Creation and validation of a digital twin model for improving almond production
Research team: Baram, Shahar, Sperling, Or, Brown, Patrick, and Wang, Lizhi
Research Funding: BARD
Students: Noam Bodin and Kai Zhong
Accurate yield prediction is important for tree growers to optimize resources and implement management strategies. Yet, predicting yield, especially in deciduous trees (e.g., almonds), is challenging since they are complex organisms in which genetics and environmental factors interact with plant physiology over multiple seasons. Our project objective is to propose an innovative avenue to interpret, predict, and improve almond yield production by creating and validating a digital twin (DT) model through field experiments on almonds. A DT refers to a virtual representation of a physical or biological object or process that can gather data from the real environment. A key innovation of the proposed DT model is its capability to combine the strengths and overcome the limitations of crop models and machine learning (ML) for crop yield prediction. The DT model consists of a new crop model, completely redesigned to be compatible with the ML paradigm's data-driven (rather than traditionally experiment-driven) training process. Its training algorithms are structured to ensure that all model parameters are explainable within the framework of tree physiology. The expected outcome of this project is a validated DT for almond orchards that will allow guiding agricultural inputs like water and nutrients to be optimized at the field level or within individual management zones to maximize yield and profitability. Thus, we propose to investigate the environmental, biological, genetic, and management factors that determine tree-level yield variability of almond orchards. We will achieve this by generating descriptive, predictive, and prescriptive insights through DT models and targeting and validating the tree-based yield gap in almonds and spatial decision support system for irrigation and fertilization to improve yield. By harnessing the capabilities of the DT methodology for yield estimation, farmers and researchers can refine their decision-making processes, optimize resource utilization, and effectively confront challenges within the almond industry. We have extensive preliminary work in DT modeling for other crops (such as maize). Over recent years, we have used data-driven ML models to develop high-accuracy almond tree-based yield and tree nitrogen content using high-resolution UAV imagery. We have collected substantial datasets from several almond orchards from various platforms, encompassing field measurements, laboratory analyses, sensor networks, high-resolution remote sensing imagery, and tree-based yield in Israel and California. This comprehensive information offers insights into genetic, management, and environmental influences on almond tree structure and function.
