Coffee agroforestry, soil carbon, and restoration of a tropical Afromontane rainforest

Coffee agroforestry systems are increasingly promoted as sustainable land-use strategies to restore degraded tropical landscapes while supporting agricultural production and climate change mitigation. However, their ability to recover soil carbon pools relative to natural forest regeneration remains uncertain, particularly in tropical Afromontane ecosystems. We hypothesized that coffee agroforestry systems partially restore soil carbon dynamics but remain functionally distinct from soils in regenerating and mature forests. To test this, we evaluated soil organic carbon, carbon stocks, microbial biomass carbon and nitrogen, soil respiration, and labile carbon across five land-use types: fallow agricultural land, two-year-old and seven-year-old coffee agroforestry systems, regenerating forest, and mature forest fragments. Soil samples were collected at 0–15 cm and 15–30 cm depths and were analyzed using standard laboratory and statistical approaches. Soil organic carbon decreased with depth and was consistently lower in coffee agroforestry systems compared with forest and fallow sites. Carbon stocks were largely similar across land uses but were significantly higher in regenerating forest. Microbial biomass and soil respiration were reduced in older agroforestry systems, indicating lower biological activity. Labile carbon was highest in forest systems and positively related to total carbon stocks. These results demonstrate that while coffee agroforestry systems can maintain soil carbon levels, they do not fully restore soil carbon dynamics and microbial functioning compared with natural forest regeneration. This study provides new insights into the role of agroforestry in soil carbon recovery and highlights the need to integrate agroforestry with forest restoration strategies to enhance soil health, ecosystem resilience, and climate change mitigation in tropical landscapes.

Land-use change in tropical regions is one of the main drivers of soil degradation and soil organic carbon (SOC) losses, with significant implications for ecosystem functioning and climate regulation (1, 2). The conversion of native forests to agricultural systems often leads to reductions in SOC stocks, disruption of soil microbial communities, and changes in key biogeochemical processes such as nutrient cycling and soil respiration (3, 4). Given the central role of soils in the global carbon cycle, understanding how different land-use systems influence soil carbon pools and associated biological processes is essential for informing sustainable land management and climate mitigation strategies.

Agroforestry systems have been increasingly recognized as land-use options with the potential to mitigate soil carbon losses while maintaining agricultural productivity (5, 6). In coffee agroforestry systems, coffee trees are integrated with shade trees, increasing litter inputs, root biomass, and soil structural stability, while also moderating microclimatic conditions that may enhance microbial activity and carbon stabilization (7, 8). Several studies have reported higher SOC stocks in agroforestry systems compared to monoculture agriculture, and in some cases SOC levels comparable to those observed in secondary or regenerating forests (9, 10). However, the magnitude and direction of these effects are context-dependent and influenced by factors such as system age, tree composition, management intensity, and environmental conditions.

In spite of knowledge gaps, coffee agroforestry systems are increasingly promoted as nature-based solutions capable of supporting agricultural production while maintaining ecosystem function, including carbon sequestration and soil restoration (5, 6, 11). However, some studies assess soil carbon dynamics across broad land-use categories without explicitly considering successional gradients or applying comparable experimental designs across systems (12, 13). Moreover, soil carbon studies often represent land-use types with a single, well-characterized site with internal spatial replication, rather than multiple sites distributed across a wider landscape (3). Tropical Afromontane ecosystems are particularly important for comparative studies of soil carbon dynamics in agroforestry and native forests as these ecosystems are highly threatened (14) and are the site of a growing emphasis on coffee production.

Mount Gorongosa in central Mozambique represents one of the most important tropical Afromontane landscapes in southeastern Africa and is recognized for its exceptional biodiversity, high conservation value, and ongoing ecosystem restoration initiatives (15–18). In recent years, shade-grown coffee agroforestry has been promoted within the Gorongosa landscape as a strategy to improve local livelihoods while reducing pressure on native forests. Because coffee cultivation is increasingly expanding across tropical montane regions, understanding its influence on soil carbon dynamics is essential for evaluating the potential contribution of coffee agroforestry to ecosystem restoration and climate-change mitigation.

