Limited effects of early snowmelt on plants, decomposers, and soil nutrients in Arctic tundra soils

Abstract In addition to warming temperatures, Arctic ecosystems are responding to climate change with earlier snowmelt and soil thaw. Earlier snowmelt has been examined infrequently in field experiments, and we lack a comprehensive look at belowground responses of the soil biogeochemical system that includes plant roots, decomposers, and soil nutrients. We experimentally advanced the timing of snowmelt in factorial combination with an open‐top chamber warming treatment over a 3‐year period and evaluated the responses of decomposers and nutrient cycling processes. We tested two alternative hypotheses: (a) Early snowmelt and warming advance the timing of root growth and nutrient uptake, altering the timing of microbial and invertebrate activity and key nutrient cycling events; and (b) loss of insulating snow cover damages plants, leading to reductions in root growth and altered biological activity. During the 3 years of our study (2010–2012), we advanced snowmelt by 4, 15, and 10 days, respectively. Despite advancing aboveground plant phenology, particularly in the year with the warmest early‐season temperatures (2012), belowground effects were primarily seen only on the first sampling date of the season or restricted to particular years or soil type. Overall, consistent and substantial responses to early snowmelt were not observed, counter to both of our hypotheses. The data on soil physical conditions, as well interannual comparisons of our results, suggest that this limited response was because of the earlier date of snowmelt that did not coincide with substantially warmer air and soil temperatures as they might in response to a natural climate event. We conclude that the interaction of snowmelt timing with soil temperatures is important to how the ecosystem will respond, but that 1‐ to 2‐week changes in timing of snowmelt alone are not enough to drive season‐long changes in soil microbial and nutrient cycling processes.


