Climate synchronises shrub growth across a high‐arctic archipelago: contrasting implications of summer and winter warming

1012 –––––––––––––––––––––––––––––––––––––––– © 2020 The Authors. Oikos published by John Wiley & Sons Ltd on behalf of Nordic Society Oikos This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. Subject Editor: Bente Graae Editor-in-Chief: Dries Bonte Accepted 29 February 2020 129: 1012–1027, 2020 doi: 10.1111/oik.07059 129 1 0 1 2 – 1027


Introduction
Global warming and its ecological implications are more pronounced in the Arctic than elsewhere (Post et al. 2009a, CAFF 2013, Larsen et al. 2014. Indeed, plant 1013 and animal populations at high latitudes are often strongly influenced by climate and year-to-year weather variations (Stenseth et al. 1999, Saether et al. 2003, Ims and Fuglei 2005, Post et al. 2009b, Hansen et al. 2013, Van der Wal and Stien 2014, Myers-Smith et al. 2015a, making Arctic species particularly susceptible to climate change (Ims and Fuglei 2005, Post et al. 2009b, Ims and Ehrich 2013, Descamps et al. 2017. Climate-induced environmental changes have already altered tundra ecosystems, with potential reinforcing feedbacks on global warming (Zhang et al. 2013). In particular, in spite of a considerable spatial heterogeneity in climate change impacts (Elmendorf et al. 2012a), studies from across the Arctic report overall positive effects of warmer summers on plant growth (Forbes et al. 2010, Van der Wal and Stien 2014, Myers-Smith et al. 2015a, Weijers et al. 2017, Ackerman et al. 2018, Bjorkman et al. 2018. Accordingly, increased vegetation productivity and shrub expansion (i.e. 'shrubification') have led to 'Arctic greening' (Elmendorf et al. 2012b, Macias-Fauria et al. 2012, Ju and Masek 2016, Vickers et al. 2016. Recently, however, large-scale vegetation damage ('Arctic browning') has also been observed and might among other reasons be related to extreme warm spells in winter, which can be accompanied by rain-on-snow events (ROS) (AMAP 2017, Pan et al. 2018, Peeters et al. 2019). In the Arctic where the snow cover is often thin, ROS events can lead to an encasement of plants in basal ice instead of snow (Bokhorst et al. 2008, Phoenix and Bjerke 2016, Bjerke et al. 2017). How such events affect variation in Arctic plants' growth, as well as vital rates such as recruitment or mortality, spatially and temporally, is still far from understood, and has not been investigated based on long-term time-series data. Climate projections forecast not only warmer summers, but also rainier winters throughout arctic regions (Hansen et al. 2014, Bintanja andAndry 2017). Therefore, disentangling the effects of such contrasting environmental drivers on spatiotemporal patterns of plant growth is crucial for our understanding of large-scale and long-term climate change effects. This is particularly important for ecosystem dynamics, since changes in plant growth could potentially trigger trophic cascades (Wookey et al. 2009).
Fluctuations in weather and climate are often correlated over large distances (Koenig 1999, Koenig 2002, Stenseth et al. 2003. Moran (1953) suggested that the dynamics of animal populations may be synchronised over similarly large spatial scales if they are strongly influenced by these environmental variables (the 'Moran effect', Royama 1992, Ranta et al. 1997). This phenomenon is seen in a large variety of taxa, ranging from large carnivores (Stenseth et al. 1999) to viruses (Pitzer et al. 2015), and has been reported for example in rodents (Bjørnstad et al. 1995, Krebs et al. 2002, Ims and Andreassen 2005, herbivores (Grenfell et al. 1998, Grøtan et al. 2005, insects (Peltonen et al. 2002, Ims et al. 2004, Sheppard et al. 2015, birds (Engen et al. 2005, Saether et al. 2007, Koenig and Liebhold 2016, fish (Östman et al. 2017) and plankton (Massie et al. 2015, Defriez et al. 2016. The concept has also been applied to and observed in plants, noticeable as a synchrony in e.g. recruitment, primary or secondary growth (i.e. radial growth) (Post 2003, Defriez and Reuman 2017, Shestakova et al. 2017. Although relatively few studies have so far examined the synchronising role of environmental fluctuations on the dynamics of plant growth, it is a rapidly expanding research area in tree species (Läänelaid et al. 2012, Koenig and Knops 2013, Shestakova et al. 2017). In many plant species, strong direct influence of climatic drivers, such as temperature, on growth can often be expected. This is especially the case in the Arctic (Elmendorf et al. 2012b, Myers-Smith et al. 2015a, Bjorkman et al. 2018, where fluctuations in primary production play an important role in shaping the dynamics of many bottom-up ecosystems i.e. primary producercontrolled ecosystems (Power 1992, Ims and Fuglei 2005, Post et al. 2009b, Wookey et al. 2009). In this biome, on the other hand, in situ time-series of annual plant biomass production covering a large spatial range are rare, because of difficult logistics associated with long-term, large-scale monitoring programs. However, this is now changing due to the rapid development of dendrochronology applicable to shrubs (Myers-Smith et al. 2015b), as well as of proxy for vegetation productivity maps from remote sensing, such as normalized difference vegetation index (NDVI) (Forbes et al. 2010, Blok et al. 2011, Macias-Fauria et al. 2012, Tape et al. 2012, Weijers et al. 2018b.