The present study aimed to evaluate the effects of land-use change on soil carbon stocks and soil biological functioning across a restoration gradient on Mount Gorongosa. Specifically, we compared mature forest, regenerating forest, fallow agricultural land, and coffee agroforestry systems of different ages with respect to the SOC concentration, carbon stocks, microbial biomass carbon and nitrogen, soil respiration, and labile carbon fractions. We hypothesized that (i) forest ecosystems would maintain higher soil carbon stocks and biological activity than agricultural systems, (ii) regenerating forest would exhibit evidence of soil carbon recovery following disturbance, and (iii) coffee agroforestry systems would have greater soil carbon and microbial activity than fallow land while remaining functionally distinct from natural forest ecosystems.

Study area

The study was conducted on Mt. Gorongosa in Gorongosa National Park (GNP), Sofala, Mozambique. Average rainfall was 2006.6 mm between 2017 and 2022. Rainfall is highly seasonal, with about 58% falling between November and April. The average temperature was 20.4 °C (16; climate data available from the GNP Science Department upon request). The soils are classified as Ferralitic (19, 20), sandy (granite-derived) and sandy to loamy-clay (gabbro-derived; 15). The particle size distribution (21) of the surface soil (0–15 cm) is 252 g kg -1 coarse sand, 297 g kg -1 fine sand, 201 g kg -1 silt, and 250 g kg -1 clay. The particle size distribution of soil at 15–30 cm depth is 240 g kg -1 coarse sand, 311 g kg -1 fine sand, 149 g kg -1 silt, and 300 g kg -1 clay. The soil is acidic with a pH (H 2 O) of 4.7 across depths (22). Effective cation exchange capacity is 4.41 cmol c kg -1 at the surface and 2.44 cmolc kg -1 between 15 and 30 cm depth (23). Base saturation is 78.8% in the surface layer and 58.8% in the lower soil layer.

Itinerant agriculture is common, and agricultural fields are typically abandoned after two years. The present study was conducted in five areas with different land uses (Figure 1). The first area was a 0.43 ha mature forest fragment (-18.48077, 34.04599, 955.4 m a.s.l.). While the exact age of the forest is unknown, satellite imagery indicates the site has been forested for at least 40 years; observations on the vertical structure, tree size, and species composition further indicate the forest is at least late secondary. The other land-use types were previously used for maize cultivation. The regenerating forest was a 0.47 ha fragment estimated to be 10 years old based on Google Earth images (-18.47878, 34.04267, 964.8 m a.s.l.). The third site was a 0.65 ha plot of agriculture land that was fallow for five years (-18.47931, 34.04122, 959 m a.s.l.). The fourth area was a 0.28 ha two-year-old coffee agroforestry system (CAFS; -18.47997, 34.04058, 960.7 m a.s.l.). The fifth site was a 0.32 ha seven-year-old CAFS (-18.48087, 34.04058, 946.5 m a.s.l.). The combined study area covered approximately 2.04 ha across the five land-use types. The mature and regenerating forest fragments as well as the fallow agricultural lands were all gently sloping with <10° incline. The two-year-old CAFS was on an ~30° incline, and the seven-year-old CAFS was on a slope of ~40°.

Coffea arabica was planted in a 2.5 x 3 m spacing scheme for an approximate density of 2222 plants ha -1 (24). Native tree species were planted between the coffee plants at a density of ~90 trees ha -1. The seven-year-old CAFS was planted in 2014, and the two-year-old CAFS was planted in 2019. Prior to planting, the area was used for maize cultivation, and originally it was a closed-canopy forest. Native shade trees in the two-year-old CAFS included Albizia adianthifolia and Khaya anthotheca. Native shade trees in the seven-year-old coffee agroforestry plantation included A. adianthifolia, Annona senegalensis, Bridelia micrantha, Erythrina lysistemon, K. anthotheca, and Trema orientale.

