In order to improve the estimation accuracy of maize, the central plain of Hebei Province was chosen as research area, and the remote sensed LAI and simulated LAI by CERES-Maize model was combined in eight typical samples from 2013 to 2018 by using the ensemble Kalman filter (EnKF) data assimilation approach. The random forest regression was used to estimate maize yield by using monitored LAI and the assimilated ones respectively. The optimal model for estimating maize yields in study area from 2013 to 2018 was selected, and the measured maize yield of the year 2015 was used to validate the accuracy of the optimal model. The results showed that the single point assimilation of eight samples using the EnKF algorithm was more consistent with the actual growth of maize. The assimilated LAIs were extended from the sampling sites to the regional scale, the phenomenon of LAIs rising and falling between adjacent pixels was reduced and the effect was better than the remote sensing inversion LAIs. Compared the yield estimation models with the monitored LAIs, the accuracy of the yield estimation models with the assimilated LAIs was improved, and the R2 was increased by 0.0245. The yield estimation model was applied to estimate maize yield in 53 counties (districts), in general, the average relative error of the estimated yield was 12.11%, and the root mean square error was 371kg/hm2, the normalized root mean square error was 6.18%. The yearly estimated yield from 2013 to 2018 in the central plain of Hebei Province was fluctuated in individual years, and the overall distribution in time was characterized by a tendency to decrease first and then increase, and the spatial distribution of maize yield was the highest in the western region of the plain, following by the north and south regions, and the lowest was in the eastern region.