Abstract Purpose Sustainable agriculture requires both high and stable crop yields. Whilegenotype-environment-management (G×E×M) interactions influence yield stability, implementingsuch understanding into practical applications demands better analytical tools. Yield StabilityZones (YSZ) effectively identify stable and unstable production areas, yet their implementationhas been constrained by data limitations and interpretability challenges. Precision agriculturenow enables the application of YSZ approaches through multi-year yield and management data,while interpretable machine learning (ML) can decode yield drivers into actionable insights. Thisstudy develops a universal framework integrating YSZ and interpretable ML to enhancedecision-making in variable agricultural environments, using citrus production as a case study. Methods Analysis of five-year yield, soil, and rainfall data (2012–2016) from a 250-ha field todevelop an YSZ framework, assess temporal yield stability and interactions by ‘comparing single-year versus multi-year data’ on a real production scenario, and integrate machine learning(decision trees) to promote interpretation of yield factors and support optimized cropmanagement. Results Significant temporal dynamics in soil-yield interactions was found. Single-year assessments fail to capture critical interannual variability in yield drivers. YSZ effectivelydelineated spatially consistent production areas, distinguishing stable high-yielding zones fromunstable regions, while decision trees identified key drivers of yield variability. Conclusion Together, these tools provide a data-driven approach to optimize crop production sustainably.Our methodology bridges a critical gap in crop analytics and offers scalable insights forprecision agriculture under dynamic production systems.
Wei et al. (2026) studied this question.