Six different N fertilization treatments using 6 Indica rice varieties were conducted in Nanchang (28°33′N, 115°57′E), Jiangxi province of south China. Two early rice hybrids, including Zhongjiazao-17 (ZJZ-17) and Tanliangyou-83 (TLY-83) and four late rice hybrids, Tianyouhuazhan (TYHZ), Yueyou-9113 (YY-9113), Xiangyou-186 (XY-186), and Wufengyou-788 (WFY-788) were used. Detailed information about soil characteristics, cropping practices, climatic conditions, and N treatments in field experiments are summarized in Table 1. Experiments were arranged using a completely randomized block design with three replications. The banks between the individual plots were covered with plastic film to prevent fertilizer penetration across treatments. Two seedlings per hill were transplanted manually in all plots and the hill spacing was 0.24 m × 0.14 m and each plot area was 4 m × 5.4 m. The N fertilizer was applied as 60% before transplanting and 40% at tilling for early rice. For late rice it was 50% before transplanting, 30% at tilling and 20% at booting. Urea was used as the N fertilizer. Phosphorus (Ca(H₂PO₄)²) and potassium (KCl) fertilizers were added to the soil with the application rates of 60 kg ha⁻¹ (P₂O₅) and 120 kg ha⁻¹ (K₂O) before transplanting.

Five hills were sampled from each plot at active tillering (AT), mid tillering (MT), stem elongation (SE), panicle initiation (PI), booting (BT), and heading (HD) stages during the vegetative phase for growth analysis. The sampling dates are detailed in Table 1. Fresh samples were separated into leaves and stems. All samples were oven-dried at 105°C for 30 min to stop metabolism and then at 80°C until constant weight was reached to determine shoot biomass (t ha⁻¹). All samples were milled and analyzed for total shoot N concentration by the micro-Kjeldahl method.

The production of shoot biomass of both early and late rice varieties was significantly affected by N application. The increase in shoot biomass followed a gradually increasing trend after transplanting for early and late rice varieties in each experiment with increasing N application rates. However, non-significant differences were observed between N application rates of 225 and 300 kg ha⁻¹ for early rice and late rice. The shoot biomass for early rice ranged from 0.21 to 10.86 t ha⁻¹ (ZJZ-17) and 0.23 to 10.50 t ha⁻¹ (TLY-83) in 2013, and 0.34 to 10.16 t ha⁻¹ (ZJZ-17) and 0.34 to 8.78 t ha⁻¹ (TLY-83) in 2014. For late rice, the shoot biomass of TYHZ ranged from 1.64 to 14.84 t ha⁻¹ in 2013 and 0.83 to 14.5 t ha⁻¹ in 2014. For YY-9113 shoot biomass ranged from 1.13 to 13.9 t ha⁻¹ in 2013 and 1.02 to 14.2 t ha⁻¹ in 2014 (Table 2). Overall, the shoot biomass of early rice varieties (ZJZ-17, TLY-83) was slightly lower than late rice varieties (TYHZ, YY-9113).

The response of plant N concentration to N application rates was generally linear and a higher rate of N application generally resulted in a higher plant N concentrations. A decline in plant N concentrations was observed with increasing shoot biomass. The Figure 1 showed that the plant N concentrations in early rice varieties ranged from 0.85 to 4.11 (ZJZ-17) and 0.85 to 4.00% (TLY-83) in 2013 and 0.70 to 3.51 (ZJZ-17) and 0.73 to 3.68% (TLY-83) in 2014. For late rice varieties, the N concentration ranged from 0.56 to 3.16 (TYHZ) and 0.56 to 2.95% (YY-9113) in 2013 and 0.82 to 3.40 (TYHZ) and 0.84 to 3.18% (YY-9113) in 2014. The early rice varieties showed higher plant N concentrations than the late rice varieties in the early growth stages and similar trends were observed in later stages.

A total of 18 data points of early rice and 24 data points of late rice obtained from experiment 1 and 2 were used to define the N_c dilution curves. The shoot biomass data for developing the N_c dilution curves ranged from 0.8 to 10.65 t ha⁻¹ for early rice and 1.37–14.48 t ha⁻¹ for late rice. The N_c dilution curves of early and late rice varieties established during present study were shown in Figure 2.

