Loading…
Reconstructing palaeoclimatic variables from fossil pollen using boosted regression trees: comparison and synthesis with other quantitative reconstruction methods
We test and analyse a new calibration method, boosted regression trees (BRTs) in palaeoclimatic reconstructions based on fossil pollen assemblages. We apply BRTs to multiple Holocene and Lateglacial pollen sequences from northern Europe, and compare their performance with two commonly-used calibrati...
Saved in:
Published in: | Quaternary science reviews 2014-03, Vol.88, p.69-81 |
---|---|
Main Authors: | , , , , , , |
Format: | Article |
Language: | English |
Subjects: | |
Citations: | Items that this one cites Items that cite this one |
Online Access: | Get full text |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | We test and analyse a new calibration method, boosted regression trees (BRTs) in palaeoclimatic reconstructions based on fossil pollen assemblages. We apply BRTs to multiple Holocene and Lateglacial pollen sequences from northern Europe, and compare their performance with two commonly-used calibration methods: weighted averaging regression (WA) and the modern-analogue technique (MAT). Using these calibration methods and fossil pollen data, we present synthetic reconstructions of Holocene summer temperature, winter temperature, and water balance changes in northern Europe. Highly consistent trends are found for summer temperature, with a distinct Holocene thermal maximum at ca 8000–4000 cal. a BP, with a mean Tjja anomaly of ca +0.7 °C at 6 ka compared to 0.5 ka. We were unable to reconstruct reliably winter temperature or water balance, due to the confounding effects of summer temperature and the great between-reconstruction variability. We find BRTs to be a promising tool for quantitative reconstructions from palaeoenvironmental proxy data. BRTs show good performance in cross-validations compared with WA and MAT, can model a variety of taxon response types, find relevant predictors and incorporate interactions between predictors, and show some robustness with non-analogue fossil assemblages.
•Boosted regression trees (BRTs) tested as a palaeoclimatic reconstruction method.•BRTs tested in reconstructions based on northern European fossil pollen sequences.•BRTs compared with two commonly-used reconstruction methods (WA and MAT).•BRTs fare a promising reconstruction method with major theoretical strengths.•Synthesis reconstructions of Holocene palaeoclimate of northern Europe are presented. |
---|---|
ISSN: | 0277-3791 1873-457X |
DOI: | 10.1016/j.quascirev.2014.01.011 |