Abstract: Aiming at the problems of high difficulty in identifying karst caves and unclear resolution limits in tunnel karst advance prediction, this paper, based on three-dimensional elastic wave theory, uses finite element numerical simulation methods to systematically study the propagation mechanism of seismic waves in complex tunnel spaces and the characteristics of wavefield separation. The study constructs five three-dimensional models with different ratios of cave diameter D to transverse wave Fresnel zone radius R (D=∞R, 5R, 2R, 1R, 0.67R) and simulates wavefield response patterns from ideal working conditions to low signal-to-noise ratio conditions. The τ-p transform technique is introduced to filter and separate strong interference signals such as direct waves and boundary reflection waves, extracting weak effective reflection signals. The results show that the intensity of cave reflection signals has a significant nonlinear relationship with D/R; when D≥R, the characteristics of cave reflection signals are clear and distinguishable; when D decreases to 0.67R, the amplitude of reflected transverse waves is attenuated to about 15% of the fault signal at the same location, which is at the identifiable critical threshold; if D<0.67R, the effective signal will be overwhelmed by background noise and cannot be identified. Based on this, D=0.67R is proposed as the theoretical minimum size for predicting caves using seismic reflection methods. Validation with the Sanyi Tunnel engineering case shows that the on-site exposed cave has a D/R value of 0.69, and its weak reflection characteristics (amplitude ratio about 16%) are highly consistent with the theoretical simulation results, achieving accurate advance prediction under near-limit resolution conditions.