Forecast Skill

As mentioned in Sect. 20.2, ENSO is the most predictable large-scale phenomenon on seasonal to interannual timescales, and is the major source of predictability. Successful predictions with a coupled seasonal forecast model are, therefore, often related to a model's ability to reproduce the slow coupled dynamics of ENSO and accurately forecast its amplitude, spatial pattern and detailed temporal evolution (Wang et al. 2008a). The skill of forecasting ENSO varies depending on the season, as well as the phase and intensity of ENSO. For example, there is usually greater skill at predicting ENSO events compared to neutral events, and predicting the growth phases of warm and cold events compared to the corresponding decaying phases (e.g., Jin et al. 2008). In terms of season, many seasonal forecast systems experience a decline in skill during the boreal spring, often referred to as the "spring predictability barrier". At this time of year, SST anomalies are particularly variable and although dynamical forecast models may have reduced skill, their advantage over persistence forecasts is at a maximum (e.g., van Olden-borgh et al. 2005; Jin et al. 2008; Wang et al. 2008a). Large multi-model projects, such as DEMETER (Palmer et al. 2004), ENSEMBLES (Weisheimer et al. 2009) and APCC/CliPAS (Wang et al. 2008a), have provided a basis for intercomparing the skill and errors from coupled models, benchmarking seasonal prediction skill and assessing progress. Weisheimer et al. (2009) report that results from the European ENSEMBLES project (using 5 European coupled models) have shown a significant reduction in the systematic SST errors (SST drift over the Pacific as the forecast progresses) compared to the previous generation project, DEMETER. For the NINO3 region (5°S-5°N; 150°W-90°W) the SST drift in DEMETER varied between +2°C and -7°C for up to 6 months lead, whereas the drift from the ENSEMBLES models was less than ±1.5°C (Weisheimer et al. 2009). They conclude that since DEMETER, the coupled models have improved significantly in terms of their physical parameterisations, resolution and initialisation. They also show that although probabilistic skill scores suggested increases in SST prediction skill in the 4-6 month forecast range in the ENSEMBLES multi-model ensemble (MME) compared to the DEMETER MME, the increases were not statistically significant, suggesting that substantially better models (perhaps with a higher resolution than available now) are required to improve upon the current skill of forecasting tropical Pacific SSTs. As an example of current skill levels, the anomaly correlation skill in predicting NINO3.4 SST anomalies (an area average over 5°N-5°S, 170°-120°W) from an ensemble of 10 coupled seasonal forecast models (for hindcasts performed over 1980-2001) is 0.86 after 6 months of the forecast (Jin et al. 2008). This level of skill from the MME is greater than from any single model, but at this lead time all models have skill greater than persistence and many of the models have anomaly correlation skills exceeding 0.8 (Jin et al. 2008).

Skill in predicting Indian Ocean SST anomalies is lower than over the Pacific. This is clear from Fig. 20.3 which shows the anomaly correlation skill of predicting SST anomalies at 6 months lead time from the POAMA (Predictive Ocean Atmosphere Model for Australia) seasonal forecast model. This is very typical of most seasonal forecast models. Prediction of the IOD is currently limited to about one season, with a strong boreal winter-spring predictability barrier (partly because the IOD is not well defined prior to June) (e.g., Luo et al. 2007; Wajsowicz 2007; Zhao and Hendon 2009). In terms of tropical Atlantic SST anomalies, current seasonal prediction models show very little skill beyond one or two months of the forecast and skill is often no better than persistence (e.g., Stockdale et al. 2006, 2011).

The forecast skill of regional surface air temperature and precipitation anomalies is strongly dependent on season and region. Skill is highest in the tropics and decreases towards middle and high latitudes, and is usually higher for temperature than precipitation (e.g., Wang et al. 2008a; Doblas-Reyes et al. 2009). At 1-month lead there is very little skill in predicting seasonal mean temperature and precipitation anomalies over land in extra-tropical regions (e.g., Wang et al. 2008a; Doblas-Reyes et al. 2009). Those extra-tropical land regions that do exhibit some skill (e.g., southern Africa and the southern United States for precipitation in DJF) are usually a result of the models capturing the atmospheric teleconnections from ENSO. Consequently, model bias and drift in the simulation of ENSO may degrade global tele-connections to regional rainfall and temperature. For example, most models exhibit a cold bias in the central equatorial Pacific and a westward drift of maximum SST variability away from the eastern Pacific with increasing lead time (e.g., Jin et al.

2008). In the POAMA seasonal forecast model, after about a season, these biases hinder the model's ability to discern between different types of ENSO events (e.g., classical east Pacific versus central Pacific events) and the teleconnection between ENSO and Australian climate is adversely affected (Hendon et al. 2009; Lim et al.

20.4 Ensemble Prediction: Representing Uncertainty

There is considerable uncertainty inherent in seasonal predictions, some natural and some due to deficiencies in the forecasting systems. Figure 20.4 shows 90 forecasts of the onset of the 1997/1998 El Nino. Each forecast was produced using the POAMA-1 model (Alves et al. 2003). The ensemble was generated by making

Fig. 20.3 SST anomaly correlation at 6 month lead time from POAMA-1.5 forecasts (left) and persistence (right). (From Wang et al. 2008b)

POAMA Nirio3.4 SST Anomaly (Olv2 SST)

POAMA Nirio3.4 SST Anomaly (Olv2 SST)

/./ fk/

f

JUL98 SEP NOV JAN 9 7 WAR MAY JUL SEP NOV JAN98

Fig. 20.4 Forecasts NINO3.4 SST anomaly during the onset of the 1997/1998 El Nino. A 90-mem-ber ensemble, where each ensemble member is generated by applying a 0.001C random perturbation to the initial SST. (From Shi et al. 2009)

JUL98 SEP NOV JAN 9 7 WAR MAY JUL SEP NOV JAN98

Fig. 20.4 Forecasts NINO3.4 SST anomaly during the onset of the 1997/1998 El Nino. A 90-mem-ber ensemble, where each ensemble member is generated by applying a 0.001C random perturbation to the initial SST. (From Shi et al. 2009)

0.001°C changes to the initial SST. These changes are physically insignificant, but because the climate system, in particular the atmosphere, is chaotic, the ensemble members can spread rapidly with time. The plot shows that while all of the forecast were for El Nino conditions, they range from a very weak El Nino with NINO3.4 SST anomalies of around 0.5°C to very strong El Nino conditions with NINO3.4 anomalies greater than 2.5°C by August. The spread in the forecasts indicates the stochastic component of the climate model, i.e. natural uncertainty and therefore the limits to predictability.

