<i>Combining UAV-based vegetation indices, canopy height and canopy coverage to improve rice yield prediction under different nitrogen levels</i>

Abstract

Abstract. Timely and accurate monitoring crop production and forecasting yield before harvesting are valuable for precision management, policy and decision making, and marketing. The spectral information such as vegetation indices (VIs) calculated from unmanned aerial vehicle (UAV)-based images have shown the potential for crop yield prediction, but the reliability and the accuracy haven‘t been satisfied yet. This study was aimed to explore the potential of fusion of UAV images-based spectral and structural information extracted from the entire growth period of rice crop to improve the grain yield prediction. Field experiments with different nitrogen (N) applications were conducted in 2017 and 2018, and ground measurements including leaf chlorophyll content (LCC), aboveground biomass (AGB) and grain yield were performed. A UAV platform carrying RGB and multispectral cameras was employed to collect high spatial-temporal resolution images of rice crops. The VIs, canopy height and canopy coverage were then extracted from RGB and multispectral images, which was used to develop random forest (RF) prediction models for grain yield. The results showed that combining normalized difference vegetation index (NDVI), normalized difference yellowness index (NDYI), canopy height and canopy coverage achieved the best prediction of grain yield with the determination coefficient (r2) of 0.85, prediction of root mean square error (RMSEP) of 0.39 t/ha, and relative RMSE (rRMSE) of 3.56% in 2017, and r2 of 0.83, RMSEP of 0.33 t/ha, and rRMSE of 2.75% in 2018, respectively, which outperformed the results in the reported studies. Furthermore, the initial heading stage was proved to be the optimal stage to evaluate the grain yield, and a model transfer strategy at the initial heading stage in 2017 was successfully implemented to predict the grain yield in 2018 with the rRMSE of 6.71%. These findings demonstrated that fusion of canopy spectral and structure information obtained from UAV-based RGB and multispectral images can improve the prediction accuracy of grain yield as well as achieve a dynamic monitoring of crop growth status. The proposed approach could provide a valuable insight for the high spatial-temporal precision management in agriculture.