Scalable Assessment of Soil Organic Carbon in Transitional Agricultural Landscapes using Multispectral UAV Remote Sensing

Scalable monitoring of Soil Organic Carbon (SOC) is imperative for climate-smart agriculture but limited by traditional sampling costs. The objective of this study was to evaluate the efficacy of UAV-mounted multispectral sensors to model SOC stocks across Integrated Crop-Livestock (ICL), Livestock-Forestry (ILF), and Native Vegetation in the Amazon-Cerrado ecotone. Soil samples (0-30 cm, n=20/area) were collected on a georeferenced grid aligned with satellite resolution. High-precision multispectral imagery was acquired using a DJI Matrice 350 RTK equipped with a MicaSense RedEdge-P sensor to derive vegetation indices (NDVI, EVI, SAVI). Statistical analyses in R included ANOVA and regression modeling. Results indicated that the ILF system significantly maximized SOC stocks (86.4 ± 2.8 Mg ha⁻¹), surpassing Native Vegetation (49.9 ± 2.3Mg ha⁻¹) and ICL (53.5 ± 2.5 Mg ha⁻¹) by over 55% (Tukey, p<0.05). Regarding modeling, while ICL showed no significant spectral correlations, the ILF system exhibited robust relationships between SOC and vegetation indices (NDVI/EVI r> 0.60). Consequently, the multiple regression model for ILF achieved high predictive accuracy (R2 = 0.81; MAPE = 5.13%), validating the arboreal component's role in enhancing both carbon sequestration and spectral detectability. We conclude that spectral modeling accuracy is positively associated with SOC levels, suggesting that the adoption of conservation systems is a prerequisite for the operational scalability of large-scale mapping technologies. Although future studies with increased sampling density are recommended to refine robustness, this protocol offers a viable tool for Monitoring, Reporting, and Verification (MRV), supporting global low-carbon targets.