Comment on bg-2021-99 Anonymous Referee # 1 Referee comment on " A new mechanistic understanding of ecophysiological patterns in a widespread alpine dwarf shrub – Refining climate-growth relationships

In this paper authors analysed the influence of environmental factors on radial growth and intra-annual growth patterns of the prostrate dwarf shrub Empetrum nigrum ssp. hermaphroditum, which is widespread in arctic and alpine ecosystems. Elevational gradients spanning 500 m were selected above treeline in a humid and continental climate region in central Norway. To determine key dates of intra-annual growth dendrometers were mounted on one major stem per plant (two regions x six samples x two region, i.e., one sample per elevation result in a total of 12 dendrometer records during four successive years). Regression and correlation analyses were applied to calculate the influence of microclimate on growth parameters. Authors found that soil moisture, solar radiation and winter conditions are the main drivers of radial stem growth of Empetrum nigrum ssp. hermaphroditum. Results of this study are unexpected, as it is generally assumed that in the temperate climatic zone growth processes at and above treeline are primarily limited by temperature during the growing season.


Introduction
Arctic and alpine ecosystems are especially sensitive to recent climate variability, with temperatures increasing twice as much as the global average in the past decades, accompanied by rising values of atmospheric CO 2 and lengthening of the growing period (e.g., Stocker, 2013). This trend has favored growth, abundance, and biomass production of numerous shrub species, resulting in widespread, yet spatially heterogenic, greening of the affected areas with potentially global effects 30 (Myers-Smith et al., 2011;Gough et al., 2015;Brodie et al., 2019;Myers-Smith et al., 2020). The observed greening has been verified using remote sensing techniques (Carlson et al., 2017) and is caused by both evergreen and broadleaved species, although in different ways (Vowles and Björk, 2019;Weijers and Löffler, 2020). In general, shrubs are considered one of the most responsive plant functional groups to climate variability (Elmendorf et al., 2012), and their expanding trend, in turn, has been associated with climatic feedbacks, such as influence on surf ace albedo and frozen-ground processes 35 (Sturm et al, 2001;Chapin et al., 2005;Blok et al., 2011). Therefore, an understanding of shrub growth-physiology and its micro-environmental drivers in these highly sensitive ecosystems is of crucial importance, and dendroecological studies on shrubs have become increasingly important over the past decade.
In recent years, temperatures and soil moisture have been identified as the most important drivers in controlling shrub 40 growth, with the highest effects caused by conditions during the growing season (Elmendorf et al., 2012;Hollesen et al., 2015;Ackerman et al., 2017;Weijers et al., 2017) and within the current year (Van der Wal and Stien, 2014). Additionally, most recent studies have suggested that snow cover and winter warming may play an important role in promoting shrub growth (Hollesen et al., 2015;Weijers et al. 2018b;Francon et al., 2020), as well as spring warming (Weijers et al., 2018b).
Yet, an increased frequency of spring freezing events might counteract these positive effects (Choler, 2018). Collectively,45 these studies agree on the fine-scale complexity of growth behavior, niche shifts, and local adaptation of shrubs in arctic and alpine regions, with a multitude of still little-understood site-related micro-environmental drivers (Graae et al., 2017;Pape and Löffler, 2017;Löffler and Pape, 2020).
High resolution data as provided by dendrometer measurements has the potential to bridge this knowledge-gap and provide 50 valuable insights into fine-scale response mechanisms in a changing environment. In tree physiology and forest sciences, dendrometers have proven to be useful measurement devices for such fine -scale monitoring of tree responses to environmental fluctuations (Breitsprecher and Bethel, 1990;Duchesne et al., 2012;Ježík et al., 2016;Van der Maaten et al., 2018;Smiljanic and Wilmking, 2018), because dendrometers can detect radial stem dimensions at hourly or even shorter intervals (Drew and Downes, 2009;Liu et al., 2018). Starting with the early designs first described in the 1930s and the 55 1940s (Reineke, 1932;Daubenmire, 1945), they have been widely used to monitor tree growth, focusing on long-term monitoring of growth responses to environmental variables (e.g., Duchesne et al., 2012;Liu et al., 2018;Van der Maaten et al., 2018), and recently a study using band-dendrometers to monitor radial stem growth of tree-like shrubs was presented https://doi.org/10.5194/bg-2021-99 Preprint. Discussion started: 10 May 2021 c Author(s) 2021. CC BY 4.0 License. (González-Rodríguez et al., 2017). Because current dendrometers can measure at a micrometer scale, they have the potential to be used on shrubs to provide fine-scale, intra-annual, continuous, and highly comparable information. Past studies, the 60 first of which were published in 2006 and 2007 (Bär et al., 2006;Bär et al., 2007), had to rely solely on sh rub-ring series (Macias-Fauria, et al., 2012;Shetti, 2018;Le Moullec et al., 2019), extraction of micro-cores, and wood anatomical analyses (Rossi et al., 2006;Weijers et al., 2010;Liang et al., 2012;Francon et al., 2020). Because trees and shrubs are both woody plants, such methodical transfers have proven successful in the past (e.g., Bär et al., 2006;Liang et al., 2012;Macias-Fauria et al., 2012). Yet, they may have to be adapted for the special morphology, growth behavior and high internal variab ility of 65 multi-stemmed shrubs (Bär et al., 2007;Buras and Wilmking, 2014;Myers-Smith et al., 2015). However, there are several insights to be gained. Because cambial activity occurs at time scales ranging from hours to days Köcher et al., 2012;Liu et al., 2018), the fine temporal resolution has the potential to provide valuable additional insights compared to traditional methods, including intra-annual and seasonal growth behavior of shrubs in alpine environments, thereby filling in existing knowledge gaps regarding plant productivity in remote ecosystems (Le Moullec et al., 2019). 70 Additionally, the time series derived from dendrometer measurements of the stem radius offer information not only on radial stem growth but also on stem water relationships with higher quality and resolution than previously attainable (Fritts, 1976;Steppe et al., 2015;Zweifel, 2016;González-Rodríguez et al., 2017). Radial increase can be a result of both swelling tissue (caused by induced water) and one-directional growth, and both can contain valuable information on cambial activity and 75 shrub responses to external factors. Thus, it can be considered one of the strengths of this approach that these factors are both visible in the data and can be simultaneously assessed (Drew and Downes, 2009;Chan et al., 2016;Zweifel, 2016).
Monitoring these inter-annual patterns could provide the opportunity to study fine-scale, inter-annual, eco-physiological mechanisms for the first time in shrubs. What has become a widely used practice in dendroecological studies on trees  has the potential to become available for shrubs as well and thus provide additional opportunities to 80 gain novel insights into those plant communities with high relevance for potential global environmental change.
In this context we monitored intra-and inter-annual variability of stem radial variations of dwarf shrubs in an alpine environment, testing this novel approach using high-precision point-dendrometer data derived from dwarf shrubs to 1) explain major growth patterns and their variation between years and specimens, 2) identify the most important micro-85 environmental drivers controlling these patterns, and 3) gain insights into potential response to environmental change. The main objective of our work is thus to gain detailed understanding of the mechanistic adaptation of one common arctic-alpine dwarf shrub (Empetrum nigrum ssp. hermaphroditum) to its micro-environment. With this we hope to bridge the gap between observed large-scale vegetational shifts, and the fine-scale physiological mechanisms driving these complex changes within the highly relevant arctic-alpine ecosystems.

