Can land degradation drive differences in the C exchange of two similar semiarid ecosystems?

The concept of land degradation stems from the loss of an ecosystem’s biological productivity, which in turn relies on several degradation processes, such as long-term loss of natural vegetation, depletion of soil nutrients, soil compaction or water and wind erosion, to which drylands are especially vulnerable. Currently, drylands occupy more than one third of the global terrestrial surface and will probably expand under future climate change scenarios. Drylands ́ key role in the global C balance has been recently demonstrated, but the effects of land degradation on C sequestration by these 20 ecosystems needs further research. In the present study, we compare net carbon exchange, together with satellite data and meteorological, ambient and vadose zone (CO2, water content and temperature) variables, between two nearby (~23 km) experimental sites representing “natural” (i.e. site of reference) and “degraded” grazed semiarid grasslands located in SE Spain, via eddy covariance measurements over 6 years, with highly variable precipitation magnitude and distribution. Results show a striking difference in the annual C balances with an average release of 196 ± 40 and -23 ± 20 g C m -2 yr -1 for 25 the “degraded” and “natural” sites, respectively. At the seasonal scale, differing patterns in net CO2 fluxes were detected over both growing and dry seasons. As expected, during the growing seasons, greater net C uptake over longer periods was observed in the “natural” site, however, much greater net C release was measured in the “degraded” site during drought periods. We tested differences in all monitored meteorological and soil variables and found it most relevant that CO2 at 1.50 m belowground was around 1000 ppm higher in the “degraded” site. Thus, we believe that subterranean ventilation of this 30 vadose zone CO2, previously observed at both sites, largely drives the differences in C dynamics between them, especially during the dry season maybe due to enhanced subsoil-atmosphere interconnectivity in the “degraded” site. Overall, the 12 site-years of data allow direct exploration of the roles of climate and land degradation in the biological and non-biological processes that ultimately control the C sequestration capacity of semiarid ecosystems. Biogeosciences Discuss., doi:10.5194/bg-2017-77, 2017 Manuscript under review for journal Biogeosciences Discussion started: 21 March 2017 c © Author(s) 2017. CC-BY 3.0 License.


Introduction
The concept of land degradation stems from the loss of an ecosystem's biological productivity, which in turn relies on several degradation processes such as long-term loss of natural vegetation, deterioration of soil quality, biodiversity de-Published by Copernicus Publications on behalf of the European Geosciences Union.
pletion or water and wind erosion (UNCCD, 1994).Drylands (arid, semiarid and dry subhumid areas), which occupy more than one-third of Earth's land surface and are inhabited by more than 2 billion people (Niemeijer et al., 2005), have been recognized as areas vulnerable to land degradation processes.In fact, they have expanded globally for the last 60 years at an estimated annual rate of 5.8 million ha in midlatitudes alone (Lal, 2001) and are projected to expand under future climate change scenarios (Feng and Fu, 2013;Cook et al., 2014), especially in the Mediterranean region, where major expansions of semiarid areas will occur (Gao and Giorgi, 2008;Feng and Fu, 2013).
Over recent decades, most research focused on land degradation has been based on remote sensing and Earth observation techniques.Much of these investigations have aimed to refine methodological issues in order to accurately track land degradation in vulnerable areas, reduce uncertainties and explain inconsistencies among studies.For instance, a wide array of satellite-derived data, such as vegetation indices, normalized surface reflectance, brightness temperature or biomass net primary production derivatives (Mbow et al., 2015), have been utilized to appraise desertification effects in the Sahel (Mbow et al., 2015;Fensholt et al., 2013) and also in other African countries such as Kenya (Omuto, 2011), Somalia (Omuto et al., 2010), South Africa (Thompson et al., 2009) or Zimbabwe (Prince et al., 2009).Likewise, desertification in the Mediterranean region has been studied through satellite imagery in Greece (Bajocco et al., 2012), Israel (Shoshany and Karnibad, 2015) and the Iberian Peninsula (del Barrio et al., 2010).However, although drylands' key role in the global carbon (C) balance has been demonstrated (Poulter et al., 2014;Ahlström et al., 2015), very few investigations have directly quantified how land degradation processes disturb the C sequestration capacity of drylands (Lal, 2001), despite it being one of the most important ecosystem services (Watanabe and Ortega, 2011).
In this regard, the few C-related desertification studies conducted over the last decade have centered on soil C dynamics.In particular, soil organic carbon (SOC) inventories have been used to explore the effects of climate, human activities and grazing pressure in desertification-prone areas of China (Feng et al., 2006) and Brazil (Schulz et al., 2016).Similarly, other investigations have evaluated soil degradation processes by means of soil CO 2 effluxes together with other biometric measurements in drylands found in China (Hou et al., 2014;Wang et al., 2007), Chile (Bown et al., 2014) and southeastern Spain (Rey et al., 2011(Rey et al., , 2017)).However, the degradation processes associated with desertification affect several subsystems as well as their interactions at multiple spatial and temporal scales.For instance, adverse effects on soil quality involve depletion of soil fertility, but also reduce soil-water storage (Mainguet and Da Silva, 1998), which, in turn, can constrain seed germination and vegetation reestablishment, modify climax vegetation, disrupt biogeochemical cycles, alter water and energy balances, and conse-quently lead to a loss of ecosystem resilience (Lal, 2001).This cascade of disturbances may result in a reduction of the C sequestration capacity of a given ecosystem, which is clearly a symptom of the loss of biological productivity, resulting in a positive feedback to global warming.Therefore, a quite suitable and holistic approach is to integrate all subsystem effects into a whole ecosystem-scale assessment when quantifying the C loss derived from land degradation.However, the use of this integrative method is mostly lacking in the available literature.
The present study is located in the southeast of Spain, which has been recognized as a hotspot of land degradation owing to the synergistic interaction of sociological and climatic factors (Puigdefábregas and Mendizabal, 1998).Our core aim is to evaluate how dryland degradation affects the dynamics of net ecosystem-atmosphere C exchange of two semiarid grasslands that represent differing degradation statuses (natural versus degraded) by means of meteorological, satellite and subsoil CO 2 measurements together with carbon and water fluxes acquired by the eddy covariance (EC) technique (Baldocchi et al., 1988).Owing to the high temporal resolution of the EC method, we can assess the effect of land degradation as a slow change or disturbance legacy in the studied ecosystems and how, in turn, it influences the ecosystems' resilience against short-term disturbances, such as climate extremes (i.e., droughts, heatwaves).
Some land degradation processes are evident when we compare the natural site with the degraded site.Accordingly, our main hypothesis is that land degradation processes can directly affect abiotic and/or biotic factors and consequently influence the biological and/or nonbiological processes that compose the net ecosystem CO 2 exchange: gross primary production, ecosystem respiration and subterranean ventilation -a nonbiological process that provokes the transfer of CO 2 -rich air from the subsoil to the atmosphere under drought and high-turbulence conditions.Firstly, the lower vegetation cover at the degraded site would entail a higher thermal and radiative stress at the surface, especially during the drought period (Rey et al., 2017).The hypothesized effects on biological processes are a direct reduction in plant productivity and respiration, and an indirect decrease in heterotrophic respiration.Secondly, the higher cover of bare soil and outcrops at the degraded site may increase the soilatmosphere interconnectivity, which can indirectly enhance the presence of advective CO 2 release through subterranean ventilation, which has been previously measured at both experimental sites (Rey et al., 2012a;López-Ballesteros et al., 2017).Thirdly, the reduced soil fertility and depth may provoke changes in microbial communities (Evans and Wallenstein, 2014) due to stronger nutrient and water limitations.Consequently, a direct decrease in heterotrophic respiration and plant productivity and respiration is expected.
Hence, our specific objectives are (1) to compare the C sequestration capacity of two semiarid ecosystems with differing degradation status, (2) to study the underlying pro-cesses (biological versus nonbiological) and influencing factors that can drive potential differences in the net C exchange of studied ecosystems, and (3) to evaluate whether degradation can modulate ecosystem responses against short-term disturbances.

