Surrounding rock classification is fundamental to tunnel support design and construction. Conventional methods are usually applied on a mileage-segment basis and are therefore unable to capture the spatial heterogeneity of surrounding rock within a single tunnel face, which may lead to overdesign or insufficient support in local areas. To address this issue, a refined advanced prediction and three-dimensional modeling method for surrounding rock classification is proposed for drill-and-blast tunneling. Measurement-while-drilling data obtained from jumbo drilling are used to investigate the relationships between drilling parameters and surrounding rock classes, and a training dataset is constructed using drilling rate, impact pressure, feed pressure, and damping pressure as input variables. A hybrid CNN-LSTM model is developed to exploit both spatial and sequential features and is used to predict the surrounding rock class at 0.02 m intervals along each borehole on the tunnel face. In addition, a grade-encoding and threshold-based reclassification scheme is introduced to convert the classification task into a continuous spatial interpolation problem, and an implicit three-dimensional modeling method based on radial basis function interpolation is employed to reconstruct the spatial distribution of surrounding rock classes ahead of the tunnel face. Validation using MWD data from an actual tunnel project shows that the proposed model achieves an accuracy of 92%, exceeding that of the CNN and LSTM models by 3% and 5%, respectively. The generated three-dimensional model provides an accurate visualization of the geological conditions ahead of the tunnel face. The proposed method can therefore improve construction safety, optimize support design and resource allocation, and reduce engineering costs.