Study sites
We conducted our study in two alpine mountain regions of central Norway. To the west, the Geiranger/Møre og Romsdal region (62°03′N; 7°15′E) is located within the slightly to markedly oceanic climatic section (O1-O2; Moen, 1999) of the inner fjords. It is characterized by humid conditions, with total annual precipitation of 1,500-2,000 mm in the valleys (Aune, 95 1993) and a mean annual ambient air temperature of 1.9 °C (range: -23.2 °C-17.2 °C) (Löffler, 2003). To the east, the Vågåmo/Oppland region (61°53′N; 9°15′E) is located within the continental climatic section (C1; Moen, 1999). The total annual precipitation is low, approximately 300-500 mm in the valleys (Kleiven, 1959) and the mean annual ambient air temperature is -1.2 °C (range: -29.2 °C-16.7 °C) (Löffler, 2003). Our ow n measurements indicated that the annual liquid precipitation was 900 mm in the west and 375 mm in the east. The additional amount of solid precipitation and its snow 100 water equivalent remains unknown, but snowdrift leads to an uneven distribution of the snowpack within the complex alpine topography (Löffler, 2007).
Across both regions, we used exposed alpine ridge sites that were randomly stratified based on elevational gradients within the framework of our long-term alpine ecosystem research project (LTAER; e.g., Löffler and Finch, 2005;Hein et al., 2014;105 Frindte et al., 2019). The elevational gradient was stratified into six elevational levels from the tree line upwards, shifted by 100 m between regions to account for slightly different conditions and the position of the tree line. In the oceanic region, we used 900, 1000, 1100, 1200, 1300, and 1400 m a.s.l. (above sea level). In the continental region, we used 1000, 1100, 1200, 1300, 1400, and 1500 m a.s.l. The micro-topographical position at the ridges likely represented the most extreme thermal regimes, with discontinuous snow cover and deeply frozen ground during winter. Our study design resulted in a total of two 110 regions × six elevational levels = 12 sites, with one specimen per site (N = 12), and 12 × four years = 48 annual dendrometer curves. A summary of total stem diameter variation measured at each site is presented in Fig. A1.

