To further improve the accuracy of winter wheat yield estimation in Guanzhong Plain of Shaanxi Province, the ensemble Kalman filter (EnKF) algorithm was used to assimilate the CERES-Wheat model simulated soil moisture at the depth of 0~20cm and leaf area index (LAI) with remote sensing observations of the vegetation temperature condition index (VTCI) and LAI, respectively. At the same time, the resonance periods between assimilated VTCI and LAI at each growth stage and yield were analysed by using the cross-wavelet transform, respectively, and the weights of assimilated VTCI and LAI at each stage were obtained by calculating the wavelet cross-correlation degrees, and then a regional yield estimation model for winter wheat based on weighted VTCI and LAI was constructed. The results showed that at the sample point scale, the assimilated VTCI and LAI can combine the effects of model simulations and remote sensing observations, and the trends were more consistent with the actual crop growth changes. At the regional scale, there were specific resonance periods between VTCI, LAI and yield for each growth stage after cross-wavelet transform, regardless of assimilation or not, respectively. It was also found that the assimilation promoted the feature extraction for key growth stages. Compared with the estimated yield model constructed without assimilation, the estimated yield model constructed with assimilation had normalized root mean square error of 13.23%, coefficient of determination of 0.50, and mean relative error of 10.58%, with a slight improvement in accuracy, and the distribution of yield estimation results from the assimilated model was closer to the official statistical yields. In summary, the regional yield estimation model combining assimilation and cross-wavelet transform can effectively improve the estimation accuracy and provide a relevant research basis for further precision agricultural management.