Advancing Quantitative Model Evaluation with Three-Dimensional Experimental Data

Abstract This study evaluates computational fluid dynamics (CFD) methods for urban atmospheric dispersion using high-resolution three-dimensional experimental data obtained through magnetic resonance imaging (MRI). Laboratory experiments were conducted in a controlled water channel featuring a 3D-printed urban array with dual contaminant injection sites. By applying magnetic resonance velocimetry (MRV) and magnetic resonance concentration (MRC), volumetric measurements of mean velocity and contaminant concentration were acquired at an isotropic resolution of 0.7 mm, yielding extensive datasets for direct, quantitative comparison with large eddy simulation (LES) results using SIERRA/Fuego. We applied established metrics-fractional mean bias, normalized mean square error, geometric mean, geometric variance, FAC2, and normalized absolute difference to assess model performance. Additionally, new metrics including local normalized difference (LND), norm of relative difference (RD), projection coefficient (PC), and a mutual information distance (D) were introduced to capture both bias and variance in the data. Sensitivity analysis using synthetic datasets informed the development of intrinsic, data-based acceptance criteria. Overall, excellent agreement between CFD predictions and MRI measurements is demonstrated, highlighting the potential of combining detailed experimental data with advanced quantitative metrics to improve the evaluation and fidelity of CFD models for urban dispersion scenarios.