Living, dead, and absent trees—How do moth outbreaks shape small‐scale patterns of soil organic matter stocks and dynamics at the Subarctic mountain birch treeline?

Mountain birch forests (Betula pubescens Ehrh. ssp. czerepanovii) at the subarctic treeline not only benefit from global warming, but are also increasingly affected by caterpillar outbreaks from foliage‐feeding geometrid moths. Both of these factors have unknown consequences on soil organic carbon (SOC) stocks and biogeochemical cycles. We measured SOC stocks down to the bedrock under living trees and under two stages of dead trees (12 and 55 years since moth outbreak) and treeless tundra in northern Finland. We also measured in‐situ soil respiration, potential SOC decomposability, biological (enzyme activities and microbial biomass), and chemical (N, mineral N, and pH) soil properties. SOC stocks were significantly higher under living trees (4.1 ± 2.1 kg m²) than in the treeless tundra (2.4 ± 0.6 kg m²), and remained at an elevated level even 12 (3.7 ± 1.7 kg m²) and 55 years (4.9 ± 3.0 kg m²) after tree death. Effects of tree status on SOC stocks decreased with increasing distance from the tree and with increasing depth, that is, a significant effect of tree status was found in the organic layer, but not in mineral soil. Soil under living trees was characterized by higher mineral N contents, microbial biomass, microbial activity, and soil respiration compared with the treeless tundra; soils under dead trees were intermediate between these two. The results suggest accelerated organic matter turnover under living trees but a positive net effect on SOC stocks. Slowed organic matter turnover and continuous supply of deadwood may explain why SOC stocks remained elevated under dead trees, despite the heavy decrease in aboveground C stocks. We conclude that the increased occurrence of moth damage with climate change would have minor effects on SOC stocks, but ultimately decrease ecosystem C stocks (49% within 55 years in this area), if the mountain birch forests will not be able to recover from the outbreaks.


| INTRODUC TI ON
Mountain birch (Betula pubescens Ehrh. ssp. czerepanovii) is the dominant tree species at the subarctic treeline of Fennoscandia, and significantly contributes to the CO 2 sink capacity of subarctic landscapes (Christensen et al., 2007). Due to climate warming and increased nitrogen (N) deposition, an advance of the treeline has been observed in recent decades (Hofgaard et al., 2012;Rundqvist et al., 2011;Tømmervik et al., 2004). Global warming has also led to a drastic increase in insect herbivory over the past decades Jepsen et al., 2007;Neuvonen et al., 1999). Insect herbivory has been shown to have immense impacts on plant growth and carbon (C) fluxes, even at background intensities (Silfver et al., 2020), but outbreak intensities in particular cause large natural disturbances. For instance, Heliasz et al. (2011) reported an 89% reduction of the C sink strength upon defoliation of mountain birch by larvae of autumn and winter moths. In Fennoscandia, outbreaks of caterpillars from foliage-feeding geometrid moths (Epirrita autumnata and Operophtera brumata) occur at intervals of around 10 years.
While mountain birch can compensate defoliation to some extent (Huttunen et al., 2012), intense damage in such nutrient-poor soils and in cold climatic conditions often leads to forest dieback, leaving large areas of deadwood behind (Tenow, 1996). Considering that 5,000 km 2 of mountain birch forest were affected in the 1960s in northern Finland alone (Nuorteva, 1963), and 10,600 km² during the 2000s in northern Fennoscandia (Jepsen et al., 2009), this issue is of quantitative relevance. High intensity summer grazing by reindeer compounds the issue by impeding the growth of new birch seedlings and basal sprouts, especially in Finnish Lapland (Biuw et al., 2014;Kumpula et al., 2011;Lehtonen & Heikkinen, 1995). The combination of grazing and moth damage induces a conversion of former forests into treeless tundra vegetation, resulting in immense reductions of tree primary production (Olsson et al., 2017), and thus, CO 2 sequestration potential.
The consequences of treeline advance or forest dieback on aboveground C storage are straightforward and frequently studied (Dahl et al., 2017;Russell et al., 2015); however, the consequences on soil organic carbon (SOC) stocks are rather uncertain. Several studies have shown that an advance of the treeline may deplete SOC stocks Hartley et al., 2012;Wilmking et al., 2006) despite C input via litter and roots. For instance, Parker et al. (2015) showed that SOC stocks of organic layers were considerably lower in forested plots (2.04 ± 0.25 kg m 2 ) than at heath plots (7.03 ± 0.79 kg m 2 ). This may be explained by a phenomenon termed the "priming effect" (Fontaine et al., 2004;Kuzyakov et al., 2000): a labile C supply from root exudates or fresh litter can stimulate microbial decomposition of older soil organic matter (SOM). In this context, it is assumed that microbes increasingly decompose SOM to acquire N from it ("microbial N-mining"; Craine et al., 2007;Moorhead & Sinsabaugh, 2006), which is particularly relevant in N-poor subarctic soils (Hicks et al., 2020).
Reversing this finding, it may be speculated that despite the decrease in tree primary production, forest dieback caused by moth outbreaks can induce an increase of SOC stocks, for at least five reasons. First, root exudation decreases, which then causes a decrease in labile C that can stimulate priming. Second, insect herbivory induces an abrupt return of available N and other nutrients to soils via frass (Kaukonen et al., 2013;Parker et al., 2017). Together with the lower N demand of dead trees, the limits on N may be released, hence N-mining and associated priming become unnecessary in soil that is no longer N-limited. Third, large amounts of organic matter are returned to the soil via decomposing roots and deadwood (Kosunen et al., 2020). Since SOM decomposes slowly in cold climates, and parts of SOM are recalcitrant or stabilized and can remain in soil for hundreds or thousands of years (Lehmann & Kleber, 2015;Schmidt et al., 2011), elevated SOC stocks may persist in the long term. Fourth, temporal patterns of deadwood decay usually follow a negative exponential curve (Mackensen & Bauhus, 2003), that is, SOM turnover rates may slow down once the most labile compounds of decaying wood have been mineralized. Fifth, insect herbivory has been found to change vegetation composition from shrub-dominated to grass-dominated plant communities (Karlsen et al., 2013), possibly associated with larger organic matter input into soil. Indeed, reduced soil CO 2 efflux has been observed immediately after a defoliation event (Parker et al., 2017) and 8 years thereafter (Sandén et al., 2020), suggesting slowed biochemical cycling. However, despite decreased turnover rates, Sandén et al. (2020) found no difference in SOC stocks 8 years after a defoliation event of subarctic mountain birch in northern Fennoscandia.
Interestingly, the opposite effect (i.e., higher resource turnover rates in the short term) have been observed after moth outbreaks (Kaukonen et al., 2013;Kristensen et al., 2018), mainly linked to increasing concentrations of soil mineral N and labile C from moth frass. According to Kaukonen et al. (2013), moth outbreaks represent a shortcut in C and N input to soils, which bypasses the main routes of carbon from plants to the soil via mycorrhizal and litterdecomposing fungi. However, it remains uncertain whether this effect persists in the long term. From a study on coniferous forests, Štursová et al. (2014) speculated that the rate of decomposition decreases as soon as the one-time litter input had been processed.
Similarly, Kristensen et al. (2018) assumed that such increased turnover rates are mainly short-lived effects, while lower turnover rates can be expected in the long term. Overall, the temporal effects of moth outbreaks and associated tree death on SOM stocks and biochemical cycling are poorly understood. In particular, the long-term effects are not well quantified.
Former studies on the effect of treeline advance or retreat suffer from the drawback that forested sites are compared with separate tundra sites, or forested sites with defoliated sites, respectively (Hartley et al., 2012;Wilmking et al., 2006). Such space-for-time substitutions are prone to confounding effects, such as differences in temperature, nutrients, or water availability between sites. In this context, it remains unknown whether specific factors (e.g., microclimate) that trigger tree growth, moth outbreaks, or tree survival can also affect SOC turnover rates, that is, a causal link between trees and lower SOC stocks is not certain. This drawback can be overcome by investigating differences between living trees, dead trees, and treeless tundra on a small spatial scale of individual trees, thereby reducing the risk of confounding factors. On a small scale,  identified decreasing SOC stocks with distance from the tree, suggesting accumulation of organic matter close to trees, which opposes the conclusions based on comparisons of separate sites.
Here, we conducted a small-scale space-for-time approach on three independent sites by comparing SOC stocks and biochemical properties of soil under four tree statuses: living trees, dead trees (12 years since moth outbreak), dead trees (55 years since moth outbreak), and treeless tundra, which co-exist at the study sites. We hypothesized that (a) living trees have lower C stocks compared to treeless tundra. Such a negative net effect of a living tree could be caused by its N-demand, which stimulates SOM turnover to such an extent that it exceeds C input from the tree. Furthermore, because it is generally assumed that SOC decomposition slows down in the absence of trees and root exudates, we hypothesized (b) that tree death causes C stocks to develop toward the C stocks of the treeless tundra.

