本研究利用三種不同的客觀分析診斷工具，測試對於辨識臺灣地區鋒面事件的能力。基於鋒面熱力參數(TFP)之空間變異所發展之診斷工具，在診斷分析時，可同時提供鋒面位置與強度等資訊，有助於推估受鋒面影響之降雨之空間分布與降雨強度變化。另外，以自組織映射圖(SOM)群聚分析技術發展之診斷工具，為非監督式學習架構，可大幅降低人為主觀判斷所導致之誤差。診斷的同時，將各類型相似環境特徵的事件加以標示，對於推估氣候變遷下，伴隨鋒面系統之劇烈天氣事件發生頻率變異提供極佳的資訊。考量到計算資源限制與降低主觀誤差等要素，我們發展基於機器學習技術之診斷工具。針對鋒面系統判定之測試結果顯示，無論是以線性或非線性模組作為核心之演算模型，均能掌握鋒面頻率變化之長期變異。傳統客觀分析方法診斷影響臺灣之鋒面系統能力較差，各季節之命中率約0.1-0.25，但誤報率較低。SOM診斷工具在命中率的表現些微優於機器學習方法，但同時伴隨著誤報率偏高的缺點。相較於傳統客觀分析方法，這兩種新的客觀診斷工具之系統判定成效皆較佳。整體而言，以SOM技術為基礎之診斷工具能診斷出大部分鋒面事件，但有過度判定的傾向。而基於機器學習方法之診斷工具，在命中率上與SOM相距不遠，但卻可大幅降低誤報率，對於長期鋒面系統頻率變化有較好的掌握能力。

In this study, we used three different objective classifiers to evaluate the performance of frontal system diagnosing in Taiwan. First, the traditional objective diagnostic method which is based on the spatial variability of frontal thermal parameters (TFP). This method can provide frontal position and intensity while diagnosing process, which can use to estimate the spatial distribution of precipitation amount and rain intensity variation. The second method be evaluated is the self-organizing map (SOM)-based classifier. It is an unsupervised learning method, which can reduce the subjective error. It can also provide the clustering results of different weather types with similar atmospheric characteristics. Finally, considering the limitations of computing resources, we developed a third method, new diagnostic tool which is based on machine learning techniques. The results of the frontal system diagnosing show that the long-term variation of the front frequency can be well represented by both linear and nonlinear kernels. The traditional objective diagnostic method has a poor ability to identify the frontal system in Taiwan. The diagnosing hit rate was only 10-20% of all events and with very low false alarm rate. In the other hand, comparing with the machine learning mode, the SOM diagnostic method has higher hit rate (70%-80%); however, the false alarm rate is also higher (20%-60%). Overall, SOM classified more frontal events than actual, with overdiagnosis issue. The diagnostic tool based on the machine learning method can greatly reduce the false alarm rate and has a better diagnostic ability for frontal frequency variations.