Crop improvement for climate change
Charlie Messina @ UFlorida
Abstract:
Adaptation of society to new climate and economic realities relies on the development of agricultural technologies such as advanced genotype-by-management technologies (GM-Tech). Crop systems are non-linear, dynamic and often manifest emergent behaviors; these are difficult to predict, and thus hamper innovation rates using current engineering paradigms. In this talk, I introduce a framework that combines symbolic AI in at the form of crop models, and subsymbolic AI in the form of a Bayesian algorithm, as a starting point to further prediction methods for developing GM-Tech. I demonstrate the approach with two maize datasets created over more than a decade of plant breeding for drought tolerance in maize. I conclude the talk by reviewing how maize adapted to climate change to motivate discussion about the need to develop a framework for crop improvement for climate change if we are going to meet societal needs for food, water and energy.
Bio:
Charlie Messina is a Professor of Predictive Breeding in the Department of Horticultural Sciences at the University of Florida and formerly a Distinguished Fellow at Corteva Agriculture and Research Scientist at DuPont. His program focuses on the development of prediction methods for agriculture and horticulture.
Summary:
Agriculture
Lots of externalities and societal costs
$1 trn value
$2 trn costs/externalities
Water use efficiency is critical but progress towards it has been slow
Plant breeding can be a great tool for adapting agriculture to make it efficient and have fewer externalities
Crop design:
Multi-dimensional framework:
Use of nitrogen, water,
Carbon intake,
Crop yield, physiology, nutritional value
Need modeling and simulation for predicting consequences of changing crop genome, treatment techniques
Phenotypes depend on both genome and environment
Causal chains go across abstraction boundaries
From gene/metabolome -> whole organism
From population/field/competition -> individual organism
Across different environments crop yield interacts strongly with different phenotypes of the plant (e.g. leaf area more critical where there is less daylight, larger plants produce more yield but only if there’s enough water)
Which phenotypes are good for growing under climate change?
Modeling crop growth
Phenotype is function of genotype, management and environment
Approach: symbolic AI
Use human-understandable structured models
Incorporate crop growth models, which are differential equations created by plant development experts
Predict configuration parameters of a crop growth model of a given plant given the genome of a particular plant
Use linear model for prediction (additive effects, no gene interaction term)(
Use Bayesian inference to predict coefficients of the linear model and their uncertainty
Then run complete model (crop growth model + linear (gene->parameters) model) across a range of environments to predict how this genotype will grow
Alternative is Genomic best linear unbiased prediction (gBLUP), which looks at the genome but doesn’t leverage crop growth models
Resulting model predicts how yield and resource use efficiency scale with environmental attributes but there is still lots of room for improving accuracy of specific predictions
Most valuable where gene/environment interactions are strongest (e.g. where there is more resource stress)
Accuracy significantly improved over genomic models that don’t use crop growth models (not not for all plant traits)
Limits on the accuracy of the current model are suspected to be caused by emergent plant behavior
Competition among nearby plants
Evidence: as more plants per acre yields start following a bi-modal pattern (more/less fit individuals competing)
Climate change
Corn-belt is seeing changes in weather: temperature, precipitation
Observation: 1990-2020 environment and management have increased yields much more than genetic gains
From 1930-1990 changes in plant hybrid genotypes has doubled crop yields
1990-2020 high gains have been observed for high-density agriculture
These are higher climate change years
Why?
Higher precipitation in the last few decades
Genotypes we developed decades ago are inappropriate for current climate
Need simulation to tease out causality and figure out best plant management and genomes for new climate conditions, in each world region