The study was conducted at a research farm of the Council for Scientific and Industrial Research (CSIR), Ghana, located at 5∘70′ N, 0∘29′ W and 49 m above sea level (a.s.l), some kilometres from Accra, Ghana (Fig. 1). The mean monthly temperature ranges from 21 to 31 ∘C, and the monthly rainfall ranges from 13 to 205 mm (annual rainfall is approximately 800 mm). The major rainy season period is from April to mid-July, with a minor rainy season in October (Fig. 2). Thus, the climate is moist semi-arid, with a growing period lasting 120–180 d (Ghana Meteorological Agency, 2019). The surface lithology is of non-carbonate sedimentary, with coarse sandy loam soils belonging to the Haplic Acrisol group (IUSS, 2015). The soil at the site had the following features at the 0–10 cm soil depth: 80.34 % sand, 12.64 % silt, 7.02 % clay, 5.86 pH (water), 0.133 % N, 2.10 parts per million (ppm) available P, 55 ppm available K and 2.61 % OM (Barnes, 1999). The native land cover of the study location is a tropical grassland savannah.

The research farm, located on the grazed–seeded grasslands (GSGs), keeps Sanga cattle, a cross between the humped Zebu-type cattle and the local West African Shorthorn known for their resistance to trypanosomiasis (adult weight range of 300–330 kg), and Djallonké sheep (adult weight range of 25–37 kg) that are managed semi-intensively. These animals are grazed rotationally on seeded pastures during the rainy season periods (April–October) and fed on conserved fodder harvested from arable fields and a fodder bank. While the exact stocking density is not recorded, typical stocking capacity around that region has been reported to be in the range of 3.5–5 LU ha-1 (Timpong-Jones et al., 2013). Compared to that, the native grasslands (NGLs) are infrequently grazed in a nomadic fashion, and no records about their stocking densities are recorded, yet grazing is limited to the dry season when feed is scarce at the GSG plots.

The site was selected due to its suitability for agriculture, its proximity to the capital city and its vegetation and climate, which are representative of the largest grassland type of Ghana (Guinea savannah). Ostensibly, the site had a uniform land usage until parts were converted for different uses in 1966 and in subsequent years (Fig. 1). Briefly, the land uses we encountered comprised arable field crops and grazed–seeded grasslands (both at field scales with at least 1 ha), and a fodder bank that also served as a forage botanical garden housing plant species brought from all over the world. Our study considered 59 species relevant to forage production (Table S1) and represented four plant functional groups: cut–use grasses (38 species), cut–use legume herbs (11 species), cut–use legume trees and shrubs (seven species) and cut–use non-legume trees and shrubs (three species) at plot scales (25–30 m2). As for a subset of species, several accessions of the same species were present, and we sampled 72 plots from the 59 species. The management practices associated with the land use types are summarized in Table 1.

Field/plot selection was based on the following two main criteria: (I) that the field was established in 1966 and (II) that the management remained relatively stable over the 50-year time frame. Information regarding the management of fields and biomass productivity of the species was obtained from documentation available at the farm. Plots/fields that underwent some changes in management or were re-established for any reason were omitted from the study. Consequently, there were no replicated plots, as in a traditional agricultural experiment, in the case of the cut–use forage species to control for variation. Hence, we categorized the forage species into functional groups, with the individual species of each functional group constituting a pseudo replicate.

Soil samples were taken from the native grasslands and arable crop fields by zone-based composite sampling, where we divided the fields into three zones (northern, central and southern). For the seeded–grazed paddocks, we randomly selected three paddocks out of the nine for soil sampling. We collected 10–20 subsamples (depending on the size of the zone), spread evenly across each zone in a zigzag pattern, to constitute a composite sample, with each composite soil constituting a replicate. In the case of the plot-scale samples, the sampling procedure followed the recommendations by Saiz and Albrechts (2016). Sampling locations were determined by roughly locating the first at the centre of the field, with three replicates laid out according to a pattern of three axes separated by 120∘ to a primary axis pointing north. Replicates were selected along these axes at an approximate midpoint between the centre of the field and its boundaries.

