In this study, the propagation time and attenuation rate distributions of each sound source grid point (200 × 200) to a microphone of an arbitrary position across the shear layer, that are required for beamforming, were reconstructed by the reduced-order model with sparse sampling for acceleration of the computation. First, the propagation time and attenuation rate distributions including the refraction of sound by the shear layer were calculated over 100 patterns of combinations of the wind speed and the microphone position, as training data. The dominant modes and optimum sampling points were discovered from the training data. Then, data-driven sparse sampling for reconstruction was applied and the propagation time and the attenuation rate from each grid point (200 × 200) to a microphone were quickly calculated for given microphone positions and wind speed. The error of the obtained calculation result is 1% or less, and the approximation by data-driven sparse sampling is concluded to be effective.