Pollen–climate transfer functions intended for temperate eastern Asia

Pollen data of 646 surface samples from northern China and Mongolia and climatic data from the relevant meteorological stations were collected in this study to develop more reliable pollen–climate transfer functions for temperate eastern Asia. Canonical correspondence analysis (CCA) was used to examine the pollen–climate relationships, and mean summer precipitation (MSP) and mean January temperature (MJaT) were inferred to be the first and second important factors controlling the spatial distribution of the surface pollen in the study area. The original dataset was screened with CCA for MSP and MJaT separately to detect anomalous samples that show the extreme values. The first screened dataset was established after excluding those anomalous samples, and the initial transfer function was generated using the weighted averaging partial least squares (WAPLS) method. The jackknife test was then applied to the WAPLS model for determining the optimum transfer function and for detecting large-residual samples, and the final transfer function was generated after removing the large-residual samples from the first screened dataset. The final dataset for MSP and MJaT consists of 428 and 419 samples, respectively. The root mean square errors of prediction for both WAPLS models are 34 mm and 2.7 °C, and the coefficients of determination are 0.85 and 0.73. This study suggests that the values of climatic parameters could be better estimated and the reliability of pollen–climate transfer functions would be significantly improved through removing anomalous and large-residual samples from the original dataset with mathematical methods.