Fig 1. Framework of the developed methodology in this study that applies systematically analyzed errors of observations and simulations in the data assimilation procedure to enhance crop status forecast. Field experiments in two successive years were conducted for the acquisition of the necessary dataset to validate this method. Daily weather data served as the forcing input of the crop model GECROS and all field observations in the first year were averaged for model calibration. Before conducting the calibration procedure, the parameters in GECROS were fixed or treated as uncertain and the error model that describes the uncertainty of crop model simulation was assumed. The uncertain parameters in GECROS and parameters in the error model were determined simultaneously by an efficient Markov Chain Monte Carlo approach (DREAM_zs). To further improve the forecast ability of the crop model, the in-season observations of leaf traits in the second year were incorporated by the commonly used data assimilation procedure, Ensemble Kalman Filter (EnKF), which integrates the sequential observations into crop model simulations of crop growth processes. Two types of observations were collected. The first one was from field destructive measurements. The second type was from the remote sensing predictions, which were regressed from the machine learning method of Gaussian Process Regression (GPR). The uncertainties of field measurements were derived from the replications in the 2nd year’s field experiment, while that of remote sensing predictions were estimated from the GPR model itself. With the systematically analyzed uncertainties of crop model simulations and observations, in-season leaf traits at the canopy level, leaf weight (W leaf), leaf nitrogen (N) content (N leaf) and leaf area index (LAI) were updated directly by EnKF, and other crop carbon and N statuses, including aboveground biomass (W above), grain weight (W grain), aboveground N content (N above) and grain N content (N grain), were updated accordingly as well. The performance of the updated states in the second year was evaluated by their in-situ measurements, accordingly.