Four-year measurement of net ecosystem gas exchange of switchgrass in a Mediterranean climate after long-term arable land use

Switchgrass ( Panicum virgatum L.) is a perennial lignocellulosic crop that has gained large interest as a feedstock for advanced biofuels. Using an eddy covariance system, we monitored the net ecosystem gas exchange in a 5-ha rainfed switchgrass crop located in the Po River Valley for four consecutive years after land-use change from annual food crops. Switchgrass absorbed 58.2 Mg CO 2 ha (cid:1) 1 year (cid:1) 1 (GPP — gross primary production), of which 24.5 (42%) were fixed by the ecosystem (NEE — net ecosystem exchange). Cumulated NEE was negative (i.e. C sink) even in the establishment year when biomass and canopy photosynthesis are considerably lower compared to the following years. Taking into account the last 3 years only (postestablishment years), mean NEE was (cid:1) 26.9 Mg CO 2 ha (cid:1) 1 year (cid:1) 1 . When discounted of the removed switchgrass biomass, ecosystem CO 2 absorption was still high and corresponded to (cid:1) 8.4 Mg CO 2 ha (cid:1) 1 year (cid:1) 1 . The estimation of the life cycle global warming effect

Carbon dioxide assimilation/emission is the balance between photosynthesis and respiration of an agricultural ecosystem (soil and plants) and, especially in the case of bioenergy feedstocks, constitutes a fundamental indicator of their sustainability. While aboveground CO 2 fixation of a crop can be easily estimated, the canopy respiration and the total belowground C are difficult to measure, especially at a large field scale. At the same time, evapotranspiration (ETc) is the second larger term after precipitation in the hydrogeological water balance and the ecosystem water use is a key component of the hydrologic cycle, being intrinsically linked to other biogeochemical cycles through vegetation (Sellers et al., 1997). Extensive introduction of a new crop as switchgrass may produce significant hydrological effects. It has been reported, for example, that perennial grasses can reduce groundwater recharge because of the longer growing seasons and water demand (Schilling, Jha, Zhang, Gassman, & Wolter, 2008) compared to annual crops. Understanding ETc levels of a perennial crop such as switchgrass can be therefore important to prevent negative consequences. Moreover, the accomplishment of the production of high quantities of biomass to support the development of the bioeconomy should be addressed by minimizing land and water competition with food crops (Dornburg et al., 2010). For example, in the Mediterranean basin biomass crops are expected to be mostly developed on marginal or abandoned land under semiarid conditions. On the whole, water use efficiency (WUE, i.e. the amount of carbon gained per consumed water) is critical in the selection of the most appropriate biomass crop and to assess the impact of the potential large deployment of a crop on local basin hydrology (VanLoocke, Twine, Zeri, & Bernacchi, 2012). Perennial species, such as switchgrass, allocate great amounts of C belowground (Dohleman, Heaton, Arundale, & Long, 2012;Kahle, Beuch, Boelcke, Leinweber, & Schulten, 2001;Neukirchen, Himken, Lammel, Czypionka-Krause, & Olfs, 1999), which should not be neglected when calculating WUE. A high WUE reflects the opportunity for a crop to provide an ecosystem service, the carbon accumulation, compared to a perceived environmental cost of this service, that is the consumed water.
Although switchgrass and, more in general, perennial grasses are recognized to be able to sequester significant amounts of soil C (up to 10.1 Mg ha À1 year À1 in the short term and averaging <1.0 Mg ha À1 year À1 in the long term; Anderson-Teixeira, Davis, Masters, & Delucia, 2009;Frank, Berdahl, Hanson, Liebig, & Johnson, 2004;Liebig, Johnson, Hanson, & Frank, 2005), little information is present in the literature on their carbon storage potential in the Mediterranean area, especially at the ecosystem level. Switchgrass was shown to be able to produce considerable amounts of biomass under a Mediterranean climate in the long term (10.0 and 13.6 Mg ha À1 year À1 in Greece and Italy, respectively; Alexopoulou et al., 2015), increasing also soil-associated organic C (SOC) content by 0.09 g kg À1 year À1 (stock increase was estimated between 0.37 and 1.58 Mg ha À1 year À1 ) in the long term (Nocentini & Monti, 2017). However, other studies in different regions have also shown C losses after conversion into switchgrass (Garten & Wullschleger, 2000;Qin, Dunn, Kwon, Mueller, & Wander, 2016;Zenone, Gelfand, Chen, Hamilton, & Robertson, 2013), with the difference often due to the land-use change, as the former land use usually influences ecosystem C stocks and subsequently the capacity of a new established crop to increase them (Fargione, Hill, Tilman, Polasky, & Hawthorne, 2008;Qin et al., 2016). Switchgrass was reported to lose SOC when replacing grassland (À1.0 Mg ha À1 year À1 , on average; Qin et al., 2016) while increasing it when replacing cropland (+1.28 Mg ha À1 year À1 , on average; Qin et al., 2016), but, for example, Zenone et al. (2013) also reported a C loss of 1.49 Mg ha À1 year À1 in the first years of switchgrass cultivation following a corn-soybean rotation in Michigan (USA).
