usability of the predictions, for example in the insurance industry.

Several limitations of the ability of the numerical systems to predict North Atlantic/European windstorms on a seasonal time scale were identified. First of all, significant skill is only found for windstorm frequency, not for intensity. Thus, the skill of seasonal predictions of windstorm climate in this region depends on the predictand used. For example, Della-Marta et al. (2010) found no significant skill for the monthly mean intensity of wintertime wind events in the 1958–2001 period (except in the first month of the predictions) in various seasonal prediction systems of the ECMWF. Owing to the different lead times, different predictand and models, and a different validation period, a direct comparison of the results is not possible; our results are, however, consistent with those reported by Della-Marta et al. (2010) with respect to generally lower skill for the 1960–2001 period and no significant skill when using a seasonally accumulated intensity measure (SSI; cf. section 5a).

As expected, it was found that the predictive skill depends on the single models used or on the composition of the multimodel ensemble. Previous studies have found a negative relation between systematic biases in seasonal prediction models and skill (i.e., the smaller the bias, the higher the skill) (e.g., Lee et al. 2010; DelSole and Shukla 2010). Our study supports those findings to some extent. The correlation between the systematic biases in the number of windstorms during DJF (Table 2) with predictive skill (Fig. 4a) is negative, although not statistically significant. On the other hand, the most successful models (CRFC and LFPW) both have a strong systematic bias in terms of the location of the wind storms. It is thus not clear which attempts to correct the systematic biases in the representation of the climatological state are most likely to result in significantly improved seasonal predictions. This statement is also consistent with the finding that the ENSEMBLES hindcasts do not generally show better skill compared with the DEMETER models (Figs. 3 and 4; Table 4), despite improvements in the horizontal and vertical resolution of the models, for example (cf. Table 1; see Palmer et al. 2004; Weisheimer et al. 2009). In contrast, some improvements of skill have been achieved in tropical regions (Weisheimer et al. 2009). However, Weisheimer et al. (2009) also mention that there are regions on the globe where systematic errors (with respect to SST) have not been improved, but that it is hardly possible to attribute certain improvements or deteriorations to specific changes in the models without targeted model runs. The one exception with considerable improvements in predictive skill is the ENSEMBLES contribution from Météo France (LFPW). This model’s components correspond to the ones used in CRFC of DEMETER. Hence, improvements in LFPW compared with CRFC [e.g., a newer version of the atmospheric component, the Global Experimental Leads and Sea Ice for Atmosphere and Ocean (GELATO) sea ice model, and a slightly different initialization] result in an increased skill of 0.17 (0.32) for the windstorm frequency in DJF (JFMA). The fact that the LFPW/CRFC models provide robust skillful predictions for wintertime windstorm frequency—both in DEMETER and ENSEMBLES and for the entire hindcast period—lends even more credibility to the forecast ability of this particular model suite.

Eventually, detailed analysis of the physical mechanisms potentially serving as sources of skill in both observational and model data (cf. Blender et al. 2003), differences in their representation in the forecast models, and their complex interplay will contribute to understanding the physical mechanisms leading to successful predictions as well as the different performances of different models. Additional work in that respect is under way. Preliminary results show that the models differ in their responses to anomalous conditions in the North Atlantic as well as in the persistence of such anomalous conditions. In particular, not only SST but also the deeper North Atlantic Ocean (e.g., ocean heat content and its persistence) could play an important role. In addition to such regional oceanic influences, remote effects of the tropical variations, which have a high seasonal predictability (e.g., Troccoli 2010; Weisheimer et al. 2009; Goddard et al. 2001), could be a relevant source of predictability. Temporally varying influences from ENSO variability have been detected in Central Europe (e.g., Knippertz et al. 2003). Such variations could be transmitted to Central Europe through the strong ENSO–PNA link and a variable PNA–NAO link, which is found both in observations and in coupled atmosphere–ocean GCMs. Like the climate models, the seasonal prediction models generally overestimate the observed strength of the link between PNA and NAO (see Fig. 6), possibly related to the exaggerated zonality of the models (e.g., Ulbrich et al. 2009). While the observed link varies with time, the seasonal prediction models produce a constantly strong link, without such variations on decadal scales. In-depth analyses of these topics would be very useful but are beyond the scope of this study.

