Phenology and carbon dioxide source/sink strength of a subalpine grassland in response to an exceptionally short snow season

Changes in snow cover depth and duration predicted by climate change scenarios are expected to strongly affect high-altitude ecosystem processes. This study investigates the effect of an exceptionally short snow season on the phenology and carbon dioxide source/sink strength of a subalpine grassland. An earlier snowmelt of more than one month caused a considerable advancement (40 days) of the beginning of the carbon uptake period (CUP) and, together with a delayed establishment of the snow season in autumn, contributed to a two-month longer CUP. The combined effect of the shorter snow season and the extended CUP led to an increase of about 100% in annual carbon net uptake. Nevertheless, the unusual environmental conditions imposed by the early snowmelt led to changes in canopy structure and functioning, with a reduction of the carbon sequestration rate during the snow-free period.


Introduction
Understanding the processes affecting the carbon dioxide (CO 2 ) exchange between the ecosystems and the atmosphere Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. is crucial to evaluate the future impacts of climate change on the biosphere and the consequent feedbacks on climate system (Cao and Woodward 1998, Denman et al 2007, Heimann and Reichstein 2008. Mountain ecosystems in the European Alps are expected to be particularly impacted by future rising temperatures, changes in precipitation patterns, duration of the snow-pack (Beniston 2005a, and by the increase in extreme events (e.g. summer and winter heat spells, summer droughts. . . ) (e.g. Beniston 2005b, Alcamo et al 2007. In seasonally snow-covered ecosystems, such as high-altitude grasslands, the snow lying on the ground fully limits the length of the growing season and overshadows the relative influence of other environmental factors (photoperiod and temperature). Future warming of the Alpine region (Foppa and Seiz 2012) will likely result in earlier snowmelt dates, thus reducing the period in which grasslands act as a net carbon source. Although several studies have investigated the effect of snow manipulation on the phenology and growth of alpine plants (Wipf and Rixen 2010), little is known about the effects of shortened snow seasons on the annual carbon budget of high-altitude ecosystems, due to the high variability of species responses. Various studies have used the eddy covariance method, which has been proven to be an effective tool for (i) measuring CO 2 exchanges at the ecosystem level and across a spectrum of time scales (Baldocchi 2003), and (ii) evaluating the impacts of extreme events on water and carbon cycle of different ecosystems (e.g. Yi et al 2012, Ciais et al 2005). Nevertheless, only a few studies (e.g. Humphreys and Lafleur 2011) have focused on the effects of changes in snow season duration on the ecosystem carbon budget.
In this study we analysed three years of continuous measurements of CO 2 exchange across the biosphere/atmosphere interface collected in a subalpine grassland by means of the eddy covariance technique. The hypothesis was tested that smaller CO 2 losses during a shorter snow cover period and more net CO 2 uptake during the subsequent longer growing period may cause an overall larger net CO 2 uptake during years with a shorter snow season. For this purpose we took advantage of a 'natural experiment' in which phenology and ecosystem CO 2 fluxes observed during a year (2011) marked by one of the shortest snow seasons on record (83 years) were compared to those observed during average years (2009,2010).
The following main questions were addressed: did the extremely short snow season increase the length of the carbon uptake period in the investigated ecosystem? If so, what is the effect on the ecosystem carbon uptake?
To answer these questions, (i) phenological indicators were extracted from the CO 2 flux time-series, (ii) the relationships between timing and length of the phenophases and the carbon balance of the ecosystem were investigated, and (iii) a modelling approach was used to disentangle the influence of functional changes from the direct effect of weather on the ecosystem carbon uptake.