| INTRODUC TI ON
Climate warming is influencing Arctic tundra ecosystems by shifting the timing of seasonal events such as snowmelt and soil thaw (Ernakovich et al., 2014;MacDonald, 2010;Xu et al., 2013). This changing seasonality is leading to biological responses such as changes in plant phenology, timing of herbivore activity, and soil microbial and nutrient dynamics (Buckeridge, Banerjee, Siciliano, & Grogan, 2013;Høye, Post, Meltofte, Schmidt, & Forchhammer, 2007). Of these, aboveground responses are better studied than those belowground in soils, and furthermore, for soil responses, surprisingly little attention has been paid to the most likely future scenario: the combination of warming and earlier snowmelt (Callaghan et al., 2011;Chapman & Walsh, 2007;Wipf & Rixen, 2010). To understand the future of Arctic tundra ecosystems, we must understand how soil organisms (including plant roots, microbes, and soil fauna) and soil processes such as decomposition and nutrient cycling will respond to warming and an earlier onset of the growing season.
In Arctic tundra soils, decomposition controls both carbon stocks (Davidson & Janssens, 2006) and nutrient availability for plants and microbes (Schimel & Bennett, 2004). Decomposition in Arctic tundra soils is driven by saprotrophs (mainly bacteria and fungi) that enzymatically degrade plant litter and soil organic matter stocks, mineralizing nutrients that are then retaken up by plants and microbes (Burns & Dick, 2002;Schimel & Bennett, 2004). Higher trophic-level organisms such as protists and microarthropods also contribute by shredding and digesting organic matter (Seastedt, 1984) or consuming saprotrophs, releasing nutrients (Clarholm, 1985). Plants take up some of these released nutrients and can also stimulate further decomposition by feeding C to microbes via rhizodeposition or transfer to mycorrhizae (Clemmensen, Michelsen, Jonasson, & Shaver, 2006;Kuzyakov, Friedel, & Stahr, 2000;Sinsabaugh & Moorhead, 1994;Weintraub, Scott-Denton, Schmidt, & Monson, 2007). All of these soil organisms and processes exhibit pronounced seasonal timings that are likely to respond to changing seasonality. Nitrogen (N) is an essential plant macronutrient in Arctic tundra ecosystems, where it often, though not always, limits both plant production and microbial activity (Mack, Schuur, Bret-Harte, Shaver, & Chapin, 2004;Melle, Wallenstein, Darrouzet-Nardi, & Weintraub, 2015;Nowinski, Trumbore, Schuur, Mack, & Shaver, 2008;Weintraub & Schimel, 2003). The main source of N in Arctic soils is the large stocks of organic N, which can be mined enzymatically F I G U R E 1 Four biogeochemical scenarios relevant to our hypotheses. Biogeochemical pools are shown as colored boxes with larger boxes indicating that that pool would be larger in that scenario than in others. Fluxes are shown as arrows, with elevated flux levels shown in red. (a) Over winter, nutrients build up in soils due to lack of plant uptake and substantial microbial activity below snow. This large pool is visible as a "pulse" of nutrients at the start of the season. (b) Midsummer, a "crash" in nutrients has been observed due to high plant and microbial demand for nutrients. (c, d) Two possible scenarios we might observe if plants are damaged by exposure to cold temperatures with early snowmelt. (c) If microbes are reliant on plant inputs (e.g., via rhizodeposition), all biotic activity may slow, leading to a delayed crash in nutrient availability. (d) If microbes continue to thrive, they may immobilize nutrients and take up C from damaged roots leading to a phenological "mismatch" in association with microbial turnover (Figure 1a; Lipson, Schmidt, & Monson, 1999). Snow depth and the timing of melt can affect the size of this pulse by altering winter soil temperatures and patterns of N release in the spring (Buckeridge et al., 2013;Edwards & Jefferies, 2013). N cycling in Arctic soils is also dynamic during the summer growing season (McLaren et al., 2017), sometimes exhibiting a nutrient "crash" in which extractable NH 4 + , NO 3 − , and amino acids fall to undetectable levels midseason as plant and microbial usage of N peaks (Figure 1b; Weintraub & Schimel, 2005). In light of these strong seasonal patterns, it is important to understand how they will respond to changing seasonal timings such as earlier snowmelt and plant phenology.
Effects of changing seasonality in Arctic ecosystems have most commonly been investigated using delayed snowmelt experiments, typically accomplished by using snow fences that deepen snowpack and delay snowmelt, resulting in several weeks delay in plant phenology and increased water input (Borner, Kielland, & Walker, 2008;Rogers, Sullivan, & Welker, 2011;Wipf & Rixen, 2010). Soil N cycling also responds to snow fence treatments: Deeper snow warms the soil, which increases nutrient availability via stimulation of microbes (Buckeridge & Grogan, 2008;Freppaz, Williams, Seastedt, & Filippa, 2012;Schimel, Bilbrough, & Welker, 2004). Snow fences also trap wind-blown litter, which could affect nutrient cycling as well (Fahnestock, Povirk, & Welker, 2000). Increased nutrient cycling has also been observed in an experiment where increasing snow depth was decoupled from delayed snowmelt (Natali, Schuur, & Rubin, 2012). A key question is whether we might see the reverse effectthat is, decreased microbial biomass and nutrient cycling activity with earlier snowmelt and soil exposure to colder air temperatures in the late spring.
In the case of advanced rather than delayed snowmelt, a multiyear snow removal experiment in a tussock tundra ecosystem near Toolik Lake, Alaska, showed that advanced snowmelt advanced the start of the growing season for plants as well as the onset of plant senescence (Khorsand Rosa et al., 2015;Starr, Oberbauer, & Pop, 2000).
There was no strong evidence of a change in net C balance due to a shift-but not extension-of the growing season (Oberbauer, Starr, & Pop, 1998) and no clear effect on plant photosynthetic capacity (Starr & Oberbauer, 2003). Results from the same early snowmelt manipulation on which we focus in this study corroborate these findings, showing advanced aboveground plant phenology (Livensperger et al., 2016). However, plant responses to early snowmelt are not always the reverse of those with delayed snowmelt. In several cases from forested and alpine systems, early snowmelt treatments have been shown to damage plants due to the loss of the insulating snow layer (Wipf, Rixen, & Mulder, 2006;Wipf, Stoeckli, & Bebi, 2009 and N cycling processes, including the nutrient pulse at snowmelt and the "crash" during peak plant physiology; (b) loss of insulating snow cover damages plants, leading to reduced root growth, with various possible outcomes. These outcomes could include reduced stimulation of microbes via rhizodeposition leading to overall reduced biotic activity and a delay in the nutrient crash ( Figure 1c); or even a phenological "mismatch" in which microbes take up carbon from damaged roots and outcompete plants for N leading to greater N immobilization in microbes (Steltzer, Landry, Painter, Anderson, & Ayres, 2009;Figure 1d). Loss of insulation could also affect the microbes directly; little is known about how these factors might interact. To test these hypotheses, we set up an early snowmelt × warming field manipulation experiment in an Arctic tundra ecosystem in which we monitored physical conditions, plant phenology, root growth, microbial biomass, soil enzyme activities, microarthropod densities, and soil nutrient dynamics ( Figure 2). With these data, we assess the effects of early snowmelt on three main components of the soil biogeochemical system: plants, decomposers, and soil nutrients. in Alaska, USA. The plant community is moist acidic tundra (MAT), a common vegetation type in the northern foothills of the Brooks Range dominated by Eriophorum vaginatum, a tussock forming sedge (Shaver & Chapin, 1991 and Salix pulchra. There is an uneven cover of organic soil 0-20 cm thick in MAT (Shaver & Chapin, 1991). The organic layer directly underneath E. vaginatum plants consists primarily of decaying roots of E. vaginatum. The soil organic matter has a mean C concentration of 45%, N concentration of 0.8%, and pH of 4.4. Due to low evapotranspiration and melting ice in the active layer of these permafrost soils, soil moisture in this ecosystem remains relatively high and constant at a level of ~0.8 g water/g wet soil ( Figure A1; Weintraub & Schimel, 2005).

