Volume 18, Issue 5 pp. 1670-1683
Primary Research Article
Full Access

Variation in soil carbon stocks and their determinants across a precipitation gradient in West Africa

Gustavo Saiz

Corresponding Author

Gustavo Saiz

School of Geography & Geosciences, University of St Andrews, St Andrews, KY16 9AL Scotland, UK

School of Earth and Environmental Sciences, James Cook University, Cairns, QLD 4870 Australia

Correspondence: Gustavo Saiz, Institute of Meteorology and Climate Research, Karlsruhe Institute of Technology, Garmisch-Partenkirchen 82467, Germany, tel. + 49 8821 183 288, fax + 49 8821 183 294, e-mail: [email protected]Search for more papers by this author
Michael I. Bird

Michael I. Bird

School of Earth and Environmental Sciences, James Cook University, Cairns, QLD 4870 Australia

Search for more papers by this author
Tomas Domingues

Tomas Domingues

School of GeoSciences, University of Edinburgh, Drummond Street, Edinburgh, EH8 9XP Scotland, UK

Search for more papers by this author
Franziska Schrodt

Franziska Schrodt

Earth and Biosphere Institute, School of Geography, University of Leeds, LS2 9JT Leeds, UK

Search for more papers by this author
Michael Schwarz

Michael Schwarz

Earth and Biosphere Institute, School of Geography, University of Leeds, LS2 9JT Leeds, UK

Search for more papers by this author
Ted R. Feldpausch

Ted R. Feldpausch

Earth and Biosphere Institute, School of Geography, University of Leeds, LS2 9JT Leeds, UK

Search for more papers by this author
Elmar Veenendaal

Elmar Veenendaal

Nature Conservation and Plant Ecology Group, Wageningen University, Droevendaalsesteeg 3a, 6700 AA, Wageningen, The Netherlands

Search for more papers by this author
Gloria Djagbletey

Gloria Djagbletey

Forestry Research Institute of Ghana, P. O. Box UP 63 Knust, Kumasi, Ghana

Search for more papers by this author
Fidele Hien

Fidele Hien

Institut de l'Environnement et de Recherches Agricoles, 04 B.P. 8645, Ouagadougou, Burkina Faso

Search for more papers by this author
Halidou Compaore

Halidou Compaore

Institut de l'Environnement et de Recherches Agricoles, 04 B.P. 8645, Ouagadougou, Burkina Faso

Search for more papers by this author
Adama Diallo

Adama Diallo

Centre National des Semences Forestières, BP 2682, Ouagadougou, Burkina Faso

Search for more papers by this author
Jon Lloyd

Jon Lloyd

School of Earth and Environmental Sciences, James Cook University, Cairns, QLD 4870 Australia

Earth and Biosphere Institute, School of Geography, University of Leeds, LS2 9JT Leeds, UK

Search for more papers by this author
First published: 28 January 2012
Citations: 33

Abstract

We examine the influence of climate, soil properties and vegetation characteristics on soil organic carbon (SOC) along a transect of West African ecosystems sampled across a precipitation gradient on contrasting soil types stretching from Ghana (15°N) to Mali (7°N). Our findings derive from a total of 1108 soil cores sampled over 14 permanent plots. The observed pattern in SOC stocks reflects the very different climatic conditions and contrasting soil properties existing along the latitudinal transect. The combined effects of these factors strongly influence vegetation structure. SOC stocks in the first 2 m of soil ranged from 20 Mg C ha−1 for a Sahelian savanna in Mali to over 120 Mg C ha−1 for a transitional forest in Ghana. The degree of interdependence between soil bulk density (SBD) and soil properties is highlighted by the strong negative relationships observed between SBD and SOC (r> 0.84). A simple predictive function capable of encompassing the effect of climate, soil properties and vegetation type on SOC stocks showed that available water and sand content taken together could explain 0.84 and 0.86 of the total variability in SOC stocks observed to 0.3 and 1.0 m depth respectively. Used in combination with a suitable climatic parameter, sand content is a good predictor of SOC stored in highly weathered dry tropical ecosystems with arguably less confounding effects than provided by clay content. There was an increased contribution of resistant SOC to the total SOC pool for lower rainfall soils, this likely being the result of more frequent fire events in the grassier savannas of the more arid regions. This work provides new insights into the mechanisms determining the distribution of carbon storage in tropical soils and should contribute significantly to the development of robust predictive models of biogeochemical cycling and vegetation dynamics in tropical regions.

Introduction

Soils constitute a major reservoir of carbon at the global scale with about a third of the total SOC (TSOC) stored in tropical regions (Schimel et al., 1994; Batjes, 1996; Davidson et al., 2000; Amundson, 2001). Although at the global scale the distribution of tropical vegetation and SOC decomposition rates are controlled by climate, edaphic and biotic factors play a fundamental role in affecting the quantity and quality of carbon inputs and decomposition processes at the local scale (Feller & Beare, 1997; Giardina & Ryan, 2000; Wynn et al., 2006; Wynn & Bird, 2007) with the SOC inventory in any soil profile being determined by the complex interplay of many factors including climate, soil texture, land use, fire frequency and topography (Bird et al., 2001). Soil texture is of paramount importance in controlling SOC storage and strongly influences nutrient availability and water retention, particularly in highly weathered soils (Tiessen et al., 1994; Silver et al., 2000). In particular, soils with a higher clay content tend to have larger SOC concentrations due to the formation of passive carbon pools via the adsorption and aggregation of soil organic matter (SOM) by clay minerals (Schimel et al., 1994; Feller & Beare, 1997).

Environmental gradients varying systematically in climate or other variables provide an excellent opportunity for both understanding mechanisms of abiotic control on ecosystem processes and study the potential impacts of global change in these ecosystems (Koch et al., 1995). The ideal conditions provided by simplified environmental gradients to maximize the interpretation of results are not, however, present in West Africa because of the heterogeneity of soils and the differing frequency of disturbance events such as fire which naturally occur in these ecosystems (Pullan, 1969; White, 1983; Bird et al., 2000). However, it is still possible to extract very valuable information about the variation in SOC stocks with a number of soil-forming factors in such circumstances (Wynn et al., 2006).

