Earlier snowmelt may lead to late season declines in plant productivity and carbon sequestration in Arctic tundra ecosystems

Climate change is affecting arctic ecosystems through temperature increase¹, hydrological changes², earlier snowmelt^3,4, and the associated increase in growing season length⁵. Annual arctic air temperature has been increasing at more than double the magnitude of the global mean air temperature increase¹, and terrestrial snow cover in June has decreased by 15.2% per decade from 1981 to 2019⁴. Warming is the main driver of the earlier start of the growing season, and the greening of the Arctic^6,7,8. Arctic greening is associated with enhanced vegetation height, biomass, cover, and abundance⁹. However, the complexity of arctic systems reveals an intricate patchwork of landscape greening and browning^8,10,11, with browning linked to a variety of stresses to vegetation⁸ including water stress^12,13. The interconnected changes in temperature, soil moisture, snowmelt timing, etc. can have important effects on the carbon sequestered by arctic ecosystems¹⁴. The reservoir of carbon in arctic soil and vegetation depends on the interaction of two main processes: (1) changes in net CO₂ uptake by vegetation; and (2) increased net loss of CO₂ (from vegetation and soil) to the atmosphere via respiration. Therefore, defining the response of both plant productivity and ecosystem respiration to environmental changes is needed to predict the sensitivity of the net CO₂ fluxes of arctic systems to climate change.

An earlier snowmelt, and a longer growing season, do not necessarily translate into more carbon sequestered by high latitude ecosystems⁵. There is a large disagreement on the response of plant productivity and the net CO₂ uptake to early snowmelt in tundra ecosystems^{15,16,17,18,19}. A warmer and longer growing season might not result in more net CO₂ uptake if CO₂ loss from respiration increases¹⁶, particularly later in the season, and surpasses the CO₂ sequestered by enhanced plant productivity in northern ecosystems^16,20. Moreover, snowmelt timing and the growing season length greatly affect hydrologic conditions of arctic soils²¹, as well as plant productivity²². Longer non-frozen periods earlier in the year²³ and earlier vegetation greening can increase evapotranspiration (ET), resulting in lower summer soil moisture^24,25,26. The complexity in the hydrology of tundra systems arises from the tight link between the water drainage and the presence and depth of permafrost. The presence of permafrost reduces vertical water losses, preventing soil drainage in northern wetlands during most of the summer despite low precipitation input²⁷. Increasing rainfall²⁸ and increased permafrost degradation can increase soil wetness in continuous permafrost regions². Further permafrost degradation (e.g. ice-wedge melting) can increase hydrologic connectivity leading to increased lateral drainage of the landscape and subsequent soil drying^2,29.

Given the importance of soil moisture in affecting the carbon balance of arctic ecosystems, and its links with snowmelt timing, in this study, we investigated the correlation between summer fluxes of CO₂ (i.e., net ecosystem exchange (NEE), gross primary productivity (GPP) ecosystem respiration (ER)), ET, and environmental drivers such as soil moisture and snowmelt timing, while controlling for the other most important drivers of photosynthesis and respiration (i.e. solar radiation and air temperature). We expected earlier snowmelt to be correlated with larger ET and lower soil moisture, particularly during peak and late season, consistent with drying associated with a longer growing season. The lower soil moisture with earlier snowmelt should result in a negative correlation between snowmelt timing and GPP, particularly during the peak and late season (when we expect the most water stress), and in a positive correlation between snowmelt timing and ER during the entire growing season. This soil moisture limitation to plant productivity should result in lower net cumulative CO₂ sequestration during the entire summer, because of lower plant productivity if these ecosystems are water-limited due to lower soil moisture with earlier snowmelt.

Testing the impact of snowmelt timing on the carbon dynamics and hydrology of tundra ecosystems