Recently, in a study from central Svalbard, Le Moullec et al. (2019) showed that annual secondary growth in the polar willow Salix polaris, a widespread dwarf shrub and important food resource for herbivores in the high-arctic tundra ecosystem of Svalbard (Rønning 1996, Bjørkvoll et al. 2009), can be used as a proxy for above-ground biomass productivity of the entire vascular plant community. Single-site studies from central parts of Svalbard have indicated that annual fluctuations in summer temperature largely determine the year-toyear variation in S. polaris secondary growth (Buchwal et al. 2013, Le Moullec et al. 2019, as well as the above-ground biomass production of vascular plants (Van der Wal and Stien 2014), i.e. the major plane of nutrition for herbivores (Bjørkvoll et al. 2009). The Svalbard tundra hosts a relatively simple, bottom-up controlled food web (Descamps et al. 2017). Here the fluctuations in population size of the only large ungulate, the Svalbard reindeer Rangifer tarandus platyrhynchus, are mainly shaped by forage availability during winter (notably density-dependent effetcs of ROS events) and productivity during summer (regulated by summer temperature, Hansen et al. 2013, Albon et al. 2017. Recent field observations and experiments also indicate that such ice encapsulation following ROS events negatively impact the growth patterns of shrubs, as well as their mortality and reproduction (Milner et al. 2016, Bjerke et al. 2017). On Svalbard, effects of climate warming are more pronounced in winter than in summer and a strong change in the winter precipitation pattern has been observed, with an increase in rainfall that is most pronounced on the west coast (Nordli et al. 2014, Peeters et al. 2019. The role of spatial heterogeneity and the geographical scale at which these contrasting climate drivers act on primary production in the Arctic, as well as their implications for the spatial synchrony of primary production, remain unexplored. Here, we take advantage of the dendrochronological tools developed for shrub species (Kolishchuk 1990) and a rare network of long-term Arctic weather station data to investigate climate-shrub growth relationships in time and space across Svalbard. First, we explore the main climate drivers of secondary growth in S. polaris, asking how and why their effects may vary across the archipelago. Second, we assess to what extent fluctuations in this proxy for annual primary production (Le Moullec et al. 2019) are spatially synchronised, and how climate and weather fluctuations contribute to this synchrony.

Study area and species
Around 60% of the land area of Svalbard (ca 74-81°N, 10-35°E) is covered by glaciers, while only ca 13% is vegetated (Johansen et al. 2012, Fig. 1). The archipelago has a relatively mild climate for its latitude and shows high inter-annual variability in temperature and precipitation (Johansen et al. 2012, Nordli et al. 2014, Van Pelt et al. 2016. Both temperature and precipitation show a negative gradient from southwest to northeast with milder climates inland of deep fjords (Van Pelt et al. 2016). These climatic gradients are strongly influenced by the West Spitsbergen Current, which delivers warm North Atlantic water along the west coast and, the colder currents of the Arctic Ocean on the east coast (Van Pelt et al. 2016). During 1962-2014, the mean annual temperature across Svalbard was −5.8 ± 0.6 (SE)°C, with winter (November-April) temperatures of −12.0 ± 2.6°C and summer (June-August) temperatures of 3.0 ± 0.8°C (Supplementary material Appendix 2 Fig. A1). The annual precipitation was 372 ± 56 mm (winter = 196 ± 44 mm, summer = 81 ± 32 mm, the calculations of these summary statistics follow the same method described below). In this period, annual mean temperature increased with a rate of 0.8°C per decade, the increase in winter temperature being stronger than that in summer temperature (1.1 ± 0.2 and 0.4 ± 0.04 respectively, Supplementary material Appendix 2 Fig. A1), while annual precipitation did not show significant temporal trends (Nordli et al. 2014).
The focal species in this study, Salix polaris, has a nearly circumpolar distribution (absent on Greenland) (< www. flora.dempstercountry.org/ >) and is the only widespread and abundant shrub found in most habitats across the Svalbard archipelago, including in polar deserts (< http://svalbardflora. no >, Rønning 1996). The circumarctic dwarf shrub Cassiope tetragona is also of great value to dendrochronological studies at high latitudes (Callaghan et al. 1989, Rayback and Henry 2005, 2006, Rozema et al. 2009, Weijers et al. 2010, 2018a, Rayback et al. 2011, Milner et al. 2018) but this species is dependent on dry soils with some snow-cover (Rønning 1996) and is hence found in a more restricted area on Svalbard than S. polaris. Also, contrary to C. tetragona, S. polaris is an important food resource for resident herbivores, i.e. the wild Svalbard reindeer and Svalbard rock ptarmigan Lagopus muta hyperborea, as well as for migratory geese Branta spp. (Van der Wal et al. 2000, Bjørkvoll et al. 2009. A particularity of the High Arctic Svalbard ecosystem is the low biomass of lichens, which therefore compose only a marginal part of the reindeer diet in winter (< 2%, Bjørkvoll et al. 2009). Instead, Svalbard reindeer graze on S. polaris year-round, with highest grazing pressure in late winter (~25%, Bjørkvoll et al. 2009).
This prostrate shrub has often only shoot tips, leaves and reproductive structures visible above-ground. The reproduction can be sexual or clonal through rhizomes (Rønning 1996). The below-ground structure usually consists of one main root which can extend several decimetres into the ground, and several thin secondary roots and shoots nested into the ground (Le Moullec et al. 2019, Fig. 2). The shoots usually emerge from the root collar complex, i.e. the oldest part of the plant, which is situated on top of the central root. Both primary and secondary growth of this species are found to be mainly driven by a positive effect of summer temperature (Buchwal et al. 2013, Van der Wal and Stien 2014, Le Moullec et al. 2019. Salix polaris shrubs usually form well-defined, however, often incomplete, annual ring growth (Buchwal et al. 2013). Ring growth in arctic shrubs is often irregular, because they are adapted to tolerate high variability in their physical environment (Crawford 2008, Wilmking et al. 2012). Nevertheless, dendrochronological tools have previously successfully been used on S. polaris at three sites in Svalbard, i.e. Hornsund (Owczarek and Opała 2016), Petuniabukta (Buchwal et al. 2013) and Semmeldalen (Le Moullec et al. 2019). The average age of analysed samples from these studies was 40 years. The oldest shrub of the study was 70 years old, from Petuniabukta.

