無母數統計降尺度模式之開發與實例應用

Abstract

近年來溫室效應所造成的氣候變遷議題日益受到重視，而全球大氣環流模式(GCM)的模擬結果，常為評估氣候變遷所造成之衝擊的重要依據。然而GCM模式的空間尺度一般均遠大於流域尺度，因此，如欲應用(GCM)的模擬結果去探討氣候變遷對流域尺度之區域水文或水資源的影響，往往需要透過降尺度模式，建立大尺度氣候因子與小尺度集水區或流域水文量間的關係。傳統降尺度方法均以某種數學函數，描述大小尺度資料間的對應關係，然而由資料顯示往往兩者間雖有某種程度之相關，惟此對應關係卻常顯示出相當大的變異性與不確定性，因而較適合以條件機率的概念來描述。因此本研究乃應用條件機率的觀念，建立一個無母數統計降尺度模式(Non-Parametric Statistic Downscaling Model, NPDM)，直接由大小尺度資料統計出，以選定之大尺度氣候因子為條件之集水區雨量的無母數條件機率分佈，再進一步以此進行大尺度氣候因子為條件下之集水區雨量合成。本研究另外與SDSM模式之雨量合成結果作比較，結果顯示在考量相同之大尺度氣候因子下，NPDM模式之合成資料較SDSM之合成結果，在統計特性上更近似觀測資料，其中又以豐水期(五到十月)之月標準差最為顯著，因此證明NPDM模式可為合適之統計降尺度模式，其合成資料可作為後續其他水資源相關評估之用。

Climate change caused by the greenhouse effect and its related issues are becoming an increasing concern. The simulation results of general circulation models (GCMs) are an important basis for evaluating the impact of climate change on a river basin. However, the spatial scale of a GCM is generally much larger than a river basin. Therefore, a downscaling model is required to relate the GCM data to basin weather information before the GCM model can be applied to studying the impact of climate change on a basin scale problem. Conventional downscaling methods use a mathematical function to describe the data relation between large scales and basin scales. Although there is some degree of correlation between the data of these different scales, the relationship is highly uncertain and random. Instead of a deterministic function, a conditional probability distribution can better describe this scale relationship. Therefore, this study proposes a novel non-parametric statistic downscaling model (NPDM) to describe the scale relationship between selected large scale factors and basin rainfall using a non-parametric conditional probability distribution. This study also compares the NPDM performance with that of the statistical downscaling model (SDSM) through synthesis of basin rainfall. The results of this comparison show that, using the same large scale climate factors, the statistical parameters such as sample mean and standard deviation of the NPDM synthesis data is closer to the statistical parameters of observation data than those produced by the SDSM. This study demonstrates that the proposed novel NPDM is an appropriate downscaling model and its data can be used to investigate the impact of climate change on regional water resource management.