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.