In this study, we addressed CO 2 and H 2 O fluxes in a 5ha switchgrass field for four consecutive years and after the land-use change from long-term cultivation of annual food crops (at least, 60 years of arable land use). Fluxes were measured through an eddy covariance tower (Owen et al., 2007) at the ecosystem level. Eddy covariance data were validated and integrated with root, soil and aboveground biomass measurements. Although the ecosystem C balance measured with the eddy covariance represents a substantial component of its life cycle global warming impact, to assess the effect on GHG emissions of establishing switchgrass, other sources, direct and indirect, were taken into account. Following methodologies from previous life cycle studies (Adler et al., 2007;Fazio & Monti, 2011), emissions produced by agronomic inputs during cultivation were assessed. Nitrous oxide (N 2 O) emissions, which can substantially impact agricultural production cycles (Reay et al., 2012), were modeled as well (Parton, Hartman, Ojima, & Schimel, 1998). Although switchgrass biomass can be used for several purposes, CO 2 credits from displacing fossil fuels by converting switchgrass harvested biomass in advanced ethanol were taken under consideration (Gelfand et al., 2011;Lynd et al., 2008;McLaughlin & Walsh, 1998). This study did not consider the indirect land-use change (ILUC) effect of displacing food production to cultivate switchgrass, as its estimation is still highly uncertain (Plevin, O'hare, Jones, Torn, & Gibbs, 2010;Zilberman, 2017). The ILUC effect represents the CO 2 emission deriving from the conversion of more land somewhere else to recover the lost commodities substituted by biofuel (Searchinger et al., 2008).
The ultimate objective of this study was the multiyear farm-scale level determination of ecosystem C and water gas exchange of switchgrass grown with conventional agricultural practices and machinery, to then include the data in an assessment of the life cycle global warming potential and WUE of advanced ethanol production. To the best of our knowledge, there are no studies reporting direct multiyear continuous measurements of the net ecosystem gas exchange of switchgrass in the Mediterranean Europe. Furthermore, this is one of the few studies that assessed the life cycle global warming potential of the land-use change from annual food crops to switchgrass integrating eddycovariance measurements, as previous studies mainly focused on the land-use change from perennial vegetation systems as set-aside grasslands (Gelfand et al., 2011. The conversion of marginal cropland to biofuel production might become common in future biomass supply districts to fulfill the demand by the transformation plants.

| Site description and management
The experiment was carried out in a 5-ha flat field of the experimental farm of the University of Bologna (44.55 N,11.41 E, 33 m a.s.l.), located in Cadriano, in the Po River Valley (North Italy). Carbon and water fluxes were monitored on switchgrass (P. virgatum L., cv. Alamo) over a 4year period (May 2012-June 2016). The soil was classified as Udic Ustochreps fine-silty, mixed mesic (USDA classification), with a pH of 7.2, and high exchangeable calcium. Soil organic carbon was 10.1 g/kg (1.6% of organic matter), total N content was 1.2 g/kg, and C/N ratio equal to 8.0. The available P and K were not limiting (57 and 161 mg/kg, respectively). Switchgrass was sown on 23 April 2012, 45 cm row spaced. Wheat (Triticum durum Desf.)sugar beet (Beta vulgaris L.) was the conventional crop rotation adopted before switchgrass plantation. Seedbed preparation included plowing in autumn, harrowing and mechanical hoeing in spring. Fertilization was 230 kg/ha of P 2 O 5 only in presowing, whereas, starting from the second year, N was applied annually at a rate of 50 kg (N) ha À1 as urea. Weeds control occurred only during the establishment by spraying Agherud Dicamba (3,6dichloro-2-methoxybenzoic acid; 1 L/ha), Harmony 50sx (tribenuron metil 50%; 10 g/ha) and Starane gold (fluroxypyr-meptyl, florasulam; 1 L/ha). Irrigation was never applied. Switchgrass was mechanically harvested in autumn (October) each year; stems were cut at~5 cm from the ground. After cutting, biomass was baled and removed from the field. After each mechanical harvest, biomass losses (residue) were measured on ten random areas (6 m 2 each) across the field. Biomass moisture concentration was determined by collecting and drying (oven-dried at 105°C until constant weight) biomass samples taken from ten bales and from residue biomass.

| Carbon and water measurements
Carbon and water vapor fluxes monitoring started on 11 May 2012. Given the relatively small size of the field, to ensure the most proper flux representation, a wind footprint analysis was performed continuously during a 8-month period, following the methodology proposed by Schuepp, Leclerc, Macpherson, and Desjardins (1990). Data confirmed that about 90% of the measured fluxes originated from the switchgrass, thus revealing the correct positioning of the tower (Figure 1). A three-dimensional sonic anemometer (Metek, GmbH, Germany) and an open path infrared absorption gas analyzer (IRGA, Li7500; LiCor Inc., USA) for measuring, respectively, carbon and water fluxes, were placed at 3.5 m height (about 2 times the height of switchgrass canopy at full elongation). Eddy measurements were coupled with a net radiometer (Kipp&Zonen CNR1) for detecting the net available energy at the canopy surface, and to calculate energy balance; flux plates (Hukseflux HFP01-10) were installed to compute the soil heat flux. Also, placed at two different soil depths (15 and 30 cm), there were four water probes (ECHO-5 Decagon) and six thermocouples to measure soil moisture and temperature. Wind, H 2 O, CO 2 and temperature were sampled at high time resolution (10 Hz) and released as 30 min integral per square meter and per second. All data were stored on a CR1000 (Campbell Sci., USA) data logger, remotely accessible. Data processing followed the guidelines of the standard EUROFLUX methodology (Aubinet et al., 1999). The half-hour mean flux values were computed in the postprocessing by covariance analysis between w and the scalars c, q and T, where w is the vertical component of the wind speed vector, c and q are the concentration of CO 2 and H 2 O, and T the air temperature. Postprocessing and quality check were applied to raw data together with specific routines, as running mean to avoid detrending problems, three angle coordinate rotations of the wind vector to remove the effects of instrument tilt or terrain irregularity on the airflow, despiking and stationarity (Derbyshire, 1995).
