Conclusion
In this study, we developed an approach that combines several disparate quantitative methods into one framework for explicitly quantifying the uncertainties of crop model simulations and remotely sensed observations, contributing to an accurate crop growth forecast while avoiding filter divergence in DA. Our results showed that calibration and uncertainty estimation of a complex crop model benefited from an MCMC approach using the adapted residual error model. The calibrated uncertain parameters in crop model performed reasonably well in validation. The GPR models for analyzing remote sensing data provided temporal and spatial predictions and corresponding uncertainties of leaf traits. Applying the quantified uncertainties into EnKF to update the leaf traits W leaf, N leafand LAI enabled the crop model simulation to agree better with the measurements, without filter divergence. More importantly, updated simulations of in-season and end-of-seasonW above, W grain,N above and N grain also performed better than those of simulations without assimilating observations. The developed method is geared toward cases where multiple crop traits are observed and in-situ crop and soil information becomes increasingly available with the rapid development of remote sensing technologies. Armed with the precise forecast of in-season crop carbon and N status, field management can be better optimized to support sustainable smart farming.