In a seasonal forecast system the ensemble spread should be commensurable to the uncertainty arising from natural stochastic processes, but this is not always the case due to errors in the forecast system. For practical reasons, the uncertainty is classified into that arising from an imperfect initial state (inital conditions uncertainty) and that arising from imperfect models (model data sampling uncertainty, model parametric uncertainty, model structural uncertainty). In dynamical seasonal prediction, ensembles are used to quantify the uncertainty (e.g., Stephenson 2008; Doblas-Reyes et al. 2009). Uncertainties in the initial conditions are taken into account by generating an ensemble from slightly different atmospheric and/or ocean analyses, where the differences are intended to reflect the uncertainty in these conditions (e.g., Vialard et al. 2005). Uncertainties in model formulation have been addressed using ensembles based on stochastic physics (Jin et al. 2007; Berner et al. 2008), perturbed parameter (Murphy et al. 2004; Stainforth et al. 2005; Collins et al. 2006) and multi-model approaches (Palmer et al. 2004; Weisheimer et al. 2009). Doblas-Reyes et al. (2009) assessed the relative merits of these three approaches using sets of seasonal and decadal hindcasts (done under the auspices of the European ENSEMBLES project; see van der Linden and Mitchell 2009). In general, they concluded that the three methods had comparable overall skill (the multi-model was slightly better for lead times up to 4 months, and the perturbed physics slightly better at longer leads). The perturbed physics and stochastic parameter methods are promising methods of sampling model uncertainty within a single model system.

Probabilistic forecasts are produced from dynamical seasonal forecasting systems by using the aforementioned ensemble of forecasts. The forecasts follow different evolutions because they are produced from perturbed initial conditions or model formulations. After the first week, the ensemble spread is large and the forecast needs to be delivered and assessed in a probabilistic fashion. Good reviews of probability forecasting in a seasonal context, including basic concepts, recalibra-tion and verification, are provided by Stephenson (2008) and Mason and Stephen-son (2008). The distribution of the ensemble members should indicate uncertainty in the forecast: if the forecasts from the ensemble members differ widely, the inferred probability distribution is also wide and the forecast is uncertain, whereas if the ensemble members are in close agreement it might suggest less uncertainty. However, in practice, forecasts from dynamical seasonal forecast models tend to be overconfident, i.e. their spread is too narrow to match the range of observed outcomes, and there is often little relationship between ensemble spread and the error in the forecast. The prime reason for this is believed to be model error (Vial-ard et al. 2005; Stockdale et al. 2010). Multi-model approaches, where ensembles from different state-of-the-art models are combined, thereby implicitly averaging out some of the model errors, generally produce more skilful forecasts than do the results from a single model (Palmer et al. 2004; Wang et al. 2008a; Weisheimer et al. 2009). A counter example of the limitation of the multi-model approach is provided by Balmaseda et al. (2010b), showing that for a given SST index, the skill of a single model can be superior to that of the multi-model product. But this is not yet the case for useful atmospheric variables such as precipitation, where reliable seasonal forecasts benefit from the multi-model approach. Multi-model forecast systems are becoming increasingly apparent in operational seasonal forecasting. For example, the APEC Climate Center (APCC) produces real-time operational climate predictions based on a well-validated multi-model multi-institute ensemble system (http://www.apcc21.org) and ECMWF has collaborated with France and the United Kingdom to produce an operational multi-model seasonal forecast system known as EUROSIP (http://www.ecmwf.int/products/forecasts/seasonal/ documentation/eurosip/).

20.5 Data Assimilation and Initialization

Dynamical seasonal prediction is essentially an initial value problem, where predictive skill comes from information contained in the initial states of the coupled system: ocean, atmosphere, land and sea-ice. Most of the skill comes from the initial conditions of the upper ocean, particularly those associated with large scale patterns of variability such as ENSO and the IOD. Assimilation of ocean observations for ocean initialisation in seasonal forecasts has become a common practice, with several institutions around the world producing routine ocean re-analyses to initialise their operational seasonal forecasts. Table 20.1, from Balmaseda et al. (2009), provides a summary of the ocean analyses used for initialisation of operational or quasi-operational seasonal forecast systems. In all these systems, the initialisation of the ocean and atmosphere is done separately, aiming at generating the best analyses of the atmosphere and ocean through comprehensive data assimilation schemes.

The simplest way to initialise the tropical ocean is to run an ocean model forced with atmospheric fluxes and with a strong relaxation of the model SST to observations. Inter-annual variability in the tropical ocean is to a large extent driven by variability in the surface wind field. This technique would be satisfactory if errors in the forcing fields and ocean model were small. However, surface flux products and ocean models are both known to have significant errors. Assimilation of ocean observations is then used to constrain the estimation of the ocean state.

In ocean assimilation, ocean sub-surface observations are ingested into an ocean model forced by prescribed atmospheric fluxes. The emphasis is on the initialisation of the upper ocean thermal structure, particularly in the tropics, where SST anomalies have a strong influence on the atmospheric circulation. Most of the initialisation systems use observed subsurface temperature (from XBT's, TAO/TRITON/ PIRATA and Argo). Some of the more recent systems also use salinity (mainly from

Table 20.1 Summary of different ocean assimilation systems used in the initialisation of operational and quasi-operational seasonal forecasts. (Based on Balmaseda et al. 2009) MRI-JMA http://ds.data.jma.go.jp/tcc/tcc/products/elnino/index.html Multi-variate 3-dimension Variational (3D-VAR). Usui et al. 2006

ORA-S3 (ECMWF System 3) http://www.ecmwf.int/products/forecasts/d/charts/ocean/real_time/ Multivariate Optimum Interpolation (OI). Balmaseda et al. 2008

POAMA -PEODAS (CAWCR, Melbourne) http://poama.bom.gov.au/research/assim/index.htm Multivariate Ensemble OI. Yin et al. 2011

GODAS (NCEP) http://www.cpc.ncep.noaa.gov/products/GODAS/ 3D-VAR. Behringer 2007

MERCATOR (Meteo France) http://bulletin.mercator-ocean.fr/html/welcome_en.jsp Multivariate reduced order Kalman filter. Pham et al. 1998 MO (MetOffice) http://www.metoffice.gov.uk/research/seasonal/ Multivariate OI. Martin et al. 2007

GMAO ODAS-1 http://gmao.gsfc.nasa.gov/research/oceanassim/ODA_vis.php

GMAO Seasonal Forecasts: http://gmao.gsfc.nasa.gov/cgi-bin/products/climateforecasts/index.cgi

OI and Ensemble Kalman Filter Keppenne et al. 2008

Argo), and altimeter derived sea-level anomalies. The latter usually needs the prescription of an external Mean Dynamic Topography, which can be a problem, and is usually taken from a model integration rather than observations. In the longer term it is hoped that it can be derived indirectly from gravity missions such as GRACE and GOCE.

Several studies have demonstrated the benefit of assimilating ocean data on the prediction of ENSO (e.g., Alves et al. 2004; Dommenget et al. 2004; Cazes-Boezio et al. 2008; Stockdale et al. 2011). The benefits are less clear in other areas, such as the equatorial Atlantic, where model errors are large. Balmaseda and Anderson (2009) evaluated three different initialisation strategies, each of which used different observational information. They showed that the ocean initialisation has a significant impact on the mean state, variability and skill of coupled forecasts at the seasonal time scale. They also showed that, using their model, the initialisation strategy that makes the most comprehensive use of the available observations leads to the best skill.