Before any sampling, surface litter was removed by hand. For each sampling plot or field, soil samples were taken (0–30 cm soil depth) using a soil probe of 1 in. (25.4 mm) diameter. Replicate samples obtained were bulked and thoroughly mixed, and a representative sample was taken in a zipped polythene bag for analyses. Due to the higher number of accessions, the total number of soil samples obtained from the cut–use forages was 72, derived from 59 species. For the estimation of soil bulk density, three to eight sets of soil samples were taken from each experimental unit (replicate plots/fields), depending on size, using a stainless steel core sampler at depths of 0–5, 5–10, 10–15, 15–20 and 20–25 cm. Soil bulk density and C stocks were estimated according to methods by Guo and Gifford (2002).

Soil samples meant for C and nutrient analysis were initially oven-dried at 30 ∘C for 48 h. Dried samples were sieved with a 2 mm sieve to remove coarse particles and plant roots. Sieved samples were milled and stored in a desiccator before analyses. Soil samples were analysed for C and total soil N (tN) with the C–N analyser (vario MAX CN; Germany), using aspirin (50 mg; N – 9.7 %; C – 34.0 %) and a standard soil sample (1 g; N – 1.2 %; C – 1.4 %) after every 10 tests of soil samples to aid in the calibration of the equipment. Bulk density was estimated after oven-drying at 105 ∘C. It was assumed that soil samples did not contain inorganic C because pH values were less than 7 and because no liming or any other amendment has been carried out during the past 50 years; therefore, total C was considered as SOC. Additional randomly performed HCl tests confirmed this assumption.

Soil pH was determined according to methods by Wiesmeier et al. (2012). Soil pH was measured directly with a pH meter (microprocessor pH/ION meter; PMX 3000; WTW) after adding 0.0125 M CaCl2 solution to each sample in the ratio of 1:2.5 (soil / CaCl2 solution). Plant available P (aP) and exchangeable K (eK) were extracted from 1 g air-dried fine soil (< 2 mm) using the Bray 2 solution, with the reagents being 0.1 M HCl and 0.03 M NH4F (Bray and Kurtz, 1945). eK was determined using flame photometry, and aP was measured calorimetrically at 882 nm (Miller and Arai, 2016) after a reaction with ammonium molybdate and the development of the molybdenum (Mo) blue colour (within 30 min).

Plant samples for CTs determination were harvested from forage species at the time of soil sampling, which was during the late annual growth stage. Sampling included leaves and leaf stalks from dicots, which consisted of legume herbs and legume and non-legume tree and shrub species (Table S1 in the Supplement). After cutting, samples were immediately cooled on ice before being freeze-dried, milled with a ball mill and stored in a freezer at -28 ∘C until further analyses. Condensed tannins (CTs; synonym – proanthocyanidins), which consisted of extractable CTs (ECTs), protein-bound tannins (PCTs) and fibre-bound tannins (FCTs) were determined according to methods prescribed by Terrill et al. (1992). Despite newer analytical techniques being available, Terrill et al. (1992) provide the benefit of separating again between protein- and fibre-bound tannins, and hence, this was considered to be most suitable method. Extractable condensed tannins (ECTs) were extracted from 20 mg plant samples using an acetone and water mixture (80:20, v/v), vortexed for 5 min and shaken on a plenary shaker (280 min-1) at 4 ∘C overnight. The samples were then centrifuged for 10 min (14 000 rpm) and decanted into 2 mL Eppendorf tubes. Residues were extracted again, using the same set-up with the plenary shaker now shaking for 3 h, and centrifuged, and the supernatant was decanted on top of the first extract. Acetone was evaporated in each case in an Eppendorf concentrator plus (Eppendorf, Hamburg, Germany) at room temperature for 90 min. Extracts were frozen overnight and, after that, freeze-dried for 24 h. A total of 1 mL of ultra-performance liquid chromatography (UPLC)-grade water was added to the freeze-dried extracts, vortexed and filtered with a polytetrafluoroethylene (PTFE) filter (0.2 µm).

