Identification of hydrologic indicators related to fish diversity and abundance: A data mining approach for fish community analysis

[1] This paper develops a new approach to identify hydrologic indicators related to fish community and generate a quantitative function between an ecological target index and the identified hydrologic indicators. The approach is based on genetic programming ( GP), a data mining method. Using the Shannon Index ( a fish community diversity index) or the number of individuals ( total abundance) of a fish community, as an ecological target, the GP identified the most ecologically relevant hydrologic indicators ( ERHIs) from 32 indicators of hydrologic alteration, for the case study site, the upper Illinois River. Robustness analysis showed that different GP runs found a similar set of ERHIs; each of the identified ERHI from different GP runs had a consistent relationship with the target index. By comparing the GP results with those from principal component analysis and autecology matrix, the three approaches identified a small number ( six) of common ERHIs. Particularly, the timing of low flow ( D-min) seems to be more relevant to the diversity of the fish community, while the magnitude of the low flow ( Q(b)) is more relevant to the total fish abundance; large rising rates result in a significant improvement of fish diversity, which is counterintuitive and against previous findings. The quantitative function developed by GP was further used to construct an indicator impact matrix ( IIM), which was demonstrated as a potentially useful tool for streamflow restoration design.