Data-driven multiscale modeling of flow in plant canopies

The interaction of airflow with plant canopies governs the exchanges of heat, moisture, and gases between the canopy and the atmosphere, influencing local weather and climate variability. Accurately capturing these interactions numerically remains challenging, even with high-fidelity models like large-eddy simulation. This limits our ability to develop improved surface-flux parameterizations for weather and climate models, which currently depend on several phenomenological assumptions that are often invalid. In this study, we explore the viability of physics-data-driven multiscale techniques based on volume-averaging to represent such a process. A novel multiscale framework for the large-eddy simulation of flow in realistic plant canopies is proposed, based on the volume-averaged Navier-Stokes equations. Resolvable terms account for spatial variations in porosity in the canopy and are discretized numerically using a finite volume multi-resolution WENO scheme. To obtain subgrid-scale fluxes, we rely on a data-driven approach. To build subgrid-scale terms, a large database of so-called unit-cell problems is first considered where the interaction between canopy elements and the airflow is resolved via a sharp interface immersed boundary method and a large-eddy simulation closure. A data-driven surrogate of the unit-cell problem is then formulated and used as a closure model for the evaluation of subgrid-scale fluxes and canopy drag in the resolved-scale model. Predictions from the proposed formulation are validated against the reference solutions from fine-resolution simulations of flow over a realistic plant canopy environment and compared with alternative state-of-the-art techniques.