Moving forward, we intend to generate a range of populations based on both phenotypic and genomic selection from this yield evaluation trial. These will then be evaluated alongside other elite material for DMY to assess if there has been any improvement. Due to the lengthy breeding process of perennial forages, it will take several years to determine whether these methods have been successful. To improve the predictive ability of the model moving forward a combination of evaluating a greater number of families and improving the quality of phenotypic data through better modeling will be imposed. Increasing the size of the training population could be facilitated without a significant increase in costs by using modern high-throughput phenotyping tools, such as dronebased remote sensing. With decreasing costs of genotyping, improved computational software and the availability of genomic resources , genomic selection is becoming increasingly available to more resource limited breeding programs like alfalfa. There is still much research required to assess whether actual yield gain can be achieved; however, these studies provide a baseline for future studies to investigate potential yield improvement.Yield is the most important trait for profitable forage production, yet the rate of genetic gain for dry matter yield in perennial forage crops is lower than the main cereal crops and has been essentially zero in alfalfa over the past 30 years . Limited resources, low heritability, square black flower bucket significant genotype by environment interaction and long selection cycles limit the rate of genetic gain in perennial forages in comparison to many annual food and feed crops .

Improvement of perennial forages is typically carried out through recurrent phenotypic selection with or without progeny testing to accumulate desired alleles at high frequency in a population . Ideally the number of families to be evaluated is very large, particularly with the advent of modern breeding methodology such as genomic selection. In reality, breeders must strike a balance between the available resources and the size and scope of breeding trials. Phenotypic evaluation of perennial forage traits requires significant investment of land, labor, and capital. Forage DMY and dormancy in alfalfa are two crucial traits that require significant resources to phenotype. The standard test for fall dormancy in alfalfa requires height measurements for each trial entry 25-30 days after the final harvest, often across multiple environments and years . To accurately assess forage yield, experimental units must be harvested, dried,and weighed to estimate dry matter content across multiple harvests and years, resulting in up to 40 total harvests over the lifetime of a trial . Further complexity is added to the breeding of perennial forages considering the diversity of evaluations trials often used, ranging from single plant evaluations to transplanted rows, seeded rows, or solid seeded swards. The choice depends on the traits of interest, the number of genotypes or families being evaluated, seed quantity, and the capital and labor resources available to the breeder, with most programs using a combination of sown and transplanted trials . Transplanted family rows are the most common as they are a cost-effective method of evaluating large numbers of trial entries for traits with high heritability. They are commonly used to screen populations for resistance or tolerance to various pests and diseases, investigating growth habit, dormancy, flowering time, and forage quality . In the past, forage yield has often been selected indirectly based on evaluation of vigor on spaced plants or short family rows . Although a useful method for evaluating other important, highly heritable traits, a poor correlation exists between these assessment methods and yield in a commercial setting . Large sown plots are commonly used for variety trials.

These trials require large quantities of seed, cover a large area, and provide phenotypic data for relatively few trial entries. In a breeding program, these trials are typically used to compare advanced breeding populations to released cultivars for key traits such as stand establishment, DMY, forage quality, flowering time, and dormancy. Although useful for obtaining phenotype data that well represents a commercial forage operation, it is usually not feasible to evaluate hundreds of families in this way. Transplanted mini-sward plots provide a compromise between family rows and large sown plots. They seek to provide a better estimation of forage DMY than family rows without the need for large quantities of seed or significant land area that large sown plots require. In these trials asmall number of plants are planted close to one another to mimic the competition observed in commercial forage stands. In recent decades, remote sensing has been widely adopted in agricultural research , offering a plethora of non-destructive vegetative data with massively reduced labor requirements. Remote sensing has the potential to address the lack of yield improvement in alfalfa and increase the rate of genetic gain for yield in other perennial forages by enabling breeders to greatly increase the size of trials without the associated increase in labor costs. This is particularly important for breeding programs looking to use genomic selection, where the size of the training population is a key component of predictive ability . Remote sensing techniques have been shown to enable accurate estimation of biomass yield in alfalfa at the field level , and at the large plot level in breeding trials . However, its accuracy has not been widely reported across the range of plot types used in forage breeding or for estimating fall dormancy in alfalfa. The overall objective of this research project was to assess the accuracy of drone-based remote sensing versus traditional phenotyping for forage biomass yield and alfalfa fall dormancy across a variety of plot types used in perennial forage breeding. The goal is to give breeders the ability to evaluate a wider range of material without the associated increase in labor and costs. In addition, we aim to provide recommendations for researchers looking to incorporate similar technology into their breeding programs.

