Recent above-ground biomass changes in central Chukotka (Russian Far East) using field sampling and Landsat satellite data

. Upscaling plant biomass distribution and dynamics is essential for estimating carbon stocks and carbon balance. In 15 this respect, the Russian Far East is among the least investigated subarctic regions despite its known vegetation sensitivity to ongoing warming. We representatively harvested above-ground biomass (AGB, separated by dominant taxa) at 40 sampling plots in central Chukotka. We used ordination to relate field-based taxa projective cover and Landsat-derived vegetation indices. A general additive model was used to link the ordination scores to AGB. We then mapped AGB for paired Landsat-derived time-slices (i.e. 2000/2001/2002 and 2016/2017), in four study regions covering a wide vegetation gradient from 20 closed-canopy larch forests to barren alpine tundra. We provide AGB estimates and changes in AGB that were previously lacking for central Chukotka at a high spatial resolution and a detailed description of taxonomical contributions. Generally, AGB in the study region ranges from 0 to 16 kg m -2 , with Cajander larch providing the highest contribution. Comparison of changes in AGB within the investigated period shows that the greatest changes (up to 1.25 kg m -2 yr -1 ) occurred in the northern taiga and in areas where land cover changed to larch closed-canopy forest. As well as the notable changes, increases in AGB 25 also occur within the land cover classes. Our estimations indicate a general increase in total AGB throughout the investigated tundra-taiga and northern taiga, whereas the tundra showed no evidence of change in AGB. as a other herb and and underestimate tundra-taiga northern Our satellite-derived estimations match the magnitude of the ground data and show greater detail in the spatial phytomass distribution for the study


