The simultaneous inference of zooplanktivorous fish and macrophyte density from sub‐fossil cladoceran assemblages: a multivariate regression tree approach

Abstract

Summary1. Quantitative palaeolimnology has traditionally sought to quantify species‐environment relationships to use alterations in biological assemblages to reflect past environmental change. Transfer functions have used regression techniques, such as weighted averaging, to define taxon optima and tolerance for a single chemical or biological variable.2. Cladoceran assemblages and their sub‐fossil remains in shallow lakes are shaped by a combination of interacting factors. Partial constrained ordination of sub‐fossil cladoceran assemblages from 39 shallow lakes (29 in Norfolk, U.K. and 10 in Denmark) indicated that both zooplanktivorous fish (ZF) density and submerged macrophyte abundance significantly influenced community composition. These dual structuring forces precluded the use of a transfer function as one of the key assumptions of this approach was not met, namely that environmental variables apart from the variable being modelled have negligible influence on species distribution or that there is a linear relationship between the two. Separate transfer functions for ZF and macrophyte abundance were developed but had poor performance diagnostics with low bootstrapped r2, high root mean square error of prediction (RMSEP) and large bias.3. To obviate the problem of multiple structuring forces a multivariate regression tree (MRT) was employed, which allows for more than one explanatory variable within a model. The MRT analysis defined six groups with discrete ranges of ZF and macrophyte densities. The technique identified critical values or ‘break points’ in ZF and macrophyte abundances which result in significant alterations in the sub‐fossil cladoceran assemblage. In addition, the MRT groups had different summer mean values for chlorophyll‐a, Secchi depth, total phosphorus and nitrate‐nitrogen.4. The predictive abilities of the model were assessed by comparing the observed versus predicted MRT group membership. In general group membership was reliably predicted, suggesting sub‐fossil cladoceran assemblages reliably reflect ZF and macrophyte density in shallow lakes. For a relatively small number of sites there were differences between the observed and predicted MRT group membership. These failures of prediction may result, at least in part, from the disparity of the time period represented by the environmental data and the surface sediment cladoceran assemblage.