Impact of correlated ground parameters on predictive capabilities for deep excavations

The predicted performance of geotechnical systems like retaining walls supporting deep excavations can deviate from observed in-situ performance due to uncertainties in ground parameters and design assumptions. Back-analysis of field monitoring data is commonly used to validate the design assumptions and the estimated geotechnical properties. Multi-supported deep excavations allow for incremental back-analysis as they progress in stages. In this study, the geotechnical properties of secant pile walls in layered stiff ground conditions in Copenhagen, Denmark are back-analysed, based on field observations of lateral wall deflections during initial construction stages. Probabilistic analysis is employed by implementing multivariate distribution functions to find parameters combinations (posterior distributions) that best describe the observed wall deflections rather than a single best-fit value. The refined properties, which are not forced to follow normal distributions, are used to predict subsequent construction stages. The impact of correlated geotechnical properties on performance prediction is evaluated, challenging the common practice in published studies that geotechnical parameters can be assumed to be independent. Comparing the two scenarios of correlated and independent parameters, cross-correlations significantly enhance efficiency, requiring only one-eighth of the analyses to achieve statistically stable results. The updated parameters also improve predictive performance for subsequent excavation stages, indicating greater accuracy and reliability.