Site description
The study was carried out in a subalpine unmanaged grassland, in the northwestern Italian Alps, from January 2009 to December 2011. The site is an abandoned pasture located a few kilometres from the village of Torgnon in the Aosta Valley region at an elevation of 2160 m asl (45 • 50 40 N, 7 • 34 41 E).
Dominant vegetation consists of Nardus stricta L., Festuca nigrescens All., Arnica montana L., Carex sempervirens Vill., Geum montanum L., Anthoxanthum alpinum L., Potentilla aurea L., Trifolium alpinum L.. The terrain slopes gently (4 • ) and the soil is classified as Cambisol (FAO/ISRIC/ISS). The site is characterized by an intra-alpine semi-continental climate, with mean annual temperature of 3.1 • C and mean annual precipitation of about 880 mm. On average, from the end of October to late May, the site is covered by a thick snow cover (90-120 cm) which limits the growing period to an average of five months.

Eddy covariance data
The eddy covariance technique was used to measure the fluxes of CO 2 and H 2 O between the ecosystem and the atmosphere. Measurement of wind speed in the three components (u, v, w) was performed by a CSAT3 three-dimensional sonic anemometer (Campbell Scientific, Inc.), while CO 2 and H 2 O vapour air densities were measured by a LI-7500 open-path infrared gas analyzer (LI-COR, Inc.). Instruments were placed 2.5 m above the ground and measurements were performed at a frequency of 10 Hz.
Eddy fluxes were obtained by computing the mean covariance between vertical wind velocity and CO 2 and H 2 O densities with a half-hour time step (Baldocchi 2003). The stored raw 10 Hz records were processed according to the Euroflux methodology (Aubinet et al 2000). The effect of temperature and humidity fluctuations on fluxes was corrected using the methodology described in Webb et al (1980) and Kramm et al (1995). Moreover, in order to correct CO 2 and H 2 O fluxes for the effect of instrument surface heating on flux measurements, the method described in Burba et al (2008) (model 4) was applied.
The storage term was estimated from the time rate of change of the CO 2 mixing ratio at the measurement height (2.5 m) and CO 2 net ecosystem exchange (NEE) was calculated as: where Fc represents the corrected flux of CO 2 and Sc is the storage term. By convention, negative fluxes represent a net mass movement from the atmosphere to the biosphere and positive values the reverse. Results from integral turbulence and stationarity tests (Foken and Wichura 1996) were combined to obtain an overall quality flag (Qc classes from 0, high quality, to 2, low quality) for each half-hour period using the standard procedure followed in Carboeurope-IP project (Mauder and Foken 2004). The eddy covariance flux footprint was determined through the analytical model of Schuepp et al (1990) and the main results of this analysis for the site were presented in Migliavacca et al (2011).
The filtering procedure applied to half-hourly CO 2 fluxes to remove data measured during unfavourable micrometeorological conditions, the evaluation of energy balance closure and the uncertainty associated with flux calculations are described in the supplementary data (available at stacks. iop.org/ERL/8/025008/mmedia). The gap-filling method (www.bgc-jena.mpg.de/∼ MDIwork/eddyproc/) described in Reichstein et al (2005) was used to produce daily, seasonal and annual sums of CO 2 exchange.