Shrub sampling and processing
In summer 2015, shrubs of S. polaris (i.e. roots, root collar and shoots) were collected from six sites in Svalbard, covering central parts as well as the eastern, western and southern parts (Table 1, Fig. 1). The sites were characterized by S. polaris forming dense cushions on pioneer vegetation terrain, such as raised beaches (Fig. 2), enabling us to sample well-defined cushions without visible injuries and, thereby, best suited for dendrochronology analysis (Myers-Smith et al. 2015b). We sampled a minimum of five shrubs per site by selecting individuals with a minimum distance of 10 m to avoid sampling from the same individual. Additionally, in the present study we also included previously published chronologies from two sites located in central Spitsbergen, namely from Petuniabukta (composed of 10 shrubs, Buchwal et al. 2013) and Semmeldalen (composed of 30 shrubs, subdivided in 10 sub-sites of 60-2000 m apart for the spatial synchrony analysis, see below, Le Moullec et al. 2019). Our sampling design gave priority to site replication over number of shrubs per site. Hence, the relatively low sample size in some sites restricts the power of our analysis at the site level.
Dwarf shrubs often exhibit eccentric growth, and rings within a single cross-section can be wedging, causing partially missing rings that can be detected by serial sectioning and comparison of so-called 'pointer years', i.e. reference rings (Kolishchuk 1990, Buras andWilmking 2014), utilizing cross-dating techniques (Douglass 1941, Fritts 1976, Pilcher 1990). Because of the limited number of shrubs sampled in each site, we applied extensive serial sectioning to each individual shrub, involving sampling and analysing multiple cross-sections acquired along main root and shoots. A detailed description of the laboratory processes and S. polaris chronology building, including the cross-dating steps, is given in Le Moullec et al. (2019). From each individual shrub we prepared five to six cross-sections of average thickness of 15-20 µm. Adjacent sections were sampled at a minimum distance of 2 cm and cuts were performed using a GSL1 microtome (Gärtner et al. 2014, Tardif andConciatori 2015). Usual cuts included one cross-section from the root collar, two to three from the central taproot (i.e. main root) and one to two cross-sections from shoots. Cross-sections were stained, dehydrated, permanently fixed and digitalized (100× magnified with a camera attached to a light microscope, Gärtner and Schweingruber 2013). Measurements of ring-widths were performed along four radii randomly selected within each quarter of a single crosssection in ImageJ (Schindelin et al. 2015). In a few cases we had to disregard parts of the cross-section that were injured or excessively merged. For standardisation purposes, the total length of the radii was also measured together with the number of remaining rings to the pith. Cross-dating, i.e. aligning each annual ring growth measurement to the corresponding calendar year, is the most essential step in constructing robust chronologies of shrub ring growth (Buchwal et al. 2013, Myers-Smith et al. 2015b). We applied cross-dating at three hierarchical levels (Myers-Smith et al. 2015b, Table 1. Description of sampling sites and Salix polaris ring growth chronology statistics. Vegetation type is extracted from Johansen et al. (2012). The chronologies contain a minimum of 10 cross-sections for the given time-span. Number of plants, cross-sections and radii measured are given, as well as the standard descriptive statistics for within-, betweenand total plant inter-series correlation (i.e. r bar.wt , r bar.bt and r bar.tot , respectively), as well as expressed population signal (EPS).  Le Moullec et al. 2019): 1) between radial measurements within cross-sections; 2) between cross-sections within a single shrub to obtain individual mean growth curves (i.e. time-series) and 3) between shrubs (minimum of five) within a site, to obtain site-specific mean growth curves (using linear mixed-effects models). After cross-dating, we truncated the mean site chronologies to contain a minimum sample depth of 10 cross-sections. No cross-dating was conducted between sites to respect spatial independency of temporal variation across Svalbard. We did not consider tree-ring measurements from the sampling year, 2015, because not all shrubs at all localities had fully completed that years' ring formation at the time of sampling. Accordingly, growth curves end in 2014. A major challenge in dendrochronology is to successfully disentangle the effect of weather from other factors, such as age-related processes, which affect trees' and shrubs' growth simultaneously. Chronology standardisation can help to account for some of these factors, and we followed the steps applied in Le Moullec et al. (2019). As trees become older, stem and root diameter increases so that inevitably, ring-width becomes thinner with increasing age for a similar amount of xylem produced (Biondi andQeadan 2008, Buras andWilmking 2014). Hence, the first standardisation step involved transforming the cross-sections' ring-width measurements to basal area increments (BAI, µm 2 ) using R ver. 3.3.2 (< www.r-project.org >), package 'dplR' (Bunn 2008), which can correct for changes in stem diameter with an increasing age (Visser 1995, Biondi andQeadan 2008). This approach assumes circular cross-sections, which is often not the case for shrubs. However, Buras and Wilmking (2014) showed that moderate eccentricity is reasonably well accounted for by averaging four radial measurements, as was done in this study. Secondly, trees and shrubs often exhibit increased ring-width development in their juvenile phase compared to in their adult phase (Briffa andMelvin 2011, Bowman et al. 2013).

Ny-Ålesund
To account for such age effects, we standardised the cross-section growth curves by the trend in the BAI found when aligning all cross-sectional growth curves from all sites by cambial age, i.e. regional curve standardization (RCS, Briffa and Melvin 2011), again using the R-package 'dplR'. RCS was applied to remove the (minor) age-related trends observed in our shrubs (Supplementary material Appendix 2 Fig. A2). Applying RCS transforms the growth curves into deviations from a mean growth trend representative of expected growth changes arising as a consequence of general shrub ageing (Helama et al. 2016), a process which also stabilizes the variances across ages. These standardisation steps resulted in dimensionless ring-width indices (RWI). Hereafter, we refer to 'ring growth' as a simplification of RWI.
