A machine-learning enabled digital-twin framework for next generation precision agriculture and forestry

Abstract

This work utilizes the modern synergy between flexible, rapid, simulations and quick assimilation of data in order to develop next-generation tools for precise biomass management of large-scale agricultural and forestry systems. Additionally, when integrated with satellite and drone-based digital elevation technologies, the results lead to digital replicas of physical systems, or so-called digital-twins, which offer a powerful framework by which to optimally manage agricultural and forestry assets. Specifically, this enables the investigation of inverse problems seeking to ascertain ideal parameter combinations, such as the number of plants/trees, plant/tree spacing, light intensity, water availability, soil resources, available planting surface area, initial seedling size, genetic variation, etc. to obtain optimal system performance. Towards this goal, a digital-twin framework is developed, consisting of a rapid computational physics engine to simulate an agricultural installation, containing thousands of growing, interacting, plants/trees. This model is then driven by a machine-learning algorithm to obtain optimal parameter sets that match observed statistical representations of a time series of growing agricultural canopy surfaces, measured by digital elevation models. Model simulations are provided to illustrate the approach and to show how such a tool can be used for large-scale biomass management.