Meteorological, radiometric and ancillary measurements
Air (Tair) and soil (Tsoil) temperature were measured respectively by a HMP45 (Vaisala Inc.) and with temperature probes type therm107 (Campbell Scientific, Inc.) at different depths (2, 10, 25, and 35 cm). Soil water content (SWC) was assessed with soil water reflectometers, model CS616 (Campbell Scientific, Inc.), and soil heat flux (G) was measured by HFP01 plates (Hukseflux). Net radiation was measured with a CNR4 (Kipp and Zonen Corp.) net radiometer. Photosynthetically active radiation (PAR) was assessed by a LI-190 (LI-COR, Inc.) sensor. Snow height (HS) was measured with a sonic snow depth sensor (SR50A, Campbell Scientific, Inc.), which was used to determine snowmelt and snow season onset dates for years 2009-2011. Long-term snowmelt, snow season onset and Tair averages were computed on the basis of data collected since 1928 at a site (Cignana, 45 • 52 31 N, 7 • 35 19 E) located nearby the Torgnon site, and at the same altitude. An automatic spectrometric system (HyperSpectral Irradiometer, Meroni et al 2011) was installed to collect high temporal resolution spectral signatures of canopy-reflected radiation. The instrument hosts a spectrometer (HR4000, OceanOptics) operating in the visible and near-infrared region of the solar spectrum (range 400-1000 nm) with a spectral resolution of 1 nm, which allows the computation of different vegetation indices. In this study the meris terrestrial chlorophyll index (MTCI) (Dash and Curran 2004) was used to infer the variation of chlorophyll content (Chl) during the growing season. MTCI values were converted to Chl concentrations using a linear regression model calibrated using Chl concentrations extracted from leaf samples collected every ten days at 12 plots during 2010 (Rossini et al 2012). The relationship (R 2 = 0.83) between MTCI and total Chl, used to estimate chlorophyll content for the three years, was: Finally, leaf area index (LAI) was determined as described in Migliavacca et al (2011).

Extraction of grassland phenophases
Information about the phenology of net CO 2 uptake was extracted from the seasonal time-series of NEE (Richardson et al 2010). We focused on different phases of the vegetative period: the beginning of the carbon uptake period (BGS cup ), the peak season, the end of the carbon uptake period (EGS cup ) and the derived length of carbon uptake period (CUP). BGS cup and EGS cup represent the dates in which the ecosystem switched from a source to a sink in spring and vice versa in autumn. Three different approaches were used to identify these dates: in the first approach we identified BGS cup and EGS cup as the first zero-crossing date after which NEE turned from daily positive values to negative ones in spring and from negative to positive values in autumn; in the second approach, zero-crossing dates were defined as above but using a moving average with a 5-day window (Richardson et al 2009); in the last approach a regression line was fitted between NEE and DOYs (day of the year) using a subset of spring and autumn data (15 days for each period). The BGS cup or EGS cup date was identified by the DOY at which the fitted line passed through 0 (Baldocchi et al 2005). The average of the BGS cup and EGS cup extracted applying the three methods was used in the analysis.

Analysis of light-response curve of photosynthesis
In order to evaluate the impact of the anomalous 2011 snowmelt on NEE, we used a modelling approach to disentangle the effect of biotic response to early spring environmental conditions from the direct effect of the growing season weather (e.g. Richardson et al 2007, Marcolla et al 2011, Wu et al 2012. The light-response curve of photosynthesis was analysed to describe the relationship between NEE and PAR in the different years. The rectangular hyperbolic light-response function (Falge et al 2001) was used: where PAR (µmol photons m −2 s −1 ) is the incident photosynthetically active radiation, A max (µmol CO2 m −2 s −1 ) is the light-saturated rate of CO 2 uptake, α (µmol CO2 /µmol photons ) is the apparent quantum yield, and R eco (µmol CO2 m −2 s −1 ) is the ecosystem respiration. A max , α, and R eco were estimated by fitting equation (2.3) to non gap-filled, half-hourly NEE with a 15-days moving window shifted each 5 days. The model parameters estimated for each year were compared and used to model NEE. In detail, by running the model with one year's PAR and the model parameters of the other years, and vice versa, using fixed model parameters and varying PAR datasets, we evaluated the biotic response (embedded in the values of the parameters) against direct effects of weather (i.e. PAR). In particular, NEE simulated during a summer period representing for all the three years a portion of the BGS cup -peak phase (DOY 176-190) was analysed.

