Abstract: Leaf area index (LAI) is one of the key indicators to measure crop growth. Accurate and efficient inversion is very important for crop monitoring and yield prediction. However, the traditional empirical model can often be confined to estimating the LAI, such as the heavy computation load and weak generalization. This study aims to inverse the silage maize LAI from the remote sensing images using a machine learning optimized PROSAIL model. The research object was taken as the LAI of the field silage maize in Minle County, Gansu Province, western China. The multi-time series Landsat-8 multi-spectral image and the LAI data were then collected in the field at the same period. Four kinds of machine learning hybrid inversion models were proposed using the global Extended Fourier Amplitude Sensitivity Test (EFAST), including MLP-PROSAIL, SVR-PROSAIL, RF-PROSAIL, and GBM-PROSAIL. The sensitivity and distribution range of important parameters were determined to implement the inversion operation. The first-order EFAST and full-order global sensitivity analysis were used to determine the contribution rate of the different parameters to the model output. The sensitivity input parameters of the PROSAIL model were accurately simulated to output the canopy reflectance spectra. Furthermore, the pre-processing was performed on the Landsat-8 multi-spectral data, including the radiometric calibration, FLAASH atmospheric correction, geometric correction and registration, and tailoring of the study area. Then, the band transformation was carried out using the Landsat-8 spectral response function. The continuous spectral reflectance of simulated output was converted into the spectral bands to match the satellite sensors. Five bands with high sensitivity were selected for the model training, considering the sensitivity difference of LAI to different bands. The geographical registration tool was combined with the inverse distance weighted interpolation strategy. The scale difference was then reduced. Bayesian hyperparameter optimization and regularization were used to optimize the different parameter types and activation functions of the models. Four improved models were obtained for training LAI and spectral data. The inversion performance of the four models was verified by the 5-fold cross-validation and the leave-one validation, including the necessary importance analysis for the band input. The performance of the four models was significantly improved after optimization. The coefficient of determination R² was more than 0.85, and the root mean square error (RMSE) was controlled within 0.80. Among them, the GBM-PROSAIL shared the highest inversion accuracy and the best-fitting performance. The R² was about 0.93, the RMSE was about 0.42, and most LAI values were within a 95% confidence interval, and almost all LAI values were included in a 95% prediction interval. The RF-PROSAIL model was followed by the RF-PROSAIL model with an R² of about 0.90 and RMSE of about 0.51. There were also a large number of LAI values within the 95% confidence interval, and only a few LAI values exceeded the 95% prediction interval boundary. The fitting accuracies R² of the rest two hybrid models were 0.85 and 0.88, respectively, and the RMSE values were 0.80 and 0.69, respectively. The spatial distribution of the multi-time series LAI inversion was drawn by the GBM-PROSAIL model in the study area. The LAI values of the silage maize at the node stage were mainly distributed between 2 and 4, which was slightly larger than the measured. This was related to the influence of LAI on the vegetation. LAI values of the silage maize in the tasseling stage were distributed mainly from 5.0 to 7.0, which was basically consistent with the measured. There was a significant difference in the LAI values of silage maize in different growth stages, indicating its growth process. The hybrid inversion model shared high performance and strong robustness. The finding can provide a strong reference for the multi-time series, large-scale crop monitoring, and yield forecasting in precision agriculture.