Introduction
Estimated global mean surface temperature has increased by 0.87 °C since pre-industrial times and continues to rise (IPCC, 2019). The Arctic is warming two to three times faster than the global annual 30 average. Here, vast amounts of terrestrial carbon are stored in the soil organic matter and living plant https://doi.org/10.5194/bg-2020-416 Preprint. Discussion started: 20 November 2020 c Author(s) 2020. CC BY 4.0 License. biomass (McGuaer et al., 2009;ACIA, 2005) and, therefore, changes in the carbon cycle potentially affected by climate change are a central issue. In the course of global warming, positive feedbacks can be observed: for example, encroachment of deep-rooted vegetation due to shrubification can lead to deeper carbon deposition and act as a potential carbon sink (Jobbágy and Jackson, 2000). Therefore, estimation 35 of above-ground biomass (AGB) stocks and detailed knowledge about the individual taxa contributing to it is of prime interest to understand whether northernmost forests and tundra also change in biomass in analogy to the widespread observed shrubification. This information is essential for modelling terrestrial carbon cycling in vulnerable high-latitude ecosystems and will help predict future carbon dynamics that may accelerate or slow down future warming. 40 Detailed (species/taxa level) estimation of AGB can provide more valuable information on ecosystem's functioning and its development than AGB estimates at a plant functional type (PFT) level. For example, a loss of specific species from one PFT can be replaced by taxa from another PFT in response to climate change even though total AGB production remains similar (Bret-Harte et al, 2008). Thus, the change in AGB between PFTs can be caused by changing species contributions within PFTs. However, many 45 studies of arctic and subarctic regions present AGB state or change at a PFT level (Räsänen et al, 2018;Berner et al, 2018;Webb et al, 2017;Walker et al, 2003). Some focus only on shrub biomass of one or more species (Vankoughnett and Grogan, 2015;Berner et al, 2014), others on tree biomass (Berner et al, 2012), or on species and PFT AGB of a one specific community (e.g. Hudson and Henry, 2009). Rarely, a study presents results of AGB on a PFT level despite sampling methods that suggest a division by 50 species in the field (Maslov et al, 2016;Chen et al, 2009). Very seldom, AGB is presented at a species/taxa level (e.g. Shaver and Chapin, 1991). In consequence, only a few estimations of species or taxon-specific AGB are available to assess species/taxa contributions.
Whereas for some Arctic regions in North America, AGB state and change have been well studied (e.g. Canada, Hudson, 2009), the Russian Far East has received less attention and AGB has never been 55 investigated in the vast areas of central Chukotka, which is our study region. The very few existing circumpolar AGB estimations that also cover these areas (Raynolds et al., 2011;Santoro and Cartus, 2019) have a coarse spatial resolution (1 km and 100 m respectively) and, therefore, show only the general AGB gradient of lowest in tundra to highest in taiga. Similarly, the circumpolar estimation of Epstein et https://doi.org/10.5194/bg-2020-416 Preprint. Discussion started: 20 November 2020 c Author(s) 2020. CC BY 4.0 License. al. (2012) covers AGB change until 2010 and shows only a general zonal pattern of change. In 60 consequence, it remains unknown how the landscape of central Chukotka, with its characteristic treeline formed by needle-leaf deciduous trees, mountainous terrain, and high diversity of vegetation communities, responds to climate warming in terms of terrestrial carbon stocks.
For vegetation and ABG investigations the remote sensing index Normalised Difference Vegetation Index (NDVI) is often used. It incorporates information from red and near infra-red regions of the light spectrum 65 that reflect plant biomass of various ecological systems (Pettorelli, 2006). In the Arctic and subarctic regions remote-sensing algorithms based on satellite derived NDVI and field measurements were used to predict the total and exclusively shrub AGB in Alaska (Epstein et al, 2008;Berner et al, 2018) and for Cajander larch in north-eastern Siberia (Berner et al, 2012). Some studies have used very high spatialresolution imagery (Räsanen et al., 2018) and hyperspectral field spectrometry for AGB investigations in 70 north-western and northern Siberia and Alaska (Bratsch, 2017), that enable spatially restricted studies on estimations of local AGB. To capture more precisely the AGB variability in our study region,  established a redundancy analysis model (RDA) that incorporates Landsat NDVI, Normalised Difference Water Index (NDWI), and Normalised Difference Snow Index (NDSI). This model, together with the extensive Landsat satellite data archive, makes it possible to assess the strength 75 and direction of AGB changes in central Chukotka over the last decades.
We used available Landsat satellite data and field data from a 2018 expedition in a statistical model for AGB mapping. The aim was to provide an estimation of AGB stocks and their change between paired time points (2000/2001/2002 to 2016/2017) at four focus areas along a tundra-taiga gradient, in central Chukotka. Our first objective was to reconstruct the AGB of each sampling plot using individual plant 80 biomass samples and their corresponding distribution within these plots. The second objective was to upscale AGB in the focus areas for the most recent time covered by Landsat-8 satellite data via statistical modelling. Finally, the third objective was to apply the developed upscaling approach to the oldest available good quality Landsat-7 acquisitions to investigate AGB changes in the focus areas.
During the expedition "Chukotka 2018" in July 2018, we inventoried 40 sample plots (Fig. 1;Biskaborn et al., 2019): five sample plots in treeless tundra (16-KP-04), 27 sample plots in the tundra-taiga ecotone (16-KP-01), and eight sample plots in northern taiga (18-BIL-01, 18-BIL-02). Fifteen-metre radius sampling plots were demarcated in the most homogeneous locations. Heterogeneity was accommodated 95 by roughly assorting vegetation into two to three vegetation types per sampling plot. Within each area of roughly estimated vegetation types we selected three representative 2 x 2 m subplots for ground-layer foliage projective cover assessment. In these subplots, a 50 x 50 cm area was selected for ground-layer ABG harvesting (major taxa and others), as well as a 10 x 10 cm area for moss and lichen biomass harvesting (Fig. 2). AGB was sampled in 38 sample plots of the 40 inventoried. 100 All biomass samples were weighed fresh in the field. In general, biomass samples with a weight of more than 15 g were subsampled to reduce the volume of biomass as there were limits to what was logistically possible to transport to the laboratory for drying. All samples were oven dried (60 °C, 24 h for groundlayer and moss and lichen samples, 48 h for shrub and tree branch samples, up to one week for tree stem discs) and weighed again. 105 Our 2018 vegetation and biomass sampling plots were consistently placed in similar vegetation communities to those investigated in 2016. Only tall dense Alnus viridis ssp. fruticosa (Rupr.) Nyman (hereafter Alnus fruticosa) shrub associations were not sampled during the expedition in 2018, which is a rare type of vegetation community that only occurs in a few places in the area of interest. Additionally, we sampled the vegetation at an old fire scar, mostly consisting of patches of tall non-creeping Salix spp. 110 shrubs with graminoids and dead, upright tree stems of Larix cajanderi Mayr. The sampling protocols for projective cover and AGB sampling are different for 1) trees (all Larix cajanderi), 2) non-creeping shrubs (Salix spp., Alnus fruticosa, Pinus pumila (Pall.) Regel), and 3) ground-layer plants (including creeping shrubs, herbs, mosses, and lichens).
Tree cover and heights of all trees were visually estimated in the 15 m radius plot after training with a 115 clinometer (SUUNTO, Finland). Detailed parameters of ten trees per 15 m radius plot were recorded: height, crown diameter, crown start, stem perimeter at basal and at 1.3 m height, and vitality. We aimed to representatively sample at least three (tall, medium, low) of these trees for ABG. Samples included, if available, needle biomass, one small living branch, one medium-sized living branch, one big living branch, one dead branch, and ideally three stem discs (basal height at 0 cm, breast-height at 130 cm, and 120 260 cm height). From the 107 trees sampled, 53 trees were fully sampled, 41 trees were sampled only from the tree trunk, and 13 trees only from branches and needles. Stem biomass was reconstructed using allometric equations (Appendix A) based on the assumption of a cone-shaped tree form. Using exponential models, we were able to reconstruct total and partial (wood, needle) ABG of all trees (separately for dead and living trees) in each 15 m radius plot. We converted our AGB estimates into 125 averages of kg m -2 for each 15 m radius plot.
Non-creeping shrub cover was estimated in the 15 m radius plot. If present, three representative shrub individuals from each species were sampled for AGB: leaf/needle and branch. The average total and partial AGB from representative shrubs were then converted to kg m -2 for each sample plot (Appendix A). 130 Ground-layer vegetation cover was estimated in 2 x 2 m representative subplots. AGB of ground-layer plants was estimated by harvesting 50 x 50 cm subplots; AGB of mosses and lichens by harvesting 10 x 10 cm subplots. By accounting for the vegetation types within each 15 m radius plot, the total average ABG of each sampled taxon was estimated in kg m -2 per sample plot (details in Appendix A).