| Study area
The study area (69°9'N, 27°9'E, 130-156 m a.s.l.) is located in Northern Finnish Lapland, 12 km south of the village Nuorgam, approximately 5 km west of Lake Pulmankijärvi (Figure 1). The mean annual temperature was −1.6°C and mean annual precipitation was 448 mm during the time period 1981-2010, measured in Utsjoki, 35 km west of the study site (Finnish Meteorological Institute, 2020). Soils in this region are Podzols on sandy texture (WRB, 2015) and are not affected by permafrost.
The area is a heavily grazed year-round pasture of the Kaldoaivi reindeer herding cooperative. The herding of such semidomesticated reindeer by the Sámi people has been practiced since centuries in this area (Kortesalmi, 2008). Reindeer densities in the Kaldoaivi area are around 2.38 reindeer/km² Biuw et al., 2014). This density remained on average at a same level since the 1950s with the exception of two short-term drops in the late 1960s and mid 1970s and a short-term peak during the 1980s with 4.5 reindeer/km² (Biuw et al., 2014; statistics from the Reindeer Herders Association for reindeer numbers during 1960-2020). The trends in reindeer population densities are quite similar in the neighboring areas in Finland and in Norway, although the overall densities in Norway are slightly lower (i.e., 1.56 reindeer/km² in Biuw et al., 2014). Traditionally, reindeer migrated between summer pastures in the coastal areas of the Arctic Ocean in Norway and winter pastures in the interior areas of Finland and Norway. Seasonal migration is still practiced in the neighboring areas in Norway. The contrasting year-round grazing in northernmost Finland results from the fact that seasonal reindeer migration is prevented by law since 1852 (Kortesalmi, 2008;Stark et al., 2021). Although rotational grazing on designated summer and winter pastures has been realized by means of fences in several Finnish districts, no such measures apply to the Kaldoaivi district (Biuw et al., 2014).
During the last century, the study area has been affected by two severe moth outbreak periods. The first occurred between 1960 and 1965, that is, approximately 55 years before sampling ("dead 55 "); the second was between 2006 and 2008, that is, approximately 12 years before sampling ("dead 12 "), as previously described (Kaukonen et al., 2013;Lehtonen & Heikkinen, 1995;Saravesi et al., 2015). Since some trees survived the respective outbreaks, the area is characterized by a mosaic that includes living trees (Figure 2a), dead but still standing trees (dead 12 , Figure 2b), largely decomposed dead trees with only a stump remaining (dead 55 , Figure 2c), and treeless tundra vegetation ( Figure 2d).
We conducted a small-scale space-for-time substitution approach to investigate the effect of living trees, dead trees, and treeless tundra on SOC stocks and dynamics. Three independent sites were selected, each approximately 1 ha in size and located approximately 2 km away from each other. Criteria for selecting sites were that (a) all tree statuses (i.e., living, dead 12 , and dead 55 , treeless tundra) were present, each with several randomly distributed individuals, (b) the altitude of the sites was similar, and (c) the sites were located on a flat area to minimize differences in climate and microclimate between and within the sites. At each site, we selected four plots of each of the four tree statuses. Each plot had a size of 3.14 m², that is, 1 m radius around the tree or plot center, in the case of treeless tundra.
Plots were chosen such that the next living or dead tree was at least 2 meters away.

| Understory vegetation
The percentage of understory vegetation cover was assessed at the species level by visual estimation within a circle of 1 m radius (Fenner, 1997). A few species were unidentifiable; these were grouped into the category of either "other grass species" or "other lichen species."

