Artificial neural network analysis on the gamma diversity of tidal flat macrobenthic communities in Korea

초록

Due to severe loss of tidal wetlands, habitat restoration became an important way of compensation for coastal developments in the west and southern coasts of Korea. For the more viable restoration and evaluation of target habitats, developing a designing tool that predicts faunal diversity was required. A diverse type of tidal flats (33 sites) located in the west and southern coasts of Korea were surveyed for macrofaunal samples and abiotic factors (e.g., tidal flat width, distribution of substratum properties, tidal range, yearly averages of air temperature and salinity). Gamma diversity, a total number of species at a site was estimated by the vertical transect sampling design. Feed-forward artificial neural network was performed with back-propagation algorithm. The number of neurons in a hidden layer was selected by the performance of the model (i.e., correlation between prediction and observation values and MSEREG). The resultant model was analyzed with neural net interpretation diagram. The most prominent feature was the importance of heterogeneity of substratum properties in a transect. Positive relationships were observed in tidal flat width, mean air temperature, salinity, tidal ranges and sand contents in a transect. The model that we have developed is applicable to design and evaluate the restoration of wetland habitats in the coastal areas of Korea.