Above-ground biomass upscaling and change derivation
A redundancy analysis (RDA) model was built with foliage projective cover of 36 taxa from the 2016 140 expedition sample plots as dependent variables and Landsat spectral indices (Normalised Difference Vegetation Index (NDVI), Normalised Difference Water Index (NDWI), Normalised Difference Snow Index (NDSI)) as predictors (Shevtsova et al., 2020a). We used the RDA model to predict RDA scores for the 40 new sample plots of the 2018 expedition. Foliage projective cover of the new sample plots covered the same taxonomical resolution and was standardised by applying a Hellinger transformation 145 (Legendre and Legendre, 2012). Every position in the ordination space describes a specific vegetation composition with a specific coverage, as well as a combination of Landsat spectral indices associated with it. Using the RDA scores, we assigned sample plots from the 2018 expedition to the four established land-cover classes using k-means classification: (1) larch closed-canopy forest, (2) forest tundra and shrub tundra, (3) graminoid tundra, (4) prostrate herb tundra and barren areas (Shevtsova et al., 2020a). 150 For predicting the total AGB for the 2018 sample plots, the RDA scores of the two first axes were used to build a generalised additive model (GAM, R package "mgcv") using Eq. (1).
where RDA1 and RDA2 are the ordination scores of the first and second axes, respectively, of the 2018 expedition data from sample plots where ABG was sampled, and s is a smooth monotonic function. The 155 parameterised GAM was subsequently used to estimate the total AGB for the four focus areas based on the RDA-scores of Landsat spectral indices (Table 1). Specifically, for each focus area the AGB was mapped for each of two time points: recent (2016 or 2017) and historical (2000, 2001 or 2002). From AGB maps with 15-16 years difference covering the same focus area, AGB change maps were produced.
The state and any change of AGB were estimated within and between land-cover classes for land-cover 160 state and change maps (Shevtsova et al., 2020a). All final estimations of AGB state are presented in kg m -2 as median with IQR.

