Interactive comment on “ The influence of C 3 and C 4 vegetation on soil organic matter dynamics in contrasting semi-natural tropical ecosystems ”

Abstract. Variations in the carbon isotopic composition of soil organic matter (SOM) in bulk and fractionated samples were used to assess the influence of C3 and C4 vegetation on SOM dynamics in semi-natural tropical ecosystems sampled along a precipitation gradient in West Africa. Differential patterns in SOM dynamics in C3/C4 mixed ecosystems occurred at various spatial scales. Relative changes in C / N ratios between two contrasting SOM fractions were used to evaluate potential site-scale differences in SOM dynamics between C3- and C4-dominated locations. These differences were strongly controlled by soil texture across the precipitation gradient, with a function driven by bulk δ13C and sand content explaining 0.63 of the observed variability. The variation of δ13C with soil depth indicated a greater accumulation of C3-derived carbon with increasing precipitation, with this trend also being strongly dependant on soil characteristics. The influence of vegetation thickening on SOM dynamics was also assessed in two adjacent, but structurally contrasting, transitional ecosystems occurring on comparable soils to minimise the confounding effects posed by climatic and edaphic factors. Radiocarbon analyses of sand-size aggregates yielded relatively short mean residence times (τ) even in deep soil layers, while the most stable SOM fraction associated with silt and clay exhibited shorter τ in the savanna woodland than in the neighbouring forest stand. These results, together with the vertical variation observed in δ13C values, strongly suggest that both ecosystems are undergoing a rapid transition towards denser closed canopy formations. However, vegetation thickening varied in intensity at each site and exerted contrasting effects on SOM dynamics. This study shows that the interdependence between biotic and abiotic factors ultimately determine whether SOM dynamics of C3- and C4-derived vegetation are at variance in ecosystems where both vegetation types coexist. The results highlight the far-reaching implications that vegetation thickening may have for the stability of deep SOM.


Introduction
A progressive "thickening" of woody vegetation in grasslands and savannas is a global phenomenon that has been widely documented (Archer et al., 2001;Boutton et al., 2009;Guillet et al., 2001;Krull et al., 2005;Liao et al., 2006;Pessenda et al., 1998).Woody thickening is being promoted by climate change, changes in fire regimes and other anthropogenic land use activities (Jackson et al., 2000;Krull et al., 2005;Silva et al., 2008) with increased woody (C 3 ) plant growth in response to continually increasing atmospheric CO 2 concentrations a likely key driver (Bond and Midgley, 2012;Buitenwerf et al., 2012;Donohue et al., 2013).The impact of these vegetation dynamics on ecosystem biogeochemistry and the global carbon cycle may be highly significant given the large extent of grass-dominated ecosystems, which represent about 30 % of primary production of all terrestrial vegetation and store 10-30 % of all soil organic carbon (SOC; Eswaran et al., 1993;Grace et al., 2006;Hall and Scurlock, 1991).Moreover, despite an increasing number of studies of terrestrial carbon dynamics, improved predictions of the impacts of future climate-driven changes on the tropical SOM pool requires a more detailed understanding of the interactions between vegetation, climate, edaphic and disturbance effects than is currently available (Archer et al., 2001;2004;Boutton et al., 2009;Jackson et al., 2000).
The use of isotopic techniques in ecological studies broadens the possibilities for better assessing soil carbon dynamics (Bernoux et al., 1998;Bird and Pousai, 1997;Bird et al., 1996;Boutton, 1996;Leifeld and Fuhrer, 2009), the documentation of past vegetation changes (Krull et al., 2005;Liao et al., 2006;Pessenda et al., 1998;Silva et al., 2008), and the reconstruction of earlier environments (Bird et al., 1996;Cerling et al., 2011).The use of the carbon isotope composition of SOM has proven to be a useful tool for investigating the influence of C 3 and C 4 vegetation on SOM dynamics (Wynn and Bird, 2007), and for identifying recent (∼ 100 years) vegetation change patterns that in the past could only be assessed by interpreting historical aerial photography or satellite imagery (Krull et al., 2005).This approach relies on the distinct carbon isotopic (δ 13 C) values of tropical grasses, employing the C 4 photosynthetic pathway (−14 ‰) compared to the δ 13 C value of trees and shrubs that utilize the C 3 photosynthetic pathway (−27 ‰; average values from Krull et al., 2007).It is feasible to obtain an integrated measure of the carbon isotopic composition of regional vegetation using SOM, given that this pool integrates the isotopic signature of the precursor biomass, thus serving as a record of changes in the ratio of tree and grass-derived organic matter input to the soil over different spatiotemporal scales (Bird et al., 2004;Lloyd et al., 2008).However, the simple interpretation of the δ 13 C value of SOM in a soil profile may not be straightforward because of fractionation effects such as those associated with microbial reprocessing of SOM, differential stabilization of SOM compounds, and the terrestrial "Seuss effect" (Bird et al., 1996;Blagodatskaya et al., 2011;Ehleringer et al., 2000;Rumpel and Kögel-Knabner, 2011).Moreover, the specific characteristics of a soil can fundamentally affect carbon isotopic dynamics through differential physico-chemical protection of SOM as influenced by the chemical, mineralogical, and textural properties of the soil (Krull et al., 2003;Sollins et al., 2009;Veldkamp, 1994).
Physical fractionation of SOM is commonly employed to simplify the complex soil matrix into discrete fractions of similar physico-chemical characteristics, as defined by aggregation, particle size, density, or a combination of these (Crow et al., 2007;Moni et al., 2012;Zimmermann et al., 2007).Each individual technique will result in a specific number of fractions with conceptually different carbon turnover times, although a common feature shared by all methods is the differentiation between mineral-associated organic matter and free particulate organic matter.The latter fraction is predominantly composed of plant-derived material in the early stages of decomposition and is generally reported to mineralise more quickly than physically protected mineral-associated fractions (Six and Jastrow, 2002;Sollins et al., 1996;Zimmermann et al., 2007).The determination of mean residence times (τ ) of SOM by radiocarbon ( 14 C) age measurements, combined with δ 13 C characterisation of vegetation change recorded in SOM has provided strong evidence of the nature and timing of tropical vegetation shifts in the past (Guillet et al., 2001;Krull et al., 2005;2007;Pessenda et al., 1998).While the use of these techniques yields useful information about SOM dynamics, there are a number of limitations and biases that need to be taken into account when interpreting the results (Creamer et al., 2011;Crow et al., 2007;Trumbore, 2009).Several authors have warned about possible biases related to deep SOM 14 C dating, advising that single analyses of bulk SOM would fail to take into account inputs of fresh organic matter from root decomposition and solubilised SOC, which could distort the true age of stabilised SOM at depth (Gaudinski et al., 2001;Herold et al., 2014;Krull et al., 2005;Wurster et al., 2010).This artefact can be minimised by implementing an SOM fractionation procedure, whereby distinct organic matter fractions are independently analysed both for δ 13 C and 14 C at different depths.Nevertheless, it remains a challenge to assess past tropical vegetation dynamics without the confounding effects posed by the interaction of varying climatic and edaphic factors.
Environmental gradients offer great opportunities for both understanding mechanisms of abiotic control on ecosystem processes and to study the potential impacts of global change (Koch et al., 1995).A good example of such a gradient is the sharp climatic gradient existing between the arid conditions characteristic of West Africa's inner continental regions and the humid environments predominant near its southern coast, which strongly influences the distribution and functioning of the wide range of natural ecosystems currently found across this expansive region (Domingues et al., 2010;Saiz et al., 2012;Schrodt et al., 2015;Veenendaal et al., 2015).However, in addition to climatic factors, SOM dynamics may vary considerably depending on the specific physical and chemical characteristics of the soil (Bruun et al., 2010;Paul et al., 2008;Plante and Conant, 2014).Certainly, West Africa presents a wide variety of soil types not directly related to contemporary climate or vegetation (Pullan, 1969), which combined with increasing anthropogenic pressure and the inherent natural heterogeneity of these ecosystems, makes any attempt to generalise SOM dynamics challenging (Saiz et al., 2012).
Field studies using the stable carbon isotopic composition of SOM can help assess the influence of C 3 and C 4 vegetation on SOM dynamics, enabling a test to determine whether there are differential patterns in their mineralisation potential as previously reported in laboratory-based studies (Wynn and Bird, 2007).The objectives of this study are (1) delineate SOM dynamics across contrasting C 3 /C 4 mixed semi-natural tropical ecosystems; (2) investigate any potential variation in tropical vegetation thickening along the precipitation transect; and (3) unambiguously evaluate the effect of vegetation thickening on SOM dynamics in two contiguous but structurally different woodland ecosystems.

