Location: Plant, Soil and Nutrition Research
Project Number: 8062-21000-053-002-A
Project Type: Cooperative Agreement
Start Date: Sep 21, 2023
End Date: Sep 20, 2024
Objective:
The objectives of this one-year cooperative agreement are in support of developing methods to breed for intercropping using an oat / pea intercrop as the model system.
These objectives are:
1. Capture and analyze images from an Unoccupied Aerial Vehicle (UAV) to estimate height and biomass of intercropping mixtures.
2. Assess the prospects for genomic prediction models to improve gain of intercropping performance using genomic prediction and quantitative genetic theory.
3. Implement and parameterize a two-species crop growth model (CGM). The CGM will not yet be optimized but we will get baseline performance of the model prior to improving parameter values.
Over the long-term, these objectives will contribute to a complete methodology for breeding for intercropping to include:
1. Genomic prediction methods to predict the performance of experimental line combinations that have not been evaluated. These methods will use measurements on a subset of the many possible combinations to predict the remaining unevaluated combinations.
2. Unoccupied Aerial Vehicle (UAV) image analysis to estimate canopy cover proportions of each species. These canopy cover proportions will also be extrapolated to biomass proportions of each species. Frequent UAV flights will enable longitudinal analysis of the intercrop mixture growth.
3. Parameterization of crop growth models (CGMs) including two species. These two-species models will enable both virtual exploration of the species space to identify characteristics of the species that will combine well and a further method of prediction of specific experimental line combinations.
Approach:
Approach for Cooperative Agreement Objective 1: Development of UAV data capture and analysis methods. In 2023 and 2024, the in collaboration with SY Jannink, the Cooperator (collectively, “we”) will plant mixtures of oat and pea. In 2023, the number of oat lines will be 48 and pea lines will be 12. In 2024 those numbers may be the same or go up if we determine that better estimates will derive from larger numbers.
In both years, we will interact with colleagues at other universities (in IL, WI, MN, ND, SD, AL) to plant similar experiments. We will collect biomass samples, sorted to the different species, from these plots at four points during the growing season. We will fly the plots with a drone prior to each of the sampling dates. Drone images will be analyzed using standard approaches that are applied to single-species plots. Plot images will also be assessed by humans to estimate the canopy cover in each plot of each species.
We will use those estimates as the response variable in supervised learning approaches using convolutional neural networks to determine the accuracy of image analysis in estimating species composition of the canopy cover. These estimates will also be correlated to measurements of biomass of each species.
Approach for Cooperative Agreement Objective 2: Application of genomic prediction to intercrop performance. In principle, there should be three sources of variation that matter for intercrop performance: the variation in performance for each species separately, and the variation due to the interaction of the experimental lines combined in each mixture. If that last source of variation is small, no new developments in genomic prediction methods will really be needed to improve intercropping performance.
Then, prediction separately within each species will be adequate and selection will simply be on “general combining ability” for the individuals of each species. If, on the other hand, interaction between the components plays an important role in intercrop performance, we will need to assess how accurately those interactions can be predicted and, if accuracy is adequate, how to use those predictions in choosing parents to cross within each species.
In the coming year, a student at Cornell will develop simulations to assess:
1. What size the interaction variance should have relative to the general combining ability variances for it to be decisive.
2. How predictions of the interaction should be used in choosing parents to cross.
Approach for Cooperative Agreement Objective 3. Parameterization and implementation of a two-species crop growth model. Such a model has been developed at Cornell. We plan to use this model. Crop growth models are complicated and we have not yet actually run this model or developed interfaces between the model and R (R will be the primary data analysis platform). The Cornell graduate student will take these initial steps during this Agreement period.