| Soil temperature and moisture
Between the beginning of July and end of September 2019, we measured soil temperature and volumetric water content (VWC) four times, parallel to the measurements of soil +understory respiration (see section 2.7). Soil temperature and VWC were measured at a depth of 5 cm. The VWC was measured at 5 locations around the respiration collar using a TDR sensor.
At each site, 2 plots per tree status were selected for temperature logging. Temperature loggers (Maxim iButton miniature temperature loggers, DS1922L) were buried at 10 cm depth and at 50 cm from the tree or plot center, in the case of treeless tundra.
Temperature loggers were inserted at the end of July 2019, and remained in the soil for almost 1 year (with 13 days missing in July 2020) in plastic bags to prevent water damage. Temperature was recorded 4 times per day. The monthly mean temperature and the mean annual temperature were calculated for each plot.

| Aboveground and belowground plant carbon stocks
For living and dead 12 trees, we measured the height of the trunk and the diameter at breast height (DBH, at 1.3 m above ground).
In case of bifurcation within the first 1.3 m, the DBH and height of each trunk were recorded separately. Using this data, we calculated aboveground biomass according to Starr et al. (1998).
For dead 55 trees, the height and diameter of the remaining stump were measured. Biomass of dead 55 trees was estimated according to Equation 1, where density was assumed to be ρ = 478 kg m 3 , as reported for living trees of B. pubescens (Heräjärvi, 2004).
Since the abovementioned equation applies to living trees, we corrected the biomass for density loss using annual decay rate constants of 4.5% upon tree death, which has been demonstrated for birch trees (Krankina & Harmon, 1995). Tree roots were estimated to be 29.2% of total tree biomass as suggested by Smith et al. (2016), that is, by dividing aboveground biomass by 70.8 and multiplying by 29.2. The proportion of carbon in all plant biomass pools was assumed to be 50% (Köster et al., 2015).   Figure 1f), and combined these to a composite sample per plot. Understory roots were isolated from the organic layer by sieving to 2 mm and subsequent washing. Such data were extrapolated to an area of 1 m radius around the plot center. C content was assumed to account for 50% of the biomass. The mineral soil contained no visible roots from understory vegetation.

| Determination of SOC stocks
To determine SOC stocks ("stock sample set," Figure 1f), 10 auger samples (2.5 cm diameter) were taken from 25, 50, 75, and 100 cm distance around the stem, or around the plot center in the case of treeless tundra (black dots in Figure 1f). The organic layer, the eluviated E horizon, and the illuviated B horizon up to a depth at which the auger could not be inserted any deeper (assuming that bedrock or large rocks were reached; cf. Parker et al., 2015) were taken separately. The thickness of each layer and the maximum depth of mineral soil were recorded for each of the 40 sampling points per plot. The auger samples were combined to form one composite sample from each distance and horizon per plot. Samples were dried at 40° immediately after sampling, subsequently sieved to 4 mm (organic) and 2 mm (mineral), and milled. The C and N contents were measured with VarioMax CN analyser (Germany). Since soils in this area are not calcareous (Hinneri et al., 1975) and our samples had a pH <6, the measured C contents were assumed to equal SOC.
Bulk density of the organic layer was measured at 3 locations 50 cm from each tree or plot center (white circles in Figure 1e) by drilling a sharp root auger (5 cm diameter) into the organic layer, measuring its thickness and resultant volume, and drying at 105°C. There was no obvious compression of the organic layer using the root auger. Bulk density of the mineral soil was determined by taking 1-2 soil cylinders Hence, calculated stocks represent total stocks, but do not refer to a standardized depth. The maximum sampled depth was 70 cm of mineral soil, but such depths were only reached occasionally and not consistently within a given plot or distance around a tree. The average depth of mineral soil across all sites and plots was 6 cm.
Stocks of total soil N were calculated analogously to SOC stocks.

| Soil biological and chemical properties
For soil properties other than C and N stocks, we took samples 50 cm from the tree or plot center ("process sample set," small grey dots in Figure 1f). According to Parker et al. (2017), this is the distance at which the effect of tree status may be the greatest. At least 5 samples were taken from each plot using an auger (5 cm diameter).
Additional samples were taken if collected soil material was not sufficient. We sampled the entire organic layer and the mineral soil up to 10 cm depth (i.e., composite of E and B layer). Samples were combined to form one composite sample per plot and soil layer (organic, mineral). Samples were kept in the fridge and sieved to 4 mm (organic) and 2 mm (mineral) within 1 week after sampling. Roots were removed, and samples were then split into separate bags and either dried (40°C), stored cold (8°C), or kept frozen (−20°).
Soil C and N contents were also measured as described above from dried soil for the process sample set.
Soil pH was measured from 1:2 soil-water slurry with a WTW electrode (Xylem Analytics, Germany).
Mineral N in the form of nitrate (NO 3 − ) and ammonium (NH 4 + ) was determined according to Carter and Gregorich (2008) from frozen soils after allowing the soil the thaw for 2 days. Mineral N was extracted from ~30 ml of soil with 100 ml 1 M KCl solution and analyzed by spectrophotometry (PerkinElmer-1420 VICTOR 3 ™, United States) according to Miranda et al. (2001). Mineral N content was calculated per g dry weight of soil.
For laboratory-based respiration measurements, frozen soils were thawed at 5°C for 2 days. An amount of soil corresponding to 8 g (organic) or 30 g (mineral soil) of dry weight was filled into incubation vessels, two replicates per sample, and subsequently rewetted to 50% of water-holding capacity (WHC) with distilled water, if necessary. Soil was pre-incubated at 20°C for 5 days. Basal respiration was then measured for 8 days, expressed as the average hourly CO 2 production per g soil or per g SOC, respectively. After measurement of basal respiration, glucose was added to each vessel at 8 mg glucose g −1 dry soil in aqueous solution (3.5 ml per vessel for organic layer, and 1.5 ml per vessel for mineral soil)-but not exceeding 60% WHC of soil. After the addition of glucose, soil respiration was measured for 100 h. This so-called substrate induced respiration (SIR) was originally developed to estimate the amount of microbial biomass in soils (Anderson & Domsch, 1978), but has subsequently often been used to assess microbial activities and characteristics (Blagodatsky et al., 2000). Here, we used it to determine microbial biomass carbon and investigate the degree of nutrient deficiency. According to Nordgren (1992) If such N addition evokes an increase in CO 2 release, this would suggest that N was the most limiting nutrient and hindered the complete mineralization of glucose. We continued measuring CO 2 evolution for 2 days. The CO 2 production was measured using an automated respirometer that allows incubating 95 samples in parallel (Respicond, Nordgren Innovations AB, Sweden). The system provides a continuous measurement of CO 2 evolution by trapping CO 2 in potassium hydroxide (KOH) (Nordgren, 1988). Decrease in electrical conductivity in KOH solution caused by CO 2 entrapment was automatically measured every 60 min (basal respiration) or 30 min (after glucose addition) by platinum electrodes, and the changes in conductivity were automatically transformed to evolution rates based on Equation (3), where A is a conductivity constant that depends on the molarity of the KOH solution, C t0 is the conductance of the fresh KOH measured at the beginning of the incubation time, and C t1 is the conductance at time t.
Enzymatic activities were expressed as nmol of DOPA g −1 soil DW h −1 .

