查詢結果分析
來源資料
相關文獻
- 偏貝氏時間序列模型分析日溫度資料和預報可行性的探討
- 溫度對鴨舌草種子發芽之影響
- 臺灣地區最高溫、最低溫長期趨勢之研究
- 溫度對種子發芽之區間設限存活分析
- 穀粒充實期間氣象因子對「臺稉9號」產量與米質性狀之影響
- 臺灣歷年溫差變化趨勢及其原因探討--以1954年至2003年為例
- 時間序列分析在品質管制上的應用
- 選擇權訂價貝氏GARCH模型及其在S&P 500指數選擇權買權價格預測上的馬可夫鏈蒙地卡羅模擬
- Comparisons of Time Series Models for the Forecasting of Albacore Commercial Harvest in the Indian Ocean
- 降雨量及地下水位之趨勢與變動分析
頁籤選單縮合
題 名 | 偏貝氏時間序列模型分析日溫度資料和預報可行性的探討=Analysis and Forecast of Daily Temperatures Using Partial Bayesian Time Series Model |
---|---|
作 者 | 李柏宏; | 書刊名 | 氣象學報 |
卷 期 | 48:2 2010.09[民99.09] |
頁 次 | 頁79-93 |
分類號 | 328.88 |
關鍵詞 | 貝氏統計; 溫度預報; 最高溫; 最低溫; 時間序列分析; Bayesian statistics; Forecast temperatures; Maximum temperatures; Minimum temperatures; Time series analysis; |
語 文 | 中文(Chinese) |
中文摘要 | 隨著數值動力模式的發展,統計降尺度預報上的探討,主要是以數值模式預報資料來進行統計預測,但數值模式預報資料面臨的問題為,數值模式初始資料的不確定性、參數化方法處理物理過程產生的誤差,以及海陸分布或是山脈地形影響等的複雜因素,這造成數值模式預報資料的不確定性,也增加了統計降尺度方法的預測誤差。因此有一部分的學者,如胡仲英(1977)、Tol(1996)、Franses等(2001),根據時間序列分析方法,探討測站觀測資料的特性,以期能對未來的資料進行合理的預測。時間序列分析方法有2個基本假設,第一是假設資料擁有長期的線性趨勢、第二是假設資料服從一平穩過程,但實際觀測到的資料大多無法滿足此2個基本假設。本文欲探討一個偏貝氏時間序列模型,以避免時間序列分析方法中的2個基本假設,使得分析者能從時間序列模型中獲得更多資料的特性。此模型是利用貝氏定 理,來放寬這2個基本假設,並將研究人員的長期觀測經驗引進模型中。另外,模型中也考慮到資料間的相關性,使得模型在使用上更有彈性,不需要像傳統的資料分析方法,忽略掉資料間的相關性。文中分析和預報臺北、臺中、高雄,和花蓮4個局屬氣象測站的日均溫、日高溫、日低溫結果。研究結果發現,偏貝氏時間序列模型在分析日溫度資料上,能提供研究者更完整的資料趨勢和週期特性因素;在預報上,和持續法的比較結果,4個測站平均都有15%到35%的正技術得分;而和氣候平均法的比較結果,以高雄站平均技術得分為-10%最差外,其餘測站則介於0到-6%之間,但大致上和氣候平均法沒有明顯差異。 |
英文摘要 | Numerical models have been developed very soon in recent years, but the problems for numerical models are the uncertainty of the initial data, physical parameterization, and the impact of topographic factors. Therefore statistical downscaling methods based on numerical model outputs will have more systematic errors which affects the accuracy of statistical forecast. Some papers (Hu 1977, Tol 1996, Franses et al. 2001) use time series analysis method to explore the characteristics of station data, which requires two assumptions, one is the time series data have a linear trend, the other is the data are realized values of a stationary time series, but most of data in the world do not satisfy the two assumptions. This paper explore the partial Bayesian time series model, which does not need the two assumptions, while is intended to generalize the two assumptions of time series analysis by the Bayes’ theorem and include the experiences of the experts to the prior probability density function. In addition, it also considers the correlation of data. Therefore, the model is more useful and provides more information of the data to the researchers. The model has been applied to daily temperatures of Taipei, Taichung, Kaohsiung, and Hualien stations from 1961 to 2009 in Taiwan. The results show the partial Bayesian time series model provides more complete information of trend and seasonal components of daily temperatures. To test the model forecast skill, 2000~2009 daily average temperatures, daily maximum temperatures, and daily minimum temperatures are forecasted and also compared with those by the two methods, persistence and optimal climate normals. The average skill scores for the four stations are 15% to 35% with persistence; the average skill score for Kaohsiung is –10%, and for other stations between 0% and –6% with optimal climate normals. Generally speaking, the partial Bayesian time series model is skillful than persistence and is not significantly different from optimal climate normals. |
本系統中英文摘要資訊取自各篇刊載內容。