• Anthes, R. A., 1977: A cumulus parameterization scheme utilizing a one-dimensional model. Mon. Wea. Rev., 105 , 270286.

  • Barnston, A. G., and Coauthors, 1994: Long-lead seasonal forecasts—Where do we stand? Bull. Amer. Meteor. Soc., 75 , 20972114.

  • ——, Glantz, M. H., and Y. He, 1999: Predictive skill of statistical and dynamical climate models in SST forecasts during the 1997–98 El Niño episode and the 1998 La Niña onset. Bull. Amer. Meteor. Soc., 80 , 217243.

    • Search Google Scholar
    • Export Citation
  • Branković, C., and T. N. Palmer, 2000: Seasonal skill and predictability of ECMWF PROVOST ensembles. Quart. J. Roy. Meteor. Soc., 126 , 20352067.

    • Search Google Scholar
    • Export Citation
  • ——, ——, and Ferranti, L., 1994: Predictability of seasonal atmospheric variations. J. Climate, 7 , 217237.

  • Colman, R. A., and B. J. McAvaney, 1995: Sensitivity of the climate response of an atmospheric general circulation model to changes in convective parameterization and horizontal resolution. J. Geophys. Res., 100 , 31553172.

    • Search Google Scholar
    • Export Citation
  • Dix, M. R., and H. G. Hunt, 1995: Chaotic influences and the problem of deterministic seasonal predictions. Int. J. Climatol., 15 , 729752.

    • Search Google Scholar
    • Export Citation
  • Drosdowsky, W., and L. Chambers, 1998: Near global sea surface temperature anomalies as precictors of Australian seasonal rainfall. BMRC Res. Rep., 65, Bureau of Meteorology, Melbourne, Australia, 98 pp.

    • Search Google Scholar
    • Export Citation
  • Epstein, E. S., 1991: On obtaining daily climatological values for monthly mean. J. Climate, 4 , 365368.

  • Fels, S. B., and M. D. Schwarzkopf, 1975: The simplified exchange approximation: A new method for radiative transfer calculations. J. Atmos. Sci., 32 , 14751488.

    • Search Google Scholar
    • Export Citation
  • Frederiksen, C. S., D. P. Rowell, R. C. Balgovind, and C. K. Folland, 1999: Multidecadal simulations of Australia rainfall variability: The role of SSTs. J. Climate, 12 , 357379.

    • Search Google Scholar
    • Export Citation
  • Graham, R. J., A. D. L. Evans, K. R. Mylne, M. S. J. Harrison, and K. B. Robertson, 2000: An assessment of seasonal predictability using general circulation models. Quart. J. Roy. Meteor. Soc., 126 , 22112240.

    • Search Google Scholar
    • Export Citation
  • Hart, T. L., W. Bourke, B. J. McAvaney, B. W. Forgan, and J. L. McGregor, 1990: Atmospheric general circulation simulations with the BMRC global spectral model: The impact of revised physical parameterizations. J. Climate, 3 , 436459.

    • Search Google Scholar
    • Export Citation
  • Harzallah, A., and R. Sadourny, 1995: Internal versus SST-forced atmospheric variability as simulated by an atmospheric general circulation model. J. Climate, 8 , 474495.

    • Search Google Scholar
    • Export Citation
  • Huffman, G. J., and Coauthors, 1997: The Global Precipitation Climatology Project (GPCP) combined precipitation dataset. Bull. Amer. Meteor. Soc., 78 , 520.

    • Search Google Scholar
    • Export Citation
  • Jones, D., 1999: The prediction of Australian land surface temperatures using near global sea surface temperature patterns. BMRC Res. Rep. 70, Bureau of Meteorology, Melbourne, Australia, 44 pp.

    • Search Google Scholar
    • Export Citation
  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77 , 437471.

  • Kininmonth, W. R., 1999: The 1997–1998 El Niño event: A scientific and technical retrospective. WMO Rep. 905, 96 pp.

  • Kumar, A., M. P. Hoerling, M. Ji, A. Leetmaa, and P. Sardeshmukh, 1996: Assessing a GCM's suitability for making seasonal predictions. J. Climate, 9 , 115129.

    • Search Google Scholar
    • Export Citation
  • Kuo, H. L., 1974: Further studies of the parameterization of the influence of cumulus convection on large-scale flow. J. Atmos. Sci., 31 , 12321240.

    • Search Google Scholar
    • Export Citation
  • Lacis, A. A., and J. E. Hansen, 1974: A parameterization for the absorption of solar radiation in the earth's atmosphere. J. Atmos. Sci., 31 , 118133.

    • Search Google Scholar
    • Export Citation
  • Louis, J-F., 1979: A parametric model of vertical eddy fluxes in the atmosphere. Bound.-Layer Meteor., 17 , 187202.

  • Manabe, S., and J. L. Holloway, 1975: The seasonal variation of the hydrologic cycle as simulated by a global model of the atmosphere. J. Geophys. Res., 80 , 16171649.

    • Search Google Scholar
    • Export Citation
  • McAvaney, B. J., and R. A. Colman, 1993: The AMIP experiment: The BMRC AGCM configuration. BMRC Res. Rep. 38, Bureau of Meteorology, Melbourne, Australia, 43 pp.

    • Search Google Scholar
    • Export Citation
  • McPhaden, M. J., 1999a: Genesis and evolution of the 1997–98 El Niño. Science, 283 , 950954.

  • ——,. 1999b: The child prodigy of 1997–98. Nature, 398 , 559562.

  • Palmer, T. N., G. J. Shutts, and R. Swinbank, 1986: Alleviation of a systematic bias in general circulation and numerical weather prediction models through an orographic wave drag parameterization. Quart. J. Roy. Meteor. Soc., 112 , 10011039.

    • Search Google Scholar
    • Export Citation
  • Parker, D. E., P. D. Jones, C. K. Folland, and A. Bevan, 1994: Interdecadal changes of surface temperature since the late nineteenth century. J. Geophys. Res., 99 , 14 37314 399.

    • Search Google Scholar
    • Export Citation
  • ——, Folland, C. K., A. Bevan, M. N. Ward, M. Jackson, and K. Maskell, 1995: Marine surface data for analysis of climatic fluctuations on interannual to century timescales. Natural Climate Variability on Decade-to-Century Timescales, National Research Council, 241–252.

    • Search Google Scholar
    • Export Citation
  • Philander, S. G., 1990: El Niño, La Niña, and the Southern Oscillation. Academic Press, 293 pp.

  • Plaut, G., and R. Vautard, 1994: Spells of low-frequency oscillations and weather regimes in the Northern Hemisphere. J. Atmos. Sci., 51 , 210236.

    • Search Google Scholar
    • Export Citation
  • Potts, J. M., C. K. Folland, I. I. Jolliffe, and D. Sexton, 1996: Revised “LEPS” scores for assessing climate model simulations and long-range forecasts. J. Climate, 9 , 3453.

    • Search Google Scholar
    • Export Citation
  • Reynolds, R. W., and T. M. Smith, 1995: A high-resolution global sea surface temperature climatology. J. Climate, 8 , 15711583.

  • Ropelewski, C. F., and M. S. Halpert, 1987: Global and regional scale precipitation patterns associated with the El Niño/Southern Oscillation (ENSO). Mon. Wea. Rev., 115 , 16061626.

    • Search Google Scholar
    • Export Citation
  • ——, and ——,. 1989: Precipitation patterns associated with the high index phase of the Southern Oscillation. J. Climate, 2 , 268284.

    • Search Google Scholar
    • Export Citation
  • Rowell, D. P., 1998: Assessing potential seasonal predictability with an ensemble of multidecadal GCM simulations. J. Climate, 11 , 109120.

    • Search Google Scholar
    • Export Citation
  • ——, Folland, C. K., K. Maskell, and M. N. Ward, 1995: Variability of summer rainfall over tropical North Africa (1906–92): Observations and modelling. Quart. J. Roy. Meteor. Soc., 121 , 669704.

    • Search Google Scholar
    • Export Citation
  • Seaman, R., W. Bourke, P. Steinle, T. Hart, G. Embery, M. Naughton, and L. Rikus, 1995: Evolution of the Bureau of Meteorology's Global Assimilation and Prediction System. Part 1: Analysis and initialisation. Aust. Meteor. Mag., 44 , 118.

    • Search Google Scholar
    • Export Citation
  • Smith, N. R., 1995: The BMRC ocean thermal analysis system. Aust. Meteor. Mag., 44 , 93110.

  • Stern, W., and K. Miyakoda, 1995: The feasibility of seasonal forecasts speculated from multiple GCM simulations. J. Climate, 8 , 10711085.

    • Search Google Scholar
    • Export Citation
  • Tiedtke, M., 1984: The sensitivity of the time mean large-scale flow to cumulus convection in the ECMWF model. Report of the Workshop on Convection in Large Scale Numerical Models, Reading, United Kingdom, ECMWF, 297–316.

    • Search Google Scholar
    • Export Citation
  • ——,. 1989: A comprehensive mass flux scheme for cumulus parameterization in large-scale models. Mon. Wea. Rev., 117 , 17791800.

  • Trenberth, K. E., 1998: Development and forecasts of the 1997–98 El Niño. CLIVAR scientific issues. CLIVAR-Exch., 3 , 414.

