This project highlights one component of a broader portfolio of center-funded research initiatives focused on developing novel approaches to complex scientific challenges.
Center Update: Elucidating the Underlying Genetics of Increased Seedling Emergence in Zoysiagrass
Submitted by: Dr. Susana Milla-Lewis (Professor, Department of Crop and Soil Sciences))
The zoysiagrass seed industry has been historically small with only a couple of cultivars being commercially available. However, there is great demand for seeded cultivars by producers and consumers alike because seeds are easier to distribute and store, and are a more cost-effective means of establishment. In this project, we aim to understand the genetic components of seed yield and seedling emergence in zoysiagrass and to identify favorable alleles that can be introduced into the breeding pipeline. For this purpose, we are evaluating a panel of 268 diverse zoysiagrass genotypes for an array of traits including inflorescence abundance, seedhead and seed morphology, seedhead and seed yield, seed fill, and germination rate. Outcomes of this project will ultimately guide breeding efforts for development of commercial cultivars. Identifying genotypes with high seedling recovery will aid in setting up the correct germplasm base for future seeded breeding efforts.
Data on flowering time, inflorescence abundance, and seedhead color were collected on a weekly basis during spring 2025. New flowering events were recorded over three months with most genotypes flowering early, but a few being delayed over three months. While 73% of genotypes flowered, 71% of those had few seedheads. At the end of May, mature inflorescences were harvested and processed for evaluation of 25 different seed and seedhead traits. Significant variation was observed among genotypes for most traits (Figure 1). PhD student Balihar Kaur is currently performing seed fill evaluations via X-raying and germination experiments. Data on seed germination will ultimately inform the most useful traits for selection of individuals with high seedling emergence potential. Additionally, because visual evaluation of inflorescence abundance and seedhead traits is tedious and labor intensive, high- throughput phenotyping methods provide opportunities for improved data collection. A deep learning model was developed to automate seedhead counting. Under controlled environment imaging, the model produced a 93% correlation between image prediction and actual counts. Over summer 2025, PhD student Stefano Fratton worked on implementing the model under field conditions. Because of differences in lighting and angle, initial validation dropped the correlation to only 50.4%. However, further model training using field images increased that value to 57.6% (Figure 2). Further validation of the model will be conducted in 2026.
Figure 1. Histogram summarizing the distribution of number of inflorescences per unit area for the zoysiagrass germplasm panel.
Figure 2. Correlation between image-based estimates and actual manual counts for artificial intelligence-based models developed to predict seedhead counts. The initial model based on greenhouse imaging is depicted in orange, while the model further trained with field images is depicted in blue.
