Agriculture, Ecosystems and Environment 199 (2014) 114–123

Soil organic carbon evolution after land abandonment along aprecipitation gradient in southern Spain

Miguel A. Gabarrón-Galeote *, Sylvain Trigalet, Bas van WesemaelGeorges Lemaître Centre for Earth and Climate Research, Earth and Life Institute, Université Catholique de Louvain, B-1348 Louvain-la-Neuve, Belgium

A R T I C L E I N F O

Article history:Received 3 June 2014Received in revised form 27 August 2014Accepted 28 August 2014Available online xxx

Keywords:Soil organic carbonLand abandonmentPrecipitation gradientMediterranean

A B S T R A C T

Land abandonment is the dominant form of land use change in the Mediterranean over the last decades,and determines the soil organic carbon (SOC) evolution during the secondary succession followingabandonment. However, the rate of succession strongly depends on climatic conditions and the extent towhich these determine the SOC dynamics is largely unknown. The aim of this study is determining thesedynamics along a precipitation gradient (1085-650-350 mm yr�1) on noncalcareous rocks in southernSpain. Fields abandoned in different periods, as verified on aerial photographs taken in 1956, 1977, 1984,1998, 2001, 2004 and 2009, were selected using a chronosequence approach. SOC was determined using aspectrometer, vegetation was described, and NDVI calculated from Landsat images. SOC and NDVIevolution were analysed subsequently. In the two wettest sites SOC increased after land abandonmentuntil it approached a plateau. Mean accumulation rates were 0.11 kg C m�2 y�1 for the wettest and0.06 kg C m�2 y�1 for the intermediate site. These sites reached the long-term state, similar to the stocksin (semi) natural fields, in c.a. 10 years (wettest) and c.a. 35 years (intermediate). SOC and NDVI followedparallel trends, so SOC stocks were mainly driven by inputs from vegetation. At the dry end of thegradient, where NDVI’s (<0.1) were very low, the SOC stocks did not respond to changes in NDVI for the50 year period.

ã 2014 Elsevier B.V. All rights reserved.

Contents lists available at ScienceDirect

Agriculture, Ecosystems and Environment

journa l homepage : www.e l sev ier .com/ loca te /agee

1. Introduction.

Soils contain approximately 1500 Pg of organic carbon, twicethe amount of carbon stored in the atmosphere and roughly threetimes the C stored in terrestrial vegetation (Raich and Potter, 1995;Lal, 2004). The amount of C stored in a soil is the result of thebalance between inputs and outputs. Inputs are determined byvegetation cover, since they are mainly plant-derived residuesdeposited aboveground as well as belowground. Outputs are Closses resulting from mineralization, leaching of dissolved organicC and erosion (La Mantia et al., 2013). The critical role of vegetationdetermining changes in soil organic carbon (SOC) stocks upon landuse change (LUC) has been widely reported (Post and Kwon, 2000;Guo and Gifford, 2002; Poeplau et al., 2011). Nearly all studiescompare SOC stocks before and after LUC. The conversion ofnatural vegetation to cropland indeed meets the characteristics ofa drastic change, but in the case of cropland abandonment, a

* Corresponding author. Tel.: +32 10472867.E-mail addresses: miguel.gabarron@uclouvain.be, miguegaba@hotmail.com

(M.A. Gabarrón-Galeote), sylvain.trigalet@uclouvain.be (S. Trigalet),bas.vanwesemael@uclouvain.be (B.v. Wesemael).

http://dx.doi.org/10.1016/j.agee.2014.08.0270167-8809/ã 2014 Elsevier B.V. All rights reserved.

secondary succession occurs with colonization by grasses in thefirst stage and woody plants in later stages. Some studies haveaddressed SOC changes along secondary successions (Knops andTillman, 2000; Rhoades et al., 2000; Zhang et al., 2007; La Mantiaet al., 2013; Novara et al., 2013). In general, a net SOC stock gainafter cropland abandonment is reported due to an increase in theinput of organic matter, above as well belowground, and anincrease in resistance to decomposition of the litter produced (Guoand Gifford, 2002). The estimated time to recovery of the originalSOC stock ranges from 5–7 years in the Dominican Republic(Templer et al., 2005), to 230 years in a sandy plain of Minnesota(Knops and Tillman, 2000). SOC dynamics differ according to theprevious land use i.e., cropland or grassland (Post and Kwon, 2000;Guo and Gifford, 2002; Templer et al., 2005). In particular, there isno agreement on the response of SOC after forest establishment ongrasslands. Some studies have reported SOC increments (Hibbardet al., 2001; McCulley et al., 2004), whereas the opposite has alsobeen documented (Guo and Gifford, 2002). Alberti et al. (2011)mentioned that precipitation is the key factor determining the typeof response after conversion of grassland to forest, indicating athreshold around 900 mm, above which a net decrease could beexpected. According to Poeplau et al. (2011), there is an initialdecrease of SOC followed by an increment, resulting in a final net

M.A. Gabarrón-Galeote et al. / Agriculture, Ecosystems and Environment 199 (2014) 114–123 115

SOC gain. Precipitation is often mentioned as one of the mainfactors affecting SOC changes after LUC (Guo and Gifford, 2002;Jackson et al., 2002; Alberti et al., 2011; Poeplau et al., 2011). Thebalance between net primary production and decomposition is thenet biome production that determines SOC sequestration ordepletion (Ciais et al., 2010). On the one hand, precipitation affectsthe net primary production, and thus the organic matter inputs tothe soil (Guzman and Al-Kaisi, 2010). This is even more evident insemiarid environments, where water is the limiting factor. On theother hand, precipitation affects soil moisture, one of the mainfactors modulating the decomposition rate of organic matter insoil, together with soil temperature (Powlson et al., 2007).

