The predictive skill of seasonal forecast arises from the slowly evolving climate processes where the signature, that noticeably influence the mean state of weather conditions, mainly resides in the ocean. The interaction of the ocean and atmosphere is therefore the minimum level of complexity required for seasonal timescale. The practice of contemporary seasonal prediction is presumably achievable with the use of two distinct GCM (Global Climate Model) configurations commonly referred to as one- and two-tiered forecasting systems based on the manner in which the atmosphere and ocean exchange information. One-tiered forecasting systems (Coupled climate models) are placed at the highest hierarchy in the science of numerical modelling in terms of complexity. They are hypothesized to represent the state of the art of seasonal forecasting which inherently renders them to be convenient for seasonal climate prediction purposes. Notwithstanding, it may be important to appraise whether or not two-tier forecasting systems (uncoupled models) offer comparable levels of skill that are currently attainable by state-of-the-art coupled climate models under a constrained computational resources environment. Such a restrictive environment is commonly found in developing countries such as South Africa. With this in mind, the study attempts to test the notion under a perfect model framework where the atmospheric global climate model is forced with the best estimate of predicted sea-surface temperature (SST), while the two systems are kept similar in all other aspects. The framework eliminates differences between the two forecasting systems due to model biases and in fact enables the discrimination of the role of coupling on seasonal forecast skill.
Due to the enormous computational resources required to develop and run an operational forecast system based on coupled models, their engagement for real-time forecasts has been negligible in South Africa. However, motivated by the recent advances in computing infrastructures in South Africa due to the establishment and maintenance of the Centre for High Performance Computing (CHPC) as well as international collaboration, the study pioneered in Africa the emergence of the South African Weather Service Coupled Model (SCM) also referred to as the ECHAM4.5-MOM3-SA. The model couples the ECHAM4.5 atmospheric general circulation model (AGCM) and Modular Ocean Model version 3 (MOM3) using the multiple program multiple data (MPMD) coupler paradigm. The model employs an atmospheric initialization strategy that is different from other models that couple the same atmosphere and ocean models. The study reveals that the South African coupled model has skill levels for ENSO (El Niño Southern Oscillation) forecasts comparable with other coupled models currently administered by international centres. Furthermore the model is also found to be skilful in predicting upper air dynamics, surface air temperature, rainfall and equatorial Indian Ocean Dipole (IOD).
In the two-tiered experiment, the AGCM is constrained by the lower boundary conditions derived from predicted SST anomalies of two ocean-atmosphere coupled general circulation models (CGCMs) combined through equal weighting. In addition, the SST uncertainty amplitude (lower and upper bounds) defined from this combination is also considered as separate forcing fields. As with the CGCM, the AGCM is initialized with the realistic state of the atmosphere and soil moisture. Results from hindcasts show that this optimized forecasting system demonstrates large-scale consistent skill improvements for surface temperature and rainfall totals relative to forcing the AGCM with persisted SST anomalies and the AMIP-2 (Atmospheric Model Intercomparison Project) type simulations. Model evaluation further reveals that the AGCM is able to forecast anomalous upper air atmospheric dynamics (circulation) over the tropics up to several months ahead. In addition, the contribution of the predicted SST, which is based on a multi-model approach, is shown to be of significant importance for optimized AGCM results. However, the AGCM appears to be weakly sensitive to soil moisture initialization which may suggest an internal weakness of the model. The study has addressed some optimization issues for atmospheric models and proposed an optimal AGCM configuration that can serve as baseline against which more advanced models can be tested.
Finally, the comparative experiments reveal that the GCM configurations widely differ in their performances and the superiority of one model over the other is mostly dependent on the ability to a priori determine an optimal global SST field for forcing the AGCM. In fact, the AGCM offers comparable predictive capabilities with the CGCM when the CGCMs skilful predicted SST evolution can in turn be used to force the AGCM. This finding supports the notion that the further enhancement of seasonal forecasting practices favours the use and further improvement of CGCMs (should computational resources be permissible) since the potential for further improvement of AGCM-based forecasts heavily depends on the improvement of CGCMs.