Seasonal and Decadal Prediction

Oscar Alves, Debra Hudson, Magdalena Balmaseda and Li Shi

Abstract Dynamical seasonal prediction has grown rapidly over the last decade or so. At present, a number of operational centres issue routine seasonal forecasts produced with coupled ocean-atmosphere models. These require real-time knowledge of the state of the global ocean since the potential for climate predictability at seasonal time scales resides mostly in information provided by the ocean initial conditions, in particular the upper thermal structure. The primary aim of the coupled model is to predict sea surface temperature variability and how this variability impacts regional climate through large scale teleconnections.

This paper reviews recent advances in dynamical seasonal prediction using coupled ocean-atmosphere models. It discusses the sources of predictability at seasonal time scales, the probabilistic nature of seasonal forecasts, the ensemble methods used to deal with it, and the current levels of skill. The ocean initialisation receives special focus, with a discussion on initialisation strategies, ocean data assimilation methods, and the role of the observing system in seasonal forecast skill.

Assimilation of observations into an ocean model forced by prescribed atmospheric fluxes is the most common practice for initialisation of the ocean component of a coupled model. Assimilation of ocean data reduces the uncertainty in the ocean estimation arising from the uncertainty in the forcing fluxes and from model errors. Although data assimilation also usually improves the skill of seasonal forecasts, its impact is often overshadowed by errors in the coupled models.

The paper also briefly discusses decadal prediction, for which there is growing demand, particularly in the context of climate change adaptation. Although decadal prediction is still in its infancy, recent development shows promising results, highlighting the role of ocean initial conditions. The initialisation of the ocean for decadal predictions is a major challenge for the next decade.

Bureau of Meteorology, Centre for Australian Weather and Climate Research (CAWCR), GPO Box 1289, Melbourne, VIC 3001, Australia e-mail: [email protected]

A. Schiller, G. B. Brassington (eds.), Operational Oceanography in the 21st Century, 513

DOI 10.1007/978-94-007-0332-2_20, © Springer Science+Business Media B.V. 2011

20.1 Introduction

Dynamical seasonal prediction has grown rapidly over the last decade or so. At present, multiple operational centres routinely issue seasonal forecasts produced with coupled ocean-atmosphere models (e.g., Fig. 20.1). The basis of dynamical seasonal prediction resides in variability driven by slow-processes in the climate system, particularly the ocean. The El Nino Southern Oscillation (ENSO) is the most prominent mode of climate variability on seasonal to interannual timescales and is the major source of predictability. The success of dynamical seasonal prediction is therefore often related to the ability to initialise and forecast ENSO, as well as capturing its teleconnections to regional climates. This paper focuses on dynamical seasonal prediction with coupled ocean-atmosphere models. Early efforts with dynamical prediction used atmosphere-only general circulation models, but today most operational centres use fully coupled ocean-atmosphere general circulation models.

The starting point for dynamical seasonal prediction is specifying the initial state of the climate system. Seasonal prediction is generally viewed as an ocean initial

Model Forecasts of ENSO from Apr 2010

Model Forecasts of ENSO from Apr 2010

JFM Mar MAM 2010


Fig. 20.1 Sample forecasts of El Nino produced by international models and assembled by the IRI. (

Dynamical Model:


Statistical Model:



JFM Mar MAM 2010


Fig. 20.1 Sample forecasts of El Nino produced by international models and assembled by the IRI. (

condition problem, but there are also benefits from realistic atmosphere (e.g., Hudson et al. 2010) and land (e.g., Koster et al. 2010) initial conditions. Data assimilation can improve forecasts by correcting the model state and/or variability, but it can also create problems such as initialisation shock. Recent studies examining the impact of ocean and atmosphere initialisation on seasonal forecast skill concluded that the most skilful initialisation scheme is that which makes the most use of the observed data, even though initial imbalances in the coupled state are generated (Balmaseda and Anderson 2009; Hudson et al. 2010). To date, initialisation of the ocean and atmosphere is done separately, although there are emerging attempts at approaching initialisation as a coupled ocean-atmosphere problem, where the component models are well-balanced. This is not trivial, particularly given the different time scales upon which the atmosphere and ocean operate.

