Abstract: In fire remote sensing monitoring, the inherent mismatch between continuous fire spread dynamics and discrete satellite observations limits the complete temporal imagery reconstruction of burned areas. To address this issue, this study proposes a unified wildfire image simulation method comprising three components: cellular-automata-based spread computation, burn-state-dependent multispectral radiance estimation, and residual correction during the fire progression phase. Without altering its internal structure, the method adapts to both forward spread simulation and boundary-constrained reconstruction by switching only the propagation reference extent. Post-fire imagery is used solely to delineate the burned-area reference extent and does not directly replace spectral observations. Taking the fire site of Xichang City, Liangshan Prefecture, Sichuan Province, China, in 2020 as the study area, simulation experiments were carried out based on Sentinel-2 MSI Level-2A surface reflectance data and existing forest fire datasets, under controlled conditions with of 0.35, wind speed of 5.0 m/s, and wind direction of 180°. The results showed that forward spread simulation and boundary-constrained reconstruction reached full coverage of the ignition-associated fuel patch at time steps 1070 and 918, respectively. For spectral consistency, forward simulation at its optimal time step (860) yielded RMSE = 0.0200 and = 0.8727, whereas constrained reconstruction at time step 918 yielded RMSE = 0.0227 and = 0.8361. For boundary consistency, constrained reconstruction achieved IoU = 0.9491, compared with 0.4804 for forward simulation. Moreover, the commission errors outside the reference extent were reduced from 290,101 to 0.