Prediction And Predictability Of Temperature Variability In The Tropical Oceans

The tropical oceans are crucial to large scale ocean-atmosphere interactions, and have profound impact on global climate variability. El Ni?o-Southern Oscillation (ENSO) and Tropical Atlantic Variabiltiy (TAV) are the most notable phenomena among Tropical Pacific (TP) and Tropical Atlantic (TA) climate variations, respectively. Prediction of tropical sea surface temperatures (SSTs) is a key issue to seasonal-to-decadal climate prediction, and has important implication to global climate and marine ecosystem studies, as well as disaster relief and economic development of many countries.In this dissertation, an intermediate ocean-atmosphere coupled model is used for retrospective predictions of TP and TA SST variability. Factors that impact SST predictability are investigated. The coupled model consists of an atmospheric general circulation model (AGCM) (CCM3), and a Zebiak-Cane type of reduced gravity ocean model (RGO), and is equipped with a novel atmospheric noise filter in order to test impact of"weather noise"on SST prediction. Ensemble forecast experiments are initialized in four seasons, and carried out for 12 months for each ensemble member.In the first part of the dissertation, ENSO prediction is conducted with an improved initialization scheme and the atmospheric noise filter. The results show that these efforts lead to an improved ENSO forecast skill. Anomaly correlation of observed and predicted Nino3 index reaches 0.71 at 6-month lead time, and 0.43 at 12-month lead time. The value of of Nino3.4 index reaches 0.75 and 0.49, respectively. The overall forecast skill is comparable to some of the most advanced ENSO prediction models. The coupled model has higher predictability of warm and cold events than near-normal events. Spacial patterns and trends of the equatorial Pacific cold tongue variability can be well captured during prediction of El Ni?o and La Ni?a events.Impacts of initial condition, weather noise and model biases on ENSO prediction are investigated based on ensemble prediction experiments. Major findings can be summarized as follows:1) Initial conditions are of significant importance to ENSO SST predictions. The initial conditions used in this dissertation were produced using the coupled model with a strong SST restoring to the observations, therefore can effectively reduce"initial shock"caused by mismatch between observations and model physics. Meanwhile, the noise filter is used in the initialization process to suppress impact of weather noise, leading to reduced initial noise in forecast. The resultant initial conditions, achieving both compatibility with model and accuracy, are shown to have major effect on improvement of short term ENSO forecast skill up to 2 seasons.2) Weather noise plays a notable role in affecting ENSO prediction skill. With appropriate initial conditions, reducing weather noise can alleviate drop of forecast skill caused by the so called"spring predictability barrier"(SPB), and help improving forecast skill in 3-4 leading seasons. The noise filter mainly takes effect through boosting signal to noise ratio (SNR) of wind stress, and improving the model's response to Bjerknes feedbacks between wind stress, thermocline and SST.3) Atmospheric model biases have severe impact on simulation of thermocline depth, which in turn limits SST prediction skill at long lead times. Systematic model biases are a common problem in many coupled models, and need to be addressed before major improvement of model forecast skill at long lead time can be achieved. Attempts were made to correct wind stress bias in the atmospheric model using a model output statistics (MOS) technique. Primary results show improvement in short term forecast up to two seasons, and are promising in overcoming the spring predictability barrier.The second part of this dissertation investigates predictability of Tropical Atlantic SSTs and the role of weather noise in that region. With the utilization of the atmospheric noise filter, it is shown that filtering weather noise generally improves forecast skill. Improvement is particularly effective in the Southern Tropical Atlantic (STA), where previous models generally failed to show any useful prediction skill. Considerably high skill is achieved in forecasting STA SST 2-3 seasons in advance. This in turn leads to a useful forecast skill of spring rainfall over the Nordeste of Brazil. The improved skill is attributed to the reduction of weather noise in surface heat fluxes. Filtering the heat flux noise enhances feedbacks between surface heat fluxes and SST in the STA region, strengthening the SNR and resulting in an improved SST predictability. Ekman process in the upper ocean is found to be important as well in SST prediction in STA, while Bjerknes feedback mechanism is less significant in this region.Finally, potential of utilization of the noise filter in SST prediction are assessed, suggestions are made that the filtering approach may be an effective method to enhance air-sea coupling and improve forecast skill. Necessity of further study is addressed to improve ENSO prediction by reducing model biases combined with assimilating observations to initialization.