Objectives We compared accuracy of the critical nitrogen dilution curve of flue-cured tobacco leaves constructed with the traditional method and by Bayesian statistics method, aiming for a simple and accurate nitrogen nutrition diagnosis method for flue-cured tobacco.

Methods A two-year field experiment, and three nitrogen application rates respectively were conducted. The dry matter accumulation and nitrogen content of flue-cured tobacco leaves at different growth days after transplanting were analyzed. According to the maximum dry matter weight of the leaves at each sampling period, all treatment samples were divided into nitrogen-limited growth groups and non-nitrogen-limited growth groups. The leaf critical nitrogen dilution curves were then established using Bayesian statistics and the traditional two-step method, respectively. The R language package rjags was used to implement the Markov Chain Monte Carlo (MCMC) algorithm to analyze the posterior distribution characteristics of the critical nitrogen dilution curve model parameters. The discriminability of the curves for nitrogen-limited and non-nitrogen-limited groups was analyzed, and the leaf critical nitrogen concentration and nitrogen nutrition index were calculated and compared with the actual observed values.

Results The application of nitrogen fertilizer significantly increased the dry matter accumulation of flue-cured tobacco leaves, with significant differences between different nitrogen treatments. The nitrogen concentration of flue-cured tobacco leaves decreased as the growth process progressed. The 95% posterior distributions of parameters A1 and A2 were 2.58−2.94 and 0.13−0.33, respectively; the uncertainty level (width of the 95% credible interval) of the fitted flue-cured tobacco leaf critical nitrogen dilution curve decreased and then increased with the increase of leaf dry matter weight, with an uncertainty level of 0.16%−0.70%. The discriminability of the flue-cured tobacco leaf critical nitrogen dilution curve constructed based on Bayesian statistics for nitrogen-limited and non-nitrogen-limited groups was 71%, which was better than the discriminability of the curve constructed by the traditional two-step method. The correlation between the leaf critical nitrogen concentration fitted by the two methods and the actual critical nitrogen concentration was basically consistent.The nitrogen nutrition index calculated by the curve based on Bayesian statistics was slightly higher than that of the two-step method, and the nitrogen nutrition index calculated by the two methods showed a high linear correlation, with a determination coefficient R² of 0.96 and a normalized root mean square error n-RMSE of 6%, indicating high stability.

Conclusions The critical nitrogen dilution curve of flue-cured tobacco leaves constructed based on Bayesian statistics is N_c=2.74×LDM^−0.22. This curve can reflect the uncertainty of the model. The effect of the Bayesian curve in fitting the critical nitrogen concentration is not significantly different from the two-step method curve, but it can better distinguish between nitrogen-limited growth groups and non-nitrogen-limited growth groups compared to the two-step method curve. Moreover, the nitrogen nutrition index calculated by the Bayesian curve has a high linear variation determination coefficient R² of 0.96 with the two-step method curve, indicating that the Bayesian curve can be used more simply and accurately to evaluate the nitrogen nutrition status of flue-cured tobacco.