Seasonal prediction is inherently uncertain and needs to be addressed in a probabilistic framework. Dynamical seasonal prediction aims to address these uncertainties and the chaotic nature of the atmosphere by producing an ensemble of forecasts. Perturbations to the initial state or model formulation generate forecasts that diverge, producing a range of possible future outcomes from which probabilistic forecasts can be produced. Ideally, generation of the ensemble should take into account uncertainties in the initial conditions (e.g., Vialard et al. 2005), as well as uncertainties associated with imperfect models (e.g., Murphy et al. 2004; Berner et al. 2008). New ocean assimilation schemes represent the uncertainty in the ocean state by producing an ensemble of ocean initial conditions (Balmaseda et al. 2008; Yin et al. 2011).

Coupled models are far from perfect and drift with forecast lead time towards the biased coupled model climate. A common approach is to remove the drift a-posteriori (e.g., Stockdale 1997). A set of retrospective forecasts (or hindcasts) is produced to provide an estimate of how the model climatology changes with lead time, and this is then used for a-posteriori calibration of the forecast results. Ideally the hindcasts should span as long a period as possible, but in practice most centres only produce hindcasts over a 15-30 year period. The hindcasts are also needed for skill assessment of the seasonal forecast system. Implicit in the production of a set of retrospective forecasts is the need for ocean initial conditions spanning the chosen hindcast period, equivalent to an ocean "reanalysis" of the historical data stream. The interannual variability represented by the ocean reanalysis (particularly due to changes in the ocean observing system) will have an impact on both forecast calibration and the assessment of skill.

This paper provides a review of dynamical seasonal prediction, with a focus on the initialisation of seasonal forecasts. Section 20.2 describes the primary drivers of seasonal prediction skill, Sect. 20.3 summarises current levels of skill and Sect. 20.4 provides some background behind ensemble prediction. Sects. 20.5 and 20.6 focus on data assimilation and initialisation and in particular the role of ocean observations. Section 20.7 provides an example of seasonal prediction in the Australian context. Section 20.8 introduces decadal prediction, which relies heavily on the ocean initialisation. Finally, a summary is provided in Sect. 20.9. Four recent review papers provide additional, more detailed reading on the topic of this chapter:

one documenting the current status of seasonal prediction and our understanding of seasonal to interannual climate variability (produced for the Copenhagen World Climate Conference 3; Stockdale et al. 2010), two focussing on the initialisation of seasonal and decadal forecasts and the role of ocean observations (Balmaseda et al. 2010a, b) and one reviewing the status of decadal prediction (Hurrell et al. 2010).

20.2 Predictability: What is the Source of Seasonal Prediction Skill?

Predictability is a feature of the climate system and cannot be changed or improved by forecast methodologies—it represents the theoretical upper limit of our prediction skill. This maximum level of predictability has not yet been achieved in seasonal forecasting: forecast skill is limited by model error, imperfect initialisation and the fact that not all the interactions in the climate system are currently fully resolved i.e. there may be sources of predictability that are unaccounted for (Kirt-man and Pirani 2009). An understanding of climate variability and its key drivers offers insight into the processes providing predictability, as well as into how model shortcomings may be limiting forecast skill. Climate variability occurs on all tim-escales. Atmospheric processes tend to vary over short timescales (less than a few days) and are a source of unpredictable noise for seasonal prediction. Processes operating over longer timescales, primarily those associated with the ocean, form the basis of seasonal predictability. Apart from the ocean, other potential sources of seasonal predictability include: the longer timescales of variability of the coupled ocean-atmosphere system, sea-ice, soil conditions, snow cover and the state of the stratosphere (Stockdale et al. 2010).

ENSO is the most prominent mode of climate variability on seasonal to interannual timescales and is the major source of predictability. Although mainly associated with coupled ocean-atmosphere variations in the tropical Pacific (Walker 1923, 1924; Bjerknes 1969), the effects of ENSO can be felt globally, with teleconnec-tions to regional temperature and precipitation in many countries (e.g., Rasmusson and Carpenter 1983; Ropelewski and Halpert 1987). For example, El Nino events are typically associated with above average rainfall in Peru and Ecuador, northern Argentina, East Africa and California, and dryer than normal conditions over Australia, southern Africa and parts of the Amazon basin. Figure 20.2 shows the sea surface temperature (SST) anomaly during December 1997, near the peak of the El Nino. This was the largest El Nino of the century, with SST anomalies peaking over 4°C in the eastern Pacific. For reviews of our understanding of ENSO and the mechanisms involved, see, for example, Neelin et al. (1998); Philander (2004) and Chang et al. (2006). The first successful prediction of ENSO with a simple coupled ocean-atmosphere dynamical model was produced by Zebiak and Cane (1987). Since then, increasingly complex and comprehensive coupled ocean-atmosphere models have been developed and dynamical prediction of ENSO is now commonplace in major operational centres.

