Dendrites are the most widely seen growth morphology during alloy solidification, and an improved understanding of dendrite growth is essential for enhancing the quality of casting products. The phase-field method is the most accurate model for reproducing dendrite growth. However, the material property data needed to perform phase-field simulations are often unavailable, particularly of alloys. Integrating phase-field simulations and time-resolved X-ray imaging is a promising strategy to overcome this issue. This novel approach is expected to reveal unknown input material properties for phase-field simulations and enable the accurate reconstruction of three-dimensional dendrite morphologies in time-resolved X-ray imaging. This study aims to develop a data assimilation system that integrates phase-field simulations and in situ X-ray imaging for columnar dendrite growth. To reduce the computational costs of data assimilation, we used the adaptive mesh refinement method for phase-field simulations and the local ensemble Kalman filter for data assimilation. In addition, parallel computing with multiple graphic processing units was performed to accelerate the data assimilation. We verified the developed data assimilation system through a numerical test termed “twin experiments,” where we used the numerical results as observation data.