The performance of three ordination methods applied to demersal fish data sets: stability and interpretability

The performance of robust principal component analysis (RPCA), detrended correspondence analysis (DCA) and non-metric multidimensional scaling (NMDS) with two demersal fish data sets were assessed in terms of their stability to bootstrap-generated sample variation and the methods ability to reflect a well-known depth gradient. Stability was assessed for both species and site orderings and configurations, using scaled rank variance (SRV) and Spearman (?) and Procrustes correlations (t0). The NMDS site and species orderings showed the highest stability. DCA performed best in terms of site ordination stability, but NMDS performed best in terms of species ordination stability. In terms of matching the expected ecological variation, NMDS was the most sensitive method for the wider-depth gradient data and DCA was the most sensitive for the narrower-depth gradient data. According to the sensitivity and informative power criteria associated with the stability assessment, t0 was the best methodological approach for site ordinations, and SRV for species ordinations.