Machine-learning based reconstructions of primary and secondary climate variables from North American and European fossil pollen data

Abstract

We test several quantitative algorithms as palaeoclimate reconstruction tools for North American and European fossil pollen data, using both classical methods and newer machine-learning approaches based on regression tree ensembles and artificial neural networks. We focus on the reconstruction of secondary climate variables (here, January temperature and annual water balance), as their comparatively small ecological influence compared to the primary variable (July temperature) presents special challenges to palaeo-reconstructions. We test the pollen–climate models using a novel and comprehensive cross-validation approach, running a series of h-block cross-validations using h values of 100–1500 km. Our study illustrates major benefits of this variable h-block cross-validation scheme, as the effect of spatial autocorrelation is minimized, while the cross-validations with increasing h values can reveal instabilities in the calibration model and approximate challenges faced in palaeo-reconstructions with poor modern analogues. We achieve well-performing calibration models for both primary and secondary climate variables, with boosted regression trees providing the overall most robust performance, while the palaeoclimate reconstructions from fossil datasets show major independent features for the primary and secondary variables. Our results suggest that with careful variable selection and consideration of ecological processes, robust reconstruction of both primary and secondary climate variables is possible.

Early Holocene Thermal Maximum recorded by branched tetraethers and pollen in Western Europe (Massif Central, France)

“…This cooling seems to have been amplified by the MAT method which has a different sensitivity compared to the WAPLS (see Section 4.1.2). More methods based on pollen data such as Inverse Modelling (Guiot et al, 2000) or Boosted regression trees (Salonen et al, 2019) can be tested to improve the accuracy of the climate reconstruction for this time period.…”

Prolonged interglacial warmth during the Last Glacial in northern Europe

“…The climate reconstruction uses a modern calibration data set consisting of 807 European lakes, derived from the European Modern Pollen Database (Davis et al 2013), with modern climate data extracted for each sample location. For further details about this calibration data set, see Salonen et al (2019). We fitted pollen-T jul calibration models to the calibration data using six generally well-performing (in modern cross-validation experiments; Salonen et al 2018Salonen et al , 2019 statistical approaches (weighted averaging (WA); weighted averaging-partial least squares (WA-PLS); maximum likelihood response surfaces (MLRC); modern analogue technique (MAT); random forest (RF); boosted regression trees (BRT)).…”

“…For further details about this calibration data set, see Salonen et al (2019). We fitted pollen-T jul calibration models to the calibration data using six generally well-performing (in modern cross-validation experiments; Salonen et al 2018Salonen et al , 2019 statistical approaches (weighted averaging (WA); weighted averaging-partial least squares (WA-PLS); maximum likelihood response surfaces (MLRC); modern analogue technique (MAT); random forest (RF); boosted regression trees (BRT)). The median of the sixmethod ensemble was calculated to summarize the individual reconstructions.…”

“…For further details about this calibration data set, see Salonen et al . (2019). We fitted pollen‐ T jul calibration models to the calibration data using six generally well‐performing (in modern cross‐validation experiments; Salonen et al .…”

Peatland Development, Vegetation History, Climate Change and Human Activity in the Valdai Uplands (Central European Russia) during the Holocene: A Multi-Proxy Palaeoecological Study

“…Pollen of aquatic plants and spores were also calculated in relation to this total. The pollen data were used to reconstruct the mean July temperature based on the pollen-climate calibration set derived from the European Modern Pollen Dataset [54] using the transfer functions developed by Salonen et al [55]. A weighted averaging-partial least-squares model (WAPLS: 3-components, RMSEP = 1.73, Maximum bias = 5.90) was applied to square-root transformed species data.…”