  • Ward, M. N., and C. K. Folland, 1991: Prediction of seasonal rainfall in the north Nordeste of Brazil using eigenvectors of sea surface temperature. Int. J. Climatol., 11 , 711743.

    • Search Google Scholar
    • Export Citation
  • Webster, P. J., A. M. Moore, J. P. Loschnigg, and R. R. Leben, 1999: Coupled ocean–atmosphere dynamics in the Indian Ocean during 1997–98. Nature, 401 , 356360.

    • Search Google Scholar
    • Export Citation
  • Zheng, X., and C. S. Frederiksen, 1999: Validating interannual variability in an ensemble of AGCM simulations. J. Climate, 12 , 23862396.

    • Search Google Scholar
    • Export Citation
  • Zwiers, F. W., 1987: A potential predictability study conducted with an atmospheric general circulation model. Mon. Wea. Rev., 115 , 29572974.

    • Search Google Scholar
    • Export Citation
  • View in gallery
    Fig. 1.

    Overall LEPS skill scores for SSTA forecasts 4, 8, and 12 weeks ahead from week 0 using persisted SSTAs during the period Mar 1997–Mar 1998. Skill scores that are positive and statistically significant at the 95% level are shown shaded

  • View in gallery
    Fig. 2.

    Correlations between the observed SOI and the NCEP–NCAR reanalysis for the period Jan 1958–Nov 1991 for precipitation, surface air temperature, 200-hPa geopotential height, and MSLP. A 3-month running mean filter has been applied to all time series. Positive correlations are shown shaded

  • View in gallery
    Fig. 3.

    As in Fig. 2, but from an ensemble of five runs of the BMRC AGCM forced by observed SSTs and sea-ice

  • View in gallery
    Fig. 4.

    Seasonal anomalies of 200-hPa geopotential height derived from the NCEP–NCAR reanalysis for the period AMJ97–MAM98. Positive anomalies are shown shaded

  • View in gallery
    Fig. 5.

    As in Fig. 4, but for the K forecasts

  • View in gallery
    Fig. 6.

    The LEPS skill score for each of the K forecasts of 200-hPa geopotential height anomalies. Positive skill is shown shaded

  • View in gallery
    Fig. 7.

    The overall LEPS skill score, significant at a 95% level, for the K forecasts (top) and M forecasts (bottom) of 200-hPa geopotential height anomalies taken over all forecasts AMJ97–AMJ98. Positive and significant skill is shown contoured and shaded. Areas of negative skill are shown blank and uncontoured

  • View in gallery
    Fig. 8.

    As in Fig. 6, but for the K forecasts of MSLP anomalies

  • View in gallery
    Fig. 9.

    As in Fig. 7, but for K forecasts and M forecasts of MSLP anomalies

  • View in gallery
    Fig. 10.

    As in Fig. 6, but for the K forecasts of surface air temperature anomalies

  • View in gallery
    Fig. 11.

    As in Fig. 7, but for K forecasts and M forecasts of surface air temperature anomalies

  • View in gallery
    Fig. 12.

    As in Fig. 6, but for the K forecasts of precipitation anomalies

  • View in gallery
    Fig. 13.

    As in Fig. 7, but for for K forecasts and M forecasts of precipitation anomalies

  • View in gallery
    Fig. 14.

    The overall skill score, significant at the 95% level, for model simulation of seasonal anomalies of precipitation, surface air temperature, 200-hPa geopotential height, and MSLP for the period Jan 1958–Nov 1991. Positive and significant skill is shown contoured and shaded. Areas of negative skill are shown blank and uncontoured

  • View in gallery
    Fig. 15.

    The fraction of the total interseasonal variance of precipitation, surface air temperature, 200-hPa geopotential height, and MSLP that is attributable to SST/sea-ice forcing

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 132 14 0
PDF Downloads 22 6 0

Dynamical Seasonal Forecasts during the 1997/98 ENSO Using Persisted SST Anomalies

Carsten S. FrederiksenBureau of Meteorology Research Centre, and Cooperative Research Centre for Southern Hemisphere Meteorology, Melbourne, Victoria, Australia

Search for other papers by Carsten S. Frederiksen in
Current site
Google Scholar
PubMed
Close
,
Huqiang ZhangBureau of Meteorology Research Centre, Melbourne, Victoria, Australia

Search for other papers by Huqiang Zhang in
Current site
Google Scholar
PubMed
Close
,
Ramesh C. BalgovindBureau of Meteorology Research Centre, Melbourne, Victoria, Australia

Search for other papers by Ramesh C. Balgovind in
Current site
Google Scholar
PubMed
Close
,
Neville NichollsBureau of Meteorology Research Centre, Melbourne, Victoria, Australia

Search for other papers by Neville Nicholls in
Current site
Google Scholar
PubMed
Close
,
Wasyl DrosdowskyBureau of Meteorology Research Centre, Melbourne, Victoria, Australia

Search for other papers by Wasyl Drosdowsky in
Current site
Google Scholar
PubMed
Close
, and
Lynda ChambersBureau of Meteorology Research Centre, Melbourne, Victoria, Australia

Search for other papers by Lynda Chambers in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

An evaluation of trial seasonal forecasts during the 1997/98 El Niño, using an atmospheric GCM forced by persisted sea surface temperature and sea-ice anomalies, is presented. Generally, forecasts of seasonal anomalies of precipitation, surface air temperature, 200-hPa geopotential height, and mean sea level pressure (MSLP) are shown to have statistically significant skill in the Tropics and subtropics, but predominantly over the oceans. Surface air temperature and 200-hPa height anomalies are also skillfully forecast over land in the 30°S–30°N latitudinal band, and, in contrast to precipitation and MSLP, also show significant skill in the extratropics. The global pattern of significant skill seems not to be oversensitive to the use of a Kuo or a mass-flux convection scheme (Tiedtke), although the global root-mean-square errors are consistently larger, in the latter case.

Results from multidecadal simulations of the model, when forced by observed sea surface temperature and sea-ice, show that the model reproduces quite well the observed global Southern Oscillation index relationships and that these go some way to explaining the skill in the model forecasts. In addition, the global patterns of skill are consistent with those seen in the model forecasts. An estimate of the role of sea surface temperature and sea-ice in forcing interseasonal climate variations, suggests that the model displays forecasts skill in those areas where this forcing plays a large, if not dominant, role. In areas where internal, or chaotic, variability plays a dominant role, the model shows little statistically significant skill.

Corresponding author address: Dr. Carsten S. Frederiksen, Bureau of Meteorology Research Centre, GPO Box 1289K, Melbourne, Victoria 3001, Australia.Email: csf@bom.gov.au

Abstract

An evaluation of trial seasonal forecasts during the 1997/98 El Niño, using an atmospheric GCM forced by persisted sea surface temperature and sea-ice anomalies, is presented. Generally, forecasts of seasonal anomalies of precipitation, surface air temperature, 200-hPa geopotential height, and mean sea level pressure (MSLP) are shown to have statistically significant skill in the Tropics and subtropics, but predominantly over the oceans. Surface air temperature and 200-hPa height anomalies are also skillfully forecast over land in the 30°S–30°N latitudinal band, and, in contrast to precipitation and MSLP, also show significant skill in the extratropics. The global pattern of significant skill seems not to be oversensitive to the use of a Kuo or a mass-flux convection scheme (Tiedtke), although the global root-mean-square errors are consistently larger, in the latter case.

Results from multidecadal simulations of the model, when forced by observed sea surface temperature and sea-ice, show that the model reproduces quite well the observed global Southern Oscillation index relationships and that these go some way to explaining the skill in the model forecasts. In addition, the global patterns of skill are consistent with those seen in the model forecasts. An estimate of the role of sea surface temperature and sea-ice in forcing interseasonal climate variations, suggests that the model displays forecasts skill in those areas where this forcing plays a large, if not dominant, role. In areas where internal, or chaotic, variability plays a dominant role, the model shows little statistically significant skill.

Corresponding author address: Dr. Carsten S. Frederiksen, Bureau of Meteorology Research Centre, GPO Box 1289K, Melbourne, Victoria 3001, Australia.Email: csf@bom.gov.au

1. Introduction

The impact of global sea surface temperatures (SSTs), and in particular tropical SSTs, on global weather and climate systems has been recognized for some time. Observational studies have identified areas around the world where climate is strongly influenced by the El Niño–Southern Oscillation (ENSO) phenomenon (e.g., Ropelewski and Halpert 1987; Philander 1990). The 1997/98 El Niño was, perhaps, one of the most severe and best observed events experienced during the twentieth century [see, e.g., the World Meteorological Organization scientific and technical retrospective prepared by Kininmonth (1999)]. Trenberth (1998) and McPhaden (1999a,b) discuss the development of this event.

Over the last decade and a half, considerable efforts have been made to increase our understanding of the physics of ENSO and its evolution. In addition to improved observational and data assimilation systems, a large number of dynamical and statistical models, of varying complexity, have been developed for use in the prediction of tropical Pacific SSTs associated with ENSO. Barnston et al. (1994) discuss the success of several such models during the 1982–93 period, and Barnston et al. (1999) evaluate the success of 15 such models in forecasting ENSO conditions during the 1997/98 event. In the latter study, the authors suggest that the skill of the models had not improved, during this latest event, beyond that seen in the earlier period 1982–93. They concluded that, for both periods, the models had useful, though not excellent, skill with the median model correlation skill, averaged over lead times of one to three seasons, approximately equal to 0.6.

