Estimating leaf nitrogen concentration considering unsynchronized maize growth stages with canopy hyperspectral technique

•Partial least squares (PLS) regression and two-band optimal combinations algorithms were compared.•Spectral models were not affected by unsynchronized maize growth stage.•Best performing spectral indices and their band combinations varied across growth stage.•FDR-PLS regression model have much better predictive performance with respect to LNC than the newly proposed 2-band index. Accurate monitoring of the leaf nitrogen concentration (LNC) in maize can provide a fundamental basis for effective N management. In recent decades, many spectral indices and algorithms can accurately estimate the crop N status during different growth stages of maize. However, the effect of unsynchronized maize growth stages on these spectral models is rarely considered. The objectives of this study were to verify the predictive ability of the published vegetation indices (VIs), the partial least squares (PLS) regression and the two-band optimal combinations algorithms, and to determine the most accurate method for assessing the LNC of unsynchronized growth stages in maize. Canopy raw and first-derivative reflectance (FDR) spectra, and destructive measurements of the maize LNC were collected in 2016–2018 in different ecological areas, unsynchronized growth stages, cultivars, plant densities, and N rates. The published VIs and new 2-band VIs and their band combinations varied across different growth stages and were not affected by unsynchronized growth stages. The red-edge chlorophyll index (CIred edge), which was identified across growth stages, performed quite well for LNC estimation, but performed unsatisfactorily when compared to the new 2-band VIs that were developed in this study. The best spectral index of the green or red and red-edge or near-infrared band combination based on FDR spectra had a good diagnostic effect for LNC of maize across the four growth stages, in which the novel normalized difference spectral indices (NDSI) effect was optimal in the V9 (9-leaf stage), VT (tasseling stage) and R1 (silking stage) stage and the novel ratio spectral indices (RSI) was the best in the R3 (milk stage) stage. Compared to PLS regression based on raw full-range hyperspectral data, the PLS regression for estimating LNC across four growth stages based on FDR full-range hyperspectral data showed a higher accuracy, with an average coefficient of determination (rval2) of 0.87 and average root mean square error (RMSEval) of 0.18. In addition, the average rval2 for the PLS regression b