1. Woody plant encroachment (WPE) is a global trend occurring in many biomes including savannas and accelerates with fire suppression. Since WPE can result in increased storage of soil organic carbon, fire management which may include fire suppression can enhance ecosystem carbon sequestration in savannas.
2. At our study site in Kruger National Park, South Africa, we used a long-term (~70-year) fire experiment to study the drivers and consequences of changes in woody (trees and shrubs) cover and soil organic carbon (C) sequestration. We surveyed four fire manipulation treatments replicated at eight locations within the park, annual high-intensity burn, triennial high (dry season) and low-intensity (wet season) burns and fire exclusion, to capture the range of fire management scenarios under consideration. Changes in woody cover were calculated over a period similar to the experiment’s duration (~80 years) using aerial photographs (1944–2018). Soils were analysed for soil organic C (SOC) and d¹³C to 30 cm, under and away from tree canopies to isolate local- and landscape-level effects of WPE on SOC.
3. The largest increases in woody cover occurred with fire exclusion. We found that plots with greater increases in woody cover also had higher SOC. Yet, trees were not the only contributor to SOC gains, such that sustained high inputs of C4-derived carbon (from grasses), even under canopies in fire suppression plots, contributed significantly to SOC. We observed little difference in SOC sequestration between the cooler triennial (wet season) burns and fire suppression.
4. Synthesis: Grass inputs to SOC remained high across the full range of woody cover created by varying burning regimes. Total SOC stocks stored from tree inputs only matched grass-derived SOC stocks after almost 70 years of fire exclusion. Our results point to C₄ grasses as a resilient contributor to SOC under altered fire regimes and further challenge the assumption that increasing tree cover, either through afforestation schemes or fire suppression, will result in large gains in C sequestration in savanna soils, even after 70 years.

Soil samples were obtained under and away from the canopies of five large (> 6 m) Sclerocarya birrea trees on each plot in 2016. Under-canopy samples were taken halfway between the tree trunk and the edge of the canopy. Away from canopy samples were taken away (> 5 m) from tree canopies. After litter was scraped away, a trowel was used to sample to 5 cm and, below this depth, a 4 cm-diameter soil auger was used to sample from 5-30 cm (at increments of 5–10 cm, 10–20 cm, and 20–30 cm) and soils kept separate for each depth. The soil was dry sieved through a 2 mm sieve to remove all roots after which a bulk sample was taken to make one sample per canopy and one away from canopy per plot per depth (two vegetation types ´ 4 replicates (strings) ´ 4 fire treatments ´ canopy/away from canopy ´ 4 depths = 256 samples). After soil sampling, soil texture was analysed at Elsenburg Laboratory (Western Cape, South Africa) following the hydrometer method for soil particle analysis (Committee, 1990).

Soil carbon and ¹³C/¹²C ratios of the soil carbon were determined using a Thermo Finnigan Delta plus XP mass spectrometer coupled with a conflo III device to a Thermo Finnigan Flash EA1112 Elemental Analyser with automatic sampler (Thermo Electron, Bremen, Germany). Although we did not expect to find carbonates in our study soils based on the pH values of the soils, we treated a subset (about 25%) of the soil samples with HCl to remove carbonates and reran these for variation in the C content, which confirmed the absence of inorganic carbon. These results were calibrated relative to Pee-Dee Belemnite as well as to correct for drift in the reference gas. The results are expressed as parts per thousand (‰) and relative to the Pee-Dee Belemnite standard are denoted by the term δ, with precision of duplicate analysis 0.1‰ (February et al., 2011). Based on the δ¹³C values of the soil and end member (mean) δ¹³C values of the grasses (-13.17‰) and trees (-27.61‰) at our study site (February & Higgins, 2010), a standard end-member mixing model was used to determine the relative proportion of C₃ (trees)- and C₄ (grass)-derived carbon in the soil. This mixing model was only applied to the 0–30 cm soil horizon (surface soils) because of unrelated fractionation processes at deeper depths causing enrichment of soil δ¹³C unrelated to the inputs from C₃ or C₄ derived carbon (Balesdent & Mariotti, 1996; Nel et al., 2018).

We calculated bulk density of the soil (Wigley et al., 2013) for each depth category and used this value to convert C concentrations to total C per volume of soil (i.e., stocks). Stocks of C per m² for each depth was calculated before summing these values and converting to total soil C per ha for soils from 0-30 cm deep (Mg ha^-1). We then incorporated the impact of localised enrichment of tree canopies on SOC by weighting our calculation of C stocks by the relative tree cover using the following equation:

plot total C (Mg ha^-1) = ((proportion woody cover ´ soil C_tree) + (proportion grass cover ´ soil C_grass)),

where grass cover equals 1 - woody cover.