Four multi-locational experiments with varied N application rates were conducted at Xinxiang and Jiaozuo from 2016 to 2018 using the six most widely grown winter wheat cultivars in the region. The detailed description of soil characteristics of experimental sites, cultivars, N rates, planting, sampling, and harvesting timing is shown in Table 1 . The weather conditions during the experimental period are shown in Figure 1 . Treatments were replicated thrice using a randomized complete block design. The size of each plot was 24 m² (6 m × 4 m). The seeding rate was 240 kg ha⁻¹. N fertilizer (urea, 46% N) was applied (50:50) as basal and top dressing before sowing and at Feekes stage 4. P (as P₂O₅) and K (as K₂O) were applied at the rates of 120 kg ha⁻¹ and 105 kg ha⁻¹, respectively. The broadcasting method of fertilizer application was used to apply fertilizer to each plot. Weeds, pests, and diseases were managed using chemical methods, with chemicals applied at recommended rates. Each plot was irrigated evenly using surface irrigation with a 4-inch plastic-coated hose. N fertilizer application was the only factor limiting crop growth and productivity.

Twenty plants were sampled from each experimental plot at different Feekes stages for the determination of plant dry matter (DM) and plant N concentration (PNC). Plant samples were oven-dried for 48 h at 80°C to get a constant weight and then weighed on an analytical balance, followed by grinding and sieving through a sieve for lab analysis. Samples of 0.15 g of ground plant biomass were digested using a mixture of H₂O₂ and H₂SO₄, and the PNC was determined using the Kjeldahl method (Bremner and Mulvancy, 1982).

Grain yield at maturity was determined by harvesting a 1 m² area from each plot and adjusting the moisture content to 14%. A thousand grain weight (TGW) was measured from 1,000 grain samples. Grain protein content was measured by multiplying the grain N concentration by a factor of 6.25, while protein yield was measured by multiplying crop yield by grain protein content. The descriptions, abbreviations, and units of all indices are described in Supplementary Table S1 .

Plant N_and was estimated by subtracting the critical plant N accumulation (PNA) from the actual PNA across different N rates at each sampling stage (Ata-Ul-Karim et al., 2013). The dilution curve of N_c concentration (N_c = 4.15DM^−0.38) of winter wheat was used to calculate the value of N_and (Yue et al., 2012).

Where N_cna is the PNA under the N_c condition (kg ha⁻¹), N_na is the actual PNA across varied N (kg ha⁻¹), and 10 is the conversion factor. If N_and is equal to 0, optimal N nutrition is observed. If N_and = 0, N nutrition was considered optimum, while N_and >0 indicated N deficiency and v<0 pointed out luxury consumption (Ata-Ul-Karim et al., 2013).

Agronomic N use efficiency is the amount of additional grain harvested per kilogram of N fertilizer applied. AE_N could be represented as G_w/N_f, where G_w is grain yield and N_f is N fertilizer application (kg ha⁻¹) (Halvorson et al., 2005). This is defined as (G_wi − G_w0)/N_f where i is the level of N fertilizer under various treatments and 0 is the N fertilizer under control treatment.

The two primary components of AE_N are: (1) N fertilizer recovery efficiency (RE_N), which describes the recovery efficiency of fertilizer N from soil (N_ti − N_t0)/N_f (Zhang et al., 2008), and (2) physiological N use efficiency (PE_N) is the measure of the ability of a plant to produce grain/biomass with N acquired at the whole plant level (G_wi − G_w0)/(N_ti − N_t0) (Zhang et al., 2008), where N_t represents plant N accumulation at harvest. Therefore, AE_N was expressed by the following equation:

The data of PNC, grain yield, grain protein content, grain protein yield, TGW, AE_N, RE_N, and PE_N were subjected to one-way analysis of variance analysis for comparing the statistical difference using the SPSS software package version 22 (SPSS Inc., Chicago, IL, USA). The difference between the data means was assessed using the least significant difference (LSD) test at the 5% level. The fixed effects included cultivar, N treatments, and site, while the random effects included block. The fitted linear model was estimated based on the least square method.

