Winter climate change is outpacing our conceptual understanding of how winter conditions regulate soil biogeochemical cycling and ultimately impact vital ecosystem services like soil carbon and nutrient retention. In seasonally snow‐covered ecosystems like northern temperate forests, increasingly inconsistent winters lead to less precipitation falling as snow, frequent midwinter snow melting, and the loss of a stable, insulative snowpack. These changes leave soils vulnerable to freezing, freeze/thaw cycling, and increasing dry/wet cycles from added snowmelt and rainwater. To uncover how these new winter soil climate conditions alter soil biogeochemistry, we introduce the DeFR❆ST (Determining Forest Responses to Snowmelt Treatments) experiment, a novel approach where we melt snow insitu throughout the winter and monitor changes to soil climate, gas exchange, and biogeochemical cycling. We installed DeFR❆ST in a New England temperate forest, an ecosystem that is part of the most significant global carbon sink and is also in the epicenter of winter climate change in the US. Experimental snow melting drove soil moisture fluctuations in addition to deep and persistent soil freezing. In turn, soils in melted plots exhibited blocked gas diffusion and lower soil oxygen availability. Oxygen limitation may have driven shifts in soil processes from high redox potential metabolisms like aerobic decomposition and nutrient mineralization towards low redox potential metabolisms like iron reduction and the dissolution of iron and carbon from organo‐mineral associations. As these changes snowball, altered soil properties and shifts in soil microbial community structure and function could reshape forest biogeochemical cycling, both in these forests and more broadly across seasonally snow‐covered ecosystems.
]]>For decades, our understanding of how climate change will alter the soil processes that regulate global biogeochemical cycling has been grounded in warming trends and growingseason dynamics. However, temperatures are increasing more rapidly in the winter than in the growing season (National Centers for Environmental Information, 2024), particularly in the ecosystems with seasonal or intermittent snow cover that constitute roughly half of the land area in the northern hemisphere (Zhang et al., 2003). In the seasonally snow-covered ecosystems of the Northeastern United States, the duration of winter weather with below-freezing temperatures and snow cover is becoming shorter (Contosta et al., 2020) and this cold weather is increasingly interrupted by winter "heat waves" that melt snow and increase the proportion of winter precipitation falling as rain (Casson et al., 2019;Sheridan & Lee, 2018). These changes inhibit the accumulation and persistence of an insulative snowpack that typically maintains stable wintertime soil temperatures and moistures (Brown & DeGaetano, 2011;Edwards et al., 2007;Groffman et al., 2006). As a result, winter climate change can lead to colder soils with greater moisture and temperature fluctuations, which may shift abiotic and biotic soil metabolisms and processes (Fig. 1). Given our reliance on soil ecosystem services (e.g. carbon sink, Bossio et al., 2020;Georgiou et al., 2022;Lal, 2005), it is imperative to better understand how winter climate change will alter soil biogeochemistry in the short and long term.
Manipulative in situ climate experiments are powerful tools that shape our understanding of how terrestrial ecosystems will feed back to climate change (Song et al., 2019). Although the majority of temperature manipulation experiments in seasonally snow-covered ecosystems neglect winter climate change by either only simulating growing season warming or maintaining constant, year-round warming (Kreyling, 2010;Sanders-DeMott & Templer, 2017), the importance of wintertime biogeochemistry research is becoming increasingly recognized (Casson et al., 2019;Contosta et al., 2019;Kreyling, 2020;Ladwig et al., 2016;Rixen et al., 2022). The main approach of winter climate change manipulations has been manually removing or physically blocking the insulative snowpack and inducing changes to soil temperature and freeze-thaw cycling (e.g. Broadbent et al., 2021;Groffman et al., 2006;Natali et al., 2011;Templer et al., 2017). Only a handful of short-term studies have both removed the snowpack and compensated for meltwater loss during or after the winter with artificial water additions (e.g. Sulkava & Huhta, 2003;Tan et al., 2014). As such, the majority of snowpack removal experiments also shovel away potential water inputs and reduce soil moisture (Zhao et al., 2022), likely failing to capture the full effect of winter climate change which could include pulses of water entering the soil from rain and snowmelt and soil frost layering (Lundberg et al., 2016).
Another established method is the use of infrared heaters to simulate winter warming and remove the snowpack (e.g. Hutchison & Henry, 2010;Kimball et al., 2008;Lí et al., 2024). However, results from an infrared snowpack manipulation experiment in a subalpine forest suggest that these heaters promote sublimation water losses of up to 100% of precipitation (Meromy et al., 2015). To our knowledge, there are no long-term ecosystem manipulation experiments that investigate how winter climate change alters temperate forest soil processes by melting snow in situ without maintaining consistently elevated temperatures or restricting soil water inputs.
