Robustness of fish assemblages derived from three hierarchical agglomerative clustering algorithms performed on Icelandic groundfish survey data

Heatmaps are used to identify species-area assemblages based on Icelandic groundfish survey data. Hierarchical agglomerative clustering algorithms are widely applied for species assemblage studies and form the basis for heatmaps. First, the robustness of fish assemblages derived from three clustering algorithms, Average, Complete, and Ward's linkage, was examined. For statistical reliability, the use of a bootstrap resampling technique to generate the confidence values for the clusters is emphasized. Two cluster validity indices were used to measure the efficiency and the quality of the clusters. To examine the stability of the results, clustering was carried out across different sample sizes and levels of data smoothing. Second, cluster analysis was carried out using a different combination of data standardization and dissimilarity measure. Ward's linkage gave the most robust fish assemblages for both modes of data analyses. Four fish assemblages were identified which could be characterized according to the depth and the geographic distribution. This algorithm was then used to generate a heatmap to determine the species-area relationships. Specific areas were characterized by the identified species groups.