The 11 sites were selected as among the longest-running tower sites in the circumpolar Arctic (including 6 to 19 years of fluxes per site and a total of 119 site-years of summer (June to August) eddy covariance CO₂ flux data, Table S1). All sites lie in the zone of continuous permafrost. The sites are representative of dominant tundra vegetation classes (wetland, graminoids, and shrub tundra), together accounting for 31% of all tundra vegetation types (Fig. 1³⁰ and Supplementary Information). Given the complex interactions among different variables (many covarying together), we used a variety of statistical analyses to identify the association between standardized anomalies of NEE, GPP, ER, and ET, and standardized anomalies of the main environmental controls during different periods of the summer corresponding to various stages in seasonal phenology (early season: June, peak season: July, and late season: August). We used a partial correlation analysis to identify if the timing of the snowmelt associates with anomalies of ET, soil moisture, NEE, GPP, ER, atmospheric vapor pressure deficit (VPD), or the Bowen ratio (the ratio between Sensible Heat (H) and Latent Heat (LE)) while statistically controlling for the main meteorological forcing such as air temperature and solar radiation (Methods). Identifying the correlation between ET (and the Bowen ratio) and snowmelt timing is a way to assess water limitation to ecosystems (in addition to testing their response to soil moisture changes), as H, and therefore, the Bowen ratio, are expected to increase with surface drying^31,32. To identify the association between snowmelt timing, the main environmental variables (i.e., air temperature and solar radiation), and NEE, GPP, ER, and ET over time, we performed a maximum covariance analysis (MCA) on the monthly median standardized anomalies from 2004 to 2019 (a time period when data for most of the sites were available). MCA allowed us to find patterns in two space–time datasets that are highly correlated using a cross-covariance matrix²⁶. We retained sites as the unit of variation (i.e., by estimating the standardized anomalies by site for each of the indicated variables, see “Methods”), hence the results of the MCA integrated the site level relationships between each of the variables over time). The goal of this analysis was to identify the most important environmental drivers associated with NEE, GPP, and ER across all the sites over time. MCA is particularly appropriate for this study as it can handle data with gaps and unequal lengths in the datasets. We also tested the relative importance of the abovementioned environmental drivers on the monthly median GPP, ER, and NEE using a linear mixed effect model, including site as a random effect to account for the site-to-site variability. The MCA and the mixed model analyses were conducted to test the relative importance of snowmelt and other variables at different times of the season. Finally, to evaluate the water balance at different times of the season, we estimated the difference between Potential Evapotranspiration (PET) and the actual ET, and the difference between precipitation (PPT) and ET for each of the sites, years, and months (e.g. June, July, and August). This study did not attempt to describe the long-term temporal changes in the anomalies of snowmelt and carbon fluxes, given the short data record available for some of the sites (i.e. less than 10 years, Table S1), but instead focused on understanding the association between environmental variables and the carbon balance at different times of the season. More details of these analyses are included in the Methods.

Influence of snowmelt timing on NEE, GPP, ER, and hydrological status of tundra ecosystems

Once statistically controlling for solar radiation and air temperature (in the partial correlation analysis, see “Methods”), we observed a significant positive relationship between the snowmelt timing anomalies and NEE anomalies (i.e. earlier snowmelt was associated with a higher net CO₂ sequestration) in June and July, but a negative correlation in August (Fig. 2a, Table 1). A significant relationship was also found between snowmelt date anomalies and GPP anomalies, with more positive GPP anomalies (i.e. higher plant productivity) with earlier snowmelt in June and July, and more negative GPP anomalies with earlier snowmelt in August (Fig. 2b, Table 1). Earlier snowmelt was associated with significantly higher ER in both June and July, but there was no significant relationship in August (Fig. 2c, Table 1), suggesting that the late-season correlation between NEE and snowmelt timing was mostly driven by the lower GPP and with earlier snowmelt in August. The MCA analysis showed that the anomalies in snowmelt timing had the highest squared covariance fraction (SCF) with the monthly median anomalies of GPP, NEE, and ER in June and July, and the lowest in August over the 2004–2019 period (Fig. 3). A similar result was observed in the linear mixed effect model, which showed a significant relationship between snowmelt date and GPP, and NEE, in all summer months, higher ({R}_{m}^{2}) between the snowmelt date and GPP in June and July, and no significant relationship between snowmelt date and ER in August (Table S3). In late season, other environmental variables had a higher covariance with the GPP, NEE, and ER anomalies than the snowmelt timing (Fig. 3, Table S3).

Our results are consistent with the discrepancy between the observed increase in the maxNDVI over the last four decades and the time-integrated (TI) NDVI which instead has plateaued in the last two decades and even decreased over the last 10 years in several northern arctic ecosystems³³. TI-NDVI considers the length of the growing season and phenological variations³⁴ and, therefore, better integrates vegetation development during the entire growing season. Moisture was shown to be important for the NDVI trends^33,35. Given the potential water limitation to summer carbon uptake in northern ecosystems^12,23,24,25, we tested if an earlier snowmelt was associated with a decrease in soil moisture, which would affect GPP and NEE. We only observed a significant correlation between soil moisture anomalies and snowmelt date anomalies in June (i.e. higher soil moisture with earlier snowmelt, Fig. S1a, Table S2), but no significant correlation in July and August (Fig. S1a, Table S2). The higher soil moisture with earlier snowmelt in June is consistent with surface inundation after snowmelt^36,37 and earlier soil thawing resulting in higher soil moisture (i.e., soil moisture is low while soils are frozen). A similar result was observed for the ET anomalies. Higher ET with earlier snowmelt in June (Fig. S1b) could be the result of surface inundation after snowmelt³². The standardized NEE anomalies were significantly correlated with the soil moisture anomalies in each of the summer months (Fig. S1d, Table S2). However, the relationship between the GPP (and ER anomalies) and soil moisture anomalies was only significant in June (Fig. S1e,f, Table S2) suggesting an earlier activation of the vegetation with earlier soil thaw (and the associated higher soil moisture). A higher water loss from ET in early season (Fig. S1b) could have resulted in the drying of the surface moss layer with the progression of the summer, which would have been consistent with the observed lower GPP and the lower net CO₂ sequestration with earlier snowmelt observed in August (Fig. 2a,b, Table 1). A potential moisture limitation to plant productivity might have been consistent also with the higher SCF of NEE, or GPP and VPD anomalies in August than in June and July (Fig. 3). However, no significant relationship between ET (or soil moisture) and snowmelt date anomalies was observed in July and August (Fig. S1a,b) contrary to what would be expected if drying occurred following earlier snowmelt. No significant relationship was found between VPD anomalies and snowmelt date anomalies in any of the summer months (P = 0.14 in a partial correlation considering air temperature and solar radiation anomalies). Finally, surface drying should result in an increase in the Bowen ratio anomalies with the progression of the summer, given that H increases with a decrease in water table and surface drying^32,38. However, the Bowen ratio showed no correlation with the standardized snowmelt date anomalies in any of the summer months (Fig. S1c, Table S2), and presented similar values in all the summer months (Fig. S2a). The lack of correlation between the soil moisture, VPD, Bowen ratio, and snowmelt date anomalies suggests that an earlier snowmelt did not result in significant surface drying in the sites of this study. The median PET-ET and PPT-ET for all years and sites included in this analysis (Fig.S2 b,c) was slightly higher in August, similar to reports by others for the Russian arctic tundra^38,39, further supporting a lack of soil moisture limitation in late season. Although these analyses do not consider runoff, which can be significant^21,26, overall our results do not suggest that an earlier snowmelt resulted in a water stress that significantly limited plant productivity in these arctic ecosystems over continuous permafrost.