Another result of this study is that skill of the forecast of windstorm frequency depends on the forecasted frequency itself as well on the frequency eventually observed. A tendency toward higher (lower) skill in observed high-frequency (medium-frequency) winters is found for both skill scores used in the study. This finding is again consistent with a tendency toward higher skill scores in NAO predictions in winters of extreme mean NAO index values (Müller et al. 2005). Possible mechanisms leading to higher skill in extreme winters could be related to anomalous North Atlantic SSTs or continental snow cover anomalies in summer/autumn. They have been related to anomalies in the atmospheric conditions in the following winter (Fletcher et al. 2009; Cassou et al. 2004; Wang et al. 2004; Qian and Saunders 2003; Saunders and Qian 2002; Czaja and Frankignoul 2002). Thus, more extreme observed anomalies, if appropriately initialized, should increase the likelihood that the forecast model will produce a response in accordance with the observation. In contrast, if the anomalies are small or counteracting, a model response is less strongly triggered.

Because our windstorm-event definition is based on the level of surface wind speed relevant to the local occurrence of losses, our study suggests potential applicability of seasonal forecasts to Europe-wide risk management. With respect to the loss potential of windstorms (Leckebusch et al. 2007), the prospect of significant skill of any quantity being related to wintertime windstorms is relevant from a risk management perspective, even if the skills derived in this study seem low. Nonetheless, a fairly low skill may potentially be exploited by users with a wide spatial and temporal scope as long as the seasonal predictions outperform simple forecasts based on climatology or persistence (García-Morales and Dubus 2007; Murnane 2004). Additionally, the tendency toward higher skill in high storm frequency winters is interesting, since from a risk perspective winters with extreme (high or low) frequency might be more relevant than winters with medium frequency. The same tendency is also noted in the predictions themselves (i.e., in the predicted storm frequency). Thus, this information can be used as a measure of confidence in a given prediction.

In conclusion, the study provides a systematic assessment of the reproduction of wintertime windstorms over the North Atlantic and Europe in several single-model ensembles and their ability to predict wintertime windstorm frequency over the North Atlantic and Europe. The study serves as a starting point in two ways. First, it shows that seasonal forecast ensembles exhibit significant predictive skill, encouraging further development of model diagnostics (especially with respect to extreme events) and improvements of seasonal forecast models. In particular, the sources of skill should be analyzed in the hindcast data and compared to observation. This would enhance the understanding of differences in model performance with possible implications for further model improvements. Second, it shows the necessity of developing foresightful procedures for risk management practices—such as better preparation of liquidity for indemnity payouts—that rely on seasonal forecasts.

Acknowledgments

ECMWF kindly provided the ERA-40 reanalysis and DEMETER data. We thank Paco Doblas-Reyes (ECMWF), Andrea Alessandri (Euro-Mediterranean Centre for Climate Change), Noel Keenlyside (Leibniz Institute of Marine Sciences at Kiel University), and Jean-Philippe Piedelièvre (Météo France) for providing ENSEMBLES data. The comments of two anonymous reviewers led to significant improvements of the manuscript.

REFERENCES

  • Blender, R., , U. Luksch, , K. Fraedrich, , and C. C. Raible, 2003: Predictability study of the observed and simulated European climate using linear regression. Quart. J. Roy. Meteor. Soc., 129, 22992313.

    • Search Google Scholar
    • Export Citation
  • Buizza, R., , and A. Hollingsworth, 2002: Storm prediction over Europe using the ECMWF Ensemble Prediction System. Meteor. Appl., 9, 289305.

