EXPERIMENTAL VALIDATION OF A METHOD FOR SELECTING ACOUSTIC MODEL PARAMETERS BASED ON ACCURACY AND COMPUTATIONAL RESOURCE CRITERIA

Abstract

This paper reports experimental results of a practical workflow for selecting numerical parameters of a time-domain acoustic wave propagation model under joint accuracy and computational feasibility constraints. In applied verification and validation tasks, the credibility of conclusions depends not only on physical modeling assumptions but also on discretization and solver settings. Overly fine spatial grids and conservative absorbing boundary settings (PML) lead to steep growth in runtime and memory demands, whereas coarse discretization may introduce numerical dispersion and boundary reflections that distort the simulated waveform and degrade agreement with experimental recordings. We use the open SWellEx-96 dataset (event S5) as the experimental reference, because it provides controlled multi-tone underwater transmissions together with auxiliary range tables that enable automatic reconstruction of the source-to-receiver distance per experiment minute. Simulations are performed using k-Wave-python (v0.4.0) in a two-dimensional configuration as a pragmatic compromise allowing systematic parameter sweeps while preserving essential wave physics. To ensure objective comparison, we implement an explicit alignment stage between measured and simulated signals: both are brought to a unified sampling rate and time window; the transmitter tone frequency f0�is determined by combining known S5 tone candidates with spectrum-based presence checking on the measured segment; time lag is estimated via cross-correlation within a narrow band around f0; and complex scaling based on analytic (Hilbert) representations is applied to compensate phase and amplitude differences that are not directly related to the propagation model quality. Accuracy is quantified using the normalized mean absolute error (NMAE) computed on the amplitude envelope of the narrowband component at f0, which reduces sensitivity to small phase mismatches and emphasizes energy consistency in the tone band. The explored parameter space is defined as a discrete set of numerical configuration values including PPW (points per wavelength defining grid step dx), PML thickness and absorption coefficient, and auxiliary domain parameters. The selection procedure is implemented in two stages: 1) a maximum-accuracy configuration providing the lowest NMAE for the selected experimental segment and serving as a baseline, and 2) an “optimal” configuration chosen as the fastest one that satisfies a practical accuracy constraint NMAE ≤ 0.2. For the considered SWellEx-96 segment, we obtain an approximately 2.96 reduction in simulation time while keeping the NMAE well below the imposed threshold, demonstrating that an admissible parameter region can be identified and that a resource-efficient configuration can be selected automatically without exceeding the required error level. The reported results support the feasibility of parameter region estimation and computationally constrained validation workflows for numerical acoustic models