Climate reconstructions based on GDGTs and pollen surface datasets from Mongolia and Siberia: Calibrations and applicability to extremely dry and cold environments

datasets from Mongolia and Siberia: Calibrations and applicability to extremely dry and cold environments Lucas Dugerdil1,2, Sébastien Joannin2, Odile Peyron2, Isabelle Jouffroy-Bapicot3, Boris Vannière3, Bazartseren Boldgiv4, and Guillemette Ménot1 1Univ. Lyon, ENS de Lyon, Université Lyon 1, CNRS, UMR 5276 LGL-TPE, F-69364, Lyon, France 2Université de Montpellier, CNRS, IRD, EPHE, UMR 5554 ISEM, Montpellier, France 3Université Bourgogne Franche Comté, CNRS UMR 6249 Laboratoire Chrono-environnement, F-25030, Besançon, France 4Ecology Group, Department of Biology, School of Arts and Sciences, National University of Mongolia, Ulaanbaatar 14201, Mongolia Correspondence: Lucas Dugerdil, lucas.dugerdil@ens-lyon.fr

. GDGT assemblages reflect bacterial activity in rivers (De Jonge et al., 2014b), soil (De Jonge et al., 2014a) or lakes (Dang et al., 2018) which is also linked to climate parameters (Weijers et al., 2007b), soil typology and vegetation cover (Davtian et al., 2016), which in turn imply land cover and land use. Accurate determinations of the relationships between GDGT assemblages and climate still need some improvements (Naafs et al., 2018) and especially on local to 55 regional scales and in extreme environments.
To evaluate the provenance and the climatic information br-GDGTs bear, several indices have been proposed in the literature (Table. S1). The first proposed to determine the origin of the GDGTs is the BIT index (Hopmans et al., 2004), followed by the III a /II a index Xiao et al. (2016). Furthermore, it has been shown both empirically Huguet et al., 2013) 60 and on cultures of pure strains (Salvador-Castel et al., 2019 in press) that organisms adjust their membrane plasticity by the degree and the position of these compounds. The br-GDGT assemblages are also a function of the bacterial species (Naafs et al., 2018) present in the environment. To monitor these changes, CBT and MBT indexes linked to environmental factors such as climate and soil parameters have been proposed (Weijers et al., 2007b;Huguet et al., 2013). Some more accurate indexes have been proposed by De Jonge et al. (2014a) to limit the multi-correlation systems such as MBT 5Me which is independent 65 of the pH and CBT 5Me which is more representative of the soil pH than the former version of the index. The statistical relevance of these indexes is a major issue in br-GDGT calibration (Crampton-Flood et al., 2019). Some regional indexes for soil temperature such as Index 1 (De Jonge et al., 2014a) and Index 2 for Chinese soils  have been explored too.
For the moisture variations, the R i/b index has been proposed as a reliable aridity proxy (Yang et al., 2014;Xie et al., 2012).
It has been shown that a linear relation exists between these GDGT indexes and some climatic features (Yang et al., 2014;Lei 70 et al., 2016). The Siberian and Mongolian surface soil samples are used to calibrate new climatic relation with GDGTs.
The aim of this study is to take advantage of new, modern surface sample datasets in Siberia and Mongolia to propose an 80 adapted calibration of pollen and bacterial biomarker proxies for cold and dry environments. For that purpose, local calibrations are compared with large calibrations to infer the influence of calibration scale and proxy types on derived climatic parameters.
Our approach is summarized in the following steps: transect. The Mongolian GIS Data is issued from a dataset ASTER (https://biosurvey.ku.edu/directory/nicholas-kotlinski), the meteorological dataset from WorldClim2 and infrastructures from public dataset (ALAGaC) (https://marine.rutgers.edu/ cfree/gis-data/mongolia-gis-data/) 5 https://doi.org/10.5194/bg-2019- 475 Preprint. Discussion started: 30 January 2020 c Author(s) 2020. CC BY 4.0 License. the dark-taiga dominated by larches (Larix sibirica) and Siberian pines (Pinus sibirica) also presents some spruces and fewer birches (Betula spp.). The Mongolian taiga is constrained to a region spanning from the Darkhad Basin to the Khentii range ( Fig. 1.A). On the north face of the Khangai piedmont, the vegetation is dominated by a mosaic of forest-steppe ecosystems: the steppe is dominated by the Artemisia spp. associated with Poaceae, Amaranthaceae, Liliaceae, Fabaceae and Apiaceae.
On these open-lands there are some patches of taiga forest, following roughly the broadside and the northern face of the crest 120 letting on to the grasslands in the valley. The two last vegetation layers through the elevation gradient is an alpine meadow dominated by Cyperaceae and Poaceae with a huge floristic biodiversity and an alpine shrubland with pioneer vegetation on the summits. On the southern slope of the range, the ecotone between the steppe and the desert vegetation extends hundreds of kilometers from the northern part of the Gobi desert (with water supplied by the Gobi lake area in the middle) to the Gobi-Altai range in the south (Klinge and Sauer, 2019). In the southernmost part of the country, the warm and dry climate conditions 125 favour desert vegetation dominated by Amaranthaceae, Nitrariaceae and Zygophyllaceae. The vegetation cover is lower than 25% and is mainly composed of short herbs, succulent plants and a few crawling shrubs.