Gaps due to unfavorable micrometeorological conditions, instrument failure and data quality check that occurred during the entire monitoring time, led to a data coverage of 66.1% of the whole period. Gap filling was then carried out on the remaining 33.9% of data with the use of the online tool "eddy covariance gap-filling and flux partitioning" of the Max Planck Institute methodology.
The gap filling of the eddy covariance and meteorological data are performed with methods similar to Falge et al. (2001), but also considering the covariation of fluxes with meteorological variables and the temporal autocorrelation of the fluxes (Reichstein et al., 2005). In this algorithm, three different conditions are identified: 1. Only the data of direct interest is missing, but the meteorological data are available for radiation (R g ), air temperature (T air ), and vapor pressure deficit (VPD). 2. Also T air or VPD are missing, but R g is available.
3. Also R g is missing.
Case 1: The missing value is replaced by the average value under similar meteorological conditions, that is with a look-up table (LUT), within a certain time window. Similar meteorological conditions are present when R g does not deviate by more than 50 W/m 2 , T air by 2.5°C, and VPD by 5.0 hPa. If no similar meteorological conditions are present within the starting time window of 7 days, the windows size is increased to 14 days.
Case 2: The same LUT approach is taken, but similar meteorological conditions can only be defined via R g within a time window of 7 days.
Case 3: The missing value is replaced by the average value at the same time of the day (1 hr), that is by the mean diurnal course. In this case the window size starts in the same day, thus a linear interpolation of available data at adjacent hours.
If after these steps the values could not be filled, the procedure is repeated with increased window sizes until the value can be filled.
Night data were inspected to detect invalid values (i.e. negative values of carbon dioxide flux, values of respiration outside of the trend), in order to avoid the underestimation of the nocturnal surface exchanges (Aubinet et al., 1999;G€ ockede et al., 2008;Twine et al., 2000). Positive CO 2 and H 2 O values will indicate gas emissions to the atmosphere. CO 2 fluxes were measured to estimate the photosynthetic uptake or gross primary production (GPP) and ecosystem respiration (R e ). GPP represents the total amount of CO 2 that is uptaken by the ecosystem through autotrophic fixation while R e represents the CO 2 released by all agroecosystem metabolic activities. The net ecosystem exchange (NEE), that is the amount of carbon which has been stored in the above-, belowground biomass and SOC, was obtained as difference between GPP and R e . We estimated switchgrass ETc through water vapor released by the system. It was calculated by dividing latent heat flux by the latent heat of vaporization. Ecosystem WUE was then calculated dividing GPP or NEE by ETc. Daily rainfall, air temperature and solar radiation were also collected by a meteorological station inside the experimental farm.

| Litter, root and soil samplings
Litter biomass, which formed because of the accumulation, year after year, of the losses from mechanical harvest, was collected in November 2015, at the end of the fourth growing season, on ten 0.25 m 2 areas, randomly distributed across the field. Subsamples were taken from each sample then oven-dried (at 105°C until constant weight) for moisture content determination.
In June 2012 and November 2015, three soil cores (70 mm /) were collected by a mechanical auger down to 0.45 m soil depth, in ten sampling areas across the field (same areas where litter was sampled). In each sampling area, the three soil cores were collected within the row, next to the row and in the interrow, in order to take into account spatial differences in belowground C. In total, 60 soil cores were collected (3 soil cores 9 10 sampling areas 9 2 years). Samples were airdried: then litter residues were manually removed, root biomass was manually separated from the soil, washed, sieved and oven-dried (105°C) for 72 hr, while the remaining soil was ground to 0.5 mm, before analyzing for organic C content and 13 C isotope abundance; 13 C abundance allowed to specifically determine switchgrass (C4)-derived SOC as C3 species were cultivated in the previous 2 years (wheat-sugar beet). After being oven-dried at 60°C, subsamples of aboveground and belowground biomass were analyzed to measure their C content. Soil subsamples (~15 mg) were pretreated with HCl to eliminate inorganic C and encapsulated thereafter. SOC was determined by an elemental analyzer (Flash 2000 CHNS/ O Analyzer, Thermo Scientific, USA), while carbon isotope composition was determined through CF-IRMS continuous flow-isotope ratio mass spectrometry (Delta V advantage, Thermo Scientific, USA). Plant material subsamples (~0.5 mg) were analyzed with the same instrumentation. To calculate C stocks variations starting from SOC concentrations in the soil, soil bulk density was measured at two depths (15 and 30 cm), using metal cylinders (volume = 98 cm 3 ) and collecting a total of 12 samples (3 samples 9 2 depths 9 2 years).

| Biogenic N 2 O emissions and CO 2 emissions from inputs and fossil offset
To obtain a full estimation of the impact of the farm-scale switchgrass field on GHG emission, biogenic N 2 O emission and the CO 2 released or taken up indirectly were also taken into account.
As our eddy covariance system was not equipped to measure N 2 O fluxes, we estimated N 2 O emission with the DAYCENT model (Parton et al., 1998). DAYCENT was calibrated and validated for Cadriano's switchgrass field, also using the data measured by the eddy covariance tower, in a recent work where a detailed description of the modeling can be found (Nocentini, Di Virgilio, & Monti, 2015).