Since ocean assimilation is important for seasonal prediction, an interesting question is: how accurate are ocean analyses from ocean assimilation systems? Figure 20.5 shows the composite El Nino evolution of heat content along the equator in the Pacific and Indian Oceans. The composite plots consist of 30 months spanning each El Nino event from -9 months (year prior to warm event), 12 months (warm event), to +9 months (year after warm event), and these are denoted as Year -1, Year 0 and Year +1 respectively. The selection criteria for El Niño/La Niña events is defined as the monthly Niño3 SST anomaly reaches or exceeds ±0.5°C for at least 5 consecutive months over the period 1982 to 2006.

Composites from two state of the art international analyses are shown to illustrate how they differ and give an indication of the level of error in the analysis. The assimilation systems used to generate each analysis are quite different and so are the forcing fields used to drive the ocean model during the re-analysis phase. The composite El Nino evolution shows El Nino peaking at the end of the year with maximum heat content anomalies in eastern Pacific. At the same time there are heat content anomalies in western Pacific, forming a strong gradient between east and west Pacific, which is driven by anomalous westerly winds (not shown). Normally during the peak of El Nino there are also easterly winds in the Indian Ocean, which leads to an east-west pattern that is the reverse of the pattern in the Pacific. The composites also show the evolution of positive heat content anomalies from the western Pacific at the beginning of the year where El Nino develops, towards the eastern Pacific through the action of equatorial Kelvin waves. There is considerable agreement between the two re-analyses, likely to be due to a reasonable observing network, particularly in the Pacific with the TOGA-TAO array and this decade with Argo.

The same is not true of salt content. Figure 20.6 compares the evolution of salt content along the equator for the same El Nino composite. One re-analysis shows significant salt anomalies throughout the equatorial Pacific during El Nino, while the other shows weaker anomalies. For example, the first re-analysis shows strong freshening in the central/west Pacific at the peak of El Nino of just over 0.1ppt,

Composite T300 Anom (5S-5N) for EINino PEODAS

Composite T300 Anom (5S-5N) for EINino GODAS

Composite T300 Anom (5S-5N) for EINino PEODAS

Composite T300 Anom (5S-5N) for EINino GODAS

40E 60E 80E 100E120E 140E 160E 180160W 140W120W100W 80W

40E 60E 80E 100E120E 140E160E 180 160W 140W120W 100W 80W

40E 60E 80E 100E120E 140E 160E 180160W 140W120W100W 80W

40E 60E 80E 100E120E 140E160E 180 160W 140W120W 100W 80W

Fig. 20.5 Evolution of a composite El Nino in two different state of the art ocean re-analyses. Each plot shows the evolution of heat content anomalies (upper 300 m) along the equator for a composite El Nino. The plot covers the period from April of the year prior to El Nino developing, to September of the year after El Nino develops. The same El Nino events were included in both composites

Fig. 20.6 As for Fig. 20.5 except for salt content of the upper 300 m

Composite S300 Anom (5S-5N) for EINino GODAS

-0.16 -0.14 -0.12 -0.1 -0.08 -0.06 -0.04 -0.02 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16

presumably due to eastward advection of fresh water associated with the anomalous westerlies. However, the second re-analysis does not show such strong anomalies, generally less than 0.04ppt. This clearly indicates that, at least at present, there are significant differences in how state-of-the-art ocean re-analyses represent inter-annual variability of salinity. It has been shown (Balmaseda and Weaver 2006) that in the absence of salinity data, the assimilation of temperature observations can increase the uncertainty in the salinity field. The salinity field can influence the seasonal forecasts by influencing the barrier layer, which acts as a reservoir of warm water (above 28°C), and can be instrumental for the development of El Niño when propagated eastward by westerly winds (Fujii et al. 2011).

Interestingly both re-analyses show similar salt content patterns in the Indian Ocean. This is probably due to the lack of salinity and temperature data in the Indian Ocean, at least before Argo. Without much temperature and salinity data, the re-analyses are simply ocean simulations driven by surface forcing, which is likely to lead to similar patterns.

There are three main ways of evaluating ocean analyses produced by data assimilation systems: (1) how well the analysis fits the assimilated observations, (2) how well the analysis fits independent observations and (3) whether the analysis leads to improved forecasts. Way (3) may not be a reliable method because if the models have significant errors a better initial state could potentially lead to a worse forecast. Way (1) is also not entirely satisfactory since it simply reflects how well the analysis fits the observations, which is mostly a function of the background and observation error variances. Way (2) is the most desirable way, but it can be difficult since usually all temperature and salinity observations are used in the assimilation. To date no assimilation system utilises ocean current data. This is one source of independent data, and Fig. 20.7 illustrates the use of ocean current data to evaluate different re-analyses. Figure 20.7 shows the correlation between re-analyses and pseudo-observed ocean surface currents (the currents are derived from altimeter data: OSCAR; Bonjean and Lagerloef 2002). Three re-analyses are assessed against the OSCAR current data. Figure 20.7a uses the PEODAS re-analysis (Yin et al. 2011) which is representative of a current generation ocean re-analysis. It makes dynamically balanced corrections to the currents based on the temperature and salinity corrections. The current corrections are based on the cross-covariances derived from a time evolving ensemble, see Yin et al. (2011) for more details. Figure 20.7b shows the correlation from a control re-analysis, i.e. the same as PEODAS except no observations are assimilated. This is essentially an ocean model forced with re-analysis surface fluxes and will do a reasonable job of representing the inter-annual variability, at least as far as it is represented in the forcing fields. Figure 20.7c uses a re-analysis from an older generation ocean assimilation system, in this case from the POAMA-1 seasonal prediction system (Alves et al. 2003). Typical of this generation, only temperature observations (and not salinity) are assimilated. However, corrections to currents are made based on the temperature corrections by assuming geostrophic balance, as in Burgers et al. (2002). In all three re-analyses no altimeter data are assimilated. The figures show that the PEODAS re-analysis produces the best correlation with the observed data in both the tropical Pacific and Indian

Fig. 20.7 Correlations between the zonal surface velocity from OSCAR and a PEODAS. b Control, and c POAMA-1. Note the non-linear correlation scale. (From Yin et al. 2011)

Oceans. Interestingly the last generation POAMA-1 system produces the worst comparison to observations, even worse than the control which uses no data. This is likely for two reasons. Firstly, salinity data are not assimilated in POAMA-1, which can lead to incorrect density profiles since density corrections are only based on temperature, which in turn can lead to wrong current increments when using the geostrophic relation. Secondly, the geostrophic relation may not be appropriate, especially for the surface current which has a significant Ekman component. While the control re-analysis does not use any observations, it does maintain a surface current that is in dynamical balance with the surface forcing and the pressure fields. These results illustrate the progress over the last decade that has led to the current state of the art in ocean data assimilation.