Characteristics of the sites
The description of the sites and sampling methodology used in the present work have been provided in detail in a previous publication focusing on the determinants of SOC stocks (Saiz et al., 2012), with further information provided in Torello-Raventos et al. (2013) and Veenendaal et al. (2015).Hence, a short summary is provided here.
The study was conducted across a latitudinal transect encompassing a wide range of semi-natural ecosystems and soils characteristic of West Africa (Figs. 1, 2).The soil sampling campaign took place in Ghana, Burkina Faso, and Mali from August to October 2006.Fourteen study sites comprised of a total of ten 1 ha and four 0.5 ha plots, were established in locations previously identified as representative of the potential natural vegetation of the region.These included National Parks, Forest Reserves, and other legally protected areas except for the Sahelian sites in Mali, which had no specific conservation status and were subject to varying degrees of grazing pressure, the latter also being the case for the most northern Sudan savanna sites of Burkina Faso.The transect was established on consistently flat terrain with less than 100 m of altitudinal variation between all sites.The northern end of the transect was dominated by Sahelian ecosystems, consisting of grassland savannas (Torello-Raventos et al., 2013) occurring on the relative nutrient-poor Arenosols on the southern border of the Sahara, receiving low mean annual precipitation (P A ∼ 0.3 m a −1 ) and subject to high rates of potential evaporation (Table A1).Further south there is a natural progression into more woodydominated savanna forms heavily influenced by a gradual increase in P A .The southern end of the transect corresponds to the more humid sites supporting semi-deciduous tall forests (P A > 1.2 m a −1 ).The variation in mean annual temperature is less than 4 • C across all sites.Saiz et al. (2012) report a wide range of soil types that are the result of the interaction of contrasting geological, climatic, and vegetation factors over extended time periods.
Estimates of the fractional vegetation cover of woody vegetation (FC w ) were obtained as described in Veenendaal et al. (2015), while estimates of the fractional vegetation of the axylale vegetation (FC a ) are as in Torello-Raventos et al. (2013).In short, the canopy area index (C), which is defined as the sum of individual canopy projected area divided by the ground area, was estimated separately for three woody strata.These strata are distinguished on the basis of stem diameter (D) at breast height (1.3 m), and individual tree height (H ).The upper (u) stratum consists of trees with d > 0.1 m, all of which were individually measured at each plot.The mid (m) stratum consists of woody vegetation of 0.1 > d > 0.025 m, which was quantified through measurements made along ten 50 m long transects.The subordinate (s) or lower stratum is made up of trees and shrubs with d < 0.025 m and H > 1.5 m, which were quantified in the same way as the mid-stratum.Subsequently, stand-level canopy projected area for each stratum (Cu, Cm, Cs) was estimated according to site-specific allometric equations presented in Torello-Raventos et al. (2013).Therefore, assum- ing a random distribution of trees and/or shrubs, the fraction of ground covered by crowns (including within-crown light gaps), which we refer here as the FC of woody vegetation (FC w ) can be estimated as FC a was visually recorded along a series of transects with a sampling intensity of 110 × 1 m 2 quadrants per plot.

Soil sampling
We made use of a stratified sampling strategy employed by Bird et al. (2004) and Wynn et al. (2006) that has been proven to be well suited for both studying the inherent spatial heterogeneity of SOM that is typically exhibited in mixed C 3 /C 4 environments, and achieving robust regional estimates of SOC inventories.This sampling approach consists of taking samples in a stratified manner near trees "Tree" (-T) samples at half canopy radius from trunks, and away from trees "Grass" (-G) samples at half the maximum distance between trees.
Surface litter was removed when present at each sampling location and three soil samples were taken at 0-0.05 m and one sample at 0-0.30 m with the aid of a stainless steel corer 40 mm inner diameter (ø).All the samples were placed in labelled zip-lock bags.Three replicate samples were collected at 0-0.05 m to smooth out local heterogeneity, which is generally more pronounced closer to the soil surface compared to deeper locations.This procedure was replicated five times at each site (both for -T and -G locations).Replicates were subsequently bulked according to location (-T versus -G) and depth (0-0.05 and 0-0.30m).In addition, a soil pit was handdug up to 2 m depth at each plot to assess soil type, provide an estimate of root biomass, and allow for the description of soil characteristics (Quesada et al., 2011).Samples were taken at 0-0.05, 0.05-0.30,0.30-0.50m and then every 0.5 up to 2 m depth (impenetrable layers permitting).