Together, soil water from rain or snowmelt and deeper soil frost formation could slow gas diffusion through the soil profile, limiting oxygen availability and dropping soil reductionoxidation (redox) potentials (Fig. 1). Redox reactions control biogeochemical cycling, and fluctuations in redox potential can profoundly change soil function (Lacroix et al., 2023). In forest soils, as oxygen availability declines and redox potential decreases, dominant soil metabolisms like aerobic decomposition and nitrification may cede to processes like denitrification or iron reduction (Kleber et al., 2021). Here, we hypothesize that midwinter snow melting causes a flush of meltwater through the soil profile and leaves soils vulnerable to freezing depending on ambient temperatures, which: 1) slows gas diffusion through the soil profile and promotes low redox, anoxic conditions; 2) inhibits aerobic metabolisms like decomposition, nutrient mineralization, and nitrification, leading to a reduction in wintertime nutrient recycling; and 3) enhances anaerobic metabolisms like manganese and iron reduction, leading to the dissolution of nutrients and organic matter bound to soil minerals. Further, we expect that these effects will 4) drive long-term changes in biotic (e.g. shifts in microbial community structure and function) and abiotic (e.g. reductions in mineral sorption capacity) factors that regulate soil carbon storage and nutrient cycling. Collectively, we hypothesize that snow melting will fundamentally change soil functions during the winter with carryover effects throughout the year and across years, as the effects of winter climate change snowball (Fig. 1).
To address these hypotheses, we introduce the DeFR❆ST (Determining Forest Responses to Snowmelt Treatments) experiment. DeFR❆ST melts the snowpack in situ to uncover how climate change introduces novel winter soil conditions that could fundamentally alter ecosystem biogeochemistry. We have located the DeFR❆ST experiment in the epicenter of winter climate change in the US-New England-to better understand the impacts of winter climate change in a relevant ecosystem. Further, the fate of the forests dominating this region is globally important, as northern temperate forest ecosystems are a powerful carbon sink (Harris et al., 2021). Here, we describe this novel snowmelt experiment and present findings from the first winter of snow melting.
Study site:
The DeFR❆ST experiment is located at the Dartmouth Skiway in Lyme, NH (43° 79'N, -72° 10'E; 325 m). Mean annual air temperature is 6.6°C, ranging from mean monthly temperatures of -13.8°C (low) in January to 27.4°C (high) in July. Mean annual rain-equivalent precipitation is 111 cm per year, with 170 cm of snow per year. All DeFR❆ST plots are located within a 30 x 40 meter forested area on a West-facing slope. Soils are shallow, Berkshire fine sandy loams (USDA NRCS Web Soil Survey, 2024) with an average rock volume of 17% in the top 30 cm and slopes ranging from 2 to 25%. The dominant tree species at the site include American beech (Fagus grandifolia), white birch (Betula papyrifera), sugar maple (Acer saccharum), and balsam fir (Abies balsamea). The site was cleared for agriculture in the early 1800s and was maintained as pastureland prior to 1956, when the Dartmouth Skiway began operation and the areas near the slopes, including our plots, were left to natural afforestation (Ben Kilham, Lyme, NH Selectboard Vice Chair). Site access and use is provided by employees of the Dartmouth Skiway.
The DeFR❆ST experiment simulates winter climate change by melting snow with aboveground heating cables (SI Methods: Snow Melting). We turn on heaters for all snow events exceeding 5 cm accumulation to melt the snowpack and allow meltwater to enter the soil. We turn off heaters after melting to prevent soil warming and leave the soil surface exposed to ambient conditions. The experimental design consists of twelve 4 x 5 m plots in a block design with four replicate blocks of three treatments: total melt (100% melt), moderate melt (50% melt), and control (0% melt; Fig. 2). We melt every snow event in the total melt plots. For the moderate melt plots, we melt snow events at the beginning and end of winter (all snow events before January 1 or, if no snowpack is established before January 1, before January 20, and after March 1) to simulate a shorter snow-covered season and aim for a 50% reduction in the standing snowpack over the winter. We do not manipulate the snowpack in the control plots.
We established the DeFR❆ST plots in August 2023, and we randomly assigned each of the three treatments within each block such that there were no significant treatment differences in measurements of soil temperature and moisture, soil respiration, or tree basal area made before the beginning of the first winter (2023-2024) of the experiment (SI Methods: Pretreatment Site Characterization). Plots are square to the hill slope, with a minimum distance of 1.5 m between plots along the same elevation (within blocks) and a minimum distance of 2.5 m between uphill/downhill plots (across blocks) to limit meltwater infiltration between plots (Fig. 2). We removed dead standing trees and large woody debris exceeding ~12 cm diameter from the plots to allow installation of heating cables and equipment.
To quantify snow melting treatments, we monitor snow depth within the plots using a combination of snow depth sensors and manual measurements. We installed snow depth sensors 1.5 m above the soil surface over a 0.5 m 2 vegetation-free area of influence that is at least 0.5 m from the edge of the plots in plots A2, A3, B3, C1, D1, and D3 (n=2 sensors per treatment).
Snow depth sensors measure the distance from the sensor to the ground or snow surface every 5 minutes, and we record the average snow depth every 30 minutes during ambient conditions and every 5 minutes during melting events with Campbell Scientific dataloggers. High-resolution snow depth data (Fig. 3b) can allow us to measure new snow accumulation and monitor snow melting in real-time, but this data is less reliable due to low replication, sensor anomalies, and a limited area of influence. We also installed graduated snow stakes at three locations in each plot.
