A Micasense RedEdge-M multispectral camera was mounted to a DJI Matrice 100 drone to collect field imagery. Flights occurred at solar noon the day prior to destructive harvests, and images were captured automatically using DJI Ground Station Pro. The drone was flown at an altitude of 15 m with 80 percent overlap between images. Raw images from each flight were built into field-scale orthomosaics and digital surface models using Pix4Dmapper version 4.3.31. The processed orthomosaics were loaded into QGIS version 3.22.1 for extraction of data at the plot level. First a grid layer was generated using the create grid function of QGIS overlaying each plot . Next a threshold layer was generated from the NDVI orthomosaic to differentiate between plants and soil. This threshold layer was then used to partition the DSM into two separate layers, a canopy DSM and a soil DSM that contained elevation information for each pixel. The zonal statistics plugin was then used to calculate the percent canopy area, mean canopy elevation, and mean soil elevation for each plot. From these statistics we calculated plant height by subtracting the average soil elevation from the average canopy elevation and a canopy volume index by multiplying the average plant height by the percent canopy area.All analyses were performed using R statistical software . A simple linear model was used to analyze alfalfa plant height from both ground measurements and remote sensing estimates in the 2018 dormancy trial. Least squares means for each entry in the trial were calculated using the emmeans package by Lenth . The Pearson correlation coefficient was then used to compare the two forms of data collection.
In the remaining trials, square plastic plant pots the Pearson correlation coefficient was again used to compare ground measured fresh weight with remote sensing estimates for biomass volume.We have flight data from a single harvest in the grass trial and alfalfa dormancy trial, from six harvests in the large sown alfalfa trial and from eleven harvests in the transplanted mini-sward alfalfa trial. The accuracy of drone-based remote sensing varied across the range of plot types and forage species, but overall, a high correlation between the two forms of data collection was observed . The Pearson correlation between the drone estimated volume index and ground measured plot fresh weight was greatest in the transplanted family rows used in the 2018 alfalfa fall dormancy trial and was the lowest in the transplanted mini-sward plots used in the 2020 genomic selection trial . In the large sown plots, there was a greater correlation in the grass trial than the alfalfa trial , although remote sensing was highly correlated with the ground measurements in both. The relationship between remote sensing and ground data appears to be linear in all instances except for the transplanted mini-sward plots. In this trial, differences in plots with high fresh weight were not as well identified by the drone as they were in plots with low fresh weight or in the other types of plots. In addition, there appears to be several instances of high biomass plots registering low volumes from the drone imagery, likely due to lodging.The process of phenotyping in plant breeding is expensive and laborious. In perennial forage breeding programs, which often have limited resources, evaluating large breeding populations is challenging due to repeated harvests across multiple years. Drone-based remote sensing offers a fast and effective method of assessing a large amount of material with little increase in labor. Incorporation of such technology in a breeding program enables breeders to increase selection intensity by increasing the scale of breeding trials and thus improving the rate of genetic gain or to replace manual measurements for laborious phenotyping tasks.
Drone estimated biomass volume serves as an effective proxy for biomass yield across a range of perennial forage breeding plot types as demonstrated in this study. Optimizing trials for the collection of remote sensing data and improving the high throughput phenotyping data analysis and curation pipelines could make the incorporation of this technology into breeding programs routine and could help to address the low rate of genetic gain observed in most perennial forage crops. Fall dormancy is a crucial trait in alfalfa that provides growers with information related to the potential of a cultivar to perform in specific environments. Alfalfa breeders must be aware of the degree of fall dormancy in their experimental populations to ensure their cultivars fit the target environments. The standard test for fall dormancy in alfalfa has a significant labor requirement and necessitates growing a dedicated trial that only provides dormancy data. Remote sensing estimates of plant height align very closely with ground-based plant height measurements for a single location and season and may be a more accurate way to assess plant height due to the large number of data points for each plant. When creating an orthomosaic from drone images, the digital surface model includes height information for each pixel; with the camera used in this trial and at a height of 15 m, each plant was represented by hundreds of data points per plant rather than a single measurement by hand of the tallest point of the plant. This will result in a more representative reading for the height of each plant and subsequently a better estimation of fall dormancy. We have previously shown that alfalfa fall dormancy characterization in sward plots is equivalent to the spaced plants used in the standard test .
