Quantifying parameter uncertainty in empirical correlations for soft clay characterisation
Preliminary estimation of the undrained shear strength of soft clays commonly relies on empirical or semi-empirical correlations incorporating soil index properties and stress history. In practical design situations where laboratory or field data are limited, such correlations are widely applied, although their reliability depends strongly on parameter uncertainty and inherent data variability. This study applies Global Sensitivity Analysis (GSA) to quantify how uncertainty in key parameters propagates through a semi-empirical undrained strength relation representative of Swedish practice. A combined Sobol variance-based analysis and Monte Carlo Filtering (MCF) approach is used to assess both parameter importance and model performance. Experimental datasets digitised from published Swedish investigations provide data-informed bounds for the input parameters. Results from both GSA methods consistently indicate that soil plasticity, represented by the liquid limit, is the dominant contributor to output variability and predictive performance. In contrast, the overconsolidation ratio and exponent parameter show limited influence. MCF further reveals that behavioural model responses are mainly associated with medium- to high-plasticity clays, corresponding to liquid limits of approximately 45-75%. The proposed framework offers a transparent basis for uncertainty assessment of empirical geotechnical correlations and provides practical insight for engineering design.