| Early snowmelt and warming treatments
We used a snowmelt advancement technique that isolates changes in snowmelt timing by using shade cloth (Steltzer et al., 2009) and crossed it with a standard open-top chamber (OTC) warming treatment (Oberbauer et al., 2007). In August 2009, we laid out five blocks of two 8 × 12 m plots at our study site ( Figure A2). The plots were demarcated in the season prior to applying the treatments in order to install minirhizotron tubes for root growth measurements and bury iButton temperature loggers (Maxim, San Jose, California, USA) in the soil prior to snowfall. The plots were parallel in arrangement, with 5-30 m between blocks and 4 m between paired plots ( Figure   A2). At the beginning of each of our study years (2010-2012), we applied early snowmelt and warming treatments. One of the plots in each block was randomly chosen for the early snowmelt treatment. Early snowmelt was achieved by deploying 8 × 12 m pieces of black 50% shade cloth fabric (Steltzer et al., 2009) until plots were snow-free ( Figure A3; we defined snow-free as <20% snow cover).
This snowmelt manipulation technique accelerates snowmelt but does not alter the amount of melting snow or increase thaw depth (Livensperger et al., 2016;Steltzer et al., 2009 Figure A4).
In each of the 10 plots, four 1 × 1 m subplots were defined by the locations of the minirhizotron tubes, which were arrayed at equally spaced intervals from the top and bottom edges (8 m sides) of the plots. In each plot, two of the subplots were randomly assigned to the warming treatment. Warming was achieved by deploying International Tundra Experiment-style passive warming OTCs (Marion, 1989, Sullivan & Welker, 2005

| Aboveground plant phenology-broadband NDVI trajectories
Leaf area development was monitored using near-surface broadband NDVI, which was measured using two radiation sensors mounted at ~50 cm height, recording a circular area of ~0.75 m 2 (Sweet, Griffin, Steltzer, Gough, & Boelman, 2015 Sweet et al. (2015). To compare season-long trends in broadband NDVI among treatments, we scaled the time series on a per-sensor basis (20 in total) to have a standard deviation of 1 and centered each series by subtracting the average of their first four readings, which represent an early-season baseline during which all plots exhibited their post-snowmelt minimum. We term the output of this scaling process "broadband NDVI trajectories" because they emphasize seasonal trends while removing the effects of plot-to-plot variation in overall broadband NDVI magnitude, which is driven mainly by how much nonvegetative material is in the viewing area (rock, soil, etc.). The unmodified broadband NDVI time series are included in the Supporting Informations for reference ( Figure A5). The offset in magnitudes among sensors visible in these unmodified series demonstrates why we took this per-sensor scaling approach.

| Aboveground plant phenology-marked individuals
To put our belowground results in context, we compare the timing of root growth to the timing of first leaf expansion on individual plants in each of the treatment plots. Data from the 2012 season were included in Livensperger et al. (2016). We use the timing of first leaf expansion because it was our most consistently recorded vari-

| Root production
Root production was estimated with a BTC-2 and for other species (Sullivan et al., 2007). Estimates of root mass were then scaled to one square meter. Though the tube depths varied ( Figure A6), this is indicative of the natural variation in depth to permafrost. All tubes were installed to maximum thaw depth during their August installation. Tubes stayed in their original positions throughout the experiment, with no appreciable heave effects observed.
Even if present, minor heave effects would be more likely to occur outside of the growing season and thus unlikely to affect our growing season production measurements. We found some evidence of under-sampling of E. vaginatum roots in the top 10 cm; in many tubes, root areas in the top 10 cm were lower than the 10-20 cm values.
This may have been due to the positioning of the minirhizotrons with respect to the E. vaginatum tussocks. To account for this, we repeated our root analyses using only the 10-20 cm depth. Trends and results were similar and so here we present only the full profile data. The vertical trends within tubes are shown in Figure A6.