    • Search Google Scholar
    • Export Citation
  • Cassou, C., , C. Deser, , L. Terray, , J. W. Hurrell, , and M. Drévillon, 2004: Summer sea surface temperature conditions in the North Atlantic and their impact upon the atmospheric circulation in early winter. J. Climate, 17, 33493363.

    • Search Google Scholar
    • Export Citation
  • Compo, G. P., , and P. D. Sardeshmukh, 2004: Storm track predictability on seasonal and decadal scales. J. Climate, 17, 37013720.

  • Conil, S., , H. Douville, , and S. Tyteca, 2009: Contribution of realistic soil moisture initial conditions to boreal summer climate predictability. Climate Dyn., 32, 7593.

    • Search Google Scholar
    • Export Citation
  • Czaja, A., , and C. Frankignoul, 2002: Observed impact of Atlantic SST anomalies on the North Atlantic oscillation. J. Climate, 15, 606623.

    • Search Google Scholar
    • Export Citation
  • Della-Marta, P. M., , M. A. Liniger, , C. Appenzeller, , D. N. Bresch, , P. Köllner-Heck, , and V. Muccione, 2010: Improved estimates of the European winter windstorm climate and the risk of reinsurance loss using climate model data. J. Appl. Meteor. Climatol., 49, 20922120.

    • Search Google Scholar
    • Export Citation
  • DelSole, T., , and J. Shukla, 2010: Model fidelity versus skill in seasonal forecasting. J. Climate, 23, 47944806.

  • Donat, M. G., , G. C. Leckebusch, , J. G. Pinto, , and U. Ulbrich, 2010: Examination of wind storms over Central Europe with respect to circulation weather types and NAO phases. Int. J. Climatol., 30, 12891300, doi:10.1002/joc.1982.

    • Search Google Scholar
    • Export Citation
  • Douville, H., 2009: Stratospheric polar vortex influence on Northern Hemisphere winter climate variability. Geophys. Res. Lett., 36, L18703, doi:10.1029/2009GL039334.

    • Search Google Scholar
    • Export Citation
  • Fischer, E. M., , S. I. Seneviratne, , P. L. Vidale, , D. Lüthi, , and C. Schär, 2007: Soil moisture–atmosphere interactions during the 2003 European summer heat wave. J. Climate, 20, 50815099.

    • Search Google Scholar
    • Export Citation
  • Fletcher, C. G., , S. C. Hardiman, , P. J. Kushner, , and J. Cohen, 2009: The dynamical response to snow cover perturbations in a large ensemble of atmospheric GCM integrations. J. Climate, 22, 12081222.

    • Search Google Scholar
    • Export Citation
  • García-Morales, M. B., , and L. Dubus, 2007: Forecasting precipitation for hydroelectric power management: How to exploit GCM’s seasonal ensemble forecasts. Int. J. Climatol., 27, 16911705.

    • Search Google Scholar
    • Export Citation
  • Goddard, L., , S. J. Mason, , S. E. Zebiak, , C. F. Ropelewski, , R. Basher, , and M. A. Cane, 2001: Current approaches to seasonal-to-interannual climate predictions. Int. J. Climatol., 21, 11111152.

    • Search Google Scholar
    • Export Citation
  • Hagedorn, R., , F. J. Doblas-Reyes, , and T. N. Palmer, 2005: The rationale behind the success of multi-model ensembles in seasonal forecasting—I. Basic concept. Tellus, 57A, 219233.

    • Search Google Scholar
    • Export Citation
  • Harrison, M., 2005: The development of seasonal and inter-annual climate forecasting. Climatic Change, 70, 201220.

  • Ineson, S., , and A. A. Scaife, 2009: The role of the stratosphere in the European climate response to El Niño. Nat. Geosci., 2, 3236.

    • Search Google Scholar
    • Export Citation
  • Jin, E. K., and Coauthors, 2008: Current status of ENSO prediction skill in coupled ocean–atmosphere models. Climate Dyn., 31, 647664, doi:10.1007/s00382-008-0397-3.

