In multi-environment trials analysis, the Finlay-Wilkinson model is popular for describing genotype-by-environment interaction. A Bayesian approach allows straightforward parameter estimation for this non-linear model and also avoids overfitting by placing distributional constraints on genotype-by-environment effects. However, existing methods fail to account for environment-specific covariates and within-field spatial correlation. In this talk, we introduce a Bayesian hierarchical Finlay-Wilkinson model that incorporates genetic, environment and spatial information. We propose novel constraint on the intrinsic spatial prior that alleviates an identifiability problem. We show practical techniques for generating predictions and making inferences with publicly available Genomes-to-Fields data.