<p>Subaerial landslide-tsunamis (SLTs) caused by mass movements into water bodies pose serious threats to infrastructure and human safety, especially in narrow reservoirs. Reliable estimation of the maximum wave arrival time (<i>t</i>_M) is essential for risk assessment and early warning, where <i>t</i>_M corresponds to the onset time of significant wave development, rather than the arrival time of the maximum crest. However, predictive formulas are scarce. This study conducted 81 laboratory experiments to investigate SLTs in narrow reservoirs, covering a range of key dimensionless parameters (e.g., the Froude number and the relative slide volume). Wavelet transform analysis was applied to normalized energy density contours to extract <i>t</i>_M, which represents the beginning of a high-energy window defined by an energy density threshold of 0.8. Empirical formulas were developed for both three-dimensional (3D) and two-dimensional (2D) wave-propagation regimes, defined by whether the wave propagation distance is smaller or larger than the reservoir width. Prediction accuracy was significantly improved compared to theoretical estimates, with most errors remaining within 50% of the measured values. Their performance was further assessed using SPH-based numerical simulations in DualSPHysics, and applicability was qualitatively demonstrated through the Gongjiafang SLT case. Although the extraction method is less suitable for idealised flume experiments, the study proposes an energy-based, wavelet-assisted method to identify <i>t</i>_M and provides empirical formulas for SLT hazard assessment in narrow reservoirs with nearly constant widths.</p>