Quantitative methods for reconstruction of the Holocene paleoclimatic characteristics and vegetation changes based on palynological data from lake and peat sediments

Abstract

The paper presents the results of a methodological study of the quantitative reconstruction of paleoclimatic characteristics (mean January and July temperatures, mean annual temperature and precipitation) and forest cover within a 20 km radius of the study site using palynological data from lake and peat sediments. The following quantitative methods have been tested: the best modern analog technique, transfer functions as weighted averaging, weighted averaging partial least square, and the Random Forest method, a meta estimator that fits a number of decision tree classifiers on various sub-samples of the dataset. A database of subfossil pollen assemblages for northern central Siberia (north of 60°C) was used as a training data set for model construction, containing 174 pollen assemblages, climatic characteristics for the sampling area, and calculations of forest cover based on remote sensing data. Leave-one-out cross validation was appy for testing of the methods. The results showed that all models had the highest coefficients of determination (R2), the smallest errors and the lowest uncertainty for the reconstruction of the mean July temperature. The developed methods were applied to two key sites in Central Siberia, located in the vicinity of Igarka and the settlement of Tura (Krasnoyarsk Region). The results obtained showed similar trends in the changes in palaeoclimatic characteristics reconstructed by different methods, but a rather large variation in temperature and precipitation values at some time intervals.