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).