During the last decades, abandonment of agricultural lands hasbeen one of the most important processes in rural areas in manyparts of Western Europe, especially in the Mediterranean region(Nunes et al., 2010). Between 1961 and 2011, 24.5% of the acreageunder annual and permanent crops in southern Europe has beenabandoned (127,450 km2) and 17.0% (35,200 km2) in Spain (FAO,2013). This phenomenon has been attributed to a complex of socio-economic factors such as globalization as well as specific policies ofthe European Union and the local governments that reduced thesubsidies for extensive farming. This resulted in an abandonmentof rainfed agriculture in the uplands and encouraged thedevelopment of irrigated agriculture and tourism along the coast(Onate and Peco, 2005; Lesschen et al., 2008). While secondarysuccession in temperate regions usually leads to the developmentof a forest., the final stage in Mediterranean environments dependson the availability of water. Grasses and shrubs mostly dominatethe later stages of the succession in semiarid lands (Bonet, 2004).Climate affects the species composition of plant communitiesalong a secondary succession and consequently is a key factor inSOC dynamic after land abandonment.

The precipitation gradient in southern Spain covers an areawith similar mean annual temperature extending from the strait ofGibraltar (mean annual precipitation (MAP) >1100 mm) to Cape of

Fig. 1. Precipitation gradient and l

Gata (MAP <200 mm). Within this gradient, changes have beendetected in the runoff generation mechanisms, in the role of soilsurface components determining hydrological response, in thewater use by vegetation and in the vegetation patterns bycomparing areas with the same lithology (Ruiz-Sinoga et al.,2010a,b; Ruiz-Sinoga et al., 2011a,b). The aim of this study is toquantify the effect of climate on SOC stock dynamics after landabandonment. We studied sites along the gradient described aboveto quantify SOC evolution after land abandonment on each site,related SOC evolution to precipitation and temperature anddetermined the role of vegetation influencing SOC evolution.

2. Material and methods

2.1. Study sites

The study was carried out in southern Spain along an East-to-West precipitation gradient (Fig. 1). The gradient ranges from thestrait of Gibraltar, with precipitation of more than 1100 mm y�1, tothe Cape of Gata with less than 200 mm y�1 (Ruiz-Sinoga et al.,2010a). Four study sites were selected: Gaucín (GAU), Almogía(ALM) and two sites near Gérgal (GER; Fig. 1 and Table 1). Thesespecific sites were selected because: (i) in spite of the differences inannual precipitation, the regime was similar; (ii) all the sites arelocated in the same biogeographical province, which ensuressimilarities in the secondary succession; (iii) the land use history issimilar and (iv) the sites are all on noncalcareous metamorphicmicaschists. The decisive factor determining the site selection wasthe different precipitation, which is the most important variableaffecting SOC in the mid latitudes at the continental to regionalscale (Adler et al., 2003; Minasny et al., 2013). Unfortunately, allabandoned fields on micaschists in GAU and ALM were originallyalmond and olive groves. As land use history explains a large part ofthe total SOC variability (Schulp and Veldkamp, 2008), wepreferred to restrict the study to abandoned cereal fields for all

ocalization of the study sites.

Table 1Main characteristics of each site. See Fig. 1 for the locations.

Gaucín (GAU) Almogía (ALM) Gérgal low (GERL) Gérgal high (GERH)

Precipitation (mm y�1) 1085 650 350 500Temperature (oC) 16.7 15.5 13 7ET0 (mm y�1) 882 820 708 630Altitude (m a.s.l) 150 400 1100 1900Lithology Marls, sandstones and unconsolidated clays Marls, sandstones and unconsolidated clays Micaschist MicaschistSlope (%) 10–20 10–20 25–35 15–35Soila Eutric Cambisols and Regosols Eutric Cambisols Regosols and Leptosols Regosols and LeptosolsSoil depth (cm) >50 >50 5–30 5–30

a Soils were classified according to the IUSS Working Group WRB (2006).

116 M.A. Gabarrón-Galeote et al. / Agriculture, Ecosystems and Environment 199 (2014) 114–123

sites, even if this meant that the fields in GAU and ALM had to befound in the neighboring lithological unit i.e., unconsolidatedmarls and clays with allochthonous blocks of sandstones.

The vegetation in GAU consists mainly of cereals and citrustrees, alternating with grasslands with shrub patches inabandoned fields. Forests are limited to remnant areas of limitedaccess, where undisturbed vegetation is dominated by Quercussuber, accompanied by Quercus faginea subs. broteroi in wet and

Fig. 2. Example of date of abandonment determination in GAU, b

shaded areas. It is a dense forest when it is in its optimum state,with the understory dominated by shrubs (Phyllirea angustifolia,Erica arborea, Pistacia lentiscus) and climbers. This site is located inthe Aljíbico biogeographical district (Rivas-Martínez et al., 1997),where net primary production of natural vegetation is 0.70 kg Cm�2 y�1. This value was obtained from the MOD17A3 product,downloaded from the Land Processes Distributed Active Archive

etween 1956 and 1977 (A), and between 1984 and 1998 (B).

Table 2Area and sample points per category in each site.

Date of abandonment GAU ALM GERL GERH

Area (Ha) Points Area (Ha) Points Area (Ha) Points Area (Ha) Points

Natural 3.9 3 2.1 5 2.0 8 1.8 856–77 14.5 10 5.4 14 2.6 12 2.0 977–84 13.3 9 3.2 8 2.6 13 1.0 584–98 21.8 14 5.8 10 0.6 3 1.9 398–01 4.1 3 4.4 8 0.8 4 – –

01–04 11.9 7 2.0 4 1.7 8 1.9 9Cropland 2.1 4 6.2 15 1.2 4 1.9 10Total 71.1 50 29.1 64 11.9 52 10.4 44

M.A. Gabarrón-Galeote et al. / Agriculture, Ecosystems and Environment 199 (2014) 114–123 117

Center of the U.S. Geological Survey (2013). It corresponds to themean between 2000 and 2010, with a 1 km resolution.

The vegetation in (semi) natural sites of ALM is a dense forestdominated by Quercus rotundifolia. It has a dense understory, withspecies such as Quercus coccifera, Chamaerops humilis and P.lentiscus. In agricultural systems, cereals and olives dominate,alternating with grasslands and shrublands in abandoned ordisturbed fields. These shrublands are dominated by Cistus spp.,Genista umbellata and Ulex parviflorus. This site belongs to theMalacitano biogeographical district (Rivas-Martínez et al., 1997),where net primary production of natural vegetation is 0.46 kg Cm�2 y�1 (U.S. Geological Survey, 2013).