Protein-bound condensed tannins (PCTs) were extracted twice from the residues by adding 10 g L-1 sodium dodecyl sulfate (SDS) and 50 g L-1 2-mercaptoethanol in 10 mM Tris–Cl adjusted to a pH of 8, vortexed for 5 min and then placed in a continually boiling water bath for 60 min and cooled on ice to room temperature. The mixture was then centrifuged for 10 min, and the supernatant was decanted into 2 mL Eppendorf tubes in triplicates. For analysis, 960 µL of n-Butanol and HCl (95:5, v/v) solution was added to 240 µL of the extract of either ECTs or PCTs, vortexed for 5 min and heated in an oven at 90 ∘C for 90 min; thereafter, it was cooled on ice and transferred to a spectrophotometer (Libra S22; Biochrom) and analysed at 550 nm to determine the CT concentrations. For the fibre-bound condensed tannin (FBCT) concentration in the samples, 1200 µL of BuOH/HCl (95:5, v/v) and 120 µL SDS was added to the residues from the extracts, vortexed for 5 min, centrifuged for 10 min and heated in an oven for 90 min at 90 ∘C before being cooled on ice to room temperature, centrifuged for 1 min and then measured in the spectrophotometer at 550 nm. Total CT (tCT) was calculated as the sum of ECTs, PCTs and FCTs for each candidate species. Annual tCT was estimated by multiplying tCT with the mean biomass yield of each functional group. The data set and a brief description of the site, materials and methods adopted to generate data are available in Nyameasem et al. (2020).

Due to lack of data regarding the initial SOC stocks of the farmland or the experimental plot, we estimated percent changes in soil C stocks using a pseudo baseline. This pseudo baseline is the natural grassland, which represents the SOC and nutrient stocks that all other sampled areas would have had they not been converted to farmland and agricultural research plots, respectively. We believe this assumption to be valid as all sampled areas were part of the same natural grassland until being converted, around the same time, to their current land use. With environmental conditions and atmospheric deposition being identical for all sites due to their proximity, the variation among plots is assumed to be a result of the land use change only. Thus, the natural grassland represents the trajectory for soil conditions as they would occur across the entire grassland region without land use change and, thus, represents the baseline against which all sites were tested. To analyse the effect of the different land use types on SOC and soil properties, we performed a one-way ANOVA using generalized linear models. Also, we tested the effects of plant functional groups (legume trees/shrubs, legume herbs and non-legume trees/shrubs) and plant parts (leaves and leaf stalks) on the polyphenol content of the forages in a two-way ANOVA, using linear mixed effect modelling, with species as a random factor. P values were estimated based on the type II sum of squares (SSs) in the case of the one-way ANOVA and type III SSs in the case of the two-way ANOVA due to the dissimilarity of the sample sizes (Fox and Weisberg, 2019). In cases where P values were significant, Tukey's post hoc tests, using the “lsmeans” function of the “multcomp” package (Bretz et al., 2011) were performed to permit pairwise comparisons of the means.

Before the ANOVA, data were checked for normality and the homogeneity of variance. In cases of abnormality or heteroscedasticity, data were log transformed or corrected using “White-adjusted heteroscedasticity-corrected standard errors” from the “car” package of R. Where data normality or the equality of variances were not confirmed even after log transformation, a non-parametric test (Kruskal–Wallis) was used, followed by Dunnett's post hoc tests to permit pairwise comparisons of means of the unequal sample sizes (Zar, 2010). Full correlations, a bivariate (linear and exponential) and a full hierarchical model were fitted to the data sets to establish relationships between the measured variables and SOC and to identify the main effects and interaction terms as predictors of SOC. Akaike's information criterion, with a small sample bias adjustment (AICC; Akaike, 1985; Burnham et al., 2011), was estimated and used to identify the best-fit model. All the statistics were performed using R (R Core Team, 2019).

The different land use types had varying effects on soil C stocks in the upper soil layer (0–30 cm), with group means ranging from 19.9 to 36.8 Mg C ha-1, and with a mean and standard deviation of 32.9 ± 10.2 Mg ha-1. Whereas SOC stocks did not differ between the agricultural soils and native grassland soils (p>0.05), grazed–seeded grassland soils had 86 % and 77 % less (p<0.01) SOC density relative to cut–use trees/shrubs and grasses, respectively (Fig. 3a). Among the cut–use forage systems, SOC was 36 % higher for cut–use grasses relative to legume herbs (p<0.05). Using the native grassland as a pseudo baseline, we observed SOC stock changes ranging from -44 % to 5 %, with a mean loss of -15 %, and the most considerable negative change occurring in grazed–seeded grassland soils (Fig. 3b). However, there appeared to be near-zero to positive changes in cut–use grass and legume trees/shrub soils. Among the cut–use fodder groups, SOC losses in soils of cut–use legume herbs were 26 % greater relative to cut–use grass soils.

