應用適合度曲線與遺傳規劃法於河川魚類棲地模擬－分類流況法

Abstract

河川生態工程為近年來治理河川的基礎上，其主要尋求接近自然的工程方法，因此若能於工程施作前，建立出良好的模擬模式，不但可提供河川生態工程方向，也能增加河川治理之效益。而河川棲地的模擬中，棲地適合度分析為重要的步驟之一，棲地適合度指數是建立目標物種與棲地環境因子之間的關係，而水理棲地模式可模擬河川斷面及計算魚類權重可用面積，兩者相輔而成為重要的河川生態分析工具。 前人於河川棲地模擬之研究，多以魚類之棲地喜好性對於棲地環境因子的相互關係建立單一模式，然而為了考量各種流況型態提供魚類不同的生態需求，例如：高溶氧的淺瀨為食物密集的區域、平靜的深潭適合當作避難所，因此需以不同標準的模擬模式來描述其行為，才能更加符合現況。本研究在新北市淡水區的大屯溪流域，進行生態的調查，並以日本禿頭鯊當作目標物種，針對流速、水深對應之魚類出現機率，來建立棲地適合度指數。其方法有三種，第一，將每個棲地環境因子分別建立出適合度值，再透過相乘而得到混合的棲地適合度指數，稱作傳統模式；第二，以遺傳規劃法的優化機制，找尋棲地環境因子之間的最佳方程式，進而求得棲地適合度指數，稱作改良模式；第三，將流速、水深以0.32m/s、0.29m為界線，劃分成四種流況，再透過遺傳規劃法搜尋最佳方程式，得到一聯合棲地適合度指數，稱作分類模式。最終再經由二維水理棲地模式River 2D模擬出流況與魚類權重可用面積之空間分佈，進一步比較三種模式的結果。 研究結果顯示，日本禿頭鯊出現頻度與流況發生頻度之相關性高達0.96，因此過去普遍使用的第二型適合度指數，並無法反映出魚類真實喜好的環境。此外，在模式的率定、驗證方面，改良模式也以均方根誤差0.0718、0.1001，較傳統模式的0.1215、0.1289還來得好，說明遺傳規劃法在考慮變數之間的關係後，確實能有較佳的預測結果，另一方面，分類模式則是以0.1127、0.1316不如改良模式，所以改良模式在採用整體資料下，其可信度及準確度還是高於分類模式。最後，在魚類權重可用面積計算之結果，發現分類模式可以避免其他兩種模式，在棲地面積空間分佈中，有低估或是均質化的現象，也期望此模式後續在探討不同魚類行為時(如：產卵、覓食)，能有更實用的價值。

River ecological engineering is the engineering method to renovate river approaching to nature in recent years. Establishing good simulation model before executing not only provides a direction for river ecological engineering, but improves the benefits of river management. During simulating river habitat, Habitat Suitability Analysis is one of the most important processes. Habitat suitability index (HSI) builds the relationship between target species and environmental factors of habitat and physical habitat model simulate the river section and calculate weighted usable area (WUA). Combining both of them become a crucial analysis tool to river ecosystem. The previous study in river habitat simulation mostly aims at the fish preference of environmental factors of habitat to build individual model. However, in order to consider different fish ecological demand in various flow conditions, for example, riffle with high oxygen is full of food sources, pool is suitable to be a shelter, it needs diverse standard simulation model for describing fish activities to approach reality. The study area is Datuan Stream located in Tamsui District, New Taipei City and the target species is monk goby (Sicyopterus japonicus). Fish presence probabilities for each velocity and water depth establish HSI. There are three methods: First, establish suitability index (SI) by each factor separately, and then multiple all SIs together to obtain a composite HSI, which called “traditional model”. Second, Search for optimal function in factors by genetic programming (GP), and obtain HSI, which called “modified model”. Third, divide into four flow conditions by velocity 0.32 (m/s) and water depth 0.29(m), and obtain united HSI, which called “classified model”. Finally, simulate river flow and calculate the spatial distribution of WUA, and then compare the result of three models. The result reveals that the correlation between frequency of monk goby presence and frequency of flow condition is up to 0.96. Therefore, Category II HSI which is the most common method can not reflect favorite environment of fish in reality. In addition, when it comes to the calibration and validation of model, the root mean square error (RMSE) of modified model is better than traditional model by 0.0718, 0.1001, and 0.1215, 0.1289. While taking the relationship between variables into consideration by GP, it has a better predictive effect. On the other hand, the RMSE of classified model is worse than modified model by 0.1127, 0.1316. All in all, the confidence and accuracy of modified model is greater than other two models. In the end, the result of calculating WUA shows that classified model could avoid underestimation or homogeneity, which may occur in other two models. While researching in different activities of fish (ex: spawning, preying), we expect classified model to be practical and valuable in the future.