| Soil pore water nutrients
We measured soil pore water nutrient concentrations nondestructively at high temporal resolution using microlysimetry allowing us to gain insight into nutrient processes that are typically not observable due to concerns over destructive sampling. High temporal resolution is important because studies that are restricted to one or several measurements per year may miss important seasonal events

| Soil cores
While the nondestructive measurements for plants and soil pore water were taken for 3 years, measurements requiring destructive soil sampling were made in the final year of the study to avoid damaging the plots. We collected soil core samples for nutrient, microbial biomass, and plant tissue analysis every 10-14 days during the summer of 2012. We sampled tussock (within an E. vaginatum canopy) and intertussock locations within the plots in the four treatments (control, early snowmelt, warming, and combined).
Collection began with spring thaw and continued until plant senescence, for a total of seven sampling dates in 2012 (25 May-14 August). Due to limited space in the warming treatment's OTCs, those treatments were sampled on only four of the seven dates.
In total for 2012, 295 samples were collected. Each sample was cored with a 5-cm-diameter metal soil corer to a depth of at least 10 cm. Within 2 hrs of collection, live plant material was removed and the top 10 cm of organic soil was homogenized by hand. Once homogenized, a 5 g (wet weight) subsample of each soil was shaken with 25 ml 0.5 M K 2 SO 4 in 50-ml centrifuge tubes on an orbital shaker table at ~120 rpm for 1 hr, then vacuum-filtered through Millipore APM 15 glass fiber filters. The extracts were then frozen for nutrient analysis. A second 5 g subsample was put into a 250-ml Erlenmeyer flask for measurement of microbial biomass.
A third 5 g subsample was used to measure water content of the soil-by-mass difference between fresh and oven-dried (60°C) soil.
Samples were also collected in control conditions just outside the plots during the 2010 and 2011 seasons. Though not presented here, those data are used to inform one analysis of early-season differences (Table 1).

| Microbial biomass
Microbial biomass carbon and nitrogen (MBC and MBN) in the soil core samples were quantified using a direct-chloroform-addition modification of the chloroform fumigation-extraction technique (Brookes, Landman, Pruden, & Jenkinson, 1985;Scott-Denton, Sparks, & Monson, 2003). In brief, 5 g (wet weight) of soil was combined with 2 ml of ethanol-free chloroform and incubated at room temperature for 24 hr in a stoppered 250-ml Erlenmeyer flask.
Following incubation, flasks were vented in a fume hood for 30 min and extracted and frozen as described above. No extraction efficiency correction factor (k EN ) was applied, as it is unknown for these soils, so values presented represent only extractable microbial biomass C and N. The nonfumigated extractions were the same K 2 SO 4 extractions that were used for nutrient analyses.

| Soil microarthropods
We estimated densities of soil-dwelling microarthropods on five dates in 2012 coinciding with five of the soil core collection dates

| Soil enzymes
Soil N-acquiring hydrolytic enzyme activity assays were conducted using high-throughput microplate assays (

| Nutrient analyses
Soil nutrient analyses for soil pore water and soil core extracts were conducted following Rinkes et al (2011 Note. Soil cores were taken shortly after soil thaw (snowmelt dates were 18, 21, and 26 May in the 3 years, respectively). The transition from late summer to fall (18 September (NO 3 − and NH 4 + ) or fluorometric (TFPA) microplate assays. NH 4 + concentrations were measured using a modified Berthelot reaction (Rhine, Sims, Mulvaney, & Pratt, 1998). NO 3 was measured using a modification of the Griess reaction (Doane & Horwath, 2003), which involves the reduction of nitrate to nitrite followed by colorimetric determination of nitrite. TFPA was measured using O-phthaldialdehyde and mercaptoethanol (Darrouzet-Nardi, Ladd, & Weintraub, 2013;Jones, 2002). Absorbance and fluorescence values were determined on a Bio-Tek Synergy HT microplate reader (Bio-Tek Inc.).
Soil pore water collected with the microlysimeters was analyzed for