Vegetation composition and above-ground biomass
In situ projective cover data of all 2018 expedition vegetation sample plots are described in Shevtsova et al. (2020b). The main vegetation communities of the study region assessed were: (1)  The predictions of the 40 new sample plots into RDA-space assigned two sample plots to the class "larch 175 closed-canopy forest", 17 sample plots to "forest tundra and shrub tundra", 13 sample plots to "graminoid tundra", and seven sample plots to "prostrate herb tundra and barren" ( Additionally, we analysed individual partial AGB of four taxa: Larix cajanderi, Alnus fruticosa, Pinus pumila, and non-creeping Salix spp. (Fig. 6). Pinus pumila had a very wide range of needle to wood mass ratios, including a ratio indicating a higher weight of needle biomass compared to wood biomass from an individual shrub. For all other investigated species this is not the case. In contrast, deciduous-needled 210 larch has the lowest weight ratio of needles to wood when compared to P. pumila, Salix spp., and A.

Upscaling above-ground biomass using GAM
In the GAM, the RDA scores are explanatory variables and total AGB is the dependent variable. The first 215 two RDA axes explain 87% of the variance in the AGB data ( Table 2). Both variables (parametric coefficient RDA1 and the smooth term s(RDA1, RDA2)) are highly significant in the model.
We plotted fitted values against residuals for the GAM model to visualise residual standard deviations (SD) for every sample plot used in the modelling (Fig. 7). There is some slight heteroscedasticity and the SD increases with an increase of absolute AGB values. The RMSE of the model is 1.08 kg.
Based on the most recent Landsat data acquisitions, the maximum total AGB estimated within our study area is found in the northern taiga in the larch closed-canopy forests (20-24 kg m -2 , 16-KP-02, Fig. 8). In the southern tundra-taiga transition (16-KP-03) maximum AGB reached 12 kg m -2 at places in a river valley that are covered by azonal dense forests. In the northern tundra-taiga (16-KP-01) maximum AGB is 4-6 kg m -2 in the forest tundra and shrub tundra. In the tundra (16-KP-04) it is 3-4 kg m -2 on the slopes 225 of rivers' valleys.

Change of above-ground biomass between 2000 and 2017 in the four focus areas
The compiled change-maps of recent ( AGB of land-cover classes that did not change within the investigated period tend to have higher values moving from the tundra to northern taiga (Fig. 10).
We find an increase in AGB for those areas where land-cover class has changed ( Table 3). The highest changes in the paired years occurred in the southern tundra-taiga (16-KP-03; +4.30 kg m -2 ) and the 270 northern taiga (16-KP-02: +4.09 kg m -2 ) associated with a change in land-cover class from forest tundra and shrub tundra to larch closed-canopy forest. The lowest AGB change rates are associated with a change in land-cover class from graminoid tundra to forest tundra and shrub tundra in the northern taiga (16-KP-02) and southern tundra-taiga (16-KP-03). In general, total AGB in the tundra focus area has not changed over the time studied (0 kg m -2 , IQR=0.2 kg m -2 ), while in the northern tundra-taiga it has increased by 0.69 kg m -2 (IQR=0.69 kg m -2 ) and by 0.44 kg m -2 (IQR=0.91 kg m -2 ) in the southern tundra-taiga. In the northern taiga total AGB has increased much more than in the other focus areas by 1.3 kg m -2 (IQR=1.4 kg m -2 ). 280 We estimated total and partial dry AGB for the 2018 expedition sample plots, which cover a wide range of vegetation associations (Shevtsova et al., 2020c;Shevtsova et al., 2020d). From these field biomass samples, AGB estimates range from 0 to 15 kg m -2 and, as expected, reflect a gradient of land-cover classes from the least vegetated prostrate herb tundra and barren areas to the larch closed-canopy forests.