Sampling was done at the edge of the mature forest fragments, and common tree species were A. adianthifolia, B. micrantha, Harungana madagascariensis, Heteropyxis natalensis, and Syzygium cordatum. Tree species in the regenerating forest fragment included A. adianthifolia, A. senegalensis, B. micrantha and H. madagascariensis. There was also a dense cover of herbaceous vegetation and invasive Lantana camara in the regenerating site. The fallow agriculture land had been abandoned for at least five years at the time of sampling. The fallow site was characterized by grass cover and scattered individuals of A. adianthifolia, A. senegalensis and B. micrantha. The area was used for maize cultivation before being abandoned. No fertilizers are used in maize fields in the region, but signs of cattle and fire were present in the study site. The abandoned agricultural land also experienced frequent burning.

Although the species composition varied among sites, the selected coffee agroforestry systems were dominated by native shade-tree species characteristic of the Mt. Gorongosa Afromontane forest ecosystem (15). Similarly, mature forest sites were selected within relatively undisturbed forest fragments representative of the regional vegetation. As vegetation composition is an intrinsic component of each land-use category, it was not controlled experimentally but is considered part of the ecological context influencing soil carbon dynamics. Fire occurrence and grazing pressure are low within the study area due to conservation regulations associated with the Gorongosa landscape, although historical disturbances may have influenced present-day soil properties. Such legacy effects are recognized as important, unmeasured drivers of soil carbon dynamics.

Experimental design and sampling

The experimental design followed a comparative, space-for-time substitution approach commonly used in soil carbon studies assessing land-use change and vegetation succession (3). Different land-use types were selected within the same landscape, including forest regeneration stages and coffee agroforestry systems. Each land-use type was represented by a well-characterized site considered representative of its category, with internal spatial replication used to capture within-site variability.

Sites representing different land-use histories were compared to infer potential patterns associated with land-use transitions. This approach is widely used in soil carbon research but does not directly measure temporal change; therefore, observed differences should be interpreted as spatial contrasts rather than demonstrated ecological trajectories through time (25).

Although each land-use category was represented by a single site, ten spatially distributed sampling points were established within each site to capture variation within a land-use type. Therefore, the study should be interpreted as a comparative site-based assessment rather than a fully replicated experimental design. Consequently, inferences should be interpreted with caution when extrapolating to broader regional contexts (26). This design is widely adopted in studies of soil organic carbon and microbial processes, particularly in tropical regions where long-term replicated experiments across multiple landscapes are rare (12, 13). By maintaining comparable soil type, climate, and topographic conditions across sites, the approach allows for robust comparisons of soil carbon pools and biological activity across land-use systems while controlling for major environmental drivers.

Soil sampling

For the determination of organic carbon (C), labile C, microbial biomass C and nitrogen (N), and soil respiration, soil samples were collected with an auger at two depths (0–15 cm and 15–30 cm). Ten sampling sites separated by >20 m were randomly selected in each land-use type, air-dried, and sieved (Ø < 2 mm particle size). For the determination of bulk density and volumetric fractions of fine earth and coarse material fractions, samples were collected following the procedures described by Cools and De Vos (27), specific for forest soils and tree/shrub species, and equivalent to ISO 11272 (bulk densities and C stocks data are presented in Supplementary Table 1).

Laboratory analyses

Soil organic C was determined by dry combustion using a Skalar Primacs SCN analyzer (Skalar, Breda, Netherlands) following the FAO (28) protocol. The fine earth fraction bulk density and volumetric percentage of the fine earth fraction were used to calculate C stocks (t ha -1) across depth (0–30 cm) following Wiesmeier et al. (29). Soil organic C stocks were calculated for each sampling location by summing the C stocks of the two sampled soil layers (0–15 cm and 15–30 cm) according to:

Where:

S O C L z is the total SOC stock (kg m −2) of both soil layers (0–15 cm and 15–30 cm) sampled in each location;

S O C i is the soil organic carbon concentration (mg g −1) of the fine-earth fraction of layer i;

B D i is the bulk density (g cm −3) of the fine-earth fraction of layer i;

L i is the thickness (cm) of layer i;

and S C i is the volumetric fraction (%) of rock fragments > 2 mm in layer i.

