Intercomparison of weather generators for hydrological applications at varied scales

氣象資料(如降雨量、溫度以及蒸發)在水文氣象方面的研究相當重要，然而，由於大部分氣象資料在紀錄上經常會有缺漏，或其年限不足以進行長期水文模擬。近幾十年來發展的氣象資料產生器(weather generators)可用來產生保有觀測資料統計特性之長期氣象資料，有效解決了上述問題。 絕大多數氣象資料產生器皆由國外學者所開發，鮮少有利用台灣地區資料進行測試，此外，氣象資料產生器種類眾多，可分為單站 (single-site) 與多站 (multi-site)，對於雨量的模擬方法不全相同；本研究旨在挑選最適合台灣地區的氣象資料產生器。 本研究利用四種氣象資料產生器，即Modified Wilks approach、Stochastic Climate Library (SCL)、Multi-site stochastic weather generator École de Technologie Supérieure (MulGETS)、以及Weather GENerator (WGEN)，模擬翡翠水庫以及八掌溪集水區內各測站之降雨，將降雨模擬結果分月份與季節進行統計特性的比較，爾後輸入HEC-HMS模式模擬流量，比較不同方法之間的模擬結果；本研究亦利用氣象資料產生器對分布於全台之多個氣象測站進行雨量模擬，以事件分析以及主成分分析比較颱風時期(七至九月)之模擬雨量空間分佈與觀測雨量之異同。不同空間尺度(集水區與全台灣)之分析結果可助於評選出最適之氣象資料產生器。 綜合所有研究區域的模擬結果，可以知道MulGETS降雨模擬時，能夠保留觀測值的空間分布，另外，在降雨量的模擬結果上，透過多種統計指標的比較，MulGETS的表現相當不錯，幾乎優於其他氣象資料產生器。而在流量模擬上，MulGETS和SCL之模擬結果相近，皆能有效保留流域內的大流量，WGEN-lumped在降雨分布較均勻的區域內也能夠合理的表現出流域的大流量。在經由各種評比後，得知MulGETS相當適合用來模擬台灣地區之天氣資料。Meteorological data such as precipitation, temperature, and evaporation are of critical importance to hydro-meteorological research. However, missing and insufficient data records are common but unfavorable conditions for long-term hydrological simulation and applications. Weather generators (WGs) are thus designed to generate weather data that preserve the observed statistical characteristics and provide a preferable period of records for various applications. Over the past few decades, the development of WGs has been evolving from single- to multi-site WGs to account for the spatial coherence of weather data. Although related literature is abundant, investigations into WGs at different climate regions are seldom found. The purpose of this study is to assess and select the most adequate WGs for hydrological applications in Taiwan. In this study, we adopted four WGs including the Modified Wilks approach, Stochastic Climate Library (SCL), Multi-site stochastic weather generator École de Technologie Supérieure (MulGETS), and Weather GENerator (WGEN). We assessed the four WGs by comparing the statistics of generated rainfall data with that of observed in the Feitsui reservoir and Bajhang river basins. We also used HEC-HMS to examine the simulated flow, driven by the generated rainfall data. The other part of this study, as a larger-scale assessment, is to generate and assess rainfall data of the 21 major weather stations over Taiwan. We applied principal component analysis (PCA) to both the observed and generated rainfall data in the typhoon period (July to September) for this assessment. The outcomes from the watershed- and island-scale assessments should be a clear indicator of the most adequate WG. In sum, MulGETS can not only preserve the similar statistical characteristics of observed rainfall at the watershed scale but also provide a reasonable spatial distribution of rainfall over Taiwan. MulGETS and SCL can retain large discharge with the highest efficiency, yet the WGEN-lumped approach can also produce large discharge with uniform spatial distribution of rainfall across the watershed. Even though MulGETS showed marked overall performance, seasonal differences in the performance of these WGs can be identified.摘要............................................i Abstract.......................................ii 目錄............................................iv 表目錄..........................................vii 圖目錄..........................................viii 第一章 緒論......................................1 1.1. 研究動機與目的...............................1 1.1.1. 研究動機...................................1 1.1.2. 研究目的....................................2 1.2. 研究架構......................................2 第二章 文獻回顧.....................................4 2.1. 氣象資料產生器面向.............................4 2.2. 水文模式面向...................................6 第三章 研究區域與數據................................8 3.1. 氣象局局屬測站.................................8 3.2. 翡翠水庫......................................10 3.3. 八掌溪流域.....................................12 第四章 研究方法.....................................15 4.1. 氣象資料產生器.................................15 4.1.1. Weather GENerator(WGEN).....................16 4.1.2. Wilks Approach..............................17 4.1.3. Modified Wilks Approach.....................21 4.1.4. Multi-site weather Generator of École de Technologie Supérieure (MulGETS)...............................24 4.1.5. Stochastic Climate Library (SCL)............27 4.2. 水文模擬系統 (HEC-HMS).........................31 4.2.1. 模式簡介.....................................31 4.2.2. 模式參數設定與檢定............................34 4.3. 主成分分析(Principal components analysis, PCA)..35 4.4. 統計指標........................................36 4.5. 相對誤差 (Relative Error, RE)...................39 第五章 結果與討論.....................................40 5.1. 中央氣象局局屬測站...............................40 5.1.1. 氣象資料產生器模擬結果.........................41 5.1.2. 降雨空間分布...................................62 5.2. 翡翠水庫........................................70 5.2.1. 氣象資料產生器模擬結果.........................70 5.2.2. 水文模式模擬結果...............................90 5.3. 八掌溪流域......................................97 5.3.1. 氣象資料產生器模擬結果.........................97 5.3.2. 水文模式模擬結果..............................117 第六章 結論與建議....................................123 6.1. 結論...........................................123 6.2. 建議...........................................124 參考文獻............................................126 附錄................................................130 附.1. 中央氣象局局屬測站雨量模擬結果..................130 附.2. 翡翠水庫雨量模擬結果...........................162 附.3. 八掌溪流域雨量模擬結果.........................194 附.4. HEC-HMS參數...................................22