The correlation between the anomalies in the August GPP and snowmelt timing is consistent with earlier senescence in northern plant species (e.g. Eriophorum vaginatum, a dominant species across these tundra types) compared to southern species growing in the same location in a common garden experiment⁴⁰. The phenotypic variation was shown to be persisting for decades⁴¹, and ecotypes may be unable to extend their effective growth period or take advantage of a longer growing season⁴⁰. Several studies across different plant functional types have shown that once plant growth is initiated after the snowmelt in northern ecosystems, it continues only for a fixed number of days until the occurrence of senescence^42,43,44. Therefore, the lower GPP in August with earlier snowmelt might not be linked to water limitation on photosynthesis later in the season, but rather to an earlier senescence arising from the endogenous rhythms of growth and senescence, that plant functional types living in these extreme conditions have developed over decades. On a broader scale, earlier senescence with an earlier start of the growing season after snowmelt in northern ecosystems is also consistent with an earlier spring zero-crossing date and an earlier autumn zero-crossing date of the mean detrended seasonal CO₂ variations at Barrow, AK, USA (NOAA ESRL: https://www.esrl.noaa.gov/gmd/ccgg/obspack/) during 2013–2017 compared to 1980–1984⁵. The spring and autumn zero-crossing date is the time when the detrended seasonal CO₂ variations intersect the zero line in spring and autumn respectively and can be used as an indicator for the start and end of the net CO₂ uptake by vegetation^45,46. On the other hand, NDVI measurements show both an earlier start of the season and a later end of season for 2008–2012 compared to 1982–1986⁵. The disagreement between the detrended seasonal atmospheric CO₂ concentration showing an earlier autumn zero-crossing date and the NDVI measurements showing a later end of the season has been explained by the increase in respiration in the fall²⁰. The disagreement between atmospheric CO₂ concentration trends (showing an earlier autumn zero-crossing date), and NDVI (showing a later end of the season,⁵ may also be explained by the challenges in using NDVI as a proxy for plant productivity in these arctic systems. The relationship between NDVI and CO₂ flux and plant productivity is highly variable and non-linear in arctic ecosystems⁴⁷. While some arctic ecosystems have shown that NDVI was strongly correlated with GPP (explaining 75% of the variation in GPP⁴⁸, other studies showed that NDVI was either not significantly correlated with GPP and NEE⁴⁹ or was only able to explain a minor fraction (maximum of 25%) of the variation in NEE and GPP in some arctic tundra ecosystems after accounting for the seasonal variation^50,51.

In conclusion, earlier snowmelt was associated with greater net CO₂ uptake and higher GPP in early and peak seasons, but with less net CO₂ uptake and lower GPP later in the summer, in the studied arctic tundra ecosystems. We did not find evidence of a late-season water limitation to GPP with earlier snowmelt. Although several hypotheses can be forwarded to explain the link between snowmelt and late season declines in plant productivity and carbon uptake, the current literature does not provide a definitive explanation (schematic Fig. 4). Future studies should investigate the potential interaction of different processes explaining the response of the carbon dynamics in the Arctic to earlier snowmelt and reconstruct the temporal changes in the carbon balance from these systems. The link between the long-term changes in the CO₂ fluxes and NDVI in tundra ecosystems needs closer examination. Studies should investigate if higher NDVI is definitively associated with higher net CO₂ uptake. Greening of the Arctic might not necessarily translate into more net CO₂ uptake, as early and peak season carbon gains might be offset by a late-season CO₂ loss, and respiration might counterbalance the increase in plant productivity. A better understanding of the processes driving these temporal changes is a fundamental step in advancing our prediction of the response of the arctic CO₂ balance to changing climate.

Source: Ecology - nature.com