The strong climatic gradient existing between the humid environments prevalent near the coast and the arid conditions found in interior continental areas exert a fundamental control on the distribution of vegetation in West Africa (White, 1983; Marks et al., 2009) with the heterogeneity of soils also reflecting the large natural diversity of this vast region. Indeed, studies conducted decades ago identified the difficulty of assigning dominant pedogenetic features to any given climate within West Africa (Thomas, 1966; Pullan, 1969). The main reason for this is the lack of an unequivocal effect of parent material as a soil-forming factor, given that West African soils have profiles that may have developed in material of mixed origin and some soils may also have been preweathered to a considerable depth. Therefore, the distribution of soils in West Africa cannot be related directly to present climate and vegetation, except perhaps for the nutrient poor soils of semiarid regions (Pullan, 1969).

A further confounding factor is that the average SOC density in West Africa is significantly lower than the global average as a result of its less favourable agroecological conditions and significant land degradation caused by human activities. Thus, the semi-arid and subhumid regions of Africa have been reported by several studies as having the largest potential for carbon sequestration in the World (Batjes, 2001; Marks et al., 2009). Many of the soils of these regions are already severely degraded and may be susceptible to further losses with climate change and increasing human pressure on the soil resource (Marks et al., 2009). Therefore, there is a clear need for observational data sets of SOC stocks and distribution, to allow for the development and validation of robust biogeochemical cycling models (Bird et al., 2001; Wynn et al., 2006). Furthermore, while we have moderate understanding about the susceptibility of SOC to degradation in tropical environments, much less is known about how the SOC fraction resistant to decomposition varies along natural precipitation gradients (Cheng et al., 2008; Lehmann et al., 2008). The assessment of this variation can provide valuable insights into the mechanisms leading to effective carbon sequestration and the likely effects of climate change on SOC pools.

The objectives of the present study are: 1. The estimation of SOC stocks at representative sites over a wide range of relatively undisturbed ecosystems characteristic of West Africa, 2. determination of relative roles of climatic, edaphic and biotic factors in determining SOC stocks and 3. evaluation of variations in the proportion of the SOC pool resistant to decomposition across the precipitation transect.

Materials and methods

Description of the West African transect- characteristics of the sites

This study was conducted along a latitudinal transect (15°–7°N) spanning nearly 1000 km and encompassing a broad array of ecosystems and soil types characteristic of West Africa (Fig. 1a). Measurements were undertaken from August to October 2006 in Ghana, Burkina Faso and Mali. Fourteen study sites, consisting of ten 1 ha plots together with four 0.5 ha plots (BFI sites), were established in areas previously identified as representative of the potential natural vegetation of the region (Table 1). Specific locations were selected because they had some degree of protection from direct human intervention, although fire was not excluded from any of the plots. These included National Parks, Forest Reserves and other legally protected areas except for the case of the Sahelian sites in Mali, which had no specific conservation status and were subject to varying degrees of grazing pressure, the latter being also the case for the most northern Sudan savanna sites of Burkina Faso.

Details are in the caption following the image
(a) Regional distribution of vegetation and (b) mean annual precipitation in West Africa. Vegetation zones adapted from White (1983). Climatic data sourced from the Climate Research Unit (CRU) - University of East Anglia, Norwich (UK), graphic interface adapted from Globalis, a software tool developed by the initiative of the United Nations Association of Norway.
Table 1. Characteristics of the sites
Site Regional classification of vegetation Canopy cover Soil type WRB Textural class FAO (USDA) Clay content kg kg−1 Sand content kg kg−1 pH ECEC mmol kg−1 N mg g−1 P mg g−1 Fe mg g−1 Al mg g−1
HOM-1 Open Sudan savanna (Sahel) 0.01 Haplic Arenosol Coarse (Sandy) 0.03 0.89 6.4 9.6 (1.7) 0.11 0.05 7.0 9.5
HOM-2 Open Sudan savanna (Sahel) 0.05 Haplic Arenosol Coarse (Sandy) 0.01 0.93 6.7 11.9 (3.8) 0.12 0.05 4.9 9.0
BBI-1 Open Sudan savanna 0.28 Haplic Luvisol Medium (Clay Loam) 0.39 0.31 5.8 39.4 (5.0) 0.30 0.08 23.8 52.9
BBI-2 Open Sudan savanna 0.51 Pisolithic Plinthosol Medium (Loam) 0.18 0.49 6.1 27.6 (3.4) 0.43 0.11 22.4 55.0
BDA-1 Open Savanna woodland 0.16 Haplic Fluvisol Medium Fine (Silty loam) 0.25 0.11 5.8 41.0 (10.7) 0.60 0.11 44.1 40.2
BDA-2 Open Savanna woodland 0.03 Acric Stagnic Plinthosol Medium (Silty loam) 0.1 0.39 5.6 26.3 (12.7) 0.32 0.07 45.5 24.3
BDA-3 Open Savanna grassland 0.00 Epipetric Stagnic Plinthosol N/A 5.6 8.7 (6.4) 0.68 0.17 65.0 30.2
MLE-1 Open Savanna woodland Guinea 0.24 Brunic Arenosol Coarse (Loamy sand) 0.04 0.81 6.1 16.8 (5.2) 0.20 0.05 7.0 11.6
BFI-1

