用統計法研究臺灣南部地區降水量之長期預報

A Study of Long-Term Precipitation Forecast for the Southern Taiwan by Using a StatisticaI Method

胡仲英(Chung-Ying Hu)

《大氣科學》 5卷1期 (1978/05) Pp. 27-34

摘要

本文用統計加法模型研究臺灣南部地區降水量長期預報的可行性。影響降水量時間數列的因素可歸納成下列四種：（1）長期趨勢（Secular trend），（2）季節變動（Seasonal variation），（3）循環變動（Cyclical variation），（4）不規則變動（Irregular variation）。本文首先用最小平方差法（Least square error method）找出代表該時間數列長期趨勢的線性函數，再由各月別降水量的中位數（Median）找出降水量的季節變動，然後用調和分析（Harmonic analysis）計算影響降水量的主要循環變動，最後用隨機線性自廻歸模式（Stochastic linear auto-regressive model）來處理不規則變動。作者用此統計加法模型作臺灣南部地區1963年至1976年每年5月至10月降水量的分級預測（Rank forecast），由校驗結果顯示此法是相當可行的。最後，作者並將此法與ARIMA時間數列預測法的校驗結果作比較，可看出此法有明顯地改進。

並列摘要

This paper presents a statistical additive model to study the feasibility of long-term precipitation forecast for the southern Taiwan area. First, the secular trend of the time series was fitted linearly by employing the least square method, then the seasonal and cyclical variations were obtained by monthly median and harmonic analysis respectively. Finally the stochastic linearly auto-regressive model was applied to derive the irregular variation of the said time series.Using this additive model we have forecasted the ranks of monthly precipitation for the southern Taiwan area from May to October during the period 1963 to 1976. The result of verifications shows that this model is quite feasible. In comparison with time series ARIMA model (Box and Jenkins, 1970) it is found that this model performs much better.