Multivariate methods and artificial neural networks in the assessment of the response of infaunal assemblages to sediment metal contamination and organic enrichment

Abstract

Changes in the structure of benthic assemblages subject to gradients of sediment metal and organic contamination are usually assessed employing traditional univariate and multivariate analyses. However, artificial neural networks (ANNs) may be able to reveal different effects of pollution and spatiotemporal variations in environmental conditions. A 4-­year annual sediment survey was performed along the Sancti Petri tidal channel (Bay of Cadiz, SW Spain) in order to compare the performance of univariate community descriptors, traditional multivariate techniques and AANs in the assessment of infaunal responses to moderate levels of sediment metal contamination, in organically enriched environments. Despite the potential difficulty to separate natural from anthropogenic stress in the Sancti Petri channel, both traditional multivariate approaches and ANNs revealed spatiotemp oral patterns of environmental and biological variables that allowed suggesting a causal relationship between them, and highlighted subsets of taxa and sediment variables as potential main drivers of those patterns identified. For instance, high values of non-­natural metals and organic content prompted high abundances of opportunists, while high values of natural metals yielded typical tolerant assemblages of organically enriched areas. The SOM ANN, combined with the K-­means clustering algorithm, allowed reaching results identical to ones obtained with the traditional multivariate approach, but needing considerably less analytical and interpretational effort. Although this ANN approach may be a promising tool for the assessment of the ecological quality of estuarine infaunal communities, further work is needed to ensure the accuracy of the method.