Species and specimens
In this study, we focused on the dwarf shrub species Empetrum nigrum ssp. hermaphroditum (hereafter E. hermaphroditum), which is abundant in the Scandes mountain chain. E. hermaphroditum is a common evergreen shrub that is almost 115 circumpolar in distribution (Bell & Tallis, 1973) and has been identified as a niche constructor species with strong direct effects on tundra communities, including a potential slowing of process rates and lowering of biodiversity with E. hermaphroditum encroachment (Bråthen et al., 2018). Because of its complex response to variation in snow cover, it is most common at positions with either shallow or relatively deep snow (Bienau et al., 2014;Bienau et al., 2016). Additionally, E.
hermaphroditum is comparatively resistant to low winter temperatures (Stushnoff and Junttila, 1986;Ogren, 2001). Löffler 120 and Pape (2020) found its occurrence promoted by temperatures of >15.5 °C in the shoot zone and >0.7 °C in the root zone. Generally, the species occurs in a wide phytogeographic range at various sites along the alpine elevational gradient. Its frequency is high at different micro-topographic positions and at high elevations the species occurs as the only exclusive dwarf shrub between a matrix of debris and graminoids. E. hermaphroditum was the first shrub species for which a chronology was successfully derived using its annual growth rings (Bär et al., 2007). 125

Dendrometric data and monitoring setup
Here, we applied a technological approach, commonly used for trees, to our multi-stemmed specimens of E. hermaphroditum, taking radial stem measurements using point dendrometers. The general idea was to use well-tested methods from dendroecology and tree growth analysis in a novel setting to further assess intra-and inter-annual variation in seasonal growth patterns and micro-environmental drivers. We mounted our dendrometers on one major above-ground stem 130 horizontal to the ground surface on randomly chosen specimens, which were as close to the assumed root collar as possible.
We avoided specific micro-positions near stones and depressions, inside the radius of other larger shrub species, and near patches of wind erosion (Fig. A2). Stem diameter data were measured at 1 min intervals using dendrometers (type DRO; Ecomatik, Dachau/Germany), where the temperature coefficient of the sensor was <0.2 µm. Because we were interested in seasonal patterns, we chose to eliminate the effects of daily short-term stem-size fluctuations from the dendrometer time 135 series, for which there are commonly used approaches . In accordance with our data we chose the "daily mean approach", averaging hourly dendrometer data using the 'dendrometeR' package (Van der Maaten et al., 2016), developed for the R statistical software (R Development Core Team, 2020).

Analysis of seasonal growth patterns
To assess inter-annual differences in seasonal growth and intra-annual growth patterns, we defined critical parameters and 140 dates of stem growth phenology, such as a) total annual growth, defined as growth -induced irreversible stem expansion (GRO), b) peak growth (maximum daily growth rate), c) growth initiation (start of the growing season), d) growth cessation (end of the growing season), and e) peak shrinking during the winter months. To separate GRO from water-related expansion and contraction caused by hydrological processes, we first excluded reversible shrinking and swelling associated wit h the stem water deficit from the original data using the approach proposed by Zweifel (2016). Assuming zero growth during 145 periods of stem shrinkage, the final growth curve was derived from the original measured data by calculating the cumulative maxima (Zweifel et al., 2014;Zweifel, 2016). The respective dates were derived from sigmoid Gompertz models fitted to these estimations. Although multiple models have been used to describe growth, the Gompertz equation is the most widely used in dendrochronological studies and explains the variations in dendrometer measurements (for trees) (e.g., Rossi et al., 2003;Rossi et al., 2006;Duchesne et al., 2012;Van der Maaten et al., 2018;Liu et al., 2019). The equation used for the 150 model was Eq (1): (1) where alpha is the upper asymptote, beta is the x-axis placement parameter, and k is the growth rate. We calculated these input parameters from our original data using the equations defined by Fekedulegn et al. (1999). To assess how we ll the 155 models fit our data, we calculated a goodness-of-fit (GoF) measure using the least-squares method with the formula Eq (2): where f is the original and f2 is the modeled stem radius.