Ensemble based data assimilation schemes, such as Ensemble Kalman Filters, provide an ensemble of analyses. The spread of the ensemble members represents the uncertainty in the estimated ocean state and the standard deviation of the ensemble spread about the ensemble mean can be considered a measure of the analyses error. Ensemble spread from the PEODAS ocean assimilation scheme (Yin et al. 2011) is shown in Fig. 20.8. The highest spread in SST (Fig. 20.8a) occurs in the eastern equatorial Pacific and along the western boundary currents, as one might expect as these are the regions of highest variability. The highest spread in surface salinity (Fig. 20.8b) occurs in regions of highest rainfall, such as along the Inter-Tropical Convergence Zone, the South Pacific Convergence Zone and the high rainfall regions of the West Pacific warm pool. Figure 20.8c shows the temperature ensemble spread at depth along the equator. Maximum spread occurs along the thermocline, the region of maximum temperature variability. Maximum salinity spread (Fig. 20.8d) occurs at the surface.

20.6 The Impact of Ocean Observations

The ocean observing system has undergone major changes over the last couple of decades. In the early 1990s the TOGA-TAO array in the tropical Pacific was introduced. This allowed the heat content of the equatorial upper ocean to be monitored on a daily basis. In the early 1990s sea level measurements from satellite altimeters became routine, although not all operational ocean data assimilation systems ingest altimeter data. During the 2000s Argo floats were introduced, and this was perhaps the biggest revolution in ocean observations for climate. Large areas of the ocean that were previously unobserved were now covered with autonomous Argo floats. Figure 20.9a shows the temperature observation density pre Argo in the Indian Ocean. Observations were mainly taken along the main shipping lanes as part of the Ship of Opportunity Program (SOOP). Large gaps remained throughout the Indian Ocean. During the Argo period (Fig. 20.9c) the temperature distribution changed dramatically, with almost every grid square experiencing at least one observation.

Perhaps the biggest impact of Argo is that it also measures salinity. For the first time there were enough salinity profiles to perform assimilation of salinity data.

Fig. 20.8 Spread of the ensemble (before assimilation) over the re-analysis period showing fields of a SST (C). b Temperature section along the equator C'C). c Sea surface salinity (psu) and d Salinity section along the equator (psu). (From Yin et al. 2011). Ensemble spread is calculated relative to a central analysis. (See Yin et al. 2011, for full details)

Fig. 20.8 Spread of the ensemble (before assimilation) over the re-analysis period showing fields of a SST (C). b Temperature section along the equator C'C). c Sea surface salinity (psu) and d Salinity section along the equator (psu). (From Yin et al. 2011). Ensemble spread is calculated relative to a central analysis. (See Yin et al. 2011, for full details)

C 30°E SO"E 70"E 90°E 1 10SE 1 JO*E 15Q"E
Fig. 20.9 The density of ocean sub-surface observations per lxl degree square per year, a and c are Temperature and b and d are Salinity, a and b are pre Argo and c and d are during the Argo period

Fig. 20.10 The impact of the TAO/TRITON and Argo data on seasonal forecast skill. Bars show the relative increase in root mean square errors of the 1-7 month forecasts of monthly SST anomalies resulting from withholding TAO/TRITON and Argo data in the initialization of JMA seasonal forecasts for different ocean areas. (From Fuji et al. 2008, where areas are defined)

Figure 20.9b shows the salinity observation density before Argo and Fig. 20.9d during Argo. The change is dramatic. Before Argo most of the Indian Ocean was unobserved. During Argo the salinity observation density is similar to that for temperature. The importance of salinity observations is discussed in Fujii et al. (2011). The results of Usui et al. (2006) indicate that only when salinity observations are assimilated is it possible to represent the strong meridional salinity gradient in the Western Equatorial Pacific, with low salinity waters north of the equator. Results also show that without the balance relationship between temperature and salinity it is not possible to represent the high salinity of the South Pacific Tropical Water, leading to the erosion of the vertical stratification and eventual degradation of the barrier layer.

The seasonal forecast skill can also be used to evaluate the ocean observing system. Fujii et al. (2011), evaluate the impact of the TAO/TRITON array and Argo float data on the JMA seasonal forecasting system by conducting data retention experiments. Their results (Fig. 20.10) show that TAO/TRITON data improves the forecast of SST in the eastern equatorial Pacific (NINO3, NINO4), and that Argo floats are essential observations for the prediction of the SST in tropical Pacific and Indian Oceans. Similar results have been obtained with the European Centre for Medium-range Weather Forecasts (ECMWF) seasonal forecasting system (Bal-maseda et al. 2007, 2009).

20.7 Seasonal Prediction in Australia

The Australian Bureau of Meteorology has produced seasonal outlooks since the late 1980s. Currently a seasonal rainfall and temperature outlook for Australia is produced operationally based on statistical links between tropical SSTs and local climate (Chambers and Drosdowsky 1999). However, it is felt that statistical ap-

TAO/TRITON ■ ARGO

NINO3 NIN

NINO12

JO34 N

NO4 NINOW WTIO

Fig. 20.10 The impact of the TAO/TRITON and Argo data on seasonal forecast skill. Bars show the relative increase in root mean square errors of the 1-7 month forecasts of monthly SST anomalies resulting from withholding TAO/TRITON and Argo data in the initialization of JMA seasonal forecasts for different ocean areas. (From Fuji et al. 2008, where areas are defined)

Fig. 20.11 POAMA monthly GBR Index (area average SST anomalies for the red box shown in the map insert) for December 2009 to May 2010 in the official outlook issued on 1 December 2009, with the distribution by quartiles of the ensemble composed of the last 30 daily forecasts. Overlaid is the ensemble mean (black). The shading indicates upper and lower climatological terciles from the POAMA v1.5 hindcasts. (http://www.bom.gov.au/oceanography/oceantemp/GBR_SST.shtml)

Fig. 20.11 POAMA monthly GBR Index (area average SST anomalies for the red box shown in the map insert) for December 2009 to May 2010 in the official outlook issued on 1 December 2009, with the distribution by quartiles of the ensemble composed of the last 30 daily forecasts. Overlaid is the ensemble mean (black). The shading indicates upper and lower climatological terciles from the POAMA v1.5 hindcasts. (http://www.bom.gov.au/oceanography/oceantemp/GBR_SST.shtml)

proaches have essentially reached the limits of their predictive ability, particularly as climate change is invalidating the assumptions of stationary that is fundamental to statistical approaches. The Bureau, in collaboration with CSIRO, has been developing successive versions of a dynamical coupled modelling system called POAMA (Predictive Ocean Atmosphere Model for Australia; http://poama.bom. gov.au). The first version was implemented in Bureau operations in 2002 and generated forecasts of ENSO-related SST indices. The POAMA system was upgraded in 2007 with version 1.5 and the operational products were extended to include forecasts of SST in the equatorial Indian Ocean (Zhao and Hendon 2009). More recently the products have been extended to give warnings of potential bleaching of coral in the Great Barrier Reef in the season ahead (e.g., Fig. 20.11; Spillman and Alves 2009). POAMA-1.5 has been shown to have high skill in prediction not only of ENSO and the IOD, but also the "flavour of ENSO", i.e. classical versus Modoki modes (Hendon et al. 2009; Lim et al. 2009). POAMA can skilfully predict tropical SST anomalies associated with ENSO two to three seasons in advance (Wang et al. 2008b) and can depict the teleconnection to Australian rainfall (Lim et al. 2009). POAMA can predict the peak phase of the occurrence of the IOD in austral spring (SON) with about four months lead time (Zhao and Hendon 2009).

