Research on potato (Solanum tuberosum L.) nitrogen nutrition diagnosis based on hyperspectral data

Reducing the overapplication of nitrogen fertilizers to potatoes (Solanum tuberosum L.) can reduce production costs and their impact on the environment. One approach to produce these impacts is to reduce overapplications of fertilizers by using the nitrogen nutrition index (NNI = plant nitrogen concentration/critical nitrogen concentration) as a basis for in-season nitrogen recommendations. The objective of this study was to create a remote sensing-based algorithm to estimate NNI. This study collected hyperspectral data (350-1830 nm) during the potato tuber formation period in 2022 and 2023. The climate regime for the study area was a mid-temperate semiarid continental monsoon; in our study, three different spectral parameter calculation methods were employed. First, the empirical vegetation index, determined through a fixed two-band calculation. Second, the optimal vegetation index, computed on a band-by-band basis. Lastly, the trilateral spectral approach, wherein the indicators are typically associated with the red edge, blue edge, and green edge. The optimum vegetation index had the highest correlation with NNI. The support vector machine, random forest (RF), and back propagation neural network models were used to create NNI prediction models. All machine learning models effectively estimated NNI, and during validation, the R-2 (coefficient of determination) was >0.700. In general, the RF model outperformed the other models and during validation had an R-2 of 0.869, a root mean square error of 0.052, and a relative error of 5.504%. This study demonstrates the scalability, simplicity, and cost-effectiveness of combining hyperspectral technology and machine learning for rapid potato NNI estimation.