SOM fractionation procedure
Soil samples were fractionated using a combination of physical sieving and density separation following a procedure that has been described in detail by a number of publications (Dondini et al., 2009;Wurster et al., 2010;Zimmermann et al., 2007).Briefly, thirty grams of dry-sieved soil (< 2 mm) was added to 161 mL of ultrapurified water (Milli-Q, Millipore Corp., Massachusetts, USA), and dispersed using a calibrated ultrasonic probe-type (VC 750, Sonics & Materials Inc., Newtown, CT, USA) with an output-energy of 22 J mL −1 .The soil solution was then wet sieved through a 53 µm mesh until the rinsing water was clear, and the size fractions were subsequently dried at 40 • C. All the particulate material passing the 53 µm mesh sieve corresponded to the silt and clay fraction (s+c).The material > 53 µm containing sand and water stable aggregates (heavy fraction -HF) was separated from the light fraction (LF) by means of static dense media separation (Wurster et al., 2010) prepared at 1.87 g cm −3 using sodium polytungstate (Sometu-Europe ™ , Berlin, Germany).All the resultant fractions were then washed and filtered at 0.45 µm with ultrapure water to remove any traces of salt, dried at 40 • C, and the weight of each fraction was determined before further analysis.

Analytical methods and calculations
Stable isotope composition and elemental abundances of carbon and nitrogen were determined in duplicate in powdered samples using a Costech Elemental Analyzer fitted with a zero-blank auto-sampler coupled via a ConFloIII to a Ther-moFinnigan DeltaPlus-XL using Continuous-Flow Isotope Ratio Mass Spectrometry (CF-IRMS) at the University of St Andrews Facility for Earth and Environmental Analysis stable isotope laboratory (U.K.).Precisions (S.D.) on internal standards for elemental carbon and nitrogen abundances and stable carbon isotopic composition were better than 0.09 % and 0.2 ‰ respectively.
The relative change in C / N ratios (Rc) for the very stable (s+c) fraction in relation to the relatively fresh particulate organic matter represented by the light fraction (LF) was determined for each sampling location (-G and -T) according to the following equations: where C/N s+c and C/N LF are the C / N ratios of the s+c and LF fractions respectively, and the subscripts (-G or -T) indicate the sampling location.Subsequently, the difference in the relative change in C / N ratios for the two contrasting SOM fractions between both sampling locations ( C/N * G-T ) was calculated for each site according to Eq. ( 3): Radiocarbon analyses were conducted at the Accelerator Mass Spectrometry facility managed by the Australian Nuclear Science and Technology Organisation (ANSTO) in Kirrawee, NSW, Australia. 14C measurement efforts were concentrated at the two transitional sites (BFI-02 and 04) on depth intervals where δ 13 C values of bulk SOM showed a significant shift (0.3-0.5 m), and also in the deepest studied interval (1.5-2.0 m) in order to determine the τ of mineralbound deep SOM fractions.Here we use radiocarbon age of each fraction as a proxy for its average τ .Radiocarbon ages were calculated according to Stuiver and Polach (1977).Calculation of τ for samples classified as "Modern" (pMC > 100 %) was determined making use of the model presented by Harkness et al. (1986).The reference 14 CO 2 atmospheric data for the Northern Hemisphere in 2006 (date of collection) are those published in Hua and Barbetti (2004) and Levin et al. (2008).Simple interpolation was used to quantify the τ where the data fell between two points on the model.Furthermore, soil texture, pH, and effective cation exchange capacity (ECEC) were determined for different depth intervals to help explain potential variations in τ between the two transitional sites (BFI-02 and 04).Particle size distribution was determined gravimetrically as described by Reeuwijk (2002).Soil pH was measured using a digital pH meter in a 2 : 1 water:soil solution.The CEC was determined by inductively coupled plasma optical emission spectrometry (ICP-OES) extraction of soils using dilute unbuffered Silver-Thiourea for Al, K, Mg, Ca and Na as described by (Quesada et al., 2011), and ECEC was calculated as the sum of these bases.

Stable carbon isotopic composition of SOM across the precipitation transect
Associated with the gradual shift in vegetation from the relatively open savannas found in the interior of the continent in Mali to the dense forests near the Atlantic coast of Ghana (Figs. 1, 2), were changes in the carbon stable isotopic composition of SOM along the transect (Fig. 3a).Shallow soil samples (0-0.05m) showed distinctly high δ 13 C values in grass-dominated environments of the north, in contrast with the lower δ 13 C values observed in the forest ecosystems occurring in the more moist, southern end of the transect.Within each sampling site, higher δ 13 C values were consistently obtained at -G sampling locations.The difference between -T and -G sampling locations at each site was consistently larger at savanna sites compared to the difference observed in forests (Fig. 3a).Nonetheless, it is worth noting the strong effect of tree clumping in δ 13 C at sites BDA 01-02, as well as the relatively large difference observed between the two sampling locations at the KOG-01 site, which reflects the vegetation composition of the southernmost savanna site sampled.SOC contents and δ 13 C values ranged from 1.5 mg C g −1 and −13.7 ‰, respectively, in one of the Sahelian sites (HOM-01), to 55.5 mg C g −1 and −28.5 ‰ in the semideciduous dry forest (ASU-01) at the south end of the transect (Fig. 4).Carbon contents were generally higher at -T locations, compared to -G locations, while the opposite was true for δ 13 C values.