The mean (± sd) OM, tN, aP and eK concentrations observed for the site were 1.32 ± 0.41 %, 0.057 ± 0.02 %, 10.4 ± 11.9 mg kg-1 soil and 68.1 ± 36.7 mg kg-1 soil, respectively. Whereas tN, aP and eK concentrations among the native grassland and agricultural soils (p>0.05), cut–use grasses soils contained 37 % more N compared to legume herb soils (Table 2). Arable crop soils contained 5, 7 and 12 times more aP density compared with cut–use grass, herbs and seeded–grazed soils (p<0.001), respectively (Table 2), while eK density in arable crop soils was 5 and 4 times higher (p<0.05) compared with seeded–grazed and legume herb soils, respectively (Table 2). CN ratio was 27 % higher in native grassland soils (p<0.05) compared with seeded–grazed and was lower (p<0.05) for grazed–seeded grasslands compared with non-legume trees/shrubs. It appears that C:N ratio values were not different among the forage species (p>0.05). Soil pH ranged from moderately acidic to neutral and was affected by the land use types, was more acidic in the cut–use herb soils relative to the grasses, arable crop and native grassland soils, and was more acid for seeded–grazed fields relative to arable crop soils (p<0.05).

The CT distribution in the forages was affected by plant functional groupings (p<0.001) but not by the part of plant analysed or by plant group–plant part interaction terms (p>0.05). Accordingly, the functional group effect on CT distribution in the forages is presented in Table 3. ECTs, PBCTs and FCTs in the legume herbs were 4, 3 and 2 times, respectively, lower (p<0.001) compared with legume trees/shrubs but not with non-legume trees/shrubs (p>0.05). However, tCT per dry matter (DM) and total annual tCT yield were 2–3 times and 4–6 times, respectively, higher (p<0.001) in both trees/shrubs compared with the herbs (Table 3).

Pairwise correlation between the measured variables suggested significant (p<0.01) positive associations between SOC and all the soil chemical properties (Table 4), with the strongest association between SOC and tN (r – 0.86). Although the CT variables correlated positively with SOC and tN, the associations were weak and insignificant, except for annual tCT yield (Table 4). Whereas tN, eK and C:N ratio related exponentially with SOC, aP and soil pH related linearly with SOC (Fig. 4a–e). Generally, higher tN,aP and soil pH appeared to be associated with higher SOC stocks; however, there appears to be a weaker association between SOC and tN at higher soil N concentrations. Whereas the response of SOC to tN was similar for non-legume trees/shrubs, grasses and legume herbs (mean slope – 10.5), it was about 36 % higher compared with legume trees/shrubs (slope – 14); see Fig. 5. Higher soil K density was associated with higher SOC stocks, but amounts beyond 16 Mg ha-1 appeared to depress SOC accumulation, whilst the relationship between SOC and C:N ratio appeared positive only beyond a ratio of 15 (Fig. 4c–d). SOC related exponentially with tCT yield, with generally higher biomass CTs associating positively with SOC at increased rates beyond 200 kg ha-1 (Fig. 6). A full regression model predicting SOC density using the measured soil and plant chemical variables as predictors showed significant effects of tN, aP, eK and tCT and aP × pH and tN × aP × eK interaction terms (p<0.05) in the model, with tN having the greatest effect (Table 5).