This experiment was carried out across several trials previously established as part of the UC Davis forage breeding program located on the UC Davis Plant Sciences Farm in Davis, CA on a Yolo silt clay loam . It is a Mediterranean environment with hot, dry summers, cool winters and moderate annual rainfall which falls predominantly in the cooler months from November-March . Soil tests were conducted prior to planting to adjust P, K, and pH according to soil test recommendations. The trials consist of three alfalfa breeding trials and a forage grass variety trial.The trial consisted of 72 released cultivars, experimental cultivars, germplasm populations, and eleven standard test check cultivars . Plants were germinated in 128-cell flats in the greenhouse in February before transplanting to the field in April 2018. This experiment consisted of four replications laid out in a randomized complete block design. Plots consisted of a single row of 25 plants spaced 30 cm apart with a 90 cm gap between plots and 60 cm spacing between rows. Fertilizer was applied to maintain P and K at appropriate levels for a high yielding alfalfa stand, with weeds and insect pests monitored and control measures applied when necessary. Plants were initially watered using sprinkler irrigation until fully established, following which they were flood irrigated to satisfy full evapotranspiration requirements.This trial contained 80 half-sib families of an experimental population UC2588 that had been selected for tolerance to lygus feeding. We had had sufficient seed of each family to plant solid seeded plots. This experiment was established following the NAAIC standard procedures for variety yield trials . It consisted of two replications laid out in a randomized complete block design with ten rows and twenty ranges. Plots were 1 m x 3 m and were drilled using a small plot planter at a seeding rate of 15 kg ha-1 with 1.5 m gaps between ranges. UC Impalo was sown as a border between ranges and around the exterior of the trial. As with the 2018 dormancy trial, crop nutrient demand, weeds and pests were monitored and adjusted when necessary. Sprinklers were used immediately after sowing to get the trial established, square black flower bucket wholesale followed by flood irrigation to meet water demand.This trial included a total of 198 entries of which 193 were half-sib families from two closely related elite UC Davis populations derived from various UC Davis germplasm that underwent selection for root rot and other stresses in El Centro and Davis, California. In addition, three cultivars: Highline, UC Impalo and CUF 101 were included as repeated checks and the remaining two entries were balanced bulks from each of the two populations . The trial was sown in the greenhouse in March 2020 and transplanted two months later in early May at two locations on the UC Davis research farm in Davis, California. Each site has the same layout consisting of two replicates with 7 rows and 29 ranges for a total of 203 plots per rep, 812 plots overall. Plots consisted of 24 plants laid out in a regular 4 × 6 grid with 20 cm spaces between plants. There was a 30 cm space between rows and a 110 cm space between ranges to allow room for mechanical harvesting.

This trial was managed as a high-yielding alfalfa stand, soil tests were conducted each year, with amendments made accordingly. The trial was established using sprinkler irrigation, which was switched to flood irrigation after plants were well established. Irrigation water was added to roughly match crop ET. Weeds were managed by a combination of manual removal and herbicides, and insect pests were monitored and controlled with insecticide application as necessary, primarily for alfalfa weevil control in spring.A grass variety trial containing 88 cultivars was sown in October 2020. Plots are 1.5 m x 4.5 m and were drilled using a small plot planter. Table 1 outlines the seeding rates used for each species. The trial was separated by species with two blocks of tall fescue, two blocks of orchard grass, one block of timothy and reed canary grass, and the remaining species in the final block. Each block contained four rows of plots with 14 ranges. The blocks are separated by borders of either tall fescue, timothy, or orchard grass to allow irrigation pipes to be laid across the field without lying on top of the plots. This trial was irrigated by sprinklers on a weekly to biweekly basis as needed to approximate ET demand. N, P and K levels were monitored, and fertilizer applied when necessary. N was applied at 100 kg ha-1 in spring and again after first harvest.Plant height measurements for the alfalfa fall dormancy standard test were measured following the protocol outlined by Teuber et al. . Twenty-five days after the final fall harvest, the natural plant height was measured on each of the 25 plants per plot. Natural plant height was deemed to be the distance from the soil surface to the top of the tallest stem as the plant stands in the field . The measurements were then averaged over the whole plot to generate a single data point for each plot. Biomass yield data were collected using a small self-propelled plot harvester. Harvests occurred in alfalfa when the field had reached 10% bloom with the first harvest usually occurring in late March/April and the final harvest in October. For alfalfa trials, subsamples were taken during each harvest, weighed wet, dried for at least 4d at 60C, and weighed dry to adjust moisture percentage. Several subsamples were taken from each replication as composite samples from all entries, rather than for every entry, and the average dry matter was used to adjust the wet weights. In the grass trial, harvests occurred when the most plots of tall fescue and orchard grass had reached the late boot stage, with the first harvest in April and subsequent harvests every 6-8 weeks for a total of four harvests per year. All species were harvested at the same time for logistical reasons, even though this was likely not ideal for individual species . All forage was clipped uniformly at 7.5 cm, weighted, and removed from the trial area. Subsamples were taken from every plot in the grass trials, weighed wet, dried for at least 4d at 60C, and weighed dry to adjust moisture percentage.Prior to remote sensing data collection, the borders surrounding the trial and between plots were mown. Drone flights and preliminary image processing were conducted following methods modified from Parker et al. .