The successful forecast of the climate impacts of ENSO, especially over remotely teleconnected regions of the globe, such as, for example, Australia, North America, and northeast Brazil, is another matter again. Once an event has established itself, statistical methods provide a means of prediction of such impacts. Thus, for example, statistical approaches, based upon relationships between Australian rainfall and the Southern Oscillation index (SOI) and global SST anomalies (SSTAs), have recently been developed to make seasonal forecasts (Drosdowsky and Chambers 1998). Similar methods have also been developed for forecasting surface temperatures (Jones 1999). These techniques have shown significant forecasting skills over some parts of the Australian continent. However, the impact of the SST forcing on the climate variability is dependent on the position, timing, and strength of these boundary forcings. For this reason, it is important to investigate the feasibility of employing both stand-alone atmospheric general circulation models (AGCMs) and coupled ocean–atmosphere models in providing dynamical seasonal forecasts.

Theoretically, dynamically extended seasonal forecasts, beyond the limit of predictability of individual synoptic weather systems (typically 10–15 days), are possible because atmospheric boundary forcings, such as, for example, SST, sea-ice, and soil moisture, which evolve on slower timescales than that of the synoptic systems, may in places substantially determine the averaged weather features on a seasonal (1–3-month averages) timescale. However, even on this timescale and for seasonal averages, the climate system is still inherently chaotic and it is of interest to determine how much of the climate variability, at different geographical locations, is in fact predictable. A number of studies have considered the question of potential long-range predictability of climate variability. These have involved ensembles of multidecadal simulations using AGCMs with observed SSTs and different initial conditions (e.g., Dix and Hunt 1995; Harzallah and Sadourny 1995; Kumar et al. 1996; Rowell et al. 1995; Stern and Miyakoda 1995; Rowell 1998; Zwiers 1987; Zheng and Frederiksen 1999). With such an approach, it has been possible to separate interannual variability into chaotic components (due to the sensitivity to initial conditions) and a potentially predictable component (based on the ensemble average). Thus, it is possible to identify areas where different climate variables are potentially predictable. In brief, these studies showed high predictability over the Tropics in an approximate 30°S–30°N latitude band. In the mid- and high latitudes the predictability is generally weak and seasonally dependent.

Recently, the question of variations in seasonal-forecasting skill and predictability has been addressed in the European Union Prediction of Climate Variations on Seasonal to Interannual Timescales (PROVOST) project. This project involves integrations of atmospheric models with prescribed SSTs for the period 1979–93. Results from these studies also show that predictability tends to be largest in the Tropics and when ENSO is strong (see, e.g., Branković and Palmer 2000; Graham et al. 2000; see also Brankovic et al. 1994).

At the beginning of 1997 we started to investigate the use of an AGCM, forced by predicted SSTs, in a quasi-operational seasonal forecast mode, to determine to what extent climate anomalies might be predictable. Extended forecasts, 1 to 3 months ahead, were routinely presented each month, over a period of some 18 months, to the Bureau of Meteorology (BoM) National Climate Centre (NCC) and considered in the preparation of their seasonal outlook report. In this paper, we present a global analysis and evaluation of these trial forecasts.

The plan of this paper is as follows. In section 2, we describe the AGCM; in section 3 we discuss the experimental details, including the verifying observed datasets, the experimental design, the skill scores, and SST forecast scheme used. The extent to which observed SOI relationships are reproduced in our AGCM is considered in section 4, and our experimental forecasts are described and evaluated in section 5. In section 6, we discuss the model skill in extended multidecadal simulations with observed SSTs and sea-ice, and consider the question of potential predictability of interseasonal climate variations. Our conclusions are in section 7.

2. Model description

The AGCM used in this study is the Bureau of Meteorology Research Centre (BMRC) Climate Model (version 3.7). It is a spectral model with rhomboidal R31 horizontal resolution, together with a 17 sigma level vertical representation. Boundary layer and vertical diffusion parameterizations are after Louis (1979). The model has the option of using two convection schemes: a version of the Tiedtke mass-flux convection scheme (Tiedtke 1989) and one where deep convection is simulated by a modified Kuo scheme (Kuo 1974; Anthes 1977) with shallow convection parameterized in terms of the model's vertical diffusion scheme (Tiedtke 1984). Colman and McAvaney (1995) discuss the implementation of these convection schemes in the BMRC model. Gravity wave drag follows the formulation of Palmer et al. (1986). Shortwave and longwave radiation schemes are based, respectively, on the methods of Lacis and Hansen (1974) and Fels and Schwarzkopf (1975) with modifications. Soil temperature is based on heat storage from three layers with a “no flux” boundary condition. A single-layer “bucket” model (Manabe and Holloway 1975) is used to represent soil moisture storage. Climatological soil moisture is used in the model initialization. The carbon dioxide concentration was fixed at 345 ppmv. Further details of the BMRC model dynamics and physics are given in Hart et al. (1990) and McAvaney and Colman (1993) and references therein.

In this study, we present results from a series of model forecasts, during the period April 1997–June 1998, involving six member ensembles and with the above model configuration using both convection schemes to assess the sensitivity of the results, especially precipitation, to the convective parameterization. Additional results from an ensemble of five multidecadal simulations (1950–91; Frederiksen et al. 1999) using a nine vertical sigma level version of the model utilizing only the Kuo convective scheme will also be presented to assess its performance over a longer period. These runs were forced by the observed Global Sea-Ice and Sea Surface Temperature Dataset version 1.1 (GISST1.1; Parker et al. 1994, 1995).

3. Experimental details

a. Observed datasets

In assessing the forecasts of this study, we have used the daily averaged National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis dataset (1958–98; Kalnay et al. 1996) as the proxy observed for each climate variable. Monthly and seasonal anomalies of the NCEP–NCAR analysis, during the forecast period, were calculated with respect to a 40-yr climatology derived from NCEP–NCAR daily values for the period 1958–97. In defining a forecast skill score (see section 3c below), we have also used this 40-yr dataset to derive the cumulative probability distribution for each climate anomaly under consideration.

The use of the reanalysis data for verification of surface variables, especially precipitation, perhaps needs some justification. We did compare the reanalysis precipitation data with the Global Precipitation Climate Project dataset (Huffmann et al. 1997) and, while we did note some local/regional bias between the two, they were very similar on the large scale. We also compared the reanalysis data with Australian station data and again found it to be acceptable for our purposes, especially since our main interest here is in anomalies and not the total field value. The reanalysis data also provides coherent descriptions between surface variables and dynamic fields. It also gives us a longer dataset for use in applying the linear error in probability space (LEPS) scores Potts et al. (1996).

Weekly SSTs for the forecast period were taken from the Australian BoM ocean thermal analysis system (Smith 1995). SSTAs were calculated from these using the Reynolds and Smith (1995) monthly SST climatology. The Reynolds and Smith (1995) 40-yr (1950–89) monthly SSTA dataset was also used to define the SSTA cumulative probability distribution used to quantify the skill of the persisted SSTAs as a forecast 1–3 months ahead.

Finally, initial conditions for each ensemble member of the runs were taken from the BoM Global Assimilation and Prediction system (Seaman et al. 1995) at 6-hourly intervals and near the time when the weekly SSTAs were obtained.

b. Experimental design

Model forecasts were initiated each month from March 1997 to March 1998, covering the developing and mature phases of the 1997/98 El Niño event. These forecasts were conducted in a quasi-operational mode, with the results presented each month at the BoM NCC's monthly seasonal outlook meeting. Typically, a series of forecasts would commence in the second week of each month (designated month zero) and forecasts would be made for the next three succeeding months. A simple SSTA forecasting approach was used employing the latest observed weekly SSTAs, for the first week of month zero, which would be persisted into the forecast period. These SSTAs were then superimposed onto a 20-week climatology extending into the forecast period. This 20-week climatology was formed from the monthly climatology of Reynolds and Smith (1995) using the harmonic analysis technique of Epstein (1991) to derive daily SST values. The skill of using such SST forecasts is discussed below in section 3d.

In evaluating the model forecasts, we shall concentrate only on the skill of the model in predicting climate (seasonal) anomalies with respect to an ensemble of model control runs forced by a weekly SST climatology, during the forecast period. For convenience, we shall refer to the forecasts conducted with the Kuo convection scheme as the K forecasts and those using the mass-flux scheme as the M forecasts.

An ensemble approach was employed to overcome the model's sensitivity to initial conditions. In each set of experiments, an ensemble of six runs was generated with different initial conditions from the 6-hourly BMRC analysis system near the time when the weekly SSTAs were obtained. Thus, for example, the forecasts for May–June–July 1997 were started in the second week of April 1997 by using the observed SSTA from the week 7–13 April together with initial atmospheric conditions from 0500, 1100, 1700, 2300 UTC of 13 April and 0500, 1100 UTC of 14 April. We were restricted to six runs per forecast and control because of the quasi-operational nature of the project and limited computing resources. For both sets of forecasts, the monthly averaged anomalies were calculated as the differences between ensemble averages of the six runs using forecast SSTs against the ensemble averages with climatological SSTs. That is, two parallel sets of integrations were conducted; these being identical in every sense to the forecast set except that they were forced by climatological SSTs instead of persisted SSTAs.