160
We determined growth initiation (onset) and cessation (offset) from this modeled curve to ensure that the main growth phase was being captured. Both dates were defined as the time when 20% and 90% of the total annual modeled growth occurred, respectively. We chose these thresholds following careful testing and, in accordance with our data, they were slightly higher than the thresholds used in similar studies for trees (e.g., Van der Maaten-Theunissen et al., 2013;Van der Maaten et al., 2018;Drew and Downes, 2018). We first calculated these growth-defining parameters for each curve to assess the variability 165 between sampled specimens. Some of our specimens did not experience any growth or irreversible stem expansion in specific years. Similar years with no or little growth have been detected in other dwarf shrubs, for example, in Salix arctica by Polunin (1955). Buchwal et al. (2013) assumed such mechanisms might be related to carbon allocation and could be irregular along the stem because 170 growth is not homogenously allocated within the different plant segments. In accordance with their findings for Salix polaris, the specimens might preferentially allocate resources to less exposed parts (e.g., roots) in these years. In general, such partial dormancy (Preece et al., 2012) might reflect insufficient resources for the homogenous growth of the entire plant. We calculated non-growing seasons for these years, and the analyses proceeded separately, excluding them from most of our calculations. 175

Environmental data collection and growth conditions
To identify the thermal constraints of our species at the critical location, we measured soil temperatures (°C) at a depth of 15 cm below the ground surface within the root zone (hereafter "T RZ ") and the air temperatures at a location 15 cm above the ground surface, which was within the shoot zone (hereafter "T SZ "), at all sites. Temperatures were measured at 1 min intervals and recorded as hourly means using ONSET's HOBO loggers (type H21-002) and type S-TMB-002 temperature 180 sensors (±0.2 °C accuracy). For the T SZ measurements, the sensors were equipped with passively ventilated radiation shields.
Moreover, to identify the soil moisture constraints in the root zone of our specimens, we measured the volumetric soil water content (m³/m³) 15 cm below the soil surface (hereafter SM RZ ) at all sites. The uncalibrated SM RZ was measured at 1 min intervals and recorded as hourly means using type S-SMD-M005 soil moisture sensors (±3% accuracy). Additionally, we https://doi.org/10.5194/bg-2021-99 Preprint. Discussion started: 10 May 2021 c Author(s) 2021. CC BY 4.0 License. measured the shoot zone global radiation (W/m²) at 1 cm above the ground surface (hereafter GR SZ ) using a type S-LIB-185 M003 silicon pyranometer (±10 W/m² accuracy). Our data covered a period of 4 full calendar years from January 1, 2015, to December 31, 2018. Missing data did not occur at the chosen sites. The different near -surface regimes of T RZ , T SZ , SM RZ , and GR SZ at our micro-topographical sites are illustrated in Fig. 1. From these raw data, we calculated a set of variables defining the micro-environmental conditions for each specimen based on the expected effects of different growth mechanisms. Averaged values for all sites (N=12) are summarized in Table 1. 190   In all four observed years, our sampled specimens experienced the highest temperatures during summer (June to September), with a slightly earlier temperature rise in 2018, beginning in May. Maximum temperatures were reached from July to September (Fig. 2). In 2018, the highest mean temperatures were expressed because of exceptionally high summer temperatures (GDD10 = 57), whereas 2015 and 2017 both experienced shorter periods of high temperatures. Temperatures 205 varied slightly among sites (Table 1), and as expected, the shoot and root zone temperature curves were well coupled.
Additionally, we detected slight variability between the two studied regions, but overall simil ar seasonal temperature patterns on the measured micro-scale (Fig. 2), which differs from the regional climate signal described above. Global radiation showed similar patterns at all sites as well, following the course of the astronomic sun angle, with a mid-summer maximum; however, there were large variations according to cloud coverage. As such, 2018 experienced a short period of 210 temperature and radiation decrease during summer (Fig. 1).
We did not explicitly measure data regarding snow cover, but calculated snow cover from the daily shoot zone temperature amplitude and validated those calculations using radiation sensor measurements. We assumed that a daily amplitude of less than 5% of the maximum amplitude reached throughout the year indicated that a layer of snow restricted daily air 215 temperature fluctuations at the measured height of 15 cm. The respective periods were therefore defined as snow -covered.
The period in which our specimens were snow-covered varied considerably between the monitored winters and between sites. The winter of 2015/2016 had comparatively little snow, whereas 2017/2018 was snow-covered for the longest period.
However, because of the chosen ridge positions, most of our monitored sites did not experience long periods of snow cover.
Nonetheless, snow and its presumed effects, such as mitigating extreme negative temperatures by acting as an isolating 220 barrier and reducing the effects of frost (Körner, 2003;Bienau et al., 2014) might play a role in influencing the growth response and was therefore included in our analysis.

Potential micro-environmental drivers of seasonal growth patterns
To relate the collected data and analyze the influence of the micro -environmental conditions on the observed growth patterns, we performed a set of multiple and partial regressions and correlation analyses, calculating partial R squared and Pearson's correlation coefficients. Regressions were performed in R (R Development Core Team, 2020) and tested for 230 multicollinearity among the independent variables that would have affected the regression outcome using several measures of collinearity, which are implemented in the mctest package (Imdad and Aslam, 2018). This included the determinant of the correlation matrix (Cooley and Lohnes, 1971), Farrar test of chi-square for presence of multicollinearity (Farrar and Glauber,  (Kovàcs et al., 2005), Sum of lambda inverse (Chatterjee and Price, 1977) values, Theil's indicator (Theil, 1971), and condition number (Belsley et al., 1980). A regression model was discarded if more than two of the six 235 calculated measures indicated collinearity among the independent variables.