| Data analysis
The effect of the early snowmelt, warming, and combined treatments was evaluated by evaluating the effect size of these treatments on each variable of interest (Cumming, 2013;Nakagawa & Cuthill, 2007).
For most measured variables, effect sizes are reported as treatment/ control ratios (denoted T r ) with 95% confidence intervals estimated by bootstrapping (Carpenter & Bithell, 2000). In cases where ratios are not appropriate, such as the day of year for phenological events, differences between the treatments (treatment-control) are calculated, also with 95% bootstrapped confidence intervals. We calculated effect sizes for each measurement date, allowing us to evaluate changing effects of the treatments throughout the season using focused comparisons with the control treatment as opposed to pooling our nonhomogenous variances across the whole season (Rosenthal & Rosnow, 1985). For several of the more frequent measurements (soil pore water nutrients and NDVI), we also used seasonal averages based on the mean values across multiple dates, retaining the replication level of n = 5 for each treatment. Among all analyses, the only exception to the n = 5 level of replication is the minirhizotron root biomass data in which two minirhizotron tubes per plot were treated as independent for a total of n = 10. We made this decision because the spatial distances between tubes in the same plots were similar to those between plots ( Figure A2), no substantial blocking effects were observed, and high variation among tubes in the same plots ( Figure   A6) made plot averages undesirable.

| Physical conditions
The our three thaw depth measurements did not reveal clear differences in thaw depth among treatments ( Figure A4).

| First leaf expansion in marked individuals
In 2010, the short snowmelt advancement of 4 days slightly ad-

| NDVI-based assessment of phenology
In 2010 and 2011, we saw relatively small differences in broadband NDVI trajectories among treatments, while in 2012, differences were larger ( Figure 5). In examining broadband NDVI averages across the month of June (representing the period leading up to peak physiology), the largest differences in all years were in the combined

| Root production
At the end of the 2011 season, cumulative E. vaginatum root production in the control plots was 83 ± 29 g/m 2 (mean ± SE, n = 10) and 77 ± 36 g/m 2 in 2012 ( Figure 6a). As an example of the range among minirhizotron tubes, in 2011 the 10 control tubes ranged from 0 to 307 g/m 2 ( Figure A6). Non-E. vaginatum roots had lower production than the E. vaginatum roots (e.g., 2011 controls = 57 ± 19 g/m 2 ; 2012 controls = 29 ± 9 g/m 2 ). In 2011, cumulative non-E. vaginatum fine root production in the 10 control tubes ranged from 0 to 188 g/ m 2 . In 2011, mean non-E. vaginatum root biomass was lower in early snowmelt and combined treatments than in the control treatment, but the differences were poorly constrained for comparisons of the individual treatments with the control (n = 10, early snowmelt

| Microarthropods
On the first measurement date, 7 June 2012, total soil microarthropod density, as measured by the sum of the dominant groups of organisms (Collembolans, Oribatid mites, and Predatory mites), was the lowest of the season in both treatments (Figure 8;

| Soil pore water nitrogen
In 2010 and 2012, average seasonal soil pore water labile N (NO 3 + + NH 4 + + TFPA) was low (<10 µM; Figure 9). With low values in all treatments, differences between treatments and controls were likewise low, showing a well-constrained lack of effect (treatment-control = <5 µM for most treatments in all 3 years).
However, in 2011, seasonal averages were higher and available la-

| Exoenzyme activities
For most of the season, measurable activities were present for the hydrolytic enzymes we assayed (medians across all dates and treatments: BG: 548 nmol g −1 soil hr −1 , NAG: 321 nmol g −1 hr −1 , LAP: 16 nmol g −1 hr −1 ). Oxidative enzymes (phenoloxidase), though not directly comparable to the hydrolytic enzyme rates, showed a median rate of ~420 nmol g −1 hr −1 . We observed a peak, particularly in the warming treatment, on one date, 11 July 2012; this occurred at the same time that there was a low point in the hydrolytic enzyme activities (Figure 11)

| Interannual variation in N cycling
Warmer fall and winter temperatures during our second measurement season (following the 2010-2011 fall and winter;

| D ISCUSS I ON
Our two hypotheses were (a) early snowmelt advances seasonal timing of soil processes; and (b) early snowmelt damages plants, leading to several possible belowground effects. To assess these hypotheses, we discuss responses of three components of the soil biogeochemical system in our experiment: plants, decomposers, and soil nutrients. In discussing these components, we make the case that F I G U R E 4 Mean date of first leaf expansion ± SE by treatment (n = 5). The transition from >20% snow cover in the controls to snow-free is shown by the shading in the same way as Figure 3 F I G U R E 5 Scaled broadband NDVI trajectories from 20 "mantis" instruments employing radiation sensors with a ~0.75 m 2 viewing area (mean ± SE). Treatments were measured simultaneously but are offset slightly to avoid overplotting F I G U R E 6 Cumulative root production to date from minirhizotron measurements in 2011 and 2012. Lines and bars show mean ± SE by treatment (n = 10). Treatments were measured simultaneously but are offset slightly to avoid overplotting. Due to destructive harvests in the open-top chamber plots, only the control and early snowmelt plots were monitored in 2012 F I G U R E 7 Microbial biomass C (MBC) and microbial biomass N (MBN) measured using the chloroform fumigation technique (mean ± SE) during the 2012 season for all four of our treatments (n = 5). Treatments were measured simultaneously but are offset slightly to avoid overplotting there was not strong support for either hypothesis; instead, belowground responses to our early snowmelt treatment were limited in the face of the relatively small changes in soil conditions that our treatments created.