As a standard quality check for ring growth data we used the following selected descriptive statistics: average correlation between growth series of all cross-sections from the same shrub (rbar wt ), the average inter-series correlation between growth series from different shrubs within a site (rbar bt ), the total average correlation between all crosssections within a site (rbar tot ), as well as the expressed population signal (EPS) for the standardised ring growth chronologies established for each study site (Table 1). EPS is an evaluation of how closely the observed mean chronology (based on the finite sample collection) represents a hypothetical mean based on an infinitive number of crosssections (Wigley et al. 1984). These descriptive statistics were obtained using the R-packages 'dplR' (Bunn 2008) and 'detrendR' (Campelo et al. 2012). In addition, we reported the average Gini-coefficient and first-order autocorrelation coefficient, calculated at the shrub level with the R-functions 'gini' and 'acf '.
Annual ring growth was first averaged (i.e. arithmetic mean) over radial measurements obtained from within each cross-section, and then between the cross-sections of a shrub, to obtain a growth curve for the individual shrub. This resulted in the construction of growth curves from 30 individuals from these six sites. Additionally, the previously published studies from Petuniabukta (Buchwal et al. 2013) and Semmeldalen (Le Moullec et al. 2019) contained 10 and 30 individual growth curves respectively. The S. polaris chronology for Petuniabukta was published as non-standardised, i.e. raw ring-width chronology (Buchwal et al. 2013). Therefore, to obtain comparable scales, we mean centred time-series of all individual shrubs from all sites by dividing their annual growth by the shrub's overall mean growth. Finally, to minimize the risk of detecting spurious correlations with climatic variables due to common trends, linear detrending across years was applied to the climate data (see below) and to all ring growth time-series of each individual shrub by extracting the residuals from a linear model of growth as a function of years.
As a summary, we used 70 individual growth curves from eight sites (Fig. 1), spread along a gradient of distances from 60 m (sub-sites within Semmeldalen) to 292 km (between Sørkapp and Ny-Ålesund). The maximum and minimum length of ring growth time-series was 66 (i.e. Petuniabukta and Hornsund) and 26 years (i.e. Sørkapp), respectively (Table 1).

Climate variables
A set of climate variables likely to influence S. polaris growth on Svalbard were selected a priori (see Supplementary material Appendix 2 Table A1 for detailed information on weather station data). Summer temperature was previously documented as an important driver of S. polaris growth at two sites in central Svalbard (i.e. Petunibukta and Semmeldalen, Buchwal et al. 2013, Le Moullec et al. 2019, and summer precipitation seems to promote ring growth in southern regions (i.e. Hornsund, Opała- Owczarek et al. 2018). Amount and type of winter precipitation likely also affect shrub growth in the Arctic: snow, often redistributed by wind, may be beneficial to plants in terms of increased insulation (Hallinger et al. 2010) and increased mineralization rates (Blok et al. 2015), but may shorten the growing season by delaying melt-out (Schmidt et al. 2006, Semenchuk et al. 2016. Spring onset, which is also temperature-dependent, may influence plant performance at high latitudes (Karlsen et al. 2014). Furthermore, ROS may lead to basal ice formation that can injure plants by inducing cell death due to frost damage or because of stress-and anoxia induced metabolite accumulations (Crawford et al. 1994, Bokhorst et al. 2010, Bjerke et al. 2017. Accordingly, we calculated monthly mean air temperature (°C, from 17 weather stations, Supplementary material Appendix 2 Table A1) and total precipitation (mm, from eight weather stations, Supplementary material Appendix 2 Table A1) during summer, as well as total precipitation (mm) falling as snow (i.e. at a temperature < 1°C) or rain (i.e. ROS, at a temperature ≥ 1°C) during winter (November-April, extracted as daily data from <www.eklima.met.no>). ROS was on the natural logarithmic scale after adding one unit to avoid transforming values of 0. A proxy for spring onset was calculated as the Julian day when the smoothed daily temperature (over 10 days) crossed 0°C and stayed above 0°C for ≥ 10 days (Le Moullec et al. 2019). Summer was here defined as July, the month reflecting the peak growing season across the whole archipelago (Supplementary material Appendix 2 Fig. A3). Climate variables' time-series from all weather stations are presented in Supplementary material Appendix 2 Fig. A4, starting from year 1962, which corresponded to the year when Petunibukta and Hornsund's time-series start to overlap.
We also examined whether large-scale climatic patterns impacted plant growth by using summer and winter Arctic Oscillation (AO, 1950(AO, -2014 indices (available on < www. ncdc.noaa.gov/teleconnections/ao/ >, Aanes et al. 2002, Stenseth et al. 2003. The AO is an index measuring a difference in atmospheric pressure measured at sea-level between 20°N and arctic regions (Thompson and Wallace 1998). When atmospheric pressure in the Arctic is low, but high pressure dominates temperate regions, this characterizes positive AO phases. In parts of the Arctic, positive AO index generally results in warm, but stormy, winters with early snowmelt, while summers are often cold, rainy and cloudy (Aanes et al. 2002). Such weather conditions are expected to reduce shrub growth (Aanes et al. 2002, Welker et al. 2005, Weijers et al. 2017. However, large-scale AO does not result in uniform changes of local weather conditions across the Arctic and can result in opposite weather conditions between e.g. Greenland and Fennoscandia (Stenseth et al. 2003, Wang andKey 2003). Since Svalbard is situated at the interface between these regions, there is limited correlation between air temperature and AO (Wang and Key 2003). Additionally, sea ice can influence coastal vegetation productivity by cold air breeze from the sea, we investigated the relation between shrub secondary growth and sea ice cover time-series at the onset of growing season, i.e. June (Forchhammer 2017, Macias-Fauria et al. 2017. We used sea ice on a regional scale (instead of on an arctic-wide scale), from five coastal areas of Svalbard during 1979-2014, extracted from Prop et al. (2015).

