Source: TEXAS A&M UNIVERSITY submitted to
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
Funding Source
Reporting Frequency
Accession No.
Grant No.
Project No.
Proposal No.
Multistate No.
Program Code
Project Start Date
Mar 15, 2017
Project End Date
Mar 14, 2020
Grant Year
Project Director
Murray, S.
Recipient Organization
Performing Department
Sponsored Research Services
Non Technical Summary
High throughput field phenotyping (HTFP) using unmanned aerial vehicle (UAV) or ground vehicle systems, equipped with sensors, is a promising new approach for plant breeders and plant scientists to measure different varieties and select the best ones. HTFP approaches are expected to help screen more varieties, for more measurable characteristics, faster and perhaps more accurately than current human-based measurements. These approaches will be useful to improve crops for yield, yield stability, stress-resistance, quality, safety or environmental impact, benefitting farmers, industry, society and the environment. Despite exciting possibilities from these approaches, there are many gaps in knowledge and tools for plant breeders to actually use HTFP approaches in decision making. In this project, new tools for HTFP data will be developed that will make the technology easier to use, from detection through action. The HTFP approaches will be developed and tested using largest and most diverse US corn experiment to date, the Genomes to Fields GxE experiment (G2F-GxE). Because the G2F-GxE will be grown under nine heat and drought stress environments, HTFP will be used to identify varieties and genetics with different tolerances to stress. The major outputs of this project will be big-data methods for HTFP, developed and deployed within the statistical computing environment 'R', and improved characterization of the maize G2F-GxE experiment. Student training, software, presentations, publications, and knowledge useful to public and private breeders will also result. Although HTFP is yet unproven, if successful, it is expected that the private and public sectors will adopt some or all of the methods to improve their own breeding processes; this will result in new industry jobs and more efficient agricultural production benefitting both farmers and society.?
Animal Health Component
Research Effort Categories

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
Goals / Objectives
The long-term goals of this project, applicable to all crops and researchers are [1] to facilitate plant breeders, and allied disciplines across species, to routinely and easily make actionable decisions and new discoveries using high-throughput field phenotyping (HTFP) data. The long-term goals relevant to maize breeding research and the Genomes to Fields (G2F) project are: [2] to automate routine measurements in maize, [3] identify novel phenotypic signatures of eliteness under stress that are infeasible or impossible to evaluate without HTFP tools and [4] to begin to apply these tools in a selection process.The supporting objectives of this project, are to: [1] Evaluate the Genomes to Fields Genomes by Environments (G2F-GxE) experiment under three different conditions of stress for three years. [2] Collect temporal data on the G2F-GxE experiment across three different HTFP platforms, including unmanned aerial vehicles (UAVs) and ground vehicles, and extract morphological and spectral measurements to compare entries. Of most value to the larger plant breeding community [3] develop methodology and approaches to analyze collected HTFP data to determine the value of extracted phenotypes and connection with genotypes. And finally, to make these tools available with helpful documentation to the larger plant research and breeding communities, and train other researchers to use these tools.?