1967), Red indicator
After testing our chosen growth parameters and their relationship to total annual growth by performing a partial regression analysis (Table 2), we analyzed the environment-growth relationship by correlating the growth parameters (total annual growth, growth initiation, peak growth, growth cessation, start of the shrinking period, peak shrinking, and day of year 240 (DOY) at which peak shrinking occurred) with the potential micro-environmental drivers listed in Table 3, using Pearson's correlation coefficient. For each group of potential drivers (means, maxima, days at which the maxima were reached for shoot and root zone temperatures, global radiation, and soil moisture), we performed a multiple regression analysis, predicting each of the growth parameters from a fitted linear regression model. Additionally, we correlated our raw environmental data with the daily growth for each season, performing rolling correlations by aggregating our hourly data and 245 calculating averages for preceding periods, ranging from days to years.

Intra-annual growth patterns
In general, the seasonal variability in the stem diameter was well explained by non-linear, sigmoid regressions (Gompertz 255 curves) with a GoF between 0.90 and 0.94 (Fig. 3), and all specimens experienced distinct growing seasons. Moreover, our data revealed a distinct phase of stem radial contraction (shrinking) following the growing phase toward the end of the year, starting in October, with remarkably little variation in timing between years (on average from the 287 th to the 311 th Julian day). In most cases, the stem radius remained below the previously achieved maximum for the entire winter and started to increase again with the following year´s growing season (Fig. 3). The timing of this shrinking period was significantly linked 260 to the day when peak growth occurred (R = 0.50, p = 0.004), as well as to growth initiation (R = 0.40, p = 0.023) and cessation (R = 0.52, p = 0.0023).   (Table 4 and Fig. 3). Interestingly, year-to-year variability and patterns in total growth were similar in the two studied regions (Fig. A3). Furthermore, our data showed slight differences in seasonal growth patterns between the two regions, but no clear overall patterns beyond the high inter-specimen variability observed in the whole dataset (Fig. A4). Our chosen growth parameters, growth initiation, peak growth, and growth cessation, together explained 93% of the variance in total 275 annual growth, with peak growth having by far the greatest influence. For stem contraction (shrinking), the same parameters explained 79% of the variance (Table 2 and Fig. A5), suggesting that the observed winter shrinking in E. hermaphroditum might be linked to growth during the growing season.

Micro-environmental drivers of growth patterns
We investigated how the specific micro-environmental conditions were correlated with the growth parameters described above using a set of potential micro-environmental drivers for each site. Together, all 25 chosen micro-environmental 285 parameters (Table 3) were able to explain the total annual growth to a very large extent. Separate regression models for each micro-environmental variable revealed an overall high explanatory power for global radiation and soil moisture, on total annual growth and peak growth, compared to temperatures (Fig. 4). However, correlating temporarily aggregated data revealed that all measured environmental parameters influenced stem radial variation, but at different time scales, with radiation taking a longer time to show effects on stem increment. Comparisons of seasons also showed that micro -290 environmental influences had strong explanatory power during spring when the plants experienced their main growing phase. Conversely, the explanatory power was comparatively low during autumn, when stem shrinking was detected (Fig. 5).
The correlations with our micro-environmental parameters (Fig. 6) indicated that a multitude of micro-environmental drivers influenced growth, yet, to a very low degree and with high inter-annual variation. Maximum soil moisture was significantly 295 correlated with total annual growth (R = 0.47). Minimum soil moisture was positively correlated with growth cessation (R = 0.56), as were annual mean soil moisture (R = 0.40) and the sum (R = 0.39), indicating a strong overall influence of soil moisture on growth processes throughout all four seasons. Growth initiation, on the other hand, was linked to spring radiation (R = -0.37) and winter temperatures (R = 0.40). Furthermore, growth cessation was negatively correlated with maximum shoot zone temperatures (R = -0.46), mean summer temperatures (R = -0.40), and GDD5 (R = -0.40). This result 300 suggested that high peak temperatures during the summer might delay growth cessation by the thermal promotion of carbohydrate storage, enabling growth continuity even under unfavorable thermal constraints during autumn (Fig. 6).