Material and methods
In this study, we analyzed 12 site-years of EC data, enhanced vegetation index (EVI) time series and monitored ambient variables registered over the same period (2009)(2010)(2011)(2012)(2013)(2014)(2015) at two experimental sites.Additionally, we used subsoil CO 2 , moisture and temperature data obtained during 2014-2015.

Experimental site description
The study area is located in southeastern Spain, the driest part of Europe.The two experimental sites, Amoladeras (36 • 50 5 M, 2 • 15 1 W) and Balsa Blanca (36 • 56 26.0 N, 2 • 01 58.8 W), are found within the Cabo de Gata-Níjar Natural Park (Almería, Spain; Fig. 1) and are quite similar in terms of climate and ecosystem type.Both sites show a desert climate, according to the Köppen classification (Bwh; Kottek et al., 2006), with a mean annual temperature of 18 • C and mean annual precipitation of approximately 220 mm.
The ecosystem type corresponds to espartal, a Mediterranean semiarid grassland where the dominant species is Machrochloa tenacissima.This ecosystem type is widespread in the western Mediterranean region: in Cabo de Gata-Níjar Natural Park, a large fraction of agricultural areas that were abandoned over 1957-1994 resulted in espartal ecosystems (Alados et al., 2004(Alados et al., , 2011)).The functioning of both experimental sites can be divided into two main periods.On one hand, the growing season usually extends from late autumn to early spring, when the temperature starts to rise and water resources have not yet become scarce (López-Ballesteros et al., 2016;Serrano-Ortiz et al., 2014).On the other hand, a long period of hydric stress, with high temperatures and scarce precipitation, results in a prolonged dry season that usually begins in May-June and ends in September-October, when the first autumn rainfall events occur.Additionally, water inputs derived from relevant dewfall episodes, which have been previously reported in the area (Uclés et al., 2014), can rehydrate soil and plants at night and in the early morning hours.
Regarding the topographic, geologic and edaphic characteristics, both sites are located on an alluvial fan, where the main geological materials consist of quaternary conglomerates and Neogene-Quaternary sediments cemented by lime (caliche) (Rodríguez-Fernández et al., 2015) on slopes of 2-6 % (Rey et al., 2017), so no significant runoff occurs.Additionally, both sites present petrocalcic horizons.However, altitude and soil type differ.While Balsa Blanca (hereinafter BB) is located at an altitude of 208 m and has Mollic and Lithic Leptosols (Calcaric), Amoladeras (hereinafter AMO) is situated closer to sea level, at 65 m and presents less developed soils Lithic Leptosol (Calcaric; Table 1).
Overall, as stated by Rey et al. (2011), these two experimental sites represent different degradation stages owing to their differing soil characteristics and surface fractions (Table 1).While BB has deeper and more fertile soils and higher vegetation cover, AMO shows thinner and poorer soils and has half of Balsa Blanca's vegetation cover.Therefore, in accordance with Rey et al. (2011Rey et al. ( , 2017)), we considered that BB represents the natural site, currently being a representative ecosystem of the area, while AMO represents a degraded site with respect to BB.The stronger degradation effects observed in AMO (degraded site) compared to BB (natural site) are probably due to its proximity to populated areas.The main factor provoking degradation in this Mediterranean area was the increase in rural population from the beginning of the 20th century to the late 1950s (Grove and Rackham, 2001).At that time, timber extraction and the use of tussock fiber for textile manufacturing and extensive farming were common economic activities that likely increased anthropic pressure on the degraded site.Afterwards, rural exodus during the middle of the century involved the abandonment of these agriculture practices.However, although degradation drivers are not currently active, their effects are still observable in the area corresponding to a case of "relict" degradation (Puigdefábregas and Mendizábal, 2004).Additionally, flux measurements acquired under lowturbulence conditions were excluded from the analysis by using a friction velocity (u * ) threshold according to the approach proposed by Reichstein et al. (2005).The average u * thresholds for the entire study period (i.e., 2009-2015) were 0.11 and 0.16 m s −1 , for AMO and BB.Therefore, the total annual fractions of missing half-hourly net CO 2 fluxes accounted for 33 ± 3 and 29 ± 6 % of nighttime data and 8 ± 6 and 14 ± 5 % of daytime data, for AMO and BB, respectively.Missing data were gap-filled by means of the marginal distribution approach proposed by Reichstein et al. (2005) and uncertainty derived from the gap-filling procedure was calculated by using the variance of the measured data, which were calculated by introducing artificial gaps and repeating the standard gap-filling procedure.Twice the standard deviation of sums of total data were taken as the uncertainty for the several aggregating time periods we used in the analysis.The annual cumulative C balance was estimated, when possible, by integrating gap-filled half-hourly net CO 2 fluxes of good quality (0 and 1 quality flags, according to Mauder and Foken, 2004) over a hydrological year.
In order to test the validity of both EC stations, we assessed the energy balance closure (Moncrieff et al., 1997) by computing the linear regression of half-hourly turbulent energy fluxes (sensible and latent heat fluxes; H + LE; W m −2 ) against available energy (net radiation less the soil heat flux; R n − G; W m −2 ) using the whole of the 6-year database.The storage term in the soil heat flux was included in the estimates, while in the cases of sensible and latent heat fluxes, this term was negligible given the short height of the vegetation (∼ 50 cm).The resulting slopes were 0.873 ± 0.002 (R 2 = 0.907) and 0.875 ± 0.001 (R 2 = 0.920) for AMO and BB, respectively.