Climate effect analysis
We used the available weather data in Svalbard (Supplementary  material Appendix 2 Table A1) to estimate weather variables across Svalbard for each year, using a linear mixed-effects model with year as fixed effect and weather station as random effect on intercept (Supplementary material Appendix 2 Fig.  A4). The same approach was used for June sea ice extent from five areas around Svalbard, providing a general estimate at the Svalbard scale. All climate variables were linearly detrended for the time-span of each individual shrub's time-series to avoid correlations due to common trend. For this climategrowth analysis, the 10 sub-sites in the Semmeldalen valley were treated as one site, resulting in a total of eight sites across Svalbard (Fig. 1).
We performed three different model selection procedures, in which we fitted linear mixed-effect models in the R package 'lmer' (function 'lme4', Bates et al. 2015). Advantages of using linear mixed-effects models in dendrochronological studies following a hierarchical sampling design are detailed in Le Moullec et al. (2019). In the first model selection ('Svalbard scale'), we explored which weather variables explain S. polaris ring growth the best at a Svalbard scale. The full model was composed of additive effects of the weather variables and the previous year's ring growth (i.e. taking into account a possible first-order autocorrelation at the shrub level) (Supplementary material Appendix 2 Table A2). This first step enabled us to identify important environmental variables acting at large spatial scales, and thus potentially acting as synchronising variables. Second, we performed a model selection ('regional scale'), where, opposed to at the 'Svalbard scale', we tested for differences in ring growth responses between sites. In this model selection, site was included as an interaction term with the different weather variables and the previous year's ring growth (Supplementary material Appendix  2 Table A3). To avoid overparameterization, a maximum of six terms (excluding intercept) were allowed in the candidate models. The third model selection ('climate proxy') investigated the potential influence of summer and winter AO indices and fluctuations in June sea ice extent on ring growth across Svalbard (Supplementary material Appendix 2 Table  A4). Because these variables are assumed to reflect weather conditions (Stenseth et al. 2002, Macias-Fauria et al. 2017, the investigation of these climate proxies was conducted in a separate model selection to avoid possible collinearity with the weather variables. For all three model selection procedures described above, the full model and all of its possible subsets were fitted as candidate models (see all proposed models at the 'Svalbard scale' Supplementary material Appendix 2 Table A2, at the 'regional scale' Supplementary material Appendix 2 Table A3 and for the 'climate proxy' Supplementary material Appendix 2 Table A4). Year was always included as random effect on the intercept to account for dependency in the response variable due to spatially correlated annual environmental conditions not captured by the fixed effects. Shrubs within each site cannot be viewed as being independent (due to shared environment and through cross-dating), but since growth data were detrended (i.e. mean of zero for all plants), the effect of plant on the intercept was zero, and shrub ID was therefore not included as random intercept effect. No signs of a substantial random slope effect of plant were found (analysis not presented). There was no serious collinearity between weather variables (r < 0.5, Supplementary material Appendix 2 Table  A5) and climate proxies (r < 0.3, Supplementary material Appendix 2 Table A6). Candidate models were compared using the Akaike information criterion corrected for sample size (AICc, Burnham and Anderson 2002), obtained from models fitted using maximum likelihood (ML), facilitated by the 'dredge' function from the R package MuMIn (Barton 2013). Parameter estimates were subsequently obtained for the top models (i.e. models with the lowest AICc), fitted with restricted maximum likelihood and corresponding 95% confidence intervals were associated to the mean estimates from 1000 bootstrap iterations ('bootMer' function in r package 'lme4', Table 2). Residual distributions were investigated for normality and homoscedasticity.

Spatial synchrony analysis
Environmental conditions can co-fluctuate over large distances and thereby synchronise primary production. Thus, we analysed spatial synchrony of shrub secondary growth and the contribution of climate to this synchrony. The correlation function 'Sncf ' for spatiotemporal data from the R package 'ncf ' (Bjørnstad 2019) was used to visualize the synchrony in average ring growth chronologies for all independently cross-dated sites (n = 18 sub-sites, since the 10 sub-sites from Semmeldalen were treated separately in this analysis) as a function of distance between them. The 'Sncf ' function uses a non-parametric smoothing spline to continuously model synchrony as a function of distance between sites (formulae in Bjørnstad et al. 1999) with associated 95% bootstrapped confidence envelopes (using 1000 resamples and two degrees of freedom). According to Bjørnstad et al. (1999), the spatial scale of synchrony can be defined as the distance at which the synchrony is no longer significantly different from zero, for example. the distance at which the lower confidence interval of the 'Sncf ' smoothing spline crosses zero (Nieminen 2015, Eberhart-Phillips et al. 2016). The regional average synchrony (ρ), i.e. the average of the pairwise correlations across all distances, was calculated using a non-parametric bootstrap also providing the associated 95% confidence interval (i.e. 2.5% and 97.5% quantiles, Supplementary material Appendix 1). Compared with the non-parametric bootstrap implemented in the 'Sncf ' function, the estimates were similar, but with higher precision (Supplementary material Appendix 2 Table A7).
We calculated the contribution to the average regional synchrony of the main climatic drivers by first accounting for their effect on ring growth using ordinary linear models, for their maximum overlapping time-span in each site. We calculated the spatial scale of synchrony and regional average synchrony (ρ res , Supplementary material Appendix 1) using the residuals from these linear models. Note that since we are using residuals, adding year as random effect in the linear models described above would remove the very thing we are interested in, i.e. the remaining synchrony in these. To evaluate the statistical significance of contribution of the climate variables to the regional average synchrony, we used the 95% confidence interval of the difference in non-parametric bootstrapped replicates (Supplementary material Appendix 1) of average regional synchrony (n = 1000) before (ρ) and after (ρ res ) accounting for the climate variable. We also computed the spatial synchrony of the most important climate variable, based on the six most complete weather station timeseries from 1979 to 2014 (Svalbard airport, Barentsburg, Svea, Ny-Ålesund, Hornsund and Hopen, Supplementary material Appendix 2 Table A1), and fitted a linear regression across distances. Similar analyses of regional spatial synchrony and the contribution of climate variables to this synchrony across Svalbard have been applied in Hansen et al. (2019) [reindeer synchrony] and Peeters et al. (2019) [rain-on-snow synchrony].