Microbial biomass C and N (MBC and MBN) were measured at 0–15 cm because land use is expected to influence soil microbes in shallow soils (30). Microbial biomass C and MBN were measured using the fumigation-extraction method with efficiency coefficients of K EC = 0.45 for C and K EN = 0.54 for N (31, 32). Carbon in the K 2 SO 4 extract was quantified with a near infrared detector, and N in the K 2 SO 4 extract was quantified using chemiluminescence. Both measurements were performed after combustion at 950 °C in a Formac analyzer (Skalar, Breda, Netherlands).

Water holding capacity (WHC) was determined for three randomly chosen samples from each land-use type. Soil respiration was then measured from samples collected at 0–15 cm depth using the Isermeyer method, modified by (33). Fifty grams of dry soil were brought to 55% of their WHC and placed in 100 mL plastic bottles. Each flask was placed in a 1 L Mason jar containing a plastic flask with 25 mL of water to maintain soil moisture and a second flask with 25 mL of NaOH (0.05 M) to capture CO 2. Six controls were prepared with flasks of water without soil. Jars were hermetically sealed and incubated for three days at 25 °C. After incubation, NaOH samples were removed for CO 2 analysis with a near-infrared detector using a Formac elemental analyzer (Skalar, Breda, Netherlands).

Labile C was determined from samples collected at 0–15 cm depth following Weil et al. (34). Briefly, 5 g of soil were extracted in 20 mL of 0.02 M KMnO 4 in 1 M CaCl2 (pH 7.2). The soil suspension was shaken on a reciprocating shaker for 2 min and allowed to stand for 10 min. Two hundred µL of the supernatant were then diluted in 10 mL deionized water, and the absorbance of the unreacted KMnO 4 was measured with a spectrophotometer (Helios Omega UV-VIS) at 550 nm. Labile C contents were calculated using a calibration curve based in KMnO 4 standard solutions.

All laboratory analyses were conducted using standardized procedures, and quality control measures were applied throughout sample preparation and analysis to ensure consistency and reproducibility.

Data analysis

Soil organic C was analyzed using a two-way ANOVA with land use, soil depth, and their interaction as categorical predictors. Carbon stocks were analyzed using ANOVA with land use as the categorical predictor variable. Microbial biomass C and MBN were modeled with ANCOVA using land use as a categorical predictor and C stocks, labile C, and soil respiration as covariates. Labile C was analyzed with ANCOVA using land use as a categorical predictor and C stocks, MBC, MBN, and soil respiration as covariates. Lastly, soil respiration was modeled with an ANCOVA based on land use and C stocks, labile C, MBC, and MBN as covariates. Prior to analyses, normality was tested with the Shapiro-Wilk test, and homogeneity of variance was verified with the Levene test. Microbial biomass C was log-transformed for normality; all other data met assumptions of normality and homogeneity of variance. Tukey tests were performed following GLMs to determine differences between levels of land use based on an alpha of 0.05. Each land-use category had ten replicates except the two-year-old CAFS which had nine replicates.

Prior to model fitting, multicollinearity among explanatory variables was evaluated using variance inflation factors (VIF) and a correlation matrix. No VIF values exceeded 1.3, and no correlations were > 0.45, indicating collinearity was not a problem in our dataset (Supplementary Table 2). Model simplification followed a backward stepwise procedure based on Akaike Information Criterion (AIC), which identifies parsimonious models while balancing explanatory power and model complexity (35). Potential spatial autocorrelation of sampling sites between and within land-use types was examined with Moran’s Index to check for pseudoreplication of sampling points across the landscape. The Ape package was used to calculate Moran’s Index (36). Analyses were performed using R version 4.1.1 (37).

Spatial autocorrelation of sampling points

Analysis of spatial autocorrelation indicated that measured soil variables did not exhibit significant spatial dependence among sampling points within land-use systems. Moran’s Index revealed spatial autocorrelation between land-use types, e.g., sampling points in fallow land were more similar to each other than to those in CAFS. However, within each land-use type, sampling locations were spatially independent (Supplementary Tables 3, 4; Supplementary Figure 1). These results confirm that spatial autocorrelation did not bias the observed differences among land-use systems and that the sampling design was appropriate for subsequent statistical analyses. The spatial distribution of local indicators of spatial association (LISA) for soil organic carbon is presented in Figure 2.