Flux partitioning
In order to partition net CO 2 ecosystem exchange into gross primary production (GPP) and ecosystem respiration (R eco ), we first modeled the ventilative CO 2 efflux by adapting the approach proposed by Pérez-Priego et al. ( 2013) using the results of previous studies at both sites (López-Ballesteros et al., 2016, 2017).Essentially, we aimed to isolate those moments when subterranean ventilation (V n ) dominates the net CO 2 fluxes (F c ) and biological fluxes are negligible.These moments correspond to daytime hours during the ex- tremely dry periods.Accordingly, data were selected using the following conditions: (i) net radiation > 10 W m −2 , (ii) 8 < daily averaged Bowen ratio < 10, and (iii) daily soil water content (in bare soil) < 10th percentile (in AMO) and < 20th percentile (in BB).A less restrictive threshold was used in BB in order to get enough data to build the V n model, since long-term data gaps occurred at this site during the summer seasons of 2012, 2014 and 2015.Afterwards, in order to build the linear model of V n , these selected F c data (maximum quality; QC flag = 0) were related to the friction velocity (u * ).
As the results show (Table 3; Fig. S1 in the Supplement), the V n model is uniquely valid for AMO.Therefore, we only applied the V n model to AMO data, particularly during those periods when ventilation (but not exclusively) occurs according to previous research (López-Ballesteros et al., 2017).Hence, the model was applied when (i) net radiation > 10 W m −2 , (ii) daily averaged Bowen ratio > 4, (iii) daily soil water content (in bare soil) < 0.01 m 3 m −3 , and (iv) σ swc (daily variance of soil water content in bare soil) < 5 × 10 −6 (m 3 m −3 ) 2 .We use those moments with very low σ swc in order to discern R eco increases caused by rain pulses (Birch effect) from V n fluxes during the dry season.Then, the modeled ventilative fluxes were subtracted from the measured net CO 2 exchange to obtain the CO 2 flux corresponding only to biological processes (i.e., biological F c ; see Fig. S2).
Finally, the partitioning approach proposed by Lasslop et al. (2010) was applied to the biological F c for both sites in order to obtain GPP and R eco fluxes.We chose this approach given the determinant influence of hydric stress, in this case atmospheric drought (assessed via VPD), on the physiology of Machrochloa tenacissima, the dominant plant species of the studied semiarid ecosystems (Pugnaire et al., 1996;López-Ballesteros et al., 2016).

Enhanced vegetation index data series
We used EVI data acquired by the Moderate Resolution Imaging Spectroradiometer (MODIS), which is on board the Earth Observing System-Terra platform, in order to track vegetation dynamics at both experimental sites.The nominal resolution of EVI products (code MOD13Q1) is 250 m at nadir and temporal resolution corresponds to 16-day compositing periods.The spatial coordinates used for AMO and BB were 36.8340• N, −2.2526 • E and 36.9394• N, −2.0341 • E, respectively.

Vadose zone measurements
Subsoil CO 2 molar fraction, temperature and volumetric water content were measured at 0.05 and 1.50 m below the surface (Table 2) from January 2014 to August 2015 at both experimental sites.In the case of the shallower CO 2 sensor, it was installed vertically with an in-soil adapter (211921GM, Vaisala, Inc., Finland) to avoid water entering.Subsoil CO 2 molar fractions were sampled every 30 s and 5 min averages were stored in a data logger (CR3000 and CR1000, CSI; for AMO and BB, respectively).The deeper CO 2 sensor was equipped with a soil adapter for horizontal positioning (215519, Vaisala, Inc., Finland) and consisted of a PTFE filter to protect to the CO 2 sensor from water.It was buried in the summer of 2013 and the measurements were made every 30 s and stored as 5 min averages in a data logger (CR1000 and CR23X Campbell Sci., Logan, UT, USA, for AMO and BB, respectively).All CO 2 molar fraction records were corrected for variations in soil temperature and atmospheric pressure.

