Project topics and materials
Get Complete Project Material File(s) Now! »
Overview of the SAWS climate prediction system
Locally, seasonal forecast practice started in the early 1990s (e.g., Landman and Mason, 1999; Mason, 1998). The monthly to seasonal predictability of extratropical atmosphere is relatively lower than the tropics owing to strong hydrodynamical instabilities associated with baroclinicity that exits in the middle latitudes (Holton, 2004). In spite of the southern Africa subcontinent being strongly affected by extratropical instabilities, observational and modelling studies succeeded to show the existence of potential predictability (e.g. Klopper et al., 1997; Landman and Goddard, 2002; Reason and Rouault, 2005, Tennant and Hewitson, 2002). All these modelling studies are mainly founded on the knowledge of the tropical heat modulation on the mid-latitude circulation particularly during the austral mid-summer when its signature becomes noticeable (Shukla 1981; Mason et al., 1996).
In South Africa, the use of AGCMs in operational seasonal forecasts became more visible when South African based institutions such as the University of Cape Town, University of Pretoria, the Council for Scientific and Industrial Research (CSIR) and the South African Weather Service (SAWS) began to display their forecasts on the website of the Global Forecasting Centre for Southern Africa (www.GFCSA.net) in 2003. See Landman (2014) for a description of some of the seasonal forecasting efforts in South Africa.
Nonetheless, the SAWS experience with regard to the use of AGCMs date back to 1995. In this historical context, the Center for Ocean–Land–Atmosphere Studies global spectral model (COLA; Kirtman et al., 1997) was the first model to be operationally introduced on the SAWS first Super Computer (SV1 Cray; Tennant, 2003). After more than a decade of service, the COLA AGCM was concurrently replaced by the ECHAM4.5 AGCM (Roeckner et al., 1996) in 2007 when the Cray Super computer was substituted by the NEC SX8 Super Computer. Since then one of the major advancements in the area of seasonal forecast development in the region includes the SAWS’s acquisition of Global Producing Centre for Long-Range Forecasts status from the World Meteorological Organisation (WMO).
With the inception of the Centre for High Performance for Computing (CHPC) in South Africa, the computational infrastructure in this country has grown exponentially. This drastic computational development also stimulated the emergence of ocean-atmosphere global circulation models (OAGCMs) locally (Beraki et al., 2014). In fact, the expansion and optimization of the SAWS global climate prediction system would have been impossible without this computational support. The CHPC was initiated by the Department of Science and Technology and its aim is to provide High Performance Computing (HPC) facilities and expertise for research in South Africa.
Preparation of the boundary conditions
Preparation of the boundary conditions
Multi-model ensemble mean SST anomalies are used as input to derive the lower boundary forcing for the AGCM. Hindcasts of SST anomalies obtained from two CGCMs are objectively combined (equal weights applied to ensemble averaged hindcasts) to produce a multi-model ensemble (MME) of global SST anomalies. The CGCMs are the SAWS coupled model (SCM; Beraki et al., 2014) and NCEP CFS v2 (Climate Forecasting System Version 2; Saha et al., 2014). The advantage of using MME mainly arises from the fact that CGCMs differ in their performance under different conditions and are presumably nearly always better than any of the individual ensembles (Doblas-Reyes et al., 2000; Krishnamurti et al., 2000). In fact, the most striking benefit obtained from MME is the skill-filtering property in regions or seasons when the performance of the individual models varies widely (Graham et al., 2000). Coupled model SST forecasts are used since it has been shown that using such forecasts in order to force the ECHAM4.5 AGCM is an improvement over using statistical SST forecasts to force the model (Landman et al., 2014). Figure 1 shows that forecast quality of ENSO is enhanced as a result of model combination, particularly at the extended lead-times over several months. The latter supports the notion of using MME SST forcing in the AGCM’s configuration.
Forcing the AGCM with all ensemble realizations of the participating models and multi-model ensemble mean is computationally inhibiting. First of all, each boundary condition forcing causes a double fold increase in the AGCM integrations by the size of atmospheric ICs as it is also important to account for uncertainties arising from the atmospheric states. Second, it is hardly possible to maintain consistency between a comprehensive set of hindcasts and real-time model forecasts because the number of participating models in the MMS may change as models are either added to or withdrawn from the system. Another drawback with such an initial and boundary value configuration is that deficiencies in the true representation of the uncertainty amplitude can result because these models sometimes tend to reveal a great deal of resemblance among themselves while at other times they are vastly differ from one another. Hence the prescription of the SST scenarios in a manner that optimizes the representation of the uncertainty envelope as well as taking into account computational limitations is vital. Besides, this strategy minimizes the risk of redoing costly hindcast runs because the reconstruction of the retroactive forecasts of the AGCM is only needed when there is a significant shift in the uncertainty envelope of the SSTs themselves rather than changes in the participating models per se. To achieve our goal, the AGCM (ScA and ScB; Table 1) is forced with only three SST scenarios which comprise the multi-model ensemble-mean and its amplitude band estimated from the uncertainty term (S. Mason 2010, personal communication).
The uncertainty term was identified independently for all lead-months (lead-0 to lead-8) from the historical standard error. In this context, a slight perturbation was applied on the SST retroactive forecasts using empirical orthogonal function analysis (EOF; Lorenz, 1956; North, 1984). The EOF analysis searches optimum directions that explain the maximum variation that deviates from the reference. In this case, the first normalized EOF mode is retained to describe the uncertainty term assuming that the variance is best explained by the most dominant EOF mode. A similar variant of SST prescription is also used to derive one of the operational AGCMs at the International Research Institute for climate and society (IRI; T. Barnston 2007, personal communication).
1. Introduction
-Research problem
-Objectives and aims
-Thesis outline
2. Dynamical seasonal climate prediction using an ocean-atmosphere coupled climate model: description and evaluation
-Preface
-Dynamical seasonal climate prediction using an ocean-atmosphere coupled climate model developed in partnership between South Africa and the IRI
-Synopsis
3. The Impact of Optimization on the Predictive Skill of a two-tiered Seasonal Forecasting System
-Preface
-Global Dynamical Forecasting System Conditioned to Robust Initial and Boundary Forcings: Seasonal Context
-Synopsis
4. On the comparison between seasonal predictive skill of global circulation models: coupled versus uncoupled
-Preface
-On the comparison between seasonal predictive skill of global circulation models: coupled versus uncoupled
-Synopsis
5. Summary and conclusions
References
