Fuzzy neural network model for habitat prediction and HEP for habitat quality estimation focusing on Japanese medaka (Oryzias latipes) in agricultural canals

In the field of agriculture, development of evaluation techniques for environmental changes is urgently required for the purpose of finding a balance between growth in agricultural productivity and environmental considerations. The habitat evaluation procedures (HEP) constitute one technique for habitat assessment. While HEP is widely applied to estimate both habitat quality and quantity in an environment, it appears to be necessary to develop an accurate habitat prediction model in order to evaluate environments precisely. In fact, habitat selection by fish is affected by complicated interaction between multiple environmental factors, which makes it difficult to relate physical environments to habitat preference. In the present study, we utilize artificial neural networks (ANNs), which are commonly applied to model complex systems, to predict the habitat selection of Japanese medaka (Oryzias latipes) in agricultural canals. Considering the essential vagueness of fish behavior, fuzzy membership functions are introduced into the input layer, which advances ANN to fuzzy neural network (FNN). In addition, symmetric triangular fuzzy numbers are employed to account for uncertainty in measurement errors and dispersions of physical environment. The FNN model precisely predicts the habitat preference of Japanese medaka in an agricultural canal, and the results show a good agreement between the calculated and observed habitat suitability indices (HSI). Finally, the habitat quality of two different reaches at the same point in time is compared using HEP, with a view of suitable habitat for Japanese medaka.