Bioclimate Systems
In the central steppe-forest biome the vegetation is marked by an ecotone with short grassland controlled by grazing in the 130 valley and larches on the slopes. The forest is gathered in patches constituting between 10% and 20% of the total vegetation cover. There are also some patches of Salix and Betula riparian forests among the sub-alpine meadows on the upper part of the range. This vegetation is characteristic of the northern border of the Palaearctic steppe biome (Wesche et al., 2016). This biome is characterised by a range of 800 to 1600 m.a.s.l, a Mean Annual Air Temperature (MAAT, Fig. 1.C) between -2 and 2 • C and a Mean Annual Precipitations (MAP, fig. 1.B) from 180 to 400 mm.yr −1 (Wesche et al. (2016) based on Hijmans 135 et al., 2005). In Mongolia, even if the MAP are very low (M AP M ongolia ∈ [50; 500]mm.yr −1 ), the major part of the water available for plants is delivered during late spring and early summer, in contrast to Mediterranean and European steppes (Bone et al., 2015;Wesche et al., 2016). These seasons are the optimal plant growth periods. Mongolian summer precipitations are controlled by the East Asian Summer Monsoon system (EASM) instead of the Westerlies' winter precipitation stocked onto the Sayan and Altai range (Fig. 2, An et al., 2008).

Pollen Analysis and Transfer Functions
Different chemical processes were performed on the samples: the mosses were deflocculated by KOH and filtered by 250µm and 10µm sieves to eliminate the vegetation pieces and the clay particles. Then, acetolysis was performed to destroy biological tocol to remove all the carbonate and silicate components. All the residuals were finally concentrated in glycerol and mounted between slide and lamella. The pollen counts were carried out with a Leica DM1000 LED microscope on a 40× magnification lens. The total pollen count size was determined by the asymptotic behaviour of the rarefaction curve. This diagram was plotted during the pollen count using PolSais 2.0, software developed in Python 2.7 for this study. The rarefaction curve was fitted with 150 a logarithmic regression analysis. The counter was suspended whenever the regression curve reached a flatter step (Birks et al., 1992). A threshold for the local derivation at dx/dy = 0.05 was set. The total count is usually around n ∈ [350; 500] grains for steppe or forest and n ∈ [250; 300] for desert slides.
Among all of the pollen-inferred climate methods, the MAT and the WAPLS were applied in this study on 4 different 155 modern pollen datasets. The MAT consists of the selection of a limited number of analogue surface pollen assemblages with their associated climatic values. (Jackson and Williams, 2004); while the WAPLS uses a Weight Average correlation method on a limited number of Principal Components connecting the surface pollen fraction to the climate parameters associated (Ter Braak and Juggins, 1993;Ter Braak et al., 1993). The first dataset, called New Mongolian-Siberian Database (NMSDB), is composed of pollen surface samples analysed in this study (N = 49, Fig. 2). The second one is the same subset aggregated to 160 the larger Eurasian Pollen Dataset (EAPDB) compiled by Peyron et al. (2013Peyron et al. ( , 2017. From this dataset of 3191 pollen sample sites, a pollen-Plant Functional Type method was applied to determine the biome for each sample according to the actual pollen rain (Prentice et al., 1996;Peyron et al., 1998). Then, only the Cold Steppe (COST) dominant samples were extracted from the main dataset and aggregated with the NMSDB to produce the COSTDB (N = 482 sites). Finally, a scale-intermediate dataset of samples located within the Mongolian border merged with the Mongolian New dataset is presented as MDB (N = 151 sites).

