Description: The composition of planktonic foraminiferal (PF) calcite is routinely used to reconstruct climate variability. However, PF ecology leaves a large imprint on the proxy signal: seasonal and vertical habitats of PF species vary spatially, causing variable offsets from annual mean surface conditions recorded by sedimentary assemblages. PF seasonality changes with temperature in a way that minimises the environmental change that individual species experience and it is not unlikely that changes in depth habitat also result from such habitat tracking. While this behaviour could lead to an underestimation of spatial or temporal trends as well as of variability in proxy records, most palaeoceanographic studies are (implicitly) based on the assumption of a constant habitat. Up to now, the effect of habitat tracking on foraminifera proxy records has not yet been formally quantified on a global scale. Here we attempt to characterise this effect on the amplitude of environmental change recorded in sedimentary PF using core top δ18O data from six species. We find that the offset from mean annual near-surface δ18O values varies with temperature, with PF δ18O indicating warmer than mean conditions in colder waters (on average by −0.1 ‰ (equivalent to 0.4 °C) per °C), thus providing a first-order quantification of the degree of underestimation due to habitat tracking. We use an empirical model to estimate the contribution of seasonality to the observed difference between PF and annual mean δ18O and use the residual Δδ18O to assess trends in calcification depth. Our analysis indicates that given an observation-based model parametrisation calcification depth increases with temperature in all species and sensitivity analysis suggests that a temperature-related seasonal habitat adjustment is essential to explain the observed isotope signal. Habitat tracking can thus lead to a significant reduction in the amplitude of recorded environmental change. However, we show that this behaviour is predictable. This allows accounting for habitat tracking, enabling more meaningful reconstructions and improved data–model comparison.

Description: The species composition of many groups of marine plankton appears well predicted by sea surface temperature (SST). Consequently, fossil plankton assemblages have been widely used to reconstruct past SST. Most applications of this approach make use of the highest possible taxonomic resolution. However, not all species are sensitive to temperature, and their distribution may be governed by other parameters. There are thus reasons to question the merit of including information about all species, both for transfer function performance and for its effect on reconstructions. Here we investigate the effect of species selection on planktonic foraminifera transfer functions. We assess species importance for transfer function models using a random forest technique and evaluate the performance of models with an increasing number of species. Irrespective of using models that use the entire training set (weighted averaging) or models that use only a subset of the training set (modern analogue technique), we find that the majority of foraminifera species does not carry useful information for temperature reconstruction. Less than one-third of the species in the training set is required to provide a temperature estimate with a prediction error comparable to a transfer function that uses all species in the training set. However, species selection matters for paleotemperature estimates. We find that transfer function models with a different number of species but with the same error may yield different reconstructions of sea surface temperature when applied to the same fossil assemblages. This ambiguity in the reconstructions implies that fossil assemblage change reflects a combination of temperature and other environmental factors. The contribution of the additional factors is site and time specific, indicating ecological and geological complexity in the formation of the sedimentary assemblages. The possibility of obtaining multiple different reconstructions from a single sediment record presents a previously unrecognized source of uncertainty for sea surface temperature estimates based on planktonic foraminifera assemblages. This uncertainty can be evaluated by determining the sensitivity of the reconstructions to species pruning.