Climate reconstruction for the Late Pliocene and Early Pleistocene - with quantitative paleo-climate methods for the Netherlands

Abstract

This guided research aims to reconstruct the climate during the Late Pliocene / Early Pleistocene for the Netherlands based on pollen assemblage data with quantitative paleo-climatic reconstruction methods. The Late Pliocene and Early Pleistocene are important Epochs because of their characteristics by cyclic growth and decay of terrestrial ice sheets related to the Milankovitch cycles. The Late Pliocene through Early Pleistocene is a frequently targeted interval for paleo-environmental reconstructions because it is considered as an analogue for future climate change. The focus of this guided research is on the Netherlands. Three different terrestrial sections are used from different parts of the Netherlands: Noordwijk, Petten and Hank. The Noordwijk section is from the south-western part of the Netherlands and has a time span from 2.61 – 2.11 Ma years ago, Early Pleistocene. The Petten section has an age range of 1.35 – 2.41 Ma years ago, Early Pleistocene and the Hank section originates from the Late Pliocene (5.45 – 2.48 Ma years ago). The modern pollen-based quantitative paleo-climatology has undergone the development of a diverse array of statistical techniques to transform fossil assemblage data into past climate estimates. Three main approaches for quantitative reconstruction of past climates from bio-stratigraphical data are known: indicator species approach, assemblage approach and multi-variate transfer function approach. The multi-variate calibration function approach involves underlying statistical models with global estimation of parametric functions for all the taxa present. The analysing program C2 is used to reconstruct the winter temperature and total precipitation. Pollen-climate transfer functions were calculated using WA with inverse deshrinking and 3- component WA-PLS. Calibration set species data values were square-root transformed for WA and WA-PLS regression to reduce noise in the data. Performance statistics were computed for each transfer function, these include the coefficient of determination (R2), Root Mean Square Error of Predication (RMSEP) and maximum bias. Both reconstructions are compared with the percentage deciduous trees in the samples and known data from Dearing Crampton-Flood et al., (2020) for the Late Pliocene and Zagwijn (1992) for Early Pleistocene. From these results, WA seems to fit the best in Late Pliocene data and WA-PLS fits best in the younger data. WA seems to fit better in the Late Pliocene and WA-PLS fits better for the Early Pleistocene. The problem for WA-PLS is the that it tries to fit all the variation in the reconstruction, for older data this is not the right way because not all pollen taxa do exist today in the study area. Some mistakes inevitable if WA-PLS is used for older data than the Pleistocene. Further research requires more datapoints to make a better connection with other reconstructions.