Statistical analysis
All meteorological and soil variables monitored at each site were compared through computation of the nonparametric two-sided Wilcoxon summed rank test in order to detect those factors/variables influencing potentially distinct ecosystem functioning between sites.This test was chosen because the variables used satisfied the independence and continuity assumptions but not all were normally distributed.The confidence level used was 95 %.The effect size was evaluated using the median of the difference between the samples (AMO minus BB), which was expressed as a standardized value (divided by its standard deviation; Diff st ; dimensionless) in order to be able to compare results among different variables.This analysis was performed by using three different periods: the entire study period, the period from May to September and the period from May to September only during the daytime.These periods were selected given their demonstrated coincidence with high relevance of nonbiolog-  ical processes.All calculations were performed using R software version 3.2.5.Additionally, in order to include the relationship between pressure and subsoil CO 2 variations as a potential factor influencing net CO 2 exchange (Sánchez-Cañete et al., 2013), we first calculated, separately for each site, Spearman correlation coefficients to determine the time step (6, 12, 24 or 72 h) with the highest correlation between the differential transformation of pressure and the subsoil CO 2 molar fraction at 1.50 m.

Ambient conditions
Over the study period, the wettest hydrological year was 2009-2010, with an annual precipitation of ∼ 500 mm (ca.twice the mean annual precipitation for both sites over the study period, Fig. 2).In contrast, the driest year was 2013-2014, with an annual precipitation of ∼ 100 mm for both sites, less than half the annual average precipitation registered at AMO and BB.Generally, months with precipitation exceeding 20 mm occurred from the beginning of autumn to midwinter.However, in 2009-2010, 2010-2011, 2012-2013 and 2014-2015, relevant rain events were registered during the spring months.By contrast, in 2013-2014, precipitation was always below 20 mm with the exception of November and December, for both sites, and June, in the case of AMO (Fig. 2a).Commonly, while maximum precipitation usually occurred from November to February, there was a remarkable drought period over the summer months (June-August) when it scarcely ever rained (Fig. 2).
Regarding air temperature (T air ) patterns, monthly averaged T air ranged from 9.6 and 8.1 • C to 27.6 and 27.9 • C in AMO and BB, respectively, over the entire study period.Based on half-hourly averaged data, minimum and maximum T air values registered were 0.1 and 37.9 • C in AMO, and −1.3 and 39.9 • C, in BB, respectively.On one hand, those months with T air above 15 • C usually corresponded approximately to April-November.Additionally, August was the month with the highest average T air at both sites, with T air ranges of 25.2-27.6• C at AMO and 24.9-27.9• C at BB, respectively (Fig. 2), over the study period.On the other hand, the lowest monthly average T air usually occurred in January but sometimes also in December and February, with 11.2-12.3• C at AMO and 8.1-14.1 • C at BB.

Annual carbon balances
The comparison of the annual C balance among sites was only possible for 3 hydrological years, 2009-2010, 2010-2011 and 2012-2013, due to long-term data gaps existing in BB during other years.The annual cumulative net CO 2 exchange was always positive for AMO (i.e., net C release), whereas BB was neutral or even acted as a C sink over the 3 years (Fig. 3).For example, in 2009-2010, the net C uptake measured in BB equated to 32 ± 10 g C m −2 , while in AMO, a total amount of 185 ± 10 g C m −2 was released into the atmosphere (Fig. 3a).The year with the largest difference between sites was 2010-2011, with an annual C release of 240 ± 8 and annual C uptake of −38 ± 10 g C m −2 in AMO and BB, respectively (Fig. 3b).Likewise, 2012-2013 was the year in which the lowest CO 2 release was measured in AMO with 163 ± 7 g C m −2 , while a neutral C balance was measured in BB with 0 ± 8 g C m −2 (Fig. 3c).
Overall, a positive and saturating trend was observed at both sites during autumn months until December-February when cumulative net CO 2 releases started to decline.The autumn net CO 2 release (i.e., positive values) was usually higher in AMO than in BB, except for 2012-2013, and the declining slope was always higher in BB, meaning greater net C uptake rates.Although the pattern of the cumulative net CO 2 exchange showed differences between sites over autumn, winter and spring months, stronger discrepancies were found during summer droughts.From April-May to August, BB showed neutral behavior, while a remarkably positive trend was observed in AMO, denoting a large net CO 2 release.

