4.5 Prospect of updated crop status forecast for field-level in-season management
Since most remote sensing algorithms have been adapted to large areas, making them applicable to the field-level is a priority now, especially in countries like China (Weiss et al., 2020) where the agricultural system is dominated by millions of smallholders and N is always overdosed (Cui et al., 2018). Spatial and temporal predictions ofN leaf (Fig. 5) are crucial for determining the management of fertilization timing to be performed at specific growth stages (Weiss et al., 2020). However, the use of remote sensing data solely is not sufficient to quantify top-dressing requirements (Weiss et al., 2020), due to the manifold interactions in the soil-crop-atmosphere continuum. Assimilating supplementary information from a crop model with remote sensing data has been identified as one of the most promising methods to make field management decisions (Jin et al., 2018; Weiss et al., 2020). For instance, Baret et al. (2007) demonstrated that an optimized in-season N application map can be generated by assimilating remotely sensed LAI and N above into the crop model STICS, in which the N application rate in each map pixel (20×20 m2) was optimized by maximizing the productivity, using the historical meteorological data between the time of decision and harvest as the unknown future weather conditions. With optimized real-time fertilizer management, severe environmental issues caused by overfertilization can be reduced, thereby enabling smart farming and sustainable agricultural production (Berger et al., 2020b).