Savanna woodland

Transition Zone

0.30 Haplic Alisol Coarse (Sandy loam) 0.11 0.72 7.0 26.5 (5.4) 0.70 0.13 16.1 23.3
BFI-2

Savanna woodland

Transition Zone

0.60 Brunic Arenosol Coarse (Sandy loam) 0.09 0.71 5.3 12.2 (5.7) 0.67 0.12 15.4 22.8
BFI-3

Semideciduous dry forest

Transition Zone

0.72 Haplic Nitosol Medium (Sandy clay loam) 0.2 0.61 5.7 38.0 (5.3) 1.40 0.19 22.4 41.8
BFI-4

Semideciduous dry forest

Transition Zone

0.80 Haplic Nitosol Medium (Sandy Loam) 0.05 0.65 6.7 31.8 (8.8) 1.42 0.24 14.7 30.7
KOG-1 Savanna woodland Transition Zone 0.42 Haplic Arenosol Coarse (Loamy sand) 0.03 0.77 5.3 9.1 (6.3) 0.24 0.04 2.8 4.2
ASU-1 Semideciduous dry forest 0.50

Endofluvic

Cambisol

Medium (Loam) 0.17 0.43 4.9 29.9 (8.8) 1.31 0.15 18.2 27.0
  • All soil-related values are based on the 0.0–0.30 m interval, except for BDA-3, which had a 0.19 m average maximum depth. For the regional classification of vegetation and calculation of canopy covers, the reader is referred to Domingues et al. (2010). Numbers in brackets within the ECEC column are standard deviations from the mean (n = 5).

The sites were set out over consistently flat terrain with less than 100 m of altitudinal variation between them. The northern point of the transect was dominated by Sahelian ecosystems, a very open savanna occurring on nutrient poor arenosols in the Southern border of the Sahara desert (Fig. 1a, Table 1). This ecosystem is characterized by low rainfall with a mean annual precipitation (Pa) typically 200–400 mm a−1, and high rates of potential evaporation. Further south, there is a natural progression into more tree-dominated forms of vegetation heavily influenced by the gradual increase in mean Pa (Fig. 1b). The Southern end of transect corresponds to the wettest sites studied (>1200 mm a−1) supporting semi-deciduous tropical forest. The variation in mean annual temperature was less than 4 °C across the transect. Climatic data for each site were extracted from the WorldClim database with 1 km2 spatial resolution (Hijmans et al., 2005).

We made use of a single climatic index effectively used in large scale environmental studies (Berry & Roderick, 2002; Wynn et al., 2006). This annual water availability index (W* in mm a−1) is the difference between Pa and the mean annual amount of water that would be evaporated if all of the global solar radiation received at the surface was used to evaporate water. The index was calculated as:
urn:x-wiley:13541013:media:gcb2657:gcb2657-math-0001(1)
where Pa is mean annual precipitation rate, Qs is mean annual global solar radiation in J m−2 a−1, ρw is the density of liquid water (~1000 kg m−3 at 25 °C) and L is the latent heat of evaporation of water (~2.5 × 10J kg−1 H2O at 25 °C). This formulation provides an index of water availability to plants, although it does not take into account runoff, surface albedo, and longwave radiation fluxes into and away from the surface.

The wide range of soil types reported in Table 1 reflects contrasts in geology, climate and vegetation integrated over extended time periods. Soils were classified according to the World Reference Base (WRB) (IUSS (International Union of Soil Science) Working Group WRB, 2006). For a detailed description of the sites, see Domingues et al. (2010).

Soil sampling

To overcome the heterogeneity in amount and stable isotopic signature of SOM in mixed C3/C4 environments (where grasses of the C4 photosynthetic pathway coexist with trees and shrubs of the C3 photosynthetic pathway), we collected soil samples following a stratified random sampling strategy that has proved well-suited to such environments (Bird et al., 2004; Wynn et al., 2006). This 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) to best account for the inherent heterogeneity of SOC characteristic of these ecosystems. At each of these locations, surface litter was removed when present and three samples at 0–0.05 m and one sample at 0–0.30 m were taken with the aid of a stainless steel corer 40 mm inner diameter (ø) before being placed in labelled zip-lock bags. This procedure was replicated five times at each site (both for T and G locations). Replicates were subsequently bulked according to location (T vs. G) and depth (0–0.05 m and 0–0.30 m) as this procedure has been shown to be a cost-effective technique for smoothing out local heterogeneity and for achieving robust regional estimates of SOC inventories (Bird et al., 2004; Wynn et al., 2006). At some sites, the individual samples were independently analysed and compared against results obtained pooling these samples, which further confirmed the soundness of the bulking procedure. The number of samples collected using this stratified sampling design totalled 560.

Using the procedures detailed in Quesada et al. (2010), deep soil augering was also carried out within each plot in the near vicinity of 5 of the locations described above and samples taken at 0–0.05 m, 0.05–0.10 m, 0.10–0.20 m, 0.20–0.30 m, 0.30–0.50 m and then every 0.5 m up to 2 m depth (impenetrable layers permitting). These samples were used for determinations of pH, cation exchange capacity (CEC) and elemental abundance of carbon and nitrogen. The total number of samples collected by deep augering was 434. For each plot, a soil pit was hand-dug to 2 m to assess soil type and describe soil characteristics (Quesada et al., 2011). In this exposed soil profile, samples were also collected at the same depth intervals as above to allow for analyses of soil colour, consistency, particle size distribution and bulk density. The latter was achieved by using specifically designed rings (ø = 80 mm). The number of samples collected at the soil pits totalled 112, not including samples specifically collected for soil bulk density determination. Samples have been archived at the University of St Andrews (Scotland, UK).

Sample preparation and bulk density determinations

Samples collected at 0–0.05 and 0–0.30 m using the steel corer were weighed in their sealed bags, clumps broken by hand and then oven dried at 40 °C to constant weight. An aliquot of these samples was then oven dried at 105 °C for 4 h which allowed for the calculation of soil bulk density (SBD). Samples were then dry sieved to 2 mm and gravel and root content >2 mm was determined by weight. The set of samples specifically collected in the soil pit for determination of bulk density was also dried at 105 °C. In all cases, calculation of SBD included fractions >2 mm. The impact of including gravel and roots >2 mm on the calculation of SOC stocks is dealt with separately in this work. Please refer to Supplementary Information.