Seasonal and diurnal net CO 2 exchanges
Long-term data loss occurred in BB during the springs of 2011-2012, 2013-2014 and 2014-2015 and summers of 2013-2014 and 2014-2015, when annual C balances could not be estimated.However, by observing the available seasonal data, it is noticeable that maximum and minimum seasonal net CO 2 exchanges were very different between sites (Fig. 4).On one hand, maximum seasonal net CO 2 uptake was measured during winter (December-February) in AMO and over spring (March-May) in BB, when peak net CO 2 uptake equated to −31 g C m −2 (winter 2011-2012) and −105 g C m −2 (spring 2010-2011) in AMO and BB, respectively.Additionally, net CO 2 uptake was only observed during three winters in the case of AMO, whereas it was frequently measured during both winter and spring in BB.On the other hand, the cumulative net CO 2 release into the atmosphere occurred over all seasons in AMO but acutely in summer, when maximum seasonal net CO 2 release was al-  Major intersite differences in autumn occurred in the first and last years of study, when ETs were 23 and 12 % higher at BB (Fig. 5).
When comparing daily-scale net CO 2 exchange and ET fluxes with enhanced vegetation index (EVI) data, we notice some similarities in the general patterns of both sites (Fig. 6).There was a common annual pattern in which the highest values of EVI coincided with maximum net CO 2 uptake rates (i.e., negative net CO 2 fluxes), which in turn, corresponded to peak ET fluxes.Additionally, a decreasing trend in EVI over the 6 years of study was also noticeable for both sites.However, some intersite and interannual differences were evident (Fig. 6).
On one hand, there were two main differences between sites.Firstly, extreme net CO 2 release was measured uniquely in AMO during summer months (June-August), when maximum net CO 2 fluxes ranging from 31 to 68 g C m −2 were measured (Fig. 6b).Over the study period, the monthly net CO 2 exchange of AMO during dry seasons was up to 100 times higher than in BB (in August 2013), since monthly net CO 2 fluxes measured in BB were much lower, from −8 to 16 g C m −2 (Fig. 6b).Besides the striking differences in summer net CO 2 exchange between sites, minor discrepancies were also found in ET fluxes and EVI for the same drought periods.In this regard, monthly averaged ET over the dry season equated to 13 ± 4 and 10 ± 4 mm for AMO and BB, respectively, and EVI was on average 4 % higher in BB than in AMO (Fig. 6a and c).The second intersite difference was the greater net CO 2 uptake over longer periods measured in BB.In particular, the period during which the ecosystems acted as C sinks lasted on average 38 days longer in BB than in AMO annually (Table 4).Accordingly, the annual amount of C fixation ranged from 6-59 g C m −2 at AMO and 15-129 g C m −2 at BB, with the annual averaged net C uptake in BB 162 % higher than at AMO (Table 4).Consequently, peak EVI values were usually observed during March-April for both sites; however, over winter and spring months (growing period), EVI measured at BB was 3-37 % higher than AMO, with the largest intersite differences in 2009-2010 and 2014-2015 (Fig. 6a).Likewise, monthly averaged ET fluxes measured at BB over winter and spring months (December-May) were from 3 to 24 % larger than those measured at AMO.Additionally, the growing period of the driest year (2013-2014) corresponded to the lowest monthly ET fluxes and the least difference between sites.On the other hand, differences in the interannual variability of EVI were found between years.To be precise, 2009-2010 and 2013-2014 were the years with maximum and minimum annual precipitation and EVI observations, respectively, for both sites.In 2009-2010, EVI observations were 28 and 20 % higher than the 6-year averaged values in BB and AMO.In the driest year, 2013-2014, growing season (winter-spring) EVI was reduced to 35 and 28 %.Nevertheless, the largest difference between sites in winter-spring EVI observations was found in 2014-2015, following the driest year, when BB showed a pattern very similar to those registered over the years prior to the dry spell, while AMO still presented EVI values 21 % below the 6-year average (Fig. 6a).

Biological net CO 2 exchange, gross primary production, ecosystem respiration and water use efficiency
The results of the "biological" annual C balance were obtained for AMO by applying the ventilation model and are in accordance with our hypotheses.Annual C emission was always measured at AMO, whereas BB acted as a neutral and mild C sink.On average, AMO emitted 32 g C m −2 more than BB.On a monthly timescale, net CO 2 fluxes during autumn were, on average, ∼ 4 times higher at BB, except in the last study year, when the net CO 2 emission at AMO was 21 times greater than at the natural site.However, during winter and spring months, net CO 2 uptake was generally higher at the BB (Fig. 7a).On average, during the 6 years of study, GPP, R eco and water use efficiency (WUE) were 9, 2 and 10 times higher at BB compared to AMO.Firstly, GPP was always higher at BB compared to AMO (Fig. 7c).Major differences occurred in autumn 2014-2015, when monthly cumulative GPP at BB was 32 times higher on average.Similarly, R eco was generally higher, up to ∼ 8 times at BB (October 2014).However, respiratory fluxes were occasionally greater at AMO, from 2 to 31 % higher, during spring and winter months of all studied years except 2013-2014 (Fig. 7b).Maximum intersite differences in GPP and R eco were found in winter and autumn 2014-2015, following the driest year, when monthly GPP was ∼ 30 times higher at BB compared to AMO.Similarly, monthly R eco was ∼ 5 times greater at BB. Intersite differences in partitioned fluxes could not be assessed during spring months due to the lack of data from BB.Secondly, WUE was lower at AMO during the entire study period, when maximum and minimum differences coincided with the highest and lowest differences in GPP between sites.On average, monthly WUE was 6 and 1.5 times higher at BB during winter and spring.Major intersite differences were found in autumn and winter 2014-2015 (Fig. 7d).

