Aim The bias in modern North American pollen assemblages by landscape disturbance from Euro-American settlement has long been overlooked in the construction of pollen–climate transfer functions. Our aim is to examine this problem and to develop an unbiased pre-settlement pollen–climate transfer function, and to test its performance and inference power in comparison with commonly used techniques.
Location Minnesota, USA, is of palaeoclimatic interest because within the state are located two continental-scale ecotones, controlled by temperature and available moisture. Shifts of these ecotones can be tracked using palaeoecological techniques.
Methods Using a data set of pre-settlement pollen assemblages from 133 lakes, which were coupled to climate data from the earliest instrumental records (i.e. 1895–1924), a pre-settlement pollen–climate data set was developed that lacked the influence of anthropogenic landscape disturbance. A corresponding modern pollen data set (from lake sediment core tops) and a modern climate (i.e. 1961–90) data set were also developed. The two pollen sets were compared to demonstrate the effects of landscape disturbance from human activities. Ordination (redundancy analysis with Monte Carlo permutation tests) and regression techniques (generalized linear modelling) were used to establish the relationships between the early instrumental climate variables and pre-settlement pollen assemblages and individual taxa, respectively. Transfer functions for the most suitable climate variables (i.e. those forming a minimal set of non-collinear climate variables that explained the greatest amount of pollen variance) were developed from the pre-settlement data set using bootstrapping.
Results Comparison of pre-settlement pollen and modern pollen showed an over-representation of Ambrosia, Chenopodiaceae and Poaceae, and an under-representation of arboreal taxa (e.g. Pinus, Quercus, Ostrya) in the modern assemblages. Not surprisingly, ordination and regression techniques showed a strong relationship between the early instrumental climate variables and pre-settlement pollen assemblages and taxa. Transfer functions were developed for May and February mean temperature and available moisture. Pre-settlement transfer functions substantially improved the root mean squared error by 37–72% in comparison with modern transfer functions inferring pre-settlement conditions, suggesting that the modern transfer functions have poorer predictive abilities.
Main conclusions For climatic reconstructions, there can be a serious distortion of inferences based solely on modern pollen–climate data sets in regions where anthropogenic landscape disturbance has occurred. By using historical climate data, coupled with pre-disturbance pollen assemblages, robust transfer functions for temperature and effective moisture were developed.