The Amazon rainforest is one of the world’s largest terrestrial above- and below-ground carbon (C) pools (Feldpausch et al., 2012; Ometto et al., 2014). With intensifying land-use change and severe drought, C emissions due to fire are increasing (Aragão et al., 2018), thereby reversing a multi-decadal C sink (Phillips et al., 2009; Brienen et al., 2015; Feldpausch et al., 2016). The central and northwestern regions of Amazonia, with high annual precipitation and short or no dry season, are resistant to burning in non-drought years (Pontes-Lopes et al., 2021), but in the transitional forests of southern Amazonia, a combination of lower annual precipitation, a pronounced dry season, and proximity to flammable Cerrado vegetation, increase the probability of fire in the presence of ignition sources (Balch et al., 2008; Nogueira et al., 2019; Prestes et al., 2020). The conditions promoting fire and the proportion of closed-canopy forest cover have varied over time, with palaeo-records for south and southwestern Amazonia indicating drier conditions during the mid-Holocene (Burbridge et al., 2004; Kury et al., 2021), followed by an expansion of closed-canopy forest over the past 2000 years (Burbridge et al., 2004).

Across the southern Amazon forest-Cerrado zone of ecological tension (ZOT), therefore, fire has been a long-term driver of vegetation dynamics (Mayle et al., 2000, 2007; Kury et al., 2021) and has contributed to the transformation of a proportion of above-ground biomass (AGB) into charcoal (Koele et al., 2017). The development of soil pyrogenic carbon (PyC), a soil organic carbon (SOC) fraction formed by the incomplete combustion of plant biomass, represents a SOC fraction resistant to decomposition that may persist for millennia (Bird et al., 2015). Determining the drivers of variation in soil PyC stocks is important for understanding past fire history, the contribution of PyC to total SOC stocks, and determining SOC sensitivity to current and future fire regimes and climate change.

Global estimates suggest an annual PyC production of 50–270 Tg (Kuhlbusch and Crutzen, 1995; Wei et al., 2018), with vegetation burning being responsible for up to 28 Tg year^–1. Estimated inputs of PyC by atmospheric deposition from fossil fuel and biomass burning aerosols in Amazonia (∼6 kg km^–2 yr^–1) are low (Coppola et al., 2019), indicating many thousands of years would be required to develop the current Amazonian soil PyC stock, estimated to be 2.67 Pg over 0–100 cm soil depth (Koele et al., 2017). Therefore, the majority of soil PyC found in forests with no history of recent fire is likely due to in situ historical fires and/or local edaphic or topographic conditions favoring the long-term accumulation of soil PyC. Radiocarbon dating of macro-charcoal fragments from the soil indicates Amazon forests have periodically burned throughout the Holocene (Sanford et al., 1985; Goulart et al., 2017; McMichael and Bush, 2019; Feldpausch et al., 2022), with some old-growth forests having fire return intervals of hundreds of years and having experienced no fire for the last ∼800 years (Feldpausch et al., 2022).

The PyC fraction of SOC is more resistant to degradation than other SOC forms due to its refractory polyaromatic composition, as well as potential interactions with the clay mineral matrix through the formation of organo-mineral associations (Czimczik and Masiello, 2007). However, there is still uncertainty about how long PyC persists in soils, or how global warming will affect soil PyC stocks. Some studies have shown that most of the larger PyC particles remain in the soil after fire (Suman et al., 1997; Gustafsson and Gschwend, 1998; Masiello, 2004). Depending on the temperature of formation and organic matter of origin of PyC, mineral phase, and soil texture, PyC can remain in the soil much longer than other forms of SOC, and in some cases for thousands of years (Singh et al., 2014; Bird et al., 2015). Dissolved and fine particulate PyC fractions may be translocated to deeper soil depths and lost to erosion over years to decades, especially with heavy rainfall (Major et al., 2010), and these effects may be greater in sandy soils and forests with steeper slopes (Rumpel et al., 2006).

