Abstract: Site investigation data are often sparse due to limited project budgets. Using sparse site investigation data to quantify soil parameter inevitably results in significant statistical uncertainties. Meanwhile, the soil parameter database is expanding rapidly due to advancements in geotechnical investigation equipment and information technology. Effectively utilizing the big data in the soil parameter database to quantify soil parameter uncertainties at three-dimensional sparse measurement sites remains a challenging problem. This study proposes a big data assimilation technique to effectively quantify soil parameter uncertainties at three-dimensional sparse measurement sites. The proposed method first constructs a probability distribution model of soil parameters using limited sparse site investigation data. This model considers the autocorrelation of soil data in both horizontal and vertical directions across different soundings. Secondly, a probability distribution model based on the soil parameter database is constructed using Gibbs sampler. Based on the two probability distribution models, the hybrid Bayesian theory is employed to integrate the statistical information of the big data into the site-specific data. A hybrid probability model of soil parameters for the sparse measurement site is derived after assimilating the big data. The incomplete borehole data are simulated to be complete data that satisfy the lattice structure using the hybrid probability density function. The proposed method uses Kronecker product to decompose the large autocorrelation matrix and achieves the efficient reconstruction of the 3D site. Finally, the proposed method is demonstrated using a simulated virtual site and a site in Texas, USA. Results indicate that the proposed method effectively utilizes geotechnical big data. Assimilating geotechnical big data significantly reduces soil parameter uncertainties at three-dimensional sparse measurement sites. The proposed method provides a valuable tool for quantifying soil parameter uncertainties at sparse measurement sites.

Key words: site investigation, uncertainty, geotechnical big data, Gibbs sampler, data assimilation