Differences in meteorological and soil variables between sites
Results from the two-sided Wilcoxon summed rank test (Table 5) showed significant differences (p-value < 0.05) between sites in most of the monitored meteorological variables.The few exceptions were the friction velocity (u * ) when using the entire study period, the maximum wind speed registered every half-hour (WS max ) when analyzing May-September data, and the wind speed (WS) and precipitation when assessing daytime May-September data   (2009)(2010)(2011)(2012)(2013)(2014)(2015) for Amoladeras (dark red) and Balsa Blanca (grey).Missing bars correspond to long-term data losses.(Table 5).The large number of observations (n ranged from 21 410 to 205 751) produced highly significant results (Table 5).Hence, the standardized difference between the samples (Diff st ) allowed us to quantitatively explore the differences between sites.Relevant differences (Diff st > 1) were found only for the subsoil CO 2 molar fraction measured at 1.50 m depth (CO 2,1.50 m ) for all periods, and during May-September months even when using only daytime data (Table 6).In particular, CO 2,1.50 m was always higher in AMO, from 889 to 1109 ppm (Tables S1-S3).Additionally, volumetric water content at 0.05 m depth (VWC 0.05 m ) was also higher in AMO compared to BB but only during summer months (Table 6), when absolute differences were very small, ranging from 0.028 to 0.037 m 3 m −3 (Tables S1-S3).
In contrast, subsoil CO 2 molar fraction measured at 0.05 m depth (CO 2,0.05 m ) was from 89 to 150 ppm higher in BB when analyzing dry season (May-September) daytime data (Tables 6 and S1-S3).
The temporal dynamics of subsoil CO 2 molar fractions revealed similar annual patterns between sites; generally, however, CO 2,0.05 m was higher in BB, from 6 to 88 %, while CO 2,1.50 m was always greater in AMO, from 31 to 97 % (Fig. 8).On one hand, the maximum monthly averaged values of CO 2,0.05 m were registered in autumn, particularly in November and October which had 642 and 1120 ppm in AMO and BB, whereas minimum values occurred in September and August with 373 and 400 ppm at each site (Fig. 8a).On the other hand, peak monthly averaged values of CO 2,1.50 m occurred in July for both sites, with 2751 and 1602 ppm in AMO and BB, although relatively high CO 2,1.50 m was also measured during November in BB.In contrast, minimum values were observed in December and February, with 1364 and 735 ppm in AMO and BB, respectively (Fig. 8b).
Finally, results of the Spearman correlation analysis between net CO 2 exchange and belowground CO 2 at 1.50 m depth (CO 2,1.50 m ) showed a positive relationship between both absolute variables, which was stronger for AMO compared to BB with Spearman correlation coefficients (r s ) of 0.30 and 0.11, respectively (Table 7).In contrast, a negative relationship was found between pressure and CO 2,1.50 m with higher r s in AMO compared to BB (Table 7).Additionally, r s were maxima at 12 h intervals for CO 2,1.50 m and pressure increments (dP 12 h ) at AMO, and at 6 h intervals for CO 2,1.50 m and pressure increments (dP 6 h ) at BB, with r s equal to −0.87 and −0.63.

Discussion
Our results verify that land degradation affects the C sequestration capacity of semiarid ecosystems, since relevant differences between sites were observed during the growing season, when greater net C uptake over longer periods was observed at the natural site (BB).However, contrary to what we previously hypothesized, a much greater net C release was measured at the degraded site (AMO) over drought periods due to the predominance of subterranean ventilation (López-Ballesteros et al., 2017).In fact, the great difference in annual C budgets between sites (Fig. 3) was largely related to this process resulting in an average release of 196 ± 40 and −23 ± 20 g C m −2 yr −1 for the degraded and natural sites, respectively.Even when assessing only the biological net CO 2 exchange, by subtracting the nonbiological CO 2 flux when feasible, the degraded site emitted 32 g C m −2 more than the natural site.In this regard, the ecosystems' functioning could be divided into three different phases.The first phase corresponded to the autumn months, when the first rainfall events after the dry summer (i.e., rain pulses) activated the soil microbiota, triggering respiratory CO 2 emissions as previously measured at the same experimental sites (López-Ballesteros et al., 2016;Rey et al., 2017).During this phase, maximum net CO 2 emission was observed at the natural site, which exceeded the biological net CO 2 flux observed at the degraded site (Fig. 7a).In fact, as hypothesized, R eco was generally higher at the natural site during autumn months (Fig. 7b), probably due to the greater pool of organic carbon (Table 1) and the potential differences in microbial communities (Rey et al., 2017) between sites.The second phase comprised the growing period when plants photosynthesized and also respired along with microorganisms under milder temperatures and better hydric conditions.During this phase, larger net CO 2 uptake was measured at the natural site, with 162 % more than at the degraded site (Table 4) due to the higher vegetation cover and more fertile soils (Table 1) of the natural site.Accordingly, GPP estimates were, on average, 9 times higher at the natural site (Fig. 7c) and moreover, these results were supported by the lower ET, WUE and EVI values obtained in the degraded site during winter and spring months over the entire study period (Figs. 5,6 and 7d).The third phase consisted of the dormancy period when water scarcity and high temperatures constrained biological activity.During this period, a neutral C balance was observed at the natural site, while extreme CO 2 release was measured at the degraded site, as expected (Figs. 4 and 6b), where ventilative fluxes were dominant.
In order to detect potential factors driving the observed differences in the C balances, we checked whether soil and meteorological variables differed between sites.Our results demonstrated that some factors typically influencing GPP, R eco , and hence net ecosystem CO 2 exchange, such as photosynthetic photon flux density (PPFD; Michaelis andMenten, 1913), precipitation (Berner et al., 2017;Jongen et al., 2011), vapor pressure deficit (VPD; Lasslop et al., 2010) and soil and air temperature (Lloyd and Taylor, 1994), did not differ between sites (Tables 5 and 6).Conversely, some differences were found for shallow soil volumetric water content (VWC 0.05 m ) during dry seasons (Table 6), when VWC 0.05 m was 2 times higher in AMO than in BB, but absolute differences were slight, from 0.028 to 0.037 m 3 m −3 Table 6.Results of the two-sided Wilcoxon summed rank test used to assess differences among soil variables measured at each experimental site over all periods, from May to September and from May to September during daytime, separately.Medians of the difference between the samples (Amoladeras minus Balsa Blanca) in standardized terms (Diff st ) and number of observations are detailed.Significant results (p-value < 0.05) are denoted with asterisks, and bold values represent those variables with Diff st between sites above 1.(Tables S1-S3).Hence, although the important influence of soil moisture in both GPP and R eco is known (Tang and Baldocchi, 2005), we believe that differences in VWC 0.05 m are not enough relevant to cause the differing ecosystem functioning observed over the drought period.Additionally, we think that this intersite difference in VWC 0.05 m could be instrumental or due to the spatial variability of VWC 0.05 m derived from the heterogeneity of soil morphological characteristics, since we only used one sensor at each site.Similarly, significant differences were not detected in several variables linked to subterranean ventilation, such as the friction velocity (u * ; Kowalski et al., 2008), wind speed (WS; Rey et al., 2012a), half-hourly maximum wind speed (WS max ) and net radiation, which has been positively correlated to ventilative CO 2 fluxes (López-Ballesteros et al., 2017), when using the analysis periods in which this process is supposed to be relevant, namely daytime hours during the dry seasons (Table 5).

