Sequential calibration of soil parameters using a two-step surrogate for high-dimensional geotechnical outputs

The observational method enables engineers to update uncertain soil parameters as new monitoring data become available. This process heavily relies on the model’s ability to deliver prompt and reliable results, enabling inverse analysis in geotechnical applications. Surrogate models are commonly used to approximate computationally costly simulations and accelerate calibration. However, the high-dimensional spatial nature of geotechnical outputs complicates the use of traditional surrogate models. To address this challenge for model calibration, this study adopts a type of reduced-order modeling (ROM) approach for high-dimensional outputs. Specifically, a two-step approach is used: (1) dimensionality reduction to extract features from geotechnical outputs and map the original output space into a lower-dimensional space, and (2) constructing a surrogate model directly within the reduced space. This approach improves surrogate model accuracy while maintaining the computational efficiency required. Combined with sequential Bayesian inference, the method is then evaluated for a laterally loaded pile problem. Results demonstrate the method’s effectiveness in handling high-dimensional geotechnical outputs during model calibration. • A two-step surrogate modeling framework for high-dimensional geotechnical outputs. • Integration with sequential Bayesian inversion for real-time parameter calibration. • Applied to a real offshore wind turbine foundation case with improved efficiency gains. • Enables global sensitivity analysis and reveals stage-dependent parameter influence.