Establishment and variation of soil inorganic nitrogen content estimation model based on maize leaf spectral characteristics

Objectives The color of crop leaves directly reflects the nutrient supply of soil. We studied the relationship between intensity of leaf chromospectrum related to N nutrition and the content of soil inorganic nitrogen, in order to establish a model to realize real-time monitoring of nitrogen contents in crops and soils by hyperspectral technology.

Methods A maize (Zea mays L. cv Zhengdan 958) experiment was conducted in 2017 and 2018. Six nitrogen application levels were setup, they were 0, 60, 120, 180, 240 and 300 kg/hm², respectively. At the jointing, booting, anthesis-silking and filling stages, leaves hyperspectral reflectance and soil inorganic N content were measured, and the relationship model between spectral reflectance of leaves and soil inorganic nitrogen contents was established, and the precision of the prediction by the spectral parameters model was evaluated.

Results The soil inorganic N content decreased with the growth period, and topdressing of N fertilizer significantly improved the soil inorganic N content in the middle and late growing stage of maize. The intensity of spectral reflectance of leaves at the jointing and anthesis-silking stages was negatively correlated with the soil inorganic N content in the visible light range, and positively correlated in the near-infrared band. There was a negative correlation in the visible band at the booting stage and no obvious correlation at the filling stages. In the spectral parameter model, the optimal spectral index for the prediction of soil inorganic N content at the four growth stages was RVI-2, RSI (534, 726), RSI (567, 519) and RVI-2, respectively, and the R² values of verification sets of the regression model were 0.642, 0.749, 0.696 and 0.540, respectively. In the PLSR prediction model, the R² values of the verification of the diagnostic model were 0.876, 0.838, 0.765 and 0.595, and the RPD (ratio of percent deviation) were 2.140, 2.077, 2.002, 1.369, respectively.

Conclusions Satisfactory prediction accuracy in soil inorganic nitrogen content is acquired by the PLSR prediction model based on leaf spectral reflectance at the jointing, booting and anthesis-silking stages of maize. The two-year field experiment has proved its feasibility to diagnose soil inorganic N content.