Equivalent CO 2 emissions from the agronomic inputs applied to the switchgrass field were estimated. Taking into account all field operations carried out and materials (i.e. seeds, fertilizers, herbicides) used (Table 1), and after interviewing the workers of the experimental farm about the machines used and the time and diesel spent per hectare for each operation, we were able to perform an analysis with the life cycle software SimaPro 8.0 (PR e Consultants, Amersfoort, The Netherlands). Processes already present in the Ecoinvent 3.0 database were modified to better reflect the field operations observed in terms of diesel use and machines consumption. At last, equivalent CO 2 emissions caused by each process were calculated employing the IPCC 2013 GWP methodology (IPCC, 2014). Emissions generated during switchgrass establishment were annualized, assuming that the economic life of the stand was 10 years (Qin, Mohan, El-Halwagi, Cornforth, & McCarl, 2006). Also, the CO 2 indirect uptake (credit) deriving from displacing fossil fuels was estimated, as the lignocellulosic harvested biomass could be used to produce  advanced ethanol. Lynd et al. (2008) reported that 282 L of EtOH can be produced from 1 ton of dry biomass. Thus, considering an energy density for advanced (lignocellulosic) ethanol of 21.1 MJ/L and C savings equal to 90 g CO 2 MJ À1 of energy produced (Gelfand et al., 2011), offset credits could be calculated.

| Flux terminology
The ecosystem CO 2 autotrophic fixation per unit of time and area is GPP (negative sign). The ecosystem CO 2 respiration per unit of time and area is R e (positive sign). The difference between these two fluxes results in the net CO 2 exchange of the ecosystem with the atmosphere (NEE): The cumulated NEE is then subtracted of the C removed through biomass harvest to obtain the net ecosystem C balance (NECB), which represents, for the switchgrass ecosystem, the sum of litter, root and change in soil-associated C: The water vapor released by the ecosystem represents the ETc. The ecosystem WUE was calculated as the amount of C fixed per unit of water released: The life cycle GHG balance was estimated by accounting for all direct and indirect emissions: NECB, biogenic N 2 O emission, cultivation emissions (from agronomic inputs), and fossil fuel offset credits obtained by producing renewable energy with the switchgrass harvested biomass:

| Environmental conditions
The experimental site falls within the Mediterranean North agroclimatic zone as described by Metzger, Bunce, Jongman, Mucher, and Watkins (2005). Local long-term measured annual precipitation (30 years) corresponded to 708 mm, while 30-year maximum, minimum and mean T air were 18.2, 8.2 and 13.2°C, respectively. Weather and soil temperature and moisture data for the period of the experiment are reported in Table 2, while Figure 2 shows the time course of temperatures and rainfalls. Mean T air during investigation was slightly higher than the long-term value, ranging from 13.8 to 14.8°C, with 2013 being colder than other years. T air varied the most during the growing seasons (April-October) through the years, from 19.4 to 22.1°C (Table 2). Soil temperature (T soil ) significantly decreased from first to second year, probably due to the cooling effect of the tall and dense established switchgrass. Annual precipitation was similar in 2013 and 2014, lower in 2015 and significantly lower in 2012; on average, it was close to the long-term value for the site (721 mm). On the other hand, distribution of monthly precipitation was also different among years ( Figure 2): for example, although showing similar values of annual precipitation, the years 2013 and 2014 had the driest and the rainiest growing seasons, respectively (Table 2).
Soil water content greatly varied among years and within the same growing season. Observing cumulative ETc and precipitation, the site apparently experienced a negative water deficit (i.e. the difference between T A B L E 2 Yearly and growing season mean air and soil temperature, mean soil water content (SWC; upper 30 cm), cumulative precipitation and evapotranspiration (ETc), and water deficit (WD) at the experimental field precipitation and ETc) only during the second growing season (Table 2); however, during summer, in July and August, water deficit assumed sometimes negative values also in other years.

| Energy balance closure
The ecosystem energy balance closure (EBC) is an important test to evaluate the reliability of the eddy covariance data, as an unbalance leads to underestimate CO 2 fluxes (Mauder et al., 2007;Moncrieff, Malhi, & Leuning, 1996). EBC requires that the sum of the estimated latent (LE) and sensible (H) heat flux are equivalent to all other energy sinks and sources, as expressed in the following equation: where R n is the net radiation, G is the heat flux into the soil, S is the rate of change of heat storage (air and biomass) between the soil surface and the level of the eddy covariance instrumentation, and Q is the sum of all additional energy sources and sinks. In a typical manner, Q is neglected as a small term.
Only high-quality fluxes of H and LE were used and EBC was calculated only when all four measurements (H, LE, R n , and G) were available. The fraction of half-hourly values of turbulent fluxes (H + LE) and available energy (R n À G) were 0.84, 0.70, 0.77 and 0.71 from 2012 to 2015, respectively (Figure 3), indicating that, despite a likely underestimation of G, the eddy covariance F I G U R E 2 Monthly average air temperature and shortwave incoming radiation (R s ), and cumulated monthly precipitation and evapotranspiration (ETc) during the 4 years of the experiment measurements can be considered as reliable, also compared with other studies (Mauder et al., 2007;Twine et al., 2000). The heat contained in the biomass can be included in the balance (Wilson et al., 2002), while punctual measurements of soil temperatures and moisture should be repeated in several locations over the field to obtain a more representative value of G (Meyers, 2004).