The (semi) natural vegetation of GERL is a forest dominated byQ. rotundifolia and characterized by low biodiversity. Someimportant species are Retama sphaerocarpa, Cytisus grandiflorus,G. umbellata and Cytisus fontanesii. This vegetation is scarce now inthe area, as large parts were afforested in the mid-twentiethcentury with Pinus halepensis. Vegetation in disturbed sites iscommonly a shrubland dominated by R. sphaerocarpa and G.umbellata. This site is located in the Almeriense biogeographicaldistrict (Rivas-Martínez et al.,1997), where net primary productionof natural vegetation is 0.27 kg C m�2 y�1 (U.S. Geological Survey,2013).

The (semi) natural vegetation in GERH is a shrubland dominatedby Juniperus communis, Juniperus sabina, Genista versicolor andCytisus galianoi. This community is formed by creeping or pillow-shaped shrubs with a high coverage (>80%). Vegetation inabandoned croplands is sparse and composed of isolatedindividuals or stands of G. versicolor and C. galianoi. This site islocated in the Filábrico district (Rivas-Martínez et al., 1997), wherenet primary production of natural vegetation is 0.39 kg C m�2 y�1

(U.S. Geological Survey, 2013).

2.2. Sampling strategy

The study was carried out using a chronosequence approach.This “space for time” method has been widely used in SOCdynamics studies given the usually slow response of this propertyto environmental changes (West and Post, 2002; Cerri et al., 2003;

Table 3Matrix of accuracy of the land use classification of the sampling points.

ClassificationGAU ALM

Real use Crop Nat Crop

Current Crop 12 0 26

Nat 8 104 0

1956 Crop 73 0 98

Nat 27 24 8

Crop, croplands; Nat, (semi) natural.

Salomé et al., 2010; De Baets et al., 2013). In each of the four sites,the optimum sampling density was estimated using SOC data froma previous sampling carried out in 2009. As these data followed anormal distribution Eq. (1) could be applied to calculate theoptimum number of samples:

ns ¼ u1�a=2 � Sr

!(1)

where ns is the optimum number of samples to take in a given area,u1 �a/2 is the (1 � a/2) quantile of the standard normal distribu-tion, S� is a prior estimate of the standard deviation of SOC in thearea and r is the absolute error limit. In our case the absolute errorlimit was 0.2 kg C m�2 and u1 �a/2 was 2, corresponding to aconfidence interval of 95% (a = 0.05). Since the area of fields in theprevious sampling campaign (A) was known, the same ratio ns/Awas applied to the new sites. The previous sampling was restrictedto natural areas, where the variability in soil properties is expectedto be larger than in croplands, so the ns estimated is expected to bevalid for the abandoned fields. Within each site the age ofindividual fields was reconstructed. Abandonment is defined as thecessation of ploughing and sowing, although other uses andactivities such as grazing may continue (Bonet, 2004). The selectedabandoned areas were previously cultivated with cereals. Aerialphotographs, taken in 1956, 1977, 1984, 1998, 2001, 2004 and2009 were compared to determine the date of abandonment(Fig. 2). The time series of the available recent photographsallowed a more precise dating of the abandonment from 2001 to2009. This comparison was done on differences in texture observedin the photographs and on the appearance of patches of naturalvegetation that contrasted with the homogeneity of croplands. Wetested the accuracy of the selection method for two specific dates:(i) the land use observed on the photos of 1956 was compared to aland use map (Junta de Andalucía, 2009) and (ii) the land use onthe photos of 2009 was checked against the current land use onsite. All the points wrongly classified were dismissed. For eachcategory, the date of abandonment was considered to be the meanof the dates of the photos under the old and new land use. We didnot find fields abandoned between 2004 and 2009. In addition, tworeference categories were determined: cultivated with cereals and

GERL GERH

Nat Crop Nat Crop Nat

10 10 7 10 095 3 69 0 49

0 56 0 47 025 4 25 4 8

Table 4Mean � standard deviation of SOC (kg m�2) and NDVI for each site and abandonment category.

Category (date of abandonment) GAU ALM GERL GERH

SOC NDVI SOC NDVI SOC NDVI SOC NDVI

(Semi) natural 4.11�0.80 0.45�0.04 4.19�0.55 0.24�0.02 2.14�0.86 0.25�0.04 2.76�0.56 0.14�0.021956–1977 4.33�1.34 0.37�0.07 3.28�0.79 0.25�0.02 1.67�1.33 0.17�0.09 2.34�0.83 0.10�0.021977–1984 4.29�1.00 0.30�0.04 2.39�0.53 0.21�0.02 0.95�0.31 0.17�0.03 2.20�0.90 0.09�0.021984–1998 4.38�0.94 0.32�0.05 1.93�0.49 0.19�0.03 0.87�0.48 0.19�0.02 2.00�0.28 0.08�0.021998–2001 5.12�1.42 0.30�0.03 2.17�0.43 0.18�0.03 2.03�0.53 0.18�0.02 – –

2001–2004 4.49�1.03 0.34�0.02 2.47�0.88 0.23�0.04 1.71�0.39 0.24�0.05 2.61�0.58 0.07�0.02Croplands 2.98�1.01 0.16�0.03 1.72�0.53 0.16�0.04 2.01�0.92 0.23�0.03 2.66�0.50 0.02�0.03

118 M.A. Gabarrón-Galeote et al. / Agriculture, Ecosystems and Environment 199 (2014) 114–123

under (semi) natural conditions both at least since 1956. Since therequired result is a parameter of the frequency distribution of thetarget variable (mean of SOC stocks), and the sampling area wasdivided according to the date of abandonment, a stratified randomsampling was applied (de Gruijter et al., 2006). Initially five fieldsper category and site were selected in GAU, ALM and the two sitesof GER together, but due to differences in lithology, impossibility ofaccess or recent land-use changes, some of these fields weredismissed and a lower number were finally sampled (seeSupplementary material 1). The random location of each samplingpoint was assigned with the ArcGis tool for sampling design. If apoint was impossible to sample a reserve point was used (seeSupplementary material 1). On each of the 210 resulting samplingpoints, two undisturbed samples were collected in Kopeckycylinders in order to calculate the bulk density (BD; Table 2).Moreover, a sample of c.a. 500 gr was collected at slices of 10 cmuntil a depth of 30 cm. When the soil was shallower than 30 cm, itwas sampled until the bedrock.

