Analyzing the Stability of P‐Wave Seismogram Method for Estimating $VS30$⁠: Insights from Two Accelerometer Arrays in Greece

We study the applicability and limitations of the single‐station P‐wave seismogram method for estimating $VS30$ by analyzing data from two accelerometer arrays in Greece. Although the method (and its variants) is usually applied at regional and local networks using small earthquake data (M 2–5), we employ a rather small number of stations (25) from the two arrays to validate the method against detailed geophysical models available for these test sites. We also examine the method applicability to downhole accelerometers, not considered in similar studies. The results show a rather systematic overestimation of the time‐averaged (to depth z) shear‐wave velocity, $VSZ$⁠, for most stations. We employ a Monte Carlo approach with synthetic data to quantify the effect of hypocentral errors and explore possible correlations between the ray parameter, p, and $VSZ$⁠. We find a clear correlation between the epicentral error and extremely large $VSZ$‐values (even >5 km/s). Removing records with low ray parameters (p <0.1 s/km) significantly improves the result stability. We also identify a correlation between the earthquake local magnitude $ML$ and the estimated $VSZ$‐values, in agreement with theoretical considerations, and propose a correction for this effect. Using reference geophysical models for the two arrays, we confirm that the method exploration depth z increases with the pulse duration of the source time function (⁠$τp$⁠), though with a larger average $τp=0.15s$ value for M∼3.0 events than previously suggested. Application of the proposed corrections leads to $VSZ$‐ and $VS30$‐values (estimated from $VSZ$‐to‐$VS30$ semi‐empirical relations) within ∼21% of the reference array velocity models. These rather small $VS30$ deviations from reference values, compared to similar studies, suggest that the P‐wave seismogram method can provide reliable $VS30$ estimates for existing large‐scale recording networks for which measured values are not available.