Fig. 20.2 Sea surface temperature anomalies during December 1997

Low-frequency coupled ocean-atmosphere variations in the Indian and Atlantic Oceans, although less dominant than the Pacific, can also drive temperature and precipitation anomalies on seasonal timescales across the globe (e.g., Goddard and Graham 1999; Folland et al. 2001; Rodwell and Folland 2002; Saji and Yamagata 2003; Kushnir et al. 2006; Ummenhofer et al. 2009). The Indian Ocean Dipole (IOD) has been identified as a low frequency coupled mode of variability in the tropical Indian Ocean (Saji et al. 1999; Webster et al. 1999). In Fig. 20.2 an IOD event can be seen in the Indian Ocean, with negative SST anomalies in the east off the Java-Sumatra coast and positive anomalies in the west. IOD events, like the one in Fig. 20.2, are often triggered by easterly wind anomalies as a result of the atmospheric response to the development of El Nino. The IOD is much less predictable (practically and theoretically) than ENSO (e.g., Luo et al. 2007; Wajsowicz 2007; Zhao and Hendon 2009), largely due to weaker surface-subsurface ocean coupling, strong interactions with the Australian-Asian monsoon and intraseasonal oscillations causing chaotic forcings in both the ocean and atmosphere (Zhao and Hendon

2009). Although the IOD is a measure of the difference between the western and eastern parts of the equatorial Indian Ocean, these two components are not always related and the skill of each component can be different. The lack of skill of the IOD is mainly due to a lack of skill in predicting the SST in the eastern component of the IOD. Other modes of atmospheric variability (not necessarily related to oceanic forcing) that may provide predictive skill on seasonal timescales, include the Northern Annular and Southern Annular modes (NAM and SAM), the Pacific North American (PNA) pattern and the North Atlantic Oscillation (NAO) (Stockdale et al.

The land surface is a potential source of seasonal predictability, primarily associated with soil moisture memory in the earth-atmosphere system (e.g., Fennessy and Shukla 1999; Koster and Suarez 2003; Seneviratne et al. 2006; Koster et al. 2004, 2010), although anomalous snow cover/amount may also be important (e.g., Fletcher et al. 2009). The coordinated approach of the Global Land-Atmosphere Coupling Experiment (GLACE; Koster et al. 2006, 2010), using a variety of state-of-the-art seasonal forecasting systems, has significantly improved our understanding of the role of land surface processes in seasonal prediction.

Fig. 20.2 Sea surface temperature anomalies during December 1997

There have also been suggestions that the stratosphere could make a contribution to seasonal prediction skill in the troposphere, particularly in the Northern Hemisphere (e.g., Baldwin and Dunkerton 2001; Ineson and Scaife 2008; Bell et al. 2009; Cagnazzo and Manzini 2009). However, most contemporary seasonal prediction models have a poorly resolved stratosphere and do not give a realistic representation of stratospheric circulation (Maycock et al. 2009). A recent study by Marshall and Scaife (2009) suggests that improving the resolution of the Quasi-Biennial Oscillation (QBO), a dominant mode of variability in the tropical stratosphere, could improve seasonal prediction of QBO-induced surface anomalies over Europe.

Was this article helpful?

0 0
Guide to Alternative Fuels

Guide to Alternative Fuels

Your Alternative Fuel Solution for Saving Money, Reducing Oil Dependency, and Helping the Planet. Ethanol is an alternative to gasoline. The use of ethanol has been demonstrated to reduce greenhouse emissions slightly as compared to gasoline. Through this ebook, you are going to learn what you will need to know why choosing an alternative fuel may benefit you and your future.

Get My Free Ebook

Post a comment