Variables
However, although no turbulence and wind speed intersite differences were found, interconnectivity of soil pores and fractures is probably higher at the degraded site (Table 1) due to its higher gravel and rock fractions (Table 1), which could lead to an enhanced penetration of eddies within the vadose zone (Pérez-Priego et al., 2013).Apart from that, outstanding differences between sites were observed in subsoil CO 2 molar fractions measured at 0.05 and 1.50 m depths (CO 2,0.05 m and CO 2,1.50 m , respectively; Table 6).On one hand, CO 2,0.05 m was generally higher at the natural site given its lower degradation level, which probably promotes higher root respiration and microbial activity supported by higher vegetation density and soil fertility (Table 1), especially during spring (Fig. 8), as also pointed out by Oyonarte et al. (2012).On the other hand, CO 2,1.50 m values were acutely higher at the degraded site, by up to 1000 ppm compared to the natural site (Tables S1- ).Therefore, we suggest that CO 2,1.50 m is the main factor responsible for the intersite differences in net CO 2 fluxes over the dry season.In this regard, previous research has suggested two potential origins of this vadose zone CO 2 : geological degassing (Rey et al., 2012b) and/or subterranean translocation of CO 2 in both gaseous and aqueous phases (López-Ballesteros et al., 2017).However, not only does the amount of subsoil CO 2 matter but also how effective its transport is, since both determine the net CO 2 release from the vadose zone to the atmosphere.In this context, Oyonarte et al. (2012) found, in the same study area (Cabo de Gata-Níjar Natural Park), that soils with degradation symptoms, such as lower SOC, depleted biological activity, coarser texture and worse structure, showed higher soil CO 2 effluxes over the dry season.Additionally, soil CO 2 effluxes measured during summer months correlated positively with the fraction of rock outcrops, suggesting that deteriorated soil physical conditions actually enhanced vertical transfer of CO 2 -rich air from the subsoil to the atmosphere (Oyonarte et al., 2012).
In fact, correlation analysis between CO 2,1.50 m and net CO 2 exchange/atmospheric pressure (Table 7) showed a stronger relationship between these variables at the degraded site.In this sense, ecosystem degradation could provoke a greater exposure of subsoil CO 2 to the pressure effect, as described by Sánchez-Cañete et al. ( 2013), probably due to a higher fraction of bare soil, coarser structure, differing porosity type and/or thinner soil depth (Table 1).
Regarding EVI data, we found a discrepancy between GPP estimates and EVI values since, contrary to what is observed in EVI results, we observed that GPP was always higher at the natural site compared to the degraded site.We think that this is due to the different spatial scales defining each measurement.MODIS pixels have an area of ∼ 6.25 ha, while the eddy covariance footprint corresponds to a smaller area of ∼ 1 ha.Therefore, there is an EVI uncertainty that stems from the influence of other surface elements apart from vegetation, such as bare soil or outcrops within the pixel, which is our case.In fact, previous studies confirm the discrepancy between MODIS-and EC-derived GPP estimates, especially on sparse vegetation areas with low productivity (Gilabert et al., 2015).However, EVI data have allowed us to complement our findings based on CO 2 fluxes, especially when EC data losses occurred.For instance, the declining trend observed from 2009-2010 to the end of the study period, for both sites, was not noticeable from EC data alone (Fig. 6).This long-term decrease in EVI may be related to a gradual drying following the wettest year (2009)(2010), when extraordinarily high precipitation occurred (twice the mean annual precipitation for both sites over the study period).This EVI pattern also suggests a pulse-like behavior of ecosystem vegetation over the interannual timescale.
Moreover, in addition to demonstrating that degradation can influence the biological activity of ecosystems' vegetation, EVI results also showed that the degradation level can modulate how an ecosystem responds to a short-term disturbance.A clear example is the dry spell experienced in 2013-2014, when a reduction in EVI was measured during the growing season at both sites, i.e., 35 and 28 % at the natural and degraded sites, respectively.However, a year later (2014)(2015), EVI values below the 6-year average were observed only at the degraded site (21 % lower; Fig. 6a) and major intersite differences were found for GPP, R eco and WUE during autumn and winter months (Fig. 7b-d).Accordingly, the natural site seemed to be more resilient than the degraded site against the short-term disturbance, since the effect of drought persisted in AMO even during the following year, while BB recovered to a pre-perturbation state within the same period (Fig. 6a).As a result, ecosystem resilience (Holling, 1973) was weakened by long-term disturbances such as land degradation, making degraded ecosystems more vulnerable to climate extremes (Reichstein et al., 2013).In this sense, mitigation policies to confront land degradation should be focused on prevention programs since ecosystem restoration does not recover complete ecosystem functionality (Lal, 2001;Moreno-Mateos et al., 2017).Moreover, even after several decades, relict degradation legacies can remain (Alados et al., 2011).
The present study can be seen as a step forward to better understanding the effect of land degradation on the intricate network of multiscale processes, factors and structures that define ecosystems' biological productivity and ultimately control their C balances.Despite some limitations, such as long-term data gaps, this research demonstrates that continuous ecosystem-scale EC observations remain crucial to comprehending how climate and land use change can modify the C sequestration capacity of ecosystems.In fact, annual average net CO 2 exchanges of 196 ± 40 (C release) and −23 ± 20 g C m −2 yr −1 (C fixation) for the degraded and natural (i.e., site of reference) sites were measured, respectively.Additionally, larger net CO 2 uptake over longer periods was observed at the natural site, with 162 % higher C compared to the degraded site, whereas a much greater net CO 2 release was measured at the degraded site during drought periods.Accordingly, the estimates of gross primary production, ecosystem respiration and water use efficiency were, on average, 9, 2 and 10 times higher at the natural site.Future research should be based on the continuity of long-term monitoring stations, such as eddy covariance stations, in order to calibrate and validate satellite data, reduce uncertainties in the relationships between ecosystem productivity, land degradation and climate change and, finally, to improve the predictive ability of current terrestrial C models.
Data availability.The eddy covariance data are available in the European Fluxes Database Cluster (http://www.europe-fluxdata.eu)where experimental sites have the codes Es-Amo and Es-Agu.Other data can be obtained by contacting the corresponding author.
Author contributions.FD, CO, AK and PSO designed the experiment.ALB, PSO, EPSC and MRM calibrated the sensors, collected the data and maintained the field instrumentation.EPSC designed subsoil data acquisition system and MRM processed subsoil data.ALB processed the eddy covariance data, made the figures and tables and wrote the manuscript.All authors reviewed the manuscript.
Competing interests.The authors declare that they have no conflict of interest.Special issue statement.This article is part of the special issue "Ecosystem processes and functioning across current and future dryness gradients in arid and semi-arid lands".It is not associated with a conference.