Variation in δ 13 C values and C / N ratios in bulk soil and SOM fractions
Analyses of bulk soil samples showed that dry forests and woody savannas typically had lower C / N ratios than grassdominated ecosystems (Fig. 5a).Likewise, for the LF the relationship between δ 13 C and C / N ratios showed lower C / N values being associated with woodier ecosystems (Fig. 5b); though with the notable exception of the most northerly Sahelian grassland savannas occurring in very sandy soils (HOM sites), which were characterised by very low C / N ratios.Consistently higher C / N ratios were observed in -T compared to -G sampling locations, with the differences between them becoming larger as δ 13 C values became more positive.
The stable s+c fraction consistently exhibited lower C / N ratios compared to LF at each site (data not shown).However, the relative difference in C / N ratios for these two contrasting SOM fractions generally differed between the (-G and -T) sampling locations at each site (Fig. 6a).Sites showing positive C/N * G-T values indicate that "Grass" sampling locations had a greater relative reduction in C / N ratios than their "Tree" counterparts, while the opposite was true for negative C/N * G-T values.The datum with the low- Grass, Tree, and Clumps of trees).Sites are ordered by decreasing latitude; (b) relationship between the weighted average stable carbon isotopic composition of all sample locations and both the fractional vegetation cover (FC) of all woody vegetation taller than 1.5 m, and the axylale vegetation (grass and herbs).The gradation intensities of the axylale symbols correspond to the relative contribution of C 4 species over the total axylale vegetation.Accordingly, white, grey and dark-grey symbols represent > 0.66, 0-66-0.33,and < 0.33 of that contribution respectively.Regressions have the form FC = a+b× (δ 13 C) and the regression coefficients (r 2 ) are 0.81 and 0.93 for the woody (solid line) and axylale (dotted line) vegetation respectively; p < 0.05 in both regressions.The pure grassland stand (BDA-03) was not included in the regressions.est δ 13 C and a neutral C/N * G-T value corresponds to a dry semideciduous forest with no C 4 vegetation (ASU-01).The relatively large variation in C/N * G-T values across the transect appeared to be strongly controlled by soil textural characteristics (Fig. 6a) with a function driven by sand content and the weighted soil δ 13 C explaining 0.63 of this variation (Fig. 6b).Here the weighted soil δ 13 C was calculated for each site according to woody fractional vegetation cover shown in Fig. 3b.

Stable carbon isotopic composition of SOM with depth across the transect
The differences in δ 13 C values between samples collected within the 0.05-0.3m depth interval and the topmost soil layer (0-0.05m) were relatively small for most sites across the precipitation gradient (< 2 ‰ ; Fig. 7).The exception to this trend occurred at transitional sites, which exist on the boundary between naturally occurring forest and savannas at the wettest end of the transect (KOG-01 and BFI sites).Similarly, the differences in δ 13 C between the 0.3-0.5 m depth interval and the topmost soil layer (0-0.05m) were also the largest in these transitional ecosystems.These differences were consistently positive along the transect, except for the case of the Sahelian ecosystems (HOM sites) and KOG-01, a savanna woodland occurring on a very sandy soil.Figure 8 shows the variation of δ 13 C values in soil profiles spanning 2 m depth for four selected sites, which illustrates the range of past changes in vegetation that may have occurred at some sites across the transect.A slight 13 C enrichment with depth is observed in the semideciduous forest at the wettest end of the transect (ASU-01, Fig. 1).Both BFI-02 and BFI-04 show a very large increase in δ 13 C values with depth.By contrast, a Sahelian site (HOM-01) showed a gradual decrease in δ 13 C values with soil depth.(BFI-02 and BFI-04).Both sites display varying degrees of increase in SOM δ 13 C values with soil depth.The low δ 13 C values observed in the top soil layers (0-0.05m) agree well with the current presence of nearly closed canopy woodlands in which C 4 vegetation is either absent (BFI-04), or virtually absent (BFI-02).There were, however, differences in the extent to which δ 13 C increased with soil depth at each site, although both ecosystems showed a sharp increase in δ 13 C values within the first 0.5 m.This trend was most acute for the woodland savanna (BFI-02), which exhibited values as high as −18.2 ‰ at 0.3-0.5 m, representing an absolute difference of more than 4 ‰ for that depth interval between the two sites.At deeper locations δ 13 C was relatively invariant for the dry forest, while there was a gradual change in δ 13 C towards lower values in the woodland savanna, nearly matching the trend exhibited by the dry forest.

Shifts in
Our results show that SOC contents, root biomass, and C / N ratios decreased with depth (Table 1).Moreover, radiocarbon analyses confirmed that high C / N ratios observed near the surface represented relatively young (with three out of four dates reflecting the contribution of modern carbon fixed after nuclear weapons testing in the late 1950s; Table A2).Furthermore, as is shown in Figure A1 there was

Stable carbon isotopic composition of SOM across the precipitation transect
The broad range of biotic and abiotic conditions present across the precipitation gradient was reflected in a wide diversity of vegetation structural formation types (Fig. 2).Besides climate, the inherent characteristics of the soil play a dominant role in determining both plant nutrient availability and the amount of water available for plant growth, making it one of the most important factors controlling the type of vegetation at a given site (Bloomfield et al., 2014;Quesada et al., 2012;Saiz et al., 2012).In turn, vegetation type strongly controls the quantity and quality of organic inputs returning to the soil, thus exerting a strong influence on its carbon storage potential (Post et al., 1982;Wynn and Bird, 2007).The gradual depletion in 13 C associated with increasing P A reflects current vegetation patterns well, with high δ 13 C values corresponding to grass-dominated savannas (Fig. 3a), showing an axylale (herbs and grasses) layer containing a variable mixture of C 3 and C 4 species (Fig. 3b; see also Torello-Raventos et al. 2013).
The relationship between δ 13 C and SOC content in mixed C 3 /C 4 systems has been reported to be linear by several local studies (Bird et al., 2004(Bird et al., , 2000)), with this attributed to differences in the input rates and turnover times of woody and grass-derived carbon.In the present work the relationship was, however, non-linear (Fig. 4), suggesting that one or several additional factors may have been involved in generating the observed pattern at a broad scale.In our study, the northern arid sites are heavily limited by low water availability with a rainy season of less than 4 months (Schrodt et al., 2015) which undoubtedly not only reduces organic inputs into the soil but likely the rate of OM decomposition as well.The latter is confirmed by previous research conducted over broad latitudinal gradients, which has shown SOM turnover rates to be heavily controlled by climate (Bird and Pousai, 1997;Bird et al., 1996).Besides climatic factors, the contrasting characteristics of the soils studied have been shown to have a strong influence on the potential preservation of SOM (Saiz et al., 2012).It is therefore plausible that the contrasting textural and mineralogical characteristics of these soils may have a major effect on both the physical protection of particulate carbon and the chemical stabilization of 13 C enriched microbial metabolites (Dümig et al., 2013;Šantrůčková et al., 2000).Such a strong influence would explain the comparatively high δ 13 C and SOC values observed in the iron-rich, silty loams of the BDA sites in particular (Saiz et al., 2012).