We manually record snow depth when visiting the site for other measurements or maintenance, usually weekly. These data are more accurate to quantify the standing snowpack across all plots, but low-resolution measurements prevent accurate measurements of new snowfall, particularly as snowpack depth constantly changes with gravity compaction and ambient snow melting (Fig. 3a,b).
We characterize ambient snow conditions (snow accumulation, snowpack depth, and snow water equivalent) at nearby reference locations outside the plots. To supplement in-plot measurements of snow event accumulation, we capture hourly trail camera pictures of graduated snow stakes at four reference locations around the plots. To measure the standing snowpack depth and snow water equivalent, we core the snowpack with a 4.13 cm diameter stainless steel tube every 1-2 weeks at six reference locations around the plots. We destructively sample outside the plots to limit snowpack disturbance within the plots; thus, these measurements only reflect the ambient (unmelted) snowpack.
To measure how snow melting alters soil climate, we monitor soil temperature and moisture across the soil profile in each plot with two types of automated soil sensors. First, we installed 30 cm GroPoint Profile soil climate sensors (2625-N-T-2, Riot Technology Corp) in each plot in November 2023. These sensors measure soil temperature at 3.5 cm, 10 cm, 20 cm, and 30 cm depths and measure soil moisture across 0-15 cm and 15-30 cm depths. We use CR1000x dataloggers to measure these soil climate profile sensors every 5 minutes and record the average values every 30 minutes. Rocky soils interfered with soil moisture measurements, so these sensors were reinstalled in May 2024 in soil slurries to ensure a continuous soil connection along the length of the probes. As such, soil moisture timeseries data at both depths starts in June 2024. Second, we also installed soil climate dataloggers (TMS-4, TOMST) in each plot in November 2023. These sensors record hourly measurements of air temperature and topsoil moisture and temperature (6 cm depth). This data cannot be accessed in real-time, and we collect data from these dataloggers seasonally.
To investigate how snow melting alters soil CO2 production and soil oxygen and methane uptake, we measure surface flux and collect gas samples across the soil profile. We take measurements of surface soil CO2 fluxes using infrared gas analyzers (IRGAs). PVC soil respiration collars (83 cm 2 ) are permanently installed in two locations per plot, and we use a portable IRGA (EGM-5, PP Systems) to manually measure net surface CO2 fluxes every 1-2 weeks. Surface fluxes can only be measured across all treatments during the snow-free season.
To evaluate diffusion and model CO2 production and methane and oxygen uptake, we collect and analyze soil gas samples across the soil profile. We use gas concentrations across depths, in tandem with measurements of soil moisture, bulk density, and rock volume to estimate soil porosity, to model a diffusion gradient and calculate gas production or uptake as described in Hicks Pries et al., 2017 andHicks Pries et al., 2020. In October 2023, we installed 6.3 mm stainless steel gas wells at 5, 10, 20, and 30 cm depths in two locations in each plot. After clearing the gas well headspace, we collect 30 mL (5 cm depths) or 60 mL (10-30 cm depths) samples with a syringe and transport them back to the lab for analysis within six hours. We similarly collect known gas standard curves to normalize measurements and correct for drift due to equipment or gas diffusion over this time. We inject samples through an O2 meter (Fibox 4, PreSens-Precision Sensing GmbH) in line with either a portable IRGA CO2 analyzer (EGM-5, PP Systems) to measure CO2 or a gas sample isotope analyzer (G2201, Picarro) to measure CO2, CH4, 13 CO2, and 13 CH4. We take these measurements every 1-2 weeks during the winter and spring and monthly thereafter. Due to the longer time required for Picarro measurements (5 min/sample) compared to the portable IRGA (1 min/sample), we collect Picarro measurements once a month in the winter and spring and once per season in the summer and fall for only the control and total melt plots.
To measure how snow melting alters carbon, nutrients, and metals dissolved in soil water, we sample and analyze soil porewater from suction lysimeters (WS1900, Soil Moisture) installed at 30 cm depths in each plot. One lysimeter per plot was installed in November 2023 and a second lysimeter per plot was installed in August 2024. We collect soil water samples every 1-2 weeks in the winter and spring and monthly thereafter. We apply a vacuum to the lysimeters 48-72 hours before collection to draw soil pore water into the sampler through a porous ceramic cup and extract these samples in the field with a vacuum pump. We filter samples to 0.45um and preserve subsamples in the field with nitric acid to 2-3 pH to prevent dissolved metal precipitation. We return non-acidified and acidified samples to the lab and start analyses or freeze samples within four hours of collection.
We measure dissolved inorganic and organic carbon concentrations from non-acidified soil water samples with a total organic carbon analyzer (M5310C, Sievers). We measure dissolved metals concentrations from field-acidified soil water samples on an ICP-MS (ICP-OES SPECTRO ARCOS, SPECTRO Ametek). We measure phosphate concentrations on subsamples preserved with sulfuric acid prior to analysis by flow injection (Lachat QuickChem 8500 AE, Hach). We measure ammonium and combined nitrite + nitrate concentrations from non-acidified soil water samples with colorimetric assays using the salicylate-nitroprusside method and vanadium(III) reduction, respectively (Doane & Horwáth, 2003;Verdouw et al., 1978) followed by microplate spectrophotometry (Spark10M, TECAN).