Intra-annual growth patterns
In this study, we demonstrated that our focus species Empetrum hermaphroditum displayed distinct annual growth patterns in response to near-ground environmental drivers and in close accordance with distinct conditions caused by local topography. These unique micro-environmental characteristics of the studied ridge positions are a feature of the 325 heterogenous topography that characterizes alpine terrain (Scherrer and Körner, 2011). They include high exposure to global radiation and very little snow cover during the winter months, associated with very low temperatures (Wundram et al., 2010), as well as a layer of lichens, keeping soil moisture and reducing the danger of summer drought (Löffler, 2005). These conditions varied comparatively little between the study regions (Fig. 2), and lead to very similar seasonal growth patterns ( Fig. A4), suggesting that their influence on growth conditions was stronger than prominent regional environmental signals, 330 which were not reflected in our growth data, as suggested by Bär et al. (2008).
Thus, we confirm findings by Bienau et al., (2014), suggesting that E. hermaphroditum had consistent response patterns to micro-environmental drivers and our results endorse the crucial role of topography in determining growth response (Ropars et al., 2015). In contrast to the oceanic-continental gradient, our study showed high intra-plant growth variability (Fig. A1  335 and Fig. A3), which has been previously described in E. hermaphroditum (Bär et al., 2008) and could be a result of the nanoscale of internal growth variability within the multi-stemmed plant itself (Bär et al., 2007).
Our findings regarding seasonally differentiated response to near ground environmental conditions (Fig. 5) highlight the importance of winter conditions for early growth. This indicates that for our sampled evergreen species at the chosen sites, 340 which experienced only short periods of snow cover that otherwise would be likely to influence the growth response, the degree to which photosynthetic activity was energetically effective in synthesizing carbohydrates during the winter months was especially important. Such continued activity was found in E. hermaphroditum, as well as several other evergreen shrub species before (e.g. Bienau et al., 2014;Wyka and Oleksyn, 2014;Blok et al., 2015). Long and severe ground frosts might limit access to soil moisture, and frost-triggered droughts might result in tissue damage caused by an internal water deficit. 345 At the same time low temperatures prevented cell production and differentiation, resulting in a carbon overflow (Körner, 2015;Saccone et al., 2017), which gives E. hermaphroditum the ability to reduce wintertime losses of carbon (Starr and Oberbauer, 2003), start growth activity as soon as liquid water is available in the root zone, and acquire nitrogen early in the season. This ability to use additional photosynthetic opportunities throughout the year is similar to mechanisms found in Juniper thurifera (Gimeno et al., 2012). Global radiation plays a major role here because light can promote photosynthesis, 350 while nutrient uptake is severely restricted (Saccone et al., 2017). Accordingly, we identified winter and spring radiation a s a strong driver, which caused the start of the growth phase (Fig. 5). The ability to preserve resources produced during the winter months would give E. hermaphroditum a local advantage compared to broadleaved species at the same sites.

Total annual growth
Our results showed that total annual growth was largely determined by peak growth, indicating that the overall duration of the growth phase was less important for overall growth than the daily growth rate. Thus, total annual growth can be interpreted as a function of daily growth. Comparing the environmental drivers, annual growth can be best predicted by soil moisture and, to some extent, by global radiation, whereas shoot and root temperatures have minor explanatory power. In 360 inter-annual comparison, this was especially evident in 2018, the year experiencing the highest temperatures. Despite these conditions, the average total growth was considerably lower than in 2016 (17 μm compared to 65 μm). A possible explanation might lie within the spring conditions (March, April, and May) (Fig. 4). In 2018, temperatures rose to comparatively high values in April and early May (Fig. 2), with soil temperatures above 0 °C and consequent thawing processes evident in our soil moisture data. This early warm phase was followed by a short cold snap and ground frost. This 365 would explain the early growth start in 2018, and the absence of high growth rates despite favorable summer conditions, indicating that an early growth start may not be efficient in terms of total growth, if conditions in early summer prevent the survival of the formed cells. For most snow-free ridges, budburst and flowering are not influenced by snowmelt and can, therefore, occur early on, causing high vulnerability to late frost events (Weijers et al., 2018a). This is in accordance with the findings of Choler (2018) and Weijers et al. (2018b), who suggested the strong influence of freezing temperatures in spring, 370 counteracting improved conditions during summer. Here, it is worth noting that the contrasting year, 2016, experienced the highest number of days with soil temperatures above 5 °C and temperatures rose quickly and steadily to that threshold without reaching continuously higher values during the summer (Fig. 1). This could indicate that optimum growth conditions lie within a soil temperature span of 5 to 10 °C, which is in accordance with previously reported temperature thresholds for alpine plant distribution (Körner and Paulsen, 2004;Rossi et al., 2008;Steppe et al., 2015). On the other hand, our regression 375 analysis indicated no direct relationship between total growth and near-surface temperatures and no clear thermal growth limit, suggesting more complex connections, probably influenced by the evident temperature extremes at our chosen sites (Körner and Hiltbrunner, 2018). Thus, in our studied alpine environment, we cannot confirm high temperatures as the main general driver of shrub growth, as was assumed in several previous studies (i.e., Elmendorf et al., 2012;Hollesen et al., 2015;Ackerman et al., 2017;Weijers et al., 2018b). 380 Overall, we found high variability in annual growth between specimens and years, with some specimens experiencing zero growth in more than one year. This occurred in 31 % of the dendrometer curves, mostly in 2017 (15 %). We therefore attributed these dormant years to comparatively long periods of snow cover during the previous winter, which might have prevented E. hermaphroditum from the photosynthetic activity and resource accumulation, and thus, may have limited a 385 crucial precondition of growth success. This assumption was supported by a highly negative correlation between stem diameter variation and the number of snow-free days (R = -0.60, p = 0.024, Fig. A6) during these years. The effects of winter snow cover on shrub growth are a critical topic in arctic and alpine ecology, with findings ranging from positive (Blok et al., 2015;Addis and Bret-Harte, 2019) to negative (Schmidt et al., 2010) growth responses, depending on snow depth and vegetation type. In accordance with Buchwal et al., (2013), we assumed that during years of no apparent radial growth, 390 dwarf shrubs might invest in root growth instead of shoot growth to prepare for the following winter. This ability to reduce cambial activity to a minimum and cease above-ground wood formation is a trait common among woody plants (Wilmking et al., 2012). However, this implies that our studied specimens were locally adapted to snow-free conditions at their exposed positions, and consequently, might not be able to cope with such unexpected growth conditions.