| Fluxes of CO 2
The GPP increased significantly from the first to the second year and, to a lesser extent, from the second to the third year, while it slightly decreased in the fourth year ( Figure 4; Table 3). Within the growing season, GPP ranged from À37.8 (establishment year) to À58.8 (third year) Mg CO 2 ha À1 . Ecosystem respiration (R e ) also increased but less than GPP (from 20.2 to 28.3 Mg CO 2 ha À1 ), thus resulting in an overall increment of NEE, which almost doubled in the third year compared to the establishment year (Table 3). Fingerprint plot ( Figure 5) allows visualizing the seasonal and daily patterns of CO 2 capture by the ecosystem (NEE): the darker the color, the higher the C storage; the larger the blue area, the higher the C stored. Daily and seasonal patterns of NEE differed over the 4 years. Higher uptake of CO 2 was mostly in 2014, followed by 2013, 2015 and 2012. Within the same day, as expected, NEE was usually higher at midday, then decreased until a release of CO 2 that generally occurred from 6 p.m. to   Figure 6) clearly showed that the difference between the establishment year and the following years was higher than 4 Mg CO 2 ha À1 at the beginning of the vegetative season, but that it decreased as the switchgrass reached maturity, until reaching similar levels in September. Maximum CO 2 monthly rate accumulation was 8.1 Mg CO 2 ha À1 in June 2013 ( Figure 6).

| Aboveground biomass and net ecosystem carbon balance
Total harvested (mechanically) aboveground biomass (dry weight) was 6.9, 13.0, 10.5 and 12.1 Mg ha À1 year À1 in 2012, 2013, 2014 and 2015, respectively. From the second year on, yields reached the same biomass yield level of a nearby established switchgrass field in a hilly area, which production ranged from 7.9 to 11.3 Mg/ha (Di Virgilio, Monti, & Venturi, 2007). It should be recognized, however, that these values are an underestimation of the "canopy net accumulation rate" as not all biomass was recovered by the bailing machine. About 1.2, 3.6, 5.3 and 3.6 Mg (d.w.) ha À1 were measured as residual biomass left on the field after bailing in 2012, 2013, 2014 and 2015, respectively. Harvest biomass losses were higher in 2014 due to lodging. The amount of aboveground biomass removed after harvest ranged from 66 (2014) to 85% (2012), while the amount of C removed from the ecosystem with harvests ranged from 53 (2014) to 71% (2015) of the NEE. The NEE can be subtracted of the C removed with harvested biomass to define the net ecosystem carbon balance (NECB; Chapin et al., 2006), which represents an indication of the sequestered CO 2 (in the litter biomass, root biomass or SOC). NECB was negative in all years increasing from À4.8 to À13.5 Mg CO 2 eq ha À1 from 2012 to 2014 (À7.2 Mg CO 2 eq ha À1 in 2015) ( Table 3). The 4-year cumulative NECB was À33.4 Mg CO 2 ha À1 . Over the last 3 years, mean NECB was À9.5 Mg CO 2 ha À1 year À1 , thus 98% higher than in the establishment year.
In November 2015, after four growing seasons, switchgrass root biomass in the upper 0.45 m soil layer was 8.35 Mg/ha, while cumulative (4 years) litter biomass on the ground was 12.7 Mg/ha. The C content of switchgrass fresh material was 41.1 AE 1.8 and 40.3 AE 3.5% (belowand aboveground biomass, respectively). We, therefore, estimated that 94% of the NECB was stocked into these two C pools, and only the remaining 6% was sequestered as SOC (0.14 Mg ha À1 year À1 ). The sampling depth adopted was considered sufficient to factor in the majority of roots and SOC deposition as the measured SOC change between June 2012 and November 2015 corresponded to 0.13 Mg ha À1 year À1 . However, considering the evolution of 13 C abundance in the soil, it was calculated that, after four growing seasons, switchgrass-derived SOC corresponded to 9.45 Mg/ha.

| Life cycle global warming potential
Considering 10 years of life span for the switchgrass stand, annualized emissions from the use of agronomic inputs were 733 kg CO 2 eq ha À1 year À1 . Nitrogen fertilization and harvest operations (cutting and baling) impacted most on emissions (23% and 60% of total emissions, respectively; Supporting information Table S1).
The displacement of fossil fuels produced CO 2 credits corresponding to À5.7 Mg ha À1 year À1 , after switchgrass harvested biomass was converted in advanced ethanol.
Simulated N 2 O emissions resulted low (0.4-1.0 Mg CO 2 eq ha À1 year À1 , respectively in the establishment year or in the following years) compared to the ecosystem CO 2 uptake.

| Fluxes of water and water components
Similar ETc patterns were observed over the first 3 years, while in the fourth year ETc was significantly lower, likely due to a summer drought (Figure 2).
Evapotranspiration increased until May and started to decline from late June, and significantly decreased from July, in parallel with GPP and NEE. Other studies showed ETc to decline later in the season, for example in August (Skinner & Adler, 2010). In our study, ETc started to F I G U R E 4 Cumulated gross primary production (GPP) and ecosystem respiration (R e ) in the 4 years of switchgrass cultivation. Values refer to daily cumulative fluxes; negative values indicate atmospheric CO 2 uptake by the ecosystem while positive values indicate CO 2 emission decrease in August only in the establishment year. Average monthly ETc during the growing season ranged from 39.6 (2015) to 59.1 mm (2014). Cumulative annual ETc accounted for 55%, 56%, 59% and 50% of the annual precipitation (Table 2), respectively in the 4 years, in agreement with other studies (Abraha et al., 2015;Skinner & Adler, 2010). December showed the lowest cumulative ETc: 5.5, 8.1, 7.7 and 4.2 mm, respectively in the 4 years.