Project Methods
The first objective will involve evaluating the genomes to fields genomes by environments (G2F-GxE) experiment. A 500 plot G2F-GxE hybrid panel will be grown under three different stress condition trials (planting date and irrigation treatments) in College Station, TX for each year of the project. Specific hybrid entries for the 500 plots of the G2F-GxE project will be chosen by a G2F-GxE sub-committee based on seed availability, community proposed experimental goals and geographical region. Weather data will be collected approximately every 30 minutes and field data will include days to anthesis, days to silk, terminal plant height (tassel and flag leaf), terminal ear height, and combine measured yield (grain weight per plot, moisture, test weight). Opportunistic rating notes will be taken in years when tests demonstrate strong visual segregating phenotypes including lodging, southern rust, southern corn leaf blight, leaf rolling, staygreen and, less often, common rust, northern corn leaf blight and charcoal rot; these provide opportunities to investigate unmanned aerial vehicle (UAV) and ground vehicle HTFP for biotic stress signatures. Data will be made available according to the online G2F data release guidelines. These tests will be evaluated for quality by variance component and repeatability estimates compared within and across replicates of the three plantings (and three years). Genetic mapping (a GWAS approach) will be performed in year three for alleles detected by HTFP. The second objective will collect temporal data on the G2F-GxE experiment across three different HTFP platforms and extract morphological and spectral measurements to compare entries. One team will use a fixed-wing UAV to collect aerial data using RGB and, as appropriate, near-infrared sensors. A second team will use a multi-rotor UAV to collect RGB morphological data, with the potential of multispectral and LiDAR data in years 1 & 2. Both teams will use ground control points for georeferencing and for spectral correction and have pipelines for mosaicking and making data available. A third team will use the ground vehicle with point sensors, as explained previously, with potential of using an unmanned ground vehicle by the second project year. Each team will strive to collect sensing data once per week throughout the growing season from planting through harvest. Images will be used to generate orthomosaics and digital elevation models for downstream analysis. Ground vehicle point sensor data will be output from a data logger as a matrix with GPS location, ready to analyze. The current best metric of success for UAV's are complete mosaicked images of the fields on a weekly basis throughout the growing season, at a high enough resolution to identify plants (~ 3cm) and ideally to resolve tassels (<1cm). Other success metrics will include correlations between the two UAV measurements, the ground vehicle measurement, and manual field measurements in addition to the genetic variance component and repeatability estimates within a mixed model framework within all measurement types. The third objective will develop methodology and approaches to analyze collected HTFP data, determine the value of extracted phenotypes and connect these with genotype. The statistical computing software R will be used to develop multiple tools including the automated methods to: A) Define and export plot coordinate boundaries (i.e. drawing polygons with names) from a few GPS points and a breeders field book; B) Automate plot calling and visualization for ground vehicle GPS datapoints; C) conduct outlier checking and analysis that leverages relationships between measurement platforms, sensors, time points and plots; and D) visualize and analyze plant growth curves. One basic metric of success is if genotypes can be differentiated using these approaches and if there is good consistency between replicates. There are few published examples of using these approaches in plant breeding so the ultimate long-term metric of success is if we or others adapt these measurements into our breeding programs. HTFP technologies and tools should 1) help to automate the collection of data that is already being collected and 2) provide phenotypes that were previously unable to be collected. It cannot be determined if these technologies will work until they are further developed and appropriate tools are available. If successful these tools will reduce labor, increase the ability to grow larger populations, and potentially identify new phenotypes that can lead to more accurate field breeding selection. For a commodity such as maize, the ultimate metric is for a cultivar to be developed with higher yield, higher yield stability, and/or higher farmer profit under diverse environments. To evaluate this, all HTFP phenotypes must and will be compared to combine harvested yield measurements. Finally, we will make these tools available with helpful documentation and train others to use them. Pipeline and analysis tools will be used and distributed in a phased rollout that allows troubleshooting beginning internally across the Texas A&M maize breeding program then by other G2F-GxE cooperators and TAMU breeders of other crops before a wide public release. If any tools or approaches are successful at improving end goals (yield, stress resistance, sustainability, nutrition, etc.) the larger plant breeding community will likely adapt them rapidly. The recipes of workflows and procedures used will be documented and distributed in a white paper at the end of the first year (potentially earlier on the project website) and similar information will also be included in the first publication targeted for the beginning of the second year. The R scripts will be well annotated so that researchers can easily modify them for their specific needs and specific experiments, and made available on the website as well. We will host a field day and workshop over two summers where students and researchers can go through the entire workflow.?