Peak growth 395
We found peak growth, the maximum daily growth rate, was closely linked to soil moisture and usually occurred in connection with the soil moisture maximum, highlighting the overall importance of the root zone soil moisture as the key driver of growth in E. hermaphroditum (Fig. 6). This was evident in the strong contraction and expansion patterns of the stems, most likely controlled by active or passive water level variability within the plant, linked to the extreme thawing and freezing processes prominent at our study sites. The predominant role of peak growth in controlling total growth suggested 400 that the shrubs were usually able to invest in new cells following the cell water level rise caused by thawing conditions, most likely affected by prior carbohydrate storage (see above).
In trees, the timing of maximum growth in cold environments has been linked to day length (Rossi et al., 2006;Duchesne et al., 2012). This cannot be confirmed for our monitored shrubs because of the high variabilit y between specimens, showing a 405 far broader range in the Julian day at which peak growth occurred than observed in trees. Therefore, peak growth is most likely controlled by other factors in shrubs than those assumed for trees, where available soil moisture is not limiting during the photosynthetically active period in spring and early summer. Instead, our results suggest that soil moisture availability played a key role.

Growth initiation and cessation 410
We found the overall link of growth initiation and cessation to the micro-environment comparably low, indicating that growth duration was most likely influenced by a multitude of environmental variables with large differences between years and sites ( Fig. 4 and Fig. 6). From the positive correlation of growth initiation with global radiation during spring we concluded that growth initiation might be driven by the constantly increasing radiation with the astronomic rise of the angle of the sun. At this time of the year, energy transfer from global radiation into thermal heat was low, but radiation was high enough for photosynthetic activity, which might explain the decoupling of thermal and radiation drivers of growth initiation.
Furthermore, growth initiation in E. hermaphroditum was positively linked to winter temperatures in our study (Fig. 6), indicating that low winter temperatures were correlated with an early start of the growing season, in contrast to common assumptions relating high late winter temperatures to an early growth start (Dolezal et al., 2020). This highlights the influence of high radiation on energy storage during periods of an absent snow cover, which is usually accompanied by low 420 temperatures, whereas mild winters are often associated with cloudy, humid weath er, and snow cover on the ridges. For growth cessation, decreasing day length was determined an unlikely trigger (Heide, 1985), because of high inter-stem variability. Instead, shoot zone temperatures played a role in determining when xylogenesis ceased, but a critical temperature threshold, as present in many trees Rossi et al., 2008;Deslauriers et al., 2008) and found for xylem growth of alpine rhododendron shrubs (Li et al., 2016), could not be determined for E. hermaphroditum. Instead, we 425 identified soil moisture as the main driver for the end of the growing season.Yet, we could not determine a minimum soil moisture threshold, which would lead to growth cessation. A contrasting pattern was shown in 2016, with a prolonged growing phase after dry conditions during early summer. Thus, we concluded that E. hermaphroditum growth was strongly dependent on the availability of water, which is why dry conditions during spring and early summer were less crucial and did not lead to immediate growth cessation, because evaporation was not fully active at these times of the year leaving enough 430 exploitable moisture available. This accounts for the species dependency on a damp climate and high rainfall (Bell and Tallis, 1973). In a close relationship with the end of growth, the stem diameter started to shrink, marking the beginning of the winter shrinking period.

