I have a dataset with different wheat lines in multiple environments. For each line various traits are measured (e.g. grain yield). I am trying to set up various genomic prediction models (linear mixed models) in order to predict these phenotypical traits based on the plant genomes, as well as the environment in which they are growing.
For the different environment, various environmental covariates (EC) were measured, like precipitation, temperature, etc. I would like to incorporate these covariates into the prediction models by creating a covariance matrix that represents the similarity between the different environments based on the different covariates. This would allow for the borrowing of information between the environments and genotype lines, when used in combination with a G-matrix.
This is based on a paper by Jaquin et al. (2014): https://doi.org/10.1007/s00122-013-2243-1 (See page 4: Extending G‑BLUP with addition of environmental covariates)
Consequently, the vector w = Wγ follows a multivariate normal density with null mean and a (2) yijk = µ + Ei + gj + εijk covariance matrix proportional to omega whose entries are computed as those of the G-matrix but using ECs instead of markers. This covariance structure describes the similarity between environmental conditions in a similar way that G describes genetic similarity between lines.
Here they mention that they compute the environmental covariance structure in the same way as the G-matrix. But I'm unsure how this is done, because the covariates are continuous and not frequencies or count data. What would be a feasible way to go about this? Or would it be better to just compute a 'normal' similarity matrix that is based on the euclidean distance for example?
