Crop yield increases are not on track to prevent mid-century food disruptions from population growth alone, before even considering climate change. Recognized mitigation strategies include (1) accelerating crop breeding improvements and (2) better on-farm production systems. Both require enhanced growth and performance prediction for new crop varieties under new management schemes like precision agriculture in new environments that might not yet exist. Within breeding programs, the “new crop varieties” will also not yet exist and key questions are, “What genetics will achieve the best on-farm performance and how can such plants be bred most quickly from current ones”?

Mathematical biology has three relevant major approaches. Sadly, the formulations have little in common, disjoint research communities, and none can solve the practical problems of crop breeding and farming alone. However, their strengths and weaknesses are highly complementary, suggesting the question, “How can they be synergized?”  Also, newly emergent technology allows digital crop tracking by subsurface sensors to camera-carrying unmanned aerial drones. The many possible sensors create the question, “What should be measured?” Bookending the size scale from aerial photography, it is rapidly becoming feasible to concurrently monitor thousands of genes’ activity levels as they, in effect, cybernetically control a plant’s hour-by-hour responses to its immediate surroundings. This begs the question, “How do we best exploit this data tsunami?”

This project has built a framework for fusing and answering these questions simultaneously. Thus, creating predictive mathematical algorithms and supporting field measurement systems becomes an integrated, engineering co-design process. The project also sought to enlarge the workforce knowledgeable in the framework, and, importantly, to increase the participants’ (Kansas State University, Langston University, and Oklahoma State University) ability to compete for related funding.

The project has developed two differential equation models that summarize certain genetically-controlled (i.e., DNA linked) wheat plant processes. The first tracks through time the growing plant’s overall weight, seed weight, and leaf area plus the self-imposed activity slowdown that protects it against winter cold. The second additionally follows soil water changes due to rainfall, runoff, evaporation, uptake by roots, and drainage below the root zone.

A complication is that differential equation crop modeling has long and unknowingly suffered from an underappreciation of the subtleties inherent in determining the true data needs for model development and use. The project has solved this problem with a test whose patentability is being assessed prior to publication. It would not be an overstatement to say that this test enables the co-design paradigm.

To facilitate future modeling and competing for needed funds, the project collected a treasure trove of data on wheat varieties usable in genetic studies planted for three years at two breeding testbeds in each state.

Models are often limited by poor subsurface data, but new equipment with multiple sensors measured soil properties at an unheard-of spatial density. The project also built an autonomous robot able to traverse breeding plots almost continuously to potentially quantify soil moisture depth profiles. Its sensor exploits magnetism, so the metal-free robot is made from carbon fiber tubing with 3D-printed plastic joints. Its wheels are on stilts, and it navigates the 9-inch alleys between individual plots. Cameras plus novel AI steering prevent the wheels from overrunning the wheat due to surface irregularities. The AI is fast enough for the robot to maintain a productive speed.

The project’s aerial imaging overflight frequency was also novel. Subject to pilot availability, flights occurred as often as two to three per week for extended periods, producing a superb data set for plant monitoring.

In order to correlate model process predictions with gene activity levels the project collected thousands of gene activity time series from multiple varieties at multiple sites in multiple years. This is thought to be the largest such wheat data set ever assembled.

Workforce development occurred throughout the project, but activities at the historically black Langston University were especially noteworthy. Via task-based, learning-by-doing, undergraduate teams acquired skills in sensor design, construction, and communication; programming; and data science pipelines.

All student tasks were structured so as to contribute to the project. This lent realism and motivation because the students knew that what they were doing was important. When COVID struck, their lab shut down. In response, the project purchased each student a backpack containing a laptop, WIFI components, a robotics kit, an instrument controller circuit board, several sensors, and tools. Working over Zoom, students designed an instrument system that each one built and used to continuously monitor a wheat plant growing in a window pot wherever they were sheltering. This “Backpack Laboratory” garnered national attention by being featured on the NSF’s Director’s Office’s “Broadening Participation” webpage and mentioned in his opening remarks at the next NSF/EPSCoR Principal Investigators meeting.

Last Modified: 04/17/2024
Modified by: Stephen M Welch