Use of artificial neural networks for habitat unit composition modeling

Understanding the fish assemblage composition and assessing habitat preferences is an important requirement for ecological engineering practices for river restoration. This study collects fish and habitat data in the Tsengwen River Basin and Kaoping River Basin to provide criteria for fish habitat restoration projects. Fish samples were collected used prepositioned areal electrofishing devices designed for repeated grid sampling. The electrofishing data were used to identify habitat preferences and develop fish community-habitat models. A fuzzy rule-based model was used to characterize habitat units as deep pool, deep riffle, shallow riffle, and shallow pool based on measured depth and current velocity. The artificial neural networks then is used to model the connection between the autecology matrix analysis values and surveyed habitat unit compositions. The utility of the model was furthered using historical fisheries data in addition to the data collected as part of the study. The fish community-habitat model results provide a reference condition that can be used to guide stream restoration and ecological engineering decisions aimed at maintaining the natural ecological integrity and diversity of Taiwanese rivers.