Integrating Enviromics to Predict Performance and Guide Clonal Deployment in Eucalyptus spp.

ABSTRACT Genotype by environment interaction (G×E) remains a central challenge for tree breeding, as it is difficult to extrapolate trial results to untested sites and complicates confident genotype deployment. Enviromics, by integrating environmental covariates into predictive models, offers a way to overcome these limitations and guide clonal deployment. This study evaluated an enviromic framework applied to 15 Eucalyptus spp. clones across 5,189 inventory plots located in southern Brazil. 10,000 Engineered Enviromic Markers (EEMs) were built from 3,869 soil, climate, and remote sensing covariates, using random forest and integrated into a mixed-model ensemble to predict mean annual increment standardized at seven years (MAI7) across the Target Population of Environments (TPE). Predictive accuracy was assessed through a Leave-One-Region-Out cross-validation procedure. The enviromic model achieved a higher performance than a baseline G×E model, with Pearson and Spearman correlations above 0.90 and a root mean squared error (RMSE) of 3.09, compared to 0.45–0.39 correlations and RMSE of 8.73 for the baseline. Spatial predictions enabled the delineation of breeding zones that minimized G×E, while also revealing regions with high discriminant power for testing new genotypes. We also applied a two-step clonal deployment procedure combining enviromic predictions with a frost-risk penalization map, refining recommendations for frost-prone areas. When comparing recommended versus planted clones in inventory plots, the framework indicated an average expected productivity gain of 13.4%. These results demonstrate the potential of enviromics as a decision-support tool for clonal deployment, enhancing productivity while accounting for environmental risks, and paving the way for future multi-omics integration.