Analysis of covariance (ANCOVA) at the 90% confidence interval was used to define the significance of N_c dilution curves of early rice and late rice. The values Sig. of slope and intercept in Table 3 for early rice (ZJZ-17 and TLY-83) are greater than 0.1 showed that there is no statistical difference between the two varieties for early rice and late rice.

The results indicated that growth rate and varieties don't significantly affect N_c for each rice type. However, significant differences were observed between early and late rice. The data of different varieties for each rice type were pooled, and the unified dilution curves for early and late rice were determined (Figure 3).

Where N_cEarly and N_cLate are the N_c concentration in shoot biomass expressed in % DM for early and late rice, respectively, W is the shoot biomass (t ha⁻¹). Parameter a for early rice (3.37) was lower than late rice (3.69). In contrast, parameter b of early rice (0.44) was higher than that of late rice (0.34).

The 95% confidence interval of the plant N concentration was 2.92–4.74 and 2.92–3.78% DM for a shoot biomass of 0.8 and 1.37 t ha⁻¹, for early and late rice, respectively. For shoot biomass 9.72 (early rice) and 14.79 (late rice) t ha⁻¹, the 95% confidence interval of the N_c concentration in aerial tissues was 1.06–1.32 and 1.37–1.64% DM for early and late rice, respectively (Figure 3). Eight data points ranging from 0.51 to 1.2 t ha⁻¹ and 12 data points ranging from 0.26 to 0.80 t ha⁻¹ for early and late rice were used to determine the constant N concentration for early and late rice respectively. The constant N_c concentrations, 3.31%DM (1.04 t ha⁻¹) and 3.15%DM (1.6 t ha⁻¹), respectively for early and late rice were calculated as the mean value between the minimum N concentration of N non-limiting points and the maximum N concentration of N limiting points.

The N_c dilution curves of early and late rice were validated with a dataset obtained from independent experiments (Exp. 3). The data points under different N treatments from the independent experiments were categorized into N limiting or non-N-limiting growth conditions based on significant (P ≤ 0.1) differences in shoot biomass for each sampling date, site, and year. Treatments were considered N-limiting when shoot biomass significantly increased with increasing N supply, while non-N-limiting treatments had no significant increase in biomass with increasing N supply (LSD < 0.1). Data points acquired from N limiting treatments were positioned approximately below the critical curves while those of non-N-limiting treatments were positioned close to or above the critical curves (Figure 4). The N_c dilution curves of early and late rice differentiate well between the N limiting and non-N-limiting conditions.

Due to an obvious variation in plant N concentration for a given shoot biomass, N maximum (N_max) and minimum (N_min) curves were determined for both rice types.

The data points from the highest N treatments represent the maximum N dilution curve (N_max), and the zero N application represents the minimum N dilution curve (N_min) (Figure 4).

Plant N concentration varies between 1.96% below and 1.19% above the N_c dilution curve for early rice, 1.61% below and 1.09% above the N_c dilution curve for late rice.

The results showed significant differences across the N treatments, growing seasons, crop growth stages, and early and late rice. The NNI values ranged from 0.49 to 1.51 and 0.45 to 1.53 for early and late rice, respectively (Figure 5). The NNI values of early and late rice varieties increased under different N_c treatments till 35 days after transplanting and then gradually decreased. The NNI values for early rice under N0 were generally below 1. For N2 to N4 values were generally above 1 and fluctuated near 1 for N1. In contrast, the NNI values for late rice showed differences between the two experiments. NNIs of N3 were closed to 1 in Exp. 1, but in Exp. 2, N2 was the optimal treatment.

The RY was expressed as a function of NNI_int at different growth periods tillering (T), tillering to jointing (T-J), tillering to booting (T-B), and tillering to harvest (T-H) of early and late rice, and the relationships between NNI_int and RY of different periods indicated a close linkage between two parameters for early and late rice. The results showed that NNI_int and RY had a significant linear relationship and RY increased with increasing NNI_int (Figure 6). R² of early and late rice models were between 0.62–0.73 and 0.77–0.89. Slight differences were observed between the slopes and intercepts of the linear functions at different stages and between the early and late rice models.

