Please use this identifier to cite or link to this item:
|Title:||Artificial neural networks as an indicator search engine: The visualization of natural and man-caused taxa variability|
|Abstract:||© 2015 Elsevier Ltd. One of the main challenges in selecting suitable biological indicators of environmental degradation is to recognize the stressor-specific response signal and to separate it from the natural background variability, which can be accomplished by setting an appropriate statistical design, with an output that enables understanding of the recorded indicator signal. In this study we used artificial neural networks (self organizing map (SOM) and geo-self-organizing map (Geo-SOM)) to model and visualize the variability in the chironomid community of the Danube basin, as a model for large non-wadeable rivers. Geo-SOM analysis visualized the longitudinal distribution of significant parameters defining different spatial-distributional types of anthropogenic disturbance. Chironomidae larvae, sampled in both shallow (river bank) and deep (middle) parts of the river, emphasized hydromorphological degradation and zinc as the most important stressing factors, with chlorophyll-a and suspended solids as accompanying variables influencing the community structure. Substrate specificity was shown to be a relevant factor influencing the variability within chironomid community structure bound to natural causes. Geo-SOM analysis also visualized the longitudinal distribution of chironomid taxa, following the distribution patterns of significant disturbance factors. The Kruskal-Wallis test validated 25 potential indicators for the shore area and 11 for the deep water area, which significantly changed their frequencies and abundances between classes with different extents of degradation. Due to its high taxonomical and ecological diversity, the Chironomidae family is a significant source of potential stress-specific indicators, which should be recognized and included in the future in relevant bioassessment methods. The artificial neural network could be a powerful tool for selecting reliable indicators to explain the variability found in the ecosystem and enable it to be specified and patterned together with environmental degradation.|
|Appears in Collections:||Faculty of Science, Kragujevac|
Files in This Item:
|29.86 kB||Adobe PDF|
Items in SCIDAR are protected by copyright, with all rights reserved, unless otherwise indicated.