Modeling temporal genetic variability using mixed models improves yield stability and selection efficiency in

Introduction: Increasing climatic variability challenges Coffea canephora breeding programs to identify genotypes that combine high productivity with temporal stability across contrasting seasons. Methods: We evaluated 44 genotypes across four consecutive crop seasons (2022-2025) in eastern Minas Gerais, Brazil, using mixed linear models (REML/BLUP) with seven alternative variance-covariance structures for genetic and residual effects. Results: The flexible model M7 (unstructured genetic covariance matrix with year-specific residual variances) provided the best fit (lowest AIC and BIC). Plot-level heritability ranged from 0.64 to 0.68, genotype-mean heritability from 0.84 to 0.86, repeatability was 0.89, and selective accuracy ranged from 0.92 to 0.94. Genetic correlations revealed atypical behavior in 2023, driven by heat stress during grain filling. Relative selection efficiency increased cumulatively by 4.4% after four years. Genotypes Bicudo, A1, and AD1 combined the highest predicted genotypic values with elevated persistence indices. Discussion: Flexible mixed model approaches improve the reliability of genetic evaluation and support resilience-oriented selection strategies in C. canephora, enabling accelerated genetic gain and identification of superior genotypes adapted to variable cultivation conditions.