Progress 03/15/17 to 03/14/18

Target Audience:Immediate target audience remains the plant breeding, genetic mapping, and maize research communities which include the spectrum of graduate students through senior scientists in both the public and private sectors. Graduate and undergraduate students that formally and informally participate in the project, students that attend the field days and workshops, and students that are interested in learning about the emerging areas of high-throughput field phenotyping (HTFP) are an important audience for advancing this interdisciplinary science. A surprising additional audience has been the popular press, who seem to take great interest in using unmanned aerial vehicles in plant breeding. In the medium term, companies that sell seed, technology or agricultural services are an audience of users who could benefit from and contribute to this research, and the students trained in this research. As research needs and regulation barriers are identified, policy and decision makers may become an audience. More distantly, consumers and society will likely be interested in and benefit from this research; breeding better varieties that have higher and more stable yield, grown using less land, with fewer inputs, while providing more ecosystem services and minimizing the degradation of land allows all of society to benefit. Changes/Problems:The main changes have been due to weather and technical delays. Our ground vehicle was of little use in 2017 because the field was nearly always too muddy to use it. In 2018 we have had challenges in finding student labor to collect routine data with it successfully, and have chosen to focus more effort on the unmanned aerial vehicles (UAVs) which we believe has more potential to help us to predict yield earlier in the season. On the UAVs, our improved workflows have identified some new technical issues (primarily trouble in mosaicking homogenous sections of field and artifacts in the mosaics); these issues were due to high image resolution, high genotype homogeneity, a larger field size and scope of data collection; these issues were further exacerbated by not enough ground control points and permitting weeds to grow in the field which covered static features (e.g. the soil). When combined with our improved sensitive analysis methods we have identified a number of flight dates we are now uncomfortable with the current mosaics and are working to address and improve these. For 2017 flights these are being addressed through different analytical methods. For the current years 2018 flights these are being addressed with an improved experimental design and protocol (add more GCPs, add more driveways/ alleys in the field, and to better control weeds). The issues experienced have led us to be approximately one year delayed in the dissemination of mosaics and in hosting a how-to workshop. New hires and more seniority and growth of key graduate personnel will likely advance these issues for next year. What opportunities for training and professional development has the project provided?Students are taking classes in relevant fields to conduct this research. Students and post-doctoral scholars have attended and presented at multiple national and international meetings. Training materials for the broader research, education, and extension community have been developed or are under development but have not yet been disseminated. How have the results been disseminated to communities of interest?To date the results have primarily been disseminated through presentations at local, national and international meetings; through publications, and through informal communications. The PI is recently or currently in leadership roles in various scientific societies (North American Plant Phenotyping Network, NAPPN; Crop Science Society of America, CSSA), in peer-reviewed journals (The Plant Phenome Journal, Crop Science) relevant to the project, which have been used as forums to share findings, issues and excitement from this project. What do you plan to do during the next reporting period to accomplish the goals?In the next reporting period we will grow our final season of three environment and management conditions of the G2F experiment with associated agronomic and unmanned vehicle data collection. We will further our data and analysis workflows. We will focus more on data cleaning and the subsequent dissemination of data and methods through presentations, and publications. We will host workshops to train others in these methods. We will finish multiple graduate students associated with this project.

What was accomplished under these goals? We successfully grew and collected data on three environment and management conditions, each with two replicates of the 250+ genotype (G) maize Genomes to Fields (G2F) experiment in 2017, and have planted again in 2018. These environments (E) and management (M) conditions included early planted irrigated and well fertilized (optimal), early planted dryland (non-irrigated and reduced fertilizer [sub-optimal]), and late planted irrigated and well fertilized (heat stress). These treatments provide relevant GxExM contrasts to evaluate genetic effects across producer realistic conditions. Standard agronomic data (flowering times, terminal plant and ear height, stand counts, grain yield, test weight and grain moisture) were collected across these 1500+ plots. We also flew fixed and rotary wing aircraft over all plots throughout growth and have over 40 successful mosaics of our fields. We have extracted plant height and some vegetation indices so far over each plot and are working to extract additional features. We have developed and improved workflows and software; we have begun documenting these for broader distribution through presentations, online and ultimately through peer-reviewed publication. We have encountered a number of issues in the process and are documenting these as well. We have involved and trained many undergraduate, graduate, and some post-PhD level personnel in agronomy, plant breeding, remote sensing, statistical analysis, and the transdisciplinary integration of these areas for high-throughput field phenotyping (HTP / HTFP).