We observed a wide range of SOC stocks (17–64 Mg SOC ha-1) across the land use types at the study site. This is in agreement with previously reported ranges of 10–50 Mg SOC ha-1 for vegetation gradients varying from the Sudanese–Sahelian savannah to a subtropical forest (Saiz et al., 2012; Bessah et al., 2016). Similar to previous reports (Olson, 2013; Bessah et al., 2016), the use of grassland for agriculture resulted in both positive and negative responses by soil C dynamics. Tan et al. (2009) projected changes of -4 % to -23 % in soil C stocks of cultivated savannahs of Ghana on a 100 year scale, depending on the climate change scenarios. Hence, the 15 % loss of SOC in 50 years appears to be on the higher side. Nevertheless, there was an indication of C sequestration in the case of cut–use legume trees/shrub production. Similarly, a previous study (Shanmugam et al., 2018) reported a mean annual change of 0.67 ± 0.95 Mg C ha-1 for tropical mineral soils of secondary woodland or savannah converted to cultivated pasture or cropland. Therefore, our observations are not surprising as the type of land management adopted after converting native vegetation affects soil C cycling and determines the ultimate C storage potential of soils (Amézquita et al., 2008; Stahl et al., 2016). From our study, SOC stocks of arable crop farming did not differ from the native grassland. Arable crop production is known to deplete SOC due to the tillage effect, which tends to facilitate SOC cycling. Among the three arable crop fields considered in this study, only one of them occasionally adopted conventional tillage. On that particular field, however, crop production was on a rotational basis and occurred only occasionally, i.e. tillage occurred only infrequently. On the other crop fields, soils were manually minimum tilled using simple farm implements such as hoes and sometimes herbicidal weed control. Thus, this is, of course, a much less invasive technique compared to regular ploughing (but a representative for many farms in that area), and hence, it appears that the net effect of the mixed management practices on the arable crop farms did not impact SOC stocks significantly.

The low SOC stock in grazed–seeded grasslands was slightly above half (20 vs. 35 t SOC ha-1) that of cut–use grasses and legume trees/shrubs, suggesting a utilization effect. The effect size of grazing on SOC stocks might be dependent on the interactions of several factors, including grazing intensity (McSherry and Ritchie, 2013). Overgrazing has been widely reported as being responsible for the loss of SOC and a decrease in soil fertility in West Africa (Schönbach et al., 2011; Saiz et al., 2012). Generally, higher grazing intensities decrease soil C and N by direct removal of above-ground herbaceous biomass, thus reducing the potential of CO2 fixation in photosynthetic tissue and reducing below-ground C inputs through lower root production and higher root litter turnover (Semmartin et al., 2010). On the other hand, moderate grazing may increase tiller density and above-ground productivity, particularly in C4-dominated grasslands (McSherry and Ritchie, 2013) compared to cut–use systems, thereby leading to higher C inputs because of additions from crop residues and leaf senescence, assuming appropriate supplies of abiotic factors such as irrigation and nutrient supply. Thus, reducing grazing intensity may not only help protect the above-ground biomass of grasslands but it may also lead to improved soil structure and enhanced accumulation of organic C (Xu et al., 2018).

The effect of plant functional differences on SOC stocks was evident in this study. Under similar growing conditions, we observed higher C stocks in fodder grass fields (36 % higher; p<0.01) compared to legume herb fields. Earlier reports (Alonso et al., 2012; Shanmugam et al., 2018) suggested that species composition of grasslands influences the quantity and quality of OM input and of C sequestration in the soil. Dry matter productivity is an essential factor that influences the accumulation of soil C. Above-ground biomass productivity is higher for fodder trees/shrubs relative to that of grasses and lowest in herbaceous legumes (Table 1). SOC stocks under the cut–use forages appear to follow a similar trend (Table 2; Fig. 3). Thus, the higher SOC stocks observed under legume trees/shrubs relative to legume herbs could be partly attributed to the greater ability of legume trees/shrubs to produce relatively high amounts of biomass (Barnes and Addo-Kwarfo, 1996; Adjolohoun et al., 2008) and to maintain soil fertility (as shown in Table 2), control runoff-related soil erosion (Franzel et al., 2014) and maintain SOM.

Grasslands containing more C4 grass species were reported to store more SOC compared to grasslands with more C3 and legume species (Yang et al., 2019). The grass species considered in this study consisted mainly of C4 grasses, whereas the legume herbs were mainly C3 shallow-rooted plants. C4 plants possess morphogenic and architectural traits that enhance their ability to out-perform C3 plants, particularly under harsh environmental conditions (Lattanzi, 2010). They tend to develop a high leaf area index, which enhances their ability to capture light, N and water compared to their C3 counterparts. Besides, C4 plants have higher photosynthetic efficiency in the use of water and N, and they yield higher quantum productivity (Taylor et al., 2010) than C3 plants. The inherently lower Rubisco concentration and the more lignified tissue in C4 plants limit organic matter decomposition, thus producing more recalcitrant organic C (Yang et al., 2019).