| Plant responses
Because no root production data were available in 2010 and the snowmelt advancement period was small (4 days), we begin by discussing plant responses in 2011. We will subsequently return to the Microarthropod density, cm -3 F I G U R E 9 Soil pore water labile N nitrogen (NO 3 + + NH 4 + + TFPA) mean ± SE by treatment (n = 5). Treatments were measured simultaneously but are offset slightly to avoid overplotting. One outlier (392 µM) is shown by an arrow in the 2011 tussock panel , TFPA, and their sum (labile N) in intertussock soil cores (left) and Eriophorum vaginatum tussock soil cores (right). Note differing scales on the vertical axis. Treatments are offset to avoid overplotting though all treatments were sampled on the same date F I G U R E 11 Exoenzyme activities in intertussock soil cores (left) and Eriophorum vaginatum tussock soil cores (right). The four enzymes measured are shown in the right panels. Note differing scales. Treatments are offset to avoid overplotting though all treatments were sampled on the same date effects likely occurs because plants cannot take advantage of early snowmelt due to unfavorable conditions during that time (Bokhorst, Bjerke, Tømmervik, Callaghan, & Phoenix, 2009;Wipf et al., 2009).
During the 15-day snowmelt advancement period in 2011, our snow-free plots were exposed to very cold air temperatures for ~10 days, suggesting that temperatures during an early snow-free period determines whether plants can begin to grow earlier. Similar interannual variation in the size of the effect of early snowmelt was observed in a previous multiyear early snowmelt study (Khorsand Rosa et al., 2015). In that study, bud break occurred several days earlier depending on species, but in some years, there appeared to be little difference. Air temperature was cited as a strong predictor of phenology, and this is in line with the interannual differences we see here.
Plant root production showed evidence of reduction in 2011 when we pooled the two warming treatments together, providing some support for our second hypothesis that early snowmelt can negatively affect root growth. This finding is consistent with other snow removal studies in showing that cold temperatures during early snowmelt years can damage plants (Inouye, 2008;Wipf et al., 2006Wipf et al., , 2009. However, aboveground plant phenology and NDVI did not show the same type of negative response to early snowmelt (it was weakly positive), which is counter to what we would expect since aboveground tissues are often more sensitive to frost damage than roots (Larcher, Kainmüller, & Wagner, 2010). This lack of aboveground effect suggests that damage to aboveground parts of the plant was not the mechanism behind the reduced root growth.
Although we are not sure of the mechanism, one possibility is the fluctuations we observed in soil temperatures. During the 15-day advancement of snowmelt, the minimum soil temperature did not drop below the controls, as was seen in a similar early snowmelt experiment (Wipf et al., 2009), but soils did refreeze and undergo daily temperature cycles. Such daily temperature cycling was not seen in the controls. These freeze-thaw cycles could damage plant roots (Tierney et al., 2001). The root growth reduction effect was only present in non-E. vaginatum roots (possibly because E. vaginatum root systems are annual and the plants did not have roots at that time), suggesting that in our system, different species may be more tolerant of exposure to early frost conditions than others.
In 2012, the ecosystem exhibited a markedly different response to our treatments, with a larger advancement of aboveground plant phenology in response to early snowmelt and warming treatments.
Unlike in 2011, the NDVI responses were noticeable, and although small in magnitude due to the narrow range created by our measurement approach and calculation as trajectories instead of raw values, the data do suggest a difference in the timing of vegetation greening.
These results suggest that in years with warmer early spring temperatures such as 2012, plants can take better advantage of early snowmelt by leafing out earlier (Livensperger et al., 2016). The acceleration of phenology in our OTC treatments is also consistent with other OTC warming experiments that have shown earlier bud break and flowering Henry & Molau, 1997). However, despite this much larger aboveground plant response, we did not see evidence for a commensurate effect on root production in either tussock or intertussock roots. Part of the reason may be that root growth peaks later than shoot growth, more toward the middle of the season (Kummerow & Russell, 1980). By the time roots were reaching maximum growth, the effects of the early snowmelt may no longer have had a notable influence.
Given the lack of evidence for a root growth effect in 2012 when there was a much larger advancement in snowmelt, it is unlikely that a large plant root response was present in 2010, either, when we had a smaller advancement in snowmelt (4 days

