4.3 Performance of the analyzed system uncertainties in data
assimilation
With the estimated uncertainties in crop model simulations and remote
sensing observations, the inflation factor was no longer needed in this
study. Filter divergence did not occur during the DA process with the
analyzed system uncertainties (Fig. 6). 96% off rc of leaf traits in DAfm and
76% in DArs ranged from -0.9 to 0.9, indicating at
least 10% uncertainty reduction of those leaf traits after conducting
DA (Fig. 6a-b), and NRMSE of updated leaf traits on average
decreased by 66% and 62% by DAfm and
DArs, respectively (Fig. 7a). This large improvement of
directly updated states was in line with the results of Kivi et al.
(2022) who estimated system uncertainties for EnKF while assimilating in
situ observed daily soil moisture into the crop model APSIM. Even though
the forecast accuracy of soil moisture was improved, the performance of
updated LAI and yield tended to be worse than that of simulations
without assimilating soil moisture (Kivi et al., 2022), which conflicts
with previous studies (de Wit and van Diepen, 2007; Ines et al., 2013).
As with the similar concerns of Schoups and Vrugt (2010) about the
approach of simultaneous optimization and data assimilation proposed by
Vrugt et al. (2005), crop growth simulation seems to be affected by
measured soil moisture during joint estimation of system uncertainty,
causing poorer performance in updated LAI and yield (Kivi et al., 2022).
Without such tangled system uncertainties, our results showed that the
performance of in-season updated and end-of-season forecasted crop
carbon and N status improved (Fig. 7). However, there is still scope for
further improvement. For instance, the distributions off rc of updated aboveground status tended to be
more diverged and closer to one (Fig. 6c-d), while those of updated leaf
traits were more converged and closer to zero (Fig. 6a-b), implying a
limitation in updating leaf traits for an accurate forecast of theW above and N above.
Meanwhile, the end-of-season forecast of W aboveand N above might also be impacted by their
insufficiently updated in-season status (Fig. 7b-c, S8a,c).