The relationships of modern pollen spectra to vegetation and climate along a steppe–forest–tundra transition in southern Siberia, explored by decision trees

• Published in: The Holocene
• Main Authors: Pelánková, Barbora, Kuneš, Petr, Chytrý, Milan, Jankovská, Vlasta, Ermakov, Nikolai, Svobodová-Svitavská, Helena
• Format: Article in Journal/Newspaper
• Language: English
• Published: SAGE Publications 2008
• Subjects: Tundra; Siberia
• Online Access: http://dx.doi.org/10.1177/0959683608096600; http://journals.sagepub.com/doi/pdf/10.1177/0959683608096600

We studied the relationships between surface pollen spectra, vegetation and selected climate characteristics along a strong gradient of climate continentality across the Western Sayan Mountains, southern Siberia. Representation of 111 pollen taxa in 81 surface samples from steppe, forest and tundra was related to the vegetation composition at various distances from the sampling point and to mean annual precipitation and mean July and January temperatures. These relationships were assessed by the decision tree models. The results show (1) which vegetation types are well distinguished by their pollen spectra; (2) which vegetation types are strongly similar in their pollen spectra and therefore their interpretation from fossil pollen spectra should be carefully considered; (3) tight relationship between surface pollen spectra and selected climate characteristics, which suggests that the past climatic conditions can be reasonably predicted from pollen spectra; and (4) an important role of weak pollen producers for assignment of pollen spectra to vegetation types or particular values of temperature and precipitation. We found the decision trees suitable for analysis of pollen/vegetation relationship because they (1) formally and precisely assign the pollen spectra to vegetation/landscape types or climatic variables by means of easy-to-interpret graphs; (2) identify pollen taxa that are best indicators of a particular vegetation type, landscape or climate characteristics; and (3) utilize the pollen signal of both strong and weak pollen producers. We compare the decision tree models with ordination and cluster analysis and suggest further applications.