The most skilful season for POAMA in predicting rainfall over Australia is during spring (SON), when the relationship between ENSO and Australian rainfall is strong. Fig. 20.12 shows that the skill (proportion correct) of predicting above median rainfall is high over south-eastern Australia and better than climatology over most of

Fig. 20.12 Proportion of ensemble members correctly predicting above median rainfall with (a) POAMA at LTO (b) POAMA at LT3 and (c) with the current operational statistical model (NCC model). The contour interval is 10%. The proportion correct greater than 60% is shaded. (From Lim et al. 2009)

the country at lead time 0 (LT0, i.e. forecasts initialised at the start of September and verified in SON, over the period 1980-2006) (Lim et al. 2009). This region of skill is where the teleconnection between rainfall and tropical SST is strong (Lim et al. 2009). However, operational regional rainfall and temperature forecasts at the Bureau are still based on the statistical system rather than POAMA at this point in time. Experimental rainfall products, such as probabilities of above median rainfall, from POAMA have been shown to be more skilful than those based on the statistical system based on skill measures such as the ROC score or hit rates (e.g., Fig. 20.12), but the forecast reliability is low, i.e. the forecasts are too emphatic (over-confident) often showing probabilities in excess of 90%. Work is in progress to address this reliability issue so that POAMA rainfall can form the basis for the Bureau's seasonal climate outlooks, including a pragmatic statistical correction and recalibration in the short term and investigating methods to increase ensemble spread in the long term.

A new version, POAMA-2 has been developed with improved physics and a new ocean data assimilation system, the POAMA Ensemble Ocean Data Assimilation System (PEODAS), mentioned in Sect. 20.5. A comprehensive set of hindcasts are currently being generated and the system is due to be implemented operationally towards the end of 2010. Preliminary results show a significant increase in SST skill in the Pacific Ocean in POAMA-2 compared to POAMA-1.5. Development of the POAMA-3 system is also underway, which includes a new coupled model based on the UKMO Unified Atmospheric model and the GFDL MOM4, to be run at a higher resolution than the current system. The ocean data assimilation system is also being extended to include the atmosphere and land surface, which will result in a multivariate ensemble coupled assimilation system.

The new ocean data assimilation system, PEODAS (Yin et al. 2011), is a major new development in POAMA. The system is based on multivariate ensemble optimum interpolation (Oke et al. 2005) where the background error covariance is calculated from an ensemble of ocean states. However, unlike Oke et al. (2005) which uses a static ensemble, PEODAS uses a time evolving ensemble to calculate a time dependent multivariate error covariance matrix. An ensemble is run in parallel to the main analyses by perturbing the ocean model forcing about the main analysis run, using a method developed by Alves and Robert (2005). An ocean reanalysis has been conducted from 1977 to 2007, assimilating temperature and salinity observations from the ENACT/ENSEMBLE project. During the assimilation, temperature and salinity were relaxed to monthly climatology through the water column with an e-folding time scale of 2 years. The model SST was strongly nudged to the SST product from the NCEP reanalysis with a 1-day time scale.

In Sect. 20.5 it was shown that the PEODAS ocean reanalysis is an improvement with respect to the previous POAMA version. Preliminary results also suggest that these improvements lead to better forecast skill of SST at seasonal time scales. For each reanalysis a set of hindcasts starting each month from 1980 to 2001 were produced. For the PEODAS reanalysis a 10-member ensemble was generated using the main PEODAS reanalysis. For the old POAMA reanalysis a 10-member ensemble was also generated, however this time by using the same ocean initial conditions

Fig. 20.13 NINO3.4 SST Nino3.4 Corr ALL 82-06

Fig. 20.13 NINO3.4 SST Nino3.4 Corr ALL 82-06

(since perturbed states were not available) and taking atmospheric initial conditions six hours apart.

Figure 20.13 shows the NINO3.4 forecast skill with lead time for forecasts from each set of reanalysis and based on the 10-member ensemble means. Forecasts using PEODAS initial conditions show significantly more skill than those using the old POAMA assimilation initial conditions. While the old reanalysis had a similar fit to observed temperature as the new reanalysis, the old reanalysis showed a considerably worse fit for salinity and zonal current. This result can be taken as an indication that, for the assimilation to improve forecast skill, it is important to keep the dynamical and physical balance among variables, and therefore all variables, not just those directly constrained by observations, should show consistent improvement.

20.8 Decadal Prediction

Decadal climate prediction is very much in its infancy, but has the potential to provide information enabling better adaptation to climate change. Anthropogenic climate change signals are strongly modulated by natural climate variability, particu larly variability driven by slow processes in the ocean on decadal time-scales (Hur-rell et al. 2010). There is growing evidence that, like seasonal prediction, decadal prediction is an initial-value problem, with recent results from the ENSEMBLES project (Smith et al. 2007; van der Linden and Mitchell 2009) showing that initialised decadal forecasts have the potential to provide improved information compared with traditional climate change projections. Decadal predictability originates primarily from changes in radiative forcing, including anthropogenic greenhouse gases and aerosols, and long-lived variations in the ocean. Examples of the latter include variations associated with the Pacific Decadal Oscillation (PDO; e.g., Mantua et al. 1997), the Inter-decadal Pacific Oscillation (IPO; e.g., Power et al. 1999) and the Atlantic Multidecadal Oscillation (AMO; e.g., Knight et al. 2005). The ability to predict these long term climate variations depends therefore, in part, on accurate ocean initial conditions. However, compared to seasonal prediction, decadal prediction relies on the less well observed deeper ocean. Recent improvements in the ocean observing system, in particular the advent of Argo data, offers potential for increased skill of decadal forecasts (Balmaseda et al. 2010a). The Argo data (available since 2003) are likely to be critical, for example, for making skilful predictions of the Atlantic Meridional Overturning Circulation (MOC) (Balmaseda et al. 2010a). But, a major challenge for decadal prediction is how to evaluate the hindcasts and forecasts, particularly in view of sparse historical ocean observations (Balmaseda et al. 2010a; Hurrell et al. 2010). In addition, as a result of our short observational record, the mechanisms of decadal variations are not well understood and the representation of this variability differs considerably among models (Hurrell et al. 2010). This means that the theoretical upper limit of our prediction skill on the decadal time scale is also not well established (Hurrell et al. 2010). Another challenge facing decadal prediction is how to initialise the forecasts. Current systems (Smith et al. 2007; Keenlyside et al. 2008; Pohlmann et al. 2009) use anomaly initialisation, rather than full initialisation, such that models are initialised with observed anomalies added to the model climate. This method is a way of dealing with model bias and reducing initialisation shock. However, the best approach for initialising decadal forecasts remains unclear (Hurrell et al. 2010).

