A data-mining framework for exploring the multi-relation between fish species and water quality through self-organizing map
Journal
Science of the Total Environment
Journal Volume
579
Pages
474-483
Date Issued
2017
Author(s)
Abstract
The steep slopes of rivers can easily lead to large variations in river water quality during typhoon seasons in Taiwan, which may poses significant impacts on riverine eco-hydrological environments. This study aims to investigate the relationship between fish communities and water quality by using artificial neural networks (ANNs) for comprehending the upstream eco-hydrological system in northern Taiwan. We collected a total of 276 heterogeneous datasets with 8 water quality parameters and 25 fish species from 10 sampling sites. The self-organizing feature map (SOM) was used to cluster, analyze and visualize the heterogeneous datasets. Furthermore, the structuring index (SI) was adopted to determine the relative importance of each input variable of the SOM and identify the indicator factors. The clustering results showed that the SOM could suitably reflect the spatial characteristics of fishery sampling sites. Besides, the patterns of water quality parameters and fish species could be distinguishably (visually) classified into three eco-water quality groups: 1) typical upstream freshwater fishes that depended the most on dissolved oxygen (DO); 2) typical middle-lower reach riverine freshwater fishes that depended the most on total phosphorus (TP) and ammonia nitrogen; and 3) low lands or pond (reservoirs) freshwater fishes that depended the most on water temperature, suspended solids and chemical oxygen demand. According to the results of the SI, the representative indicators of water quality parameters and fish species consisted of DO, TP and Onychostoma barbatulum. This grouping result suggested that the methodology can be used as a guiding reference to comprehensively relate ecology to water quality. Our methods offer a cost-effective alternative to more traditional methods for identifying key water quality factors relating to fish species. In addition, visualizing the constructed topological maps of the SOM could produce detailed inter-relation between water quality and the fish species of stream habitat units. © 2016 Elsevier B.V.
Other Subjects
Biochemical oxygen demand; Chemical oxygen demand; Clustering algorithms; Conformal mapping; Cost effectiveness; Data mining; Dissolved oxygen; Fish; Neural networks; Rivers; Self organizing maps; Water; Water quality; Water resources; Data mining frameworks; Eco-hydrological environments; Fish communities; Flow regimes; Heterogeneous datasets; SelfOrganizing Feature Map (SOM); Spatial characteristics; Water quality parameters; Reservoirs (water); ammonia; dissolved oxygen; nitrogen; phosphorus; phosphorus; artificial neural network; data mining; ecohydrology; finfish; freshwater ecosystem; self organization; water quality; Article; artificial neural network; chemical oxygen demand; cluster analysis; cost effectiveness analysis; data mining; fishery; freshwater fish; habitat selection; nonhuman; Onychostoma barbatulum; priority journal; self organizing map; spatial analysis; suspended particulate matter; Taiwan; water quality; water sampling; water temperature; algorithm; analysis; animal; artificial neural network; biochemical oxygen demand; ecosystem; environmental monitoring; fish; procedures; statistics and numerical data; water pollutant; water pollution; Taiwan; Pisces; Algorithms; Animals; Biological Oxygen Demand Analysis; Data Mining; Ecosystem; Environmental Monitoring; Fishes; Neural Networks (Computer); Phosphorus; Taiwan; Water Pollutants, Chemical; Water Pollution, Chemical
Type
journal article