Abstract:  

Vegetation cover was an important factor in palaeo-climate simulation.The study of pollen-vegetation cover transfer function based on modern processes provides a way to reconstruct the past vegetation cover change by using pollen stratum data.The Inner Mongolia Plateau and its adjacent area,which is located at the marginal monsoon area of northern China and has a large climatic and vegetal gradient,provides us an ideal area to study the pollen-vegetation cover transfer functions.In this paper,we investigated pollen assemblages of lake central surface sediment from 46 lakes and reservoirs,and developed transfer functions between lake surface sediment pollen assemblage and Normalized Difference Vegetation Index (NDVI) using three different methods which include weighted average partial least squares (WA-PLS),local weighted average method (LWWA) and modern analogue technique (MAT).We used leave-one-out cross-validation,bootstrapping and autocorrelation test to evaluate the performance of those three pollen-NDVI transfer functions and select the most reliable to reconstruct the Holocene vegetation cover history of Huitengxile Plateau of central Inner Mongolia.As results,the redundancy analysis (RDA) was conducted on the pollen data and environmental data according to the linear data type of the pollen data,as indicated by the gradient length (i.e.,2.047) of the detrended correspondence analysis (DCA).The RDA plot displays that the NDVI points to the positive of the first axis where forest and forest steppe sites clustered,and has smaller angles with the Pinus and Betula and larger angles with the desert pollen type such as Nitraria,Ephedra and Tamaricaeae,which suggested that the NDVI index was closely related to the spatial distribution of pollen and thus the pollen-NDVI transfer functions could be reliably developed.The leave-one-out cross-validation and the bootstrapping test showed that the LWWA_Inv (R²_jack=0.57, R²_boot=0.57) and WA-PLS-3 (R²_jack=0.56, R²_boot=0.54) models have better performance than the MAT (R²_jack=0.53, R²_boot=0.58) model.The autocorrelation test suggests that both MAT model and WA-PLS model are greatly affected by spatial autocorrelation,and thus not suitable to develop pollen-NDVI transfer function.Therefore,the LWWA Inv model is suitable to reconstruct the past vegetation cover history (NDVI) for its overall good performance,and was used to reconstructed the NDVI history of Huitengxile Plateau of Inner Mongolia by applying to a fossil pollen record of a lacustrine sections from a unnamed small crater lake.The comparison shows there are broad consistence between pollen reconstructed NDVI and other pollen-based vegetation (i.e.,biome score in this paper),which further demonstrated the reliability of our pollen-NDVI transfer functions.Furthermore,we compared those regional vegetation evolution with those non-pollen based Holocene temperature and precipitation sequences.The result shows that the pollen-based vegetation sequences were correlated well with the non-pollen based precipitations but not correlated with regional temperature sequences,which possibly indicated the Holocene vegetation change was closely responded to precipitation but temperature in marginal monsoon area of northern China.

Key words: font-size:10.5pt, Inner Mongolia Plateau')">">Inner Mongolia Plateau, lake surface sediment, pollen assemblage, vegetation coverage, font-size:10.5pt, NDVI')">">NDVI, font-size:10.5pt, transfer function')">">transfer function