|WANG, X - Kansas State University|
|POLAND, J - Kansas State University|
Submitted to: Transactions of the ASABE
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/1/2016
Publication Date: 11/1/2016
Publication URL: http://handle.nal.usda.gov/10113/63293
Citation: Wang, X., Thorp, K.R., White, J.W., French, A.N., Poland, J.A. 2016. Approaches for geospatial processing of field-based high-throughput plant phenomics data from ground vehicle platforms. Transactions of the ASABE. 59(5):1053-1067.
Interpretive Summary: Electronic sensors attached to ground vehicle platforms are useful for rapidly measuring characteristics of field crops for both management and breeding purposes. To assign geographic coordinates to the data from these sensors, GPS receivers are deployed to measure the vehicle position during data collection outings. Post-processing is required to calculate the absolute position of each sensor depending on the vehicle heading and offset distances between the GPS receiver and the sensors. To associate sensor data with cultivars or management practices at a given location, the georeferenced sensor data must also be spatially located within field plot boundaries. This study provides guidance and software tools to conduct these geoprocessing tasks for sensor data collected from ground-based vehicles. Examples of three sensor data collection vehicles are presented to compare alternative methodologies for geoprocessing the sensor data. The importance of vehicle heading and options for its measurement are discussed. Two algorithms for locating sensor data within field plots are also presented: one based on a geographic information system and another based on analysis of the sensor data itself. Research scientists from academia, industry, and government will use these techniques in crop improvement efforts to meet the food security challenges of a growing population.
Technical Abstract: Understanding the genetic basis of complex plant traits requires connecting genotype to phenotype information, known as the “G2P question.” In the last three decades, genotyping methods have become highly developed. Much less innovation has occurred for measuring plant traits (phenotyping), particularly under field conditions. This imbalance has stimulated research to develop methods for field-based high-throughput plant phenotyping (HTPP). Sensors installed on ground vehicles can provide a huge amount of potentially transformative phenotypic measurements -- orders of magnitude larger than provided by traditional phenotyping practice -- but their utility requires accurate mapping. Using geospatial processing techniques, sensor data must be consistently matched to their corresponding field plots to establish links between breeding lines and measured phenotypes. This paper examines problems and solutions for georeferencing sensor measurements from ground vehicle platforms for field-based HTPP. Using three case studies, the importance of vehicle heading for sensor positioning is examined. Three corresponding approaches for georeferencing are introduced based on different methods to estimate vehicle heading. For two of the studies, approaches to develop a field map of plot areas are addressed, where the issue is to ensure sensor positions are correctly assigned to plots. Two solutions are proposed. One uses a geographic information system to design a field map before planting, while the other adopts an algorithm that calculates plot boundaries from the georeferenced sensor measurements. An advantage of the latter approach is the accommodation of irregular planting patterns. Using the algorithm to calculate the plot boundaries of a winter wheat field in Kansas, 98.4% of the calculated plot centers were within 0.4 m of the surveyed plot centers, and all of the calculated plot centers were within 0.6 m from the surveyed plot centers. While multiple options and software tools are available for geospatial processing of field-based HTPP data, they share common problems with sensor positioning and plot delineation. The options and tools presented in this study are distinguished by their practicality, accessibility, and ability to rapidly map phenotypic data.