Evaluating the accuracy and precision of diatom-based transfer functions in Holocene sea-level studies: An example from Ho Bugt, western Denmark

  1. Lookup NU author(s)
  2. Dr Steve Juggins
Author(s)Szkornik K, Juggins S, Gehrels WR, Charman DJ
Publication type Article
JournalQuaternary Science Reviews
Year2012
Volume
Issue
Pages
ISSN (print)0277-3791
ISSN (electronic)1873-457X
Full text for this publication is not currently held within this repository. Alternative links are provided below where available.
Diatom-based transfer functions have become a widely accepted tool in the reconstruction of Holocene sea-level change from salt-marsh sequences. However, such approaches may be over-optimistic in terms of their estimates of uncertainty. This paper evaluates the accuracy and precision of a diatom-based transfer function as a tool for the reconstruction of Holocene sea-level change in two salt-marsh sequences from the Ho Bugt embayment in western Denmark, based on a modern diatom training set from the same location. The relationship between modern diatom assemblages and environmental variables in the Ho Bugt embayment was explored using Canonical Correspondence Analysis (CCA). Initial CCA results suggest that elevation is the most important variable in terms of explaining the variation in the modern diatom data (13.9% variance explained, P-value = 0.0020), however, varience partitioning (partial CCAs) demonstrates that the unique effect of elevation on modern diatom assemblages is much smaller (5.9% variation explained) and the counfounding effect of multiple environmental gradients on modern diatom distributions in the Ho Bugt embayment is quite large (7.4% variation explained). Initial transfer function results for the Ho Bugt training set (100 samples, 152 taxa) are comparable with a number of other studies from temperate salt-marsh enevironments (WA-PLS (2 component) RMSEPjack = 0.117 m, r2jack = 0.956), max biasjack = 0.182), however, when we employ a cross-validation approach that is more robust to spatial autocorrelation and pseudoreplication than traditional leave-one-out error estimates, the performance of our WA-PLS transfer functions decreases considerably (RMSEPjack = 0.234 m, r2jack = 0.794), max biasjack = -0.418). Evaluation of the resulting palaeomarsh-surface reconstructions in the two salt-marsh sequences analysed suggests that the overall trends predicted by the transfer function can be replicated. However, a lack of modern analogues, under-representation of taxa in the modern training set and the issue of multiple analogues have all contributed to reducing the reliability of the reconstructions and may account for the discrepencies between the reconstructed record and the instrumental tide gauge data. We conclude that future studies need to employ different cross-validation approaches, that are more robust to autocorrelation and pseudoreplication, and should seek to replicate trends from multiple cores and multiple sites, in addition to making comparisions with the instrumental tide gauge record, to ensure that regional-scale trends are being captured.
PublisherPergamon
Actions    Link to this publication