Abstract: Prior to geotechnical design and construction, it is quite necessary to thoroughly understand the geological conditions of the project site to ensure the safe operation of the project during subsequent stages. Reconstructing the three-dimensional geotechnical stratigraphy provides an effective approach for accurately understanding the geological conditions of the project site. Due to the complexity of geological structure spatial characteristics and the sparsity of site exploration data, it is difficult for traditional methods to reveal the boundary patterns and spatial distribution characteristics between different strata based on a small amount of site exploration data. Reconstruction of the three-dimensional stratigraphy based on limited site exploration data still faces considerable challenges. To address this problem, a series of three-dimensional distance field features is used to characterize the potential boundary characteristics between strata. This paper introduces the feature selection process of the distance field and proposes a three-dimensional stratigraphic reconstruction method based on a boundary dictionary-random forest (BD-RF) model. The effectiveness of the proposed method is validated through two engineering cases. The results indicate that the proposed method, by employing a series of three-dimensional distance field features, can effectively capture the stratigraphic distribution patterns revealed by drilling data. Compared with other methods, the proposed method has significant advantages in predicting the boundary characteristics between different strata, providing an effective analytical tool for reconstructing three-dimensional geotechnical models based on limited drilling data in practical engineering applications.

Key words: three-dimensional stratigraphy reconstruction, stratigraphic boundary, distance field feature, random forest, boundary dictionary