Soil organic carbon concentration

Soil organic carbon decreased with depth and varied significantly among land-use systems (Figure 3; Supplementary Table 5). The lowest SOC concentrations were recorded in both CAFS at the two sampled depths. Mean SOC values in the two-year-old CAFS were 58.1 g kg −1 (± 4.64 SE) at 0–15 cm and 41.2 g kg −1 (± 3.24 SE) at 15–30 cm, whereas the seven-year-old CAFS presented SOC values of 55.6 g kg −1 (± 4.53 SE) and 38.3 g kg −1 (± 3.80 SE) for the same respective depths. In contrast, SOC concentrations in fallow agricultural land and forest fragments were statistically similar. Fallow land exhibited SOC values of 67.7 g kg −1 (± 2.27 SE) at 0–15 cm and 54.3 g kg −1 (± 2.46 SE) at 15–30 cm, while regenerating forest presented SOC values of 69.7 g kg −1 (± 2.02 SE) and 49.7 g kg −1 (± 2.50 SE). Mature forest had SOC concentrations of 63.7 g kg −1 (± 2.33 SE) in surface soils and 50.8 g kg −1 (± 2.14 SE) in subsurface soils. Model selection results and corresponding AIC values for all tested models are in Supplementary Table 5.

Soil organic carbon stocks

Total soil C stocks calculated for the 0–30 cm depth interval did not differ significantly among most land-use systems (Figure 4; Supplementary Table 6). Mean C stocks were 134.7 t ha −1 (± 3.96 SE) in fallow agricultural land, 127.0 t ha −1 (± 8.11 SE) in seven-year-old coffee agroforestry systems (CAFS), and 125.0 t ha −1 (± 3.72 SE) in mature forest fragments. In contrast, regenerating forest fragments exhibited significantly higher carbon stocks, reaching 158.0 t ha −1 (± 3.56 SE). Carbon stocks in the two-year-old CAFS (133.8 t ha −1 ± 7.83 SE) were statistically similar to those observed in the regenerating forest fragment.

Microbial biomass carbon and nitrogen

Microbial biomass C and MBN varied significantly among land-use systems but were not influenced by soil depth (Figure 5; Supplementary Table 7). Microbial biomass C was lowest in the seven-year-old CAFS with a mean value of 177.0 mg C kg −1 soil (± 12.5 SE). All other land-use systems presented statistically similar MBC values, including fallow agricultural land (403.5 mg C kg −1 soil ± 56.0 SE), two-year-old CAFS (419.6 mg C kg −1 soil ± 90.7 SE), regenerating forest fragments (481.8 mg C kg −1 soil ± 51.6 SE), and mature forest fragments (405.2 mg C kg −1 soil ± 71.8 SE).

Microbial biomass N followed a similar land-use pattern but showed clear differentiation among systems (Figure 6; Supplementary Table 8). Higher MBN values were observed in regenerating forest fragments (58.6 mg N kg −1 soil ± 4.5 SE) and fallow agricultural land (63.0 mg N kg −1 soil ± 3.91 SE). In contrast, lower MBN values occurred in the seven-year-old CAFS (39.5 mg N kg −1 soil ± 3.0 SE) and mature forest fragments (34.3 mg N kg −1 soil ± 4.4 SE). The two-year-old CAFS presented intermediate MBN values (50.8 mg N kg −1 soil ± 8.2 SE; Supplementary Tables 7, 8).