Supplementry material related to this article found, in theonline version, at http://dx.doi.org/10.1016/j.agee.2014.08.027.

2.3. Sample preparation

Samples for BD were oven-dried at 90 �C, weighed and BD wascalculated using the known volume. Samples for SOC were air-dried and coarse roots and stones were first removed manually.Then they were gently crushed and sieved at 2 mm to obtain thefine earth fraction.

2.4. Soil organic carbon measurements

In order to cope with the large number of samples a rapidalternative was chosen for the bulk of the samples and calibrated tothe classical drycombustionanalysisonasubset (seeSupplementarymaterial 2). Soil reflectance was measured using an ASD Fieldspec-Pro spectrometer (ASD Inc., Boulder, CO) in the visible and nearinfrared range (Vis–NIR:350–2500 nm). A portion of the fine fractionof the sample was placed in a small container and the surface wassmoothed. The scanning was performed directly on this surface witha contact probe containing an internal light source. Four replicates ofeach sample were scanned, mixing and smoothing between them. Atotal of 568 (GAU: 150; ALM: 187; GERH: 109, GERL: 122) sampleswere measured this way. The spectral matrix was smoothed using

Table 5Correlation matrix of the continuous variables.

SOC NDVI P T ET0

SOC 1NDVI 0.45 1P 0.61 0.75 1T 0.39 0.73 0.87 1ET0 0.47 0.71 0.90 0.99 1

the Savitzky–Golay algorithm and transformed by standard normalvariate (SNV) to reduce spectral variability (light scattering). Twospectra (one in GAU and one in ALM) were excluded from furtheranalysis as they showed very different spectral characteristics. Theremaining spectral dataset was divided into a training/cross-validation set of 50 samples in GAU and ALM, 30 in GERH and35 in GERL, fordeveloping the spectroscopic models, and a predictionset with the remaining samples. The training samples were selectedusing the Kennard–Stone algorithm. Samples selected for thetraining/cross-validation set were analysed for SOC content usinga VarioMax CN dry combustion Analyzer (Elementar GmbH,Germany). Samples containing inorganic C were treated using 6 MHCl containing 3% by weight FeCl2�4H2O to measure inorganic Cfollowing Sherrod et al. (2002).

Supplementry material related to this article found, in theonline version, at http://dx.doi.org/10.1016/j.agee.2014.08.027.

Spectral data was calibrated against SOC reference values usingpartial least square regression (PLSR). The models were developedwith the caret package (Kuhn et al., 2012) of the R statisticalsoftware (R Development Core Team, 2012). In order to selectmodel parameters, the models were cross-validated using10 random partitions having a probability of selection of 4:5(i.e., leave-one-group-out cross-validation). The best set ofparameters for each model was selected as the ones producing amodel having the smallest root-mean-square error (RMSE) ofcross-validation within one standard error of the minimum RMSEof cross-validation. Once the SOC concentrations were determined,SOC stocks (kg m�2) of each layer were calculated (Eq. (2)):

SOC ¼ Cf � BD � d � ð1 � RmÞ (2)

where Cf is the organic carbon concentration (%) of the fine soil, BDis the bulk density (g cm�3), d is the thickness of the layer (m), andRm is the rock fragment fraction by mass (%).

2.5. Vegetation cover and NDVI

Vegetation cover was estimated from aerial photographs andobservations in situ taken during the soil sampling campaign. NDVIwas calculated using the Landsat 5 images, with a resolution of30 m, downloaded from the U.S. Geological Survey (USGS) archive.Twelve images were selected of the years between 2008 and 2010,one in each season and free of clouds. With these 12 images a mean

Table 6ANCOVA table of the analysis relation SOC with precipitation and date ofabandonment.

D.F. S. SQ. M. SQ. F P-value

Precipitation (P) 1 147.35 147.35 152.70 <0.001Date of abandonment (D) 6 25.68 4.28 4.44 <0.001Lithology 1 0.55 0.55 0.57 0.45Interaction P � D 6 29.51 4.92 5.10 <0.001Residuals 195 188.17 0.96

D.F., degrees of freedom; S. SQ., sum of squares; M. SQ., mean square.

Fig. 3. SOC stocks evolution in each study site. A: GAU (1085 mm); B: ALM (650 mm); C: GERL (350 mm); D: GERH (500 mm). Error bars and shaded areas (for SOC0 and SOClt)indicate the standard deviation (see Table 2 for the number of sample points). SOC0: SOC in croplands; SOClt: SOC in (semi) natural points. Different lower case letters indicatesignificant differences. No lower case letters indicates a lack of significant differences for the entire site. The letters for the reference sites (natural vegetation and cereal fields)are next to the corresponding lines.

M.A. Gabarrón-Galeote et al. / Agriculture, Ecosystems and Environment 199 (2014) 114–123 119

NDVI was calculated and taken as representative of the actualvegetation in the area. The high resolution of the Landsat imagesallowed calculating the mean NDVI for each date of abandonment.Each point (n = 210) was assigned an NDVI value, corresponding tothe pixel where the point is located. In addition, each category ofabandonment was also assigned the mean NDVI of the corre-sponding area.

2.6. Climate variables

In order to take in account the variability within each site, asingle value of yearly precipitation (P), mean annual temperature(T), and annual reference evapotranspiration (ET0) was assigned toevery sampling point. These data were taken from the mapspublished by the Andalusian Government (Junta de Andalucía,2013), covering the period 1971–2000.

2.7. Data analysis

A correlation matrix was drawn up in order to test the relationbetween the continuous variables (SOC, NDVI, T, P and ET0) for eachsampling point (n = 210). Also using the value of each point, theeffect of precipitation and date of abandonment on SOC and NDVIfor the whole gradient was tested with an analysis of covariance(ANCOVA). Lithology was introduced as a categorical variable inboth analyses to take into account the SOC and NDVI variabilityassociated to it. In order to assess the effect of abandonment on SOCand NDVI within each site a one-way ANOVA was applied. The

adjustment of data to normal distribution was tested using theKolmogorov–Smirnov test, whereas the Bartlett’s test was per-formed to determine if the data accomplished the homoscedasticitycriteria. The significance level was set at 0.05, and all analyses wereperformed using R software (R Development Core Team, 2012).