In this work, WUE was calculated by taking into account NEE and measured ETc (Table 5). Average WUE (ratio between NEE and ETc) was 1.3 (2012), 1.5 (2013), 1.6 (2014), 2.0 (2015) mg C g À1 H 2 O, with highest values occurring in August 2012, June 2013, June 2014 and July 2015 (2.3, 2.5, 2.7 and 10.5 mg C g À1 H 2 O, respectively). ETc was particularly low in July and August 2015, resulting in a decreased CO 2 uptake (Table 3)  | 475 much higher WUE for that period. Decreased ETc in summer 2015 could be the result of a prolonged lack of moisture in the soil (Table 2). NEE and GPP resulted significantly correlated with ETc (r 2 = 0.45* or 0.48*, respectively; all years' growing season daily data).
Moreover, considering the estimated use of water during transformation to ethanol (0.5 Mg H 2 O GJ À1 ; Berndes, 2002) and the evapo-transpired water during the growing season (April-October), WUE of ethanol production corresponded to 0.95, 0.71 and 1.22 L (EtOH) Mg À1 H 2 O, respectively in 2013, 2014 and 2015. However, almost all the consumed water (about 99%) was part of that received by the crop during the cultivation phase from rainfall.

| DISCUSSION
Despite a considerable lower biomass accumulation in the first year, switchgrass was however able to maintain a CO 2 storage capacity in all years. Mean NEE and NECB of switchgrass during the postestablishment years (second to fourth year) were À26.9 and À9.5 Mg CO 2 ha À1 year À1 , respectively, with NEE reaching peak monthly values of À8.1, À7.9 and À7.6 Mg CO 2 ha À1 during the 2013, 2014 and 2015 growing seasons, respectively. In the same way, in a semiarid environment in Oklahoma, Wagle and Kakani (2014) measured a peak monthly value of NEE in a young switchgrass stand of À9.9 Mg CO 2 ha À1 . Different from the present study, they estimated that NECB made switchgrass a small C source; nevertheless, their monitoring was limited to the establishment period. Lower values of yearly NECB (À1.4 Mg CO 2 ha À1 year À1 ) were reported by Skinner and Adler (2010) in a study in Pennsylvania under a continental climate and at higher altitude than our experimental site. Studies in Iowa (Al-Kaisi, Yin, & Licht, 2005) and in South Dakota (Lee, Owens, & Doolittle, 2007) reported SOC sequestration rates by mature stands from À4.4 to À14.7 Mg CO 2 ha À1 year À1 , while in the Central Illinois (Zeri et al., 2011), having the same latitude as in the present study, switchgrass C sequestration (NEE) was À9.0 Mg CO 2 ha À1 year À1 in the first year to then steeply increase in the second and third years (À16.6 and À17.8 Mg CO 2 ha À1 year À1 , respectively), in line with our findings, although lower. As reported in Skinner and Adler (2010) and Zeri et al. (2011), a stabilization of switchgrass yields and NEE to higher values starting from the second year was also observed in this study. This is an indication that establishment of switchgrass might be faster than in other bioenergy perennial crops such as miscanthus (Miscanthus 9 giganteus Greef and Deuter) and giant reed (Arundo donax L.).
Starting from late March/early April, R e tended to increase (Figure 4), together with increasing air and soil temperatures. A major determinant for R e was in fact the air temperature (r 2 = 0.62**, data not shown). Lower values of R e were recorded in this study (24.7-33.4 Mg CO 2 ha À1 year À1 ; Table 3) than measured elsewhere for other switchgrass stands (29.1-36.1 Mg CO 2 ha À1 year À1 ; Skinner & Adler, 2010), which could be linked to lower soil temperatures (Bremer, Ham, Owensby, & Knapp, 1998). Observing how R e seems strongly related to air temperature suggests that further studies in warmer climates of the southern Mediterranean are needed.
Gross primary production also increased every year from late March showing a similar pattern to R e (Figure 4). In the establishment phase, GPP was considerably lower than in the following years. While in the last 3 years the highest GPP occurred in June, in the first year it occurred in July, thus 1 month later (Figure 4). Switchgrass was reported to maximize the photosynthetic rate in the range of 30-35°C (Zeri et al., 2011). As this range was recorded only in 2012, GPP might have been limited by temperatures in the following 3 years. However, although lower than measured by Zeri et al. (2011;~À51 to À80 Mg CO 2 ha À1 year À1 ), GPP measured at our site (À44.1 to À65.0 Mg CO 2 ha À1 year À1 ; Table 3) was substantially higher than recorded in other studies on switchgrass (Skinner & Adler, 2010;Wagle & Kakani, 2014).
The ecosystem C assimilation (negative NEE) lasted 176 (from May), 182, 226 and 185 days, from first to fourth year, respectively, which is a considerable longer period than that measured by Skinner and Adler (2010) or by Zeri et al. (2011) in the northern United States (98 and 90 days, respectively). The NEE usually turned negative in the second half of March (earlier in 2014, at the end of February) and stayed negative until October. In an important way, CO 2 assimilation rates (GPP) were still substantial at the time of harvest, suggesting that it could be delayed to the end of October in order to exploit a longer accumulation period.
Concerning the C partition of NEE by measuring the C content of the above-, belowground biomass and soil in the switchgrass stand, our results seem to confirm that most of the C storage occurred in the upper soil (0-0.45 m). This is in agreement with previous studies (Garten & Wullschleger, 2000;Qin et al., 2016), although also depending on the specific land-use history of the experimental site.
