PARAMETER OPTIMIZATION OF A PLANT CANOPY MODEL USING PHYSICS-INFORMED NEURAL NETWORKS FOR REPRODUCING FLOW FIELDS AROUND SIMPLIFIED TREE SHAPES

To investigate wind damage to structures, Computational Fluid Dynamics (CFD) with realistic surroundings is effective. Trees are often modeled using simplified tree shapes with drag force coefficient and leaf area density. However, the wake may not be reproduced because it is different from the original tree shape. This study builds Physics-Informed Neural Networks (PINNs) to estimate parameters for reproducing the flow field behind a tree with a triangular pyramid-shaped tree using a simplified tree shape. As a result, CFD with estimated parameters partially predicted mean wind speed behind trees accurately, but parameter identification accuracy of the developed PINNs remains limited.