| Soil +understory respiration
At a distance of 50 cm from each tree or plot center, we drilled a PVC collar (22 cm diameter) into the soil one week before the first measurement, by cutting roots and the organic layer with a knife, if necessary.
We measured soil +understory respiration four times during July-September 2019. soil +understory respiration in this study is defined as the sum of microbial, root, and understory vegetation respiration. We define the results as soil +understory respiration-not as ecosystem respiration-because aboveground respiration of trees was excluded in the measurements.
We placed a dark chamber (20 cm diameter, 25 cm height) on top of the collar, equipped with a fan to ensure ventilation. After closing the chamber, CO 2 concentration (ppm) was allowed to equilibrate for 1 min. The CO 2 concentration within the chamber was then recorded every minute for 5 min by a Vaisala GMP252 probe (Vaisala Oyj, Finland). The CO 2 accumulation rate (ppm min −1 ) was calculated using a linear part of the recorded data, and CO 2 efflux F was calculated by Equation 5 where dC dt is the change in CO 2 concentration across measurement time (ppm min −1 ), P is atmospheric pressure (in hPa, available from the nearest meteorological station in Nuorgam and adjusted for the ~100 m difference in elevation, Finnish Meteorological Institute, 2020), R is the gas constant (8.314 J mol −1 K −1 ), T is the measurement temperature in Kelvin inside the chamber, V is the chamber volume in m 3 (0.009841 m³), and A is the chamber area in m² (0.039761 m²).
Respiration measurements of each site were completed within 2 h such that the temperature and moisture conditions were similar for all plots within a site. The order of plots per site was randomized.
Measurements of the three sites were taken within one day, except for the last measurement date when the duration of daylight was not sufficient.

| Statistics
We consider site as a block factor in ANOVA or random effect in linear mixed-effect models, respectively. We present treatmentspecific mean values of the three sites in our figures. However, since the tree status appears to have different effects on soil properties at the three independent sites, we also discuss some site-specific differences, which are shown in the Supplementary material.
To investigate the effect of tree status on C stocks, C contents, and organic layer thickness, we used linear mixed-effect models for each soil layer (organic, E, B) separately, with tree status and distance to the tree as fixed effects and plot ID nested in site as a random effect. We also tested for interactions. If the tree status effect was significant, Tukey's honest significance test was applied to further investigate differences between each tree status. Data were checked for normality of residuals by visual inspection of qqplots, and variance homogeneity was tested with a Levene test.
If assumptions were violated, the model was repeated on log-or square-transformed data. As this did not change the overall result, and because linear mixed-effect models have been shown to be robust to moderate violations of distributional assumptions (Schielzeth et al., 2020), we decided to keep the original model.
To compare the effect of tree status on overall SOC stocks within a 1 m radius around the tree or plot center, we multiplied the SOC stock (kg m²) at each distance with the area of the respective distance radius and summed them. This was done for each soil layer separately, as well as for total SOC stocks. We performed a one-way ANOVA with block design (site as block) and interaction term (tree status × site) for each soil layer separately (i.e., organic layer, E horizon, B horizon), as well as for total SOC stocks. Data were checked for normality of residuals with a Shapiro-Wilk test, and variance homogeneity was tested with a Levene test. If data were not normally distributed or did not reveal homogeneity of variances, they were log-or square-transformed.
To test for differences in soil +understory respiration between tree statuses, we performed a linear mixed-effect model with tree status, measurement date, soil temperature, and soil moisture as fixed effects, and plot ID nested in site as a random effect. We also tested for interactions. Non-significant interactions terms were removed from the model in a stepwise manner. Tukey's honest significance test was applied to further investigate differences between each tree status.
To test for differences in biochemical soil properties between tree statuses, we performed a one-way ANOVA with block design (site as block) for each soil layer (organic, mineral) separately. If effects were significant, a post-hoc Tukey's honest significance test was applied. If data were not normally distributed or did not reveal homogeneity of variances, they were log-or square-transformed.
Correlation analyses were conducted to explore relationships between the variables. In case of linear relationships and normal distribution of data, Pearson's correlation coefficient was used.
Otherwise, Spearman's rank correlation coefficient was calculated.
Principal component analysis (PCA) was conducted to visualize relationships between the large number of measured variables and differences between sites and tree statuses. PCA was performed for the organic layer and mineral soil separately. All measured variables from the "process sample set" were included, except the large number of measured enzymes, which showed no correlation with tree status. We further included the soil +understory respiration rate at the middle of the growing season (12 July). The PCAs were performed in R using the function prcomp. Data were scaled to similar standard deviation and centered to a mean of zero. Ellipses are presented at a confidence interval of 90%.
If not otherwise stated, we define significant differences as p < 0.05.
Statistics and figures were performed in R (version 3.6.1, R Core Team, 2013) and maps in ArcGIS (version 10.8, Esri).

| Data
The data that support the findings of this study are openly available in Dryad (Meyer et al., 2021).

| Understory vegetation cover
Understory vegetation was similar under living, dead 12 , and dead 55 trees. The plant community of the treeless tundra differed considerably from that under living and dead trees. In the treeless tundra, moss (Dicranum spp), lichen (Ochrolechia frigida), and black crowberry (Empetrum nigrum) were most frequent, each covering around 30% of the area. Other species occurred occasionally. Under living, dead 12 , and dead 55 trees, Empetrum nigrum accounted for 65-70% of the vegetation cover, that is, a considerably higher percentage than in treeless tundra. The remaining ~30% were mainly dominated by lingonberry (Vaccinium vitisidaea), Dicranum spp, and Ochrolechia frigida, each accounting for ~10% of the area ( Figure S1).