Recent state of above-ground biomass at the field sites
As in other subarctic and arctic vegetation studies the taxa found in our study region can be grouped into 285 similar PFTs for a convenient comparison. Thus, deciduous shrubs are largely represented by Betula nana, Vaccinium uliginosum and Salix sp., which are typical circumpolar subarctic species (Grigoryev, 1946) and are widely found, for example in the tundra in Alaska near Toolik Lake (Shaver and Chapin, 1991).
In graminoid tundra, which, by its characteristics, is comparable to tussock tundra in Alaska, deciduous shrubs contribute 33% to the total AGB (tundra, median=0.09 kg m -2 , IQR=0.05 kg m -2 ) or 9% (tundra-290 taiga, median=0.07 kg m -2 , IQR=0.05 kg m -2 ), which is similar to deciduous shrub AGB of Alaskan tussock tundra (0.09±0.02 kg m -2 ). However, in Alaska, deciduous shrub contribution to the total AGB is 16%, which is lower than the central Chukotka graminoid tundra, but higher than the graminoid tundra in the central Chukotkan tundra-taiga. Evergreen shrub taxa are also similar in our study region to those near Toolik Lake, Alaska being mainly represented by Ledum palustre, Vaccinium vitis-idaea, Dryas 295 octopetala, and Empetrum nigrum with Pinus pumila in our study region in contrast to Alaska. Evergreen shrubs generally have a lower AGB in the graminoid tundra of our study region (tundra, median=0.08, IQR=0.11; tundra-taiga, median=0.03, IQR=0.10) than in the tussock tundra of Alaska (0.17±0.02 kg m -2 ), but the percentage of this PFT is slightly higher (31%) in central Chukotka than in Alaska (24%).
In the graminoid tundra of the central Chukotka tundra-taiga, AGB of evergreen shrubs is poorly 300 represented (4%). Graminoids in our region were not separately sampled but are included as "other". The highest contribution to partial AGB in central Chukotka is from Cajander larch (Larix cajanderi), the only tree species present in the study region. Despite many studies using complex allometric equations, 325 mostly including tree height and stem diameter (e.g. Dong at al., 2020;Alexander et al., 2012;Bjarnadottir et al., 2007) to estimate AGB of an individual tree, we used only tree height because stem diameter measurements (stem perimeter) were not available for all trees. However, where measurements of tree stem diameters were available, these are shown to be highly correlated with height, which makes it rational to use only height to estimate tree AGB to avoid multicollinearity in the model. Other 330 parameters (crown height, crown width) were also measured on a subset of trees and proved to be insignificant predictors. Thus, using estimated tree height we provide coherent AGB estimation models by accounting for living state (live or dead) and ecological zone (tundra-taiga, northern taiga). We also estimated leaf and wood biomass separately and summed them up in the data processing procedure (Appendix A). These established allometric equations can be applied at a broad scale in central Chukotka 335 to a range of tree heights (up to 20 m), as covered by our study.

Recent state of above-ground biomass upscaled for central Chukotka
The and also, partly, the northern tundra-taiga is generally comparable to the AGB of the North Slope of 385 Alaska, which ranges from 0 to 4 kg m -2 (Berner et al, 2018 open forests in the Kolyma river area ranges, on average, from 0.5 to 5 kg m -2 , reaching the maximum of 6.7 kg m -2 , which is comparable with our forest tundra and shrub tundra AGB assuming a 57% representation of Larix cajanderi in this land-cover class. Many factors can influence the AGB estimates such as the number of reference samples, prediction method, remote sensing sensor type (optical, radar), as well as spatial and temporal resolution of the 395 satellite imagery and products (Fassnacht et al., 2014). Overall, a comparison with global and circumpolar AGB estimates highlights great improvements in the accuracy of the estimates and a better way to resolve a more landscape-related spatial pattern of our AGB estimates for the study region.

Change in above-ground biomass within the investigated 15-16 years in central Chukotka
We with modelling extreme temperature increases in Alaskan tundra (Hobbie and Chapin, 1998). In their study, Hobbie and Chapin (1998) conclude that, in tundra, plant biomass accumulation depends on nutrient availability and AGB will only increase if mineralisation of soil organic nutrients is stimulated 410 together with climate warming. Given differences in soil development between the focus areas of tundra, tundra-taiga and northern taiga, their conclusion may also apply to our results. In general, the comparison with circumpolar estimated AGB changes from 1982 to 2010 (Walker and Raynolds, 2018) shows that changes in AGB in our focus areas of central Chukotka between 2000 and 2017 were much faster, probably because of the stronger warming in the first decades of the 21 st century in these regions. 415 Our estimates of AGB change within our land-cover classes show that AGB change does not necessarily lead to a change in land-cover class. We assume that changes for different regions within the same stable land-cover classes could be associated with population size change, but also, likely, with changes in the plant's parameters (height, crown density etc.). This could explain why the change in AGB estimated for the graminoid tundra in the northern taiga (16-KP-02) is greater than for the tundra (16-KP-04, Fig.10).