165
The relation between each taxa and climate parameter was checked and then the MAT and WAPLS methods were applied with the Rioja package from the R environment (Juggins and Juggins, 2019).

SIG Bioclimatic Data
Because Mongolia and Siberia have relatively few weather stations ( Fig.1.A), climate parameters were extracted with R from the extrapolated climatic database WorldClim2 (Fick and Hijmans, 2017). We used Mean Annual Precipitation (MAP, Fig.1.B),

170
Mean Annual Air Temperature (MAAT, Fig.1.C), as well as temperatures and precipitations for spring, summer and winter (T spr , P spr , T sum , P sum , T win and P win ), Mean Temperature of the Coldest Month (MTCO) and the Mean Temperature of the Warmest Month (MTWA) in this study to characterize the actual climate. The elevation data and the topographic map originate from the ASTER imagery ( Fig.1.A). The biome type for each site derives from the LandCover database (Olson et al., 2001), classification and field trip observations.

GDGT Analysis
After freeze drying, about 0.6 grams of surface samples were subsampled. The Total Lipid Extract (TLE) was microwave extracted (MARS 6 CEM) with dichloromethane (DCM):MeOH (3:1) and filtered on empty SPE cartridges. The extraction was processed twice. Following Huguet et al. (2006), C 46 GDGT (GDGT with two glycerol heed groups linked by C 20 alkyl chain  (Fig.4.B, Sinninghe Damsté, 2016), as follows: To infer temperature from br-GDGT abundances, two kinds of model were applied: the linear relation between temperature  Table S2). These models were clustered into 3 categories: the type of sample, the geographical area and the statistical model. According to the type of environment from which the samples originated, there was peat, soil and lake-inferred modelling. For the geographical area of samples, we discriminated the regional model (on the country or district scale) from the global model made on the world scale. Finally, there were 2 statistical families of model: the first one was built 200 on common ratios like the MBT-CBT, and the second one was inferred with the multiple regression model. All these models were applied on the Mongolian surface samples and compared with the actual MAAT value at each site.

Statistical Analyses
GDGTs and pollen matrix were analysed with a Principal Components Analysis (PCA) to determine the axes explaining the variance within the samples. The biotic values (pollen and GDGTs) were also compared to abiotic parameters (climate, 205 elevation, location and soil features) by the way of a Redundancy Analysis (RDA). The regression models were run with the p − value < 0.05 for the relevance of the model, the R 2 for the level of correlation between the variables, the RMSE to determine the climate error of the models and Akaike's information criterion (AIC) to quantify the over-parameterization effect of multiple regression models (Arnold, 2010;Symonds and Moussalli, 2011). All the statistical analyses were performed with the Rcran project, using the ade4 package (Dray and Dufour, 2007) for multivariate analysis. All the plots were made with the 210 ggplot2 package (Wickham, 2016) or the Rioja package (Juggins and Juggins, 2019) for the stratigraphic plot and the pollen clustering using the CONISS analysis method (Grimm, 1987). 1. Light taiga dominated by Pinus sylvestris (> 70%), Pinus sibirica and very low NAP (< 5%).

Pollen -Climate Interaction
The pollen rain trends follow similar variations to bio-climate parameters in MAP, MAAT and elevation (  and Cyperaceae, Artemisia spp. and Brassicaceae percentages. MAP, fairly related to RDA 1 , rises with AP and decreases with NAP ( Fig.5.D). Finally, the elevation gradient favors Artemisia spp. and Cyperaceae for NAP and Salix spp. and Larix sibirica for AP (Fig.5.D).

MAT and WAPLS Results
To reconstruct bioclimatic parameters from pollen data, MAT and WAPLS methods were applied on the four scales, modern 245 pollen datasets and the ten climate parameters (Table 1). All these methods can be run with n ∈ [1; 10] parameters: the number of analogues for MAT and the number of components for WAPLS. The best transfer functions among all of them were selected by the following approach: in a first step, for each climate parameter the methods fitting with the higher R 2 and the lower RMSE were selected. Then, in case the highest R 2 and the lowest RMSE were not applied for the same number of analogues or components, we chose the method presenting the lower number of parameters. Despite the small number of parameters 250 relative to the number of observations, the method fits well (Arnold, 2010, table 1). MAT transfer function gives better R 2 in bigger DB than in smaller ones. Fitting increases with the diversity and the size of DB, since MAT is looking for the closest value between climate and pollen abundance. By contrast, WAPLS fits better on the local scale and especially with a smaller number of sites. In this case, the pull of data is largest and the variance is largest (Ter Braak and Juggins, 1993). WAPLS also leads to better value of RMSE than R 2 , in contrast to MAT. For temperature, pollen fits better with T spr or MTWA in Mongo-