Results
3.1. Long-term climatic conditions and weather during the study period in January to a maximum of 17 ± 3.9 • C in July (1928-2010 average ± sd). Regarding 2011, the main discrepancies compared to previous years were observed in the period March-April, characterized by a mean temperature 2.2 • C warmer than the 1928-2010 average and 2.7-3.0 • C warmer than the mean temperature of previous study years-a difference that was very likely responsible for the early snowmelt.
Soil temperature (figure 1(c)) under snow cover showed constant values above 0 • C. Immediately after snowmelt Tsoil exhibited a characteristic rise. In spring 2009 and 2010 this rise was fast and in a few days Tsoil reached temperature values similar to those of Tair. In 2011, the increase in Tsoil after snowmelt was slower.
The seasonal pattern of midday (11:00-13:00) mean values of incident PAR (figure 1(d)) was similar among the different years.
The snow-free period SWC (figure 1(e)) showed mean values of 26.3 and 25.1% in 2011 and 2010 respectively, and a lower mean of 16.4% in 2009. SWC in all years exhibited a typical peak around the day of snowmelt, when the highest seasonal value was reached.    Table 1 shows the differential contributions of the considered phenophases to the annual cumulative NEE in terms of sums and average rates (i.e. NEE sum in each phenophase divided by the number of days). The averages of the environmental variables measured in each period are also reported. The peculiarities of 2011 were: the overall shorter snow season, lower respiration during the January-snowmelt and snowmelt-BGS cup periods, a longer time for net CO 2 uptake (i.e. longer CUP), but the latter was associated with reduced rates during the developing phases (i.e. BGS cup to peak).

Phenology and CO 2 source/sink strength
The lower cumulative NEE from January to snowmelt in 2011 was mainly a result of the shorter duration of that period and of the slightly lower (16%) respiration rate (figure 2, table 1) compared to the same period in 2009 and 2010. Lower rates of CO 2 losses in 2011, together with lower mean Tsoil, Tair and midday PAR, were also observed in the period between snowmelt and BGS cup (in figure 2 the maximum cumulative NEE after snowmelt reached lower positive values in 2011). Less CO 2 was taken up in 2011 compared to 2009/2010 during the longer period between the BGS cup and the peak value because of a lower NEE rate (∼50%), as highlighted by the less steep slope in 2011 (figure 2). This period was again associated with lower mean Tair, Tsoil and PAR compared to previous years. This pattern is also evident in figure 3(a), showing the mean diurnal patterns of NEE from the BGS cup to peak period. Maximum diurnal NEE in this period reached −9.4, and −8.   During the long period separating the peak value and the EGS cup in 2011, the ecosystem CO 2 uptake was higher than in previous years, even if the average rate remained slightly lower (table 1, figure 3(b)).
Finally, respiratory rates after EGS cup were similar among years (table 1) as highlighted in figure 2 by the similar slopes of NEE. The lower cumulative CO 2 loss in 2011 during this phase was therefore mainly related to the shorter period. The results of the modelling analysis (table 2) clarified whether in 2011 the ecosystem was unable to reach NEE values and rates similar to previous years as a consequence of changes in canopy structural and physiological properties (e.g. lower Chl and LAI) due to the early development or as a direct effect of unfavourable summer weather conditions. The analysis showed that: (i) model parameters (i.e. α, A max and R eco ) obtained from light-response curve in 2011 were significantly lower compared to 2009 and 2010 (p < 0.05 Wilcoxon-Mann-Whitney test); (ii) using fixed PAR and varying physiology parameters (i.e. using PAR data of one year for all three physiology parameter sets), predicted NEE sums using the 2011 parameters were always higher (i.e. lower uptake, ranging from −41.2 to −44.29 gC m −2 ) than those obtained using 2009 (from −60.73 to −66.63 gC m −2 ) or 2010 (from −57.93 to −62.38 gC m −2 ) parameters; (iii) using fixed physiology parameters with varying PAR dataset, NEE sums simulated using the 2011 PAR dataset were always lower (i.e. higher uptake) than those obtained using 2009 or 2010 PAR dataset. This underlined that the NEE reduction observed in 2011 can be attributed to changes in grassland structural and physiological properties rather than to direct limiting weather conditions in this period.