To characterize pretreatment soils, we collected samples at 0-10 cm, 10-20 cm, 20-30 cm, and 30-40 cm mineral soil depths at three locations per plot in November 2023. We brought soils back to the lab within 4 hours of collection and froze soils at -20°C until processed. We homogenized soils by sieving through a 2 mm mesh after removing roots, and we separated representative subsamples to measure pH from field-moist soils in 1:1 or 1:2 soil:water slurry with a benchtop pH meter (S220, Mettler Toledo) and bulk %C and %N from dried (60°C), finely ground soils with an elemental analyzer (Costech ECS 410).
To assay how snowmelt treatments altered soil microbial community function, we collected samples at 0-10 cm, 10-20 cm, and 20-30 cm mineral soil depths in each plot after the end of the melting treatment season in May 2024. We homogenized soils via 2 mm sieving within 24 hours of collection. We measured potential rates of net N mineralization and nitrification by immediately extracting 2.5 g representative subsamples of soil in 2M potassium chloride (KCl) and then aerobically incubating a second 2.5 g subsample for two weeks in the dark at room temperature prior to extraction. We measured inorganic N concentrations in the soil KCl extracts as described above for dissolved nutrients in soil water. We calculated net N mineralization potential as the difference in total KCL-extractable inorganic N from the beginning to end of the 2-week incubation and net nitrification potential as the difference in nitrate + nitrite over this incubation (Waring et al., 2016).
All statistical analyses were performed with R software (version 4.4.0, 2024). We assessed the effects of the snow melting treatments on the average snowpack depth, soil moisture, and soil water measurements during and outside of the winter using linear mixed effects models and analysis of variance with plot as the random effect (lme4 package, version 1. 1-35.3, Bates et al., 2015). We also included depth in the soil profile as a factor for applicable datasets, including soil temperature, gas concentrations through the soil profile, and measurements made from collected soils. We assayed interactive effects between treatments, continuous predictors (e.g., depth in the soil profile), and continuous response variables (e.g., dissolved carbon or iron concentrations) with linear regressions. We modeled differences in soil gas concentrations as a function of depth in the soil profile, where a positive slope indicates the presence of a diffusion gradient to the atmosphere, a negative slope indicates the presence of a diffusion gradient from the atmosphere, and no slope indicates no diffusion gradient. We used post-hoc testing to compare data means or regression coefficients across treatment groups (emmeans package, version 1.10.4, Lenth et al., 2024). We used a threshold of p<0.05 for statistical significance.
We defined our winter snow melting treatment season as the date range from January 8 (first snowfall, beginning of melting treatments) until March 21 (first day with no remaining snowpack and above-freezing surface soil temperatures in all treatments) for the first year of the DeFR❆ST experiment unless otherwise stated (e.g., we compare soil moisture during the winter snow melting treatment season before frozen soils inhibited accurate soil VWC measurements, January 8 -24, 2024, Fig. 5). Because data interpretation may differ depending on how the melting treatment season is defined in our experiment, we also present data for the winter treatment season before and after soil freezing, astronomical winter (December 21, 2023-March 19, 2024), astronomical spring (March 20, 2024-June 19, 2024), and all winter and spring dates (SI Table 2).
Prior to data analyses, we removed outliers from the snowpack depth sensor data and from the soil water or extract nutrient and dissolved metal concentration data. Snow depth sensors produce short-term (on the scale of several minutes) and long term (on the scale of several days) data anomalies. We filtered short-term data anomalies from the dataset by removing datapoints that indicated infeasible snowpack accumulation or losses exceeding 10 cm/hr (extremely high snowfall rate for our region is ~4 cm/hr, NOAA). These anomalies represent less than 5% of the data. We removed soil water or extract measurements that exceeded three standard deviations of the dataset mean (dissolved iron, n=2; extractable soil ammonium, n=2; extractable soil nitrate, n=1 measurements removed).
Snow melting treatments successfully shrunk the snowpack Melting significantly altered the average depth of the uninterrupted snowpack (manual measurements, treatment: p<0.001, F2,9=65.4, Fig. 3a; sensor measurements, treatment: p=0.035, F2,3=12.3, Fig. 3b). The average snowpack depth in the control plots (10.3 ± 0.6 cm) was roughly twice that of the moderate melt plots (5.4 ± 0.5 cm) during the melting treatment season from Jan 8-Mar 20, 2024 (Fig. 3b, SI Table 2). Maximum daily snow depths exceeded 5 cm for 52.6 ± 2.3 days in the control plots, 23.4 ± 1.8 days in the moderate melt plots, and 3 ± 0 days in the total melt plots (Manual measurements, SI Table 2).
The average duration of a melting event was 22 hours, but the melting efficiency depended on ambient conditions. Fully melting the snowpack took as long as 4 hours per centimeter of snowfall when ambient air temperatures were very low (average air temperature -10.7°C from 1/17/24-1/19/24), or was as quick as 45 minutes per centimeter of snowfall when air temperatures were above freezing (average air temperature 2.1°C on 3/10/24) (SI Table 1).