Ring growth variation in time and space
All Salix polaris site time-series' overlapped for the period 1989-2014 (except Petuniabukta which overlapped until 2010), and for this period the largest and lowest average ring Table 2. Parameter estimates of the most parsimonious model for the climate analysis of Salix polaris ring growth at the 'Svalbard scale' and at the 'regional scale' (i.e. allowing for among-site heterogeneities). Growth (t − 1) = previous year's ring growth; July temperature (°C) = estimated mean regional July temperature; log(ROS) = ln-transformed estimated total regional precipitation (mm) falling as rain (≥ 1°C) from November through April. Note that because July temperature is an additive effect in the 'regional scale' model, estimates do not differ among sites. Coefficient estimates and 95% confidence intervals are obtained with restricted maximum likelihood.  (Fig. 1, 3), and average first-order autocorrelated of −0.08 [−0.47:0.31] (ranging from −0.5 to 0.4), both calculated at the shrub level. The average raw ring-width of an individual shrub was 49 ± 37 SE, with a maximum value of 220 µm for a shrub from Edgeøya in 2002. Descriptive chronology statistics showed a low within shrub variation (r bar.wt ), but a relatively high between-shrub variation (r bar.bt , Table 1). The EPS-value for sites with limited shrub replication had relatively low values, such as Kapp Linné, yet, no consistent mismatch in alignment with the other Svalbard sites' mean growth time-series was observed (Supplementary material Appendix 2 Fig. A5). We considered all eight chronologies, obtained after our elaborated crossdating procedure, as sufficiently reliable for climate effect analyses. Nevertheless, while the overall pattern of high-and low-growth years may have been captured quite well even for sites with low EPS values, the results, especially from such low-replicate sites, must still be interpreted with care, as the shrubs sampled may not represent the mean growth rate of these sites sufficiently.

Climate effects on ring growth
The correlation between the mean ring growth and mean July temperature over Svalbard was fairly high ( Table A8). Accordingly, the linear mixed-effect model showed that July temperature positively influenced ring growth at the 'Svalbard scale' (for a comparison with non-detrended data, Supplementary material Appendix 2 Table A9) and similarly at the 'regional scale' (i.e. no interaction with site, Table 2). Moreover, all models with ΔAICc < 2 included July temperature (Supplementary material Appendix 2 Table A10).
The top model of the 'regional scale' model selection also contained ROS in interaction with site, meaning that this variable influenced ring growth differently among sites (Table  2). Results from the analysis at the 'local scale' supported that two (i.e. Ny-Ålesund and Kapp Linné) out of four sites with an established weather station close by showed a negative effect of ROS (Supplementary material Appendix 2 Table  A11, Fig. A6). The effect of the previous year's growth differed among sites both in strength and sign (Table 2). Overall in Svalbard, the effect of growth of the previous year was negative ( Table 2). Note that this effect was masked by the autocorrelation in time-series, before temporal linear detrending was performed (Supplementary material Appendix 2 Table  A9). Year explained 9% of the total random variation (number of years n = 52) in both the top model at the 'Svalbard scale' and the 'regional scale'.
Ring growth did not show any consistent response to the detrended large-scale climate proxies. The top-ranked model only included previous year's ring growth, without any of these variables (intercept = −1.68 × 10 -03 ± 0.04, slope = −0.14 ± 0.02, t = −5.91, p ≤ 0.001). Additionally, including any of the climate proxies did not enhance model fit (ΔAICc > 2 for all models). Note also that none of the large-scale climate proxy variables were significantly correlated with July temperature or ROS. Hence, the correlation between summer AO and July temperature (linearly detrended) was r = −0.04 [−0.23:0.31]; the correlation between June sea-ice and July temperature was r = −0.33 [−0.60:0.01]; and the correlation between winter AO and ROS was r = −0.20 [−0.45:0.08].

Synchrony analysis
From the climate effect analysis, July temperature was considered the main potentially synchronising climate variable ( Table 2 A7). The average regional synchrony was ρ T° = 0.49 [0.42:0.56] (mean [95% CI], non-parametric bootstrap, Supplementary material Appendix 1) and the spatial scale of synchrony was estimated to 350 km (Supplementary material Appendix 2 Fig. A7). Ring growth also showed significant spatial synchrony across large distances: the average regional synchrony was ρ = 0.24 [0.21:0.27] (Supplementary material Appendix 1), and the spatial scale of synchrony was estimated to be 183 km (Fig. 4a). Average regional synchrony in model residuals when accounting for the effect of July temperature remained significant but was reduced to ρ res = 0.18 [0.15:0.22] (Supplementary material Appendix 1), and the spatial scale of synchrony was estimated to 159 km (Fig. 4b). Hence, the average regional synchrony after accounting for July temperature was significantly reduced by 0.054 [0.039:0.069] (Fig. 4c). ROS average regional synchrony was ρ ROS = 0.62 [0.48:0.76] (see also Peeters et al. 2019). However, accounting for the effect of ROS (instead of July temperature) did not reduce the regional average synchrony (ρ res = 0.24 [0.21:0.27], i.e. similar estimates as ρ).

Discussion
By applying dendrochronological tools to the high-arctic dwarf shrub Salix polaris, this study has demonstrated how spatial autocorrelation in weather fluctuations synchronises shrub secondary growth over large distances in a hotspot for climate change. We found contrasting effects of different weather variables associated with climate warming during summer versus winter. Summer temperature had an overall strong positive effect on ring growth (Table 2, Fig. 3) and, hence, explained a significant part of the spatial synchrony in secondary growth (Fig. 4). On the contrary, ROS events associated with winter warming caused reduced ring growth, yet only in some relatively wet and mild coastal sites ( Table 2).