Discussion
In seasonally snow-covered ecosystems, earlier snowmelt and later establishment of snow cover potentially reduce the continuous off-season CO 2 losses and may result in longer periods of CO 2 uptake and growth. In this study we took advantage of a natural experiment (i.e. an exceptionally short snow season) to test the hypothesis that shorter snow-covered periods may enhance the CUP and the annual net CO 2 uptake of subalpine grasslands. Results confirmed this assumption. Nevertheless, two different effects were observed: less CO 2 losses took place during the shorter snow season and lower CO 2 uptake rates occurred during the longer CUP.
Did the extremely short snow season increase the CUP in the investigated ecosystem? We observed that a variation of more than one month (43 days) in the date of snowmelt caused a similar shift (35-40 days) in the beginning of the CUP. However, the earlier snowmelt was followed by a slower increase of the biological activity compared to years with average snowmelt. This observation was supported by the larger time lag between the date of snowmelt and the beginning of the subsequent phenological phases in 2011, i.e.: the ecosystem processes were slower in reaching the different thresholds considered.
The date of snowmelt imposes a clear physical limit to canopy spring development of high-altitude grasslands. When the snowpack-imposed decoupling between vegetation, ambient light and temperature finishes, biological processes quickly take advantage of favourable weather conditions optimizing the short snow-free period available for growth (Körner 2003, Monson et al 2005. This dynamic pattern is typical of warm-season vegetation types, for which photosynthesis in spring recovers generally faster than the senescence at the end of the season (Gu et al 2009). In alpine ecosystems, extreme climate events, such as particularly warm spring spells and the subsequent early snowmelt, could change this typical pattern. The result is a lengthening of the CUP but also an exposure of the vegetation to early spring unfavourable weather conditions (e.g. lower PAR and temperature colder than usual) and an increased risk of cold damage. When snowmelt occurred around the end of May in 2009 and 2010, weather conditions (Tair and Tsoil, photoperiod, PAR) at the study site were already at optimal level and, as a consequence, the up-regulation of photosynthetic activity was fast. On the contrary, the early snowmelt recorded at the beginning of April 2011 led to the advancement of each phenophase, but caused the ecosystem to face less favourable weather conditions, typical of an earlier time of the year, characterized by shorter day-length, lower PAR and colder temperatures.
While the effect of recent warming trends on the onset of plant activity in spring has been outlined in several works (Richardson et al 2013), uncertainties exist on how climate change may affect autumn phenology and CO 2 fluxes (e.g. Wu et al 2013, Piao et al 2008). Moreover, the end of CUP variability and its environmental drivers are poorly investigated within grassland sites. In 2011 we observed a late snow onset in autumn, due to a prolonged period of warm temperatures and absence of precipitation. As a consequence, the ecosystem turned to a source about 20 days later than in previous years, further contributing to the increase of the CUP duration.
What is the effect on the ecosystem CO 2 uptake? The two-month longer CUP observed in 2011 (figure 5(a)) did not lead to higher, but rather similar seasonal cumulative carbon uptake compared to average years, as a result of the compensation between a longer CUP and a lower NEE rate (figure 5(c), table 1). In 2011 daily NEE rate during the CUP was generally lower compared to previous years, especially during the spring development, and reached a peak value lower than other years. We observed that environmental conditions during the summer period were fairly similar among years and, thus, unlikely to account for the observed differences in physiology parameters and photosynthetic rates. The lower summer CO 2 uptake appears to be the result of the biotic response of the ecosystem to an exceptional climate event: the early spring weather that the ecosystem experienced as a consequence of early snowmelt, changed the typical trajectory of canopy development and physiological responses of the ecosystem to environmental conditions. Compared to previous years, plants developed in unusual spring conditions may have adjusted their physiological responses to lower PAR, shorter day-length and lower temperature experienced during the early development