透過您的圖書館登入
IP:18.217.203.172
  • 學位論文

應用類神經網路探討衛星影像對集水區降雨量推估之影響

Watershed Rainfall Estimation from Satellite Imagery Using Neural Networks

指導教授 : 張麗秋

摘要


本研究主要目的是使用類神經網路探討衛星影像資訊對於降雨量推估的影響,因為衛星影像的資料維度相當大且非線性,導致從衛星影像中獲得可用的資訊是相當困難的,因此使用類神經網路中對於影像辨識成果極佳的自組特徵映射網路(SOM)進行衛星影像的處理。 本研究以2000∼2004年及2006年蒐集之25場颱風為例,架構石門水庫上游集水區颱風時期累積六小時之降雨預報模式。首先藉由隨機分類將颱風事件分成三種不同的組合,分別為方案一至方案三,並使用氣象衛星的紅外線頻道資料與地面雨量站資料當作輸入資料,經由建立下列六種預報模式:多變量線性迴歸模式(MLR)、倒傳遞類神經網路(BP)、自組特徵映射網路結合倒傳遞類神經網路(SOMBP)、自組特徵映射網路結合多變量線性迴歸(SOMMLR)、SOMBP再結合倒傳遞類神經網路(SOMBP+BP)以及SOMMLR再結合倒傳遞類神經網路(SOMMLR+BP),預報石門水庫上游集水區颱風時期之六小時累積雨量,以探討氣象衛星的資訊對於累積降雨預報之成效。 預報結果顯示由於六小時累積雨量資料具連續性,因此MLR模式預報結果相當不錯;而從衛星雲圖在SOM的拓樸圖上可發現SOM網路具有分辨衛星雲圖特徵的能力,同時使用地面雨量站資訊與衛星資料,有效地改善SOMBP及SOMMLR的模式預報結果,顯示衛星影像對於降雨預報上有重要的影響性,可以有效改善模式預報値。

並列摘要


The main purpose of this study is to explore the influence of satellite imagery information on rainfall estimation using artificial neural networks. However, it is often difficult to extract interpretable information from satellite images, as data dimensions are large and nonlinear. We proposed the self-organizing map (SOM), one of artificial neural network adept at pattern cognition. In this study, watershed rainfall estimation models are constructed to forecast the rainfall summation of future six hours during typhoon events. The models are based on SOM or linear regression to investigate the characteristics of satellite imagery information and its influence on rainfall estimation. The available data are hourly rainfall data of sixteen rainfall gauge stations in the Shihmen watershed from 25 typhoon events and GMS-5/MTSAT remotely sensed data are collected from 2000 to 2004 and 2006. In order to investigate the characteristics and compare the performance among the different models, we designed three cases with different sizes or amount of rainfall in training data, then constructed six different models, multivariate linear regression model (MLR), back-propagation neural network (BP), self-organizing map linking with BP (SOMBP), self-organizing map linking with linear regression (SOMMLR), SOMBP linking with BP (SOMBP+BP) and SOMMLR linking with BP linear regression (SOMMLR+BP), to estimate the future six-hour rainfall summation. The input variables have two types: the past three six-hour rainfall summations and satellite images. The results show that (1) the MLR models have nice performances when the input variable only include the past rainfall summations, (2) SOM indeed has the ability to extract patterns from satellite data, (3) SOMBP and SOMMLR can get better results when the input variables are the past rainfall summations and satellite images. The satellite imagery information is indeed helpful to improve the accurate of rainfall estimation.

參考文獻


21. 蔡孟蓉,2006,「類神經網路結合衛星影像預報颱風降雨量」,台灣大學生物環境系統工程學研究所碩士論文。
18. 葉天降、謝信良、吳石吉,2000,「簡單統計方法與台灣地區颱風降雨預測之研究(二)預測結果隨區域之分布」,大氣科學,28,263-279。
17. 葉天降、吳石吉、謝信良,1999,「簡單統計方法與台灣地區颱風降雨預測之研究(一)預測方法與台北颱風降雨之預測校驗」,大氣科學,27,395-412。
24. Cho, S. B., 1997. Neural-network classifiers for recognizing totally unconstrained handwritten numerals. IEEE Transactions on Neural Networks. 8(1):43-53
25. Chen, K. S., Wang, J. T., and Mitnik, L.M., 2001. Satellite and ground observations of the evolution of typhoon herb near taiwan, Remote Sensing of Environment 75:397-411

被引用紀錄


陳淵翔(2013)。雷達定量降水資料結合類神經網路於颱風時期降雨量與流量推估之研究〔碩士論文,淡江大學〕。華藝線上圖書館。https://doi.org/10.6846/TKU.2013.00180
許惠茵(2010)。類神經網路結合衛星影像與氣象資料於颱風雨量推估之研究〔碩士論文,淡江大學〕。華藝線上圖書館。https://doi.org/10.6846/TKU.2010.00737

延伸閱讀