    • Search Google Scholar
    • Export Citation
  • Johansson, A., 2007: Prediction skill of the NAO and PNA from daily to seasonal time scales. J. Climate, 20, 19571975.

  • Jones, P. W., 1999: First- and second-order conservative remapping schemes for grids in spherical coordinates. Mon. Wea. Rev., 127, 22042210.

    • Search Google Scholar
    • Export Citation
  • Kharin, V. V., , and F. W. Zwiers, 2003: On the ROC score of probability forecasts. J. Climate, 16, 41454150.

  • Klawa, M., , and U. Ulbrich, 2003: A model for the estimation of storm losses and the identification of severe winter storms in Germany. Nat. Hazards Earth Syst. Sci., 3, 725732.

    • Search Google Scholar
    • Export Citation
  • Knippertz, P., , U. Ulbrich, , F. Marques, , and J. Corte-Real, 2003: Decadal changes in the link between El Niño and springtime North Atlantic oscillation and European–North African rainfall. Int. J. Climatol., 23, 12931311.

    • Search Google Scholar
    • Export Citation
  • Kushnir, Y., , W. A. Robinson, , P. Chang, , and A. W. Robertson, 2006: The physical basis for predicting Atlantic sector seasonal-to-interannual climate variability. J. Climate, 19, 59495970.

    • Search Google Scholar
    • Export Citation
  • Leckebusch, G. C., , U. Ulbrich, , L. Fröhlich, , and J. G. Pinto, 2007: Property loss potentials for European midlatitude storms in a changing climate. Geophys. Res. Lett., 34, L05703, doi:10.1029/2006GL027663.

    • Search Google Scholar
    • Export Citation
  • Leckebusch, G. C., , M. Donat, , U. Ulbrich, , and J. G. Pinto, 2008a: Midlatitude cyclones and storms in an ensemble of European AOGCMs under ACC. CLIVAR Exchanges, No. 13, International CLIVAR Project Office, Southampton, United Kingdom, 3–5.

    • Search Google Scholar
    • Export Citation
  • Leckebusch, G. C., , D. Renggli, , and U. Ulbrich, 2008b: Development and application of an objective storm severity measure for the northeast Atlantic region. Meteor. Z., 17, 575587.

    • Search Google Scholar
    • Export Citation
  • Lee, J.-Y., and Coauthors, 2010: How are seasonal prediction skills related to models’ performance on mean state and annual cycle? Climate Dyn., 35, 267283.

    • Search Google Scholar
    • Export Citation
  • Luksch, U., , C. C. Raible, , R. Blender, , and K. Fraedrich, 2005: Decadal cyclone variability in the North Atlantic. Meteor. Z., 14, 747753.

    • Search Google Scholar
    • Export Citation
  • Mason, S. J., , and N. E. Graham, 1999: Conditional probabilities, relative operating characteristics, and relative operating levels. Wea. Forecasting, 14, 713725.

    • Search Google Scholar
    • Export Citation
  • Mason, S. J., , and N. E. Graham, 2002: Areas beneath the relative operating characteristics (ROC) and relative operating levels (ROL) curves: Statistical significance and interpretation. Quart. J. Roy. Meteor. Soc., 128, 21452166.

    • Search Google Scholar
    • Export Citation
  • Mathieu, P.-P., , R. T. Sutton, , B. Dong, , and M. Collins, 2004: Predictability of winter climate over the North Atlantic European region during ENSO events. J. Climate, 17, 19531974.

    • Search Google Scholar
    • Export Citation
  • Müller, W. A., , C. Appenzeller, , and C. Schär, 2005: Probabilistic seasonal prediction of the winter North Atlantic Oscillation and its impact on near surface temperature. Climate Dyn., 24, 213226.

    • Search Google Scholar
    • Export Citation
  • Munich Re, 2007: Zwischen Hoch und Tief—Wetterrisiken in Mitteleuropa. Munich Re, Order Number 302–05481, 57 pp.