The robust relationships at different crop growth periods accurately explained the variation in RY both under N-limiting and non-N-limiting growth conditions. Therefore, the present relationships between RY and NNI_int are of practical meaning for in-season estimation of grain yield in double cropping rice varieties.

The results in Table 4 show that models at each growth period offer a good prediction of yield in the independent experiments. The R², RRMSE and accuracy ranged from 0.8451 to 0.9004, 6.896 to 8.811, and 0.8174 to 1.0163%, respectively, for early rice. For late rice R², RRMSE and accuracy ranged from 0.818 to 0.9332, 4.848 to 9.421%, and 1.0131 to 1.251, respectively.

In this study, N concentration decreased with increasing shoot biomass. Results of N concentration and shoot biomass were in consensus with previous reports (Lemaire and Gastal, 1997; Ziadi et al., 2010; Ata-Ul-Karim et al., 2013). This phenomenon of decreasing N is mainly attributed to plant aging and phenology (Lemaire et al., 2007). Sheehy et al. (1998) suggested that the internal cycling of N from old to young developing tissues and an aging root system may play an important role in the rate of N accumulation and N concentration. However, in this study a sudden change in plant N concentration was observed at the jointing stage in 2014 for late rice (Figure 1). This is likely because sampling was conducted 1 week after N addition resulting in increases in N concentration. This is similar to the report of Justes et al. (1994) that N supplying resulted in an increase in the N concentration. The constant N concentration for both early and late rice was likely the result of shoot biomass increasing and the lack of competition for light among isolated plants in initial growth stage as reported by Lemaire and Gastal (1997).

There was a larger variability of N concentration for early rice than that of late rice. However, the lower shoot biomass in early rice was attributed to lower accumulated temperatures and less sunshine in the early rice season (Huang et al., 2013). The difference in N concentration in biomass can be explained by the specific value for the two rice types. The proportionality coefficient k linking N uptake and growth rate using mathematical method (Justes et al., 1994) as follows:

where N_a represents the amount of N uptake in the shoot expressed in kg ha⁻¹ and W is the shoot biomass expressed in t ha⁻¹. The k values for early rice (k_early) and late rice (k_late) calculated according to the Equations (4) and (5) were, k_early = 18.9 W^−0.44 and k_late = 24.4 W^−0.34, respectively. The results showed that k_early < k_late, validated from 0.8 to 10.65 t ha⁻¹ for early rice and from 1.37 to 14.48 t ha⁻¹ for late rice. Coefficient k depends on the shoot biomass accumulation and the N absorption rate is directly related to the growth rate and biomass (Justes et al., 1994), k_early < k_late was validated in whole vegetable growth stage, thus the biomass accumulation of early rice is lower than late rice, and we can infer that N absorption rate of early rice is lower than late rice. Thus, the results from the coefficient k imply that early rice has a lower N accumulation capacity in shoots than late rice for the same aerial biomass.

The variability of the maximum and minimum N dilution curves from the critical curve was 0.77 and 0.52 for early and late rice, respectively. These values are very similar to those for Japonica rice 0.77 (1.8% below and 1.03% above) (Ata-Ul-Karim et al., 2013). Similar variability in plant N concentration for Indica and Japonica rice show that both rice ecotypes have similar capacity for N absorption from minimum to maximum N concentrations in their shoots. The N_max curve represents the maximum capacity of N accumulation in the shoot, while the N_min curve represents the lower limit at which metabolism would cease to function (Justes et al., 1994).

Critical N dilution curves have been previously developed for Japonica and Indica rice ecotypes under different climatic conditions (Sheehy et al., 1998; Ata-Ul-Karim et al., 2013; Huang et al., 2015). The existing N_c dilution curves were compared with the newly developed curves of early and late rice in the present study (Figure 5). The parameter information of these curves is presented in Table 5.

The coefficients of N_c dilution curves for Indica rice showed significant differences between the three Indica rice types. This indicates that Indica rice has different rates of N uptake per unit biomass accumulated, which may be due to temperature difference, day length and crop growth rate. In the early rice growing season (transplanting to flowering), the average temperature was about 6°C lower than that in the late rice growing season. Previous studies have shown that temperature has a significant effect on the absorption of N in rice. Low temperatures result in reduced N uptake and relatively higher temperatures lead to higher N uptake (Shimono et al., 2012), which may explain the lower parameter a and higher parameter b for early rice. The peculiarity of IR72 is that the N concentration at early growth stage is higher than that of late rice but at late growth period the N concentration is close to late rice, then faster N declined with increasing biomass and leading to a higher parameter b.

