AG2PI Field Day #29 - Wednesday February 21, 2024

The short answer is maybe. The interesting thing about Phenomic data (beyond phenotype data) is it captures G, E, GxE, and more complicated G-effects like epistasis. We believe that environment data will still help, but our examples to draw on so far have provided little gain over phenomic data. But environment data helps a lot if you only have genomic data. However environmental data can definitely still help understand what is going on in phenomic data in response to the weather. It is definitely an exciting frontier! For Genomes to Fields (G2F), each location has a weather station supported by Iowa Corn. However, things like NASA power data is used to supplement although reviews are mixed and some collaborators think NASA power still has too many serious issues to rely on for this work.

We started with plant height in 2016 because it is highly correlated with yield (0.5-0.7) in our Texas environments. Our ag engineering colleagues also told us that it was easy. Four years later we finally had a handle on it (it was not as easy to operationalize in a breeding program as the previously published one off demonstration articles had made is seem). Yield is obviously super complicated. We need to crawl before we run. The best approach to predict yield from drone data is phenomic prediction. The insights we have gained in the interim from plant height to phenomic prediction of grain yield have been transformative! To date only a few of us have attempted this - lots will be coming out in the next three to five years based on conversations with breeders interested in adopting these tools. We have chosen a data set (2020-2021 G2F) that uses common genotypes across environments only to eliminate different genotypes as a confounding variable. Using the same genotypes does NOT allow to standardize/correct for the different dates of flight, because the phenome is a function of G+E+GxE+other measurable stuff+error unique to each minute, hour and day. The way we are attempting to standardize so environments can be compared is using modeling, Area under curve (AUC), and/or Functional Principal Component Analysis (FPCA - pretty different from PCA because of time addition).

Wind is a manageable problem. There are also two software packages that are non-orthomosaic based. Progeny/Plot Phoenix clipped plots - but that software is no longer available, and Hiphen as a fee for service (currently infeasible to scale in public sector). The non-orthomosaic approaches will be much more important as we move to single plant analysis as Dr. Jacob Washburn demonstrated at the North American Plant Phenotyping Network (NAPPN) meeting last week. Again, this is a research frontier and there are far more questions than answers.