| Soil temperature and moisture
The VWC that was measured four times between the beginning of July and end of September was on average higher at sites 2 and 3 in comparison with site 1, with significant differences in September but not in summer (Table 1). There was no significant effect of tree status on VWC.
Soil temperature at a depth of 10 cm was significantly affected by tree status: soils under living trees had lower average monthly temperatures in the summer compared with soil at the treeless tundra (Table 1). However, during the winter months, soils under living trees were on average warmer than at the treeless tundra sites, although this difference was not significant (Table 1, for details see also Figure S2). Hence, there was a significant tree status × month interaction (Table S1). At sites 1 and 3, summer temperatures under dead trees were intermediate between treeless tundra and living trees (significant tree status × site effect). On average, the soil temperatures at site 3 were lower throughout the years compared with sites 1 and 2 (Table 1).

| Aboveground and belowground carbon and nitrogen stocks
On average, the thickness of the organic layer was 5.5 ± 3.6 cm, but was significantly affected by tree status and site (Table 2, Figure 3a). Both living and dead trees had thicker organic layers than the treeless tundra. However, there was no consistent difference in organic layer thickness between living and dead trees: while site 1 revealed decreasing organic layer thickness in the order of living>dead 12 =dead 55 >treeless, dead 55 trees revealed the thickest organic layers at site 2 ( Figure S3). At site 3, organic layer thickness was generally lower, with only small differences between tree status ( Figure S3). Differences between tree status persisted up to at least 1 m from the tree, but decreased significantly with distance. Organic layer thickness did not change with distance to the plot center in the treeless tundra plots, yielding a significant tree status × distance interaction. C contents were also significantly higher under living and dead trees compared with the treeless tundra (Figure 3b). As a result, C stocks followed a similar pattern as organic layer thickness, with overall higher stocks under living and dead trees compared with the treeless tundra ( Figure 3c). This effect similarly decreased with increasing distance from the tree, and high variability between sites was seen ( Figure S3).
The average thickness of the E horizon was 2.1 ± 0.8 cm across all sites, tree status, and distances from a tree. The remaining mineral soil up to the bedrock ("B horizon") was on average 4.1 ± 3.3 cm thick. In mineral soil, neither the thickness, SOC contents nor SOC stocks were significantly different between tree statuses (Table 2).
There was no clear trend with distance from the tree ( Table 2, see also   Figures S4 and S5). TA B L E 1 Temperature, soil moisture, and tree size characteristics across sites and tree statuses. Different lowercase letters indicate differences between sites (line by line) based on two factorial ANOVA. Different capital letters in the column "all sites" indicate differences between the tree statuses (two factorial ANOVA). In case of non-significant differences, letters were not provided Volumetric water content July 12 21.5 ± 3.0 23.7 ± 6.3 23.6 ± 6.2 22.9 ± 5.4 Volumetric water content September 25 21.5 ± 5.1 b 30.9 ± 6.6 a 29.5 ± 4.6 a 27.3 ± 6.8 1 Excluding data for 13 days in July where measurements were not available.
TA B L E 2 F and P statistics of the effects of tree status and distance on layer thickness, C contents, and C stocks Total overall SOC stocks within 1 m distance (i.e., sum of SOC stocks from org + E + B, Figure 4) decreased in the order of dead 55 >living> dead 12 > treeless, with significant differences only between treeless tundra and living trees and between treeless tundra and dead 55 (Figure 4).
Overall ecosystem C stocks decreased in the order of liv-ing>dead 12 >dead 55 >treeless tundra ( Figure 4) and were mainly driven by changes in aboveground biomass, which significantly decreased upon tree death, especially at later stages.
Soil N stocks were also calculated (see supplementary information).

| Ex-situ soil respiration measurement
Basal respiration measured under standardized conditions of moisture (50% of WHC) and temperature (20°C) decreased in the order of living>dead 12 =dead 55 > treeless tundra, both in the organic ( Figure 5a) and mineral layer (Figure 5f). The effect of tree status was significant in mineral soil, but only close to significant (p = 0.09) in the organic layer. A post-hoc test revealed significant differences in mineral soil only between living trees and the treeless tundra ( Figure 5f). Although dead trees were not significantly different from living trees or the treeless tundra, there was a tendency that dead trees were in transition between living trees and treeless  Distance from the tree (cm) tundra, both in mineral soil and organic layer (Figure 5a,f). The soil respiration rate was related to C contents in mineral soil (r = 0.55, p < 0.001) but not in the organic layer (r = 0.15, p = 0.31). When expressed per g of C (i.e., SOC-specific respiration), soil respiration was not significantly different between each tree status, neither in the organic layer nor in the mineral soil (Figure 5b,g). The SOC-specific respiration was inversely related to the C/N ratio in the organic layer (r = −0.61, p < 0.001) and mineral soil (r = −0.44, p < 0.001).
Following the addition of glucose, hourly CO 2 release increased in comparison with basal respiration rate ( Figure S9). The cumulative amount of CO 2 released within 100 h after glucose addition (i.e., SIR cum ) decreased in the order of living >dead 12 =dead 55 > treeless tundra, both in the organic layer ( Figure 5c) and in mineral soil ( Figure 5h). The effect of tree status was significant. Post-hoc tests revealed that SIR cum under living trees, dead 12 , and dead 55 trees differed significantly from the treeless tundra in the organic layer. In mineral soil, only living trees differed from the treeless tundra. There was a tendency for dead trees to differ from living trees and from the treeless tundra, although these differences were not significant.
SIR cum was positively related to the concentrations of NO 3 , both in the organic layer (r = 0.30, p = 0.04) and in mineral soil (r = 0.35, p = 0.02). Adding (NH 4)2 SO 4 to the glucose-amended mineral soil induced larger CO 2 release than without N addition ( Figure S9).