3. Results and discussion

3.1. Accuracy of point selection

In order to test the accuracy of the land use classification, thepredicted land use from the aerial photographs was compared tothe land use map of 1956 (Junta de Andalucía, 2009) and in situobservations. In 1956 all the abandoned fields were supposed to becroplands (Table 3). The level of accuracy was generally high, withan overall 80% of success. As a result of the similarity betweencropland and grassland in GAU some of the fields weremisclassified. The current land use for the sample points is either(semi) natural vegetation or cropland. The overall success of theclassification was 90% and again some of the croplands in GAUwere misclassified. In this site some owners were contacted inorder to verify the date of abandonment of some of the points.

3.2. Soil organic carbon evolution

SOC stocks in croplands exceeded 2 kg m�2 in three out of foursites, (Table 4). However, in areas covered by (semi) naturalvegetation, SOC stocks were larger than 4 kg m�2 only in GAU and

Fig. 4. Evolution of the soil cover i.e., grass, shrub and bare soil after abandonment. A: GAU; B: ALM; C: GERL; D: GERH. LT: (semi) natural vegetation.

Table 7ANCOVA table of the relation NDVI with precipitation and date of abandonment.

D.F. S. SQ. M. SQ. F P-value

Precipitation (P) 1 1.14 1.14 379.38 <0.001

Date of abandonment (D) 6 0.23 0.04 12.91 <0.001Lithology 1 0.00 0.00 0.05 0.81Interaction P � D 6 0.06 0.01 3.05 0.007Residuals 195 0.59 0.003

D.F., degrees of freedom; S. SQ., sum of squares; M. SQ., mean square.

120 M.A. Gabarrón-Galeote et al. / Agriculture, Ecosystems and Environment 199 (2014) 114–123

ALM. In GERH and GERL there was little difference between SOC0

(under cropland) and SOClt (under (semi) natural vegetation) sincethe latter was lower than 3 kg m�2.

Given that the correlation coefficients between NDVI, P, T andET0 are larger than 0.7, only precipitation, in addition to date ofabandonment and lithology, was used as a factor in the ANCOVAwith SOC as dependent variable (Table 5).

Precipitation and date of abandonment significantly affectedSOC (p < 0.001, R2 = 0.48). The same is true for the interaction ofboth factors (Table 6). Thus, SOC significantly changed after landabandonment, but in different ways along the precipitationgradient. Lithology did not affect SOC stocks (p = 0.57) SOC showeda large variability within each site and category, especially in GAU.Due to the large variability there were no significant differences ofSOC along the chronosequence in GAU (Fig. 3a). However, SOCstock in the fields abandoned between 2001 and 2004 was alreadyhigher than the one measured in (semi) natural areas, and it was1.5 kg m�2 higher than in croplands. SOC was significantly affectedby abandonment only in ALM (Fig. 3b). The Tukey test split thelevels of the factor in two groups. SOC remained more or less stableuntil the category 1977–1984 and SOC stock was significantlydifferent in fields abandoned earlier and later than 1977, so SOCaccumulation was only significant after c. 35 years of abandon-ment. Thus, in GAU and ALM SOC accumulated until it reached aplateau corresponding to the long-term state (Post and Kwon,2000; Six et al., 2002). Most studies analysing SOC changes afterconversion from croplands to forest reported SOC increments (Post

and Kwon, 2000; Guo and Gifford, 2002; Stevens and vanWesemael, 2008; Boix-Fayos et al., 2009; Poeplau et al., 2011).Some studies have reported a similar SOC response along asecondary succession (Knops and Tilman, 2000; Zhang et al.,2007). Novara et al. (2013) fitted a linear model although in thiscase no reference of long term vegetation was included.

Although there were no significant differences of SOC stockbetween the categories of abandonment in GER (Fig. 3c and d), SOCdecreased after land abandonment to reach the minimum level22 years after abandonment, followed by a recovery. Despite thestrong differences in precipitation, the SOC stocks in croplands ofthree out of four sites were rather similar. In GERL and GERH, theSOC stocks in cropland are similar to the 2.5 kg m�2 reported forother semi-arid environments in south east Spain (Almagro et al.,2010; Albaladejo et al., 2013).

The accumulation rates based on the difference between SOC0

and SOClt and the time to recover this difference (10 and 35 years),

Fig. 5. NDVI evolution in each study site. A: GAU (1085 mm); B: ALM (650 mm); C: GERL (350 mm); D: GERH (500 mm). Error bars and shaded areas (for NDVI0 and NDVIlt)indicate the standard deviation (see Table 2 for the number of sample points). NDVI0: NDVI in croplands; NDVIlt: NDVI in (semi) natural vegetation. No lower case lettersindicates a lack of significant differences for the entire site. The letters for the reference sites (natural vegetation and cropland) are next to the corresponding lines.

M.A. Gabarrón-Galeote et al. / Agriculture, Ecosystems and Environment 199 (2014) 114–123 121

were 0.11 kg C m�2 y�1 for GAU and 0.06 kg C m�2 y�1 in ALM(Fig. 3) In GERL final SOC gain was almost inexistent so it can bededuced that the accumulation rate was lower than in GAU andALM. Thus, SOC accumulation rates in the three sites with similarmean annual temperature were proportional to precipitation.These values are in the range of those found in the literature. In acold climate with precipitation of 550–560 mm y�1, Zhang et al.(2007) estimated 40 years as the necessary time to recover SOCstocks after land abandonment. In similar climate conditions but insandy soils Knops and Tilman (2000) predicted that it wouldrequire 230 years. In tropical conditions, Rhoades et al. (2000) andTempler et al. (2005) documented a time of 20 and 5–7 yearsrespectively. In semiarid Mediterranean conditions, Novara et al.(2013) reported SOC increase after 45 years of abandonment. Inaddition to precipitation, one supplementary explanation for thehigh recovery rate in GAU could be the intensive grazing, which cantrigger a “pumping action” due to the growth of new plants afterthe roots die-off (Guo and Gifford, 2002).

3.3. Vegetation and NDVI

The recently abandoned areas (1998–2009) in GAU werecolonized by grasses typical of grazed areas. The height of thegrass seldom exceeded 15 cm, and the cover was more than 90%.After some years of abandonment shrubs started to grow inpatches, mostly thorny shrubs, since these are resistant to grazing.In fields abandoned between 1977 and 1998, the percentagecovered by shrubs was around 20%, and in fields abandonedbetween 1956 and 1977 this percentage was almost 90%.