Moreover, C4 plants are reported to partition C towards roots in N limited situations; thus, they have a higher ability to fix soil C under N stress situations (Sage and Pearcy, 1987; Long, 1999). Another reason could be that C4 plants have a higher concentration of amino acids and organic acids in their root exudates compared to C3 plants (Nabais et al., 2011), which could protect SOC and N from microbial decomposition. Therefore, the relative differences in biomass characteristics could be responsible for the observed differences in SOC stocks under fodder grass and legume herbs fields.

We observed a higher N concentration in the soil of cut–use grass fields compared to legume herb fields, although legumes are known to fix atmospheric N in soils. Indeed, in the semi-arid and savannah ecosystems of sub-Saharan Africa, legume herb and legume browse species are reported to fix 8–217 and 61–643 kg ha-1 yr-1 atmospheric N, respectively (Hassen et al., 2017). Accordingly, legume-based systems can produce high biomass yield even in the absence of N fertilizer application, showing a high N cycling efficiency (Schmeer et al., 2014) and generally increasing soil N concentration. However, several factors may affect N fixation in legumes, including environmental conditions, N uptake by plants and soil pH (Nutman, 1976; Sage and Pearcy, 1987). Pure grass stands may show a high fraction of C and N allocation below-ground (Loges et al., 2018), particularly when the functional traits equilibrium theory of the roots is considered (Brouwer, 1983) and the stands have amino-acid-rich roots (Nabais et al., 2011), leading, potentially, to a more excessive N and C accumulation in the soil. These attributes of grasses might explain why soils in grass plots had higher tN compared to the legume herbs. The fairly tight C:N ratio (14:1) observed for the agricultural systems reflects the stoichiometry of stable soil OM at the study site.

The site of this study has seen a general decline in soil fertility (tN and SOM), although there were marginal increases in available P and K compared with the values reported for the site by Barnes (1999). While P and K concentrations were moderate in arable crop soils, they were low in the other land use types (Bationo et al., 2018; Apal Agricultural Laboratory, 2020). The relatively high P and K values observed in the food crop fields may probably be due to their annual supply through fertilizer application (Table 1). Cut–use grasses and legume herbs appeared to deplete soil P, but this effect was not apparent for legume or non-legume trees and shrubs. This trend might be partly related to differences in the root architecture. The grasses and herbs are shallow rooted and, therefore, exploit nutrients from within the upper soil depth, while the trees/shrubs have roots below the 30 cm soil depth considered in this study and, therefore, might be less able to exploit nutrients extensively from the upper layer of the soil. The relatively low nutrient concentrations observed in the grazed–seeded grasslands matches the low C stocks (Table 2). It is particularly striking that P and K are much lower in grazed grassland despite the greater amounts of excreta deposition associated with grazing. Although we could not ascertain the stocking rate on the grazed–seeded pastures, it appears that there was frequent overgrazing of the pastures, which is a regular feature in this region, exposing the soils to nutrient loss via erosion (Bationo et al., 2018).

The soils were generally acidic, a common characteristic of tropical soils (Jayne et al., 2015). Associated effects of the <5.5 pH observed in grazed–seeded grasslands, cut–use legume herbs and non-legume trees/shrubs soils might cause aluminium and manganese toxicities. Low pH could also cause molybdenum, calcium, magnesium or potassium deficiencies and perhaps reduced microbial activity (Apal Agricultural Laboratory, 2020). Among the land use systems, cut–use legume herbs and seeded–grazed soils appeared more acidic relative to the others. Legume plants commonly form symbiotic associations with rhizobia and accumulate most of their N through symbiotic N fixation. During this process, legume plants take up more cations than anions and release more H+ ions from roots to the soil, leading to low pH values in both the rhizosphere and bulk soil (Zhao et al., 2009; Yang et al., 2016). This effect might differ between legume herbs and legume trees/shrubs due to differences in their root architecture. Grazing fields are associated with high N and C returns from animal excreta, but C and N cycles might cause acidification in grazed fields; for example, nitrate leaching might increase the concentration of H ions, which might lead to decreased soil pH (Ridley et al., 1990).