| Soil biological and chemical properties
Neither the pH value, C/N ratios, nor enzyme activities differed significantly between tree status (Table 3). Moreover, the microbial biomass (MBC) did not differ significantly between tree status in the organic layer and mineral soil (Figure 5d,i), although there was a decreasing trend in the order of living>dead 55 >dead 12 >treeless tundra (Figure 5i).
Soil N availability differed across tree status: in the organic layer, NO 3 − concentrations were significantly higher under living trees than under dead 55 and the treeless tundra; dead 12 was intermediate, although these differences were not significant (Figure 5e). This difference did not persist in mineral soil, where no significant differences were found due to the overall low NO 3 − content (Figure 5j). In all plots, NH 4 + dominated the soil mineral N pool with a minimum of 80% (Table 3), but did not differ significantly between tree status, neither in the organic layer nor in mineral soil.

| Soil +understory respiration
Soil +understory respiration was significantly higher under living and dead trees compared with the treeless tundra, but there was no significant difference between living and dead trees (Table 4, Figure 6). Nevertheless, there were remarkable difference between tree status among the three sites ( Figure S10). At site 1 and site 3, average soil +understory respiration decreased in the order of living>dead 12 >dead 55 >treeless tundra.
However, at site 2, dead 55 revealed the highest ecosystem respiration rates, even exceeding those of living trees.

| Principal component analysis
In the organic layer, the first and second principal components ex-  Carbon pool dead trees (Figure 7a), which were also immediate here. Furthermore, living and dead trees revealed larger variability in the multidimensional space, while treeless tundra clustered in a smaller area.
In mineral soil, the first and second principal components explained 34% and 19% of variance, respectively (Figure 7b). There was no clear clustering by tree status. As already observed for the organic layer, living trees revealed the largest variability in the multidimensional space, while treeless tundra clustered in a smaller area, with dead trees in between.

| DISCUSS ION
In this study, we aimed to understand the effect of treeline advance and decline on SOC stocks and dynamics. To this end, we compared SOC stocks and biochemical soil properties under living trees, treeless tundra, and two stages of dead trees (12 and 55 years since moth outbreak) in a small-scale space-for-time approach. In summary, our first hypothesis (section 4.1) of living trees having lower C stocks compared to treeless tundra was rejected. Contrary to our hypothesis, the net effect of the living tree on soil C stocks was positive. However, this does not exclude the possibility that the living tree enhances SOM decomposition via priming effects. In fact, some of our results point toward higher priming effects under living trees, but the net effect of the tree was positive, which means that the C losses from accelerated SOM decomposition under living trees did not exceed the C inputs. Our second hypothesis (section 4.2) was that tree death causes the C stocks to develop toward the C stocks of the treeless tundra. Indeed, several soil properties under dead trees were in transition between those of living trees and treeless tundra. Yet, this did not apply to SOC stocks, which remained at an elevated level comparable to that of living trees, even 55 years after the moth outbreak. Hence, also our second hypothesis was rejected.