This is a different approach from that used for example in the PROVOST studies (see, e.g., Branković and Palmer 2000). Our interest here is in isolating the impact of the SSTA on circulation, temperature, and precipitation anomalies, and the extent to which these anomalies serve as a useful forecast. Using climatological SST-forced runs allows us to attribute these anomalies to the SSTA in the event—not, in some sense, polluted by the model interannual variability. This is, of course, not possible with the observed anomalies that are dependent on the particular averaging period used to determine the observed climatology. All model anomalies were compared with anomalies derived for the forecast period using the NCEP–NCAR reanalysis for the period and the NCEP–NCAR climatologies.

c. Skill scores—LEPS

The most commonly used measures of skill, when assessing climate model simulations or forecasts, are the root-mean-square error and the anomaly correlation. In this study, we have decided to use, as our principal skill measure, the LEPS score developed by Ward and Folland (1991) and revised by Potts et al. (1996; where the advantages of this skill measure over the more commonly used root-mean-square error and the anomaly correlation are discussed). This score is related to the difference between the position of the forecast and the observation in the cumulative probability distribution of the particular climate variable under consideration.

Thus, for individual forecasts, if the position of the forecast in the cumulative distribution is Pf (ranging from 0.0 to 1.0) and that of the observation is Pυ, then the LEPS skill score (Potts et al. 1996) is defined to be
SPfPυPf2PfPυ2Pυ

From Eq. (1), it follows that, for a given observation Pυ, the largest skill score occurs when Pf = Pυ. In addition, most skill is attached to correctly forecasting at the extremes of the cumulative distribution (giving a score near 2). Correctly forecasting near the middle of the cumulative distribution (Pυ = 0.5), gives the smallest correct skill score (approximately 0.5). The worst possible score (−1.0) is assigned to forecasting at the exact opposite extreme for an extreme observation (e.g., Pυ = 0.0 and Pf = 1.0).

It is desirable to have a measure of overall skill, given a set, or ensemble, of forecasts. To achieve a skill score range from 100% to −100%, average skill (SK) may be defined (Potts et al. 1996) as
i1520-0442-14-12-2675-e2

Here the summation is over all pairs of forecasts and observations, S is the individual score for each forecast, and Sm depends on whether the numerator is positive or negative. For a positive numerator, Sm is the sum of the maximum possible scores given the observations [obtained by setting Pf = Pυ in Eq. (1)]. If the numerator is negative, Sm is the sum of the moduli of the worst possible scores given the observations, obtained by setting either Pf = 1.0 or Pf = 0.0 in Eq. (1) and taking the negative value with the largest modulus.

We shall be mainly concerned with testing the skill of the forecast anomalies. This has the advantage of removing some of the biases seen in the model simulation of climate variables. We shall also use the cumulative probability distribution derived from the 40-yr NCEP–NCAR reanalysis dataset (1958–97) for each climate anomaly we consider. The question of the significance of the model skill score also arises. To determine a 95% significant level for the overall skill score SK we have used a Monte Carlo approach in which we have generated 1000 random forecasts covering the entire period March 1997–March 1998. These forecasts have been generated by a random choice from the 40-yr NCEP–NCAR derived dataset of seasonal anomalies. The corresponding skill scores have been ranked and the 50th top score provides the 95% significant level.

d. Persistent SSTAs

In order to develop a dynamical seasonal forecasting scheme using an AGCM, we need first to predict the SSTs that will act as the lower boundary forcing for the atmospheric model. Given the SST climatology, it is only necessary to predict the SSTA from the climatological mean value. On a timescale of 1–3 months ahead, statistical models for SST prediction might be expected to outperform model-based SST predictions [Barnston et al. (1999) discuss the predictive skill of a number of dynamically based SST prediction models during the 1997/98 El Niño episode]. Before finally deciding to use persisted SSTAs, we did investigate two other statistical methods for predicting the SSTAs. One was based on a multichannel singular spectrum analysis (see, e.g., Plaut and Vautard 1994) and the other on time series analysis using a low-order autoregressive technique. However, we found that neither of these methods gave improvements in SSTA forecast skill, compared with simple persistence, for short lead times extending out to 3 months.

It is, therefore, of interest to calculate the LEPS scores for an SSTA forecasting scheme using simple persistence out to about 3 months ahead. In Fig. 1, we show the overall skill score SK for forecasts 4, 8, and 12 weeks ahead. Here, the score is based on forecasts using the SSTA from each week of the period March 1997–March 1998 to forecast for weeks 4, 8, and 12 from the week 0. The SSTAs are determined as the difference between the SST from the BoM ocean thermal analysis system (Smith 1995) and the Reynolds and Smith (1995) SST climatology. The SSTA cumulative probability, used in the skill score, is based on the 40-yr monthly anomaly dataset of Reynolds and Smith (1995) extending from 1950 to 1989, for which anomalies are given only for the region 45°S–65°N.

From Fig. 1, we see that there is high and significant skill over much of the world's oceans during the forecast period. In particular, there is very high skill in the eastern equatorial Pacific, the eastern North Pacific, the western Indian Ocean, and parts of the Atlantic. As the forecast period is extended, skill decreases in most areas with regions of negative skill appearing, especially at higher latitudes and in the western Pacific. However, there is still appreciable and significant skill in the eastern Pacific and along the Pacific rim, the western Indian Ocean, and around Australia, in those areas thought to be important regions of SST forcing during the period.

4. Modeled and observed SOI relationships

Before discussing the forecast results, it is instructive to consider the extent to which the model is able to reproduce observed SOI relationships when forced by observed SSTs over many decades. A number of studies (e.g., Ropelewski and Halpert 1987, 1989; Philander 1990) have identified regions around the globe where climate variability is strongly influenced by ENSO. Here, we investigate the observed (NCEP–NCAR) and modeled relationships between precipitation, surface air temperature (at 2 m above the surface), 200-hPa geopotential height, and mean sea level pressure (MSLP) with the corresponding observed and modeled SOI. The model results are from the 5-member ensemble runs of Frederiksen et al. (1999; see section 2). We have calculated monthly averaged anomalies, with the annual cycle removed, of each of the simulated and observed (NCEP–NCAR) climate variables, for the period 1958–91. To these, and the corresponding monthly SOI time series, we have applied a 3-month running mean filter, and have correlated the observed and simulated time series with, respectively, the observed and simulated SOI time series. Although the SOI–climate relationships tend to be seasonally dependent, this is a useful diagnostic to evaluate the model. Correlations between the corresponding observed and modeled SOI time series is better than 0.8 for each of the five runs.

Figures 2 and 3 show global plots of these correlations for the observed and modeled variables, respectively. The similarity between the observed and modeled correlation patterns and the magnitude of these correlations is quite remarkable. The pattern correlations between the observed and modeled cases are 0.56, 0.83, 0.91, and 0.93 for precipitation, surface air temperature, MSLP, and 200-hPa geopotential height, respectively.

With the exception of western and central Africa, parts of Europe, and the very high latitudes of the Northern Hemisphere, the SOI–precipitation relationship is correctly simulated over most continents. Thus, for example, a positive SOI (La Niña), is associated with enhanced rainfall over Australia, Southeast Asia, northeast Brazil, Canada, southern and eastern Africa, and India, and reduced rainfall over the United States and southern South America. However, over many of the continental areas, the linear SOI relationship explains a relatively small fraction of the rainfall variation. Consequently, as we show in the next section, internal, or chaotic, variations can play a large role in many of these locations. This may explain, for example, why over Australia the model forecasts during 1997–98 were not very skillful.

For the surface air temperature, the similarities between the observed and modeled is remarkable, again with the exception of the high latitudes of the Northern Hemisphere. Both show a correlation pattern, over the oceans, which has features of the typical ENSO SSTA pattern (La Niña phase), with negative correlations in the Indian Ocean, Atlantic Ocean, throughout most of the equatorial Pacific Ocean, and the eastern Pacific rim. Positive correlations are seen in the western, northern, and southern Pacific Ocean. Weak negative correlations between SOI and surface temperature are seen over, for example, the eastern and western coasts of Australia, parts of central and southern Africa, subtropical South America and Central America, and Canada.

The relationship between variations in the general circulation and the SOI is also very well reproduced in the model simulations. The patterns for 200-hPa geopotential height are very similar and show the Pacific North American (PNA) and Pacific South American (PSA) teleconnections, and the doubled-centered structure of the subtropical Pacific associated with ENSO. For negative SOI, these features would have opposite signs. Similarly, the shift in the Walker circulation associated with the SOI, reflected in the SOI–MSLP relationship, is also very well captured by the model. Thus, for example, during El Niño such changes are associated with an increase in MSLP over the western Pacific Ocean and a decrease in MSLP over the eastern Pacific.

