SeedLinked reposted this
Over the past year, we’ve been iteratively stress-testing how participatory on-farm trials can support robust GEI analysis, using tomatoes as a pilot at SeedLinked. This work was made possible thanks to our team Ashok Kumar, Erik Kastman, and Hans Boot, and through trials from the ADAPT program at Seed Savers Exchange, internal SeedLinked trials, and USDA seed bank trials made possible by Sarah Dohle. Starting from a simple G + E model (Y ~ Variety + Site), we progressively added climate structure, data quality controls, and season-specific stress metrics. Each step was evaluated empirically, not assumed. A few concrete results from ~2,000 tomato trial observations (2022–2025): • Variety effects alone explained ~8–14% of trait variance, confirming strong genetic signal • Adding climate as a main effect explained only ~0.2–0.4% • Introducing Variety × Climate interactions increased explained variance by ~5–7% • Cleaning and imputing phenology dates with crop-specific bounds increased detected significant interactions by +23% • Moving from full-year climate to season-specific PRISM daily data resulted in a ~16× increase in significant V×Climate interactions (from 1 to 16 detected) The biggest lesson for us: GEI signal is highly sensitive to when climate is measured and how data quality is handled. Season-specific stressors such as freeze-thaw cycles and heat dynamics consistently outperformed composite annual indices. This reinforces a core principle of participatory breeding: community-generated data can meaningfully inform adaptation and variety choice, but only if trust, phenology, and environment are treated as first-class components of the analysis. As Jacob van Etten told us more than five years ago: “You are sitting on top of a gold mine. The day you build the data analysis capacity, you will realize the power of insights shared by the community.” We are now expanding this framework beyond tomatoes into 10 key crops through hundreds of curated trials. If you are a breeder, researcher, or grower working with on-farm or participatory trials and would be interested in giving feedback or pressure-testing these assumptions, we would love to connect. More broadly, this work is a step toward a predictive seed search for specialty crops and niche markets, grounded in crowdsourced on-farm data and explicitly modeling genotype, environment, and stress rather than relying on static descriptions or single-location trials.