| Decomposer responses
The early-season treatment responses in microbial biomass and potential enzyme activities suggest that both warming and early snowmelt (at least in tussocks) caused a small advance of the seasonal microbial succession patterns that have been observed in association with snowmelt in tundra ecosystems (Lipson, Schadt, & Schmidt, 2002;Schmidt et al., 2007). This finding makes sense with the expected physiological effects of a warmer soil environment Wallenstein et al., 2009), but contrasts with one study that suggests the opposite effect (Tan, Wu, Yang, & He, 2014). The earlier onset of isothermal (0°C) conditions in the upper soil in multiple treatment years likely stimulated soil microbial communities as unfrozen water increased in soil (Mikan, Schimel, & Doyle, 2002;Monson et al., 2006). The cycling patterns in soil temperature noted above in the context of the roots may also have affected microbial activity levels and turnover (Yergeau & Kowalchuk, 2008). The reason for the lack of a detectable effect in the intertussock early snowmelt treatment is unclear, but could be linked to the weaker soil temperature response observed in the intertussock iButton temperature data. The disappearance of this early-season effect by the second measurement date suggests that microbial communities rapidly responded to the convergence of environmental conditions, an effect that could be driven by their rapid generational turnover times even in tundra soils . This small and transient effect is not consistent with either of our hypotheses.
Our data instead suggest that while early snowmelt may alter the timing of microbial activities early in the season, these effects will not cascade into season-long changes to ecosystem function.
Densities of microarthropods were low during the early-season sampling (7 June) in both the early snowmelt and control treatments, suggesting that earlier snowmelt does not necessarily lead to earlier microarthropod activity, unlike in warming experiments where more substantial responses have been documented (Dollery, Hodkinson, & Jonsdottir, 2006;Hodkinson et al., 1998;Sistla et al., 2013). The lack of treatment response could be because Arctic microarthropods are already well adapted to the freeze-thaw cycles observed in both the control and early snowmelt treatments (Konestabo, Michelsen, & Holmstrup, 2007;Sjursen, Michelsen, & Holmstrup, 2005) and because while these organisms are sensitive to low temperatures after resuming activity in the springtime, soil temperatures were still high enough for activity (Hodkinson et al., 1998). The microarthropods that we sampled are also primarily saprotrophs and thus do not rely directly upon the living plant community (Koltz, Asmus, Gough, Pressler, & Moore, 2018), which may buffer them from some of the resource-associated effects of early snowmelt. Had we sampled belowground herbivores in 2011 for example, we may have seen greater response due to reductions in root growth of some species. Populations of mites and Collembola increased between our first two measurement dates, 7 June and 18 June. Based on the limited differences in the microbial biomass among treatments after the first measurement date, it is possible that an overall return to control-like conditions in the early snowmelt plots occurred quickly enough to prevent a measurable impact on these organisms. Overall, these findings suggest that modest changes in the timing of snowmelt alone in the tundra may not have a major effect on the phenology or density of soil-dwelling microarthropods, particularly on those that emerge well past snowmelt.