Analytical methods

Soil pH was measured using a digital pH meter in a 2 : 1 water : soil solution. Particle size distribution was determined gravimetrically as described by van Reeuwijk (2002). Briefly, 10 g of soil dry sieved to 2 mm was first treated with a chemical dispersant (sodium hexametaphosphate), and then physically separated by sieving into sand (particle sizes between 0.05 and 2 mm), with the material passing through the sieve being placed in 1 L water solution and allowed to settle for 4 h at 20 °C to subsequently determine clay content (particle sizes < 0.002 mm). Silt (particle sizes between 0.002 and 0.05 mm) was obtained by mass balance from the recorded dry weights.

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). Effective cation exchange capacity (ECEC) was calculated as the sum of these bases. Total phosphorous concentration was determined by ICP-OES (Perkin Elmer 5300DV, Wellesley, MA, USA) on extracts obtained by acid digestion as described in Tiessen & Moir (1993).

Quantification of Fe and Al was performed by X-ray fluorescence (XRF) using a Spectro XLAB EDPXRF spectrometer (SPECTRO Analytical Instruments GmbH, Kleve, Germany) equipped with a Rh anode X-ray tube, with the mineralogy of the Fe and Al oxides determined using X-ray diffractometry (XRD) PW1050 (Philips Analytical, Netherlands) attached to a X-ray generator DG2 (Hiltonbrooks Ltd, Crewe, UK) at the University of St. Andrews (Scotland). Interpretation and semi-quantitative analysis of the scans were achieved using the Rietveld refinement method built-in within the Siroquant software (SIROQUANT; Sietronics Pty Ltd, Canberra, Australia).

Sample aliquots were pre-treated with 6N HCl and subsequently analysed for variation in the C content, which confirmed the absence of inorganic carbon. Elemental carbon and nitrogen were determined in duplicate using a Costech Elemental Analyzer (Costech International S.p.A., Milano, Italy) fitted with a zero-blank auto-sampler coupled via a ConFloIII to a ThermoFinnigan DeltaPlus-XL (Thermo Scientific, Waltham, MA, USA) using Continuous-Flow Isotope Ratio Mass Spectrometry at the University of St Andrews Facility for Earth and Environmental Analysis stable isotope laboratory. Precisions (SD) on internal standards for elemental carbon and nitrogen abundances were better than ± 0.09% and 0.02% respectively.

We used a modified version of the technique used by Bird & Grocke (1997) to isolate resistant SOC (RSOC). In short, 50 mL of solution made up of 0.1 M K2Cr2O7 and 2 M H2SO4 was added to 500 mg soil samples in centrifuge tubes. They were then capped and heated to 60 °C in a temperature-controlled orbital shaker for 72 h. The tubes were periodically uncapped to release evolved gases. At the end of the incubation, all samples were washed by centrifugation with distilled water thrice and then oven-dried at 60 °C. The determination of elemental RSOC was similar to the method for TSOC.

Results

Soil characteristics

Soils were moderately acid to neutral (4.9 ≤ pH ≤ 7.0) and ranged from medium to coarse texture, with a wide range of fertilities as evidenced by the ECEC varying from 9 to 41 mmol kg−1, [N] ranging from 0.1 to 1.4 mg g−1 and [P] ranging from 0.04 to 0.24 mg g−1 (Table 1).

The SBD increased with latitude along the transect (Fig. 2), but with the magnitude of this change depending on soil depth. Northernmost sites had SBD values exceeding 1500 kg m−3 at the soil surface (0.0–0.05 m), which more than doubled those observed in forests at the transect southern end. Reflecting the depth-dependent latitudinal gradient, the more arid ecosystems towards the north (Sudan/Sahelian savannas) had higher SBD in the first 0.05 m as compared to 0.30 m, while deeper soil samples towards the southern end of the transect had higher SBD than their shallower counterparts.

Details are in the caption following the image
Soil bulk density (kg m−3) ordered by decreasing latitude. T and G represent ‘Tree’ and ‘Grass’ locations respectively. Error bars are standard deviations of the means. Asterisk denotes BDA-3 for which the average soil depth was only 0.19 m.

The SBD and SOC showed strong negative relationships across the precipitation transect at both 0.0–0.05 and 0.0–0.3 m intervals. These relationships were best classified on the basis of soil texture (see Discussion). Regressions indicated different decreasing patterns in SBD with increasing SOC for each soil depth. While similar rates of change were observed within each depth interval for both textural classes (medium and coarse), analyses of covariance showed these regressions to be significantly different (P < 0.05; Fig. 3).

Details are in the caption following the image
Relationship between SBD and SOC content for (a) 0.05 m and (b) 0.30 m intervals. Closed and open circles correspond to sites with medium and coarse soil textural classes respectively, classified according to the FAO textural classes shown in Table 1. Each measured SBD is the mean of five measurements per sampling location. Error bars are standard deviations of the means. Regressions are significant at P < 0.05 level. Analyses of covariance (ancova) were performed at each depth interval to test for significant different differences between regressions (in both cases, P < 0.05). BDA-3 is not included in the regression for 0.3 m as its average soil depth was only 0.19 m.

The mineralogical composition as derived from XRD analyses revealed the strongly weathered characteristics of the soils as suggested by the large presence of quartz and kaolinite (Table 2). The semi-quantitative XRD analyses provided mineralogical abundances that correlated well with elemental contents obtained from XRF analyses. The imbalance between quantitative XRD and XRF analyses may indicate a significant proportion of amorphous or poorly crystalline material in some of the samples, especially in sites with abundant clay and iron contents (BBI and BDA sites) (Table 2).

