Modeling for Functional Traits in Bluebunch Wheatgrass Using Unmanned Aerial Vehicle (UAV) High-Throughput Phenotyping

Changing climates and shifts in community composition in the Intermountain West threaten sagebrush-steppe ecosystems, impacting both plants and animals. Restoring native grasses like bluebunch wheatgrass (Pseudoroegneria spicata; BBWG) can help ecosystems resist invasive species and recover from stress and disturbances. As a native cool-season grass, BBWG is widely used in rangeland restoration for the western U.S. Breeding programs for restoration plant materials, such as seeds for ecological projects, often require years or decades of field testing while facing staffing constraints. High- throughput phenotyping (HTP), which rapidly measures observable plant traits, offers a faster solution using drones to capture time-series data. This approach helps create predictive models for BBWG traits, streamlining plant material evaluation in trials assessing growth and restoration potential. This research explores two key questions: 1) Can drones with multispectral cameras accurately count BBWG seedlings for material evaluation? 2) Can models built with drone images and on-the-ground data estimate BBWG traits across time and sites in large trials? We used field measurements and images captured by drones, including reflectance values (indicative of plant health), spectral indices (to highlight certain features in an image), and 3D maps of plant structure, to create two models. One model estimated plant traits such as biomass, leaf area index, canopy cover, and height, while the other estimated seedling presence. We used machine learning with decision trees to build the models, training them on 70% of the data and testing them on the remaining 30% to check its accuracy. Improving plant material selection is essential as climate stress and invasive species threaten BBWG and other perennial grasses and forbs in the Intermountain West. While our drone-based methods struggled to accurately estimate seedling presence, we successfully estimated key traits related to plant growth and ecological health, including canopy cover, leaf area index, and biomass. These traits reflect ground shading, photosynthetic capacity, and total productivity. These models enable efficient plant evaluation across locations and time, reducing costs and minimizing inconsistencies in data collection, which enhances genetic trait identification and supports the development of stronger, more adaptable species for rangeland restoration.