| Soil nutrient responses
We observed two main treatment effects in soil nutrient dynamics, a transient effect, similar to microbes, in the extractable nutrients in 2012, and a season-long reduction in soil pore water N in the early snowmelt tussock plots during the 2011 season. While not undetectable or negligible, these effects were situational and do not amount to strong evidence in support of our hypotheses, which would predict larger alterations in the timing of important nutrient cycling events.
The OTC warming treatment, more so than the combined treatment, enhanced N cycling at the time of soil thaw, with higher microbial N, labile N, and N-acquiring enzyme activity. While these differences were not apparent later in the season, they indicate that previous studies in which only a few measurements were taken throughout the summer (Hobbie & Chapin, 1998;Jonasson, Havström, Jensen, & Callaghan, 1993;Kudernatsch, Fischer, Bernhardt-Romermann, & Abs, 2008) could miss some of these short-lived effects. The cause of the season-long reduction in soil pore water nutrients in 2011 is not known. It could be linked to the observed root growth reductions via a mechanism such as rhizodeposition (Cardon & Gage, 2006;Weintraub et al., 2007). However, we saw the nutrient reduction in tussock soils while the root reduction was in non-E. vaginatum roots (though on the other hand, we note that non-E. vaginatum roots such as those from V. vitis-idaea are often present in E. vaginatum tussocks). Regardless of the mechanism, because this effect was only apparent in one soil type (tussock) and was only seen in one of 3 years, it does not represent strong evidence that soil nutrient dynamics will be altered in a consistent and predictable fashion by early snowmelt.
Similar to previous studies, we saw evidence of a nutrient crash in the form of low soil-core-extractable N concentrations by our third measurement date in 2012 (Weintraub & Schimel, 2005). The soil pore water lysimeter data showed different patterns, with no clear crash. This difference in dynamics was likely due to the substantially different nutrient pool that soil pore water lysimeter measures. The soil cores capture adsorbed and possibly physically protected N in addition to the labile N captured by soil pore water lysimeters that is mobile in the soil pool (Darrouzet-Nardi & Weintraub, 2014). The crash was more evident in the depletion of this more protected pool, with the more mobile pool staying at relatively low levels for most of the season, consistent with the idea that plants and microbes will draw levels of easily available limiting nutrients down to very low levels (Hobbie & Hobbie, 2012). Despite this observation of nutrient crash dynamics in the soil core extractions, the hypothesis that the timing of the crash would change with our treatments can be rejected. In the soil core extractions, the reduction of extractable N to low levels occurred simultaneously among treatments. Furthermore, we saw no evidence of large timing differences in the soil pore water samples, with seasonal trends among treatments tracking one another well.

| Interannual variation
The interannual variations in early-season labile N, microbial biomass N, and N-acquiring enzyme activity suggest that warmer overwinter temperatures led to enhanced overwinter N cycling activity that can increase nutrient availability at the time of thaw and throughout the following growing season. The soil pore water N data were particularly striking: following a warmer 2010-2011 snow season, the 2011 soil pore water N concentrations were higher throughout the season. Previous studies have suggested that higher microbial biomass N at the end of winter can affect the total amount of N available to plants at the beginning of the growing season (Buckeridge & Grogan, 2008). Also, overwinter soil enzyme potentials are often highest at the end of winter, just before soil thaw (Wallenstein et al., 2009). The importance of overwinter temperatures in regulating soil processes is supported by the results of snow fence manipulation studies, which have shown that deeper snowpacks can enhance N cycling by warming soils (Borner et al., 2008;DeMarco, Mack, & Bret-Harte, 2011;Natali et al., 2012;. While we did not observe an association between snow depth at thaw and overwinter temperatures in 2011 versus 2012, this may be due to differences in air temperatures relative to the timing of snowfall between the 2 years. However, regardless of snow depth at thaw, the associations with warmer overwinter temperatures suggest a causal effect. Microbes can be active at temperatures below 0°C, and thus, even small increases in winter temperatures may have substantial effects on overwinter N mineralization (Mikan et al., 2002;Monson et al., 2006). These interannual differences support our conclusion that our manipulations had less impact on the belowground ecosystem than expected because they did not have a large enough effect on soil temperatures and thaw depth. Large interannual variations are big enough to drive major differences in these belowground processes, but a modest change in snowmelt timing with little concomitant change in soil temperatures for about 2 weeks was not enough.

| CON CLUS ION
We did find some effects of our treatments on belowground processes. We were able to document several clear differences among treatments, including the reduced nutrient availability in tussock soils in 2011, reduced root growth in non-E. vaginatum roots that same year, and the clear separation of treatments in soil microbial biomass, enzyme activity, and nutrient cycling on the first measurement date in 2012. However, our documentation of these small effects also made clear, by contrast, the absence of larger and more consistent effects, such as those we had hypothesized and those that we observed between years. Despite not seeing the hypothesized large belowground effects, we have demonstrated aboveground phenology responses based on interannual variation in climate, and highlighted the physical effects that may occur in soils when snow is removed early, including earlier and more dramatic diurnal temperature cycling. For tundra ecosystems, resistance of the belowground environment to aboveground conditions such as earlier snowmelt, and the rapid capability of belowground organisms to respond to new microclimate conditions could provide some buffering against incremental shifts in climate. In the context of global change in Arctic tundra ecosystems, these results suggest that modest changes in the timing of snowmelt that are not accompanied by warmer air temperature will likely have small effects on ecosystem function, particularly below ground. F I G U R E A 5 Raw NDVI data. Each line indicates the time series of NDVI readings from each of our 20 NDVI sensors. Vertical displacement of the seasonal trends is largely due to differences in the ground cover underneath the sensors (i.e., relative abundance of rock vs. plant vs. soil).