Table 2. Regression values for the function predicting TSOC using Water availability index-W* (mm a−1, x) and sand content (kg kg−1, y) at two different depths
0.30 m 1.00 m
Depth n = 13 r2 0.84 n = 12 r2 0.86
f = yo+a*x+b*y yo a b yo a b
Coefficient 16.063 0.010 −27.056 57.918 0.019 −72.901
St Error coeff 6.410 0.002 5.703 16.694 0.006 14.281
t 2.506 4.721 −4.744 3.469 3.273 −5.105
P value 0.031 0.001 0.001 0.007 0.010 0.001

Soil organic carbon

There was a large contrast in SOC stocks between sites from the Northern and Southern end of the studied transect (Fig. 4). Sahelian ecosystems contained less than 11 Mg C ha−1 in the first 0.3 m of the soil, while transitional dry forests growing in much less water-limited sites stored up to four times that amount. Greater carbon stocks and concentrations were found in the vicinity of trees at all sites compared with those observed in locations away from tree stems ‘grass’ (Fig. 4; Table S1). SOC stocks in the first 2 m of the soil ranged from 20 to over 120 Mg C ha−1 for a Sahelian and a transitional forest respectively. Overall, the average SOC contained in the first 0.05 m of the soil accounted for roughly 0.3 of that stored in the first 0.3 m. Similarly, the upper 0.3 m of the soil accounted for about 0.3 of the total 2-m OC inventory (Table S2). The inclusion of gravel and roots > 2 mm had a relatively low impact on the calculation of SOC stocks (Table S3).

Details are in the caption following the image
Soil carbon stocks (Mg C ha−1) at contrasting locations ordered by decreasing latitude. Stippled columns correspond to sampling conducted in clumps of trees (see text). Asterisk at BDA-3 denotes that soil sampling was limited to 0.19 m only.

Although there was an overall discernible increasing pattern of SOC stocks with Pa (Fig. 4), simple regressions based on either Pa or mean annual temperature could account for only a maximum of 0.52 of the variability in SOC at any depth, with these fits slightly improving when the W* of Eq. 1 was used as a predictor variable (analyses not shown). We did, however, obtain a much better fit when sand content was included along with W* as predictors of SOC. Taken together, these variables explained 0.84 and 0.86 of the total variability in SOC stocks observed to 0.3 and 1.0 m depths respectively (Table 3, Fig. 5).

Details are in the caption following the image
Predicted and measured SOC stocks (Mg C ha−1) at 0–0.3 m (n = 13) and 0–1.0 m (n = 12) across the latitudinal transect.
Table 3. Regression values for the functions predicting TSOC using Water availability index-W* (mm a−1), and either a) sand or b) clay content (kg kg−1) respectively at two different depths. Presented coefficients, associated standard errors and significance values make reference to regressions showing the highest fit where both variables were used. Cumulative r2 values from stepwise regressions analyses are shown for reference. () Indicate the highest regression fit using a single variable; Note that in the case of sand-driven equations, sand alone was a better predictor than W*. Conversely, in the case of clay-driven equations, W* alone was a better predictor than clay.
Depth 0.30 m 1.00 m
a) Sand n = 13 P < 0.0001 n = 12 P < 0.0001
Cumulative r2 0.49 0.84 0.71 0.86
f = yo + a*sand + b*(W*) yo a b yo a b
Coefficient 16.063 −27.056 0.010 57.918 −72.901 0.019
SE coeff 6.410 5.703 0.002 16.694 14.281 0.006
t 2.506 −4.744 4.721 3.469 −5.105 3.273
P value  0.031  0.001 0.001  0.007  0.001 0.010
b) Clay n = 13 P = 0.0025 n = 12 P = 0.0114
Cumulative r2 0.49 0.70 0.50 0.63
f = yo + a*(W*) + b*clay yo a b yo a b
Coefficient −8.971 0.012 45.779 −12.679 0.0268 65.417
SE coeff 7.008 0.003 17.363 19.278 0.009 36.807
t −1.280 3.920 2.637 −0.658 3.026 1.777
P value 0.229 0.003 0.025 0.527 0.014 0.109

The relative contribution of RSOC to the TSOC pool declined with increasing precipitation as is shown in Fig 6. Savanna sites showed fractional RSOC contributions larger than 0.18 of the TSOC, while dry forests consistently had lower contributions of RSOC of typically less than 0.1 Absolute RSOC values were lower than 7 Mg C ha−1 except for one savanna site with 15 Mg C ha−1 (BDA-2), which seemed to have a very intense burning regime (see Discussion).

Details are in the caption following the image
Relative contribution of Resistant SOC (RSOC) to Total SOC (TSOC) pool.

Discussion

Soil bulk density variation along the transect

Calculation of accurate SOC stocks at any given depth relies on the acquisition of SBD and SOC concentration. The former variable encompasses the whole soil fraction, while the latter is usually reported for soil constituents smaller than 2 mm (refer to Supplementary Information for a critical assessment about the inclusion of gravel and roots in SOC stocks). These variables are interdependent to some degree as it is shown by the latitudinal gradient in SBD, which is primarily the result of the interplay between SOC contents and soil properties at each site (Figs 2, 3). Transitional forests at the Southern end of the transect had the lowest values in SBD as a result of having larger content of silts and clays, and the highest SOC values (Table 1, Fig. 3, Table S1 and S2). The significance of SOC contents, soil textures and mineral compositions determining SBD along the precipitation transect is highlighted by the fact that large SBD values of the relatively carbon-poor Northern sites were not just exclusive to markedly sandy Sahelian ecosystems (i.e. HOM sites), but also occurred in relatively dry savanna sites with more loamy textures (i.e. BBI and BDA sites) (Table 1). However, the reason behind the relatively high SBD values observed in the savannas with noticeably finer soil textures is very different.

The soils of these Sudan savannas are characterized by the relatively large content of iron and aluminium oxides with net positive surface charges, which have the capacity to form surface coatings on negatively charged clay minerals (Cornell & Schwertmann, 1996; Hien et al., 2006). These coated clay particles are cemented by iron, forming sand-sized microaggregates called sesquioxides or pseudo-sands, which feel coarse-textured and result in higher SBD values than those observed in sandier soils at comparable SOC contents (Fig. 3). The strong negative relationships observed between SBD and SOC along the precipitation transect were best classified based on the soil texture, a physical property heavily influenced by soil mineral composition.