Differential patterns in SOM dynamics across contrasting C 3 /C 4 mixed ecosystems
Savannas in their natural state are generally less productive than forests due to a myriad of reasons, which include characteristically lower precipitation, the existence of periods of drought of varying severity, discrete seasonal productivities, and an overall net reduction of annual productivity due to fire (Grace et al., 2006).Studies conducted in West African savannas show that more than half of the net primary productivity of these ecosystems is attributable to C 4 grasses (Menaut and Cesar, 1979), a fact further supported by the pantropical productivity estimates for savannas reported by Lloyd et al. (2008).
In order to facilitate the study of heterogeneous C 3 /C 4 environments, we adopted a soil sampling strategy that allowed for the comparison of soil properties at systematically defined locations.This approach allows for the relative unbiased accounting of the effect of woody cover, thus minimising the potential confounding effects derived from preferential sampling of either grass or woody-dominated areas across all study sites.Indeed, there was considerable variation in carbon isotope composition of surface soil samples at the individual plot scale, as demonstrated by the large differences in δ 13 C values between -T and -G sampling locations in ecosystems where C 3 /C 4 species coexist (Fig. 3).Similarly, carbon contents were generally higher in -T locations than in their -G counterparts across all sites (Fig. 4).This, especially as it has been reported by some earlier studies (Bird et al., 2000;2004;Wynn and Bird, 2007), suggests that such variation could be mainly related to differences in the input rates and turnover times of tree and grass-derived carbon.However, both the large temporal and spatial variability in OM inputs and turnover rates, as well as the potentially large time lag between production and decomposition processes, make any comprehensive characterisation of SOM dynamics very challenging, especially under field conditions.Indeed, it is characteristically difficult to make sound generalizations of even the most fundamental properties of natural ecosystems.Such is the case for plant biomass C / N ratios, a feature considered to be highly relevant in SOM decomposition processes (Brady and Weil, 2007;Kirschbaum et al., 2008;Tian et al., 1992).
Plant C / N ratio is greatly dependent on a number of factors that include the type and age of the tissue, sampling season, and the specific characteristics of the species being sampled (i.e.drought tolerant, contents of polyphenol and lignin, etc; Abbadie et al., 2006).Large spatial variability in OM inputs is inherent in commonly heterogeneous tropical environments (Mordelet and Menaut, 1995), and while an adequate assessment of aboveground litter may be relatively straightforward, it is certainly much more difficult to obtain accurate information on the quality, quantity, and spatiotemporal variability of belowground litter dynamics (Gignoux et al., 2006).Hence, it is highly problematic to establish an unambiguous relationship between the properties of the precursor biomass and that of SOM. Figure 5 shows that dry forests and woody savannas typically had lower C / N ratios than grass-dominated ecosystems.The latter environments may also present relatively high C / N values as influenced by their comparatively larger amounts of pyrogenic carbon (Saiz et al., 2012).The exception to such a trend were the Sahelian ecosystems (HOM-01-02) which had very low C / N ratios as a result of their very low biomass (Aranibar et al., 2004), the occurrence of dominant, albeit sparse, nitrogen fixing species (i.e.Acacia sp, Zornia sp.; Hiernaux et al., 2009b), and the external nitrogen inputs from domestic grazers (Saiz et al., 2012).The LF or particulate organic matter is the soil fraction that best reflects recent organic inputs to the soil, as it includes contributions from both aboveground and belowground biomass.This fraction is widely referred to as a very labile SOM component with characteristically high C / N ratios, mainly composed of recent OM inputs that have not yet been physically protected by the soil matrix (Marin-Spiotta et al., 2009;Zimmermann et al., 2007).
We therefore made use of the relative change in C / N ratios between two contrasting SOM fractions (LF and s+c) in order to assess potential differences in SOM dynamics in grass-and tree-dominated sampling locations.Figure 6 shows a relatively large variation in C / N* G−T values across the transect, which appears to be strongly controlled by soil textural characteristics.Negative C / N* G−T values were observed for relatively coarse-textured soils across Bulk soil is defined as < 2000 µm; < 53 µm is the fraction associated to silt and clay; HF is the heavy fraction > 53 µm (specific density > 1.87 g cm −3 ); LF is the light fraction > 53 µm (specific density < 1.87 g cm −3 ).τ of selected SOM fractions are shown adjacent to the fractions and denote conventional radiocarbon ages and, in between brackets, calculated τ derived from radiocarbon analyses presented in Table A2.
the wide range of ecosystems studied, indicating that "Tree" sampling locations exhibited a greater relative change in C / N ratios between the two contrasting SOM fractions than their 'Grass' counterparts.This implies that, at those sites, the processing of SOM was potentially faster in treedominated locations compared to those dominated by grass.The role of trees in preserving soil water on coarse-textured soils in semi-arid environments may be very relevant at the canopy level due to their provision of shade and the funnelling of precipitation with associated improvements in soil water availability (Abbadie et al., 2006;Mordelet et al., 1993;Ong and Leakey, 1999).Moreover, these ecosystems traditionally suffer from a recurrent loss of herbaceous cover through fire and/or overgrazing (Saiz et al., 2015), which have an adverse impact on the amount of fresh organic inputs returned to the soil, an aspect that may be particularly severe in grass-dominated locations (Abbadie and Bismarck Nacro, 2006).These factors, together with the comparably higher SOM contents observed in tree-dominated locations (Fig. 3a), may all result in the persistence of suitable environmental conditions further promoting the activities of SOM decomposers at these localities.
In addition to growing on relatively coarse-textured soils, the transitional sites had a much lower abundance of C 4 vegetation (Fig. 3b) which seems is progressively being replaced by newly established C 3 plants as a result of woody thickening (the reader is referred to the next section for more details).This might result in low C / N_LF G values, and consequently an overall negative C/N * G-T .On the other hand, positive C/N * G-T values were only evident in open grassdominated ecosystems occurring on fine-textured soils.Besides maintaining the herbaceous cover for longer than Sahelian ecosystems (HOM sites), these soils also have a greater water retention capacity compared to sites on coarse-textured soils, which may diminish the role of trees as key factors for maintaining suitable conditions for the activities of SOM decomposers.
Further to site-specific soil stabilization mechanisms and differences in the C / N composition of the precursor biomass, it is also likely that differences in the organic biochemical composition of C 3 and C 4 -derived litter represent an inherent primary control on their respective decomposition rates (Meentemeyer, 1978;Melillo et al., 1982).Indeed, the recalcitrance of plant biomass to degradation is a function of how polymers (e.g.lignin and cellulose) crosslink and aggregate within cell walls (McCann and Carpita, 2008), and it is well recognised that the composition of cell walls of grasses greatly differs from those of most other angiosperms (Carpita and Gibeaut, 1993).Abbadie and Bismarck Nacro (2006) showed the preponderant role that grass root mineralisation plays in primary production for savannas where the axylale layer was almost exclusively composed of C 4 grass perennials, which also provides support for the grass component being very dynamic.Recent work suggests litter carbon chemistry is a key factor controlling litter decay through its effect on microbial substrate use efficiency (Cotrufo et al., 2013).While the latter is not the focus of the present work, our findings broaden the possibilities for further research to be conducted on preferential substrate utilisation of grass-derived carbon by microbial communities in mixed C 3 /C 4 ecosystems.
Laboratory-based work has reported different turnover times for tree and grass-derived carbon in soils from mixed C 3 /C 4 ecosystems (Bird and Pousai, 1997;Wynn and Bird, 2007).However, the findings of such studies cannot be universally extrapolated given that these controlled experiments could not account for the contrasting environmental conditions likely experienced by "Grass" and "Tree" sampling lo-cations that usually occur in real field settings.On the other hand, the work by Wynn and Bird (2007) agrees well with our view that, under comparable (site-scale) environmental conditions, C 4 -derived SOM decomposes faster than SOM derived from woody biomass in mixed C 3 /C 4 ecosystems.Our results further suggest that soil textural properties exert a strong influence on SOM dynamics across the ecosystems examined along the transect.We postulate that the interdependence between biotic and abiotic factors ultimately determine whether SOM dynamics of C 3 -and C 4 -derived vegetation differ from each other in ecosystems where both vegetation types coexist.