Vegetation at this stage had a similar cover as the natural areas;although species diversity remained lower (Fig. 4 ).

In the recently abandoned areas (1998–2009) in ALM, first ashrubland of low coverage developed. Moreover, seasonal grassesappeared during the wet season and their residues remained onthe soil during summer. These two formations covered almost100% of the abandoned fields during the first 10 years. Subse-quently, some shrub species started to appear. The shrub coverincreased progressively to reach almost 100% in the (semi) naturalvegetation (Fig. 4).

In GERL, the first stage after land abandonment was a shrublandthat progressively changed into a community typical for aban-doned fields dominated by G. umbellata after c. 10 years ofabandonment. The coverage of this community increased to 50%around 30 years after abandonment. Only after 45 years ofabandonment (1956–1977) shrublands dominated by Q. coccifera,with the same accompanying species, were developed. In the areaswith lower precipitation, Stipa tenacissima was the dominantspecies. Approximately 5% of the surface remained bare (Fig. 4).

In GERH, grasses colonized the fields after land abandonment.The grasses were characterized by low to medium coverage (20–40%). Colonization by shrubs was quicker in this site and stableshrub coverage, around 40%, was reached after c. 20 years,although this site was also the one with the largest bare soil cover(c. 25%). Subsequently the coverage did not change but the speciescomposition did (Fig. 4). On shallow soils the main species was G.versicolor and if the soils were more developed the dominatingspecies was C. galianoi.

NDVI depended on precipitation and date of abandonment(p < 0.001, R2 = 0.69; Table 7). According to some models (e.g., CASA

122 M.A. Gabarrón-Galeote et al. / Agriculture, Ecosystems and Environment 199 (2014) 114–123

(Field et al., 1995)) net primary production depends directly on theNDVI, so NDVI is a good indicator of primary and ecologicalproductivity, and hence of carbon input into the soil (Minasny et al.,2013). As for SOC, lithology did not affect NDVI (p = 0.81).

The significant effect of the interaction of precipitation and dateof abandonment means that the effect of abandonment on NDVIdiffered between sites. NDVI increased significantly after landabandonment in GAU and ALM, but whereas there was a steepincrease in the first 10 years after abandonment in GAU (Fig. 5a),NDVI only increased significantly after c. 35 years after abandon-ment in ALM (Fig. 5b). In GERL, although there was a decrease ofNDVI after abandonment, NDVI0 and NDVIlt were not significantlydifferent (Fig. 5c). Lastly, in GERH, NDVI increased at every step inthe chronosequence, although the values of this property werelower than in the other three sites (Fig. 5d).

3.4. Relation between SOC, NDVI and precipitation

NDVI was significantly related to SOC when we performed thecorrelation analysis with the NDVI values assigned to eachsampling point (p < 0.001, R2 = 0.2; Table 5). In fact, NDVI hasalready been successfully used to predict SOC concentration inother contexts (Bou Kheir et al., 2010; Burnham and Sletten, 2010;Kunkel et al., 2011). Both SOC and NDVI depended on precipitation(Tables 6 and 7). Precipitation has been identified as the mainfactor determining SOC stock change during the later stage(>30 years) following cropland conversion. It affects the organicmatter input, which is the major factor influencing the rate of SOCaccumulation (Deng et al., 2014). Bui et al. (2009) established thatclimate variables play a crucial role on the topsoil carbondistribution, and that the association of climate and SOC isconducted through plant primary productivity. Looking at theresponse of SOC to vegetation recovery and hence NDVI increase, itappears that SOC responds once NDVI increases clearly (Figs. 3 and5). Thus, the SOC response is the fastest in GAU where it thenremains stable as does the NDVI. In ALM, SOC response is muchslower and only increases after c. 30 years as NDVI slowly increasesfrom a minimum at 14 years. In GAU and ALM a dense grasslandcovered the soil surface just after the abandonment of the field,avoiding SOC losses by erosion and providing a source of carboninput (Fig. 4). Indeed, the conversion from cropland to grassland ischaracterized by a high rate of SOC recovery during the first yearsafter abandonment (Poeplau et al., 2011). In the driest sites, SOCdoes not respond to land abandonment as either NDVI remainsmore or less constant (GERL) or increases very slowly starting froma very low level (GERH). This increase of NDVI was not enough topromote SOC enrichment. Garten and Hanson (2006) came to asimilar conclusion studying SOC stocks along an elevationgradient. In GERL and GERH the vegetation recovery was slowerthan in the wetter sites and a significant percentage of soilremained bare after land abandonment (Fig. 4), causing a slowrecovery of NDVI and SOC and even a decrease during the firstyears (Zhang et al., 2010). These slow responses in the dry sites areconfirmed by Lesschen et al. (2008) who estimated that recovery ofvegetation and change in soil properties after land abandonmentare slow and take at least 40 years in such a semi-aridenvironment.

4. Conclusions

1. Soil organic carbon only accumulated after land abandonment inthe two wettest sites (precipitation >650 mm). SOC started toaccumulate after abandonment until SOC stock approached aplateau, corresponding to the one in (semi) natural vegetation.In the two driest sites (precipitation <450 mm) there was noSOC gain.

2. When comparing the three sites with similar temperature,precipitation was a key factor determining the accumulationrates of SOC after abandonment and the promptness of the SOCresponse: the higher the annual precipitation, the faster the SOCrecovery.

3. An increase in precipitation resulted in a higher NDVI and hencea faster SOC accumulation rate. The similarity of SOC and NDVIevolution suggests that SOC accumulation after abandonmentwas mainly an input driven process. However, at very low NDVI’s(<0.1) the SOC stocks did not respond to changes in NDVI for the50 year period.

Acknowledgments

The first author is a beneficiary of an FSR Incoming Post-doctoral Fellowship of the Académie universitaire ‘Louvain’, co-funded by the Marie Curie Actions of the European Commission.This research was funded by the Belgian Science Policy Office(belspo) in the framework of the Inter University Attraction Poleproject (P7/24): SOGLO–the soil system under global change.Finally, we thank the Physical Geography group of the University ofMalaga for the logistic support.