The total CT concentration of below 67 g kg-1 DM observed in this study is similar to values reported for tropical forages, with lower concentrations in legume herbs than for browse species (Mupangwa et al., 2000; Sottie et al., 2016). CTs may have both positive and negative effects on livestock, depending on their concentration in the plants. The nutritional implications of CTs are particularly vital in the case of tropical shrubs and tree species, as high CTs are known to impair protein availability (Jayanegara et al., 2019). Previous authors have reported higher values of more than 70 g CT kg-1 DM for some tropical shrubs and trees (Jackson et al., 1996; Rosales, 1999; Pereira et al., 2018), which invariably limits their use as fodder for ruminant livestock.

The proportion of total CTs allocated to ECTs was similar to the range of 12 %–44 % reported by Mupangwa et al. (2000) but lower than the 70 %–95 % reported for some tropical browse species (Jackson et al., 1996). Compared with other tropical forage species, the tCT shares of below 7 % indicate comparably low tannin concentrations, which indicates their suitability as potential feeds for livestock (Mupangwa et al., 2000; Jackson et al., 1996). However, further analyses on forage quality parameters and tannin composition, and on other plant secondary metabolites, might be required to give reliable estimates of their digestibility value.

Our correlation and regression analyses (Fig. 4; Tables 4 and 5) indicated significant relationships between the soil chemical properties and SOC. Similar to reports of Stahl et al. (2016), we found a close association between C and N. Nevertheless, in the literature the relationship between N and SOC is ambiguous, with evidence of different responses of soil CO2 fluxes to N levels in soils (Yan et al., 2016), including increases (Wang et al., 2015), decreases (Jiang et al., 2010) and no significant differences (Li et al., 2012). The N cycle is strongly interconnected with the C cycle as N is required for the cell growth and division and, therefore, for the accumulation of biomass (and hence of C). Whereas Six et al. (2002) attributed the positive relationship between soil C and N to protection by both macro- and microaggregates against mineralization, Waldrop et al. (2004) explained that greater N availability reduces decomposition rate of SOC by regulating the production and activity of microbial extracellular enzymes. Indeed, when N becomes available in N-limited soils, photosynthetic reaction increases, thus enhancing SOC storage, but when amounts of N are low, this may limit the CO2 fertilization effect (Lal, 2018). Accordingly, Tan et al. (2009) observed that increasing N application to about 30–60 kg ha-1 yr-1 could result in positive changes in the soil C stocks of cultivated grasslands of Ghana. In the present study, the multiple linear regression model (Table 5) showed that tN had the most significant effect on SOC density and with a greater effect in size in legume trees/shrub species compared with the legume herbs, non-legume trees/shrubs and grasses (Fig. 5).

Interaction effects involving N, P and K and pH and K were significantly associated with SOC stocks (p<0.05), suggesting that the effect of N on SOC sequestration might be dependent on the availability of P and K and on the soil pH. The associative effect of N and P on SOC density was also reported in an earlier study (Bradford et al., 2008), where P and N additions led to a more significant C sequestration in soils. In a simulated study (Li et al., 2014), a combination of N and P fertilizers increased SOC storage but reduced microbial activity and C mineralization compared to adding N or P fertilizer alone or no fertilization. In contrast, Graham et al. (2014) reported a significant release of CO2 when soil N increased. In the current study, if the relative proportions of soil nutrients impact SOC storage and soils deficient in P are less able to use N efficiently, then this might increase N losses by gaseous emissions and leaching and, thereby, contribute to depletion of SOC pools.