5. Forecasts March 1997–March 1998

In this section, we present an evaluation of results from our model seasonal forecasts. These 120-day forecasts were initialized each month for the period March 1997 to March 1998. We shall concentrate on 3-month seasonal mean anomalies and shall compare these with anomalies from the NCEP–NCAR reanalysis for this period.

a. The 200-hPa geopotential height

Results for the model forecasts of the 200-hPa geopotential height suggests that both sets of forecasts have very good skill in the low latitudes (roughly between 30°S and 30°N) and moderate skill extending into the extratropics. This can be seen, to some extent, in Figs. 4 and 5. These show the seasonal anomalies [April–May–June 1997 (AMJ97) to March–April–May 1998 (MAM98)] for the NCEP–NCAR analysis and the K forecasts. It is clear from this sequence of diagrams that both the observed and modeled show similar features, especially in the latter half of the forecast period when the El Niño is well established. However, the magnitudes of the anomalies are generally weaker in the model, especially during the first half. Thus, for example, the evolution of features such as the double-centered positive anomaly in the tropical eastern Pacific, the PNA anomaly pattern, and Southern Hemisphere wave structures, especially south of South America, are seen in both figures. These features are consistent with the SOI–200-hPa geopotential height relationships depicted in Figs. 2 and 3. There is evidence, in both figures, of the characteristic PSA anomaly pattern with a high centered southwest of South America.

By May 1997, there were clear indications in the surface and subsurface thermal structure of the tropical eastern Pacific that an El Niño was imminent. Prior to the May forecast (JJA97), neither set of model forecasts displayed much skill. This can be seen from Fig. 6, which shows the individual skill score, S, for the K forecasts. The skill of the forecasts evolves throughout the forecast period, as the Pacific and Indian Ocean SSTAs (Webster et al. 1999) establish themselves and grow in magnitude. Thus, in Fig. 6, we see that it is not until JJA that the K forecasts show coherent and quite substantial skill throughout the subtropical band 30°S–30°N. In the latter half of the period, the model shows very high skill in the Tropics/subtropics, over the North Pacific and North America (PNA pattern) and in the southern Pacific Ocean south of South America.

The overall skill score SK is shown in Fig. 7 where areas of significant positive skill (at the 95% level) are shown shaded and contoured, and areas of negative skill are shown blank and uncontoured. This figure suggests that there is significant skill over much of the globe with the most noticeable exception of the high latitudes of the Southern Hemisphere, which is also the region where the K forecasts have the largest root-mean-square (rms) errors (not shown). It should be noted that we are dealing here only with 13 forecasts and consequently it is to be expected that randomly generated forecasts may prove to be fairly skillful, and will therefore raise the 95% significance level. The same procedure applied to a larger number of forecasts would be expected to yield larger areas with statistically significant skill. However, this will not be the case in regions of low inherent predictability.

The individual skill scores for the M forecasts (not shown) show essentially the same features as seen for the K forecasts. However, there appears to be less skill in the extratropics (including the Northern Hemisphere) and over Australia, in this case. This is reflected also in the overall skill score (Fig. 7), and in the larger rms errors seen in the M forecasts (not shown) over these regions. In particular, the globally (G) averaged rms error in the M forecasts is about 14% larger than for the K forecasts. The corresponding hemispheric averages over the Southern Hemisphere (SH) and the Northern Hemisphere (NH) are about 18% and 10% larger, respectively.

b. MSLP

With significant positive SSTAs over the eastern Pacific Ocean during the El Niño event, the model successfully simulated the MSLP anomalies associated with the shift in the tropical Walker circulation. This feature was reflected in the increase of MSLP over the western Pacific Ocean and the decrease of MSLP over the eastern Pacific Ocean (not shown) for the observed, K forecasts and M forecasts of MSLP anomalies. Associated with the eastward shifting of the ascending branch of the Walker circulation, convection activities also moved toward the central Pacific Ocean. In both model results, there was a significant increase in convective cloud formation over the eastern Pacific Ocean, and a significant decrease over the western Pacific Ocean (not shown).

Figure 8 shows the individual skill scores for the K forecasts. Again, it is not until the latter half of the forecast period that large regions of coherent and substantial skill appear on these maps. Note, during the premature period, the model skill of the MSLP anomaly forecasts is generally better in the eastern Pacific Ocean than in the western Pacific Ocean. When the event was fully established, there is largest skill in the western Pacific Ocean and much of the Indian Ocean. The overall skill score is shown in Fig. 9. These results indicate that the Southern Oscillation during the course of the 1997/98 El Niño event is well forecast by the model. Good forecasting skill is seen over large areas of the eastern and western Pacific Ocean and over the South Pacific Convergence Zone. Maxima in overall skill occur in those geographical regions where SOI–MSLP correlations have the largest magnitudes (Figs. 2 and 3). There is also some skill over the tropical Atlantic Ocean, parts of the North and South Pacific Ocean, and parts of the Indian Ocean. Except for the northern half of Australia, Indonesia, parts of central Europe, the Middle East, Alaska, and eastern Russia, there is little significant skill over land and the extratropical oceans, where the root-mean-square-errors (not shown) tend to be largest. The overall skill score for the M forecasts is also given in Fig. 9 and is not much different from the K-forecast skill, although this shows some extra skill over the North Pacific, Northern America, and Siberia, and the South Atlantic Ocean. Again, global and hemispheric RMS errors are larger for the M forecasts (G: 17%; SH: 21%; NH: 12%) with largest errors in the SH.

c. Surface air temperature

The major skill in the K forecasts of surface air temperature (Fig. 10) is over the tropical oceans, particularly over the eastern Pacific Ocean, the Indian Ocean, and the Atlantic Ocean, and is directly related to the skill in the persisted SSTAs. Again the skill shows an evolving pattern, becoming progressively larger and more coherent as the El Niño strengthens. In the extratropics, there is skill over the northeastern Pacific Ocean, the southeastern Pacific Ocean, the North Atlantic, and the southern Indian Ocean. There is also appreciable skill over tropical/subtropical landmasses, notably central/southern Africa, Central America, South America, Southeast Asia, and Australia. Over Australia, the pattern of skill is remarkably similar to statistical forecasts (not shown) conducted by the Australian Bureau of Meteorology and based on SST empirical orthogonal functions (D. Jones, BoM NCC, 1999, personal communication) during the same period. The K forecasts also demonstrate skill over the North American continent, with a pattern reminiscent of a PNA-like pattern. These same features can also be seen in the overall skill score (Fig. 11). This is consistent with the pattern of SOI–surface air temperature correlations shown in Figs. 2 and 3 excepting for some regions such as the southern United States and parts of east Asia. Between 60°S and 60°N, the root-mean-square temperature errors and biases are relatively small (not shown), with the largest errors occurring at the higher latitudes.

The individual LEPS score for the M forecasts, especially in the latter half of the forecast period, are qualitatively similar to the corresponding K forecasts (Fig. 10). This is also true of the overall global LEPS scores, although the K forecasts show additional significant skill over western Canada (Fig. 11). Consequently, the differences in rms errors is not as marked with the M-forecast errors only slightly larger (G: 8%; SH: 4%; NH: 12%).

d. Precipitation

The precipitation forecasts are the least skillful of the four variables. This can be clearly seen in Fig. 12 where the individual skill scores for the K forecasts are shown. There is little appreciable skill except in the tropical eastern Pacific Ocean, tropical western Indian Ocean, and over Indonesia. This diagram also shows increasing skill as the El Niño develops with strengthening SSTAs in the eastern Pacific. The overall skill score, for all the forecasts combined, is shown in Fig. 13, and again significant skill is confined largely to the Tropics, with maxima in skill in the east tropical Pacific, over Indonesia and Southeast Asia, the western Indian Ocean and central East African coast, and northeast Brazil. Over Australia, there are some isolated areas of significant skill. Skill over the mid- and high latitudes does not generally exhibit a coherent pattern and probably reflects the more stochastic nature of the rainfall process in the extratropics.

Generally the M-forecast skill scores are qualitatively similar to Fig. 12, but with a more coherent structure in the region of the SPCZ, where the model response appears to be improved with the mass-flux convective scheme. This can also be seen in the overall skill score (Fig. 13). The corresponding root mean square errors and biases (not shown) are generally larger when the mass-flux convective scheme is used (G: 23%; SH: 22%; NH: 24%). Otherwise, there are many qualitative similarities with the K-forecast results. Indeed, Colman and McAvaney (1995) reported broad similarities between experiments using Kuo and the mass-flux schemes, although the mass-flux scheme improved the model precipitation climatology over some regions.

Although there was quite large SST forcing during the 1997/98 El Niño, some of the observed seasonal anomalies had unique features (see, e.g., Trenberth 1998). This was especially the case in some parts of the Tropics and subtropics that are generally thought of as being fairly predictable (see next section) and where a model might therefore expect to have skill. Thus, for example, while winter (JJA) rainfall in Australia during 1997 was below average over much of eastern Australia, it was not as extreme as in previous major El Niños. Similarly, summer (JJA) Indian rainfall was only marginally below average. In both these areas, the model showed little skill (see Fig. 13). Interestingly, statistical methods used by the Australian Bureau of Meteorology and based on SST empirical orthogonal functions also showed poor skill (not shown) over Australia (W. Drosdowsky 1999, unpublished manuscript) during the same period.