References

Adler, R.F., Huffman, G.J., Chang, A., Ferraro, R., Xie, P.-P., Janowiak, J., Rudolf, B.,Schneider, U., Curtis, S., Bolvin, D., 2003. The version-2 global precipitationclimatology project (GPCP) monthly precipitation analysis (1979–present). J.Hydrometeorol. 4, 1147–1167.

Albaladejo, J., Ortiz, R., Garcia-Franco, N., Ruiz Navarro, A., Almagro, M., GarciaPintado, J., Martínez-Mena, M., 2013. Land use and climate change impacts onsoil organic carbon stocks in semi-arid Spain. J. Soils Sediments 13, 265–277.

Alberti, G., Leronni, V., Piazzi, M., Petrella, F., Mairota, P., Peressotti, A., Piussi, P.,Valentini, R., Gristina, L., La Mantia, T., Novara, A., Rühl, J., 2011. Impact of woodyencroachment on soil organic carbon and nitrogen in abandoned agriculturallands along a rainfall gradient in Italy. Reg. Environ. Change 11, 917–924.

Almagro, M., López, J., Boix-Fayos, C., Albaladejo, J., Martínez-Mena, M., 2010.Belowground carbon allocation patterns in a dry Mediterranean ecosystem: acomparison of two models. Soil Biol. Biochem. 42, 1549–1557.

Boix-Fayos, C., de Vente, J., Albaladejo, J., Martinez-Mena, M., 2009. Soil carbonerosion and stock as affected by land use changes at the catchment scale inMediterranean ecosystems. Agr. Ecosyst. Environ. 133, 75–85.

Bou Kheir, R., Greve, M.H., Bøcher, P.K., Greve, M.B., Larsen, R., McCloy, K., 2010.Predictive mapping of soil organic carbon in wet cultivated lands usingclassification-tree based models: the case study of Denmark. J. Environ. Manage.91, 1150–1160.

Bonet, B., 2004. Secondary succession of semi-arid Mediterranean old-fields insouth-eastern Spain: insights for conservation and restoration of degradedlands. J. Arid. Environ. 56, 213–233.

Bui, E., Henderson, B., Viergever, K., 2009. Using knowledge discovery with datamining from the Australian Soil Resource Information System database toinform soil carbon mapping in Australia. Global Biogeochem.Cycles 23.

Burnham, H.J., Sletten, R.S., 2010. Spatial distribution of soil organic carbon innorthwest Greenland and underestimates of high Arctic carbon stores. GlobalBiogeochem. Cycles 24.

Cerri, C.E.P., Coleman, K.W., Jenkinson, D.S., Bernoux, M., Victoria, R., Cerri, C.C.,2003. Modeling soil carbon from forest and pasture ecosystems of Amazon,Brazil. Soil Sci. Soc. Am. J. 67, 1879–1887.

Ciais, P., Wattenbach, M., Vuichard, N., Smith, P., Piao, S.L., Don, A., Luyssaert, S.,Janssens, I.A., Bondeau, A., Dechow, R., Leip, A., Smith, P.C., Beer, C., Van der Werf,G.R., Gervois, S., Van Oost, K., Tomelleri, E., Freibauer, A., Schulze, E.D., 2010. TheEuropean carbon balance part 2: croplands. Global Change Biol. 16, 1409–1428.

De Baets, S., Meersmans, J., Vanacker, V., Quine, T.A., Van Oost, K., 2013. Spatialvariability and change in soil organic carbon stocks in response to recoveryfollowing land abandonment and erosion in mountainous drylands. Soil UseManage. 29, 65–76.

de Gruijter, J., Brus, D., Bierkens, M., Knotters, M., 2006. Sampling for NaturalResource Monitoring. Springer-Verlag Berlin Heidelberg, Netherlands.

Deng, L., Liu, G-b., Shangguan, Z-p., 2014. Land-use conversion and changing soilcarbon stocks in China’s ‘Grain-for-Green’ Program: a synthesis. Global ChangeBiol doi:http://dx.doi.org/10.1111/gcb.12508.

FAO, 2013. http://faostat3fao.org/faostat-gateway/go/to/download/R/RL/E(accessed 11.13.).

Field, C.B., Randerson, J.T., Malmström, C.M., 1995. Global net primary production:combining ecology and remote sensing. Remote Sens. Environ. 51, 74–88.

Garten Jr., C.T., Hanson, P.J., 2006. Measured forest soil C stocks and estimatedturnover times along an elevation gradient. Geoderma 136, 342–352.

M.A. Gabarrón-Galeote et al. / Agriculture, Ecosystems and Environment 199 (2014) 114–123 123

Guo, L.B., Gifford, R.M., 2002. Soil carbon stocks and land use change: a meta-analysis. Global Change Biol. 8, 345–360.

Guzman, J.G., Al-Kaisi, M.M., 2010. Soil carbon dynamics and carbon budget ofnewly reconstructed tall-grass prairies in south central Iowa. J. Environ. Qual.39, 136–146.

Hibbard, K.A., Archer, S., Schimel, D.S., Valentine, D.W., 2001. Biogeochemicalchanges accompanying woody plant encroachment in a subtropical Savanna.Ecology 82, 1999–2011.

IUSS Working Group WRB, 2006. World Reference Base for Soil Resources 2006, 2nded., World Soil Resources Report 103. FAO, Rome, 128 pp.

Jackson, R.B., Banner, J.L., Jobbagy, E.G., Pockman, W.T., Wall, D.H., 2002. Ecosystemcarbon loss with woody plant invasion of grasslands. Nature 418, 623–626.

Junta de Andalucía, 2009. Mapa de usos y coberturas vegetales del suelo deAndalucía 1956 escala 1:25.000. Consejería de Medio Ambiente. Junta deAndalucía.

Junta de Andalucía, 2013. Red de información ambiental REDIAM. http://www.juntadeandalucia.es/medioambiente/site/web/rediam (accessed 11.13.).

Knops, J.M.H., Tilman, D., 2000. Dynamics of soil nitrogen and carbon accumulationfor 61 years after agricultural abandonment. Ecology 81, 88–98.