The relative low amount of OC present in the top soil layer of the most arid ecosystems in this study (Sudan/Sahelian savannas) has direct implications for soil structure, and hence the higher bulk density observed at the top layer of these arid sites (Figs 2, 3), a characteristic to which cattle trampling may have also contributed. In contrast, the larger presence of clay particles at depth promoted particle aggregation and consequently the increase in overall pore space, which agrees well with the decrease in SBD observed at these arid sites. However, this pattern in SBD is reversed towards the Southern end of the transect, where deeper soils have higher bulk densities than surface soils (Figs 2, 3).

Soil organic carbon stocks along the transect

The range of West African environments studied here show a moderate capacity to store large amounts of organic carbon in the soil with the exception of Sahelian ecosystems and savanna sites growing in markedly sandy soils (sandy and loamy sands textural classes; Table 1). Indeed, where sampling to 1.0 m was possible, sites with finer soil textures invariably showed SOC values exceeding 60 Mg C ha−1 (Table S2), making the soil a much larger carbon reservoir than that of biomass as was also reported by Grace et al. (2006) in savanna environments. The SOC stocks observed at different ecosystems agree relatively well with results compiled by Post et al. (1982) who reported an average of 20, 54, 61 and 99 Mg C ha−1 at 1 m depth for tropical desert bush, tropical woodland-savanna, very dry and dry tropical forests respectively. More specifically, individual studies conducted across West Africa reflect the large degree of variability in SOC stocks for the different types of ecosystems. A study by Woomer et al. (2004a) reports a range of 11–25 Mg C ha−1 at 0.4 m in the Sahelian transition zone of Senegal, whereas Roose & Bathès (2001) observed a range of 15–46 Mg C ha−1 to 0.3 m over a rainfall gradient encompassing a Sudano-Sahelian savanna in Burkina Faso and a subequatorial forest in Ivory Coast. Moreover, Batjes (2001) calculated an average of 42–45 Mg C ha−1 for West Africa at 1 m depth making use of a global soil database. These quantities are somewhat lower than the results shown in our work (Table S2); however, their results include agricultural and more degraded biomes. The loss of SOC by conversion of natural vegetation to agricultural use is widely reported in the literature (Post & Kwon, 2000), and this region is by no means an exception to this trend (Roose & Bathès, 2001; Hien et al., 2006). Besides, West Africa is severely affected by other important factors contributing to the decline in SOC stocks, including decreases in soil fertility as a result of agricultural mismanagement and overgrazing, persistent droughts and soil erosion (Batjes, 2001; Tschakert et al., 2004; Marks et al., 2009). Work conducted by Woomer et al. (2004b) in Senegal estimated an average annual carbon loss of approximately 0.07% from the first 0.4 m of soil for the period between 1965 and 2000 after adjusting for land use/cover change and the depletion of woody biomass.

At the plot scale, the consistently greater SOC stocks observed in locations directly influenced by the presence of trees justify the use of a stratified sampling design in mixed C3/C4 environments, and may reflect both the larger content of OM inputs occurring at tree locations and/or lower decomposition rates of C3-derived material relative to C4-derived material (Wynn & Bird, 2007). Furthermore, this aspect is further highlighted by the fact that the largest SOC stock at 0.3 m observed over the entire dataset corresponded to locations that were purposely sampled because of the noticeable presence of clumps of trees growing on abandoned termite mounds (Fig. 4). Similarly, in a study conducted on a humid savanna, Mordelet et al. (1993) reported large SOC concentrations in tree clumps due to greater organic matter input beneath tree canopies.

Although there was an overall increase in SOC stocks with increasing precipitation (Fig. 4), this trend was heavily influenced by the different soil types existing along the transect (Table 1). Indeed, simple regressions based on either Pa or mean annual temperature accounted for 0.52 of the variability in SOC at any depth (analyses not shown). These results are comparable to other studies that have used linear relationships driven by single climatic or soil variables to explain SOC stocks across extensive tropical semi-arid regions (Jones, 1973; Bird et al., 2004; Wynn et al., 2006). The relatively low explanatory power of these functions, which typically explain less than 0.50 of the total variation, suggests a complex interplay of multiple factors driving the storage dynamics of SOC.

It is well established that fine textured soils have a strong effect on SOC decomposition processes by physically protecting SOM, which increases both SOC content and carbon residence time (Schimel et al., 1994; Silver et al., 2000). In this study, sandier soils (sandy and loamy sand textures) consistently had the lowest SOC stocks regardless of climate because of the low nutrient and water retention capacity as well as the poor structural characteristic of these soils while the opposite was true for soils with finer textural classes (Table 1; Fig. 4).

