Modern spore-pollen spectra of the Altai-Sayan region, their relationship with climate and transfer functions for palaeoclimate reconstructions

Quantitative reconstruction of paleoclimate based on spore-pollen data remains an important task in the study of long-term climate dynamics. The construction of transfer pollen-climate functions on a training set of modern spore-pollen spectra is an effective method for such studies, especially necessary in areas that are poorly supported by numerical reconstructions of paleoclimate, which includes Siberia. To solve this problem, a series of 145 modern spore-pollen spectra were collected during summer expedition at different years from various phytocenoses (plant functional types) representing biomes of: mountain forest, lowland forest, forest-steppe, steppe, desert steppe and alpine tundra-steppe on the territory of the Altai-Sayan mountain region and adjacent areas of the plains (Fig. 1). At each sampling point, from 1 to 6 samples were taken in the form of moss pollsters or surface detritus, geographic coordinates were noted, and a geobotanical description of the vegetation was made. After physicochemical sample preparation, spore-pollen analysis was carried out using generally accepted methods. In common 143 pollen types were identified in study set of modern spore-pollen spectra. To create the transfer pollen-climate function, first of all, we studied by using the method of multivariate statistical analysis the relationship between the composition of the obtained spore-pollen spectra and the composition of maternal phytocenoses (based on geobotanical descriptions made during the collecting of samples), as well as with climatic parameters that could influence the composition of spores-pollen spectra. The results of constrained cluster analysis (Fig. 2) showed that each group of spore-pollen spectra characteristic of a particular biome is distinguished by a separate subcluster of the cluster tree, which confirms the possibility of identifying biomes by spore-pollen spectra. In addition, specific phytocoenoses characterizing plant functionl types are also distinguished by independent clusters.

To study the general structure of the calibration set of modern spore-pollen spectra, a PCA analysis of sampling points (grouped by biomes) and pollen taxa was carried out using the “stats” package basic for “r”, as well as the “vegan” packages 2.6-4 [Oksanen et al., 2018] and “ellipse” 0.5.0 [Murdoch et al., 2018]. Based on the distribution of species and their ecology, axis 1 of the PCA biplot (Fig. 3) reflects the moisture gradient, and axis 2 is associated with the temperature gradient. The pollen types were distributed according to these gradients. The fields of the corresponding biomes are highlighted, by different marks united by colored ovals, which are shown in Fig. 3. Thus, in the most humid and warm conditions, in the upper right quarter of the PCA graph there are located pollen types character for biomes of lowland forests and mountain dark coniferous and “chern’” forests with abundance of fir (Abies sibirica), birch (Betual pendula) and linden (Tilia) with tall grass and fern grass cover. In cold and dry conditions (lower left quarter of the PCA plot) pollen types of the alpine tundra-steppe (yellow oval) and desert steppe (pink oval) biomes are located.

 To identify the influence of climatic factors on the variability of spore-pollen spectra, RDA analysis was performed on six factors: MAP - mean annual precipitation; TJAN - mean temperature of January; TJUL - mean temperature of July; MAT - mean annual temperature; Altitude and GCI - index continentality of Garchinski. Inflation Factors test showed that Altitude and GCI correlate positive with each other and strongly negative with MAT, hence they are not recommended for transfer function construction. The RDA plot (Fig. 4) revealed a positive correlation between MAP and TJAN, as well as Abies sibirica pollen and spores of ferns (Monolete), which reflects the spreading of dark coniferous tall-herb-fern mountain taiga and “chern’” forests with fir, aspen, and linden on the western macroslope of the Kuznetski Alatau Mountains in an area with maximum precipitation and milder winters with abundant snow cover. A positive correlation was found between the Altitude factor and pollen of Pinus sibirica, Betula nana and Cyperaceae, reflecting the ecological conditions of the upper part of the mountain forest belt and the subalpine belt of sparse cedar forests with thickets of Betula nana, sedges and areas of alpine meadows. The pollen of xerophytic plants, from taxa Artemisia and Chenopodiaceae is strongly correlates with GCI. Pollen of Poaceae is equally correlates with GCI and Altitude factors, reflecting the distribution of grasses in both high-mountain tundra and steppe. Tree pollen of Pinus sylvestris and Betula pendula has maximum positive correlation with MAT, while with TJAN+MAP and TJUL these species correlate less strong.  

The cluster analysis, as well as PCA and RDA analyses showed that the composition of the studied spore-pollen spectra adequately reflects not only the peculiarities of the altitudinal belts (biomes) of the vegetation cover and the composition of the parent phytocoenoses (plant functional types), but also the temperature-humidity gradients existing in this area. Consequently, despite the complex structure of the vegetation cover of the mountain region, the presented series of spore-pollen spectra can be used as a training set in the construction of transfer functions for their use in paleoreconstructions based on paleopalynological data.

Transfer function modeling was performed with the R package “rioja” 1.0-6 using the numerical methods WA, WA-PLS, MAT*, MLRC - evaluated by Bootstrap Cross-Validation to identify the strongest model for paleo reconstructions [Hall & Wilson, 1991; Payne et al., 2012]. Statistical analysis showed that for the presented set of modern spore-pollen spectra a significant models can be built for the factors TJUL, MAT, TJAN and MAP. Of the 4 types of models (based on WA, WAPLS, MAT* and MLRC methods) which we created for the 5 variables MAT, MAP, TJAN, TJUL and GCI for the presented set of modern spore-pollen spectra, the best model results were obtained by the MAT* method for the factors MAT, MAP, TJAN and GCI (Table 1). However, the transfer function model for TJUL created by the MLRC method (R2=0.7268 and RMSE=1.68°C) was the strongest. By performance characteristics our TJUL model is comparable to previously published models by other authors created for reconstruction the mean July temperature of the Arctic zone of Siberia [Klemm et al., 2013], and for reconstruction the vegetation cover characteristics such as afforestation [Tarasov et al., 2007; Zanon et al., 2018], NDVI [Liu et al., 2013; Chen et al., 2019], and fractional vegetation cover [Li et al., 2024].

Further statistical analysis of the data and comparison of the results with those published for the neighboring region of the central Tienshan Mountains [Li et al., 2024] showed that the leading climatic factor controlling the variability of spore-pollen spectra in the Altai-Sayan mountains of southern Siberia is the temperature of the growing season expressed as TJUL, while in the mountains of the central Tienshan such a factor is the annual precipitation - MAP. This reflects well the natural geographical patterns of vegetation-climate dependence in the more northern, humid and cold Altai and in the more southern hot and continental climate of Tienshan. Taking into account the different leading factors controlling the variability of modern spore-pollen spectra and vegetation in the two regions under consideration, the newly created transfer functions can be recommended for paleoclimatic reconstructions in the Altai-Sayan region.