NSF Award
2412929
Maize breeders have significantly increased crop yields by optimizing plants for higher planting densities. Further improvements in crop productivity per unit of land can be achieved by modifying the structure of individual plants and their arrangement in fields. Our current technologies allow for scanning large quantities of plants, but the acquired data is often underused. The goal of this project is to use the scanned data to create a virtual model of maize (its digital twin), which will capture the maize geometry, reflectivity, and function. The digital twin will be able to simulate real plant growth and response to the environment by careful verification against the measured data. The digital twin will be used in hypothetical scenarios of changing climatic conditions to answer "what-if" questions, providing answers for better plant architecture and planting distributions. By using AI and automatic optimization, this project will attempt to identify genetic markers and candidate genes governing variation in the same traits, enabling efforts to breed or engineer plants with optimal canopy architectures. This innovative approach will advance our understanding of plant biology and contribute to meeting global food demands.<br/><br/>This project takes an important step towards in silico optimization of maize canopy architecture. We propose to develop innovative data processing and advanced visualization tools to generate fundamental knowledge applicable to agriculture to advance food needs. Our tools will reconstruct maize into its digital twins (plant ideotypes), simulate configurations of individual plants and plant populations differing in leaf canopy-related traits, and evaluate how plant traits perform in varying environments. We will use the vast amount of gathered data from phenotyping facilities and gantry to reconstruct 3D plants into their simulation-ready digital twins, fine-tune computer simulations to visualize and optimize the plant structure and function and identify optimal canopy architectures for given sets of conditions. This work will be combined with genome-wide association study for leaf canopy architecture traits derived from 3D reconstructions of real populations to identify markers and candidate genes, enabling efforts to breed or engineer plants producing optimal canopy architectures. The results of this work will strongly impact agronomic and plant genetic research in both the public and private sectors. There is a critical need for models to predict how plant varieties will respond to different environments. The 3D interactive application will allow experimenting with complex situations at interactive frame rates on a standard desktop computer, something never achieved before. It will be connected to existing data pipelines that provide vast amounts of (often unused) data. We will develop a set of novel algorithms that reconstruct 3D maize plant shapes and functions from input data from varying sources (RGB, depth, point clouds). The developed system will also generate synthetic data suitable for AI training (labeled sets of plants and 3D geometries with proper lighting). The project will partner with The National Data Mine Network, an NSF-funded initiative and the Computer Science department at Purdue University to engage and recruit students in phenotyping, data analysis, algorithmic design, and deployment.<br/><br/>This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