GDGT Variance in the dataset
Iso-GDGTs are dominated by GDGT 0 and crenarcheol (X percent in relative abundances, respectively, in Fig.4.A). Since the majority of these molecules are thought to be produced in the lake water column , the variations of fractional abundance in the soils and moss samples are very discrete and poorly linked to climate parameters. Br-GDGT concentrations differ depending on the sample type: br-GDGT fractional abundances are consistent with each type of sample: the major compounds are the I a , II a , II a and III a and lacustrine core top sediments (blue). The punctuation marks ' and " refer to 6 and 7 methyl, respectively.
The sediment samples from Mongolian lakes are more homogeneous than the surface samples, especially when compared with the moss polsters that present a wide variability (Fig. 4B). Generally, soil samples are more relevant analogues to sediments 280 than moss polsters (Fig. 4B). This variability shows an influence of the sample type on br-GDGT responses. Sample type also bears climate information, since soil and moss polsters originate from steppe to desert environments and forest/alpine meadows, respectively.

Climate Influence on br-GDGT Indexes
The influence of the bio-climate parameters on the br-GDGT matrix variance is connected with MAP ( Fig. 5.B: RDA 1 = 10.01%).

285
The negative values show higher precipitations and uncyclized 5-Me GDGTs, such as I a , II a and III a . While the lower MAP
Caryo. match with 6 or 7-Me GDGTs, such as III a , II a , II a . The RDA 2 is slightly more connected to MAAT and elevation, also clustering the methyled and cyclized GDGTs to the higher MAAT. The correlation between chemical structure and climate parameters Huguet et al., 2013) was not strong. All the MBT, MBT', MBT 5Me and CBT, CBT' , CBT 5Me

Multi-regression Models
The Stepwise Selection Model for the climate -br-GDGT modelling was applied only on the 5 and 6 Methyls, because 7-   Fig. 5A and B. The ∆T values closest to 0 reveal the best fitting model on each point ( Fig. 7, panel 1). Then, the box-plot (Fig. 7, panel 2) summarises the best fitting model at the regional scale.

305
Among the possible methods, statistical values help to select the most reliable ones for palaeoclimate reconstruction. However, the correlation (R 2 ) and errors (RMSE) are not enough to discriminate between them to identify the most suitable ones for palaeoclimate modelling, especially for the multi-parameter methods, such as br-GDGT multi-regression models and pollen transfer functions. Indeed, the more input parameters in the method, the more accurate it is (Tables 1, 2 and Fig. 6A and 6B). All the regression models improve with parameter additions, and especially the less fitting methods improve exponentially (lower 310 limit of the R 2 area, Fig. 6B). The best R-squared-models for each parameter number (Fig. 6A) correspond to the upper limit of the R 2 area (Fig. 6B). The R 2 trend in function of the parameter number follows a logistic regression both for MAAT mr and MAP mr models. However, and especially for MAAT mr regression models, this logistic curve becomes asymptotic early, similar to the RMSE decrease. This model over-parameterization has proven to produce artefacts in ecological modelling (Arnold, 2010;Symonds and Moussalli, 2011). The issue is thus to identify the threshold in the parameter numbers selected. 315 We used Akaike's Information Criterion (AIC) to determine the better model without over-parameterization for br-GDGT regression models: the lower the AIC, the better the model ( Table 2). The trend of AIC versus the parameter number is however more complex (Fig. 6C). For MAAT mr , the regression model becomes more accurate from one to five parameters quite rapidly, but then slowly decreases. The AIC curve takes an asymmetrical hollow shape around five parameters with a steeper slope on the left side (Fig. 6A). The AIC values for MAAT mr6 and MAAT mr7 are almost identical (Fig. 6A) . The MAP mr6,7,8 have 320 almost equivalent AIC values, while the AIC curve shapes differ for the other MAP mr models (asymmetrical hollow shape around five with a steeper slope on the left side, Fig. 6A). To sum-up, the most universal models are MAAT mr5 and MAP mr7 but the closed models are also valuable in some local contexts. We need to determine the cross-values of these models to select the appropriate ones for the Siberian-Mongolian context.

