Objectives Hyper spectral remote sensing has been used for monitoring crop growth and estimating the relationship between crop growth and related physiological indexes. The red edge position is considered as a piece of spectra closely related to crop nitrogen nutrition, and often used to monitor the content of chlorophyll or nitrogen in crops.The monitoring parameters of the red edge position and the data interpret method determine the accuracy and availability of the monitoring. So the monitoring parameters in potato were optimized and the accuracy of the six often used algorithms were compared in this paper.

Methods Field experiments were conducted in the northern Yinshan of Inner Mongolia of China from 2014 to 2016. Three potato cultivars were used as tested materials and 5 or 12 nitrogen application rates were set for the experiment. The canopy hyper spectral reflectance was collected at the seedling, tuber initiation, tuber bulking, starch accumulation and harvesting stages of potatoes. The nitrogen contents were measured and calculated using there flectants with six methods. The correlation between the measured and interpreted nitrogen contents was compared among the six algorithms.

Results The double peak phenomenon in the first-order derivative spectrum of potato was more obvious at the late growth stages. The red edge position was poorly correlated to plant nitrogen concentration due to the influence of soil background and noise spectrum at the seedling stage.There was a stronger correlation between the plant nitrogen concentration and the red edge position from the tuber initiation to starch accumulation stage, and the tuber expansion stage had the highest correlation. There was no continuity between the red edge positions obtained by the largest first derivative method and Lagrange interpolation. The linear extrapolation method resulted in the highest amplitude and standard deviation of the red edge position, reaching 44.6 and 9.3 respectively, while polynomial fitting method was the second best with amplitude and standard deviation of 15.1 and 2.6. Inverted Gaussian fitting and linear four-point interpolation had small amplitude and standard deviations. The linear extrapolation method had the highest coefficient of determination (R² = 0.55) in predicting the aboveground plant nitrogen concentration among the six methods. The coefficients of determination (R²), root mean square error (RMSE) and relative error (RE%) of the correlation between the predicted value and the observed value were 0.44, 3.96 g/kg and 11.46%, respectively. Inverse Gaussian fitting method and polynomial fitting method had similar coefficient of determination (R² = 0.30). Inverted Gaussian fitting method had better prediction ability in predicting plant nitrogen concentration (R² = 0.31, RMSE = 4.33 g/kg, RE = 12.03%).

Conclusions The red edge position can detect plant nitrogen concentration of potato from tuberous formation to starch accumulation stage. In spite of a slight saturation in the estimation of plant N concentration, the influence on overall prediction is minimum. Linear extrapolation could result in a large amplitude of red edge position, thus is very sensitive to the change of plant nitrogen concentration and can minimize the influence of red edge bimodal phenomenon. Linear extrapolation method had the highest R², lowest RMSE and RE%, making it a very suitable method for calculating the position of the red edge.