  • Munich Re, NATCATSERVICE cited 2010: Significant natural disasters since 1980. [Available online at http://www.munichre.com/en/reinsurance/business/non-life/georisks/natcatservice/significant_natural_catastrophes.aspx.]

    • Search Google Scholar
    • Export Citation
  • Murnane, R. J., 2004: Climate research and reinsurance. Bull. Amer. Meteor. Soc., 85, 697707.

  • Murnane, R. J., , M. Crowe, , A. Eustis, , S. Howard, , J. Koepsell, , R. Leffler, , and R. Livezey, 2002: The weather risk management industry’s climate forecast and data needs. Bull. Amer. Meteor. Soc., 83, 11931198.

    • Search Google Scholar
    • Export Citation
  • Orsolini, Y. J., , and N. G. Kvamstø, 2009: Role of Eurasian snow cover in wintertime circulation: Decadal simulations forced with satellite observations. J. Geophys. Res., 114, D19108, doi:10.1029/2009JD012253.

    • Search Google Scholar
    • Export Citation
  • Palmer, T. N., and Coauthors, 2004: Development of a European Multimodel Ensemble System for Seasonal-to-Interannual Prediction (DEMETER). Bull. Amer. Meteor. Soc., 85, 853872.

    • Search Google Scholar
    • Export Citation
  • Palutikof, J. P., , T. Holt, , and T. J. Osborn, 2002: Seasonal forecasting of strong winds over Europe. Extended Abstracts, 16th Conf. on Probability and Statistics in the Atmospheric Sciences, Orlando, FL, Amer. Meteor. Soc., J3.12. [Available online at http://ams.confex.com/ams/pdfpapers/30343.pdf.]

    • Search Google Scholar
    • Export Citation
  • Pinto, J. G., , E. L. Fröhlich, , G. C. Leckebusch, , and U. Ulbrich, 2007: Changing European storm loss potentials under modified climate conditions according to ensemble simulations of the ECHAM5/MPI-OM1 GCM. Nat. Hazards Earth Syst. Sci., 7, 165175.

    • Search Google Scholar
    • Export Citation
  • Pinto, J. G., , S. Zacharias, , A. H. Fink, , G. C. Leckebusch, , and U. Ulbrich, 2009: Factors contributing to the development of extreme North Atlantic cyclones and their relationship with the NAO. Climate Dyn., 32, 711737, doi:10.1007/s00382-008-0396-4.

    • Search Google Scholar
    • Export Citation
  • Pinto, J. G., , M. Reyers, , and U. Ulbrich, 2011: The variable link between PNA and NAO in observations and in multi-century CGCM simulations. Climate Dyn., 36, 337354, doi:10.1007/s00382-010-0770-x.

    • Search Google Scholar
    • Export Citation
  • Qian, B. D., , and M. A. Saunders, 2003: Seasonal predictability of wintertime storminess over the North Atlantic. Geophys. Res. Lett., 30, 1698, doi:10.1029/2003GL017401.

    • Search Google Scholar
    • Export Citation
  • Rodwell, M. J., , and C. K. Folland, 2002: Atlantic air–sea interaction and seasonal predictability. Quart. J. Roy. Meteor. Soc., 128, 14131443.

    • Search Google Scholar
    • Export Citation
  • Rodwell, M. J., , and F. J. Doblas-Reyes, 2006: Medium-range, monthly, and seasonal prediction for Europe and the use of forecast information. J. Climate, 19, 60256046.

    • Search Google Scholar
    • Export Citation
  • Saunders, M. A., , and B. D. Qian, 2002: Seasonal predictability of the winter NAO from North Atlantic sea surface temperatures. Geophys. Res. Lett., 29, 2049, doi:10.1029/2002GL014952.

    • Search Google Scholar
    • Export Citation
  • Schwierz, C., , C. Appenzeller, , H. C. Davies, , M. A. Liniger, , W. Müller, , T. F. Stocker, , and M. Yoshimori, 2006: Challenges posed by and approaches to the study of seasonal-to-decadal climate variability. Climatic Change, 79, 3163.