Soil respiration

Soil respiration was significantly influenced by land use but not soil depth (Figure 7; Supplementary Table 9). The lowest respiration rates were recorded in CAFS with a mean of 55.3 mg C–CO 2 kg −1 soil day −1 (± 3.3 SE) in the two-year-old CAFS and 46.2 mg C–CO 2 kg −1 soil day −1 (± 4.8 SE) in the seven-year-old CAFS. In contrast, soil respiration rates were statistically similar among the remaining land-use systems, including fallow agricultural land (73.2 mg C–CO 2 kg −1 soil day −1 ± 2.8 SE), regenerating forest fragments (72.0 mg C–CO 2 kg −1 soil day −1 ± 2.4 SE), and mature forest fragments (82.6 mg C–CO 2 kg −1 soil day −1 ± 4.0 SE). Model selection results and corresponding AIC values for all tested models are presented in Supplementary Table 9.

Labile carbon

Labile C differed among land-use systems and was highest in mature and regenerating forest fragments, both presenting mean values of 0.6 mg C kg −1 soil (± 0.01 SE; Figure 8; Supplementary Table 10). Intermediate labile C values were observed in fallow agricultural land and seven-year-old coffee agroforestry systems (CAFS), each with means of 0.55 mg C kg −1 soil (± 0.02 SE). The lowest labile carbon levels were recorded in the two-year-old CAFS, with mean values of 0.5 mg C kg −1 soil (± 0.02 SE). Labile C was positively correlated with total soil C stocks as indicated by regression analysis with a parameter estimate of 0.14 (Supplementary Table 10).

Across all land-use categories, SOC, microbial biomass, respiration, and labile C generally declined with soil depth. Soil organic C concentrations and stocks were highest in regenerating forest. Soil organic C concentrations were lowest in the CAFS, and stocks were lowest in the fallow agricultural land, seven-year-old CAFS, and mature forest. Microbial C and N were highest in regenerating forest and fallow lands. Microbial C was also high in the two-year old CAFS and fallow site; it was lowest in the seven-year-old CAFS. Microbial N was similarly lowest in the seven-year-old CAFS, but mature forest levels were equivalently low. Soil respiration was highest in mature and regenerating forest and the fallow area, while it was lowest in the CAFS. Lastly, labile C was highest in mature and regenerating forest, intermediate in fallow sites and seven-year old CAFS, and lowest in the two-year-old CAFS. These results lend partial support for our hypotheses and provide lessons for improved C storage in CAFS. We discuss these findings in turn.

Spatial autocorrelation and experimental design

A limitation of this study is the absence of independent site-level replication within each land-use category. Although multiple sampling points were established within each site, these represent subsamples rather than true experimental replicates. As highlighted by Hurlbert (26), such designs may introduce pseudoreplication and limit the strength of ecological inference. Therefore, our findings should be interpreted as evidence from the specific sites investigated rather than as representative of all coffee agroforestry systems or forest ecosystems in the region. Future studies should include multiple independent sites for each land-use category to better separate land-use effects from site-specific environmental variability.

Nonetheless, the absence of significant spatial autocorrelation among our sampling points confirms the independence of observations within land-use systems and supports the findings of our study. Differences among land-use types are not driven by spatial clustering but reflect genuine land-use effects, consistent with best practices in soil spatial analysis (29). The comparative design applied in this study, in which each land-use type was represented by a single site with internal spatial replication, is consistent with approaches commonly used in coffee agroforestry and forest regeneration studies (6, 38, 39).

Effects of land-use change on SOC and C stocks

Data on belowground C in CAFS is of global relevance, as coffee is cultivated across the tropics, and soils store more C than aboveground biomass (40).