In addition to the influence of climate and soil properties on SOC stocks, the type of vegetation existing at a given location has a strong effect on the amount and quality of organic inputs returning to the soil, thus greatly influencing its carbon storage potential (Post et al., 1982). Within this context, it is worth considering the role of soil fertility in determining the type of vegetation across the transect. Overall, the soils studied here presented relatively low ECEC rates commonly observed in strongly weathered tropical ecosystems (Marques et al., 2004). Indeed ECEC, defined as the sum of the exchangeable cations that a soil can adsorb, is an important chemical property commonly used for assessing soil fertility (Brady & Weil, 2002; Sankaran et al., 2005). Nonetheless, the role of available soil N and P should also have important implications for plant productivity and ecosystem functioning (Domingues et al., 2010; Quesada et al., 2010). The irregular pattern in ECEC rates observed along the transect, with some Sudan savannas showing the highest values (Table 1), demonstrates that soil fertility was not the main limiting factor driving the type of vegetation occurring at a given location, except perhaps for the forests existing in the wetter end of transect where we observed larger concentrations of soil N and P compared with savannas. Therefore, precipitation, or rather the amount of water available for plant growth, may be the main factor influencing the type of vegetation occurring along the transect, which has direct implications for site productivity, and consequently for soil structure, SBD, and SOC storage. A good example of the complex interplay of factors determining the impact that soils have on vegetation and their capacity to store carbon is offered by the Sudan savannas studied in Burkina Faso. These sites had comparatively large SOC contents because of the physical protection provided by the nature of their soils (Tables 1, 2). The presence of sesquioxides greatly affected soil structure and consequently their water retention capacity, as these particles promote large inter-aggregate pores capable of draining water at the same soil water potentials as sands of comparable size, whereas the intra-aggregate micropores hold water at very high tensions (Santos et al., 1989). The consequence of this is that less water will be available for plant growth, given that these soils have relatively few pores in the size range that contains water accessible for plants (Nitzsche et al., 2008), which undoubtedly influence the type of vegetation that can be sustained. While plant-available water is the main determinant for vegetation type along this precipitation transect, there are factors other than soil fertility which may influence this distribution. These factors include both natural and anthropogenic disturbances (fire, grazing pressure) and soil physical constraints on plant growth. Such is the case of BDA-3, a grassland site growing on hardened plinthite crust occurring less than 0.2 m from the surface (Table 1). In such circumstances, root penetration by woody plants is strongly diminished and only herbaceous vegetation may develop.

Functions predicting soil organic carbon stocks

In view of the contrasting soil characteristics observed over the wide range of vegetation existing across the precipitation gradient, we established a relatively simple predictive function to predict SOC stocks at different depths driven by a combination of climate and quantifiable soil variables capable of effectively encompassing the effect of climate, soil properties and vegetation type on SOC storage.

A large number of studies have reported strong correlation between SOC and clay contents, and indeed most process-based models simulating SOM dynamics make use of this relationship as the role of clays in soil physiochemical processes is fundamental (Spain, 1990; Schimel et al., 1994; Sollins et al., 1996; Feller & Beare, 1997). However, it is difficult to find unequivocal evidence on the role of clays stabilizing SOC, given that clay may be correlated with other factors and it is not clear which ones are causative (Oades, 1988). Moreover, there are also studies that have found weak correlations between SOC and clay content in contrasting ecosystems (Percival et al., 2000; Silver et al., 2000; Bricklemyer et al., 2007). The effect of clay on SOC stocks is also dependent on the clay mineralogy of the soil (Spain, 1990; Bruun et al., 2010). Therefore, caution should be exercised when generalizing about the role played by clay content in SOC stabilization.

Most tropical systems, with the exception of those occurring in mountainous regions, wetlands or recent volcanic deposits, usually contain soils that have undergone significant heavy weathering for prolonged periods of time. Consequently, these soils contain minerals highly resistant to weathering (Table 2). As discussed above, the association of clays with aluminium and iron oxides may result in the formation of sesquioxides in certain tropical soils conferring the soil a sand-like texture, which strongly affects its water retention capacity and result in the unusual high SBD values observed for medium-textured soils (Fig. 3). Thus, these particles may exert a strong influence on these two soil properties, which are essential factors in determining SOC stocks. However, a large proportion of the constituents of sesquioxides will be accounted for as clay fraction in laboratory analyses, which may limit to some extent the predictive power of regressions driven by clay content (Table 3). In this study, a function combining sand content and available W* explained 0.84 and 0.86 of the total variability in SOC stocks observed at 0.0–0.3 and 0.0–1.0 m respectively (Table 3, Fig. 5). Used in combination with a suitable climatic parameter, sand content was a good predictor of SOC stored in highly weathered dry tropical ecosystems with arguably less confounding effects than that provided by clay content.

Resistant soil organic carbon variation along the transect

The soil sampling strategy we used in this study allowed for the comparison of soil properties at systematically defined locations. The fact that we used ‘Grass’ locations in the assessment of RSOC had a double advantage; on the one hand, it excludes the possibility of confounding effects derived from any preferential sampling of woody biomass, while on the other hand, sampling at a mid distance from trees allows for the relative unbiased account of the effect of contrasting woody covers. However, failure to determine RSOC at ‘Tree’ locations may result in an underestimation of its overall absolute value for a particular site.

Several studies using strong acid treatments to isolate RSOC have shown that poorly crystalline and amorphous mineral components are left relatively untouched, (Kleber et al., 2005; Siregar et al., 2005; Mikutta et al., 2006), thus supporting the idea that the most important determinants controlling mineral-associated RSOC in heavily weathered soil systems may not be significantly affected by the use of acid treatments. The analytical procedure we chose to isolate the RSOC fraction has already been used as a proxy for pyrogenic carbon (Bird & Grocke, 1997), although we purposefully chose not remove the mineral component by HF dissolution prior to oxidation to include OC protected by mineral associations, given that dissolution of mineral phases previous to oxidation with treatments like HF hydrolysis has been shown to release significant amounts of OC from soils containing large contents of mineral-bound OM (Kaiser et al., 2002; Gonçalves et al., 2003).

The physicochemical protection of SOC conferred by soil minerals may be an important factor contributing to RSOC; indeed, the presence of sesquioxides promoting stable aggregates may have contributed to the relatively large RSOC contribution observed in the savannas of Burkina Faso (Tables 1, 2; Fig. 6). It has been reported that the lability of SOC in tropical ecosystems when compared across contrasting soils types is significantly influenced by clay mineralogy and content of Fe and Al (hydr-) oxides, but not by clay content (Bruun et al., 2010). On the other hand, Plante et al. (2006) showed in a study conducted over two widely ranged textural gradients that biochemically protected OC in whole soil samples increased with clay content. Even though the range of textures covered in our study is much narrower than that of the abovementioned work, the role of texture influencing the rates of mineral-protected SOC, particularly in soils presenting similar mineralogy, cannot be ignored.