20.9 Summary

Today's sophisticated operational seasonal forecast systems rely on a number of interrelated components: data assimilation and initialisation, a coupled ocean-atmosphere general circulation model, ensemble generation and forecast calibration. The ocean plays a key role in each component. Predictive skill in seasonal forecasting comes from the initial state of the coupled system, particularly the upper ocean. Correctly initialising the important modes of seasonal and interannual variability, such as ENSO and the IOD, is vital. Real-time estimates of the ocean initial state have improved dramatically over the last two decades with improvements to the ocean observing network, especially from the TAO/TRITON array and Argo floats.

However, seasonal forecasting requires an ocean reanalysis going back in time in order to initialise the retrospective forecasts required for skill assessment of the forecast system and calibration of the forecasts. The non-stationarity of the ocean observing system poses huge challenges for the initialisation and verification of seasonal, as well as decadal, hindcasts and forecasts. Results have shown that the method of ocean initialisation has a significant impact on the mean state, variability and skill of the forecasts (Balmaseda and Anderson 2009). Because of deficiencies in the coupled model, the aim of producing the best initial state, closest to observed, may not produce the best forecasts. There may be long-term effects of model spin-up or initialisation shock when using observed initial conditions. Recent research suggests that the initialisation scheme that makes the most use of the observed data will produce the most skilful forecasts, even though initial imbalances in the coupled state are generated (Balmaseda and Anderson 2009). Clearly, however, the impact of the initialisation scheme is very dependent on the quality of the coupled model. Current research is addressing the prospect of "coupled assimilation", where data assimilation for the atmosphere and ocean are done by the coupled model, leading to a well-balanced initial state.

Seasonal prediction is a complex and challenging field of research and application. This paper addresses dynamical seasonal prediction using coupled ocean-atmosphere models, with particular focus on data assimilation and initialisation. The delivery, value and use of seasonal forecasts have not been discussed. It is the latter that will continue to drive future advances in coupled models, data assimilation, ensemble techniques and the ocean observing system.

Acknowlegements The authors would like to acknowledge Eun-Pa Lim, Claire Spillman, Guomin Wang and Yonghong Yin for providing some of the figures used in this paper.

References

Alves O, Robert C (2005) Tropical Pacific Ocean model error covariances from Monte Carlo simulations. Quart J Roy Meteor Soc 131:3643-3658 Alves O, Wang O, Zhong A, Smith N, Tseitkin F, Warren G, Schiller A, Godfrey JS, Meyers G (2003) POAMA: bureau of meteorology operational coupled model forecast system. National Drought Forum, Brisbane (15-16 April) Alves O, Balmaseda M, Anderson D, Stockdale T (2004) Sensitivity of dynamical seasonal forecasts to ocean initial conditions. Quart J Roy Meteor Soc 130:647-668 Baldwin MP, Dunkerton TJ (2001) Stratospheric harbingers of anomalous weather regimes. Science 294:581. doi:10.1126/science.1063315 Balmaseda MA, Weaver A (2006) Temperature, salinity, and sea-level changes: climate variability from ocean reanalyses. Paper presented at the CLIVAR/GODAE meeting on ocean synthesis evaluation, 31 August-1 September 2006, ECMWF, Reading, UK. http://www.clivar.org/orga-nization/gsop/synthesis/groups/Items3_4.ppt. Accessed 26 May 2009 Balmaseda MA, Anderson D (2009) Impact of initialisation strategies and observations on seasonal forecast skill. Geophys Res Lett 36:L01701. doi:10.1029/2008GL035561 Balmaseda MA, Anderson DLT, Vidard A (2007) Impact of Argo on analyses of the global ocean. Geophys Res Lett 34:L16605. doi:10.1029/2007GL030452

Balmaseda MA, Vidard A, Anderson D (2008) The ECMWF ORA-S3 ocean analysis system. Mon

Wea Rev 136:3018-3034 Balmaseda MA, Alves O, Arribas A, Awaji T, Behringer D, Ferry N, Fujii Y, Lee T, Rienecker M, Rosati T, Stammer D (2009) Ocean initialisation for seasonal forecasts. Oceanography 22:154-159

Balmaseda MA, Fujii Y, Alves O et al (2010a) Initialisation for Seasonal and Decadal Forecasts. In: Hall J, Harrison DE,Stammer D (eds) Proceedings of OceanObs'09: sustained ocean observations and information for society, vol 2, ESA Publication WPP-306, Venice, 21-25 September 2009

Balmaseda MA, Fujii Y, Alves O et al (2010b) Role of the ocean observing system in an end-to-end seasonal forecasting system. Plenary paper, OceanObs'09, Venice, 21-25 September 2009. http://www.oceanobs09.net/plenary/index.php Behringer DW (2007) The Global Ocean Data Assimilation System at NCEP. 11th symposium on integrated observing and assimilation systems for atmosphere, oceans, and land surface, AMS 87th Annual Meeting, San Antonio, pp 12 Bell CJ, Gray LJ, Charlton-Perez AJ, Scaife AA (2009) Stratospheric communication of ENSO

teleconnections to european winter. J Clim 22:4083-4096 Berner J, Doblas-Reyes FJ, Palmer TN, Shutts G, Weisheimer A (2008) Impact of a quasi-stochastic cellular automaton backscatter scheme on the systematic error and seasonal prediction skill of a global climate model. Philos Trans R Soc A 366:2561-2579 Bjerknes J (1969) Atmospheric teleconnections from the equatorial Pacific. Mon Wea Rev 97:163-172

Bonjean F, Lagerloef GSE (2002) Diagnostic model and analysis of surface currents in the tropical

Pacific Ocean. J Phys Oceanogr 32:2938-2954 Burgers G, Balmaseda MA, Vossepoel FC, van Oldenborgh GJ, van Leeuwen PJ (2002) Balanced

Ocean-Data Assimilation near the Equator. J Phys Oceanogr 32:2509-2519 Cagnazzo C, Manzini E (2009) Impact of the stratosphere on the winter tropospheric teleconnec-

tions between ENSO and the North Atlantic and European region. J Clim 22:1223-1238 Cazes-Boezio G, Menemenlis D, Mechoso CR (2008) Impact of ECCO ocean-state estimates on the initialisation of seasonal climate forecasts. J Clim 21:1929-1947 Chambers LE, Drosdowsky W (1999) Australian seasonal rainfall prediction using near global sea surface temperatures. AMOS Bull 12(3):51-55 Chang P, Yamagata T, Schopf P, Behera SK, Carton J, Kessler WS, Meyers G, Qu T, Schott F, Shetye S, Xie S-P (2006) Climate fluctuations of tropical coupled systems—the role of ocean dynamics. J Clim 19:5122-5174 Collins M, Booth BBB, Harris GR, Murphy JM, Sexton DMH, Webb MJ (2006) Towards quantifying uncertainty in transient climate change. Clim Dyn 27:127-147 Doblas-Reyes FJ, Weisheimer A, Deque M et al (2009) Addressing model uncertainty in seasonal and annual dynamical ensemble forecasts. Quart J R Meteor Soc 135:1538-1559 Dommenget D, Stammer D (2004) Assessing ENSO simulations and predictions using adjoint ocean state estimation. J Clim 17:4301-4315 Fennessy MJ, Shukla J (1999) Impact of initial soil wetness on seasonal atmospheric prediction.