The potential for AFS to store C depends on multiple factors including the native tree species planted, the focal crop species, the age of the plantation, planting density, and management practices (e.g., fertilization, crop residue treatments, and the harvesting regime; 41). On Mt. Gorongosa, C stocks were slightly lower in the seven- relative to the two-year-old CAFS, which contradicts typical expectations for C dynamics. In fact, other data from Mt. Gorongosa document a trend of SOC accumulation in shade-grown coffee over time, reaching 61.1 Mg ha −1 in 8-year-old plantations (42). Although the age difference between our seven-year-old coffee agroforestry system and the eight-year-old system studied by Magalhães et al. (42) is relatively small, several important ecological and methodological differences may explain the contrasting results. First, the two studies were conducted in different locations within the Gorongosa landscape, with distinct soil conditions, vegetation structure, and land-use histories. Magalhães et al. (42) investigated coffee agroforestry systems established in the buffer zone of Gorongosa National Park, whereas our study was conducted on Mount Gorongosa, an Afromontane ecosystem characterized by cooler temperatures, higher rainfall, and steeper terrain. Magalhães et al. (42) also quantified SOC to a depth of 1 m using five soil layers, whereas our study focused on the most biologically active upper 30 cm (0–15 and 15–30 cm), where management effects on microbial biomass and labile C are generally most pronounced. Therefore, differences between the two studies should be interpreted in the context of their distinct environmental settings, land-use histories, vegetation characteristics, and sampling designs, rather than being attributed solely to the one-year difference in the age of the agroforestry plots. Patterns in our data may be partially explained by soil conditions and external disturbances. For instance, site-specific factors such as soil texture, acidity, or cation exchange capacity can limit organic matter stabilization and nutrient retention (43). In addition, land-use history and legacy effects from previous cultivation, erosion, or fire exposure may have led to the long-term depletion of soil organic C prior to CAFS establishment (44).

The capacity of CAFS to retain substantial soil C stocks highlights their potential contribution to restoration and climate-change mitigation strategies. Agroforestry is increasingly recognized within international frameworks as a nature-based solution capable of enhancing C sequestration while supporting rural livelihoods (5, 11, 41). In Mozambique, where ecosystem restoration and climate adaptation are national priorities, shade-grown coffee systems may represent an effective land-use option for rehabilitating degraded agricultural landscapes, reducing pressure on natural forests, and contributing to national commitments under the Paris Agreement and restoration initiatives associated with the African Forest Landscape Restoration Initiative (AFR100).

Management practices, including the clearing of understory vegetation and removal of leaf litter, may also reduce organic inputs over time, constraining C accumulation in older systems (45). Some C may be lost at the time of establishment of an AFS, but these losses are usually followed by an accumulation of C as the AFS matures (46) unless leaf litter is cleared from CAFS plots. The decline in C stocks we measured may be due to management practices in our study site in which all dead organic matter is regularly rakes from plots. Overall, however, a meta-analysis of 73 agroforestry studies from sub-Saharan Africa indicates SOC increases in AFS (11), so with proper management CAFS hold valuable C storage potential.

The lower C stocks in the CAFS on Mt. Gorongosa relative to the fallow site may also be due the dominance of grasses and associated organic inputs in the abandoned field. Grass root biomass and exudates contribute to C sequestration in soils (47), allowing grasslands to store large amounts of C belowground (48). The presence of cattle (as was observed at our study site) adds N to the ecosystem, which also facilitates C sequestration (49). In addition, slash and burn agriculture, such as that which generated the fallow agricultural site, leads to the mineralization of cations and their incorporation into the soil, thus increasing C stocks in disturbed soils (50). Nonetheless, soil C in sites dominated by grasses may be largely labile (51), as observed on Mt. Gorongosa, and easily lost with disturbance.

Topography may have contributed to some of the observed differences among land-use types. The CAFS were located on relatively steep slopes (approximately 30-40°), whereas forest and fallow sites occurred on gentler terrain (<10°). Slope influences runoff generation, soil erosion, moisture retention, litter redistribution, and nutrient cycling, all of which affect soil C accumulation and microbial activity (52, 53). Consequently, some differences attributed to land use may also reflect topographic controls that could not be fully disentangled within the present study design. Future investigations should incorporate replicated sites across comparable slope classes or explicitly include topographic variables in statistical models.