Snow melting treatments resulted in drastic soil climate shifts during the winter. Air temperature data collected at 2 cm aboveground in each plot demonstrates how differences in the insulative snowpack maintained more stable and often warmer surface temperatures in the control plots than the melted plots, particularly at the end of January (SI Fig. 1). On average, soils in melted plots were colder (treatment: p<0.001; F2,9=31.5) and froze more deeply (treatment x depth: p<0.0001; F6,163971=294) than the control plots during the melting treatment season. The surface soils (3.5 cm) were at or below freezing (< 0.2°C) for 89 ± 21.6 days in the total melt plots, 71 ± 17 days in the moderate melt plots, and 9.6 ± 2.6 days in the control plots (Fig. 4, SI Table 2). Treatment effects were even more influential at 10 cm depths, where soils were at or below freezing for 116 ± 4.8 days in the total melt plots, 50 ± 6.7 days in the moderate melt plots, and 2.6 ± 2.6 days in the control plots (Fig. 4, SI Table 2). Only soils in the total melt plots froze at 20 cm, with 17.5 ± 17.1 days at or below freezing. Three sequential melting events drove corresponding pulses in surface soil moisture before surface soils froze in melted plots, and these fluctuations were absent in the control plots where snow events were not melted (Fig. 5).
Snow melting treatments blocked CO2 diffusion out of the soil profile when soils were frozen (treatment x date range x depth: p=0.0109, F1,856=6.52) and drove soil CO2 accumulation (treatment x date range: p<0.0001, F1,856=18.4). The confidence interval for modeled CO2 across depth overlapped zero (suggesting no diffusion gradient) after soils froze during the melting treatment season in the total melt plots (slope 16.1, CI -2.3 to 34.4 ppm/cm, Fig. 6a). By contrast, we observed consistently increasing CO2 with depth (suggesting gas diffusion from the soil to the atmosphere) in the control plots over the same date range (slope 37.5, CI 21.3 to 53.8 ppm/cm; Fig. 6a, orange vs. blue in left plot) and in the melted plots over the remaining dates (slope 68.2, CI 57.2 to 79.3 ppm/cm; Fig. 6a, orange in left vs. right plot). Further, the average CO2 concentration across the soil profile in the total melt plots when the soils were frozen (3901 ± 304 ppm) was more than double that of either the control plots over the same date range (1480 ± 208 ppm; p<0.0001) or the total melt plots during the remaining dates when the soil was not frozen (1818 ± 384 ppm; p<0.001; Fig. 6a).
Snow melting treatments reduced soil O2 concentrations (treatment x date range: p<0.0001, F1,800=45). Unlike CO2, O2 declined with depth in all models (suggesting gas diffusion from the atmosphere into the soil profile; SI Table 4). However, the average O2 concentration across the soil profile was significantly lower for the total melt plots when soils were frozen (20.3 ± 0.03 %) than the control plots over the same date range (20.65 ± 0.03 %; p<0.0001) or than the total melt plots during the remaining dates when the soil was not frozen (20.59 ± 0.02 %; p<0.0001; Fig. 6b).
Snow melting treatments led to emerging patterns in indicators of winter soil biogeochemical processes including aerobic decomposition, microbial nutrient recycling, and organo-mineral destabilization.
Soil CO2 production, an indicator for aerobic decomposition, was significantly higher for the control than total melted plots at the beginning of the melting treatment season before gas diffusion was blocked (treatment: p<0.01, F2,77=6.47, SI Fig. 2). We also observed a trend, although not significant (treatment: p=0.19), toward inhibited surface CO2 fluxes in the melted treatments from the only measurement that was possible during the melting treatment season (SI Fig. 3).
Mineralization of inorganic N, an indicator of microbial nutrient recycling, was positive in control plots (0.24 ± 0.15 ug N g soil -1 day -1 ) and tended to be higher than that of soils from melted treatments (total melt: 0.05 ± 0.10 ug N g soil -1 day -1 ; moderate melt: -0.30 ± 0.13 ug N g soil -1 day -1 ) at the end of the winter (treatment: p=0.06, F2,9=3.86; Fig. 7a). Nitrification potential tended to be lower for the moderate melt soils than the control soils (SI Fig. 5). Soil porewater in control plots tended to have reduced ammonium and nitrate concentrations compared to melted plots during the winter (SI Fig. 6).
Correlations between dissolved carbon and iron, an indicator of organo-mineral dissolution from iron reduction, depended on melting treatment and season (dissolved iron x treatment x season interaction: p<0.0001, F2,120=13.7; Fig. 7b). Dissolved iron and dissolved carbon in soil water samples collected during the winter season were only related in the melted treatments, and in the control plots, the slope's confidence interval overlapped zero suggesting no iron-carbon correlation (total melt: slope 214, CI 159 to 269 ppm/ppm; moderate melt: slope 46.2, CI 28.1 to 64.3 ppm/ppm ; control: slope 22.1, CI -2.60 to 46.9 ppm/ppm). We observed no significant differences between the treatments outside of the winter (SI Fig. 7).