325
The cross-values of the nine best MAAT mr regression models ( Fig. 7A1 and 7A2) and the best MAP mr regression models Both on MAAT mr and MAP mr models, the 95 % interval shrinks with parameter addition, but the mean values do not necessarily get closer to the climate parameter measured value (the dashed line in Fig. 7.A2 and B2). Therefore, if the tests on 335 the AIC point toward the MAAT mr4 and the MAP mr7 regression models, the back-cross plots suggest the MAAT mr3 and the MAP mr6 regression models provide the best estimates for climate reconstruction in lacustrine archives (∆MAP = 0 and best fitting temperature for the mean value of all samples, Fig. 7.B2 and Fig. 7.B1).

Global vs. Local Calibration
Whatever proxy is used, when reconstructing temperatures and precipitation from past archives in a given location, there is the Similarly, for pollen transfer functions, the geographic range of the surface samples on which the calibration relies is a rel-355 evant parameter to take into account for the trustworthiness of the paleoclimate reconstructions. The choice of the maximum value of this geographic range has been discussed previously for vegetation modelling, for example, the Relevant Source Area of Pollen (RSAP, Hellman et al., 2009a, b;Bunting and Hjelle, 2010;Prentice, 1985). For MAT and WAPLS regression models, the same issue holds true. The responses of the eight over-represented taxa to climate parameters are different in the three geographic ranges. The linear tendency allows for checking the main trends between taxa distribution and climate parameters,        (Table 1). On these subsets, the WAPLS RMSE and R-square values are even higher than for the MAT transfer function. The major difficulty resides in the reconstructions of precipitation. Even if the RMSE and R 2 values are higher for all models of MAP than MAAT, the influence of precipitation on vegetation cover is not well understood. In Mongolia it is clear that the precipitation controls the treeline in mountainous areas (Klinge and Sauer, 2019) and the global openness in the steppe -forest ecotone (Wesche et al., 2016) as well as human land-use (Tian et al., 2014), but the risk of autocorrelation between 375 MAAT and MAP signals is important, even if the RMSE and R 2 values are higher for MAP regression models than for MAAT ones (Telford and Birks, 2009;Cao et al., 2014).

Issues in Modelling Mongolian Extreme Bioclimate
Firstly, the commonly used br-GDGT indexes (MBT and CBT) are not relevant for arid areas with MAP < 500mm.yr −1 because of the relation between low soil water content and soil br-GDGT preservation and conservation interferes in the br-380 GDGT / climate parameters (Dang et al., 2016). Moreover, the main issue in climate modelling in Mongolia is the narrow relationship between MAAT and MAP. Because of the climatic gradient from dry deserts in the southern latitudes to wet taiga forests in the northern ones, MAAT and MAP maps are strongly anti-correlated ( Fig. 1 B and C). This correlation could also be a bias resulting from the interpolation method of the WorldClim2 database. In fact, there are very few weather stations ( Fig. 1.A, Fick and Hijmans, 2017) on the large Mongolian plateau area and a great diversity of mountain ranges interrupting 385 them. Moreover, the relevance of the interpolation models suffers from the transition threshold made by Mongolia between the EASM and the Siberian Westerlies (Fig.2, An et al., 2008).
However, both GDGT and pollen models show that the precipitation calibrations are more reliable than temperature ones (Tables 2, 1, Figures 3, 6 and 7), reflecting that the Siberian-Mongolian system seems to be controlled by precipitation. This 390 dominance of precipitation could be due to seasonality. Even if the br-GDGT production is considered to be mainly linked to MAAT (Weijers et al., 2007a, b;Peterse et al., 2012), the high pressure Mongolian climate system (Zheng et al., 2004;An et al., 2008) favors a strong seasonal contrast: almost all the precipitation and the positive temperature values happen during the summer (Wesche et al., 2016). Consequently, for the NMSDB pollen models (Table 1). The seasonal parameters such as MTWA, T sum and P sum better describe the GDGT variability than MAAT and MAP. It is the opposite on EAPDB and 395 COSTDB models. The Mongolian permafrost persists half the year in the northern part of the country (Sharkhuu, 2003) and acts on vegetation cover and pollen production (Klinge et al., 2018). Furthermore, the effects of frozen soils on soil bacterial communities and GDGT production are thought to be important (Kusch et al., 2019). This system leads to a quasi equivalence between MAP and P Sum while MAAT is torn apart by the large T Sum − T win contrast. The MAP appears then to be the most reliable climate parameter for Siberian-Mongolian climate studies under the threshold of about 5 • C.

