Abstract
<jats:p>This paper proposes a practical validation method for numerical hydroacoustic wave-propagation models using field measurements in situations where the “raw” model output and the sensor recordings have different sampling rates. Such a mismatch is typical for finite-difference time-domain (FDTD) simulations: the model time step (and thus the effective sam-pling rate) is constrained by grid spacing and medium properties, whereas the measurement system samples with a fixed hardware rate. As a result, direct overlay and point-wise com-parison in the time domain becomes mathematically incorrect and may lead to misleading conclusions. The goal of the study is to develop an automated comparison procedure that (i) consistently matches sampling rates, (ii) extracts a physically meaningful and informative signal component, and (iii) enables stable quantitative assessment of model-to-measurement agreement.The proposed pipeline starts with selecting a target sampling rate that guarantees a power-of-two record length, which is required for efficient FFT processing and helps stan-dardize subsequent spectral estimation. Depending on the relationship between the original and target rates, the method applies either integer decimation or fractional resampling. To prevent aliasing, resampling is combined with anti-aliasing FIR filtering; the implementation is organized to avoid unnecessary computations for samples that would be discarded after decimation. After resampling, the signal is transferred from the time domain to the frequency domain. To reduce spectral leakage caused by finite windowing, a Blackman window is ap-plied prior to the FFT. Instead of comparing full broadband waveforms that may be strongly affected by ambient noise, multipath interference, and sensor artifacts, the method extracts the amplitude at the known transmitter (carrier) frequency and forms comparable amplitude time series for the model and the field data.Before computing any accuracy metrics, the algorithm performs a set of automatic cor-rectness checks aimed at rejecting numerically unreliable configurations and non-informative comparisons. These checks include: spatial resolution adequacy (sufficient number of grid points per wavelength), verification of the CFL stability condition (Courant number), detec-tion of a pronounced spectral peak at the target frequency, and polarity/anti-phase detection based on the sign of the zero-lag correlation to avoid falsely “good” agreement due to sign inversion. Model quality is then quantified using complementary measures computed on nor-malized amplitude series: absolute deviation (ΔE), a relative ratio coefficient (K), and mean squared error (MSE). A series of validation experiments demonstrates that isolating the tar-get frequency significantly reduces the influence of broadband noise and yields a more stable and informative assessment compared with evaluation on raw time-domain signals. The method is intended for integration into software validation toolchains where reproducibility, automation, and robustness to heterogeneous sampling conditions are critical.</jats:p>