For Japonica rice dilution curves, parameter b is approximately 0.3 and obviously lower than that of early rice and IR72. Parameter a from a previous study conducted by Huang et al. (2015) was 2.77, nearly 35% less N accumulation at the early growth stage than Indica rice and Japonica rice of Ata-Ul-Karim et al. (2013) (Figure 7). This may due to the fact that typical Indica varieties and Japonica varieties have significant differences in several factors such as photosynthesis (Ji and Jiao, 2001), metabolism (Hu et al., 2014). Previous studies have shown that plant N content in Indica rice is significantly higher than Japonica rice at the maximum tillering stage (Yoshida et al., 2006). However, early rice N content is particular lower than Japonica rice of Ata-Ul-Karim et al. (2013) due to lower temperatures and solar radiation. And Japonica rice in high latitudes has a slower growth rate than Indica rice owing to genotype and environmental conditions resulting in a slower decrease of N uptake per unit biomass accumulation in Japonica (Ata-Ul-Karim et al., 2013).

Nitrogen nutrition index based on the N_c dilution curve is a useful tool to diagnose deficient and non-deficient N nutrition status for crops (Lemaire et al., 2008). The NNI values observed in this study were similar to those obtained for corn (0.3–1.35) (Ziadi et al., 2008), durum wheat (0.25–1.5) (Debaeke et al., 2006) and for rice (0.6–1.05) (Ata-Ul-Karim et al., 2013). NNI values were at a maximum at jointing stage. This is due to N concentration increasing after topdressing at tillering, leading to a high value of NNI, which implies the NNI may cause volatility with fertilizer or change of soil N condition. This result is similar to Lemaire and Gastal (1997).

Nitrogen nutrition index can also be used to assess the effect of N on crop yield (Jeuffroy and Recous, 1999). Higher yields are always associated with high plant N, however, excessive N fertilizer cannot always increase crop yield (Ying et al., 1998). In this study, positive linear correlation between NNI_int and RY was estimated at different growth periods. Minor differences between early and late rice imply the NNI_int are stable at each period. In Table 3, the RRMSE are under 10% and the accuracies are closed to 1.0 illustrating models of each period have good yield prediction accuracy across different years. Several attempts have been made for in-season estimation of RY on the basis of NNI in spring wheat, corn, sunflower, barley and rice (Ziadi et al., 2008, 2010; Debaeke et al., 2012; Yuan et al., 2016; Zhao et al., 2016b; Ata-Ul-Karim et al., 2016a,b, 2017a,b), but the relationships derived in previous studies were based on instantaneous NNI, either using the averaged NNI data at different crop growth stages or at particular crop growth stage, which may lead to an over or underestimation of grain yield in N deficiency or luxury consumption nutrition conditions (Lemaire and Gastal, 1997). According to Lemaire et al. (2008), using NNI_int can avoid the drawback of instantaneous NNI and average values of NNI. Therefore, NNI_int may be more reliable for predicting grain yield.

Nitrogen nutrition index can be a good indicator for diagnosing and directing N fertilization during vegetative growth period to improve crop yields. Moreover, the relationship between NNI_int and RY further indicates the effectiveness of the Nc dilution curve as a diagnostic tool. NNI values can also be integrated into the growth models such as RiceGrow (Tang et al., 2009) and CERES-Rape (Gabrielle et al., 1998). However, real-time determination of shoot biomass and plant N concentration are the obstacle in directly using the NNI as a pre-diagnosis tool. Modern real-time diagnostic techniques such as satellite imagery (Cohen et al., 2010; Huang et al., 2015), chlorophyll meter (Yuan et al., 2016; Zhao et al., 2016a), and hyperspectral imaging based on UAV (Unmanned Aerial Vehicle) (Pölönen et al., 2006) and ground based canopy reflectance (Liu et al., 2015) can be used to determine plant biomass and plant N concentration non-destructively.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.