The deriving of sensitive waveband for the estimation of plant nitrogen concentration in potato based on hyperspectral indices

Objectives Nitrogen nutrition diagnosis based on spectral indices is one of the ways to rapidly obtain nitrogen nutrition status of crops. Among them, the ratio and normalized spectral indices constructed by visible light and near-infrared have important significance for estimating the nitrogen nutrition status of crops. In order to evaluate the feasibility of the existing ratio and the normalized spectral index in the diagnosis of nitrogen concentration in potato plants during the critical growth period, the problems of indices saturation and data dispersion in the diagnosis process of nitrogen nutrition were solved.

Methods From 2014 to 2016, four multi-point field experiments with different nitrogen fertilizer application rates were conducted in Wuchuan County and Siziwang banner, Inner Mongolia. The vine and tuber samples of potato in the tests field and adjacent farmlands were collected to analyze the nitrogen content, and the spectral data were collected 50–80 cm above the potato canopy during potato growth stages of tuber initiation, tuber bulking and starch accumulation. The accuracy of the model was verified by farmland potato data. Correlation and estimation models of 12 published ratios and normalized spectral indices and band-optimized spectral indices with plant nitrogen concentration were built in the critical growth periods of potato.

Results The distribution range of nitrogen concentration in potato plants was 1.89%–4.69%, the average nitrogen concentration was 3.30%, and the coefficient of variation was 18.75%. The data of the verification set were from farmland. And the distribution range of nitrogen concentration in potato plants was 2.00%–4.92%, the average nitrogen concentration was 3.34%, and the coefficient of variation was 19.27%. The experimental results showed that the bands of blue violet at 400–450 nm and red edge at 690–720 nm were the sensitive bands for nitrogen concentration estimation of potato plants. Some existing spectral indexes could be used for nitrogen concentration estimation of potato plants, but the absence of blue violet band greatly reduced the accuracy of estimation. The optimal spectral indices positions of RSI and NDSI were 430 nm, 694 nm and 426 nm, 694 nm, respectively. The linear estimation model of nitrogen concentration of potato plants based on the optimized spectral indices NDSI (426 nm, 694 nm) was y=−6.87x+6.08, with the highest determination coefficient R² (0.68). The linear estimation model of RSI spectral indices with potato plant nitrogen concentration was y =−1.11x+ 5.92 (R² =0.65), Compared with the existing ratio and normalized spectral indices, optimization of spectral indices RSI and NDSI overcomed the high nitrogen concentration spectrum index under the condition of low saturation caused by sensitivity phenomenon, and significantly improved the potato plants linear modeling effect of nitrogen concentration in potato plant. The verification data of farmers’ field showed that the estimated value of the estimation model was close to the 1∶1 line with the measured value, and the NDSI spectral indices estimation model had the best verification effect, with the mean relative error RE and root mean square error RMSE being 10.58% and 0.42%, respectively.

Conclusions In this study, the nitrogen concentration sensitive bands of potato plants with ratio and normalized spectral index are determined by band optimization algorithm. Using blue-violet light 400–450 nm and red edge 690–720 nm to estimate nitrogen concentration of potato plants can improve the index sensitivity and data dispersion problem in the diagnosis process of high nitrogen concentration of potato plant, and improve the accuracy of nitrogen nutrition diagnosis of potato plant.