The theoretical basis, according to linear mixed effect regression models (Table 5), suggested tCT concentration in above-ground biomass as being one of the potential factors that might affect SOC storage at the site (Table 5). This observation partly confirms our hypothesis that above-ground CT concentration might be influential in C storage. Although CT concentrations in the soils were not considered in this study, we speculate that the higher CT content in legume trees/shrubs (Table 3) might have slowed the decomposition rate of soil OM (Dong et al., 2016) to promote SOC accumulation (Kraus et al., 2003; Halvorson et al., 2011; Tamura and Tharayil, 2014; Chomel et al., 2016; Adamczyk et al., 2016, 2017; Kagiya et al., 2019). By extension, the higher potential of the trees/shrubs to produce more mass and more CTs per unit area might explain the higher SOC stocks under trees/shrubs relative to those for the herbs. However, the bivariate regression analysis showed that positive associations between tCT and SOC only occurred at concentrations above 200 kg ha-1 yr-1 (45 mg kg-1 DM). Although tropical herbs contained relatively higher CTs than their grass counterparts, the CT concentration might not be potent enough to impact SOC storage, as shown in this study (tCT = 10 ± 8; <45 mg kg-1 DM). Although tropical herbs contained higher amounts of CTs than their grass counterparts, SOC density was lower in soils under the herbs compared with grasses. Even when the CT concentration in the biomass is low, it might be that, in the rhizosphere, it is higher due to accumulation. Below-ground biotic and abiotic conditions, including the presence of some fungi (e.g. Basidiomycetes and Ascomycetes) and bacteria species (e.g. Pseudomonas), temperature, pH, oxygen and substrate availability, are factors that affect the degradation of phenolic compounds in soils (Min et al., 2015). Although the effect of pH on polyphenol degradation is not consistent, under laboratory conditions low soil pH (<5) was shown to be optimal for extracellular enzymes involved in degrading phenolic compounds in soils (Min et al., 2015). In our study, soil pH was lower in soils under the legume herbs compared with the grasses (Table 2). This low soil pH might have facilitated a faster degradation of the potentially accumulated CTs and mitigated the assumed effect of CTs on SOC.

We observed significant positive associations between annual CT yield and all the measured soil parameters. A previous study (Mudau et al., 2007) demonstrated that, regardless of the season, application of N, P and K fertilizers increased the total polyphenols in bush tea (Athrixia phylicoides L.), suggesting that N, P and K limitation might affect the synthesis of CTs by plants and for SOC storage. The low nutrient density in the soils might partly explain the lower concentration of CTs in the forages compared to earlier reports on similar species (Jackson et al., 1996; Rosales, 1999; Pereira et al., 2018). Thus, the proportion of extractable CTs and bound CTs, even within species, might change with plant maturity and environmental conditions (Jackson et al., 1996) and might influence the feeding value of the forage. Consequently, increased soil fertility might have some implications for CT concentration in forages and, thus, for livestock feeding and SOC storage.

We speculate that, at the farm level where climatic and soil properties are relatively uniform, differences in land use might influence the quantity and quality of organic matter added to the soil. These variables might have induced plant–soil feedbacks to regulate soil C cycling at the micro level, causing local variations in C stocks (Post et al., 1982; Saiz et al., 2012; Chen et al., 2018).

This study has shown that soil nutrients have implications for SOC input and removal and, by extension, for primary plant productivity and ecosystem functioning (Post et al., 2012; Marques et al., 2016). However, most Ghanaian soils are inherently infertile (Bationo et al., 2018) as result of leaching, soil erosion by rainfall (0.6–0.9 t ha-1) and the fact that nutrients removed by crop harvest are not replaced by the corresponding amount of plant nutrients (Bonsu, 1979; Bationo et al., 2018). Consequently, the low concentration of soil nutrients across the fields may partly explain the low SOC stocks observed for soils in the study site. These fragile soils would require sustainable forms of agricultural land use systems to ensure higher below-ground net primary productivity (NPP) for increased C stocks.

The forage legume herbs showed the lowest yields (Table 1), indicating low bioactivity in comparison with grasses, whereas the legume shrubs had the highest yields and, thus, captured a lot of N from the biomass. In addition, soils under the shrubs contained similar tN compared with grasses, suggesting that most of the tested forage legume herbs were not suitable species with regards to biomass yield for forage and soil carbon sequestration. Legume shrubs should be preferred, as they provide more forage and ecosystem services. Invariably, our findings further extend the benefits of cut–use trees/shrubs as they already contribute to food security, incomes and livelihoods in Africa. These plants are deep rooted, resistant to drought and maintain high protein levels during the dry season when high-quality feed is scarce (Wambugu et al., 2011).