Vegetation shifts along the precipitation transect
The differences in δ 13 C values between the surface (0-0.05m) and the 0.3-0.5 m interval provide evidence of recent vegetation shifts across the precipitation gradient (Fig. 7).Even after considering the relatively small 13 C enrichment with depth typically observed in semi-arid soils (Bird et al., 2004;Krull et al., 2005), our results suggest a significant increase in woody vegetation at the wetter sites along the transect.The trend in isotope composition with depth is also heavily dependant on the characteristics of the soils, which have also been identified as one of the main determinants of vegetation type observed at each site (Saiz et al., 2012).Vegetation thickening reached its maximum in the transitional ecosystems (BFI sites; Fig. 2), which constitute the natural border between savanna and dry forests on the transect.The vegetation dynamics occurring at each end of the precipitation gradient were quite different (Fig. 8).On the one hand, the wettest end of the transect corresponded to a semideciduous dry forest (ASU-01) showing quite a stable δ 13 C composition of SOM with depth, which agrees well with the long-term persistence of a closed canopy ecosystem.On the other hand, the arid Sahelian ecosystems studied (HOM sites) may have experienced a relatively recent reduction in woody cover as evidenced from the large enrichment in 13 C towards the surface.There may be several reasons behind this potential thinning of woody biomass at the driest sites, with a combination of overgrazing, fuel harvesting, fires, and above all the severe droughts suffered over the past few decades, being the most likely causes (Krull et al., 2007).However, Sahelian ecosystems are known to have a large resilience to drought and grazing, to the extent that woody plant population dynamics can largely vary between nearby sites as a result of contrasting substrates, grazing intensities, land use history, and species composition (Hiernaux et al., 2009a;Mougin et al., 2009).
Vegetation shifts at a given site may alter the quantity and quality of OM inputs into the soil, as different vegetation types are likely to have distinct biochemical and physiological characteristics.Therefore, these vegetation dynamics will have significant impacts on the total production and allocation of biomass and on its mineralization potential (Melillo et al., 1982;Wynn and Bird, 2007).Moreover, fundamental differences in δ 13 C fractionation dynamics have been reported for different soils and types of organic matter (Krull and Skjemstad, 2003), and it is therefore of paramount importance to unambiguously evaluate the impact that vegetation shifts may have exerted on SOM dynamics in the past.

