本研究先以調和分析法分析台灣地區氣候資料，求取其週期性變化和長期趨勢，並以
各地之自迴歸積分移動平均時序統計模式(ARIMA Model) 模擬各地區氣候變化，建立
台灣地區定量預報之氣候變遷模式。
文中詳細討論調和分析法與時序統計模式，並利用台灣地區北、中、南三測站之84年
氣候資料進行分析。研究結果顯示：(1) 調和分析法為診斷氣候變遷最佳之方法，但
不適用於以外延法進行預報。(2) 研究三區的降水量發現南北較差較大，東北季風之
影響到達恒春地區已近強弩之末，因此冬天十分乾燥。(3) 氣壓、氣溫之變化為中緯
度綜觀天氣系統所控制，均有一增高之長期趨勢。(4) 分析降水量顯出有不規則之變
化，正確之氣候預報模式必須具備分析此種不規則變化的能力。(5) 時序統計最佳模
式的預報值與實際觀測值相較近似符合，並能掌握各種不規則之變化。
本研究以1972～1980年間108 個月和36個季節的預報值作為模式之驗證，計算其預報
誤差之配置均為常態分布。逐月氣壓之預報誤差絕對值平均為0.70mmHg(1mmHg =
1.333mb)，有72次預報的誤差小於0.50mmHg；逐季氣壓之預報誤差絕對值平均為0.75
℃，有82次預報誤差小於0.40mmHg。逐月氣溫之預報誤差絕對值平均為0.75℃，有82
次預報的誤差小於0.50℃；逐季氣溫之預報誤差絕對值平均為0.52℃，有31次預報的
誤差小於0.40℃。逐月總降水量預報誤差絕對值平均為38.0mm，有68次預報的誤差小
於25.0mm；逐季總降水量預報誤差之絕對值平均為30.0mm，有25次預報的誤差小於20
.0mm，而極端誤差發生次數佔全部預報次數的2.78％。
本研究具有地區共同性，對於不同地區之氣候變化，可以相類之模式進行預報與分析
。此外本研究亦討論時序統計模式在台灣地區氣候變化分析和長期預報可能的發展。
///////
Harmonic analysis was first applied to the climatic datain Taiwan in this
study in order to see its periodic changes and secular trends. Later the
ARIMA model has been introduced to verify the degree of success on
quantitative forecasting of the climatic changes in this area.
Theorectical basises on the harmonic analysis and the ARIMA model have
been discussed to some extent, some views on the application of ARIMA
model have been drawn at the following points:
(1) Harmonic analysis is suitable to find any periodic changes on climate
as a whole, but it doesn''t seem adjustable to make a forecast through
extrapolation.
(2) Rainfall fluctuations are found to be bigger than the other elements.
It is the reason why floods and droughts are frequently happened in
Taiwan. The strengty of winter monsoons was weak when it swang over
Hengchun area and caused a dry winter.
(3) The synoptic pattems in midlatitude revealed to have an increase on
both pressure and temperature changes in long-range tendencies.
(4) Irregularities on the abrupt change of rainfall were particularly
noticeable. An additional factor must be inserted into the formula before
a more precise climatic forecast can be made.
(5) The calculated ARIMA model had been evaluated on observations in mant
ways and they are believed to be the proper instruments.
Monthly and seasonal forecasts had been performed during the period of
1972-1980. Their standard errors form normals are presented in
tabulation. The average of absolute errors in monthly pressure
forecasting is 0.70 mmHg(1 mmHg = 1.333 mb); among them ,72 errors are
less than 0.50mmHg. The average of absolute errors in seasonal pressure
forecasting is 0.50 mmHg;30 errors are less than 0.40 mmHg. The average
of absloute errors in monthly temperature forecasting is 0.75℃;82 errors
are less than 0.50℃. The average of absolute errors in seasonal
temperature forecasting is 0.52℃;31 errors are less than 0.40℃. The
average of absolute errors in total monthly rainfall forecasting is 38.0
mm; 25 errors are less than 20.0mm. The frequency of extreme errors is
2.78%of all forecasting errors.
The simulated ARIMA model in this paper may be applied to other areas on
making any forecasts or analysis on climatic changes on consequence of
they have the similarity in commom. The development of the ARIMA model in
future may be improved for long-range forecasting in Taiwan area when more
fruitful data are implemented.