Figure 1 .
Figure 1.Location (a) and photographs (b) of the experimental sites.Green area represents the Cabo de Gata-Níjar Natural Park (Almería, Spain).

Figure 2 .
Figure 2. Daily averages of air temperature (T air ) and precipitation in (a) Amoladeras and (b) Balsa Blanca.

Figure 3 .
Figure3.Cumulative annual net CO 2 exchange over the 3 hydrological years without long periods of missing data at both experimental sites, Amoladeras (red lines) and Balsa Blanca (black lines).Negative values denote net carbon uptake, while positive values denote net carbon release.Thin lines indicate uncertainty derived from the gap-filling procedure.

Figure 4 .
Figure 4. Cumulative seasonal net CO 2 exchange over the study period in both experimental sites.Negative values denote net carbon uptake, while positive values denote net carbon release.In the case of Balsa Blanca, missing bars correspond to long-term data losses (> 50 % data).

Figure 5 .
Figure5.Cumulative seasonal evapotranspiration fluxes (ET) over the study period at both experimental sites.In the case of Balsa Blanca, missing bars correspond to long-term data losses (> 50 % data).Error bars denote uncertainty derived from the gap-filling procedure.

Figure 6 .
Figure 6.Time series of (a) enhanced vegetation index (EVI), (b) daily net CO 2 exchange and (c) daily evapotranspiration fluxes measured in Amoladeras (red lines and dots) and Balsa Blanca (black lines and dots) over 6 hydrological years (2009-2015).Long-term data losses correspond to periods of several months when ET and CO 2 fluxes are absent.

Figure 8 .
Figure 8. Box-and-whisker plots of CO 2 molar fractions measured at (a) 0.05 m and (b) 1.50 m belowground in Amoladeras (red boxes) and Balsa Blanca (grey boxes) from January 2014 to August 2015.The box extends from the first (Q 1 ) to the third quartiles (Q 3 ) and the central line represents the median (50 % percentile).Dots represent outliers; the upper whisker is located at the smaller of the maximum value and Q 3 + 1.5 IQR (interquartile range), and the lower whisker is located at the larger of the minimum value and Q 1 − 1.5 IQR.

Table 2 .
Variables measured, sensors used and their installation height in Amoladeras and Balsa Blanca experimental sites.

Table 3 .
Linear regression results between half-hourly net CO 2 fluxes of maximum quality (QC flag = 0) and friction velocity (u * ) used to model subterranean ventilation.

Table 4 .
Number of days with daily net CO 2 uptake and the related total C absorbed for every hydrological year and every field site of the study.Asterisks denote those years with abundant data losses (∼ 30 % data).

Table 5 .
Results of the two-sided Wilcoxon summed rank test used to assess differences among meteorological variables measured at each experimental site over all periods, from May to September and from May to September during daytime, separately.Medians of the difference between the samples (Amoladeras minus Balsa Blanca) in standardized terms (Diff st ) and number of observations are detailed.Significant results (p-value < 0.05) are denoted with asterisks, and bold values represent those variables with Diff st between sites above 1.

Table 7 .
Spearman correlation coefficients (r s ) for every paired simple correlation among maximum quality net CO 2 exchange fluxes (µmol m −2 s −1 ), absolute and differential pressure (hPa) at 6, 12, 24 and 72 h time step and absolute and differential CO 2 molar fraction measured at 1.50 m below ground (ppm) at the same time steps.Bold values represent the highest correlation coefficients, while shaded ones denotes nonsignificant relationships (p-values > 0.05).