Estimating rice yield by assimilating UAV-derived plant nitrogen concentration into the DSSAT model: Evaluation at different assimilation time windows

Abstract

Accurately simulating crop growth and estimating yield play a paramount role in policy decision making and agriculture management. To improve the accuracy of rice yield estimation at the field scale, plant nitrogen concentration (PNC), a highly nitrogen-related variable, was derived from the UAV-based imagery through a two-year rice experiment by using the random forest (RF) algorithm; then the PNC data was assimilated into the CERES-Rice model with a particle swarm optimization (PSO) method. In this study, the sensitivity of assimilation process to the acquisition time of UAV remotely sensed PNC data, including the vegetative stage, heading stage, ripening stage and the whole growth stage, was investigated. The results showed that the relationship between the estimated and measured PNC after data assimilation at the vegetative stage was stronger than that at other growth stages. Moreover, assimilating PNC at the early growth stage could better simulate the dynamics of PNC with no nitrogen (N) stress. Due to the assumption of soil homogeneity under various N fertilization treatments, all data assimilation strategies had the tendency to overestimate PNC in the N stress condition. In addition, the accuracy of yield estimation obtained by assimilating PNC at the vegetative stage was the highest. Accordingly, the assimilation of PNC data at the early rice growth stage could provide a great potential for improving yield estimation.