In contrast, the model successfully forecast the highly anomalous heavy rainfall in the Horn of Africa region and the very severe droughts over Indonesia. The exceptionally high SSTs in the equatorial Indian Ocean are thought to be largely responsible for the heavy rains in the former region, and the model appears to have responded correctly to this lower boundary forcing. Trenberth (1998) suggests that the major convection in this area had an impact on the divergent atmospheric circulation that may have altered the teleconnections and thus influenced Australia, India, and southern Asia. The presence of strong intraseasonal oscillations [Madden–Julian oscillations (MJOs)] during JJA appear to have overcome the large-scale tendency for dry conditions (Trenberth 1998). Given that models tend to underestimate the MJO, this may be a factor in why our forecasts lacked skill over Australia and India.

6. Multidecadal simulations with observed SSTs and sea-ice

It is instructive to consider the results in section 5 in the light of the model's skill in simulating seasonal anomalies when forced by observed SSTs and sea-ice over many decades. It is also useful to estimate how “potentially” predictable (Rowell et al. 1995; Rowell 1998) interseasonal variations, such as those considered in the previous section, are when the external (SST/sea-ice) forcing is prescribed from observations. Such estimates provide, in a sense, an upper predictive bound on our seasonal forecasts forced with predicted SSTAs and highlight the importance of internal, or chaotic, variations in some geographical locations. To this end, we have used the five-member ensemble of multidecadal simulations (1950–91) of Frederiksen et al. (1999) forced by GISST1.1 and the NCEP–NCAR reanalysis data for the period 1958–97, and have analyzed the common period 1958–91.

a. Model skill (1958–91)

Here, we present plots of the overall skill score SK (Fig. 14), significant at the 95% level (using Monte Carlo techniques), for the model simulation of all seasonal climate anomalies for JFM, FMA, …, DJF (taken together) for the period January 1958–November 1991; that is, averaged over 405 seasonal means JFM58–SON91. We only consider the ensemble average of all five runs, and seasonal anomalies for both the NCEP–NCAR dataset and the model are with respect to seasonal means averaged over the period 1961–90. As above, we concentrate on only four climate variables; precipitation, surface air temperature, 200-hPa geopotential height, and the MSLP. Again, regions of significant positive skill are shown shaded and contoured.

Significant and coherent skill in the simulation of precipitation anomalies is largely confined to the subtropical band 30°S–30°N. As was the case with the rainfall forecasts (Fig. 13), maxima in skill occur in the eastern Pacific Ocean, the Indonesian archipelago, and eastern Africa/western Indian Ocean. There is also significant skill over much of Australia, the southwestern United States, southern Africa, along the eastern coast of Africa, the Middle East, parts of Southeast Asia, and northeast Brazil. In the extratropics there appears to be no systematic pattern of skill. There is some seasonality in the skill (not shown). Thus, while the maximum in the eastern Pacific Ocean persists during most of the year, the maxima in skill over the Indonesian archipelago and eastern Africa/western Indian Ocean are largest during austral winter and spring. Similarly, over Australia, there is some significant skill over northern Queensland and parts of the central east of the continent also mainly during austral winter and spring. In austral autumn and winter there is also significant skill in a band stretching from the northwest across the continent.

The simulation of surface air temperature anomalies is skillful over most of the global oceans. This is due mainly to the response of the near-surface temperatures to the prescribed SSTs. As in Fig. 11, for the forecasts, maxima in skill occur in the eastern Pacific Ocean, the North Pacific, the Indian Ocean, and the Atlantic Ocean. Over land, the pattern of significant skill is qualitatively similar to that seen in Fig. 11, with skill largely confined to the 30°S–30°N latitudinal band but with some skill over North America, Asia Minor, Europe, Japan, Australia, New Zealand, and southern South America. There is no significant skill over large parts of Antarctica, continental Europe/Asia, and the central United States. The location of the skill maxima also seems to have some dependence on the annual cycle of solar radiation. This is especially true for the Atlantic Ocean where maximum in skill occur in the summer hemisphere, and in the subtropics during spring and autumn (not shown).

The model simulation of the 200-hPa geopotential height field is significantly skillful globally and, as in Fig. 7, has maximum skill along the equator and between 30°S and 30°N. Figure 7 shows significant and high skill in the Southern Ocean south of South America and over North America that is not seen to the same extent in Fig. 14. This can be partly explained by the fact that the simulations here involve both ENSO and non-ENSO years. The tropical SST forcing of the PNA and PSA teleconnections in the 200-hPa geopotential height helps explain the high forecast skill seen in our forecasts during 1997–98. Nonetheless, overall, there is a good qualitative correspondence between the distribution in skill scores seen in our multidecadal simulations and that seen in our forecasts. This is also true for the MSLP, where there are consistent similarities between Fig. 8 and Fig. 14. Both show significant skill predominantly in a subtropical band, with maxima generally in the eastern and western Pacific Ocean, and Atlantic Ocean.

b. Potential predictability of interseasonal variations

Given the similarity between the geographical distribution of skill in both the multidecadal simulations above and the forecasts of section 5, it is of interest to try to explain this distribution in terms of the role of SSTs and sea-ice in forcing interseasonal variations. To this end, we have adapted the analysis of variance (ANOVA) technique of Rowell et al. (1995) and Rowell (1998) to distinguish between the roles of external forcing and internal, or chaotic, variability in the interseasonal variation of the seasonal anomalies, of the multidecadal simulations above. Here we have used the 167 seasonal mean anomalies (MAM, JJA, SON, DJF), with the seasonal cycle removed, for the period 1950–91 from each of the five runs.

Rowell et al. (1995) and Rowell (1998) investigate interannual variations of seasonal means but the technique can equally well be applied to interseasonal variations, which are more appropriate here given the way we have defined our overall skill score. The technique assumes that seasonal anomalies can be represented as consisting of an SST/sea-ice forced component and one due to internal variability. While the components themselves cannot be explicitly determined, their contribution to the total variance may be estimated (Rowell et al. 1995; Rowell 1998). Thus, if we represent the seasonal anomaly time series as {xst, t = 1, …,T} for a given run s (s = 1, S), with T = 167 and S = 5, we can estimate the contributions to the total interseasonal variance due to internal variability (ε) and SST/sea-ice forcing (β) by
i1520-0442-14-12-2675-e3
Here, we have used the convention that an average over a given subscript is represented by an “open circle.” Thus, xso represents the average over all seasons and xoo the average over all seasons and all simulations. The total interseasonal variance is just the sum (ε + β).

In Fig. 15, we show, for each of our four climate variables, the fraction of the total interseasonal variance that is attributable to the SST/sea-ice forcing, and is therefore potentially predictable if the SST/sea-ice forcing were perfectly predictable (Rowell et al. 1995; Rowell 1998). There is a remarkable resemblance between the geographical distribution of the skill scores in Fig. 14 and the SST/sea-ice forced variance in Fig. 15, with areas of high/low skill corresponding, generally, to areas of high/low SST/sea-ice forcing. If, for convenience we rescale the total skill scores in Fig. 14 so that they lie between [0, 100] [i.e., we use (SK/2 + 50) instead of SK], we can calculate pattern correlations between corresponding plots in these two figures. The pattern correlations are 0.90, 0.84, 0.95, 0.83, respectively, for the 200-hPa geopotential height, MSLP, the surface air temperature, and precipitation. Rescaling and correlating the total skill scores of our forecasts (Figs. 7, 9, 11, 13) with Fig. 15 results in corresponding pattern correlations of 0.90, 0.85, 0.88, and 0.79 for the K runs, and 0.90, 0.86, 0.88, and 0.78 for the M runs.

It would appear, therefore, that the model has skill in those areas where SST/sea-ice forcing plays a large, perhaps dominant role. For precipitation, this tends to be predominantly over the tropical/subtropical parts of the oceans, with local maxima located in the eastern equatorial Pacific Ocean, to the north of Australia, throughout the Indonesian archipelago, in the tropical/subtropical Atlantic Ocean, and throughout the Indian Ocean. Over land, better than 20% of the variation in rainfall over eastern- and western-central Africa, northwest Australia, Indonesia, northeast Brazil, and central America is SST/sea-ice forced. However, over land, rainfall variation is predominantly due to internal variability, in this model. In contrast, for the other three climate variables, there are land areas in the 30°S–30°N latitudinal band for which SST/sea-ice forcing plays a dominant role. For the surface air temperature, the SST/sea-ice forcing also extends into the midlatitude oceans of both hemispheres.

7. Conclusions

The main aim of this study has been to assess the use of a stand-alone AGCM, forced by persisted sea surface temperature anomalies, in forecasting seasonal climate anomalies one to three months ahead. To this end, a series of quasi-operational forecasts were conducted using the BMRC AGCM throughout the 1997/98 El Niño period, covering mainly the development and mature phases of this event. A simple SSTA forecast scheme was used where the latest observed SSTA persists into the forecast period. This approach was shown to have significant skill (based on LEPS scores) throughout the forecast period during the 1997/98 El Niño. This was especially true in the eastern equatorial Pacific Ocean, the eastern North Pacific Ocean, the western Indian Ocean, and parts of the Atlantic.