Data acquired from field experiments in 2016–2017 were used to develop the relationship between N_and and AE_N and its components, while the data acquired in 2017–2018 were used to validate the newly developed relationships. Root mean square error (RMSE), relative root mean square error (RRMSE), and a 1:1 plot were used to determine the robustness of newly developed models between the estimated and observed values of AE_N, RE_N, and PE_N. RMSE and RRMSE were calculated as follows:

where n represents the sample number, and P_i and O_i represent estimated and observed values, respectively. Õ is the average value of the observed values.

Plant N concentration increased with the increase in N input rates, and a significant difference was observed for PNC under different N treatments and growth stages ( Figure 2 ). PNC significantly declined from Feekes stage 6 to Feekes stage 11.4 with the growth of winter wheat during the 2016–2017 season at Xinxiang and Jiaozuo ( Supplementary Tables S2, S3 ). The maximum value of PNC was observed at Feekes stage 6 for both cultivars, while the minimum value of PNC was observed at Feekes stage 11.4. PNC ranged from 1.19% to 4.18% and 1.09% to 3.46% for ZM22 and ZM27, respectively, during the 2016–2017 season of Xinxiang ( Figures 2A, B ), while PNC during the 2016–2017 season of Jiaozuo ranged from 1.18% to 4.05% and 1.18% to 3.76% for ZM22 and ZM27, respectively ( Figures 2C, D ). Overall, ZM22 showed higher PNC than ZM27 across different N treatments, growth stages, seasons, and experimental sites.

Grain yield, grain protein content, grain protein yield, and TGW were not affected (p<0.05) by the interactive effect of cultivar × N rate, cultivar × site, N rate × site, and cultivar × N rate × site. The grain yield of both cultivars was affected by the application of N fertilizer ( Table 2 ), and maximum grain yield was observed with ZM27. ZM27 showed 10% higher grain yield than ZM22. The grain yield of both cultivars declined under maximum N application (300 kg ha⁻¹) during the 2016–2017 wheat growing season. Higher grain protein yield and grain protein content were observed for ZM22 as compared to ZM27. Grain protein yield and protein content of ZM22 were 20% and 8% higher than those of ZM27, respectively. Besides, grain protein yield and grain protein content were also affected by N application. Grain protein yield and grain protein content of both cultivars showed an average increase of 19% and 137%, respectively, as compared with the N0 treatment. However, little variation was observed in grain protein yield and grain protein content across the sites. Grain protein content was slightly higher at Xinxiang than at Jiaozuo, while grain protein yield was higher at Jiaozuo. Additionally, the TGW of both cultivars was not affected by fertilizer application rates.

The accumulated nitrogen deficit varied from −65.73 to 104.37 kg ha⁻¹ under different N rates across seasons, cultivars, sites, and growth stages of winter wheat. There were substantial differences in N_and across different N treatments and cultivars ( Figure 3 ). The N_and showed a declining trend with increasing N application rates, while it increased gradually towards advancing maturity. This increase towards advancing maturity was obvious under N-limiting treatments (N0, N75, and N150), and it reached its maximum value at Feekes stage 11.4. Conversely, this increase was minor under optimal N conditions (N225), while under non-N-limiting treatment (N300), excess N accumulation was observed until Feekes stage 11.4. The N_and ranged from 104.37 to −44.42 kg ha⁻¹ and 93.07 to −40.64 kg ha⁻¹ during the 2016 to 2017 season at Xinxiang for ZM22 and ZM27, respectively, while it ranged from 91.97 to −38.27 kg ha⁻¹ and 102.08 to −65.73 kg ha⁻¹ during the 2016 to 2017 season at Jiaozuo for ZM22 and ZM27, respectively. Non-significant differences were observed for N_and across different cultivars and sites (p-value = 0.817). The N_and values were higher than 0 for N0, N75, and N150 treatments, were almost equal to 0 for the N225 treatment, and were lower than 0 for the N300 treatment across sites and seasons.

