High-resolution mapping of soil carbon stocks in the western Amazon

Abstract

Global soil carbon maps are essential to understanding the global carbon cycle and supporting policy decisions, but their uncertainty in remote areas with limited data remains a significant challenge. Assessing and quantifying soil carbon at a regional level can shed light on existing uncertainties in global maps, providing more dependable information for stakeholders. Therefore, we aimed to model and map the soil organic carbon (SOC) stock in the Western Amazon, state of Rondônia, which witnessed a loss of 30% of its native coverage in the last 35 years. We used information from almost three thousand soil profiles that have not been included in the modeling of global and national existing soil maps. SOC stocks were stratified at 0–5, 5–15, 15–30, 30–60, and 60–100 cm depths and we selected the environmental predictors based on correlation < |0.95| and importance (recursive feature elimination). To reduce uncertainties and select the best model, we tested six different machine learning algorithms and ran the models 100 times for each depth with different subsets of samples, resulting in maps of mean, quantiles (Q05 and Q95), and uncertainty (Coefficient of variation). The Random Forest model presented the best performance, and the soil class was one of the predictors that most influenced the SOC stock, especially in the superficial layers (0–30 cm). The largest SOC stocks are in the southern region (protected areas), at low altitudes and places with seasonal flooding dynamics, and the smallest values are found in the central region, associated with the high-weathered Latossolos. Our results show that national and global maps overestimate SOC stocks (up +100 Mg C ha−1) in the central region with lower SOC stocks, and underestimate (− 400 Mg C ha−1) in the southern region with higher SOC stocks. These higher SOC stocks are concentrated in protected areas, but this information was hidden until now since global and national SOC maps could not identify these hotspots. Although global and national maps are handy for first assessment in locations without specific information, regional soil sampling and mapping should be used in hotspots of soil carbon sequestration and land use changes to ensure the continuous protection of these areas.