- 學位論文
發展以自組織映射圖網路為基礎之雨量繁衍模式於未來降雨推估
Development of A Self-Organizing Map Based Daily Precipitation Generator for Future Projection
摘要
對於許多水文設計，日降雨資料為重要的資訊。本研究以自組織映射圖網路(Self-organizing map, SOM)為基礎建置日雨量繁衍模式，使用大氣環流模式(general circulation model, GCM)大尺度大氣因子模擬結果作為預測因子(predictor)，推估未來氣候情境下小尺度地面降雨變化趨勢。 本研究發展的模式包含兩部份：SOM為基礎之天氣型態分類和雨量繁衍。首先蒐集與降雨相關之大氣因子，使用自組織映射圖網路群聚分析法對大氣因子作分類，所分出之每個類別即為一種天氣型態。進而，計算各天氣型態的統計特性：天氣型態出現頻率、天氣型態維持天數、天氣型態轉換機率。接著分析各天氣型態內之日降雨資料，計算其降雨統計特性：降雨機率、濕天平均降雨量。最後，利用前兩步驟計算得到的統計特性以建構雨量繁衍模式。 此外本研究於兩種假設情況下，將所建構的雨量繁衍模式應用於評估氣候變遷下未來降雨量的變化。於第一種情況，假設天氣型態之統計特性在未來情境下需要修正，而降雨統計特性則保持與基期相同。於第二種情況，假設天氣型態之統計特性與降雨統計特性在未來情境下皆需要修正。在各情況下，根據GCM未來大氣因子模擬值，以本雨量繁衍模式產生未來日降雨資料。以實際案例評估本研究所建構之模式表現，顯示使用本模式所繁衍之日雨量資料可以保有觀測資料之統計特性，本模式確可用來繁衍小尺度集水區降雨資料。因此，本研究亦根據GCM的未來模擬結果，產生未來之日降雨推估資料。應用到石門水庫集水區顯示，無論是模擬情況一或是情況二，未來降雨於冬季與夏季降雨有增加的趨勢，而春秋兩季降雨有減少的趨勢。
並列摘要
Daily precipitation data are frequently required for many hydrological applications. In this study, based on self-organizing map (SOM), a daily precipitation generator is proposed. The large-scale atmospheric variables simulated by General Circulation Models (GCMs) are used as predictors to estimate the local-scale precipitation under different scenarios. The proposed generator consists of two parts: the SOM-based weather type classification and the precipitation generation. Firstly, the atmospheric variables that have great influence on precipitation are clustered using SOM to indentify different weather types. The statistics of each weather type (the frequency, the spell duration, and the weather-type transformation probability) are calculated. Secondly, for each weather type, the statistics of corresponding daily precipitation data (the probability of precipitation and the average wet-day precipitation amount) are also calculated. Thirdly, according to the aforementioned statistics, the precipitation generator is constructed. Additionally, in order to apply the proposed generator to assess the impact of future climate change on precipitation, two different conditions are considered herein. For the first condition, the statistics of each weather type will change, but the statistics of precipitation in each weather type will remain the same in the future. For the second condition, both the statistics of each weather type and those of precipitation in each weather type will change in the future. For each condition, the proposed generator is used to generate future daily precipitation according to the future simulations of atmospheric variables from GCMs. An actual application to the Shi-Men Reservoir Watershed is conducted to evaluate the performance of the proposed generator and to assess the future variation of precipitation. Results show that the synthetic daily precipitation data generated by the proposed generator can preserve the statistical characteristics of the observed data, and hence the proposed generator can be used to generate the local-scale precipitation. As to the future precipitation in the study watershed, for both aforementioned conditions, the precipitation will increase in winter and summer and decrease in spring and fall.