Abstract: To improve the accuracy of stratum prediction during tunnel construction and achieve dynamic updating of geological models throughout the excavation process, a multi-lithology probabilistic modeling method based on Dynamic Indicator Kriging (DIK) is proposed. An initial geological probability model was constructed using borehole data. During excavation, lithology probability fields and spatial correlation structures were continuously updated based on geological information revealed at the tunnel face. An anisotropic variogram was adopted to characterize the strong longitudinal continuity and vertical heterogeneity of surrounding rock. Maximum a posteriori (MAP) classification and information entropy were used to quantitatively evaluate prediction results and uncertainty. Application to the Zhaibeishan Tunnel section showed that the proposed method reconstructed the main stratigraphic framework using sparse borehole data. As excavation data were progressively incorporated, model accuracy improved continuously, lithological boundaries became clearer, and uncertainty decreased significantly. The results indicate that Dynamic Indicator Kriging enables rolling updates and reliable prediction of tunnel geological models, and provides effective support for geological information integration, risk identification, and intelligent decision-making during construction.