Conclusions
We successfully used field-based AGB data and Landsat satellite data in statistical modelling to map recent (2016/2017) and historical (2000/2001/2002) states of AGB in four focus areas along a tundrataiga gradient in central Chukotka. The total AGB values consist of major taxon-specific (and other) estimates that allow us to achieve a more detailed picture of AGB change and to reveal changes in major 425 species contributions from areas with diverse ecology. In addition, we were able to analyse changes in AGB together with changes in land-cover classes.
AGB of the investigated areas in the field ranged from 0 to 16 kg m -2 . Taxa making the most contribution to AGB in our study region include Cajander larch (Larix cajanderi) in forest stands, and dwarf birch, IQR=1.14 kg m -2 ), but are rare at the landscape level and are azonal. Thus, an expansion of forest would make the strongest change to total ABG, but it is still unclear how fast taiga could colonise tundra areas 435 in the upcoming decades. Nevertheless, taxon-specific estimations allow us to separate tree biomass from other vegetation forms, expanding the usefulness of our study to treeline migration assessment and forest management in the study region. in the spatial phytomass distribution for the study region.
We found that the greatest AGB changes occurred in the northern taiga, particularly in the larch closed-450 canopy forest class (+4.09 kg m -2 ), which also has the highest AGB and most favourable environment for the expansion of Larix cajanderi which contributes highly (92% on average) to AGB. The less favourable environments in the tundra-taiga and tundra would need more time to adapt to recent climate changes.
We found changes in AGB that are not only associated with changes in land-cover classes, but also within areas with no changes in land-cover class. This could indicate either that vegetation composition changes 455 are not yet prominent enough to trigger a change in land-cover class, or that there has been a change in plant properties (height, crown diameter, leaf size etc.) within the investigated period.
Overall, our mapped AGB of recent and historical times in central Chukotka are of value in helping to understand regional ecosystem dynamics as well as circumpolar processes, especially in the light of recent climate changes. The specific parameterisation of plant biomass from central Chukotka make our AGB 460 maps most suitable for the region and more precise in terms of spatial resolution than global and circumpolar estimations of AGB. Future uses of our AGB state and change maps could include modelling of carbon stocks and investigating habitat changes in the area. Knowing the recent and historical AGB distribution and the contributing taxa is useful for modelling studies that aim to project future AGB changes, as well as for policy-making, particularly in relation to mitigation of climate-change impacts 465 and conservation.

Appendix A. Sampling and above-ground biomass (AGB) calculation protocol for field data
Here we present the step-by-step protocol for harvesting and calculating ground-layer AGB for a 30 Calculation for Pinus pumila shrub AGB. We sampled three (small, medium, big) individual pine plants on each 30 x 30 m sample plot that contained the species. With the following steps we calculated the AGB for each individual plant:

1)
Woody AGB of all small living branches (g): where is dry weight of subsample of a small branch wood; , or are size of an individual plant; is the number of small branches, or is the fresh weight of a whole sample or subsample of a small branch wood, respectively.

2)
Needle AGB of all small living branches (g):

3)
Woody AGB of all big living branches (g): where is dry weight of subsample of a big branch wood, is the number of big branches, or 500 is the fresh weight of a whole sample or subsample of a big branch wood, respectively. 4) Woody AGB of all dead branches (g): where is dry weight of subsample of a big branch wood, is the number of dead branches, or is the fresh weight of a whole sample or subsample of a dead branch wood, respectively. 505

5)
Average AGB of small living branch wood (across the three different-sized samples, g): where is the average dry weight for only the woody part of a plant, is number of cones, and is cones biomass.

10)
Average volume of a shrub crown (cm 3 ): where , and is height of a small, medium and big plant respectively; 1 and 2 are two measurements of a diameter of a crown perpendicular directions.