The archipelago-wide positive summer temperature effect was expected, since it has been previously indicated to be the major driver of biomass and secondary growth of other vascular plants in several sites across the Arctic, both from field based and remote sensing studies (Forbes et al. 2010, Blok et al. 2011, Macias-Fauria et al. 2012, Van der Wal and Stien 2014, Myers-Smith et al. 2015a, Weijers et al. 2017, 2018a, b, Ackerman et al. 2018. While it is the temperature of a plant's tissue that regulates the plant hormone triggering xylogenesis and ring formation (i.e. auxin) (Wilmking et al. 2012), summer air temperature is one of many proxies for plant temperature that can be used (Körner and Hiltbrunner 2018). The dwarf shrub S. polaris' secondary growth has previously been demonstrated to be strongly influenced by summer temperature in two single sites (i.e. Petuniabukta and Semmeldalen, Buchwal et al. 2013, Le Moullec et al. 2019, and in the current study we found supporting evidence for its summer temperature sensivity across Svalbard. Furthermore, dendrochronological studies performed on another dwarf shrub, Cassiope tetragona, also found a close relationship between growth and summer temperature in several sites located in central and western Spitsbergen (Callaghan et al. 1989, Aanes et al. 2002, Rozema et al. 2009, Weijers et al. 2010, Blok et al. 2015, Milner et al. 2018. Although the correlation between shrub growth and July temperature in our study was fairly high, it was lower Figure 4. Spatial synchrony in Salix polaris ring growth and the contribution of July temperature to this synchrony pattern. (a) Pairwise correlations (ρ) between ring growth chronologies for all sub-sites (n = 18) against distance, and the spatial non-parametric correlation functions (Sncf, thick black line) with 95% confidence interval (thin black lines) for ring growth. Horizontal black dotted line shows zero correlation as reference. (b) Pairwise correlations in residual ring growth (ρ res ) against distance, when accounting for detrended July temperature. (c) The frequency distribution of the non-parameteric bootstrap replicates (n = 1000) of the average regional synchrony in ring growth (blue bars) and of the average regional synchrony in residual ring growth when accounting for detrended July temperature (red bars). The difference between the two frequency distributions was significant with ρ − ρ res = 0.054 [0.039:0.069]. The average regional synchronies were represented by corresponding blue and red dashed lines. than correlations reported in some local studies conducted on C. tetragona on Svalbard (Weijers et al. 2010(Weijers et al. , 2012. This is possibly because of the high individual variation in S. polaris secondary growth (as will be discussed more in-depth later). Also, this study is conducted at an archipelago-wide scale and the weather data across Svalbard cannot capture microclimatic variations in the same way as local studies can (Armbruster et al. 2007, Körner andHiltbrunner 2018).
The Arctic climate is rapidly changing, particularly during the winter season, and weather events such as warm spells and ROS are already increasing in frequency (Rennert et al. 2009, Hansen et al. 2014, Larsen et al. 2014, Moore 2016, AMAP 2017, Bintanja and Andry 2017, Pan et al. 2018, Peeters et al. 2019. Although the most dramatic effects of ROS have been found in large herbivores (Rennert et al. 2009, Hansen et al. 2011, Forbes et al. 2016, Berger et al. 2018, the consequences of ROS for primary production are still not well understood. Weijers et al. (2010) noted that the relation between July temperature and shrub growth was reduced following winters with numerous thawing days (temperature > 0°C, January-April), and Milner et al. (2016) found that basal ice due to experimental ROS had negative effects on shoot survival and flowering in C. tetragona, thereby indirectly promoting growth in surviving shoots. Other field observations and remote sensing indicate broader plant community-level effects of ROS, possibly explaining part of the recently observed 'browning of the Arctic' (Bokhorst et al. 2011, Phoenix and Bjerke 2016, Bjerke et al. 2017). This could result in a reduced positive effect of increasing summer temperatures on primary production (Vickers et al. 2016).
A negative effect of ROS-events on primary production was found at some, but not all, sites. Two of these sites are located along the west coast, on flat plains close to the sea, exposed to milder and rainier winter climate. Such locations facilitate the conditions that frequently cause formation of thick and continuous basal ice after ROS events (Supplementary material Appendix 1 Fig. A8) (Van Pelt et al. 2016, Peeters et al. 2019. Basal ice occurrence and thickness vary according to the small-scale topography and its accumulated snow amount, which can impact individual shrubs differently. Large rain-on-snow amounts could even lead to total ablation of snow/ice cover, although this will mainly occur on very exposed ridges with negligible snow cover. However, Peeters et al. (2019) showed, using similar weather records as used in our study, that these records predicted field-based basal ice measurements well. These site differences and possibly individual differences to ROS exposure potentially introduces spatial heterogeneity that influences the co-fluctuations of primary production. Accordingly, ROS did not contribute to the observed regional synchrony in ring growth in the past. The magnitude and spatial extent of ROS is forecasted to increase (Bintanja and Andry 2017), so that these sites could represent future conditions across larger areas on Svalbard and the Arctic in general, potentially changing the relative importance of different seasons in their influence on spatiotemporal dynamics of shrub growth. However, note that one of the two coastal sites where we detected adverse effects of ROS at the local scale had a fairly low EPS-value and hence we encourage further in-depth investigations into this matter.