In general, the model forecasts showed significant skill in forecasting seasonal anomalies of precipitation, surface air temperature, 200-hPa geopotential height, and MSLP over large areas of the Tropics and subtropics, but predominantly over oceans in the case of precipitation and MSLP. There was little systematic skill in the rainfall and MSLP forecasts in the extratropics. In contrast, skill in surface temperature and 200-hPa geopotential height forecasts tropical/subtropical landmasses and extended into the extratropics of both hemispheres.

Model forecasts were able to capture the Southern Oscillation and changes in tropical convection associated with the observed eastward shift of the ascending branch of the Walker circulation. Important teleconnections in the global circulation, such as the PNA and PSA anomaly patterns were also skillfully forecast. Interestingly, model skill for all climate variables evolved throughout the forecast period, showing increasingly more widespread and larger skill with the maturing of the El Niño.

The model forecasts showed some sensitivities to the convection schemes used. However, there was not much different in the overall patterns of skill as determined by combined LEPS scores. There was, however, a tendency for the forecast using the mass flux scheme to have larger root-mean-square errors. This was especially the case with precipitation but only marginally so with surface air temperature.

An analysis of the model's ability to simulate climate anomalies, over the period 1958–91, when forced with observed SSTs and sea-ice, showed patterns of skill consistent with those seen in the model forecasts discussed above. The model was also able to reproduce the observed SOI relationships during this period. These relationships were shown to explain, to a large extent, the skill seen in the forecasts, especially in those regions highly correlated with the SOI.

Finally, in this study, we have looked at the roles played by SST/sea-ice forcing and internal dynamics, or chaos, in forcing climate variations, as a means to understanding what might be the limits of predictability. That proportion of the climate variation that is SST/sea-ice forced is potentially predictable if the SST forcing were perfectly predictable. Using an ANOVA technique adapted to seasonal data for the period 1950–91, we have shown that SST/sea-ice forcing predominates mainly in the 30°S–30°N latitude band for all four climatic variables considered here. That is, climate variations are potentially predictable mainly in this region. In the extratropics, internal variability, in our model, tends to play the dominant role, except for surface temperature which also showed high SST forcing in the extratropics. Again, as with the SOI relationships, these results go some way to explaining the skill of model forecasts during 1997/98 and in the multidecadal simulations, with regions of high/low skill generally corresponding to regions of high/low SST/sea-ice forcing.

One question that arises from this study is how much skill was lost due to the error in the SST forecast scheme used. To some extent this was addressed by the similarity of the patterns of skill seen in the longer 1958–91 simulations forced with observed SSTs. However, we have also conducted some supplementary experiments forcing the model with observed weekly SSTAs for the same 1997/98 forecast period. While this does lead to improved LEPS scores especially over the tropical Atlantic Ocean and Indian Ocean, the overall pattern of the geographical locations of forecasting skill is very similar between the two sets of experiments.

Continued efforts are being taken to study the model forecasts in more detail to see if improvements can be made in future. One area we hope to investigate is the impact of soil moisture anomalies on the seasonal forecasts. Here, the initialization of soil moisture was with climatology. This may not be sufficient, and some longer spinup strategy may be necessary to initialize the soil moisture to more realistic values. We also hope to investigate moisture fluxes during the 1997/98 El Niño to try to understand how these varied from earlier episodes, and what influence this may have had on global patterns, especially of rainfall, during this period.

Acknowledgments

We are indebted to Bryant McAvaney, Bill Bourke, and members of their modelling groups for maintaining and developing the BMRC AGCM. We would also like to thank Vuk Vojisavljevic and Alex Kariko for their technical support. This study was partly funded by a grant from the Australian Land and Water Research and Development Corporation (LWRRDC).

REFERENCES

  • Anthes, R. A., 1977: A cumulus parameterization scheme utilizing a one-dimensional model. Mon. Wea. Rev., 105 , 270286.

  • Barnston, A. G., and Coauthors, 1994: Long-lead seasonal forecasts—Where do we stand? Bull. Amer. Meteor. Soc., 75 , 20972114.

  • ——, Glantz, M. H., and Y. He, 1999: Predictive skill of statistical and dynamical climate models in SST forecasts during the 1997–98 El Niño episode and the 1998 La Niña onset. Bull. Amer. Meteor. Soc., 80 , 217243.

    • Search Google Scholar
    • Export Citation
  • Branković, C., and T. N. Palmer, 2000: Seasonal skill and predictability of ECMWF PROVOST ensembles. Quart. J. Roy. Meteor. Soc., 126 , 20352067.

    • Search Google Scholar
    • Export Citation
  • ——, ——, and Ferranti, L., 1994: Predictability of seasonal atmospheric variations. J. Climate, 7 , 217237.

  • Colman, R. A., and B. J. McAvaney, 1995: Sensitivity of the climate response of an atmospheric general circulation model to changes in convective parameterization and horizontal resolution. J. Geophys. Res., 100 , 31553172.

    • Search Google Scholar
    • Export Citation
  • Dix, M. R., and H. G. Hunt, 1995: Chaotic influences and the problem of deterministic seasonal predictions. Int. J. Climatol., 15 , 729752.

    • Search Google Scholar
    • Export Citation
  • Drosdowsky, W., and L. Chambers, 1998: Near global sea surface temperature anomalies as precictors of Australian seasonal rainfall. BMRC Res. Rep., 65, Bureau of Meteorology, Melbourne, Australia, 98 pp.

    • Search Google Scholar
    • Export Citation
  • Epstein, E. S., 1991: On obtaining daily climatological values for monthly mean. J. Climate, 4 , 365368.

  • Fels, S. B., and M. D. Schwarzkopf, 1975: The simplified exchange approximation: A new method for radiative transfer calculations. J. Atmos. Sci., 32 , 14751488.

    • Search Google Scholar
    • Export Citation
  • Frederiksen, C. S., D. P. Rowell, R. C. Balgovind, and C. K. Folland, 1999: Multidecadal simulations of Australia rainfall variability: The role of SSTs. J. Climate, 12 , 357379.

    • Search Google Scholar
    • Export Citation
  • Graham, R. J., A. D. L. Evans, K. R. Mylne, M. S. J. Harrison, and K. B. Robertson, 2000: An assessment of seasonal predictability using general circulation models. Quart. J. Roy. Meteor. Soc., 126 , 22112240.

    • Search Google Scholar
    • Export Citation
  • Hart, T. L., W. Bourke, B. J. McAvaney, B. W. Forgan, and J. L. McGregor, 1990: Atmospheric general circulation simulations with the BMRC global spectral model: The impact of revised physical parameterizations. J. Climate, 3 , 436459.

    • Search Google Scholar
    • Export Citation
  • Harzallah, A., and R. Sadourny, 1995: Internal versus SST-forced atmospheric variability as simulated by an atmospheric general circulation model. J. Climate, 8 , 474495.

    • Search Google Scholar
    • Export Citation
  • Huffman, G. J., and Coauthors, 1997: The Global Precipitation Climatology Project (GPCP) combined precipitation dataset. Bull. Amer. Meteor. Soc., 78 , 520.

    • Search Google Scholar
    • Export Citation
  • Jones, D., 1999: The prediction of Australian land surface temperatures using near global sea surface temperature patterns. BMRC Res. Rep. 70, Bureau of Meteorology, Melbourne, Australia, 44 pp.

    • Search Google Scholar
    • Export Citation
  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77 , 437471.

  • Kininmonth, W. R., 1999: The 1997–1998 El Niño event: A scientific and technical retrospective. WMO Rep. 905, 96 pp.

  • Kumar, A., M. P. Hoerling, M. Ji, A. Leetmaa, and P. Sardeshmukh, 1996: Assessing a GCM's suitability for making seasonal predictions. J. Climate, 9 , 115129.

    • Search Google Scholar
    • Export Citation
  • Kuo, H. L., 1974: Further studies of the parameterization of the influence of cumulus convection on large-scale flow. J. Atmos. Sci., 31 , 12321240.

    • Search Google Scholar
    • Export Citation
  • Lacis, A. A., and J. E. Hansen, 1974: A parameterization for the absorption of solar radiation in the earth's atmosphere. J. Atmos. Sci., 31 , 118133.

    • Search Google Scholar
    • Export Citation
  • Louis, J-F., 1979: A parametric model of vertical eddy fluxes in the atmosphere. Bound.-Layer Meteor., 17 , 187202.

  • Manabe, S., and J. L. Holloway, 1975: The seasonal variation of the hydrologic cycle as simulated by a global model of the atmosphere. J. Geophys. Res., 80 , 16171649.

    • Search Google Scholar
    • Export Citation
  • McAvaney, B. J., and R. A. Colman, 1993: The AMIP experiment: The BMRC AGCM configuration. BMRC Res. Rep. 38, Bureau of Meteorology, Melbourne, Australia, 43 pp.

    • Search Google Scholar
    • Export Citation
  • McPhaden, M. J., 1999a: Genesis and evolution of the 1997–98 El Niño. Science, 283 , 950954.

  • ——,. 1999b: The child prodigy of 1997–98. Nature, 398 , 559562.

  • Palmer, T. N., G. J. Shutts, and R. Swinbank, 1986: Alleviation of a systematic bias in general circulation and numerical weather prediction models through an orographic wave drag parameterization. Quart. J. Roy. Meteor. Soc., 112 , 10011039.

