應用類神經網路推估溪流之生物多樣性

Abstract

隨著人口成長、土地利用需求快速增加，人類擷取自然環境資源已遠超過自然界自行修復的速率，導致生態系結構逐漸惡化、生態物種歧異度降低，進而減少環境原本能提供的物質與服務。近年來由於生態環境復育意識的提昇，人類在追求物質的滿足之餘，慢慢開始重視人類與生態環境的共存關係，探討人類行為對生態環境所造成的嚴重衝擊。 河川流量管理即為一兼顧人類使用需求及河川生態系統需求之理念，將生態觀念融入河川流量經營管理之中，以達到人類與河川生態系統共存的理想，然而基於對大自然有限的了解，往往無法得知河川生態系統的實際需求，也無法明確地以數值或方程式的方式表示，因此本研究建立將自組特徵映射網路(Self-Organizing Feature Map， SOM)與輻狀基底函數類神經(Radial Basis Function Neural Networks，RBFNN)網結合之自組特徵輻狀基底類神經網路(SORBNN)架構，透過此模式利用台灣生態水文指標系統(Taiwan Ecohydrology Index System，TEIS)中的指標推估溪流生態的生物多樣性，其中以魚類科別的歧異度指數代表生物多樣性(Bio-diversity)。本研究選取全台灣河川中未受人為控制、流量資料長度大於二十年且其上下游10公里內有魚類調查資料之流量站作為研究測站，結果顯示模式除了具有對流量資料進行型態判別分類的能力外，也能快速、有效率且準確推估生物多樣性。

As the population and demand for land use rapidly increased, the use of environmental resources has exceeded the rate of naturalization that might result in the degeneracy of ecological structure and the decrease of the diversities of species which could reduce the resources provided by environment. Due to the raise of eco-environmental restoration concept in the past several years, people gradually pay attention to the coexistence relationship between human being and eco-environment and the impacts of human activities on eco-environment. Stream flow management is the idea that combines the concept of ecology and provides the demand for both human and river ecosystem. Base on the limited understanding of nature, it’s hard to get acquainted with actual demands of river ecosystem and represent it by numerical methods or formula. Therefore, this study combines Self-Organizing Feature Map (SOM) and Radial Basis Function Neural Networks (RBFNN) into Self-Organizing Radial Basis Neural Networks (SORBNN). By this model, it can be estimated “river bio-diversity” by using the index of Taiwan Ecohydrology Index System and the diversities of fish families are to be the index of bio-diversity. In this research, the stream flow data which are only collected with records more than 20 years and without anthropogenic control would be tested. The result shows that this model not only can categorize the stream flow data but also can estimate the bio-diversity quickly, efficiently and precisely.