  • Type: Journal Articles Status: Published Year Published: 2018 Citation: Malambo, L., S.C. Popescu*, S.C. Murray, E. Putman, N.A. Pugh, D.W. Horne, and M. Vidrine. 2018. Multitemporal field-based plant height estimation using 3D point clouds generated from small unmanned aerial systems high-resolution imagery. International Journal of Applied Earth Observation and Geoinformation, 64, 31-42.
  • Type: Journal Articles Status: Published Year Published: 2018 Citation: Pugh, N.A.; D.W. Horne; S.C. Murray, G. Carvalho Jr, L. Malambo, J. Jung, A. Chang, M. Maeda, S. Popescu, G. Richardson, T. Chu, M.J. Starek, M.J. Brewer, and W.L. Rooney*. 2018. Temporal estimates of crop growth in sorghum and maize breeding enabled by unmanned aerial systems. The Plant Phenome 1. doi:10.2135/tppj2017.08.0006
  • Type: Other Status: Published Year Published: 2017 Citation: Murray, S.C.*. 2017. Optical Sensors Advancing Precision in Agricultural Production. Photonics Spectra. 51:49+
  • Type: Conference Papers and Presentations Status: Published Year Published: 2018 Citation: Anderson, S.L.*, L. Malambo, S. Popescu, and S.C. Murray. 2018. Exploring the Genetic Variation in Maize Height Utilizing Unmanned Aerial Systems (UAS). 2018 Genomes to Fields USDA NIFA FACT Meeting, Ames, IA. 1/28-30/2018. (poster)
  • Type: Conference Papers and Presentations Status: Published Year Published: 2018 Citation: Cruzato, N.P.*, S.C. Murray, D. Cope, A. Chang, J. Jung, S.L. Anderson, and C. Ratcliff. 2018. Spectral and Three-Dimensional High-Throughput Phenotypes As Indicators of Plant Variability in a Maize Breeding Program. 2018 Genomes to Fields USDA NIFA FACT Meeting, Ames, IA. 1/28-30/2018. (poster)
  • Type: Conference Papers and Presentations Status: Published Year Published: 2018 Citation: Ratcliff, C.*, S.C. Murray, G2F Collaborators. 2018. Comparison of Growing Degree Days and Crop Heat Units of Maize Hybrids in Texas for the Accurate Prediction of Anthesis and Silking. 2018 Genomes to Fields USDA NIFA FACT Meeting, Ames, IA. 1/28-30/2018. (poster)
  • Type: Conference Papers and Presentations Status: Published Year Published: 2017 Citation: Cruzato, N.P.*, S.C. Murray, D. Cope, A. Chang, J. Jung, S.L. Anderson, and C. Ratcliff. 2017. Spectral and Three-Dimensional High-Throughput Phenotypes as Indicators of Plant Variability in a Maize Breeding Program. 2017 Annual Meeting. Tampa, FL 10/22-25/2018. (oral and poster)
  • Type: Conference Papers and Presentations Status: Published Year Published: 2017 Citation: Anderson, S.L.*, L. Malambo, S. Popescu, and S.C. Murray. 2017. Exploring the Genetic Variation in Maize Height Utilizing Unmanned Aerial Systems (UAS). 2017 Annual Meeting. Tampa, FL 10/22-25/2017. (oral and poster)
  • Type: Conference Papers and Presentations Status: Published Year Published: 2017 Citation: Murray, S.C.* 2018. Transdisciplinary Frontiers in Field Based Phenomics. UF Plant Science Symposium, Gainsville, FL. 1/25-26/2017.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2017 Citation: Murray, S.C., S. Popescu, D. Cope, L. Malambo, S. Anderson, N. Cruzato, C. Ratcliff. 2017. Aerial and Ground Phenotyping Analytical Tool Development for Plant Breeders Using the Maize G2F project. National Association of Plant Breeders (NAPB) Annual Meeting, Davis, CA 8/7-10/2017
  • Type: Conference Papers and Presentations Status: Published Year Published: 2018 Citation: Murray, S.C.* 2018. Field Phenomics; the Next Difference Maker in Crop Improvement. Big Data Driven Agriculture USDA NIFA FACT Workshop, Washington DC. 2/26-27/2018.