Abstract: To address the issues of insufficient coverage from single monitoring methods and the significant differences in spatiotemporal resolution among multi-source monitoring data—which are difficult to fuse—in the safety monitoring of rock-fill dams, this paper proposes a multi-source monitoring data fusion method based on transfer learning to obtain the spatiotemporal continuous distribution of external deformation in rock-fill dams. Taking the Lianghekou core-wall rock-fill dam as a case study, a spatiotemporal information-driven neural network fusion model was constructed. By employing a transfer learning strategy to balance the differences in data volume and accuracy among multiple monitoring sources, the method fuses wide-area, areal InSAR time-series data with high-precision point-based leveling data, thereby achieving complementarity between “wide-area, low-precision” and “discrete-point, high-precision” monitoring information. The results demonstrate that the proposed method can robustly reconstruct the external deformation field of the dam even under conditions of significant differences in the characteristics of multi-source monitoring data. It effectively reveals the response patterns and spatial distribution characteristics of rock-fill dam deformation in response to water level fluctuations during the initial impoundment phase, providing reliable technical support for the analysis of rock-fill dam deformation evolution and the development of monitoring and sensing systems.