Agronomic N fertilizer use efficiency and its components (RE_N and PE_N) were not affected (p<0.05) by the interactive effect of cultivar × N rate, cultivar × site, N rate × site, and cultivar × N rate × site. AE_N, RE_N, and PE_N were affected by N fertilizer application ( Table 3 ). The values of AE_N, PE_N, and RE_N ranged from 12.5 to 23.56 kg kg⁻¹, 23.51 to 37.39 kg kg⁻¹, and 31.6% to 55.7%, respectively, across different N rates during the 2016 to 2017 season ( Table 3 ). AE_N, RE_N, and PE_N were higher at 75 kg N ha⁻¹ than those at 300 kg N ha⁻¹, which indicates that AE_N, RE_N, and PE_N declined with increasing N supply. ZM27 showed higher values of AE_N and PE_N than ZM22. The ZM27 of AE_N and PE_N were 42% and 37.6% higher than those of ZM22, while the RE_N of ZM22 was 5% lower than that of ZM27. However, non-significant differences were observed for AE_N, RE_N, and PE_N of wheat cultivars at both sites; the minor differences might be attributed to experimental error ( Table 3 ).

Accumulated nitrogen deficit was positively correlated with AE_N, RE_N, and PE_N for both cultivars during the 2016 to 2017 seasons at Xinxiang and Jiaozuo ( Figures 4 – 6 ). The robustness of these relationships was validated using data acquired from an independent experiment from 2017 to 2018. The RMSE of 3.43 kg kg⁻¹, 4.22%, and 3.67 kg kg⁻¹ and RRMSE of 17.53%, 12.46%, and 13.17% were observed for AE_N, RE_N, and PE_N, respectively, while the values of R² were 0.67, 0.73, and 0.85 for AE_N, RE_N, and PE_N, respectively, indicating a good relationship between the observed and predicted AE_N, RE_N, and PE_N values ( Figure 7 ).

The nitrogen fertilizer application rate affects PNC during the growth period of winter wheat. PNC generally increases with increasing N application rates. Plants can accumulate excessive N nutrition in their organs (leaf and stem). A significant increase in plant dry matter accumulation and leaf area expansion with an increase in PNC under N-limiting conditions has been previously reported; however, excessive N application beyond the threshold level (non-limiting N conditions) poses negative impacts on plant growth (Yao et al., 2014; Zhao et al., 2014). Additionally, when soil N level is high, the result shows little increase in PNC or yield as N fertilizer increases in some studies (Feng et al., 2008). This might be because plants tend to absorb soil N, not fertilizer N (Grant et al., 1999; Lafond et al., 2008).

The increase in grain yield from N0 to N225 treatment in this study indicated that the winter wheat plant has very high N requirements ( Table 2 ). However, the effect of N fertilizer on yield changes with site and year. An obvious change in yield under different soil types, climates, and cultivars in south Asia has been previously reported (Ladha et al., 2003). The change in winter wheat yield was mainly attributed to the change in yield components (the number of spikes per plant and grain weight per plant), and there was a significant effect of N application on the formation of yield components in winter wheat (Zhao et al., 2020). Excessive N application does not always warrant higher grain yield but decreases N use efficiency and poses negative impacts on the environment due to environmental loss of applied N up to ∼180 kg ha⁻¹ yr⁻¹ (Vitousek et al., 2009). The recommended N application rates of 120–170 kg N ha⁻¹ to balance grain yield, NUE, and N loss in the North China Plain have been previously reported (Cui et al., 2009; Ma et al., 2019).

The effect of N application rates was more obvious on grain protein yield than grain protein content, which was attributed to the impacts of N fertilizer on grain yield and protein content. The grain protein content increased irrespective of N supply conditions; however, the increase in grain protein content was minor under N non-limiting treatments (Zhao et al., 2020). The non-significant effect of N fertilizer on TGW agreed with previous reports (Yan et al., 2019).

The variability in N_and with N rates and crop growth stages in this study was in consensus with a study on rice (Ata-Ul-Karim et al., 2013). N_and is a suitable index for in-season crop N deficit monitoring and can potentially be used to quantify the amount of supplemental N fertilization scheduled during crop growth. N_and has advantages over other indices such as PNC (Guillard et al., 2021), DM (Zhu et al., 2003), and chlorophyll concentration (Li et al., 2019) as it contains more information on crop growth conditions by integrating PNC with plant dry matter while diagnosing plant N status (Lemaire et al., 2008; Ziadi et al., 2008; Ata-Ul-Karim et al., 2017a).

