In-season maize nitrogen status diagnosis and yield prediction from unmanned aerial vehicle (UAV) mounted hyperspectral images

This study explores how hyperspectral sensors mounted on unmanned aerial vehicles (UAVs) may assist in maize nitrogen status monitoring and yield prediction using traditional regression and machine learning models. Three field experiments were conducted with three soil types (alluvial soil, black soil, and aeolian sand soil) and various nitrogen (N) fertilizer treatments (0, 168, 240, 270 and 312 kg N/ha) in Lishu County, Northeast China. Hyperspectral images obtained from sensors mounted on UAVs were collected to monitor the nitrogen nutrition index (NNI) at the jointing, silking, and maturity stages and the grain yield of maize in 2019 and 2020. In comparison to the prediction performance of the partial least squares regression (PLSR) and random forest (RF) regression models, 13 narrowband vegetation indices (VIs) and N application rates were employed as predictors to determine N status at the field scale. The results revealed that most VIs were significantly correlated with the NNI and yield at different stages and that the Maccioni index was the most influential predictor for both the NNI and yield estimation based on the relative importance calculation results of the different predictors. Compared with the PLSR model, the NNI and yield were better estimated by the RF model, except for yield estimation at the jointing and silking stages. The best performance for maize NNI and yield estimation was achieved at the silking stage and maturity stage, respectively. Based on the relationships between the NNI and yield, the NNI intervals were 0.91–1.30 for alluvial and black soil and 0.67–0.72 for aeolian sandy soil, with the goal of achieving the target yield. This study highlights the significance of introducing UAV technologies in providing a field-scale data-driven approach for crop nitrogen monitoring and yield prediction information to farmers and policy-makers for better precision crop management.