The ZOT between the two largest Brazilian biomes, the Amazon forests and the Cerrado, is located in a seasonal climate, with a dry period of up to 6 months, average temperature above 25°C, and annual precipitation predominantly between 1400 and 2000 mm (Silva et al., 2008). While an increase in temperatures with global warming (IPCC, 2018) is likely to increase the rate of respiration of SOC to the atmosphere (Davidson et al., 2006), for recalcitrant PyC fractions losses could be much lower since PyC stocks are more resistant to degradation in response to warming (Lehmann et al., 2008). The more open vegetation and the low stratum that dries during the dry season, that are typical of the Cerrado (Ribeiro and Walter, 2008), favor the entry of fires in transition forests at higher frequency than moist forests (Balch et al., 2008) and therefore, likely the production of PyC. Although these seasonal evergreen forests show relative fire resistance, this resistance significantly changes with an increase in fire frequency, and intensity, exacerbated by extreme weather events and land-use change (Balch et al., 2008; Brando et al., 2020). Thus, these factors, combined with the precipitation gradient and differences in vegetation structure such as height (Barbier et al., 2010; Feldpausch et al., 2011), biomass (Fauset et al., 2015), and composition (Morandi et al., 2016) across the transition may interact to affect soil PyC stocks.

Some factors could also act ambiguously to affect SOC and PyC stocks across the ZOT. For example, AGB stocks are generally greater with higher precipitation, providing more fuel for fire and potentially more charcoal during burning (Fearnside et al., 2001). However, higher AGB and higher precipitation can maintain higher sub-canopy moisture, preventing combustion of dead leaves and small branches that act as fine fuel and thereby reduce the capacity of these components to carry a fire (Ray et al., 2005). Likewise, soils with more clay can protect PyC from decomposition through the formation of organo-mineral compounds (Glaser et al., 2003); however, the lower moisture content (Metcalfe et al., 2007) of sandier soils may slow the decomposition of SOC and PyC. Therefore, a combination of edaphic, climatic, and vegetation factors may interact to affect soil PyC. Thus, we hypothesize that forests located in regions with lower annual precipitation, combined with more clayey soils, will have higher PyC than sandy soils located in regions with higher precipitation. Furthermore, we hypothesize that PyC stocks will vary with AGB, as AGB directly reflects the amount of material that can be converted to PyC during burning (Fearnside et al., 2001; Preston and Schmidt, 2006). Our study aims to evaluate the interaction of multiple drivers in controlling the distribution and stocks of forest PyC across the Amazon-Cerrado transition. We address the following questions: (1) How much soil PyC is stored in forests across the Amazon-Cerrado transition? (2) How do PyC stocks relate to soil texture, forest AGB, and climate? (3) How does the distance of each sampled location to the current Cerrado biome border influence forest soil PyC?

The PyC stocks in the soils of the forests spanning the ZOT ranged between 0.9 and 6.8 Mg ha⁻¹ to 0–30 cm depth (with an average of 2.3 ± 1.5) (Figure 2), with PyC representing 4.3% of the total soil organic carbon (SOC) to 30 cm depth. The most parsimonious model, based on AICc, included soil clay and above-ground biomass as the main predictors of PyC, and these together explained 71% of the variation [F_{(2, 16)} = 20; r² = 0.71; p < 0.001, intercept = −2.01] (Table 2). When evaluating these variables in isolation, only clay explained a significant part of the model variance, however, with a loss in the explanatory power [F_{(1, 17)} = 13.46; r² = 0.44; p = 0.001, intercept = 0.30] and in the estimative value, of 0.07 to 0.06 Mg of PyC for each unit of clay in the model.

Next, we evaluated the effect of removing the ADE plots from the models. As these sites are known to have been altered by pre-Columbian peoples, the ADE plots represent outliers compared to the PyC stocks found in non-ADE soils, and non-ADE plots represent the vast majority of soils across the ZOT. After the removal of the ADE plots, soil PyC stocks ranged between 0.9 and 3.8 (mean 1.9 ± 0.8 Mg/ha PyC) to 0–30 cm depth. The PyC, on average, continued to represent ∼ 4% of the total soil carbon inventory to 30 cm depth (Supplementary material 1). The most parsimonious model included soil sand content and biomass as the main predictors of PyC and these explained 72% (r²) of the variation. When evaluating these variables in isolation, only sand explained a significant part of the model variance (Table 2), with a slight loss in the explanatory power and estimation (−0.81 to −0.004) of the model [F_(1,14) = 26.14; r² = 0.65; p < 0.001, intercept = 4.42, Table 2 and Figure 3]. That is, there is a gain in the predictive power of PyC when evaluating the sand in the presence of the variable biomass; however, the variable sand has a predictive capacity three times that of biomass to explain the amount of PyC.

