To realize the regional winter wheat yield estimation accurately, efficiently and in real-time, Shiqiao Village, Qi County, Hebi City, Henan Province, was taken as the study area. The ensemble Kalman filter (EnKF) was used to assimilate the time-series leaf area index (LAI)，which were estimated by the PROSAIL radiation transfer model, into PyWOFOST crop growth model to estimate a certain number of winter wheat site yield points with different growth. And those site yield points provided training data for random forest regression (RFR) algorithm to establish machine learning model. Finally, the established machine learning model and the time-series optical remote sensing images of Sentinel-2 with 10m resolution were used to estimate the regional winter wheat yield, so as to realize the application of coupling crop growth model and machine learning algorithm, and establish a new regional winter wheat yield estimation mode. Based on Sobol parameter sensitivity analysis algorithm, the sensitivity parameters of TWSO and LAImax were quantified. The TDWI, TBASE, CVS and CVL sensitivity parameters related to LAImax were optimized by time-series LAI data and particle swarm optimization (PSO) algorithm. And inputting them into the PyWOFOST model, using the EnKF algorithm and time-series LAI data to adjust the AMAXTB1, TDWI, TSUMEM, and CVO sensitivity parameters of TWSO to improve the accuracy of the singlepoint yield estimation. Compared with the site yield points, the R2, RMSE, MAE, and Bias of estimation were 0.8665, 468.64kg/hm2, 385.70kg/hm2 and 103.08, respectively, providing accurate site points yield of training data for establishing the RFR region yield estimation algorithm.