The DeFR❆ST experimental manipulation successfully melted the snowpack (Fig. 3), with cascading impacts on soil conditions and biogeochemical processes. Snow melting led to a reduction in winter soil temperatures and deeper soil freezing (Fig. 4) and more variable midwinter soil moisture conditions resulting from melting events (Fig. 5). These changes to soil climate inhibited gas diffusion through the soil profile and led to lower soil oxygen levels in the melted plots during the winter (Fig. 6). In turn, we observed emerging patterns in microbial respiration, nutrient recycling, and dissolved carbon and metals in soil water (Fig. 7) that support our hypotheses that winter climate change shifts soil redox conditions and may drive fundamental changes to soil biogeochemical cycling.
Our experimental approach using aboveground heating cables allowed us to realistically simulate midwinter snow melting. Snow loss rates with experimental melting (ranging from 0.3-1.3 cm snow loss/hr, SI Table 1) were comparable to the fastest control snowpack reduction rate we observed (~1 cm/hr, Fig. 3) and fell within the range of observations from a nearby snow monitoring site in Corinth, VT (up to 32 cm snow loss/day or 1.33 cm snow loss/hr, Global Historical Climatology Network Daily (GHCNd)). Melting events left some snow remaining between melting cables because our cables are not embedded in a mass substrate like soil to facilitate heat distribution (e.g. Fig. 3c). However, snow loss was consistent across the plots and any remaining snow between cables rarely persisted beyond a few days after experimental melting, likely due to ambient melting, snow sublimation, or a local decrease in albedo from having half or more of the soil surface exposed. Similarly, results from a whole-ecosystem warming experiment show how declining albedo under limited snow cover increases near surface air temperature (Richardson et al., 2024).
The snow melting treatments supported our hypotheses for how winter climate change will impact the soil environment (Fig. 1). A deeper control snowpack buffered changes in nearsoil air temperature (SI Fig. 1), and snowpack melting increased the depth and extent of soil freezing (Fig. 4). This pattern aligns with the established potential for colder winter soils in a warming world (Groffman et al., 2001) and consistent findings across snow removal experiments (Kosolapova & Altshuler, 2024;Lí et al., 2024;Sanders-DeMott & Templer, 2017;Z. Zhao et al., 2022). These meta-analyses also identify concurrent drier soils with snow removal, but we show that melting treatments promoted soil moisture fluctuations during three sequential melting events at the beginning of the season (Fig. 5). In future years of the experiment, our improved soil moisture measurements down to 30 cm may allow us to better characterize patterns that were not measurable when soils froze deeper than our topsoil moisture sensors (Fig. 4). These induced environmental changes could reshape soil microbial communities and function (e.g. de Nijs et al., 2019;Lí et al., 2024;Pastore et al., 2023). Soil moisture fluctuations have drastic and lingering impacts on soil microbial community structure and activity that can even exceed the effects of freeze/thaw cycles (Meisner et al., 2021). Further, elevated soil water content from snowmelt or winter rain can amplify the impact of soil freezing and freeze/thaw cycling as water expansion upon freezing can mechanically shear roots and fungal hyphae, lyse microbial cells, fragment soil aggregates, and reduce soil pore connectivity (Matzner et al., 2008). Repeated water additions to frozen soils could promote the formation of concrete rather than granular soil frost, which determines whether freezing alters soil nutrient cycling (Patel et al., 2018).
Meltwater likely contributed to the formation of solidly frozen soil layers with a low permeability to gas diffusion in the first winter of our experiment, as suggested by our observations of diffusion gradient deterioration, soil CO2 accumulation, and soil O2 depletion (Fig. 6).
Multiple lines of evidence support our hypothesis that snow melting treatments shift soil biogeochemical processes by inducing changes to soil redox potential. First, we observed reduced oxygen levels through the soil profile (Fig. 6b). While these changes seem small (~3% O2 reduction), these data represent 30 or 60 mL bulk soil air samples collected from soil volumes that could be as large as 500 mL depending on localized porosity and soil moisture. Anaerobic soil conditions exist in microsites throughout the soil matrix where O2 consumption exceeds new O2 supply (Lacroix et al., 2021;Tiedje et al., 1984) and these microsites are not fully captured by bulk O2 measurements (Keiluweit et al., 2018). Here, we argue that blocked gas diffusion and a significant drop in the bulk O2 levels when soils are frozen in melted plots is a strong indication of a climate-driven promotion of anaerobic microsites with low redox potential. Second, we observed patterns in biogeochemical processes that could result from altered redox conditions. Soils in the melted plots tended to exhibit less CO2 production (SI Figs. 2, 3) and inhibited N mineralization (Fig. 7a) and nitrification (SI Fig. 5) potentials. While these patterns could be attributed to slowed microbial activity in colder soils (Fig. 4), they also align with a shift away from high redox potential, oxygen-reducing metabolisms like aerobic decomposition and nitrification. Further, our data suggests the potential for snow melting to enhance low-redox processes like iron (III) reduction. Iron oxides in soils have a strong organic matter sorption capacity but are susceptible to reductive iron dissolution (Coward et al., 2017;Li et al., 2023), where the typically persistent organic matter bound to these minerals can be liberated in lowredox conditions (Chen et al., 2020;Hall & Silver, 2013;Lovley, 1991;Pan et al., 2016;Patzner et al., 2020). Here, our observation that dissolved organic carbon and iron in soil water samples from the winter treatment season are only correlated in the melted treatments (Fig. 7b) supports our hypothesis that snow melting treatments promote the dissolution of iron minerals and mineral associated organic matter by driving redox fluctuations and low redox conditions (Fig. 1). In future winters, we plan to directly measure soil redox potential throughout the winter to mechanistically link treatment effects on soil climate with changes in soil biogeochemical processes resulting from redox changes.