    • Search Google Scholar
    • Export Citation
  • Shongwe, M. E., , C. A. T. Ferro, , C. A. S. Coelho, , and G. J. Van Oldenborgh, 2007: Predictability of cold spring seasons in Europe. Mon. Wea. Rev., 135, 41854201.

    • Search Google Scholar
    • Export Citation
  • Toth, Z., , O. Talagrand, , G. Candille, , and Y. Zhu, 2003: Probability and ensemble forecasts. Forecast Verification: A Practitioner’s Guide in Atmospheric Science, I. T. Jolliffe and D. B. Stephenson, Eds., John Wiley & Sons, 137–163.

    • Search Google Scholar
    • Export Citation
  • Troccoli, A., 2010: Seasonal climate forecasting. Meteor. Appl., 17, 251268, doi:10.1002/met.184.

  • Troccoli, A., , M. Harrison, , D. L. T. Anderson, , and S. J. Mason, Eds., 2008: Seasonal Climate: Forecasting and Managing Risk. NATO Science Series, Springer Academic Publishers, 467 pp.

    • Search Google Scholar
    • Export Citation
  • Ulbrich, U., , G. C. Leckebusch, , and J. G. Pinto, 2009: Extra-tropical cyclones in the present and future climate: A review. Theor. Appl. Climatol., 96, 117131.

    • Search Google Scholar
    • Export Citation
  • Uppala, S. M., and Coauthors, 2005: The ERA-40 Re-Analysis. Quart. J. Roy. Meteor. Soc., 131, 29613012.

  • Van Oldenborgh, G. J., 2005: Comments on “Predictability of winter climate over the North Atlantic European region during ENSO events.” J. Climate, 18, 27702772.

    • Search Google Scholar
    • Export Citation
  • Vitart, F., 2006: Seasonal forecasting of tropical storm frequency using a multi-model ensemble. Quart. J. Roy. Meteor. Soc., 132, 647666.

    • Search Google Scholar
    • Export Citation
  • Vitart, F., and Coauthors, 2007: Dynamically-based seasonal forecasts of Atlantic tropical storm activity issued in June by EUROSIP. Geophys. Res. Lett., 34, L16815, doi:10.1029/2007GL030740.

    • Search Google Scholar
    • Export Citation
  • Wang, B., and Coauthors, 2009: Advance and prospectus of seasonal prediction: Assessment of the APCC/CliPAS 14-model ensemble retrospective seasonal prediction (1980–2004). Climate Dyn., 33, 93117, doi:10.1007/s00382-008-0460-0.

    • Search Google Scholar
    • Export Citation
  • Wang, W., , B. T. Anderson, , R. K. Kaufmann, , and R. B. Myneni, 2004: The relation between the North Atlantic Oscillation and SSTs in the North Atlantic basin. J. Climate, 17, 47524759.

    • Search Google Scholar
    • Export Citation
  • Weigel, A. P., , M. A. Liniger, , and C. Appenzeller, 2007: The discrete Brier and ranked probability skill scores. Mon. Wea. Rev., 135, 118124.

    • Search Google Scholar
    • Export Citation
  • Weigel, A. P., , D. Baggenstos, , M. A. Liniger, , F. Vitart, , and C. Appenzeller, 2008: Probabilistic verification of monthly temperature forecasts. Mon. Wea. Rev., 136, 51625182.

    • Search Google Scholar
    • Export Citation
  • Weisheimer, A., and Coauthors, 2009: ENSEMBLES: A new multi-model ensemble for seasonal-to-annual predictions—Skill and progress beyond DEMETER in forecasting tropical Pacific SST. Geophys. Res. Lett., 36, L21711, doi:10.1029/2009GL040896.

    • Search Google Scholar
    • Export Citation
  • Wilks, D. S., 2006: Statistical Methods in the Atmospheric Sciences: An Introduction. International Geophysics Series, Vol. 91, Elsevier, 475 pp.

    • Search Google Scholar
    • Export Citation
Save