Microbial biomass carbon and nitrogen

In general, the conversion of natural areas to agriculture reduces the amount of microbial biomass in the soil (54), while the restoration of agricultural sites may increase microbial biomass (55). Microbial biomass C is highly sensitive to litter inputs, root exudates, and organic matter quality (31, 43), and higher microbial biomass in forest regeneration reflects favorable microclimatic conditions and greater substrate availability (30, 31). Again, a different pattern emerged from our data; MBC was lowest in the seven-year-old CAFS and equivalent in all other land-use types. The low MBC in the seven-year-old CAFS may reflect changes in labile C resulting from land-use change (44) and may be due to management practices that limit organic inputs and create an unfavorable environment for microbial communities. The lower values observed in older coffee systems may indicate reduced organic matter turnover and altered microbial habitat conditions compared with regenerating and mature forests. Differences in the composition of plant communities may have also contributed to patterns in soil C and microbial properties. Tree species influence litter quantity, litter quality, root biomass production, and rhizosphere processes, which collectively regulate soil organic matter accumulation and microbial activity (8, 9). Consequently, part of the variation observed among land-use categories may also reflect differences in plant community composition.

Disturbance per se does not therefore predict the direction of changes in microbial communities, and habitat quality should be evaluated post-disturbance to understand how microbial biomass and associated C dynamics may change. MBC and MBN may also be related to factors such as the age of the agroforestry system, the species planted, and water availability (43, 55–58). For example, in forests, dense vegetation and high organic matter inputs increase soil moisture, promoting microbial activity in soils (59). The decline of both MBC and MBN with soil depth across all land-use systems highlights the close coupling between microbial biomass and SOC availability, a pattern widely observed in tropical and temperate soils (30).

Soil respiration as a reflection of microbial activity and C availability

Soil respiration was also lowest in the CAFS on Mt. Gorongosa. Similar to the low C stocks in the CAFS, this may be due to the clearing of vegetation between trees, resulting in lower biomass inputs than in forest and fallow land. Furthermore, soil respiration is often low in AFS. Reduced soil respiration results from tree cover that promotes microclimate stability and increases soil moisture; this microclimate stability contributes to reduced microbial activity (56, 60–62).

The CAFS respiration rates suggest active C cycling is supported by continuous organic inputs. As soil respiration represents a C flux rather than a C pool, its decline with depth mirrors reductions in microbial biomass and labile C availability (33, 63). Respiration is positively correlated with labile C (64), and the availability of labile C increases the flux of CO 2 to the atmosphere (65). In support of this relationship, both labile C and respiration were highest in the forest fragments and lowest in the two-year-old CAFS on Mt. Gorongosa.

Labile carbon as an indicator of short-term soil C dynamics

Labile C reflects recent organic matter inputs and is an early indicator of changes in soil C dynamics and soil quality (34). Labile C fractions were highly sensitive to land-use type and soil depth, with higher values in mature and regenerating forests and lower values in early successional systems.

The intermediate labile C levels observed in the fallow site and the seven-year-old CAFS suggest that greater organic inputs contribute to the maintenance of active C pools. Similar patterns have been reported in other agroforestry systems where management practices influence short-term C availability without necessarily affecting long-term SOC stocks (39).

Implications for coffee agroforestry management and soil C conservation

The intermediate performance of coffee agroforestry systems across C pools and biological indicators underscores their potential role in climate-smart land management. While agroforestry systems may not fully replicate forest soil conditions, they represent a viable compromise between agricultural production and soil C conservation (6). Differences in the C storage potential seen in our data and other studies indicate that management is key to maximizing the climate mitigation role of CAFS. By maintaining continuous vegetation cover and organic inputs, CAFS can contribute to the recovery of soil C stocks and sustain soil biological activity, particularly when compared to early successional or simplified agricultural systems (8, 38).

This study evaluated how contrasting land-use systems influence soil C dynamics and microbial functioning in a tropical Afromontane landscape in central Mozambique. Soil organic C concentrations declined with depth across all land uses, while regenerating forest maintained the highest soil C stocks and biologically active C fractions. Coffee agroforestry systems retained substantial soil C but generally supported lower microbial biomass and respiration than forest ecosystems. These findings indicate that coffee agroforestry can contribute to C conservation and partial restoration of soil functions in human-modified landscapes, although CAFS do not fully replicate the ecological characteristics of natural forests. Future research should incorporate replicated sites, longer-term monitoring, tests of management practices, and assessments of microbial community composition to better understand the mechanisms governing soil C recovery under agroforestry-based restoration.