B1113
Title: Understanding crop vulnerability to soil moisture extreme conditions
Authors: Veronica Berrocal - University of California, Irvine (United States) [presenting]
Abstract: Agronomists have been studying the relationships between crops and climate for decades. Using container and plot-level experiments, they have investigated how yield depends on soil moisture, temperature, humidity, sunlight, and their interactions. In almost all cases, it is found that climate drivers relate to yields non-linearly, with associations that are crop- and growth-stage specific. These insights constitute the theoretical backbone of numerical crop yield prediction models, typically referred to as process-based models. Despite being able to emulate the main physiological processes of crop growth and development, process-based models are quite limited in their scope. Thus, statistical crop yield prediction models are preferred over process-based models, especially when the goal is to generate regional predictions of crop yield. At the same time, statistical crop models typically overlook the insights provided by container and plot-level experiments, and they oversimplify the relationship between crop yield and climate drivers. A Bayesian hierarchical spatio-temporal model is proposed that models the spatially and temporally varying effect and the non-linear interaction of climate drivers on crop yield using an approach that builds on functional data analysis methods and employs Gaussian processes. A key insight of the model is the ability to identify time periods during the growing season during which the crop is more vulnerable to the effect of climate factors.