    • Search Google Scholar
    • Export Citation
  • Parker, D. E., P. D. Jones, C. K. Folland, and A. Bevan, 1994: Interdecadal changes of surface temperature since the late nineteenth century. J. Geophys. Res., 99 , 14 37314 399.

    • Search Google Scholar
    • Export Citation
  • ——, Folland, C. K., A. Bevan, M. N. Ward, M. Jackson, and K. Maskell, 1995: Marine surface data for analysis of climatic fluctuations on interannual to century timescales. Natural Climate Variability on Decade-to-Century Timescales, National Research Council, 241–252.

    • Search Google Scholar
    • Export Citation
  • Philander, S. G., 1990: El Niño, La Niña, and the Southern Oscillation. Academic Press, 293 pp.

  • Plaut, G., and R. Vautard, 1994: Spells of low-frequency oscillations and weather regimes in the Northern Hemisphere. J. Atmos. Sci., 51 , 210236.

    • Search Google Scholar
    • Export Citation
  • Potts, J. M., C. K. Folland, I. I. Jolliffe, and D. Sexton, 1996: Revised “LEPS” scores for assessing climate model simulations and long-range forecasts. J. Climate, 9 , 3453.

    • Search Google Scholar
    • Export Citation
  • Reynolds, R. W., and T. M. Smith, 1995: A high-resolution global sea surface temperature climatology. J. Climate, 8 , 15711583.

  • Ropelewski, C. F., and M. S. Halpert, 1987: Global and regional scale precipitation patterns associated with the El Niño/Southern Oscillation (ENSO). Mon. Wea. Rev., 115 , 16061626.

    • Search Google Scholar
    • Export Citation
  • ——, and ——,. 1989: Precipitation patterns associated with the high index phase of the Southern Oscillation. J. Climate, 2 , 268284.

    • Search Google Scholar
    • Export Citation
  • Rowell, D. P., 1998: Assessing potential seasonal predictability with an ensemble of multidecadal GCM simulations. J. Climate, 11 , 109120.

    • Search Google Scholar
    • Export Citation
  • ——, Folland, C. K., K. Maskell, and M. N. Ward, 1995: Variability of summer rainfall over tropical North Africa (1906–92): Observations and modelling. Quart. J. Roy. Meteor. Soc., 121 , 669704.

    • Search Google Scholar
    • Export Citation
  • Seaman, R., W. Bourke, P. Steinle, T. Hart, G. Embery, M. Naughton, and L. Rikus, 1995: Evolution of the Bureau of Meteorology's Global Assimilation and Prediction System. Part 1: Analysis and initialisation. Aust. Meteor. Mag., 44 , 118.

    • Search Google Scholar
    • Export Citation
  • Smith, N. R., 1995: The BMRC ocean thermal analysis system. Aust. Meteor. Mag., 44 , 93110.

  • Stern, W., and K. Miyakoda, 1995: The feasibility of seasonal forecasts speculated from multiple GCM simulations. J. Climate, 8 , 10711085.

    • Search Google Scholar
    • Export Citation
  • Tiedtke, M., 1984: The sensitivity of the time mean large-scale flow to cumulus convection in the ECMWF model. Report of the Workshop on Convection in Large Scale Numerical Models, Reading, United Kingdom, ECMWF, 297–316.

    • Search Google Scholar
    • Export Citation
  • ——,. 1989: A comprehensive mass flux scheme for cumulus parameterization in large-scale models. Mon. Wea. Rev., 117 , 17791800.

  • Trenberth, K. E., 1998: Development and forecasts of the 1997–98 El Niño. CLIVAR scientific issues. CLIVAR-Exch., 3 , 414.

  • Ward, M. N., and C. K. Folland, 1991: Prediction of seasonal rainfall in the north Nordeste of Brazil using eigenvectors of sea surface temperature. Int. J. Climatol., 11 , 711743.

    • Search Google Scholar
    • Export Citation
  • Webster, P. J., A. M. Moore, J. P. Loschnigg, and R. R. Leben, 1999: Coupled ocean–atmosphere dynamics in the Indian Ocean during 1997–98. Nature, 401 , 356360.

    • Search Google Scholar
    • Export Citation
  • Zheng, X., and C. S. Frederiksen, 1999: Validating interannual variability in an ensemble of AGCM simulations. J. Climate, 12 , 23862396.

    • Search Google Scholar
    • Export Citation
  • Zwiers, F. W., 1987: A potential predictability study conducted with an atmospheric general circulation model. Mon. Wea. Rev., 115 , 29572974.

    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

Overall LEPS skill scores for SSTA forecasts 4, 8, and 12 weeks ahead from week 0 using persisted SSTAs during the period Mar 1997–Mar 1998. Skill scores that are positive and statistically significant at the 95% level are shown shaded

Citation: Journal of Climate 14, 12; 10.1175/1520-0442(2001)014<2675:DSFDTE>2.0.CO;2

Fig. 2.
Fig. 2.

Correlations between the observed SOI and the NCEP–NCAR reanalysis for the period Jan 1958–Nov 1991 for precipitation, surface air temperature, 200-hPa geopotential height, and MSLP. A 3-month running mean filter has been applied to all time series. Positive correlations are shown shaded

Citation: Journal of Climate 14, 12; 10.1175/1520-0442(2001)014<2675:DSFDTE>2.0.CO;2

Fig. 3.
Fig. 3.

As in Fig. 2, but from an ensemble of five runs of the BMRC AGCM forced by observed SSTs and sea-ice

Citation: Journal of Climate 14, 12; 10.1175/1520-0442(2001)014<2675:DSFDTE>2.0.CO;2

Fig. 4.
Fig. 4.

Seasonal anomalies of 200-hPa geopotential height derived from the NCEP–NCAR reanalysis for the period AMJ97–MAM98. Positive anomalies are shown shaded

Citation: Journal of Climate 14, 12; 10.1175/1520-0442(2001)014<2675:DSFDTE>2.0.CO;2

Fig. 5.
Fig. 5.

As in Fig. 4, but for the K forecasts

Citation: Journal of Climate 14, 12; 10.1175/1520-0442(2001)014<2675:DSFDTE>2.0.CO;2

Fig. 6.
Fig. 6.

The LEPS skill score for each of the K forecasts of 200-hPa geopotential height anomalies. Positive skill is shown shaded

Citation: Journal of Climate 14, 12; 10.1175/1520-0442(2001)014<2675:DSFDTE>2.0.CO;2

Fig. 7.
Fig. 7.

The overall LEPS skill score, significant at a 95% level, for the K forecasts (top) and M forecasts (bottom) of 200-hPa geopotential height anomalies taken over all forecasts AMJ97–AMJ98. Positive and significant skill is shown contoured and shaded. Areas of negative skill are shown blank and uncontoured

Citation: Journal of Climate 14, 12; 10.1175/1520-0442(2001)014<2675:DSFDTE>2.0.CO;2

Fig. 8.
Fig. 8.

As in Fig. 6, but for the K forecasts of MSLP anomalies

Citation: Journal of Climate 14, 12; 10.1175/1520-0442(2001)014<2675:DSFDTE>2.0.CO;2

Fig. 9.
Fig. 9.

As in Fig. 7, but for K forecasts and M forecasts of MSLP anomalies

Citation: Journal of Climate 14, 12; 10.1175/1520-0442(2001)014<2675:DSFDTE>2.0.CO;2

Fig. 10.
Fig. 10.

As in Fig. 6, but for the K forecasts of surface air temperature anomalies

Citation: Journal of Climate 14, 12; 10.1175/1520-0442(2001)014<2675:DSFDTE>2.0.CO;2

Fig. 11.
Fig. 11.

As in Fig. 7, but for K forecasts and M forecasts of surface air temperature anomalies

Citation: Journal of Climate 14, 12; 10.1175/1520-0442(2001)014<2675:DSFDTE>2.0.CO;2

Fig. 12.
Fig. 12.

As in Fig. 6, but for the K forecasts of precipitation anomalies

Citation: Journal of Climate 14, 12; 10.1175/1520-0442(2001)014<2675:DSFDTE>2.0.CO;2

Fig. 13.
Fig. 13.

As in Fig. 7, but for for K forecasts and M forecasts of precipitation anomalies

Citation: Journal of Climate 14, 12; 10.1175/1520-0442(2001)014<2675:DSFDTE>2.0.CO;2

Fig. 14.
Fig. 14.

The overall skill score, significant at the 95% level, for model simulation of seasonal anomalies of precipitation, surface air temperature, 200-hPa geopotential height, and MSLP for the period Jan 1958–Nov 1991. Positive and significant skill is shown contoured and shaded. Areas of negative skill are shown blank and uncontoured

Citation: Journal of Climate 14, 12; 10.1175/1520-0442(2001)014<2675:DSFDTE>2.0.CO;2

Fig. 15.
Fig. 15.

The fraction of the total interseasonal variance of precipitation, surface air temperature, 200-hPa geopotential height, and MSLP that is attributable to SST/sea-ice forcing

Citation: Journal of Climate 14, 12; 10.1175/1520-0442(2001)014<2675:DSFDTE>2.0.CO;2

Save