A Framework for Uncertainty Quantification in One-Dimensional Plant Canopy Flow

摘要

Although the fidelity of computational-fluid-dynamics (CFD) models for the study of flow in plant canopies has significantly increased over the past decades, the inability to exactly measure the canopy structure and its material and physiological properties introduces a degree of uncertainty in model results that is often difficult to quantify. The present work addresses this problem by proposing a Bayesian uncertainty quantification (UQ) framework for evaluating the impact of uncertain canopy geometry on selected microscale flow statistics (the quantities of interest, QoIs, of the problem). The framework links available in-situ measurements of flow statistics to the uncertainty stemming from foliage spatial distribution and orientation, as well as from the aerodynamic plant response. The uncertainty is first characterized via a Markov chain Monte Carlo procedure, and then propagated to the QoIs through the Monte Carlo sampling method, which returns mean profiles and two-standard-deviation-(2SD-)intervals for the QoIs. The UQ framework relies on a one-dimensional CFD solver to simulate the flow over the Duke Forest, located near Durham, North Carolina, USA. Model results are compared against a standard deterministic solution in terms of mean velocity, Reynolds stress and turbulence-kinetic-energy profiles, as well as canopy aerodynamic parameters. For the considered QoIs, it is found that the 2SD-intervals obtained with the UQ procedure cover 80% of the experimental intervals, whereas the deterministic solution overlaps with only 47% of them. Overall, this study highlights the potential of UQ to advance CFD capabilities for predicting exchange processes between realistic plant canopies and the surrounding atmosphere.