透過您的圖書館登入
IP:3.149.249.68
  • 期刊

應用支撐向量機與輻狀基底神經網路修正QPESUMS雨量

QPESUMS Rainfall Adjustment Using Support Vector Machines and Radial Basis Function Neural Network

摘要


本研究分別應用支撐向量機與輻狀基底函數神經網路,建立QPESUMS(quantitative precipitation estimation and segregation using multiples sensor)估計雨量之空間修正模式,目的為提昇空間估計雨量的準確性,並保有QPESUMS雨量具高解析度及涵蓋區域完整的優點。研究區域為台灣西南部,以七場颱風降雨事件進行模式的率定與驗證。模式建立階段採地面雨量站的觀測降雨為目標輸出值,輸入因子為待修正空間雨量位置的原始QPESUMS雨量及地理空間因子(包含坐標、高程及距雷達站的直線距離),並組合五種不同的輸入向量探討輸入因子對於模式的影響程度。經分析結果與地面雨量站觀測雨量的比較得知,模式的輸入向量包含愈多地理空間因子愈能有效修正QPESUMS雨量;模式驗證結果顯示,支撐向量機與輻狀基底函數神經網路均能有效提昇QPESUMS雨量的準確性,證實本研究發展的空間雨量修正方法具實用性,尤以支撐向量機模式能大幅減少QPESUMS雨量在山區的估計誤差;本研究最後將修正模式應用於研究區域內QPESUMS估計降雨的全面修正,並探討其於空間上的改善成果。

並列摘要


This study develops a spatial rainfall adjustment model using support vector machine and the radial basis function neural network, respectively. This study aims to enhance the accuracy of QPESUMS (quantitative precipitation estimation and segregation using multiples sensor) rainfall estimation, and to preserve the advantage of high spatial resolution and large coverage area of QPESUMS rainfall. Different input vectors to the adjustment model are combinations of the original QPESUMS rainfall estimates and the topographic variables (i.e., the coordinates, the elevation and the distance to the radar station). Analytical results show that including more topographic variables in the input vector can improve the model performance. Validation results pertaining to typhoon events reveal that the adjustment models can reduce QPESUMS rainfall errors. Moreover, the support vector machine outperforms the radial basis function neural network, especially in the mountain regions. Finally, the adjustment model was applied to the full coverage of the study area to demonstrate the spatial rainfall distribution of adjusted rainfall.

延伸閱讀