研究報告
| 學年 | 101 |
|---|---|
| 學期 | 1 |
| 出版（發表）日期 | 2012-08-01 |
| 作品名稱 | 多型態類神經網路整合異質資料於颱洪時期區域降雨與淹水推估 |
| 作品名稱（其他語言） | Multiple Neural Networks with Heterogenous Data for Regional Rainfall and Inundation Estimation during Typhoon Period |
| 著者 | 張麗秋 |
| 單位 | 淡江大學水資源及環境工程學系 |
| 描述 | 計畫編號：NSC101-2313-B032-002-MY3
 研究期間：201208~201307
 研究經費：1,010,000 |
| 委託單位 | 行政院國家科學委員會 |
| 摘要 | 近年來氣候變遷與土地過度開發利用下，造成日益頻繁與嚴重的洪水已成為全球 的重要議題之一。臺灣被列為同時暴露於颱風、洪水、地震三項以上天然災害之土地 面積與人口比例高居世界第一，極待國人共同努力克服環境災害。在面對氣候變遷、 大洪水時代的來臨，探討颱風對於區域降雨與淹水之影響，以及區域淹水特性分析與 推估，以因應這個洪水發生頻繁的時代，以減少水患災害風險、水資源永續發展之目 的。 本研究區域以宜蘭縣與蘭陽溪流域為主，提出三年計畫目的是從颱風路徑與氣候 資料、二維淹水模擬模式之淹水歷程模擬資料與衛星影像資料等三種異質資料探討颱 風對宜蘭地區淹水程度之影響，並互相提供降雨、淹水相關資料以為模式修正之參考。 理論方面擬採用倒傳遞類神經網路(BPNN)、自組特徵映射網路(SOM)與型態影像學， 可互相結合與發揮各理論模式之特性與所長。第一年為探討颱風路徑特性，以SOM 進 行路徑分類、分析其與降雨量、河川流量之關係，並以BPNN 建立流量或雨量預測模 式；第二年為探討區域淹水之空間特性，擬以SOM 進行區域淹水分布圖之分類，瞭解 區域淹水空間分布之特性，將龐大的淹水模擬資料儲存於SOM 拓樸層中，其神經元作 為BPNN 建置降雨-區域淹水分布預測模式之輸出指標；第三年為探討合成孔徑雷達 (SAR)衛星影像結合BPNN 建構淹水辨識模式對淹水範圍之辨識能力，並擬以型態影像 學演算法修正雜訊，期以提高辨識正確率。 In the last decades, the increasing frequency and severity of floods globally was cause by climate change and land overuse. In Taiwan, the ratios of land and population exposed at relatively high risk from three natural hazards (typhoons, floods and earthquakes) are the highest in the world. When facing climate change and the deluge era, we investigate the impacts of typhoons on both regional rainfall and inundation, and conduct the characteristics analysis and estimation of regional inundation so that respond to the era of the frequent occurrence of flood and achieve the purposes of flood disaster risk reduction and sustainable development of water resources. The three-year proposal will primarily focus on Yilan County and Lanyang River. The purpose of this study is to investigate the impacts of typhoons on the inundation levels of the Yilan County from three kinds of heterogeneous data including typhoon tracking information (tracks, pressure, wind), inundation hydrographs of two-dimensional simulation model and satellite image data; therefore, interactively provide the amount of rainfall and inundation data for cross-calibration of the models. The Back-Propagation Neural Network (BPNN), Self-Organizing Map (SOM) and Morphology are proposed to mutually integrate with each other and provide their features and advantages. The first year research will explore the characteristics of typhoon tracks by using the SOM to analyze the relationship among typhoon tracks, rainfall and streamflow; then, use BPNN to build rainfall/streamflow forecasting models. In the second year, we will investigate the spatial characteristics of regional inundation. The SOM is used to classify the distribution of regional inundation and provide the characteristics of the spatial distribution for regional inundation. The huge amount of inundation simulation data can be stored in the topology layer of the SOM and its neuron can be used as the output of the BPNN forecasting model for the rainfall-regional inundation distribution. In the third year, the BPNN model will be constructed to identify the inundation extents from Synthetic Aperture Radar (SAR) satellite images. Through the morphological operations, we also expect that the noises can be removed and the correct percentages of identification are improved. |
| 關鍵字 | 自組特徵映射圖; 倒傳遞類神經網路; 型態影像學; 颱風路徑; 合成孔徑雷達; 衛星影像; 淹水分布預測; 淹水辨識; self-organizing map; back-propagation neural network; Morphology; typhoon tracks; synthetic aperture radar; satellite image; inundation distribution forecast; inundation identification |
| 語言 | zh_TW |
| 相關連結 |
機構典藏連結 ( http://tkuir.lib.tku.edu.tw:8080/dspace/handle/987654321/101435 ) |