Effect of vegetation thickening on SOM dynamics in two neighbouring transitional ecosystems
As has been previously discussed, it is difficult to assess past vegetation dynamics in isolation from the confounding effects posed by the interactions between climatic and edaphic factors.However, a Natural Reserve in central Ghana provided the ideal conditions to study potential shifts in vegetation in two structurally different ecosystems occurring under the same climatic conditions given that the sites are < 1 km from each other.The two sites are classified as transitional ecosystems, each with vegetation widely divergent from the other (Table 1).BFI-02 is a savanna woodland with a relatively sparse tree canopy cover (0.6) that allows the development of a thin grass stratum.BFI-04 is a semideciduous dry tall forest with an almost complete tree canopy closure (0.8).
For a detailed floristic composition of these sites, refer to Domingues et al. (2010) and Torello-Raventos et al. (2013).
The detailed study of these two sites revealed a close concordance in δ 13 C values along the soil profile between bulk SOM and the fractions associated with silt and clay (s+c), and to a lesser degree to sand-size aggregates (HF; Fig. 9).The only obvious exception to this general trend was the surface layer (0-0.05m), where a significant proportion of the OM may not be associated with mineral phases.The observed decrease in both SOC contents and C / N ratios with depth at both sites (Table 1) is a commonly reported observation indicating that deep SOM is usually highly processed by microbes (Rumpel and Kögel-Knabner, 2011).It is also worth noting the large differences in C / N ratios between LF and the rest of the fractions, which confirm their contrasting degrees of decomposition.
Fractionation of OM down soil profiles has previously been used to isolate fresh OM inputs derived from current vegetation (Krull et al., 2005;Marin-Spiotta et al., 2009).Indeed, the LF had significantly lower δ 13 C values compared to the bulk soil, with this difference increasing with depth at both sites.This is consistent with higher contemporary C 3 vegetation input throughout the soil profile than was the case at some time in the past.On the other hand, the increasing δ 13 C values with soil depth is another commonly reported feature that may be explained by a combined effect of several factors, which include OM decomposition processes (i.e., progressive significance of microbial and fungal decay to the SOM pool), the influence of carbon fixed from the higher δ 13 C atmosphere that existed prior to significant fossil fuel burning, and stabilization mechanisms influenced by specific properties of a given soil (Bird et al., 1996;Ehleringer et Krull et al., 2003).However, the magnitude of the increase in δ 13 C values, and more specifically the large differences observed in δ 13 C between fresh organic matter (LF) and other fractions, are critical indicators that the vegetation at these sites is undergoing significant change (Krull et al., 2005).
It is worth reflecting on the very different τ observed between the two sites.Both ecosystems present highly dynamic SOM processes, highlighted by the relatively short τ of sand-size aggregates (> 53 µm HF) even at considerable depth (1.5-2.0 m; Fig. 9).Interestingly, the most stable SOM fraction associated with silt and clay (< 53 µm s+c) has a shorter radiocarbon age in the savanna woodland (1450 y BP) than in the pure C 3 dry forest site (3445 y BP).The main reason for the contrasting τ observed in comparable fractions between the two sites may be the different input rates of belowground OM.There is increasing evidence that SOM turnover is mainly controlled by its availability to decomposers (Don et al., 2013;Dungait et al., 2012).These sites are characterised by a low abundance of SOM, particularly at depth, and it is likely that woody encroachment has led to an increase in recent organic matter inputs below the topmost soil as a result of the larger/deeper root systems characteristic of woody vegetation (Boutton et al., 2009).This process may have been more pronounced in the savanna woodland (BFI-02 site) with the previously more abundant grassland stratum being gradually replaced with woody vegetation, as demonstrated by the large shift in δ 13 C values towards the surface (Fig. 9).While both sites show a strong reduction in root biomass with depth, this trend is more obvious in the case of the dry tall forest (BFI-04), which shows a lower belowground biomass content over the entire soil profile compared to the savanna woodland (BFI-02; Table 1).This observation is in agreement with findings reported by Lawson et al. (1970Lawson et al. ( , 1968) ) in work conducted in a savanna and a tropical deciduous forest in Ghana.Butler et al. (2012) also compared two structurally contrasting savannas, and found that the more open canopy savanna had a proportionally higher carbon investment belowground.Recent findings reported by Don et al. (2013) show that any increase in the carbon concentration down the soil profile decreases the distance between decomposer and substrate, which may increase accessibility and SOM turnover.Therefore, the enhanced contribution in belowground organic matter inputs in the savanna woodland as a result of woody encroachment could largely explain its comparatively shorter τ across all soil fractions compared to the dry forest (Figs. 9, A1).
It is well established that the physical protection of SOM by aggregation mechanisms plays a fundamental role in soil carbon stabilisation (Denef et al., 2001;Six et al., 2004).Root-derived particulate organic matter has a significant control on aggregate dynamics (Six et al., 2004) as the stabilization and de-stabilization of macroaggregates in soils is strongly linked to the incorporation and biodegradation of fresh plant debris (Puget et al., 2000).This fresh OM is even-tually redistributed among aggregate classes through their destruction and re-formation.Compared to grasses, the comparatively more recalcitrant OM input from woody vegetation can promote lower microbial substrate use efficiency, which may have a negative effect on aggregate stability (Cotrufo et al., 2013).Furthermore, a combination of δ 13 C and 14 C measurements has shown that even very stable soil structures, such as iron nodules containing occluded carbon, do not act as closed systems with respect to organic carbon exchange (Bird et al., 1994).These observations together with the findings presented in this study, further suggest that carbon transfers within the soil matrix are highly dynamic, especially if impacted by recent shifts in vegetation type (Guidi et al., 2014).It is therefore highly likely that the different τ observed between the two sites is strongly influenced by vegetation-related factors, especially when considering that there are only minor differences in soil physical properties (i.e.texture, mineralogy -cf.Table 1 in Saiz et al., 2012), which are, by themselves, insufficient to explain such contrasting τ between the two sites.Therefore, making the reasonable assumption that both stands have been exposed to comparable deposition and erosion regimes; our results strongly suggest that both ecosystems are undergoing a rapid transition from open woodlands to denser canopy formations.However, such vegetation thickening varies in intensity at each site, and this exerts contrasting effects on their SOM dynamics (Fig. A1).

Conclusions
The first objective of this study was to assess the influence of C 3 and C 4 vegetation on SOM dynamics in semi-natural tropical ecosystems sampled along a precipitation gradient in West Africa.This work shows that the interdependence between biotic and abiotic factors ultimately determine whether SOM dynamics of C 3 -and C 4 -derived vegetation are at variance in ecosystems where both vegetation types coexist.Our results suggest that soil textural properties exert a strong influence on the contrasting SOM dynamics observed across the precipitation gradient.Accordingly, C 4 -derived SOM decomposes faster than SOM derived from woody biomass in mixed C 3 /C 4 ecosystems, provided comparable (site-scale) environmental conditions exist.This is in agreement with previous research conducted under controlled environmental conditions (Wynn and Bird, 2007).Moreover, studies conducted in West African savannas as well as pantropical productivity estimates both show that more than half of the net primary productivity of these ecosystems is attributable to C 4 grasses (Lloyd et al., 2008;Menaut and Cesar, 1979).This agrees well with the notion that, at least in non-coarse textured soils, SOC sequestration potential per unit productivity must be inherently lower for C 4 dominated locations, which is further confirmed by the comparatively lower SOC concentrations observed at grass-dominated locations (Saiz et al., 2012).
The second objective of our study was to investigate potential variations in tropical vegetation shifts along the precipitation transect.Vegetation thickening was significant at the more humid sites, in the zones of tension where forest and savanna formations coexist.Our findings reveal that current environmental conditions favour the expansion of C 3 species over their C 4 counterparts in the more mesic savanna ecosystems of West Africa.Such vegetation dynamics pose significant impacts not only on the total production and allocation of biomass, but also on its mineralization potential.
The third objective was to unambiguously evaluate the effect of vegetation thickening on SOM dynamics in two contiguous but structurally contrasting transitional ecosystems occurring on comparable soils.Such a setting was chosen to minimise confounding effects posed by climatic and edaphic factors as fundamental differences in δ 13 C fractionation dynamics have been reported for different soils and types of organic matter (Krull and Skjemstad, 2003).Radiocarbon dating of SOM fractions together with the vertical variation in δ 13 C values strongly suggest that both ecosystems are undergoing a rapid transition towards denser closed canopy formations.However, vegetation thickening varied in intensity at each site and exerted contrasting effects on SOM dynamics.This study further highlights the far-reaching implications that vegetation thickening has for the stability of deep SOC, which has been shown to be heavily controlled by fresh organic inputs (Fontaine et al., 2007).It also confirms that SOM pools that have been stabilized for centuries to millennia may be susceptible to abrupt change when soils cross pedogenic thresholds associated with rapid shifts in vegetation (Trumbore, 2009).
Our findings have significant implications for a lot of carbon cycle science that exploits the carbon isotopic difference of C 3 /C 4 photosynthetic pathways.This includes research work that uses the variation in soil δ 13 C values to unravel past vegetation shifts and their impact on SOC storage, paleoenvironmental interpretations, and modelling of ecosystem carbon budgets (Lloyd et al., 2008;Saiz et al., 2015;Wynn and Bird, 2007).