and were unable to capitalize on later growing season weather conditions (Monson et al 2005). The snowmelt observed in 2011 is the third earliest snowmelt in 83 years and hence represents an unusual event for plants that are very likely acclimated to a narrower range of weather conditions. This observation was also supported by LAI and Chl data, since the ecosystem modifies its photosynthetic capacity through variations in LAI and chlorophyll content in relation to changes in limiting factors (Dawson et al 2003). To our knowledge there are few studies highlighting similar findings. Indeed, although shifts in structural or reproductive phenology as a result of variations in snow cover depth and duration have clearly been described in several experimental studies of snow manipulation (e.g. Wipf and Rixen 2010), results concerning the effects on growth and productivity are less clear and differ among species, growth forms and habitats (snowbeds, fellfields, meadows. . . ). Moreover, most studies have focused on the effect of a delayed rather than an earlier snowmelt (Wipf and Rixen 2010). For example Wipf and Rixen (2010) found that a delayed snowmelt decreased productivity (peak season biomass). On the contrary there is evidence (Walker et al 1995, Wipf et al 2009) that the growth of some alpine species was reduced in years with advanced snowmelt as a consequence of unfavourable conditions that plants experienced in early spring. Moreover, if reproductive and vegetative phases are partially influenced by different environmental factors, an early snowmelt can determine in some species an early flowering causing differences in reproductive and vegetative resource allocation compared to standard snowmelt date (Inouye 2008, Körner andBasler 2010), resulting in a decrease in vegetative development.
Changes in snow cover amount and duration cause also a change in nutrient supply in spring (Körner 2003). Our results suggest that the timing of soluble N availability may be more likely of concern than the total amount of N in melted water (Smith et al 2012): 2010 and 2011 had indeed similar HS and SWC at the time of snowmelt, but different NEE rates and canopy properties (LAI and Chl) during the CUP. Since there is evidence that plants may take up N maximally after snowmelt (Jaeger et al 1999, Bardgett et al 2007, nutrient uptake in 2011 may have been limited by the low Tsoil following early snowmelt (Karlsson and Nordell 1996). Moreover, Brooks and Williams (1999) suggested that the availability of soil N is strongly regulated by the timing and duration of the snow cover, with higher N immobilization under long-lasting snowpacks and consequent higher net N conservation in the soil pool. Following this interpretation, less N may have been immobilized in soil during shorter winter 2011 and subsequently available in early spring 2011, thus contributing to suboptimal conditions for plant development.
We hypothesized that shorter snow seasons potentially enhance the annual net CO 2 uptake of seasonally snowcovered ecosystems. Since the two-month longer CUP alone cannot explain the twice as high annual uptake observed in 2011 ( figure 5(b)), explanations were found in the short 2011 off-season period (i.e. snow-covered and non-snow-covered periods with continuous respiration, i.e. NEE >0) during which the ecosystem lost less CO 2 than in previous years. The off-season period lasted 245 and 242 days in 2009 and 2010 and was 58-52 days shorter in 2011 ( figure 5(a)). The shortened snow season translated in a 40% reduction of the off-season cumulative CO 2 loss compared to average years. Conversely, differences in CUP cumulative NEE amounted to less than 5%, since, despite the longer CUP, the average 2011 NEE rate was lower (figure 5(c)) compared to 2009/2010. Taken together, these observations confirm that the higher annual net CO 2 uptake in 2011 was mainly caused by a shorter period of off-season respiration rather than by an enhanced CO 2 uptake during a longer CUP.

Conclusion
During a year characterized by an extremely short snow season, a 100% increase in the annual net CO 2 uptake was observed at a subalpine grassland. The larger carbon sink was attributable to smaller cumulative CO 2 loss during the shorter snow season, as lower CO 2 uptake rates during the longer CUP resulted in similar cumulative NEE during the vegetation period as compared to average years. If an increase of future occurrence of events such as the observed is assumed, this trade-off between reduced CO 2 losses during shorter winters and lower uptake during longer summers will be crucial in determining the annual NEE.