Transferability of SPT-N–Vs Correlations Across a Multi-Country Database: Variability, Validation and Implications for Seismic Site Classification

Correlations between the standard penetration test blow count (SPT-N) and shear wave velocity (Vs) are widely used for seismic site classification of buildings, but their transferability to new sites is rarely tested. A database of 2666 SPT-N–Vs pairs from 13 studies in six countries was compiled and screened to 2045 pairs in four soil groups. Eighty-two published equations were re-evaluated, bias-corrected power-law equations were developed, and country–method correction factors and thirteen machine learning models were benchmarked under random 10-fold and leave-one-source-out validation. Under leave-one-source-out validation, a pooled SPT-N-only power law (RMSE = 53.5 m/s) and the group-specific power law (54.8 m/s) were the most transferable of the benchmarked models, correction factors did not improve transfer to withheld sources, and the machine learning models, including those tuned with source-grouped validation, did not improve on them. A mixed-effects analysis showed that 54–71% of the variance in the clay, silt and sand groups arises from differences between studies. More than half of the layers with reported Vs < 180 m/s were predicted as NEHRP class D, and the Vs30 uncertainty corresponds to about 10–26% for the assumed multi-layer profiles in long-period site amplification. For new sites, σlnVs ≈ 0.22–0.25 is recommended, while direct Vs measurements remain necessary for design.