現今茶農預測茶葉採收期時,往往需仰賴其主觀經驗進行預測。然而茶葉生長狀況往往受溫度、雨量、剪枝活動及海拔高度等不確定因素影響,造成茶葉對應之採收日期並不恆定,以致於茶農不易準確預測、掌握茶葉採收日期。另一方面,茶農除需耗費精力觀察茶葉生長狀況外,亦需花費大量時間等待茶葉成熟後再判斷茶葉採收日期、進行茶葉採收之相關準備,此舉甚為耗時且徒增茶農之工作負荷。因此,本研究乃提出一套以「茶葉採收期影響因子」為基礎之「茶葉採收期預測模式」,此模式可適用各地不同海拔高度之茶園,以供各地茶農準確且有效地預測茶葉採收時間,並充分進行茶葉採收之相關準備。 為達成前述目的,本研究乃先釐清影響茶葉採收期之各種因素,並依所確定之影響因子建構對應之彙整及量化模式(即「茶葉採收期影響因子之彙整及量化模式」)。之後,本研究再以此彙整及量化模式為基礎,發展多組可預測茶葉採收期之多元多次迴歸模式,並從中選取一組最適之茶葉採收期預測模式(即「決定多元多次茶葉採收期預測迴歸模式」)。最後,本研究再以各地茶園之歷史採收期資料為實務案例,驗證所發展之模式可行性。而由驗證結果分析可知,本研究所提出之「茶葉採收期預測模式」可有效地推論茶葉採收期日數(第二階段模式績效驗證之長期學習趨勢下推論的平均誤差率為5.88%),顯示此模式確實可作為茶農預測茶葉採收期之決策參考。 整體而言,本透過此「茶葉採收期預測模式」,茶農進行茶葉採收期之預測時僅需輸入目標茶園之雨量、溫度、剪枝活動及海拔高度等相關資訊,即可求得此模式下所對應之茶葉採收期,以提供茶農預作各項準備(含臨時人力招募)之參考。
Nowadays, tea farmers often rely on their subjective experience and judgment to predict the harvest time of tea. However, the growth of tea is affected by a lot of uncertain factors such as temperature, amount of rainfall, pruning activity, and sea-level altitude. All the factors result in uncertainty on tea harvesting period and the tea farmers can hardly predict and control the harvest time. In addition to spending time to observe the growing condition of tea leaves, tea farmers have to spend much time to wait tea leaves becoming mature so as to prepare for harvesting jobs. This farming cycle is quite time-consuming and is also a heavy burden to tea farmers. Thus, based on the factors of tea harvest period, this research develops a tea harvest prediction model, which can be applied to any tea gardens in different sea-level altitudes; tea farmers may use this model to accurately and effectively predict the harvest time. This research firstly clarified various factors that have effects on the harvest time. Secondly, this research established a quantitative model according to the factors. Afterward, on the basis of the quantitative model, this research developed several multivariate multiple regression models to predict the harvest time. One of the predictable multivariate multiple regression models was selected as it is the most suitable for predicting the harvest time. At the final stage, this research took some historical records of tea harvesting period collected from tea gardens as cases to verify feasibility of the model. The verification result indicates that the harvest prediction model proposed by this research can efficiently determine the number of day for tea harvesting. That is, the model can be a reference tool for tea farmers to predict the harvest time.