Abstract: The spatial distribution of xenon and iodine within a pressurized water reactor core critically influences neutronic calculations and operational safety, particularly during power transients where reactivity changes rapidly. Due to the high computational cost of neutronics programs and the inability to directly measure xenon-iodine concentrations in the core, accurate and real-time estimation of the high-dimensional physical field of the core remains a formidable challenge. This study addresses these limitations by developing a novel framework that couples a control-oriented reduced-order model with four-dimensional variational data assimilation (4D-Var) to enable rapid and precise reconstruction of initial xenon-iodine distributions. The dynamic mode decomposition with control (DMDc) was employed to construct a linear reduced-order model capable of explicitly incorporating external operational inputs, such as control rod movements and boron concentration adjustments, into the system dynamics. Unlike standard dynamic mode decomposition (DMD), which assumes autonomous evolution, the DMDc formulation can capture the impact of time-varying conditions on xenon-iodine transients. The analytical gradient expression of the 4D-Var optimization problem coupled with DMDc with respect to the reduced-order state coefficients (also known as initial amplitude of modes) was derived, transforming the estimation problem of high-dimensional physical fields into a low-dimensional optimization problem and significantly improving computational efficiency. The methodology was validated using the HPR1000 reactor design as a testbed, generating synthetic transient scenarios via the CORCA-3D code under various control rod and boron modulation schemes. These scenarios included both constant and fluctuating operational parameters to rigorously test the model’s robustness against external perturbations. Numerical experiments demonstrate that the DMDc model substantially outperforms standard DMD in predicting core state evolution under active control interventions. While standard DMD fails to reflect changes induced by control rod insertion or withdrawal, the DMDc model accurately reproduces the trajectory of the axial offset (AO) and local power distributions, achieving a maximum k_eff prediction error of 27 pcm in complex transient cases. The integration of DMDc into the 4D-Var framework yields exceptional computational performance. The reduced-order model inference accelerates calculations by a factor of approximately 4 500 compared to full-order simulations, reducing a 48 hours transient simulation from over twelve minutes to tens of milliseconds. Consequently, the entire data assimilation process, including iterative optimization via the L-BFGS-B algorithm, completes within eight seconds. The results indicate that the proposed method effectively corrects initial background errors, reducing the deviation of k_eff from −337 pcm in the unassimilated background field to −96 pcm in the analyzed field at the initial time step. Furthermore, the assimilated xenon and iodine distributions exhibit strong agreement with reference solutions, with root mean square error (RMSE) remaining below 1.0% for xenon and 0.4% for iodine across the transient period. The reconstructed physical fields lead to significant improvements in subsequent predictions of axial offset, confirming the efficacy of the approach for online core monitoring. This work establishes that coupling control-aware reduced-order modeling with variational data assimilation provides a viable, high-efficiency solution for real-time reactor state estimation, offering a promising tool for online monitoring in advanced nuclear energy systems.