查詢結果分析
來源資料
頁籤選單縮合
| 題 名 | 聖嬰/反聖嬰(El Nino/La Nina)統計預報之發展=A CCA Model for El Nino/La Nina Prediction |
|---|---|
| 作 者 | 陳孟詩; 盧孟明; | 書刊名 | 氣象學報 |
| 卷 期 | 44:4 2002.12[民91.12] |
| 頁 次 | 頁25-39 |
| 分類號 | 328.8 |
| 關鍵詞 | 聖嬰; 正準相關分析; 統計預報模式; ENSO; Canonical correlation analysis; Statistical model; |
| 語 文 | 中文(Chinese) |
| 中文摘要 | 本研究利用正準相關分析法發展聖嬰/反聖嬰統計預報模式,預報因子為前四季之季平均海平面氣壓場,預報對象則為赤道地區季平均海溫。 研究結果顯示,模式發展期之冬半年預報表現較夏半年好,就短期氣候預報而言,雖然持續法較正準相關模式佳,但中長期氣候預報則是正準相關模式優於持續法。正準相關模式亦有能力預報海溫之年際變化及季節變化,但有較冷誤差,預報聖嬰比反聖嬰要來得好,雖然模式對主要的聖嬰/反聖嬰事件開始時間預報落後,但對其結束時間則掌握地相當不錯。由相關係數及均方根誤差評估模式之預報技術顯示,不同海溫指標之預報以Nino3.4最好,而Nino3.4冬季之領先1季預報優於其他領先預報,若考慮應用至實際預報作業之時效需求,則Nino3.4領先2季預報之表現於冬半年較佳。 本研究亦針對正準相關模式進行敏感度測試,結果發現改變預報因子區域及解析度、預報對象選取8個海溫指標、預報對象扣除表現較差的2個海溫指標及資料前置處理未含去趨勢化,並未改善模式表現,預報因子加入預報對象本身只在領先3季預報略有改進,而模式發展期改為1960-1999年對模式預報改進幫助較大,因此實際預報作業建議採用最近40年為模式發展期。 檢視模式於預報實驗期之表現,正準相關模式對於聖嬰/反聖嬰事件之強度預報不足,而轉變時間之掌握梢為落後,雖然表現不及ENSO-CLIPER(Landsea and Knafr,2000),但較持續法佳。 |
| 英文摘要 | The El Nino/La Nina prediction model using Canonical Correlation Analysis (CCA) is established. Seasonal mean sea level pressure (SLP) of prior four seasons is used to predict seasonal mean tropical sea surface temperature (SSD. During the training period (1956-1995), the model has better performance in winter than in summer. The cross-validation results show that persistence method is better than CCA within 6-month forecast, but CCA is more skillful in longer lead-time forecast. CCA model is also capable of predicting the interannual and seasonal variations of SSTs, though with cold error. In general, El Nino prediction is better than La Nina. Although the forecast of El Nino/La Nina beginning falls behind the observation, the forecast of ending does not. Based on correlation and root mean square error (RMSE) evaluation, Nino3.4 is an ENSO index being best predicted. For one season lead forecast, the correlation of the predicted and the observed Nino3.4 in winter is higher than 0.9. For two seasons lead forecast, the correlation can be as high as 0.89. Several sensitivity tests have also been studied, namely, the sensitivity of the spatial coverage in the predictor and predictand, the effects of detrend and the selection of training period. We find that the prediction results are most sensitive to the selection of training period. It is suggested that it is better to use the latest 40 years as the training period in real forecasts. The model hindcast results during 1996-2000 show that the predicted intensity of ENSO is weaker, and the phase transition time is delayed compared with the observation. Although we do not find that CCA is better than ENSO-CLIPER, we find it outperforms the persistence method. |
本系統中英文摘要資訊取自各篇刊載內容。