400
To reduce the signal/noise ratio, a wider diversity of sample sites should be added as initial inputs in the models. This raises the question of the availability of reliable samples in desert areas. The soil samples in the steppe to desert biomes are often very dry and these over-oxic soil conditions are the worst for both pollen preservation (Li et al., 2005;Xu et al., 2009) and GDGT production (Dang et al., 2016). br-GDGT concentrations in moss polsters and sediments are thus higher than in soils 405 in our database.
The soil of the Gobi desert also has a high salinity level which is also a parameter of control on br-GDGT fractional abundance (Zang et al., 2018). Even if the impact on sporopollenin is not well understood, the salinity also affects pollen conservation in soils (Reddy and Goss, 1971;Gul and Ahmad, 2006). This taphonomic bias (also climatically induced) could For instance, in a tropical context, temperature values are too high to be linearly linked to fractional abundances (Pérez-Angel et al., 2019). Considering pollen-climate relationships, the inferior limit of pollen percentage is critical: for the majority of pollen types, whenever the MAAT or MAP reaches a very high or low threshold, the pollen fraction approaches zero (Fig. 8).
These limit areas need to be investigated closely, which legitimises the local calibration methods.

Conclusions
The palaeoenvironmental and palaeoclimatic signals present several uncertainties which can misguide the interpretation of past variations. This study shows how both a multi-proxy approach and accurate calibration are important in combating these biases. We propose a new calibration for Mean Annual Precipitation (MAP) and Mean Annual Air Temperature (MAAT) from br-GDGTs as well as a new pollen surface database available for transfer functions. The correlations between pollen rain and 425 climate on one hand and br-GDGT soil production and climate on the other are visible but are still mitigated by the complex climate system of arid central Asia and the diversity of soils and ecosystems. Precisely, each of our proxies seems to be more narrowly linked to precipitation (MAP) than temperature (MAAT) counter to the majority of calibrations in the literature.
The nature of the samples considered (soil, moss polster and sediment top-core) also greatly affected these correlations. The calibration work in the extreme bio-climates of the Siberian basin and Mongolian plateau is difficult because of the low 430 range of climate values, despite the climate diversity ranging from cold and slightly wet (north) to the arid and warm (south).
The MAAT and MAP values do not remarkably spread in the vectorial space, which makes harder to distinguish the linear correlation against variance noise. Moreover, this range of values is close to the lower saturation limit of the proxies, which makes the accurate local calibration tricky but necessary. The local calibrations also suffer from the reduced size and small geographic extent of the dataset. The vegetation cover, extending from a high cover taiga forest to nude soil desert cover, also 435 buffers the climate signal and the GDGT / pollen response. The correlations between climate parameters and GDGT / pollen proportion are therefore lower than they could be at the global scale. Nonetheless, and despite the lower correlation of the local calibration, these local approaches appear to be more accurate to fit with the actual climate parameters than the global ones: both for pollen function transfer and br-GDGT multiple regression models. These positive model results have to be considered in front of over-parameterization limits. Too many parameters in MR-GDGT models or in pollen MAT or WAPLS transfer 440 function can add artificially to the linear relation between climate and proxies and lead to misinterpretation of palaeoclimate records. Akaike's information criterion associated with RMSE and R 2 values is a fair way to select the best climate model.
We encourage the wider application of this local multi-proxy calibration for a more accurate constraint of these central Asian climatic systems, a crucial improvement to properly model the fluctuations of the Monsoon Line since the Optimum Holocene.

445
The new calibration of climate reconstruction for Mongolia and Siberia presented in this study is based up on the New Mongolia-Siberia Data Base (NMSDB). Location, ecosystems as well as sample type are provided in Table ( Table A3. Synthesis of the formulae for the main indexes br-GDGT fractional abundances.   Figure A1. A : Cross-plot for MAPmr model. The parcels are sorted by increasing parameter number.
B : Cross-plot for selected climate MBT-CBT models.