| El Nino-Southern Oscillation (ENSO) is one of the strongest climate signals on the interannual time scale, acting as the dominant mode of large scale air-sea coupled interaction in the tropical Pacific Ocean. ENSO has direct impacts on the atmosphere and ocean over the Pacific region, and it plays an important role in modulating the climate anomalies over the middle latitude and even the global regions through teleconnections as well. Consequently, the study of ENSO predictability not only has important scientific significance on the understanding of the occurrence and development of ENSO, but also provides important theoretical foundation for improving the forecast skill of tropical Pacific climate variability and even the global weather and climate anomalies.By using the outputs of several CMIP5 models, we investigated the spatial characteristics and season-dependent evolution of the initial errors that could yield a prominent spring predictability barrier (SPB) phenomenon for ENSO events, and identified the sensitive area for targeted observation associated with ENSO forecasts. On this basis, a new "sensitive area based multi-model ensemble forecast method" was proposed by combining the sensitive area for targeted observation with multi-model ensemble forecast schemes. This new method was applied to the hindcast experiments of the tropical Pacific SST and the zonal wind in lower troposphere and proved to have considerably high forecast skill and very low computational expense. The main results are summarized as follows:1) The initial errors that could cause an obvious SPB for El Nino events and their evolutions were analyzed by using three CMIP5 model outputs (i.e. FGOALS-g2, BCC-CSM1.1 and NorESMl-M model). The results indicated that for the SST forecasts of El Nino events in these three models, the significant growth of prediction errors usually occurs in spring or the beginning of summer, as the typical SPB phenomenon arises. Further study showed that the initial errors yielding a significant SPB have particular spatial structure, which can be classified into two types:type-1 initial errors often present an SST pattern with positive errors in the eastern equatorial Pacific (5°-5°N; 150°W-90°W), and the subsurface temperature has a dipolar pattern with positive anomalies in the eastern equatorial Pacific and negative anomalies in the central-western equatorial Pacific, the signals are concentrated at depths of 100-155 m in the central-western equatorial Pacific (5°S-5°N; 130°E-180°E); type-2 errors have a spatial structure opposite to that of type-1. Then the evolutions of the two types of initial errors for strong and weak El Nino events are investigated. For strong El Nino events, both types of initial errors which have opposite spatial structure ultimately lead to the weaker predicted El Nino events or missed events; while for weak El Nino events, the prediction errors caused by the two types of initial errors are nearly opposite. The former could be attributed to the greater nonlinear effects in strong events; in contrast, the nonlinear effects in the latter are much weaker.2) The relationship between the initial errors that cause the SPB of El Nino forecasts and the precursors for ENSO events were investigated. It was concluded that for both the strong and weak El Nino events, the spatial structure of the initial errors yielding the SPB bears a high resemblance to that of the precursors, and their spatial patterns show localized characteristics with large values concentrating at the surface of eastern tropical Pacific and the subsurface 100-155 m in the central-western equatorial Pacific. This indicated that targeted observations in the localized sensitive areas could improve the accuracy of initial fields, and it would also be helpful in capturing the signals of the precursory disturbances. The relationship between the initial errors in the sensitive areas and the ENSO prediction errors were further investigated. The results indicated that the smaller initial errors in sensitive regions are key to higher forecast skill, and vice versa. Therefore, the large value regions are much likely to be the sensitive areas for ENSO forecasts.3) Previous research has indicated that the ENSO forecasts also have strong sensitivity to the model tendency errors over the eastern equatorial Pacific region. If the model uncertainties over this region can be firstly reduced, the ENSO forecast skill may be dramatically improved. We applied the sensitive area information to the multi-model ensemble forecast method, and proposed a new "sensitive area based multi-model ensemble forecast method". The advantage of this method is that it has comparable forecast skill with traditional multi-model ensemble methods, and evidently decreases the computational cost. In the long-term hindcast experiments of the tropical Pacific SST, we employed the superensemble prediction method (SUP) with higher skill in sensitive areas but kept the relatively simple and lower-skill bias ensemble mean method (BREM) used in other regions, namely the proposed sensitive area based multi-model ensemble forecast method. The results showed that for the forecasts of SST over the entire tropical Pacific, the proposed method has very close prediction skill to that of the SUP in the forecast period, while the computational cost is only one-fourth of that of the SUP. The results not only demonstrated the effectiveness of this new method, but also verified that the SST forecasts for the Pacific area are very sensitive to the model errors over the equatorial eastern Pacific region.4) The target observation sensitive region of the tropical Pacific zonal wind was presumably deduced by the interaction between the sensitive region of the tropical Pacific SST and the tropical wind field, namely the central equatorial Pacific. The validity of the sensitive region in the forecasts of tropical Pacific zonal wind field was further verified by using the exhaustive method. In the hindcast experiments of the wind field, the experimental setup that the forecast skill of the tropical Pacific wind using the SUP with higher skill in the central equatorial Pacific (5°S-5°N; 157.5°E-142.5°W) and the BREM in other regions (namely the "sensitive area based multi-model ensemble forecast method") was employed to forecast the wind field. It was found that the new scheme has very close skill compared to the SUP used over the whole region, and significantly reduces the calculation cost. Therefore, the wind forecasts for the Pacific Ocean are very sensitive to the model errors over the central equatorial Pacific as well. The physical and dynamical mechanism was further clarified to support the results. In all, the above hindcast experiments and mechanical explanation verified the sensitive area of the tropical Pacific low level wind forecasts.Abover all, the "sensitive area based multi-model ensemble forecast method" introduced in this dissertation is based on the theoretical analysis and numerical hindcast experiments, thus has a solid theoretical foundation and potential application value. This method is expected to play an important role in operational forecasting in the future. |
PDF Full Text Request