| Effect of living birch trees on ecosystem carbon stocks and soil organic matter turnover
Assuming that priming is a relevant mechanism under living trees, we hypothesized that accelerated SOC turnover leads to smaller SOC  stocks under living trees compared to treeless tundra (Hartley et al., 2012;Parker et al., 2015;Wilmking et al., 2006). Contrary to this hypothesis, we found that SOC stocks were on average significantly higher under living trees compared with treeless tundra (Figure 3c).
Higher SOC stocks were still evident even 1 m from the tree, although these differences decreased with increasing distance. Most studies that have reported lower SOC stocks under living trees have compared forested sites with separate treeless sites, for example, along latitudinal or altitudinal transects or at locations with considerable distance between each other. In such approaches, SOC stocks could be also affected by differences in temperature, moisture conditions, soil characteristics, and/or aboveground and belowground litter input (e.g., Rapalee et al., 1998;Sjögersten & Wookey, 2004). In this regard, a causal link between trees and lower SOC stocks is not certain. With our single tree scale approach, we could and Wang et al. (2015) also found that positive priming mainly takes place in the mineral soil, but not or at lower rates in the organic layer.
Hence, the effect of higher SOC input under living trees may be cancelled out by priming in mineral soil.
Numerous studies have clearly demonstrated that priming effects occur in the field under living trees (Adamczyk et al., 2019;Clemmensen et al., 2021;Hartley et al., 2012;Parker et al., 2015).
Although we did not directly measure priming, we have several results that point toward accelerated SOM decomposition under living trees at our study sites, as discussed in detail below.
Since priming is notoriously difficult to measure under field conditions, it is often investigated in incubation experiments after the addition of 13 C labeled substrates, which allows native SOC mineralization to be distinguished from mineralization of labile C (Karhu et al., 2016). Such experiments only indicate a potential priming effect under standardized C supply, but do not reflect the naturally occurring priming intensity in soil where labile C supply varies between living trees, dead trees, and treeless tundra. Yet, priming has been reported to be driven by microbial N mining, that is, by microbial TA B L E 4 F and P statistics for ecosystem respiration measurements. Significant effects are highlighted in bold  labile C supply (e.g., from fresh litter and root exudates) increases microbial N demand, eventually increasing the activity of extracellular enzymes that decompose SOM for N acquisition. Hence, mineral N contents may provide hints for priming. Indeed, we found higher NO 3 − contents under living trees compared with treeless tundra. The finding of higher N availability under living trees was also corroborated by incubation experiments after adding glucose: soils taken under living trees were able to mineralize larger amounts of glucose, which, according to Nordgren (1992) and Meyer et al. (2018) suggests higher amounts of available nutrients, in this case N (as tested by N additions, Figure S9). Other studies have also shown that living birch trees increase soil N availability (Mikola et al., 2018;Silfver et al., 2020). We interpret this finding as an indicator of N mining, which is presumably coupled with accelerated turnover of SOM, that is, with priming. However, higher mineral N contents may also be interpreted inversely: higher mineral N contents under living trees may indicate less need of microbes to decompose SOM for N acquisition (Craine et al., 2007), that is, it suppresses priming. In light of the very N-poor subarctic soils (Hicks et al., 2020), it seems contradictory to find an excess of mineral N despite the greater N demand of birch roots. Hence, the excess mineral N under living trees most likely results from accelerated SOM decomposition close to the tree. This is further indicated by largest soil +understory respiration rates and microbial biomass under living trees.
Besides higher mineral N contents, we also found elevated N stocks under living trees than under treeless tundra. As both C and N are major components of SOM, higher N stocks under living trees are not necessarily surprising, since SOC stocks were also high. In such nutrient-limited regions, it is unlikely that the soil beneath the tree contained sufficient N for accumulating such high amounts of aboveground and belowground organic matter stocks.
While larger mineral N contents can be explained by accelerated SOM decomposition close to the tree, another mechanism must be responsible for elevated N stocks. Although N fixation might be a potential source of N (Rousk & Michelsen, 2017), we do not think that this can explain the excess of N under living trees, especially because mosses as important N fixing species (Zackrisson et al., 2009) are more abundant in the treeless tundra ( Figure S1).
Notably, birch trees form symbioses with ectomycorrhizal (ECM) fungi that can extend their hyphae several meters (Simard & Durall, 2004) while "mining" the soil for N by decomposing SOM. Ellipses show the 90% confidence interval for the respective tree status. Thickness refers to the thickness of the organic layer or mineral soil, respectively; C and N refer to concentrations (%); C stock is the calculated SOC stock at 50 cm distance; Soil resp./C is the SOC-specific respiration rate; SIR cum is the cumulative CO 2 evolution within 100 h after glucose addition; Soil resp. is the soil respiration rate under laboratory conditions; Field resp. is soil +understory respiration measured in the field; and MBC is microbial biomass carbon [Colour figure can be viewed at wileyonlinelibrary.com] priming (Paterson et al., 2016;Read & Perez-Moreno, 2003). As a result, litterfall concentrates N that has been acquired from a larger volume of soil, to a smaller area around it. Hence, elevated N stocks below the tree may indeed be indicative of N mining, a process which has also been observed in other mountain birch forests in Fennoscandia . Yet, locations of excess N stocks do not necessarily match locations where N was mined and the associated priming took place. We suggest that N mining and priming are not exclusively coupled in a spatial way and hypothesize that elevated N stocks, in parts, could originate from SOM decomposition further away from the tree.
We found further indications that priming-induced SOC losses can be considerable, even if this was on average not the case in our study area. Our results revealed considerable variations in SOC stocks and dynamics between the three investigated sites ( Figure   S3). The finding that site 3 had low SOC stocks under living trees was remarkable, as tree biomass and thus C input to soil was largest compared with the other sites. Also, respiration rates under living trees were at a similar or even higher level at site 3 compared with the other sites ( Figure S10), even though site 3 was characterized by lower SOC stocks, similar potential SOC decomposability, and colder temperatures. Hence, root respiration may have a higher contribution to soil +understory respiration at site 3 than in other sites. The amount of root respiration is correlated with the amount of root exudation (Sun et al., 2017), which is one of the main C sources that cause priming (Shahzad et al., 2015). Hence, it can be speculated that larger priming effects are responsible for the low SOC stocks under living trees at site 3. Thus, we suggest that in some cases the high amount of N required by large trees can stimulate SOM decomposition to such an extent that it compensates for elevated SOC inputs (cf. Hartley et al., 2012), making it necessary to conduct these studies on several independent sites simultaneously. It has been reported that C input by the understory vegetation can make a substantial contribution to total soil C input in forests, especially in the north. For instance, Liski et al. (2006) (Ding et al., 2019). The finding of greater understory root biomass under living trees in comparison with treeless tundra supports this assumption ( Figure 4). Hence, the understory vegetation may obscure the direct effect of trees on SOC stocks, that is, the magnitude of accelerated SOC turnover by birch trees may be underestimated.
Overall, living trees increased the heterogeneity of SOM stocks and dynamics in comparison with treeless tundra. The PCA plot indicates that the treeless tundra clusters in a narrower range in the multidimensional space compared to living trees. This was especially true for mineral soil, where treeless tundra revealed little variation.
We assume that differences in tree properties (e.g., age, size, root system, etc.) induce a different degree of priming, N mineralization, and SOC accumulation. This has also been shown by , suggesting large variations in respiration and mycorrhizosphere productivity between studied trees. Thus, we conclude that trees generate soil heterogeneity, ultimately increasing patchiness of subarctic landscapes.