Switchgrass-derived SOC in four growing seasons corresponded to 9.45 Mg/ha. Given the net SOC change in four growing seasons (0.52 Mg/ha), it derives that about 8.9 Mg/ha of older SOC (C3-derived) was lost by the ecosystem, and this emission corresponded to the 27% of total R e from establishment to the end of the fourth growing season (November 2015). In an interesting manner, in a nearby long-term switchgrass trial where the former crop had been poplar (Populus spp.; Nocentini & Monti, 2017), different SOC dynamics were observed, likely due to distinct land-use change effects. In the first 3 years after poplar conversion into switchgrass, poplar-derived SOC increased by 0.15 g kg À1 year À1 while the increment in switchgrass-derived SOC was insignificant (Nocentini & Monti, 2017), because of the persistence of poplar residues. On the contrary, in the present study, residues from annual crops were largely emitted in the first 3 years after conversion into switchgrass, allowing a substantial switchgrassderived SOC deposition (in November 2015, 14% of total SOC was switchgrass-derived). The net SOC increase was quite low in the first years of switchgrass production, nevertheless it should be taken into account that the soil C stock at establishment was substantial (67.0 Mg/ha in the top 0.45 m), likely due to the residue management normally applied at the experimental farm (i.e. annual crops' residue reintegration) combined with the fine texture of the soil which tends to bind organic matter more strongly and thus stabilize it more compared to sandy textured soils (Parton, Schimel, Cole, & Ojima, 1987).
After 4 years of cultivation, most of the sequestered C (94%) was then still stocked into the more biochemically active litter and root pools. Nevertheless, these C pools will, more or less substantially, contribute to SOC in the following years or when, after 10 years, the soil will be tilled before restarting a new switchgrass (or other perennial crop) cycle (during which new litter and root C pools are going to form). Indeed, even before tillage, litter can be incorporated into the soil by the activity of the microfauna (Agostini, Gregory, & Richter, 2015), and, further, in perennial herbaceous systems, coarser roots can persist in the soil for years before being mineralized, so that their incorporation into the SOC pool can occur when the stand is already mature (Agostini et al., 2015).
Life cycle CO 2 emission was estimated À12.4 Mg CO 2 eq ha À1 year À1 in the first 4 years of switchgrass cultivation. This is a very high CO 2 reduction potential if compared to similar studies (Gelfand et al., 2011Skinner & Adler, 2010;Zenone et al., 2013;Zeri et al., 2011), although its higher significance can be F I G U R E 6 Cumulated net ecosystem exchange (NEE) and cumulated net ecosystem carbon balance (NECB) in the 4 years of switchgrass cultivation. NECB represents the NEE discounted by the carbon removed with biomass harvest; the lower the values the higher the CO 2 uptake by the ecosystem largely explained by the high biomass production and, more in general, by the high sink capacity (NEE) attained by switchgrass in the present study. The main reasons for the discrepancy are likely the longer growing season and accumulation potential in the warmer Mediterranean environment and the land-use change from annual food crops. For example, Skinner and Adler (2010) and Zeri et al. (2011) measured less than 100 days with continuous negative NEE in the northern USA, while, at Cadriano site, on average, switchgrass acted as a net sink for 198 days in a row. If excluding the establishment year, mean biomass harvest yields at Cadriano site were 11.9 Mg ha À1 year À1 , while they were 4.9 and 5.6 Mg ha À1 year À1 in Skinner and Adler (2010) and Zenone et al. (2013), respectively. In Zeri et al. (2011) aboveground biomass yield of switchgrass was high (11.5 Mg ha À1 year À1 ) and, similarly to the present study, the measured NECB was À7.3 Mg CO 2 ha À1 year À1 ; also in their study switchgrass was cultivated after the former land-use was annual food crops (oats). Gelfand et al. (2013) measured aboveground yields of about 15 Mg ha À1 year À1 for their successional grass system and their estimation of life cycle GHG emission is slightly lower than found here (À9.2 Mg CO 2 eq ha À1 year À1 ), as the fossil fuel offset credits were over 6 Mg CO 2 eq ha À1 year À1 , but the NECB was not as impactful (À4.0 Mg CO 2 ha À1 year À1 ), although significant; in their study the land-use change was from unmanaged marginal lands, thus the potential to accumulate ecosystem C lower. Zenone et al. (2013) included a scenario where switchgrass followed annual crops, but, surprisingly, in their study this land-use change produced CO 2 emissions rather than acting as sink (NECB = 5.5 Mg CO 2 ha À1 year À1 ), even after including fossil fuel offset credits in the balance (4.1 Mg CO 2 eq ha À1 year À1 ). At their site switchgrass harvested biomass was low, probably due to a slow establishment phase and, however, their measurement were limited to the first 2 years of switchgrass. Furthermore, in Zenone et al. (2013) the land had been converted to a corn-soybean rotation in the last few decades before land-use change to switchgrass, whereas it had been a grassland, probably accumulating high C stocks, before then.
Our estimated emission from agronomic inputs use (733 kg CO 2 eq ha À1 year À1 ) is in line with previous studies in the same region (686 kg CO 2 eq ha À1 year À1 ; Fazio & Monti, 2011), and with studies carried out elsewhere (635 kg CO 2 eq ha À1 year À1 ; Gelfand et al., 2013).
Although not insignificant, estimated N 2 O emissions (0.9 Mg CO 2 eq ha À1 year À1 ) did not have a substantial impact on the Life cycle GHG balance of switchgrass (Table 4). Also, Skinner and Adler (2010) reported low N 2 O emissions in switchgrass ranging from 0.3 to 0.5 Mg CO 2 eq ha À1 year À1 , thus slightly influencing NEE. This was also confirmed by Don et al. (2012) that found perennial biofuel grasses to emit 40%-99% less N 2 O than annual cropping systems.