J Clim 12(11):3167-3180 Fletcher CG, Hardiman SC, Kushner PJ, Cohen J (2009) The dynamical response to snow cover perturbations in a large ensemble of atmospheric GCM integrations. J Clim 22:1208-1222 Folland CK, Colman AW, Rowell DP, Davey MK (2001) Predictability of Northeast Brazil rainfall and real-time forecast skill. J Clim 14:1937-1958 (1987-1998) Fujii Y, Matsumoto S, Kamachi M, Ishizaki S (2011) Estimation of the equatorial Pacific salinity field using ocean data assimilation system. Adv Geosci (In Press) Goddard L, Graham NE (1999) The importance of the Indian Ocean for simulating rainfall anomalies over eastern and southern Africa. J Geophys Res 104:19099-19116 Hendon HH, Lim E, Wang G, Alves O, Hudson D (2009) Prospects for predicting two flavors of El Niño. Geophys Res Lett. doi:10.1029/2009GL040100

Hudson D, Alves O, Hendon HH, Wang G (2010) The impact of atmospheric initialisation on seasonal prediction of tropical Pacific SST. Clim Dyn. doi:10.1007/s00382-010-0763-9 Hurrell J, Delworth TL, Danabasoglu G et al (2010) Decadal climate prediction: opportunities and challenges. In: Hall J, Harrison DE, Stammer D (eds) Proceedings of 0cean0bs'09: sustained ocean observations and information for society, vol 2. ESA Publication WPP-306, Venice, 21-25 September 2009

Ineson S, Scaife AA (2008) The role of the stratosphere in the European climate response to El

Nino. Nat Geosci 2:32-36 Jin F-F, Lin L, Timmermann A, Zhao J (2007) Ensemblemean dynamics of the ENSO recharge oscillator under statedependent stochastic forcing. Geophys Res Lett 34:L03807. doi:10.1029/2006GL027372 Jin EK, Kinter JL III, Wang B et al (2008) Current status of ENSO prediction skill in coupled ocean-atmosphere models. Clim Dyn 31:647-664. doi:10.1007/s00382-008-0397-3 Keenlyside N, Latif M, Jungclaus J, Kornblueh L, Roeckner E (2008) Advancing decadal-scale climate prediction in the North Atlantic Sector. Nature 453:84-88 Keppenne CL, Rienecker MM, Jacob JP, Kovach R (2008) Error covariance modeling in the GMAO ocean ensemble kalman filter. Mon Wea Rev 136:2964-2982. doi:10.1175/2007MWR2243.1 Kirtman BP, Pirani A (2009) The state of the art of seasonal prediction: outcomes and recommendations from the first world climate research program workshop on seasonal prediction. Bull Am Meteor Soc 90:455-458 Knight JR, Allan RJ, Folland CK et al (2005) A signature of persistent natural thermohaline circulation cycles in observed climate. Geophys Res Lett 32:L20708. doi:1029/2005GL024233 Koster RD, Suarez MJ (2003) Impact of land surface initialisation on seasonal precipitation and temperature prediction. J Hydrometeor 4:408-423 Koster RD, Suarez MJ, Liu P et al (2004) Realistic initialisation of land surface states: impacts on subseasonal forecast skill. J Hydrometeor 5:1049-1063 Koster RD,Guo Z, Dirmeyer PA et al (2006) GLACE: The global land-atmosphere coupling experiment. Part I: overview. J Hydrometeor 7:590-610 Koster RD, Mahanama SPP, Yamada TJ et al (2010) Contribution of land surface initialisation to subseasonal forecast skill: first results from a multi-model experiment. Geophys Res Lett 37:L02402. doi:10.1029/2009GL041677 Kushnir Y, Robinson WA, Chang P, Robertson AW (2006) The physical basis for predicting Atlantic sector seasonal-to-interannual climate variability. J Clim 19:5949-5970 Lim E-P, Hendon HH, Hudson H, Wang G, Alves O (2009) Dynamical forecasts of inter-El Niño variations of tropical SST and Australian spring rainfall. Mon Wea Rev 137:3796-3810 Luo JJ, Masson S, Behera S, Yamagata T (2007) Experimental forecasts of the Indian ocean dipole using a coupled OAGCM. J Clim 20:2178-2190 Mantua NM, Hare SR, Zhang Y, Wallace JM, Francis RC (1997) A Pacific interdecadal climate oscillation with impacts on salmon production. Bull Am Meteor Soc 78:1069-1079 Marshall AG, Scaife AA (2009) Impact of the QBO on surface winter climate. J Geophys Res

114:D18110. doi:10.1029/2009JD011737 Mason SJ, Stephenson D (2008) How do we know whether seasonal climate forecasts are any good? In: Troccoli A, Harrison M, Anderson DLT, Mason SJ (eds) Seasonal climate: forecasting and managing risk. NATO Science Series. Springer, Dordrecht, pp 467 Martin MJ, Hines A, Bell MJ (2007) Data assimilation in the FOAM operational short-range ocean forecasting system: a description of the scheme and its impact. Quart J R Meteor Soc 133:981-995

Maycock AC, Keeley SPE, Charlton-Perez AJ, Doblas-Reyes FJ (2009) Stratospheric circulation in seasonal forecasting models: implications for seasonal prediction. Clim Dyn. doi:10.1007/ s00382-009-0665-x

Murphy JM, Sexton DMH, Barnett DN, Jones GS, Webb MJ,Collins M, Stainforth DA (2004) Quantification of modelling uncertainties in a large ensemble of climate change simulations. Nature 430:768-772

Neelin D, Battisti DS, Hirst AC, Jin F-F, Wakata Y, Yamagata T, Zebiak S (1998) ENSO theory. J

