AbstractWe applied two widely-used methods for data partitioning - constrained incremental sum-of-squares (CONISS) and Optimal Partitioning (OP) along with two supplementary methods, a Kohonen artificial neural network (self-organising map, SOM) and the indicator value (IndVal) index, for the quantitative analysis of subfossil chironomid assemblages from a palaeolake in Central Poland. The samples, taken from 79 core depths, were divided into 5-11 groups (five by SOM, seven by CONISS, 11 by OP), for which different numbers of indicator taxa were determined with the use of the IndVal index (18 for CONISS, 15 for SOM, 11 for OP). Only six indicator taxa were common to all three methods. The number of highly specific (p < 0.001) taxa was highest for SOM. Only the SOM analysis clearly reflected the rate of the changes in chironomid assemblages, which occurred rapidly in the Late Glacial (as a result of greater climate variability) and slowly in the Holocene (as a reflection of slow long-term changes in the local habitat, such as paludification). In summary, we recommend using SOM and the IndVal index in combination with CONISS and/or OP in order to detect different aspects of temporal variability in complex multivariate palaeoecological data.
Subfossil chironomid assemblages in sediments of Meiliang Bay,Lake Taihu and the trophic reconstruction
