Development of digital phenotyping methods for breeding applications in stress phenotyping, spatial adjustments, and end season yield prediction

Thumbnail Image
Date
2023-05
Authors
Carroll, Matthew Eden
Major Professor
Advisor
Singh, Asheesh K
Sarkar, Soumik
Cannon, Steven
Dixon, Phillip M
Singh, Arti
Committee Member
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Plant scientists now have a large array of sensors and phenotyping vehicles, which they are able to exploit to collect large volumes of high dimensional data with minimal human labor. These phenotyping methods can be an effective tool, if there are tools available to extract and process the data that has been collected. With the increasing usability and advances in machine learning (ML) techniques, ML paired with high throughput phenotyping (HTP) can help plant scientists and breeders address key issues. Continuing to develop highly productive cultivars with more insights into the drivers that effect important traits such as stress tolerance, and end season yield will be critical for breeders to be aware of for future breeding efforts. Best practices and previous work is important for researchers to be aware of as they start working with HTP technologies. One of the most commonly used HTP platforms used are unmanned aerial vehicles (UAVs), due to their relatively low cost, and speed at which they are able to traverse a field. We present a review article that covers the latest in UAV plant science research as well as provide best practices for data collection, data management, and downstream analysis. The intention of this review is to provide new users a complete overview of the UAV landscape and what they will need in order to successfully use UAVs for their research goals and objectives. Continuing work with UAVs has shown that they can be an effective tool for plant stress phenotyping. Building from previous work we explored different flight altitudes and the effect it had on the accuracy of classification of iron deficiency chlorosis( IDC), as well as for the ability to map the genetic loci that effect the severity ratings that are observed. In addition to this we looked at the utility of mapping secondary traits such as canopy growth and development which have not been investigated in previous work. IDC is commonly associated with high pH soils in the Midwest. When breeders are looking at the yield trends observed within a breeding trial they very rarely have a single soil property that is driving nonuniform field conditions. We show that using high resolution soil maps breeders can adjust for yield trends directly, instead of relying on neighbor based adjustments. This process can give breeders additional insights into the micro environments which plots are grown in, and provide more precise estimates of genotypic values of breeding lines. The last chapter explores the use of a ground based phenotyping system to predict end season yield. Breeding programs strive for high quality data, but the geographical extent of a breeding program typically requires that large amounts of time and effort are spent to collect yield data at outside locations. The seed harvest from these plots, is only used for data and is never used to propagate the following season's seed. we investigate the usability of rovers with digital cameras to predict end season yield by estimating the number of pods in a given plot. Future work would allow programs to optimize harvesting efforts and reduce the labor and time to generate the same level of high quality data.
Series Number
Journal Issue
Is Version Of
Versions
Series
Academic or Administrative Unit
Type
dissertation
Comments
Rights Statement
Copyright
Funding
DOI
Supplemental Resources
Source