Our results here suggest that modifying the fall dormancy standard test to enable remote-sensing height data collected as a matter of course in variety trials could be feasible, saving plant breeders time and money in evaluating dormancy response and providing growers with a more precise dormancy estimate. Although remote sensing has significant potential in forage breeding programs as they currently exist, a number of design and management modifications could be made to future breeding trials to maximize the quality of data being collected. First, anything that affects the soil level or plant height will be reflected in the data. Lodging was the most significant problem that we encountered over the course of this study and is a major issue concerning plant height or volume estimates from multi-spectral aerial imagery as it results in the underestimation of true biomass. Selecting a harvest interval that minimizes lodging is one method to mitigate this issue. Additionally, breeders can adjust flight schedules in anticipation of weather events that may result in lodging, such as high winds or heavy rain. As well as causing lodging, wind can influence the quality of aerial images. Plants that are moving in the wind may cause anomalies when stitching the raw images into an orthomosaic; therefore, weather monitoring is an important consideration in remote sensing as well as having a flexible flight schedule. Machinery traffic and mammalian pests are the main causeof soil level issues that we encountered. Wheel ruts from machinery may lead to overestimation of height and biomass, square pot plastic while mounding from mammalian pests such as gophers and ground squirrels will have the opposite effect. Thus, controlling mammalian pests and avoiding traffic on trials when the ground is soft will result in better remote sensing estimates. The method of data collection used in this study requires some reference to ground level within each plot and designing trials to enable clear ground level identification will improve the data collection process. Large gaps between plots will ensure that a ground reference can be found throughout the trial. A system that we recommend is that used in our grass trial, where alleys were mown between plots prior to flying. This has the advantage of maintaining the selection pressure from competition that plants would experience in a commercial field, while also providing a solid baseline for ground level. It also has the additional benefits of limiting the impact of traffic on the soil level and controlling weeds surrounding the plots. Finally, although not essential, ensuring trials are arrayed on a regular grid with straight lines and even plot sizes will streamline the data extraction process. Overlaying a grid in QGIS requires significant manual adjustment for trial layouts that are not uniform so additional care when planning and planting a trial, or ideally, using a GPS equipped planter will greatly simplify the pipeline. Though beyond the scope of this paper, a host of alternative applications for remote sensing data beyond plant height and volume estimation can be imagined. The non-destructive nature of remote sensing means that a breeder could measure the entire growth cycle of forage crops to make better informed selection decisions. There are wide range of vegetation indices other than NDVI to highlight various properties of vegetation that are not observable to the naked eye.
Forage quality is another key trait that breeders must evaluate for which different vegetation indices may prove useful. Remote sensing imagery also serves as a digital archive of trials that can be revisited to better understand trends in trial development over time.Despite the substantial range of applications for remote sensing data in perennial forage breeding, there are a number of limitations. High throughput data collection, storage, and processing require hardware and software investment and the knowledge to develop an analytical pipeline to extract actionable information. Also, weather plays a substantial role in data collection . Finally depending on the camera and drone, there may be a large initial investment to get set up. Remote sensing offers a promising method to reduce the costs of phenotyping in perennial forage breeding programs with the accurate estimation of important traits. The results from this study suggest that breeders could increase the size of breeding trials without a proportional increase in labor and consequently increase the rate of genetic gain for forage yield Breeders also have a new method of assessing fall dormancy in alfalfa that requires significantly less labor than traditional phenotyping. Ground-measurements supplemented with remote sensing data opens the door for smaller, resource limited breeding programs to adopt new methodology such as genomic selection to help bridge the gap in genetic gain between perennial forages and alternative crops thus ensuring continued inclusion in crop rotations.By value, tomato is the fourth most important commercial crop globally . Tomato is a rich source of minerals, vitamins, and phytochemicals. Post harvest deterioration is among the major challenges for fruit industry, accounting for up to 50% of harvested losses . The primary cause of post harvest deterioration is fruit softening, which decreases fruit shelf-life and increases susceptibility to pathogens . Theoretically, regulating the rate of softening would extend shelf-life and increase pathogen resistance and be an effective strategy to reduce post harvest losses . Fruit softening is result of destruction of the fruits wall’s structural polysaccharides and reduction in intercellular cell wall adhesion . The main components of the cell wall include cellulose, hemicellulose, pectin, and a small amount of protein . Due to the complex composition and structure, many enzymes have been reported to catalyse the fruit softening. The role of polygalacturonase , pectin methyl esterase , bgalactanase, expansin, and pectate lyases regulating fruit texture has been well investigated. Downregulation of the PG and PME genes does not affect tomato fruit softening . Silencing the PL gene in tomato delayed fruits softening and reduced susceptibility to grey mould, implying prolonging fruit shelf-life by genetic modification of cell wall-modifying enzymes is a potential approach . Ascorbic acid , vitamin C, a crucial compound is present in most living organisms . In higher plants, AsA functions as an antioxidant and enzymatic cofactor, playing a crucial role in multiple physiological processes including photoprotection, cell expansion and division, ethylene biosynthesis and abiotic stress responses . As a result of these critical functions in plants and its benefits to human health, AsA biosynthesis, recycling, and accumulation in plants have been extensively investigated. The current consensus is that the L-galactose pathway is the primary pathway for AsA accumulation in higher plants. The structural genes have been identified. L-galactose 1-phosphate phosphatase catalyses the conversion of L-galactose 1-phosphate to Lgalactose in AsA synthesis . It has been reported that the expression patterns of GPP are associated with AsA content in apple and tomato plants under abiotic stress . In Arabidopsis, the VTC4 gene encodes an enzyme catalysing the similarreaction with the GPP enzyme in AsA biosynthesis .