None of the components (climatic, environmental, or edaphic), pure or shared, was important in explaining soil PyC variation with or without ADE plots based on the partial regression models (Figures 4A,B and Table 3). However, when we removed the ADE plots, there was a small increase in explanatory power of the pure edaphic component (Figure 4B). The fractions not explained by the models (residuals) were relatively high (Figures 4A,B).

Studies over the past decade have begun to reveal the effects of historical fire on structurally intact forests with no history of recent fire throughout the Amazon Basin (Feldpausch et al., 2022), including higher than predicted PyC stocks (Bird et al., 2015; Koele et al., 2017). However, there remains large uncertainty about how soil PyC varies spatially (Carvalho et al., 2018), and the drivers of spatial variability, especially throughout the ZOT where fire risk is greater and fire has potentially affected vegetation and soils for millennia. Here, we show that soil PyC storage in the southern Amazon ZOT varies approximately eight fold (Figure 2), at local and regional scales of 1540 km, and these stocks are, on average, similar to humid regions of Amazonia where historical fire may have been less prevalent (Koele et al., 2017). Edaphic factors were one of the strongest predictors of soil PyC stocks, which follows expectations based on studies of biochar application and the capacity of clay to protect PyC through the formation of organo-mineral complexes (Czimczik and Masiello, 2007). Although there was substantial variation in T_A (4°C) and P_A (900 mm/yr) across the geographical focus on the ZOT, which are known to affect soil respiration rates (Davidson and Janssens, 2006; Lehmann et al., 2008), these variables were unimportant in predicting soil PyC. The inferred main drivers of soil PyC stocks differed when including the ADE plots rich in PyC, with biomass and soil clay (−88 correlation with sand, Supplementary material 3) content then being important drivers. This was due to biomass stocks on average being higher in ADE than the surrounding non-ADE forests in southern, but not eastern, Amazonia (Oliveira et al., 2020).

This study shows that there is only a moderate amount of PyC (∼ 4% of total SOC) in the forest soils in the southern Amazonian ZOT. A much greater fraction of soil PyC in the ADE plots reinforces the importance of long-term conservation of ADE areas as a recalcitrant carbon stock. Soil PyC stocks in non-ADE forest plots are largely driven by edaphic properties. A sharp reduction in soil PyC in sandier soils and higher PyC in clayey soils highlights the physicochemical role of soil in the mineralization or the protection of soil PyC and its loss from the upper soil by illuviation to deeper soil levels. Biomass combined with edaphic variables explained part of the soil PyC variation, but only when including ADE forests. It will be important to include new forest types and soil samples at greater depths to more fully understand the variation observed in this study. Our surprising findings that soil PyC stocks are no greater in the ZOT than in the wetter regions of Amazonia, and that PyC stocks are not greater near the fire-dependent cerrado, require further investigation to understand variation in past fire history and factors that affect PyC formation and loss. This study improves understanding of the drivers of PyC across the ZOT, providing important new information to manage and conserve soils and to improve models to assess the impact of climate change on SOC.

The original contributions presented in this study are included in the article/Supplementary material, further inquiries can be directed to the corresponding author/s.

EO, TF, BM, and BM-J designed the study. EO analyzed the data. EO, AM, CAQ, LA, and PM analyzed the sampled soil. MB analyzed the soil PyC. CAQ analyzed the soil physicochemical properties. BM, PM, LA, OLP and AM contributed to the plot vegetation data. EO wrote the manuscript with TF, BM, and BM-J and with input from all co-authors. All authors contributed to the article and approved the submitted version.