The Enviromic marker
Gustavo E. Marcatti and Rafael T. Resende
Abstract: We formalize “enviromic markers” as modeling units parallel to DNA markers, but herein for genotype–environment (G × E) prediction. Four operational premises (linearity; site potential; heterogeneous favorability; and envirotypic covariates (ECs)–genotype-dependence) are presented to enable their use in linear mixed models and also to motivate four construction strategies: (i) using raw environmental covariates as linear markers; (ii) applying transformations to capture mild nonlinearities; (iii) deriving ecophysiological functions; and (iv) engineering markers with AI models which learn nonlinear environment phenotype mappings for linear downstream use. Environmental data quality control is detailed, including checks of spatial coverage and resolution, variance within the TPE, collinearity control, and spatial/temporal validation without leakage. Envirome data are linked with GIS to compute environmental kernels, quantify covariate shifts, and deliver pixel-level predictions with uncertainty diagnostics. The framework clarifies assumptions and standardizes the use of enviromic markers for predictive breeding analyses.