臺灣地區降雨模擬的動力與統計降尺度的比較與分析

Abstract

過往直接比較動力降尺度與統計降尺度的研究較少，大部分研究進行降尺度比較或評估時，通常是以實際的觀測資料為基準，評估各個降尺度方法。動力降尺度方法藉由趨近真實世界的資料提高氣候模式的解析度，獲得高解析度的氣候推估資料，運算出的結果能呈現不同地理與氣候特徵。統計降尺度方法則利用過往的觀測資料與氣候模式模擬資料，以回歸和統計分析方法建立兩者之間的統計關係。相較於動力降尺度方法，在電腦計算的要求相對較低，方法上也更加簡單；但是統計降尺度模擬的結果沒有辦法呈現物理過程，並且缺乏適合的物理解釋，應用時需要假設未來的氣候特徵在統計上是穩定的；然而近年來氣候變遷影響下，這個統計穩定的假設通常是無法滿足且無法證明。對此本研究以「理想模式」作為實驗架構，使用動力降尺度資料代替降尺度中的觀測資料，驗證統計降尺度方法，並以全球高解析度模式資料作為參考，比較動力與統計降尺度在不同季節、極端降雨之降尺度成效。研究結果顯示統計降尺度在梅雨和夏季容易受到高解析度模式資料影響，與動力降尺度差異較大，而統計降尺度與動力降尺度在冬季較為接近；然而，極端降雨主要集中在梅雨和夏季，對於系統性降雨如颱風、梅雨等，統計降尺度與動力降尺度具有接近表現，但它依然受模式資料影響為主；極端降雨能觀察到冬季的成效比梅雨和夏季高，趨近動力降尺度的比例相比所有天數降低。

There have been few studies that directly compared dynamical downscaling and statistical downscaling in the past. When most studies conduct downscaling comparisons or evaluations, they usually evaluate various downscaling methods based on actual observation data.The dynamical downscaling method improves the resolution of climate models by approaching real-world data and obtains high-resolution climate estimation data. The calculated results can present different geographical and climate characteristics. The statistical downscaling method uses past observation data and climate model simulation data to establish the statistical relationship between regression and statistical analysis methods. Compared with the dynamical downscaling method, the computer calculation requirements are relatively low and the method is simpler. However, the results of statistical downscaling simulations can’t represent physical processes and lack suitable physical explanations. It is necessary to assume that future climate characteristics are statistically stable. Nevertheless, this assumption of statistical stability is often unsatisfied and cannot be proven under the influence of climate change in recent years. In this regard, this study uses the"perfect model" as the experimental framework. Dynamical downscaling data are used to replace the observation data in downscaling to verify the statistical downscaling method. We also use global high-resolution model data as a reference to compare the effect of dynamical and statistical downscaling in different seasons and extreme rainfall. The research results show that statistical downscaling is easily affected by high-resolution model data during the Meiyu and summer periods, and is differentfrom dynamical downscaling. Statistical downscaling and dynamical downscaling are relatively close in winter. Extreme rainfall is mainly concentrated in Meiyu and summer. For systematic rainfall such as typhoons, Meiyu, etc., statistical downscaling and dynamical downscaling have similar performances, but it’s still affected by model data. Extreme rainfall is more effective in winter than in Meiyu and summer. The proportion of approach dynamical downscaling decreases compared to all days.