Accumulated N deficit can be exploited for in-season estimation of crop N status and to fine-tune the supplemental N scheduling during the growth period as it is crop-specific, precise, and theoretically sound in relation to actual crop growth (Ata-Ul-Karim et al., 2017b). Being based on the N dilution theory, N_and is not only theoretically sound but also relevant agronomically as an N diagnostic tool to quantify the in-season crop N status. Results indicated that the winter wheat plant invested a higher proportion of dry matter in the stem (low organ N concentration) after Feekes stage 6 as compared to the leaf (high organ N concentration) to ensure optimal plant height and capture more light. Additionally, the decline in the N concentration of shaded leaves per unit leaf area contributes to optimizing plant N allocation in relation to light distribution for improving canopy photosynthesis (Lemaire et al., 2008).

The lower variability of AE_N and its components across cultivars and sites might be due to similar climatic conditions and management practices at both sites ( Figure 1 ). The decrease in AE_N under N non-limiting treatments indicated that higher N fertilizer losses occurred under higher N application rate treatments. The variability of AE_N and PE_N between cultivars in this study was in consensus with inter- and intra-specific genetic variability for N fertilizer use among crop cultivars and species (Wu et al., 2006; Salim and Raza, 2020). The higher values of AE_N and PE_N of ZM27 indicated that a higher amount of grain yield was produced per unit of N fertilizer applied as compared to ZM22. Both cultivars showed non-significant differences in RE_N, which indicates that AE_N and PE_N were mainly affected by the characteristics of the cultivars rather than RE_N, which might be affected by external environments. Plant N uptake under field conditions is also affected by the high spatial and temporal heterogeneity in soil N availability (Finger et al., 2019). The lower RE_N at a high level of soil N was due to the lower absorption of fertilizer N by winter wheat plants. The accumulated NO⁻³-N in the soil poses an obvious effect on the N fertilizer use efficiency as a supplement of soil N has a very high replacement value for N fertilizer (Cassman et al., 2003). The proportion of N applied that is not absorbed by the plant under non-limiting N conditions increases the risk of N leaching due to the movement of applied N to deeper soil layers during the crop growth period (Villar-Mir et al., 2002).

The significantly positive relationships of N_and with AE_N, RE_N, and PE_N indicated that N_and can be potentially applied to predict AE_N, RE_N, and PE_N during the crop growth period. N_and can also be used for the estimation of crop N requirements, which can be calculated as the difference between critical N accumulation and actual N accumulation (Eq. (2)). N_and has successfully differentiated the N sub-optimal optimal and supra-optimal N growth conditions ( Figure 3 ), and N_and = 0 indicates that N nutrition is optimum, while N_and >0 indicates N deficiency and<0 points out luxury consumption (Ata-Ul-Karim et al., 2013). The relationships between these characteristics can also contribute to a better understanding of N fertilizer management for increasing agronomic N use efficiency. The high yield was the result of better exploitation of N, which is accompanied by lower AE_N with the increase in N application. RE_N describes the recovery efficiency of fertilizer N by the crop from the soil. RE_N can serve as an important parameter for strategizing precise and quantitative N application technologies. However, it is affected by various factors (soil type, N application, agro-climatic conditions, yield level, and cultivars; Conant et al., 2013). RE_N was approximately 40% in this study when N_and was equal to 0. A previous study reported that 40% of the RE_N of winter wheat could be considered a critical value in a high-yield field (Xu, 2010). RE_N will decrease with the decline of N_and when N_and is lower than 0 due to lower absorption of fertilizer N from the soil by the plant because of excessive N accumulation by the plant. PE_N is considered the most important component of AE_N, as PE_N accounts for more of the variation in AE_N than RE_N (Dordas, 2008). PE_N generally remains relatively stable during the crop growth period and is little affected by grain yield. A cultivar with a higher RE_N should be selected for increasing AE_N (Novoa and Loomis, 1981; Raun and Johnson, 1999). Many trade-offs exist between the different components of AE_N as they are influenced by several factors.