Figure 1 .
Figure 1.Geographical locations of the studied sites in West Africa.The map interface is adapted from Globalis, a software tool developed by the initiative of the United Nations Association of Norway.

Figure 2 .
Figure 2. Selected examples of ecosystem types and soil profiles occurring over the precipitation gradient.

Figure 3 .
Figure 3. (a) Stable carbon isotope composition of soil samples taken from the 0-0.05 m interval at different sampling locations (i.e.Grass, Tree, and Clumps of trees).Sites are ordered by decreasing latitude; (b) relationship between the weighted average stable carbon isotopic composition of all sample locations and both the fractional vegetation cover (FC) of all woody vegetation taller than 1.5 m, and the axylale vegetation (grass and herbs).The gradation intensities of the axylale symbols correspond to the relative contribution of C 4 species over the total axylale vegetation.Accordingly, white, grey and dark-grey symbols represent > 0.66, 0-66-0.33,and < 0.33 of that contribution respectively.Regressions have the form FC = a+b× (δ 13 C) and the regression coefficients (r 2 ) are 0.81 and 0.93 for the woody (solid line) and axylale (dotted line) vegetation respectively; p < 0.05 in both regressions.The pure grassland stand (BDA-03) was not included in the regressions.
Figure9shows δ 13 C values of bulk and fractionated SOM at different depths for two contrasting transitional ecosystems

Figure 5 .
Figure5.Relationship between δ 13 C and C / N values across the gradient at both "Tree" and "Grass" sampling locations for (a) bulk soil samples, and (b) the light fraction (LF).Solid and dotted regression lines denote "Tree" and "Grass" sampling locations respectively.In red are the Sahelian ecosystems (HOM sites), which have not been included in the regressions.All samples derive from the 0-0.3 m depth interval.The regressions have the form C / N = a + b× (δ 13 C) and the regression coefficients (r 2 ) are 0.3 and 0.37 for the Tree and Grass locations respectively for bulk soil samples (5a); while r 2 are 0.83 and 0.63 for the Tree and Grass locations respectively for LF samples (5b); p < 0.05 in all regressions.

Figure 6 .
Figure 6.(a) Relationship between the weighted average soil δ 13 C calculated for each site and the difference in the relative change in C / N values for two contrasting SOM fractions (see text for details) between -G and -T sampling locations ( C / N * G−T ).The size of each data point is proportional to sand content.Grey dots denote transitional sites (Saiz et al., 2012; Fig. 7).(b) Measured and predicted C / N * G−T as a function of weighted average δ 13 C and sand content (sc).The regression takes the form C / N* G−T = 0.441+ 0.011 (δ 13 C) -0.410 (sc); r 2 = 0.63, P < 0.05, n = 13.

Figure 7 .
Figure 7. Differences in soil δ 13 C values between both the 0.05-0.3and 0.3-0.5 m depth intervals and the topmost soil layer (0-0.05m) across the precipitation gradient.The dashed line represents mean annual precipitation (P A ). Sites are ordered by decreasing latitude.The shaded region contains the transitional sites.The asterisk at BDA-3 denotes that soil sampling was limited to 0.19 m only.

Figure 8 .
Figure 8. Variation in soil δ 13 C values with depth for selected ecosystems demonstrating distinctive vegetation dynamics across the precipitation gradient.

Figure 9 .
Figure 9. Stable carbon isotope composition of bulk and fractionated SOM, and mean residence times (τ ) for two transitional ecosystems in central Ghana; (a) savanna woodland (BFI-02), (b) semi-deciduous dry forest (BFI-04).Bulk soil is defined as < 2000 µm; < 53 µm is the fraction associated to silt and clay; HF is the heavy fraction > 53 µm (specific density > 1.87 g cm −3 ); LF is the light fraction > 53 µm (specific density < 1.87 g cm −3 ).τ of selected SOM fractions are shown adjacent to the fractions and denote conventional radiocarbon ages and, in between brackets, calculated τ derived from radiocarbon analyses presented in TableA2.

Figure A1 .
Figure A1.Comparison of 14 C data (pMC) and C / N ratios presented in Table 1 for the fractionated SOM samples.Circles indicate shallow samples (0.3-0.5 m) and triangles denote deep samples (1.5-2.0 m).< 53 µm is the fraction associated to silt and clay; HF is the heavy fraction > 53 µm (specific density > 1.87 g cm −3 ).The largest C / N value of each paired symbol consistently represents BFI-02.

Table 1 .
Soil properties determined for different depth intervals at both studied sites.Numbers in brackets denote standard deviation from the means (n = 5).Analyses were conducted on dry samples sieved to 2 mm.Abundance of root fragments < 0.02 m was assessed over the total dry mass of each individual sample prior to any sieving.b Soil fractions were obtained from a composite sample made up from all five individual samples at each depth interval.Bulk: composite sample subject to fractionation; s + c : fraction < 53 µm associated to silt and clay; HF: Heavy fraction > 53 µm and > 1.87 g cm −3 ; LF: Light fraction < 1.87 g cm −3 . a al.,

Table A2 .
Radiocarbon results(AINSE GraFink et al., 2004);Fink et al., 2004)and calculated mean residence times (τ ). 13 C values relate solely to the graphite derived from the fraction that was used for the radiocarbon measurement.Results from both percent modern carbon (pMC) and conventional radiocarbon age are reported with 1σ errors (±) attached. δ