11)
Average wood AGB of Pinus pumila (g m -2 ): where is the average woody mass of a plant per m 2 . 525

12)
Average needle AGB of Pinus pumila (g m -2 ): where is the average needles' mass of a plant per m 2 .

13)
Total average AGB of Pinus pumila shrub on a 30 x 30 m sample plot (kg m -2 ): Calculation for Alnus fruticosa and Salix sp. shrubs AGB. We sampled three (small, medium, big) individuals as for Pinus pumila at each plot if present. Calculations are similar as for pine, but include not only big and small branches, but also medium 535 branches.
Calculation for Larix cajanderi AGB. Larix cajanderi trees were representatively subsampled at the following parts: living branches (small, medium, big), dead branches, needles from small branches, stem (ideally three tree discs at 0, 1.3, and 2.6 m heights), and cones. Total AGB of an individual tree (g) from the field survey of 2018 expedition was calculated as following: 540 where is total dry AGB of a tree, is dry weight of biomass of branches and leaves, is dry weight of stem biomass.
2) = * + * + * + where is the number of small branches, is the small branch dry biomass, is the small branch needles dry biomass, is the number of medium branches, is medium branch dry biomass, is number of big branches, is dry biomass of big branches, is number of dead branches, is dead branch biomass, is number of cones, and is cones biomass. 3) where is volume ( − is a base of a tree stem from 0 to 130 cm, − is a middle part of a tree stem from 130 to 260 cm, C is a top part of a tree stem from 260 to the top), is the wood density of a tree part (base, middle or top).

4)
where is the wood density of tree disc A and is the wood density of tree disc B.

5)
where is the wood density of a tree disc C.
where is the volume of a tree disc sampled at 0 cm tree stem height, is dry weight of a tree disc sampled at 0 cm tree stem height, ℎ is height of a tree disc sampled at 0 cm tree stem height, is diameter of a tree disc sampled at 0 cm tree stem height, is diameter of a circular hole in the central part of a disc (if present), and and are length and 560 https://doi.org/10.5194/bg-2020-416 Preprint. Discussion started: 20 November 2020 c Author(s) 2020. CC BY 4.0 License.
average width of a crack in the tree disc, respectively (if present). and are calculated by analogy with . 7) Calculation of volume of a tree part (base, middle or top) varies depending on presence or absence of a central hole in the tree stem.
Scenario 1: A hole in the tree disc is absent Dz = 0: 565 where − is the volume of a tree stem part from 0 (A) to 130 cm (B), is diameter of disc A, and is diameter of disc B.
where is the volume of a top part of a tree stem from 260 cm to the full height of a tree ( ) and is the diameter of disc C. 570 Scenario 2: A hole in the tree disc is present only in disc A Dz ≠ 0 (only A): where is the diameter of a central circular hole in disc A.
by analogy with Scenario 1.
where is the diameter of a central circular hole in disc B.
by analogy with Scenario 1.
The next step in estimation of Larix cajanderi AGB was to estimate it for the 30x30 m sample plot, limited to tree height as a 580 predictor. We differentiated between allometric equations to estimate partial individual larch AGB from trees from two ecological regions (tundra-taiga and northern taiga).
To assess the different models for different regions we used a Wilcoxon rank sum test on measurements of tree stem perimeters.
It showed significant differences between basal perimeter and perimeter at 1.3 m height of trees from 16-KP-01 (tundra-taiga, 178 samples) and BIL-18 (northern taiga, 74 samples) (Fig. B1). In both cases, tree basal perimeter (p=0.007453) and tree 585 perimeter at 1.3 m (p=0.03014) in the tundra-taiga is statistically greater than in northern taiga. Since individual trees are significantly different in the two regions, different AGB-prediction models are required for the tundra-taiga and northern taiga focus areas.
where AGB is above ground biomass and is tree height in cm . where is number of trees on the 15 m radius sample plot, is the needle biomass of a tree, and is the woody biomass of a tree.

Data availability statement
The data that support the findings of this study are published in the PANGAEA® Data Repository for Earth and Environmental  https://doi.org/10.5194/bg-2020-416 Preprint. Discussion started: 20 November 2020 c Author(s) 2020. CC BY 4.0 License.