A self-organizing radial basis network for estimating riverine fish diversity

In aquatic ecosystems, particularly rivers, hydrology plays a key role in structuring and maintaining habitats and flow regimes that influence ecological sustainability. Flow regime assessment in Taiwan has been facilitated recently by the Taiwan Eco-hydrologic Indicator System (TEIS). In this study, the self-organizing feature map (SOM) and radial basis function (RBF) neural network are combined to produce a self-organizing radial basis network (SORBN) that takes the advantages of both methods for strengthening the power of presentation and reliability of estimation. The SORBN is proposed to estimate the diversity of fish communities based on the TEIS and historic fish community composition at 36 locations in Taiwan. The discharge data are available for a minimum of 20. years. Data analysis applying a moving average method to the TEIS statistics is used to reflect the effects of antecedent flow conditions on fish diversity. Results indicate the hybrid SORBN not only effectively categorizes stream flow data but also reasonably identifies relationships between flow regime and fish community diversity. Results are encouraging so that it is possible to better relate flow and ecosystem conditions, and the proposed method can be used to quantify how flow influences river ecosystems.