海灘地形受到海氣象與沙源等因素影響經常在改變，要預測其變化一直不是簡單的問題。海洋大學蕭松山教授的研究團隊從2004年至今一直進行北海岸鹽寮海灘監測，他們的結果似乎顯示鹽寮沙灘係一動態平衡海岸，呈現季節性變化之趨勢，並與颱風侵襲有緊密關係。此海灘地形的演變與那些自然因子有關連甚至是否可以預測其演變乃是重要課題。目前雖可用數值模式模擬甚至預測，然而海灘地形變化的數值模擬仍有其困難度，故本研究希望以資料探勘的方法，針對已測量及將來繼續會測量之鹽寮沙灘北中南三個斷面變化的相關資料，尋找出與斷面變化的強關聯因子，藉由強關聯分析能分別精確的掌握每一個斷面與侵淤量變化，進而建立一鹽寮沙灘變遷的預測模式。本研究為三年計畫，第一年將蒐集鹽寮北中南三個斷面海灘侵淤量變化的相關資訊，並分析尋找出與每個斷面海灘變化的強關聯項，藉由強關聯的分析定義出探勘模式的輸入項或其轉換函數。由於根據過去研究發現鹽寮地區沙灘變化與颱風襲擊有緊密關係，故本研究計劃於第二年將針對颱風時期沙灘變化，建置一颱風時期鹽寮沙灘變化的預測模式。第三年則計畫將所得到的強關聯項，經適當的轉換後做為資料探勘的主要輸入依據，並使有探勘技術建置任一時期鹽寮海灘變遷的預測模式。我們已嘗試將資料探勘演算法運用於師範大學林雪美教授提供的福隆海灘面積資料，初步結果顯示以決策樹演算法所建置的模式有不錯的效果。這顯示本計畫所提的方法應該可行。The topography or morphology of a beach is influenced by meteorological and oceanic forces as well as the supply of sand. Thus, beaches constantly change their shapes. It is still difficult to predict the evolution of beaches. The research team lead by Professor Song Shan Hsiao has done beach monitoring at the Yenliao beach on the northern coast since 2004. They found that the beach is at a dynamic equilibrium. Their data seem to suggest that the changes in the beach profiles and beach width were due to seasonal effects as well as the actions of typhoons. It is important to know how the beach changes with what external parameters and how we can predict its change. Although numerical modeling can be used to simulate changes in the beach topography, there are still some difficulties for the numeral methods to work right. One way to answer these questions is to use data mining techniques, and this is the objective of this proposed study. We will collect data for the profiles of the Yenliao beach, which Professor Hsiao’s team has obtained since 2004 and will continue to measure, and other relevant information such as meteorological data, off-shore wave data, and Shunshi river flow and sediment load data to determine which parameters have strong correlations with changes of beach profile and erosion or accretion. Then we will attempt to find a predictive model for the beach evolution. This is a three-year project. In the first year, relevant data will be collected and dominant factors affecting the beach profile changes will be identified by association rules. These will be the input factors for our predictive model. In the second year, we will study how the various typhoon parameters (tracks, center pressure, etc.) effect the changes in the beach, so that a predictive model for the typhoon effects on the beach can be established. In the third year we will use factors found by association rules with appropriate conversions as the main input factors to build a predictive model for the beach evolution based on data mining techniques. We have tested our data mining technique on building a model for the variation of Fulong beach area. Our initial results showed that using decision tree we can obtain some satisfactory results. This means that the method proposed in this study is a viable one for building predictive models for changes in beach profile or topography.
artificial neural network