The synchronising effect of summer temperature on primary production represents an analogy to the theoretically expected (Moran 1953) and observed (Grenfell et al. 1998, Saether et al. 2007) 'Moran effect' in animal population dynamics. Although this study is the first to conduct such spatial synchrony analysis from dendrochronological data on a relatively large spatial scale in the high-Arctic, a recent study on C. tetragona (Milner et al. 2018) correlated retrospective shrub growth in neighboring valleys (including Semmeldalen), but at a smaller spatial scale. The correlation at the maximum distance (~30 km) corresponded approximatly to our findings at similar distances for S. polaris. At much larger scales, using estimates from NDVI, Defriez and Reuman (2017) demonstrated an effect of temperature on spatial synchrony in vegetation productivity across large parts of the globe, yet excluding high latitudes. Together, these studies strongly indicate a generalizable pattern that fluctuations in weather, notably summer temperature, cause large-scale synchronisation of plant growth dynamics across biomes. To our knowledge, this study is the first to quantify, in situ, the role of weather in generating archipelagowide spatial synchrony in plant growth in the high-Arctic. Important insights from remote sensing studies have previously been gained from analysing spatiotemporal gradients of arctic primary production driven by weather (e.g. land surface temperature) (Raynolds et al. 2008, Forbes et al. 2010, Walker et al. 2012, but no studies exist on spatial synchrony per se. The lack of such studies is likely due to few in situ plant growth time-series at large spatial scales combined with the low resolution of available weather station records at high latitudes. Svalbard is unique within the High Arctic with respect to availability of such data (Nordli et al. 2014, Supplementary material Appendix 2 Table A1). This, combined with the recent development of dendrochronological tools applied to shrubs (Schweingruber and Poschold 2005, Buchwal 2014, Myers-Smith et al. 2015b, enabled us to detect the synchronising role of a single-month weather variable (July temperature). When such in situ weather data are missing, regional weather proxies can be used as an alternative to detect important large-scale patterns of biotic dynamics on the tundra (Aanes et al. 2002, Forchhammer 2017, Macias-Fauria et al. 2017, Weijers et al. 2017, Buchwal et al. 2019). However, it seems that these regional variables were not able to capture the mechanistically important weather variables for ring growth in Svalbard well (Polyakov et al. 2003, Stenseth et al. 2003, Wang and Key 2003. Population synchrony is expected to be lower than that of the environmental drivers (Saether et al. 2007) due to the wide array of microclimates, habitats and associated individual variation. In our case, July temperature had a stronger spatial autocorrelation across Svalbard (ρ T° = 0.49) than ring growth in S. polaris (ρ = 0.24). One reason for this is the high heterogeneity in individual growth (Crawford 2008), caused in part by substantial irregularities such as missing and wedging rings (Buchwal et al. 2013) and measurable as high between-plant variation (r bar.bt ). The observed individual heterogeneity was likely amplified by our restricted sample sizes and possible exposure to grazing. Furthermore, temperature effects at the microhabitat level are important (Scherrer and Körner 2011) and snow cover and soil moisture at such small scales can interact with the effect of summer air temperature on individual growth (Hallinger et al. 2010, Ackerman et al. 2017, Weijers et al. 2017. We also observed heterogeneity in the strength and even sign of delayed growth effects, both between sites and between individuals. This likely reflects different tradeoffs in resource allocation between growth and e.g. damage repair or reproduction, which is a highly individual process (Dormann andVan der Wal 2002).
Nonetheless, the synchronising effect of summer temperature was still clearly detectable, explaining one fourth of the observed regional synchrony of ring growth. The remaining observed synchrony is likely a combination of different climatic variables affecting growth simultaneously (Supplementary material Appendix 2 Table A10, A11), and during other parts of spring and summer than July (Weijers et al. 2018b). Furthermore, the temporal linear detrending we performed is a conservative approach, increasing the confidence with which the results from this work can be reported but may have reduced the strength of the detected climate-growth relationships (Supplementary material Appendix 2 Table A9). Other factors that can influence population synchrony include trophic interactions (Bjørnstad et al. 1999, Liebhold et al. 2004). However, the single most abundant herbivore (in terms of biomass), the Svalbard reindeer, is highly sedentary, and does not migrate between sites (Tyler and Øritsland 1989) and a significant effects of top-down regulation on annual plant growth in this system is not expected , Descamps et al. 2017, Le Moullec et al. 2019.
Bottom-up effects of spatial synchrony in primary production are expected to have large implications for ecosystem-level dynamics through cascading effects (Haynes et al. 2009, Post et al. 2009b, Wookey et al. 2009). The evidence of climate-driven spatial synchrony in S. polaris secondary growth found in the present study, calls for large-scale investigations of spatial synchrony in tundra community dynamics. Secondary growth dynamics of S. polaris from multiple sites also contain valuable information on the spatial synchrony in wood deposited below-ground (shoots are nested into the ground, Le Moullec et al. 2019) and, hence, their contribution to the carbon pool size contained therein (Buchwal et al. 2013, Babst et al. 2014, Iversen et al. 2015. Regionally replicated plot-design experiments, similar to the International Tundra Experiment (ITEX) project Molau 1997, Elmendorf et al. 2012a, b), could test how different vascular plant species cope with rainy winters (i.e. experimentally encapsulated in basal ice), in addition to, and in interaction with, simulated summer warming. Such experiments simulating ROS and basal ice encasement found species-specific effects on some growth and reproductive traits, however, there was also an overall high tolerance to experimental icing (Preece and Phoenix 2014, Milner et al. 2016, Bjerke et al. 2018. To expand our spatial synchrony study to larger parts of the Arctic, in situ dendrochronological data from multiple species can be combined with experimental icing data and remote-sensing vegetation productivity index maps (Forbes et al. 2010, Blok et al. 2011, Macias-Fauria et al. 2012, Weijers et al. 2018b. Furthermore, in areas with ROS data available from local weather stations, alternatively downscaled climate models, the larger-scale implications of (potentially negative) effects of ROS on secondary shrub growth, and how this influences spatial synchrony across environmental gradients (e.g. coast versus inland), could be tested through remote sensing (e.g. observed as 'browning', Bjerke 2016, Bjerke et al. 2017). We believe that such large-scale studies may contribute to a holistic understanding of the role of spatial synchrony in primary production in arctic community-level dynamics in both time and space.