Methane (CH 4 ) emission was not assessed in this study, but it was found to have a very limited impact on Life cycle GHG emission of biofuel perennial systems (Gelfand et al., 2011. The present study did not include ILUC emissions in switchgrass GHG balance, as their estimation is uncertain (Plevin et al., 2010). If adopting the factors and formulas from Plevin et al. (2010) and using our estimated switchgrass ethanol yields, ILUC emissions would vary between 13 and 456 g CO 2 MJ À1 , thus from very low to highly impactful. Nonetheless, Zilberman (2017), in a recent commentary, using empirical evidence based on history, argued that ILUC might just be "much ado about (almost) nothing." Moreover, given the progressive abandonment of cropland observed in Europe in the last decades (Ramankutty & Foley, 1999), it can be hypothesized that converting marginal cropland to switchgrass in the Mediterranean might not generate much ILUC effect.
Regarding water use, almost 60% of annual precipitation was used by the crop, while most of the ETc (up to 93%) occurred within the growing season only (Table 2). Hence, switchgrass growth was apparently not limited by water availability. It should also be noted that the mean annual depth of the water table in this area is 140 cm, thus quite shallow. This gives the possibility to a deep-rooted crop such as switchgrass to use the belowground water reserve if facing short drought periods (Abraha et al., 2015).
Measured WUEs were higher than WUEs observed in miscanthus or giant reed in the Mediterranean area (Triana, Nassi o Di Nasso, Ragaglini, Roncucci, & Bonari, 2014), two perennial lignocellulosic crops of commercial interest T A B L E 4 Summary of Life cycle CO 2 emission from a farmscale switchgrass stand in North Italy monitored for four consecutive years. The net ecosystem C balance (NECB) was measured by an eddy-covariance system and by carrying out biomass harvest every year. N 2 O emissions were estimated with the DAYCENT model. Emissions from agronomic input use were estimated using a dedicated software (SimaPro 8.0 that could compete with switchgrass for the same land. A correlation between CO 2 uptake (GPP, NEE) and ETc was found. A strong correlation in switchgrass was also found by Skinner and Adler (2010) between GPP and ETc (r 2 = 0.87). The linear correlation of ETc with NEE and GPP (Table 5) suggested a strong linkage between C fixation and water loss over the seasons. The results from this study can provide a strong base to estimate potential environmental benefits of the large-scale deployment of switchgrass in the Mediterranean region. Nocentini et al. (2015), using the Corine Land Cover database (European Environment Agency, 2010), estimated that 1.21 Mha of marginal cropland might be used for switchgrass cultivation in the Mediterranean. It is assumed here that most of the marginal cropland identified by Nocentini et al. (2015) had been, as the experimental site in this study, under arable land use for many decades, as, opposite to agricultural expansion, a progressive abandonment of cropland in Europe (and the Mediterranean; Greco & Bellini, 2010) has been observed since the 1960s (about 20 Mha of abandoned cropland; Ramankutty & Foley, 1999), mainly in marginal areas (Hatna & Bakker, 2011). Hence, if we hypothesize À12.4 Mg CO 2 ha À1 year À1 absorbed during switchgrass life cycle, switchgrass could potentially capture, in the whole region, 15.0 Mt CO 2 during each year after establishment. Ten years of switchgrass production (i.e. the economic life span normally considered for a switchgrass stand) may therefore offset EU's CO 2 equivalent emissions (4,451.8 Mt CO 2 year À1 ; EURO-STAT, 2015) by 0.33% every year; this value is highly relevant if weighted on the 20% reduction goal set by EU for the period 1990-2020 (Delbeke & Vis, 2016). Nonetheless, it must be considered that this is likely an overestimation as the NECB and ethanol yields might reduce in time due to ecosystem carbon saturation and stand maturation, especially between the sixth and tenth year of production (Arundale et al., 2014;Eichelmann, Wagner-Riddle, Warland, Deen, & Voroney, 2016;Nocentini et al., 2015). In a possible scenario, switchgrass ecosystem may have reached C equilibrium at the fourth year, so that from the fifth to the tenth years of cultivation the NECB factor would equal zero in the life cycle balance, and this would reduce the life cycle emission to À7.4 Mg CO 2 eq ha À1 year À1 and the EU's GHG reduction to 0.20% every year. If this is true, at the same time, more CO 2 would probably be saved, if considering the foregone emission of the former annual crops cultivation (i.e. higher emission from agronomic inputs and higher N 2 O emissions) compared to switchgrass perennial system to which it was converted (Adler et al., 2007;Don et al., 2012;Fazio & Monti, 2011). At last, although highly uncertain (Plevin et al., 2010) and unlikely to be very impactful in the "old" Europe (Zilberman, 2017), especially after marginal cropland conversion (Hatna & Bakker, 2011;Ramankutty & Foley, 1999), ILUC emissions could totally offset the GHG benefits brought by switchgrass large-scale cultivation in Mediterranean marginal cropland. Therefore, more accurate methods to determine ILUC are necessary as soon as possible, although not yet at the horizon (Plevin et al., 2010).
T A B L E 5 Correlation factors between cumulative evapotranspiration (ETc) and net ecosystem exchange (NEE) or gross primary production (GPP). ETc-to-NEE ratio indicates the amount of evapo-transpired water for the net fixation of CO 2 at ecosystem level, and ETc-to-GPP ratio indicates the amount of evapo-transpired water for the absorption of CO 2 by the ecosystem