本研究之田間試驗係在位於臺南縣新化鎮之農委會畜產試驗所農場進行，自2002-2004年的尼羅草(Acroceras macrum cv. Taishigrass No. 1)9生長作季期間不定期的取樣地上部鮮重(牧草鮮產量)，同時亦收集鄰近一級農業氣象測站量測之作季內每小時最高、平均與最低氣溫。其中6生長作季資料供作建立牧草生產量與氣溫關係模式之用，另3作季資料則作為模式之驗證。根據試驗結果，牧草鮮產量與生長作季累積每小時氣溫具有顯著相關，適用於二次曲線函數，無論採用每小時最高、平均或最低氣溫作為模式參數之決定係數(R^2)皆高於0.66 (P<0.0001)。以每小時平均氣溫為參數之迴歸方程式進行模式驗證，發現其估值可以解釋預測牧草產量之93%變異。結果亦顯示，日間及夜間累積之每小時平均氣溫亦相關於牧草生產量，若改以日間與夜間累積氣溫之溫差，將可提高所建立模式之準確性。惟經驗證此一模式之適用性，發現產量預測結果較差。綜合試驗結果，顯示本研究以氣溫建立預測尼羅草牧草鮮產量之關係模式具有一定可行性及適用性，惟為獲得較佳生產量預測之準確性，宜收集更廣環境變異資料融入於模式之改進及適用性之驗證。

The aboveground fresh weights (forage fresh yield) were measured and the maximum, mean, and minimum hourly air temperatures were recorded for nilegrass (Acroceras macrum cv. Taishi No. 1) vegetation grown in the experimental pasture of Taiwan Livestock Research Institute (Hsinhua, Taiwan) during the nine growing seasons in 2002-2004. Data from 6 seasons were used to establish the forage yield-air temperature relationships and the rest data from the other 3 seasons were used for models' validation. As the results shown, the relationships between forage fresh yield and accumulated hourly air temperatures were found fitted to a quadratic function with R^2 value greater than 0.66 (P-value less than 0.0001), irrespective to use maximum, mean, or minimum hourly air temperature as temperature variable. When using the regression equation with mean hourly air temperature as input for validation, result demonstrated its applicability in accounting for more than 93% variability of yield prediction. Results further indicated that the accumulated mean air temperatures of daytime and nighttime hours were also corresponded to forage fresh yield. The relationship can be improved by substituting with the values of temperature difference between daytime and nighttime hours, yet the applicability was in a lesser extent. As a result, the established regression equations using air temperature as a predictor have proved their feasibility and applicability as tools for predicting forage yield. However, more data covering a wider range of environmental variation is needed for models' improvement and validation if an accurate yield prediction if of concern.