Chemically recalcitrant carbon compounds are also known to be significant contributors to the abundance of RSOC (Cheng et al., 2008; Lehmann et al., 2008). The mechanisms of stabilization of OM in forest subsoils were investigated by Mikutta et al. (2006) who reported an average contribution of 27% of the total stable SOC attributable to chemically recalcitrance of OC. However, that study was not conducted in fire-prone savannas, which have been shown to present a relatively large presence of recalcitrant substances derived from incomplete combustion of biomass (i.e. charcoal) (Bird & Grocke, 1997; Lehmann et al., 2008). Aromatic substances like lignin have traditionally been considered as important controlling factors over the formation and stabilization of SOC; however, recent studies have challenged this view (Thevenot et al., 2010). In particular, mechanisms behind their stabilization and turnover in soils remain open to debate. Furthermore, there are a limited number of studies dealing with lignin dynamics in dry tropical ecosystems, and those reported show relatively low lignin contents compared with other systems such as agricultural and temperate forests (Guggenberger et al., 1995; Thevenot et al., 2010). Hence, we hypothesize that the role of lignin determining the amount of RSOC in these ecosystems is far less significant than that of pyrogenic carbon. The savanna site with the highest RSOC contribution to TSOC provided good evidence for fire being the main factor behind the high content of RSOC observed in savanna environments not only because of the noticeably low numbers of trees that were able to reach maturity (Table 1) but also because this site had an extraordinary large presence of Cochlospermum planchonii, a pyrophyllic shrub associated with very frequent fires (Devineau et al., 2010). Therefore, we postulate that the decreasing trend in the contribution of RSOC to TSOC with increasing precipitation is mainly the result of more frequent fire events characteristic of savanna ecosystems (Sankaran et al., 2005; Grace et al., 2006; Furley et al., 2008), which is in agreement with the higher abundance of macroscopic charcoal fragments we noted in the soils of these ecosystems.

Significance of the findings for soil carbon studies in tropical ecosystems

We assessed the influence of climate, soil properties and vegetation characteristics on soil organic carbon (SOC) storage in a key geographical area with considerable potential for SOC sequestration. The soil sampling strategy used in this study allowed for the comparison of soil properties at systematically defined locations. The strong control by vegetation at the plot level was shown by the contrasting values in SOC contents observed between sampling locations. The large observed variation in SOC stocks reflects the very different climatic conditions existing along the transect, which together with soil properties, strongly determined the contrasting type of vegetation occurring at those sites. The degree of interdependence between SBD and soil properties for the range of soils covered in this work is highlighted by the strong negative relationships observed between SBD and SOC along the transect. Early studies dealing with SOC in tropical ecosystems usually reported soil carbon abundances without information on SBD (e.g. Jones, 1973; Kadeba, 1978). Therefore, it has not been possible to convert those results to inventories. However, provided that information on the basic textural characteristics of those soils is available, one can make use of the SOC-SBD relationships reported here to ascertain SBD for the range of mineral soils included in the present study. This can be achieved within a reasonable degree of accuracy as at least 0.84 of the variability gets explained. Therefore, the use of these relationships may allow the calculation of SOC stocks for those studies, which may provide useful baselines for research work dealing with changes in SOC stocks over time.

Used in combination with a suitable climatic parameter, such as available water, sand content is a reliable predictor of SOC stored in highly weathered dry tropical ecosystems with arguably less confounding effects than are associated with the use of clay content as a predictor. The presence of sesquioxides at some of the studied sites resulted in an ‘apparently coarse’ texture, which strongly influenced both the high SBD values observed and the amount of water available to plants. The latter played a fundamental role influencing the type of vegetation occurring along the transect, which has direct implications for the amount and quality of organic inputs returning to the system, and consequently on soil structure, SBD and SOC storage. Factors influencing the type of vegetation observed along this precipitation transect included soil fertility, soil physical constraints for plant growth and both natural and anthropogenic disturbances (e.g. grazing pressure, fire).

We suggest that the observed decreasing trend in the contribution of RSOC to TSOC pool with increasing precipitation was mainly the result of more frequent fire events characteristic of savanna ecosystems. Global coupled climate carbon cycle model simulations predict net losses in SOC stocks in West Africa as a result of increased heterotrophic soil respiration and reduced precipitation (Friedlingstein et al., 2010). These models have identified that SOC losses will be more significant in humid coastal regions, while SOC pools will show lower susceptibility in more arid regions. The greater relative proportion of RSOC in savannas further confirms that the resilience of these ecosystems to SOC loss is larger than that of forests. While the present study stresses the relevance of West African soil properties in SOC storage, our findings reinforce the view that semi-arid ecosystems offer a significant opportunity for soil carbon sequestration because of their large area and relatively low human populations (Tschakert et al., 2004; Marks et al., 2009). This work will contribute to the development of robust predictive models of biogeochemical cycling and vegetation dynamics in semi-arid tropical regions.

Acknowledgements

We gratefully acknowledge the many generous individuals who took part in the making of this work either in the field or in the UK. Sandra Lopez, Chris Wurster, Michael Zimmermann, Philippa Ascough and Paul Nelson provided very helpful discussions. Angus Calder and Martin Gilpin dedicatedly overviewed some laboratory analyses. We thank Dr Eric Mougin and colleagues for their hospitality and assistance in Mali. Field work assistance was provided by Markus Fink, Osei, Simon, Samuel, Wareh Zakaria, Oliver Phillips and Patrick Meir. The Forest Research Institute of Ghana (FORIG) provided vehicles, drivers, lab space and local support through the diligent work of Kester Mensah. The Scientific Research Station of the Dreyer Foundation in Dano (Burkina Faso) provided excellent accommodation and support. This work was funded as part of the UK National Environment Research Council - Tropical Biomes in Transition (TROBIT) consortium via research grant NE/D01185x/1 to the University of St. Andrews (UK).

      The full text of this article hosted at iucr.org is unavailable due to technical difficulties.