Evaluación de un modelo isotópico para rastrear el origen geográfico de la soja en Brasil

The expansion of soybean cultivation in Brazil highlights the need for robust, auditable traceability tools. We present an end-to-end workflow for plant-origin assignment encompassing sample collection, stable isotope analysis (δ¹³C, δ¹⁵N, δ¹⁸O, δD), isoscape construction, probabilistic assignment, cropland masking, and external validation with blinded samples. Predictive maps were generated using regression-kriging. δ¹⁸O and δD showed strong spatial structure suitable for provenance inference, whereas δ¹³C and δ¹⁵N exhibited weak spatial autocorrelation and limited interpolative value. In blinded tests, site-level log₁₀BF for 50-km circles indicated reliable origin confirmation mainly in the north-northeast and southern Rio Grande do Sul. Performance in central-southern Brazil was weak, with log₁₀BF near zero or negative and frequent exclusion of true locations. Even in successful regions, false positives persisted, particularly at lower probability thresholds. Although not yet ready for operational deployment, the workflow establishes a transparent basis for isotope-based origin verification and a roadmap to improving deforestation-free traceability.