| Effect of dead trees on ecosystem carbon stocks and soil organic matter turnover
We originally hypothesized that tree death causes the soil C stocks under the tree to develop toward the C stocks of the treeless tundra, that is, to be in transition between living trees and treeless tundra.
Contrary to our expectations, we did not find a consistent development back to tundra-like SOC stocks within 55 years after tree death, that is, SOC stocks under dead trees did not significantly differ from those under living trees. In a short-term study, Sandén et al.
(2020) also found no change in SOC stocks within 8 years after tree death. Similarly, no change in SOC stocks was found within 4 years after tree death in boreal Picea abies stands caused by a bark beetle outbreak (Kosunen et al., 2020). A continual supply of decaying wood seems to maintain SOC stocks at comparatively high levels, even decades after tree dieback.
In addition to continual supply of decaying wood, elevated SOC stocks under dead trees in comparison with treeless tundra may be further explained by the understory vegetation. Karlsen et al. (2013) and Sandén et al. (2020) reported that moth outbreaks affect the understory vegetation toward a grass-dominated community with consequently labile litter input to soils. We could not confirm this observation in our study, possibly because the appearance of grasses is a short-lived effect caused by the sudden nutrient supply during and shortly after an outbreak event (Vuojala-Magga & Turunen, 2015).
Here, the understory community remained unaffected by tree death and resembled that of living trees. Thus, the amount of soil C inputs from understory litter and roots may be similar under living and dead trees, but clearly different from treeless tundra. This can be of quantitative relevance: according to Liski et al. (2006), the understory vegetation was estimated to account for 26% of the total soil C inputs in Finnish forests. This contribution may be even higher in northern birch forests, that is, 35% according to Ding et al. (2019).
Moreover, the composition of SOM may be dominated by recalcitrant litter input from Empetrum nigrum (Tybirk et al., 2000), both under living and dead trees, eventually resulting in accumulation of SOM. The combination of ongoing deadwood supply and unaltered C input from understory vegetation may partly explain the unaltered SOC stocks upon tree death.
A further explanation for the unchanged SOC stocks may be a slowed SOM decomposition, which can be assumed for several reasons. First, quality of SOM may change toward a larger proportion of recalcitrant compounds once the most labile compounds of decaying wood have been mineralized (Mackensen & Bauhus, 2003).
This, however, was not observed here. Instead, the SOC-normalized respiration did not decrease upon tree death, which indicates similar potential decomposability of SOC under living and dead trees (Fang & Moncrieff, 2005;Mikan et al., 2002). This finding may be partly explained by the unchanged understory vegetation (see above), thereby attenuating the effect of tree death on overall SOC decomposability. Second, the need of vegetation and microorganisms for N may decrease in the absence of trees, tree exudates (Bengtson et al., 2012), and mycorrhiza (Paterson et al., 2016;Read & Perez-Moreno, 2003). Together with the excess of available N in the soil both before (i.e., living trees, Figure 5) and immediately after the moth outbreak event (Kaukonen et al., 2013), N deficiency was likely alleviated after tree death, which may restrict SOM decomposition for the N acquisition, that is, N mining. Indeed, we found lower NO 3 contents and SIR cum under dead trees compared with living trees, which points to smaller amounts of available nutrients. This indicates that SOM turnover-but not potential decomposability of SOM-decreased upon tree death. Lower soil +understory respiration and MBC under dead trees compared with living trees further support this assumption. In summary, the combination of ongoing deadwood supply, unaltered C input from understory vegetation, and slowed SOM turnover may explain why SOC mineralization could not exceed SOC input for at least 55 years after tree death.
Our results of lower N availability, MBC, and soil respiration under dead trees oppose previous observations that mainly covered the short-term effects of moth outbreaks. Such studies mainly referred to the deposition of insect frass, which was reported to be more easily available to microbes than senesced litter (Kristensen et al., 2018), thereby enhancing SOM turnover (Kaukonen et al., 2013). Increases in available N and dissolved organic carbon were also reported, with stimulating effects on MBC and ultimately soil respiration (Kaukonen et al., 2013;Kristensen et al., 2018;Sandén et al., 2020). Here, we did not observe any of these short-term effects that have been observed after moth outbreaks, neither after 12 nor 55 years since the outbreak. This contradiction to previous studies is likely a result of the different time scale considered. Kristensen et al. (2018) speculated that moth outbreaks cause a short-term increase, but long-term decrease in decomposition rates of SOM. Hence, the stimulation of microbial activity upon tree death is likely a short-lived effect triggered by a sudden C and N supply that is not reflected in our data.
Although we found indications for decreasing SOM turnover under dead trees, it did not reach the level of treeless tundra, even after 55 years. Enzyme activities reflect unaltered C or N acquisition strategies of soil microbes upon tree death. Indeed, leaching of dissolved organic carbon from deadwood has been reported, thereby stimulating MBC and respiration (Bantle et al., 2014;Peršoh & Borken, 2017). This effect may persist in the long term.
For instance, Minnich et al. (2020) observed elevated MBC and SOC mineralization rates in soil beneath deadwood even 8 years after experimentally placing logs on the forest floor. Furthermore, several studies showed that fungi transfer N from soil to the decaying wood (Oleveira Longa et al., 2018;Philpott et al., 2014;Rinne et al., 2017), hence N mining may continue upon tree death. Such processes may likely be more prominent in our long-term study with more deadwood laying on the forest floor compared to studies that focused on short-term effects of moth outbreaks with standing trees. Hence, N mining and associated priming may continue for a long time after tree death, although at lower rates, and mainly triggered by deadwood instead of root exudates as a source of labile C. This transitional state of dead trees is also visualized by the PCA. In the organic layer, living trees clustered in the lower part of the plot, while treeless tundra was on the upper part and dead trees were in between ( Figure 7a). Hence, tree death is accompanied by a transition from a rather N-rich soil with high microbial activity to a N-poor soil with low microbial activity. These changes in soil properties upon tree death indicate a shift in characteristics of biogeochemical cycling toward a state comparable with treeless tundra. Yet, even after 55 years, SOM turnover was still elevated in comparison to treeless tundra.
While SOC stocks remained at an elevated level even decades after moth outbreaks, it should be highlighted that the overall ecosystem carbon stock was clearly reduced by tree death, mainly caused by a decline in aboveground biomass. In such heavily grazed areas, reindeer grazing is a crucial factor in regulating birch growth (Stark et al., 2021), thereby possibly also determining whether birch trees recover or die after moth defoliation (Lehtonen & Heikkinen, 1995). It may also restrict growth of new trees (Biuw et al., 2014;Olofsson et al., 2009). There is clear evidence from published studies that regeneration of birch forest is hampered on the year-round grazed areas on the Finnish side, compared to Norway, where the reindeer migrate seasonally between summer and winter range (Biuw et al., 2014;Kumpula et al., 2011 in the interior birch forests. In our study area, in contrast, the area is a year-round pasture for reindeer (Biuw et al., 2014). However, it remains uncertain to what extent the lack of forest recovery on our sites was affected by reindeer densities and on the other hand by historical developments in the study area, preventing the reindeer from going into separate summer and winter pastures. In the circumpolar Arctic, it is more common for the reindeer to migrate seasonally between winter and summer ranges (Stark et al., 2021), which affects the generalization of our findings over a larger geographic scale. In summary, we conclude that in this area, the increased occurrence of moth damage would have minor effects on SOC stocks, but would decrease ecosystem C stocks if the mountain birch forests will not be able to recover from the outbreaks. However, we cannot exclude the possibility that the effect of moth outbreaks on ecosystem C stocks are different in regions that are less heavily affected by grazing, which calls for a comparison of areas with different grazing intensities.

| CON CLUS IONS
Our results indicate that trees strengthen the subarctic carbon sink in the Pulmankijärvi area of Finnish Lapland by increasing aboveground and belowground carbon stocks. Forest dieback by severe moth damage may become more frequent in the future due to global warming, and may decrease the total ecosystem C stocks. However, although aboveground biomass decreased upon tree death, SOC stocks under dead trees remained at comparable levels to living trees for at least 55 years. We conclude that the CO 2 -related positive feedback of forest disturbance on climate change might be small and very slow in the subarctic region, although only with respect to soil, but not to aboveground stocks. Importantly, our data also highlight considerable spatial variability in the effect of tree status on SOC stocks. This clearly warns against comparing separate sites of living forests, dead forests, and treeless tundra when making predictions about the effect of treeline advancement or recession. We consider the small-scale approach as a more reasonable space-for-time substitution in such patchy landscapes. While our study represents a comprehensive survey of the net effects of mountain birch tree deaths on ecosystem soil C stocks in the Pulmankijärvi area, similar small-scale studies are required in other regions of Fennoscandia to study how generalizable our findings are.