Geophys Res 103:14261-14290 Oke PR, Schiller A, Griffin DA, Brassington GB (2005) Ensemble data assimilation for an eddy-

resolving ocean model of the Australian region. Quart J R Meteor Soc 131:3301-3311 Oldenborgh GJ van, Balmaseda MA, Ferranti L, Stockdale TN, Anderson DLT (2005) Did the ECMWF seasonal forecast model outperform a statistical model over the last 15 years? J Clim 18:2960-2969

Palmer TN, Alessandri A, Andersen U et al (2004) Development of a European multimodel ensemble system for seasonal-to-interannual prediction (DEMETER). Bull Am Meteor Soc 85:853-872

Pham DT, Verron J, Roubaud MC (1998) A singular evolutive extended Kalman filter for data assimilation in oceanography. J Mar Syst 16:323-340 Philander SG (2004) Our affair with El nino. Princeton University Press, Princeton, pp 275 Pohlmann H, Jungclaus J, Marotzke J, Köhl A, Stammer D (2009) Improving predictability through the initialization of a coupled climate model with global oceanic reanalysis. J Clim 22:3926-3938

Power S, Casey T, Folland C, Colman A, Mehta V (1999) Inter-decadal modulation of the impact of ENSO on Australia. Clim Dyn 15:319-324 Rasmusson EM, Carpenter TH (1983) The relationship between eastern equatorial Pacific SSTs and rainfall over India and Sri Lanka. Mon Wea Rev 111:517-528 Rodwell MJ, Folland CK (2002) Atlantic air-sea interaction and seasonal predictability. Quart J R

Meteor Soc 128:1413-1443 Ropelewski CF, Halpert MS (1987) Global and Regional Scale Precipitation Patterns Associated with the El Niño/Southern Oscillation. Mon Wea Rev 115:1606-1626 Saji NH, Yamagata T (2003) Possible impacts of Indian Ocean Dipole mode events on global climate. Clim Res 25:151-169 Saji, NH, Goswami BN, Vinayachandran PN, Yamagata T (1999) A dipole mode in the tropical

Indian Ocean. Nature 401:360-363 Seneviratne SI, Koster RD, Guo Z et al (2006) Soil moisture memory in agcm simulations: analysis of global land-atmosphere coupling experiment (GLACE) data. J Hydrometeor 7:1090-1112 Shi L, Alves O, Hendon HH, Wang G, Anderson D (2009) The role of stochastic forcing in ensemble forecasts of the 1997/98 El Niño. J Clim 22:2526-2540 Smith D, Cusack S, Colman A, Folland C, Harris G, Murphy J (2007) Improved surface temperature prediction for the coming decade from a global circulation model. Science 317:796-799 Spillman CM, Alves O (2009) Dynamical seasonal prediction of summer sea surface temperatures in the Great Barrier Reef. Coral Reefs. doi:10.1007/s00338-008-0438-8 Stainforth DA, Aina T, Christensen C, Collins M, Faull N, Frame DJ, Kettleborough JA, Knight S, Martin A, Murphy JM, Piani C, Sexton D, Smith LA, Spicer RA, Thorpe AJ, Allen MR (2005) Uncertainty in predictions of the climate response to rising levels of greenhouse gases. Nature 433:403-406

Stephenson D (2008) An Introduction to Probability Forecasting. In: Troccoli A, Harrison M, Anderson DLT and Mason SJ (eds) Seasonal climate: forecasting and managing risk. NATO Science Series. Springer, Dordrecht, pp 467 Stockdale TN (1997) Coupled ocean-atmosphere forecasts in the presence of climate drift. Mon Wea Rev 125:809-818

Stockdale TN, Balmaseda MA, Vidard A (2006) Tropical Atlantic SST prediction with coupled ocean-atmosphere GCMS. J Clim 19:6047-6061 Stockdale TN, Alves O, Boer G et al (2010) Understanding and predicting seasonal to interan-nual climate variability—the producer perspective. White Paper for WCC3. Draft. http://www. wcc3.org/sessions.php?session_list=WS-3 Stockdale TN, Anderson DLT, Balmaseda MA, Doblas-Reyes F, Ferranti L, Mogensen K, Palmer TN, Molteni F, Vitart F (2011). ECMWF Seasonal forecast system 3 and its prediction of sea surface temperature. Clim Dyn (In Press)

Ummenhofer CC, England MH, McIntosh PC, Meyers GA, Pook MJ, Risbey JS, Gupta AS, Taschetto AS (2009) What causes southeast Australia's worst droughts? Geophys Res Lett. doi:10.1029/2008GL036801 Usui N, Ishizaki S, Fujii Y, Tsujino H, Yasuda T, Kamachi M (2006) Meteorological research institute multivariate ocean variational estimation (MOVE) system: some early results. Adv Space Res 37:806-822

van der Linden P, Mitchell JFB (eds) (2009) ENSEMBLES: Climate change and its impacts: summary of research and results from the ENSEMBLES project. Met Office Hadley Centre, Exeter, pp 160

Vialard J, Vitart F, Balmaseda M, Stockdale T, Anderson D (2005) An ensemble generation method for seasonal forecasting with an ocean-atmosphere coupled model. Mon Wea Rev 133:441-453 Wajsowicz RC (2007) Seasonal-to-interannual forecasting of tropical Indian Ocean sea surface temperature anomalies: potential predictability and barriers. J Clim 20:3320-3343 Walker G (1923) Correlation in seasonal variations of weather VIII. A preliminary study of world weather. Mem Indian Meteorol Dept 24(4):75-131 Walker GT (1924) Correlation in seasonal variations of weather IX. Mem Indian Meteorol Dept 24(9):275-332

Wang B, Lee J-Y, Kang I-S, et al (2008a) Advance and prospectus of seasonal prediction: assessment of the APCC/CliPAS 14-model ensemble retrospective seasonal prediction (1980-2004). Clim Dyn. doi:10.1007/s00382-008-0460-0 Wang G, Alves O, Hudson D, Hendon H, Liu G, Tseitkin F (2008b) SST skill assessment from the new POAMA-1.5 System. BMRC Res Lett 8:2-6 (Bureau of Meteorology, Australia) Webster PJ, Moore AM, Loschnigg JP, Leben RR (1999) Coupled ocean-atmosphere dynamics in the Indian Ocean during 1997-1998. Nature 401:356-360 Weisheimer A, Doblas-Reyes FJ, Palmer TN et al (2009) ENSEMBLES: a new multi-model ensemble for seasonal-to-annual predictions—Skill and progress beyond DEMETER in forecasting tropical Pacific SSTs. Geophys Res Lett 36(21):L21711 Yin Y, Alves O, Oke PR (2011) An ensemble ocean data assimilation system for seasonal prediction. Mon Wea Rev. doi:10.1175/2010MWR3419.1 Zebiak SE, Cane MA (1987) A model El nino-southern oscillation. Mon Wea Rev 115:2262-2278 Zhao M, Hendon HH (2009) Representation and prediction of the Indian Ocean dipole in the POAMA seasonal forecast model. Quart J R Meteor Soc 135(639):337-352

Part VII Evaluation

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