Stem shrinking
The phase of stem shrinking during the winter months is usually not present in trees and might, therefore, be interpreted as a 435 unique feature of shrub growth, which we documented for the first time in this study. The reason why this mechanism has not been described earlier might be attributable to the methods for shrub growth measurements used in the past, which were insufficient to document intra-annual variability at the appropriate scale. High-precision point dendrometers can reveal these patterns in growth variability, demonstrating the large amount of additional information gained from this method compared to traditional measuring methods. 440 In trees, radial stem shrinkage has been related to sap flow and tree water content (Zweifel et al., 2006;Tian et al., 2019) . As our observed shrinking phase occurred during periods of extremely low temperatures during winter and was negatively correlated with root zone temperatures during these months, we assumed a relationship with the high subzero temperatures and consequent freezing conditions present at our study sites. As such, stem shrinkage could be interpreted as a result of 445 freezing processes causing living cell shrinkage because of water losses, as commonly observed in trees and other woody plants (Neuner, 2014;Charra-Vaskou et al., 2016). As subzero temperatures are a major environmental stress factor, alpine shrubs have developed a strategy to avoid frost damage, especially where the protective effects of snow cover are missing (Kuprian et al., 2014;Neuner, 2014). We assume that E. hermaphroditum similarly uses cell dehydration to actively protect living cells from the consequences of freezing (e.g., ice nucleation), causing the radial stem contraction evident in our data. 450 The start of the shrinking period was linked to the day when peak growth occured, but finding a singular event causing the shrinking process to start proved difficult without significant connections with the first frost events or autumnal soil moisture declines. We concluded that surviving during extreme winters was the main principle governing E. hermaphroditum growth when existing at alpine ridge positions, causing unique adaptations to local micro-site conditions.

Conclusions 455
This study showed that high-precision dendrometers are suitable measuring instruments for identifying annual growth patterns in dwarf shrubs. For our focal species Empetrum hermaphroditum, the method yielded several novel insights into the phenology and growth physiology. From our analysis, we concluded that as an evergreen shrub at exposed and therefore mostly snow-free positions, E. hermaphroditum appeared capable of continuing photosynthetic activity throughout the year and to thus aggregate resources for use in early cell formation, as observed in other evergreen plants (Wyka and Oleksyn, 460 2014). This provided a competitive advantage over deciduous species in the same habitat, limiting the risk of losing resources through competition. To sustain the continued metabolism throughout the year, we found that E. hermaphroditum was highly dependent on available moisture, and thus, highly adapted to local microsite conditions. Furthermore, our findings confirmed the positive effects of temperatures on shoot growth in E. hermaphroditum to some extent (Chapin and Shaver, 1985;Shevtsova et al., 1997;Bråthen et al., 2018), yet, a clear link between near-surface thermal conditions and 465 growth was lacking, with the overall growth mechanism defined by moisture availability and solar radiation. Hence, temperatures mainly play a role in freezing and thawing processes, on an intra-annual scale. We can thus confirm that while there is a link between shrub growth and warming conditions, it is most likely not uniform and highly variable over spatial and temporal scales (Elemendorf et al., 2012).

470
Overall, the fine-scale data provided by dendrometer measurements proved highly important, since it allowed for a detailed growth analysis, showing a growth mechanism, which is highly adapted to the local micro-environmental conditions at our studied exposed ridge positions, explaining the wide distribution and competitive ability of the species at these sites (Bienau et al., 2014;Bienau et al., 2016;Löffler and Pape, 2020) and highlighted the prominent role of the micro -environment in controlling growth processes (Zellweger et al., 2020). In a changing climatic regime, this might become a disadvantage, 475 complicating adaptation to warming winters and longer snow-covered periods coupled with prolonged dry periods during summer (Hollesen et al., 2015;Weijers, Pape et al., 2018), which might cause early growth cessation. In accordance with previous studies (Milner et al., 2016;Virtanen et al., 2016;Wheeler et al., 2016;Saccone et al., 2017), our results suggested that winter conditions and altered snow regimes represented one of the most serious threats to evergreen shrub growth in tundra ecosystems. We conclude that because of the high local adaptation and dependency on specific winter radiation 480 conditions and soil moisture availability of this species, E. hermaphroditum will not be able to persist at exposed positions in a changing climate or respond with longer periods of dormancy to warming conditions. This will potentially promote the spread of competing deciduous species and thus contribute to the arctic-alpine greening trend.

Data availability
All underlying data and statistical codes pertinent to the results presented in this publication will be made available in a data 485 publication DOI in "ERDKUNDE---Archive for Scientific Geography" (https://www.erdkunde.uni-bonn.de).

Author contribution
JL had the idea, designed the research platform, conducted the field work, and together with RP ran the long -term project.
SD analyzed the data, lead the writing of the manuscript and arranged the figures, with contributions from RP and JL.

Competing interests 490
The authors declare that they have no conflict of interest.