While we posit that melting snow may impose more drastic winter redox fluctuations in upland soils than observed with snow manipulations that restrict soil water inputs, the potential for redox conditions to alter soil biogeochemistry can be found in other winter climate change studies. Some snow removal experiments have found elevated winter soil CO2 concentrations and/or large pulses of CO2 release upon spring thaw (e.g. Gavazov et al., 2017;Monson et al., 2006). These patterns are often attributed to mechanisms like the birch effect or accelerated microbial turnover under freeze-thaw cycles, but also align with soil CO2 accumulation from inhibited diffusion out of the soil that could simultaneously limit soil O2 supply (Fig. 6; Maljanen et al., 2007;Raz-Yaseef et al., 2017). Further, soil hypoxia has been observed in ice encasement experiments where snow is compressed and watered to form ice layers within the snowpack that block gas diffusion (Martz et al., 2016;Watkins et al, 2025;Milner et al., 2016). In turn, multiple studies demonstrate the potential for low-redox, anaerobic biogeochemical processes in upland soils with winter climate change. For example, winter N2O fluxes, a product of biogeochemical reactions that occur in suboxic or fluctuating redox conditions, increase in response to soil freezing with snow removal (Blankinship & Hart, 2012;Groffman et al., 2006Groffman et al., , 2011;;Maljanen et al., 2010;Vankoughnett & Henry, 2014) and snowmelt (Morse et al., 2015;Risk et al., 2013).
Denitrifier abundance has been shown to increase with snow removal and soil freezing (Jusselme et al., 2016;Sharma et al., 2006). While snow removal studies have generally found that soil freezing increased inorganic nitrogen concentrations, results have been highly variable for NO3 - (Fitzhugh et al., 2001;Groffman et al., 2001Groffman et al., , 2006;;Patel et al., 2018). This variability may be due to differences in the soil volume affected by and the degree of reducing conditions needed for denitrification and dissimilatory nitrate reduction to ammonia. Snow removal can also decrease CH4 uptake (Borken et al., 2006;Groffman et al., 2006), a mostly aerobic process, suggesting lower redox conditions and diffusion limitations.
We have not found published studies investigating the potential for iron reduction in snow manipulation experiments, which often investigate different primary questions (Rixen et al., 2022). However, the potential for organic matter destabilization due to iron reduction under dynamic redox conditions in upland soils has become increasingly recognized (Hall et al., 2016;Lacroix et al., 2025;Q. Zhao et al., 2017). Studies in permafrost and tropical forest soils demonstrate how oxygen limitation during intermittent events like permafrost thaw or intense rain can mobilize iron and carbon (Barcellos et al., 2018;Patzner et al., 2020). In temperate or boreal forest ecosystems, reducing soil conditions in winter have been linked with the export of dissolved organic carbon and/or iron in streamwater (Ekström et al., 2016;Knorr, 2013;Tittel et al., 2022). Across a meta-analysis of snow manipulation experiments, snow removal increased soil dissolved organic carbon, which has been attributed to freeze-thaw cycling and aggregate dispersal but could also partially result from organo-mineral destabilization (Hua et al., 2024, but see Zhao et al., 2022). Ultimately, our soil porewater data, which reflects active and rapidly changing biogeochemical processes, may be an early indicator for treatment-induced shifts in soil organic matter that typically require longer timescales to detect due to spatial heterogeneity and sampling limitations (Post et al., 2001;Schrumpf et al., 2011;Stanley et al., 2023).
If sustained, the snowmelt responses we observed in the first winter of the experiment demonstrate the potential for winter climate change to fundamentally alter soil biogeochemical cycling in ways that could impair forest soil ecosystem services and alter greenhouse gas emissions. For example, impaired aerobic winter decomposition may both limit microbial mineralization of nutrients to fuel plant productivity in the growing season (Fig. 7a; Blankinship & Hart, 2012) and also reduce soil CO2 emissions during the winter (SI Fig. 3). Low-redox iron mineral dissolution could enhance losses of associated carbon and nutrients through soil water leaching, and prolonged redox fluctuations could limit the future formation of new mineralassociated organic carbon by increasing iron mineral crystallinity and reducing specific surface areas for sorption (Hall et al., 2018;Mikutta et al., 2024;Thompson et al., 2006). This process could also stimulate soil CO2 emissions during the growing season from renewed microbial access to labile organic matter (Dong et al., 2023). Over time, soil freezing and redox fluctuations could select for more stress-tolerant, but slower growing, microbial communities, reducing biogeochemical cycling rates (Garcia et al., 2020). A shift towards more anaerobic microbial metabolisms and communities could also increase emissions of more potent greenhouse gases. For example, low redox could decrease aerobic methanotrophy reducing soil CH4 uptake and even increase anaerobic methanogenesis (Megonigal & Guenther, 2008).
