The generation of low-frequency (≤ 40 Hz) acoustic waves (T-waves) by undersea earthquakes below a flat abyssal plain is not yet fully understood. To model the generation and propagation of PN-waves (horizontally in the crust and vertically in the ocean) and of T-waves over a rough sea bottom, we use a 2D spectral finite-element code (SPECFEM2D). The model includes a solid layer (Earth crust) overlain by a fluid ocean, and separated by a sinusoidal crust/water interface (seafloor roughness). Synthetic T-waves propagate as Rayleigh modes with their expected vertical and long-range horizontal propagation. In the simulated PN-wave spectrum, resonance peaks appear at frequencies predicted by the analytical solution for Rayleigh modes. The same peaks are observed in actual T-wave records from an antenna of two hydrophones at different, generated by a large magnitude earthquake 987 km away. The T-waves spectrum from the shallowest hydrophone shows an energy gap in the 1-4 Hz frequency range that can be explained by modal propagation. The resonance peaks in the observed PN-wave spectrum are also well predicted by the analytical solution.