Please use this identifier to cite or link to this item:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/57634
Title: | 基於雨量型態的統計降尺度方法研究 A statistical downscaling method based on rainfall patterns |
Authors: | Jyue-Ting Wu 吳爵廷 |
Advisor: | 林國峰(Gwo-Fong Lin) |
Keyword: | 氣候變遷,降尺度,去勢波動分析法,支援向量機,改良式自組織線性輸出映射圖, Climate Change,Downscaling,Detrended Fluctuation Analysis,Support Vector Machine,Improved Self-organizing Linear Output Map, |
Publication Year : | 2014 |
Degree: | 碩士 |
Abstract: | 氣候變遷議題日趨重要,研究中常使用大氣環流模式(General Circulation Models, GCMs)配合降尺度方法映射(Project)未來水文變化。降尺度方法主要分為兩種:動力降尺度和統計降尺度,其中被廣泛使用的統計降尺度法時常低估極端值。因此,本研究為了改善模式準確度,提出一個新的統計降尺度方法,並用此法映射台灣未來降雨趨勢。
本方法分成日雨量分類階段與日雨量推估階段。日雨量分類階段先以去勢波動分析(Detrended Fluctuation Analysis, DFA)定義極端雨量。接著建構極端雨量與濕雨量兩個分類模式。其使用日尺度的大氣因子資料,判斷該日的雨量型態為極端雨量、一般雨量或乾雨量(零雨量)。於日雨量推估階段,將日尺度大氣因子資料分別和極端雨量及一般雨量建立極端雨量與一般雨量兩個迴歸模式,以推估日雨量值。最後將極端雨量、一般雨量及乾雨量合成為日雨量序列。本研究引進兩種類神經網路,日雨量分類階段使用支援向量機(Support Vector Machine, SVM);日雨量推估階段使用改良式自組織線性輸出映射圖(Improved Self-organizing Linear Output Map, ISOLO)。模式建立使用到台灣12個雨量測站的日雨量資料與再分析資料(NCEP/NCAR reanalysis data)的大氣因子。 結果發現,本研究提出的基於雨量型態的統計降尺度方法,其推估結果十分接近歷史值,相關性高且誤差降低。於未來雨量映射時,使用IPCC (Intergovernmental Panel on Climate Change)第四次評估報告提出之CGCM3.1模式資料,選擇其A2與B1情境中期(2046-2065)與長期(2081-2100)資料映射各測站雨量變化情形。結果顯示,台灣各區域呈現不同趨勢。石門水庫地區中期與長期雨量皆減少。高山地區中期濕季雨量增加,長期濕季雨量顯著增加。平原地區中期雨量變化不大,長期濕季、乾季雨量皆增加。 總結而言,本研究提出的基於雨量型態的統計降尺度方法表現優異。日雨量分類階段,準確度高且穩定;日雨量推估階段,極端值明顯改善。證實本研究提出的統計降尺度方法解決低估問題,可配合GCM模式的未來情境映射其它區域水文變化。 Statistical downscaling methods are widely used to downscale the outputs of General Circulation Models (GCMs). However, the statistical downscaling methods tend to underestimate, especially the extreme values. To solve this problem, a new statistical downscaling method based on rainfall patterns is proposed in this study. Daily rainfall data from 12 rainfall stations in Taiwan are used. In addition, the large-scale weather factors obtained from NCEP/NCAR reanalysis project and GCMs runs are employed. The proposed method is used to assess the impacts of specific climate change scenarios on future rainfall. The proposed method is composed of two steps. The first step is daily rainfall classification step in which two classification models, extreme rainfall and wet rainfall, are established. These two models are used to identify the rainfall patterns (extreme rainfall, normal rainfall and dry rainfall) of the daily weather factors. The Detrended Fluctuation Analysis (DFA) is used to define the extreme rainfall. The classification models establish the relationship between NCEP/NCAR weather factors and rainfall patterns using Support Vector Machine (SVM). The second step is daily rainfall estimation step. Improved Self-Organizing Linear Output Map (ISOLO) is used to estimate the rainfall amounts for three different rainfall patterns. The changes of downscaled rainfall based on the outputs of 20C3M and the projections of GCMs for A2 and B1 scenarios over the study area are investigated. The results demonstrate that the proposed method provides reliable and accurate rainfall-pattern classification. In addition, the improvement of the estimation of daily rainfall series is significant, especially for the extreme rainfall. In conclusion, the proposed method is practical and suitable for the future projections. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/57634 |
Fulltext Rights: | 有償授權 |
Appears in Collections: | 土木工程學系 |
Files in This Item:
File | Size | Format | |
---|---|---|---|
ntu-103-1.pdf Restricted Access | 2.37 MB | Adobe PDF |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.