Fluctuating redox could also lead to higher rates of incomplete denitrification and potential N2O emissions (Butterbach-Bahl et al., 2013;Groffman et al., 2006). Our continuing work at the DeFR❆ST experiment will help quantify the direction, magnitude, and timescale of forest responses and feedbacks to winter climate change.
Despite operating in the warmest winter on record in New Hampshire in its inaugural season (National Centers for Environmental Information, 2024) and relying on a snowpack from only two snowfalls at the beginning of winter (Jan. 8/9 and Jan.16, 2024, SI Table 1), the DeFR❆ST experiment effectively simulated how snow melting and reductions in snowpack depth and duration transform soil environmental conditions. Data from the control plots in this experiment demonstrates the relevance of simulating winter climate changes at our experimental location. A winter 'heat wave' with above-freezing average daily temperatures from February 25-28 fully removed the remaining snowpack in the control plots (Fig. 3, SI Fig. 1) and a subsequent dip in air temperatures (SI Fig. 1) induced mild soil freezing in the surface soil of the control plots (Fig. 4c). While less severe than our imposed melting treatment effects, this observation underscores the relevance of our research and the importance of studying the impact of more frequent midwinter snow melting in this region.
Uncovering how winter climate change impacts the mechanisms regulating soil and forest ecosystem services is critical, given the rapid and ongoing loss of snowy winter conditions and the regional and global importance of these seasonally snow-covered ecosystems (Burakowski et al., 2022;Contosta et al., 2019;Gottlieb & Mankin, 2024;Studd et al., 2021).
Collectively, we demonstrate the ability of the DeFR❆ST experiment to realistically simulate midwinter snow melting and we present strong evidence that this manipulation alters the soil environment in ways that reshape soil biogeochemical cycling. As this experiment continues, our ability to capture short-and long-term responses to midwinter snow melting will strengthen our predictive understanding of how changing winter climate conditions alter the function of midand high-latitude ecosystems over the next century.
Fig. 1: Hypothesized mechanisms for how accelerating winter climate change may alter forest soil biogeochemical processes. On the left, we depict a consistently cold winter with an insulative snowpack that maintains a stable wintertime soil climate. As winter climate change accelerates, more variable wintertime weather drives midwinter snow melting events that result in a flush of meltwater through the soil profile and the loss of an insulative snowpack on the right. These changes leave soils vulnerable to deep, concrete soil frost layers depending on ambient air temperatures, which can reduce soil oxygen (O2) availability and drive fluctuating or lower redox potentials (Eh). Inhibited aerobic organic matter decomposition, nitrification, and methanotrophy respectively decrease microbial decomposition and carbon dioxide (CO2) respiration, slow inorganic nitrogen (N) recycling and nitrate leaching, and reduce soil methane (CH4) uptake compared to soils in a consistently cold winter on the left. Further, lower soil redox potentials promote anaerobic biogeochemical processes like increased denitrification and nitrous oxide (N2O) production, methanogenesis and CH4 production, or iron reduction and the dissolution of organo-mineral complexes.
Longer-term, winter climate change may drive persistent shifts in soil physical structure and mineralogy, soil organic carbon and nutrient availability, and soil microbial community structure and function. Created in https://BioRender.com Fig. 3: Melting treatments shrunk the snowpack. 3a: Manual measurements of snowpack depth for the total melt (orange), moderate melt (gold), and control (blue) plots. Melting events in the total melt plots are shaded in gray; 3b: Continuous sensor measurements of snowpack depth for the total melt (orange), moderate melt (gold), and control (blue) plots. Sensor data includes anomalies where no new snow events occurred (e.g. Feb 28-Mar 10; Mar 13-Mar 23). Melting events in the total melt plots are shaded in gray. 3c:
In-progress melting of a shallow (<10 cm) snow event at -8°C air temperature; 3d: Complete melting of a deep (>30 cm) snow event at 0°C air temperature. Temperature data is depicted with a manually modified color scale to reflect ecologically significant thresholds, where below-zero soil temperatures are white, transition-zone soil temperatures around 0ºC are depicted in blue, and above-freezing soil temperatures are depicted with a linear color scale in brown. Fig. 5: Sequential manipulated snow melting events (gray shaded regions) promoted fluctuations in topsoil (6cm) moisture in the total melt plots (orange) that were absent in the unmanipulated control (blue). Soil volumetric water content data is presented during the winter before freezing conditions (beginning on Jan 24, 2024) in the melted plots impeded soil water content measurements by the shallow TMS-4 sensors. Post-hoc tests show that the total melt data during the melting treatment season are significantly different from all remaining data for both CO2 (p<0.001) and O2 (p<0.0001) concentrations. Fig. 7: Melting treatments altered indicators of winter biogeochemical process changes. 7a: Potential N mineralization rates, where positive = net production or negative = net uptake of inorganic N by soil microbes over two week lab incubation. Soils were collected at the end of the winter from each plot to a depth of 30 cm, and there were no significant depth or treatment x depth effects, so data is shown across depth. 7b: The relationship between dissolved iron (ppm) and dissolved organic carbon (ppm) for soil water samples collected during the winter season (Jan 8-Mar 20, 2024).