Improving crop model accuracy in the development of regional irrigation and nitrogen schedules by using data assimilation and spatial clustering algorithms

Crop growth models have been used to develop irrigation and nitrogen schedules (INSs). However, differences in crop cultivar coefficients are often ignored in the development of regional INSs. This study aimed to formulate suitable INSs under spatial heterogeneities in crop cultivar coefficients. Therefore, we propose two strategies for retrieving maize cultivar coefficients by using a data assimilation algorithm and spatial clustering algorithm. The first strategy involves determining the cultivar coefficients for each simulation unit in the examined region and then assigning different cultivar coefficients to the different clusters obtained using a spatial clustering algorithm, with the cultivar coefficients employed as clustering characteristics (CAs). The second strategy involves assigning different cultivar coefficients to the different clusters obtained using a spatial clustering algorithm on the basis of cultivar coefficients and geographical characteristics (CAGCs). By using observational data, the accuracy of cultivar coefficients CAs and CAGCs was compared with that of commonly used regional representative coefficients (RRs) and sub-region representative coefficients (SRRs). Furthermore, we formulated INSs with these four coefficients by using yield maximization as the objective and water use efficiency (WUE) and nitrogen use efficiency (NUE) as constraints. We examined the differences between the INSs, yields, WUE values, and NUE values obtained using each of the aforementioned four coefficients and those obtained using a point optimization approach. The results revealed that the highest accuracy in the simulation of the regional leaf area index, yield, and phenological stage was exhibited by the CAGCs-based strategy, followed by the CAs-based strategy. The RRs-based and SRRs-based strategies produced considerable errors. Crucially, the INSs obtained using the CAGCs-based strategy were more similar to those obtained through point optimization and more reasonable than were the INSs obtained using the other three strategies. In addition, more accurate yield, WUE, and NUE values were obtained with the CAGCs-based strategy than with the other three coefficients-based strategies. The results of this study indicate that the combination of a data assimilation algorithm and spatial clustering algorithm can improve the application potential of crop models in agricultural systems.