Kuhn, M., Wing, J., Weston, S., Williams, A., Keefer, C., Engelhardt. A., 2012. Caret:classification and regression training. R package, version 5.15. http://cran.r-project.org/web/packages/caret/ (accessed 08.12.).

Kunkel, M.L., Flores, A.N., Smith, T.J., McNamara, J.P., Benner, S.G., 2011. A simplifiedapproach for estimating soil carbon and nitrogen stocks in semi-arid complexterrain. Geoderma 165, 1–11.

La Mantia, T., Gristina, L., Rivaldo, E., Pasta, S., Novara, A., Rühl, J., 2013. The effects ofpost-pasture woody plant colonization on soil and aboveground litter carbonand nitrogen along a bioclimatic transect. iForest 6, 238–246.

Lal, R., 2004. Soil carbon sequestration impacts on global climate change and foodsecurity. Science 304, 1623–1627.

Lesschen, J.P., Cammeraat, L.H., Kooijman, A.M., van Wesemael, B., 2008.Development of spatial heterogeneity in vegetation and soil properties afterland abandonment in a semi-arid ecosystem. J. Arid Environ. 72, 2082–2092.

McCulley, R.L., Archer, S.R., Boutton, T.W., Hons, F.M., Zuberer, D.A., 2004. Soilrespiration and nutrient cycling in wooded communities developing ingrassland. Ecology 85, 2804–2817.

Minasny, B., McBratney, A.B., Malone, B.P., Wheeler, I., 2013. Chapter one–digitalmapping of soil carbon. Adv. Agron. 113, 1–47.

Novara, A., Gristina, L., La Mantia, T., Rühl, J., 2013. Carbon dynamics of soil organicmatter in bulk soil and aggregate fraction during secondary succession in aMediterranean environment. Geoderma 193–194, 213–221.

Nunes, A.N., Coelho, C.O.A., De Almeida, A.C., Figueiredo, A., 2010. Soil erosion andhydrological response to land abandonment in a central inland area of Portugal.Land Degrad. Dev. 21, 260–273.

Onate, J.J., Peco, B., 2005. Policy impact on desertification: stakeholders’ perceptionsin southeast Spain. Land Use Policy 22, 103–114.

Poeplau, C., Don, A., Vesterdal, L., Leifeld, J., Van Wesemael, B., Schumacher, J.,Gensior, A., 2011. Temporal dynamics of soil organic carbon after land-usechange in the temperate zone–carbon response functions as a model approach.Global Change Biol. 17, 2415–2427.

Post, W.M., Kwon, K.C., 2000. Soil carbon sequestration and land-use change:processes and potential. Global Change Biol. 6, 317–327.

Powlson, D., Smith, P., De Nobili, M., 2007. Soil organic matter. In: Gregory, P.J.,Nortcliff, F. (Eds.), Soil Conditions and Plant Growth. Wiley-Blackwell, Chicester,UK, pp. 86–131.

R Development Core Team, 2012. R: a language and environment for statisticalcomputing. R Foundation for Statistical Computing. http://www.r-project.org(accessed 11.13.).

Raich, J.W., Potter, C.S., 1995. Global patterns of carbon dioxide emissions from soils.Global Biogeochem. Cycles 9, 23–36.

Rhoades, C.C., Eckert, G.E., Coleman, D.C., 2000. Soil carbon differences among forestagriculture and secondary vegetation in lower Montane Ecuador. Ecol. Appl. 10,497–505.

Rivas-Martínez, S., Asensi, A., Díez-Garretas, B., Molero, J., Valle, F., 1997.Biogeographical synthesis of Andalusia (southern Spain). J. Biogeogr. 24,915–928.

Ruiz-Sinoga, J.D., García-Marín, R., Martínez-Murillo, J.F., Gabarrón-Galeote, M.A.,2010a. Pluviometric gradient incidence and the hydrological behaviour of soilsurface components (southern Spain). Land Degrad. Develop. 21, 484–495.

Ruiz-Sinoga, J.D., Martínez-Murillo, J.F., Gabarrón-Galeote, M.A., García-Marín, R.,2010b. Effects of exposure scrub position and soil surface components on thehydrological response in small plots in southern Spain. Ecohydrology 3,402–412.

Ruiz-Sinoga, J.D., Gabarrón-Galeote, M.A., Martínez-Murillo, J.F., García-Marín, R.,2011a. Vegetation strategies for soil water consumption along a pluviometricgradient in southern Spain. Catena 84, 12–20.

Ruiz-Sinoga, J.D., Martínez-Murillo, J.F., Gabarrón-Galeote, M.A., García-Marín, R.,2011b. The effects of soil moisture variability on the vegetation pattern inMediterranean abandoned fields (southern Spain). Catena 85, 1–11.

Salomé, C., Nunan, N., Pouteau, V., Lerch, T.Z., Chenu, C., 2010. Carbon dynamics intopsoil and in subsoil may be controlled by different regulatory mechanisms.Global Change Biol. 16, 416–426.

Sherrod, L.A., Dunn, G., Peterson, G.A., Kolberg, R.L., 2002. Inorganic carbon analysisby modified pressure-calcimeter method. Soil Sci. Soc. Am. J 66, 299–305.

Schulp, C.J.E., Veldkamp, A., 2008. Long-term landscape-land use interactions asexplaining factor for soil organic matter variability in Dutch agriculturallandscapes. Geoderma 146, 457–465.

Six, J., Conant, R.T., Paul, E.A., Paustian, K., 2002. Stabilization mechanisms of soilorganic matter: implications for C-saturation of soils. Plant Soil 241, 155–176.

Templer, P.H., Groffman, P.M., Flecker, A.S., Power, A.G., 2005. Land use change andsoil nutrient transformations in the Los Haitises region of the DominicanRepublic. Soil Biol. Biochem. 37, 215–225.

U.S. Geological Survey, 2013. https://lpdaac.usgs.gov/products/modis_products_ta-ble/mod17a3 (accessed 11.13.).

West, T.O., Post, W.M., 2002. Soil organic carbon sequestration rates by tillage andcrop rotation: a global data analysis. Soil Sci. Soc. Am. J. 66, 1930–1946.

Zhang, J.B., Song, C.C., Wang, S.M., 2007. Dynamics of soil organic carbon and itsfractions after abandonment of cultivated wetlands in northeast China. Soil Till.Res. 96, 350–360.