Comparison of boosted regression trees vs WA-PLS regression on diatom-inferred glacial-interglacial climate reconstruction in Lake Tiancai (southwest China)

Abstract

Quantitative paleoclimate reconstructions based on biological fossils over glacial-interglacial timescales are a major source of information on long-term climate variability. However, such reconstructions can present major methodological challenges, as calibration methods based on modern climatic and biological patterns may yield biased results, especially when data are particularly sensitive to the influence of secondary environmental variables in the training set or fossil assemblage. A machine-learning technique, boosted regression tree (BRT), is compared with a weighted averaging-partial least squares (WA-PLS) regression in a diatom-based summer (mean July) temperature transfer function ..