Barley (Hordeum vulgare L.) is a highly valued cereal around the world but especially in Australia. In the 2013 world food commodity ranking, barley was the 12th most important and the 4th major cereal right behind wheat, rice, and maize (Food and Agricultural Organization (FAO), 2016). The 2014 world production was estimated at 144 million tons, with Australia ranked 4th for its 9.1 million tons (FAO, 2016). Approximately 60% of Australian barley grains is exported as malt, food, and feed with the largest being feed barley (Barley Australia, 2016). The Australian barley industry can maintain or improve its relative dominance in terms of productivity and environmental sustainability by considering nutrient management strategies that enhance N efficiency. Some of these strategies were grounded on the reflectance properties of crop canopies and their correlations with agronomic quantities such as crop biomass, grain yield, and crop N content. These correlations have been used to develop N fertilizer algorithms for in-season fertilization of cereal crops with the aim of maximizing yield and minimizing N losses.

Various investigations have established the strong relationship between barley canopy reflectance and grain yield. One of the earliest studies on barley yield prediction using canopy reflectance was reported in the early 1980s (Pinter Jr et al., 1981). With a handheld radiometer, daily normalized difference vegetation index (NDVI) values of barley canopy were collected at Feekes growth stage (entire ear out of sheath and flowering commencing) to complete senescence. From these daily NDVI values which is a baseline, the NDVI value of a completely senesced barley canopy was subtracted to derive an index that correlated with barley grain yield. This was an advance over a previously cumbersome method that used cumulated green leaf area index from barley heading until maturity to predict grain yield (Watson et al., 1963). Notwithstanding, because this model relied on reflectance measurement at flowering, its use was restrained in predicting grain yield. In the late 1980s, another study compared reflectance factors of single spectral bands and ratios of spectral bands ranging from 400 to 2300 nm on barley (Kleman and Fagerlund, 1987). Although the Infra-red/red ratio strongly correlated with grain yield in the middle of the season, the correlation was inconsistent afterwards. As an alternative approach to improve this correlation, climatic data were suggested and included as input parameters (Kleman and Fagerlund, 1987). Interestingly, about a decade later, a yield prediction model for winter wheat that included growing degree day (GDD) was developed (Raun et al., 2001).

Another vegetation index reported to correlate with barley grain yield is the transformed chlorophyll absorption in reflectance index (TCARI) (Pettersson and Eckersten, 2007). Reflectance was measured at early stem elongation using a hand-held passive sensor. This indicates that TCARI could be used for early yield prediction in barley. However, this also reveals that NDVI, as a sole vegetation index, has not been extensively investigated on barley at early growth stages. Considering that a number of variable rate technology equipment use NDVI, establishing a relationship between NDVI and barley grain yield at early stages of development under Australian conditions would contribute to enhanced barley N efficiency.

An important application of early canopy sensing correlation with yield and N content is the design of in-season N fertilizer algorithms. The N fertilizer algorithm that seems to stand out among  others  is  the  N fertilizer algorithm or NFA (Lukina et al., 2001). It has been refined by several other contributors (Freeman et al., 2003; Raun et al., 2005) and evaluated in terms of higher N efficiency and financial returns on such cereal crops as wheat, corn and rice. The NFA was reported to improve the N use efficiency (NUE) by at least 15 and 9%, respectively in winter wheat and corn (Raun, 2002; Tubana et al., 2008). The starting point toward devising N fertilization optimization algorithm (NFOA) developed an index that predicted potential yield based on in-season NDVI readings.

In winter wheat, the potential yield could be predicted at earlier stages of development as Feekes growth stages 4 and 5 (or tillering) by combining NDVI values with GDD (Raun et al., 2001). The estimated yield (EY) was obtained by adding NDVI values at Feekes growth stages 4 and 5 and dividing them by the cumulative GDD between the two measurements. Later, another yield prediction index, the in-season estimate of grain yield (INSEY), was developed (Lukina et al., 2001). It appeared that grain yield prediction could be improved if the estimated yield was obtained by using the cumulative GDD from sowing to sensing as a divisor. Sensing could be done from Feekes growth stages 4 to 6 and there would not be any significant improvement in the yield prediction regardless of the number of sensing performed within that window. Moreover, yield prediction would be less affected regardless of the NDVI being used solely or with a divisor (the cumulative GDD from sowing to sensing).

Once the yield prediction index or INSEY was established, different approaches for determining N requirements were proposed. For instance, a 5-step approach that required the prediction of the potential grain yield (using the INSEY index), grain N content, grain N uptake and early-season plant N uptake was proposed (Lukina et al., 2001); whereas a 7-step approach included, on top of the foregoing, the calculation of the predicted yield using the predicted potential grain yield, the response index and the forage N uptake (Raun et al., 2002). To refine the algorithm, consideration was equally given to variations in plant-stand densities (Raun et al., 2005; Teal et al., 2006). These approaches can be acted on and adapted in devising customized N fertilizer recommendations provided all inputs for building an algorithm are supplied.

The overall objective of this study was to improve barley N use efficiency through canopy sensing under Australia biophysical conditions by developing two algorithms for in-season N fertilizer prescriptions for barley grain and forage production. Upon validation, these algorithms may be used for N fertilizer recommendations on barley in Queensland (Australia). Possible outcomes of using the algorithm may include input cost reductions, limited environmental contamination, higher quality grains and forage and better financial returns. 

Experimental site and design

The experiment was conducted at the research facility of the University of Queensland located in Gatton Campus during the winter barley growing season, from June to October, 2016. The previous crop for the trial area was forage sorghum with no added fertilization in order to substantially deplete the soil N level. Soil test analyses were undertaken prior to sowing and the results are presented in Table 1. Based on these results, some nutrient deficiencies were corrected. Climatic conditions prevailing throughout the experiment were gathered by a weather station located within the premises of the research facilities. Downloaded from the Australian Bureau of Meteorology website (BOM, 2016), monthly averages of these climatic conditions are summarized in Table 2.

In-season estimate of yield

Number of days from sowing to sensing where GDD is larger than zero. All over the growing season daily temperatures were larger than the GDD.

Response index

where RI: response index; NDVI_rich: NDVI measured in a N rich strip, and N is non-limiting; NDVI_sowing: NDVI measured in the plot with the sowing N rate. The sowing N rate can range from zero to the level where the N is non-limiting.

Normalized difference vegetation index

where NDVI: normalized difference vegetation index; R: canopy reflectance in the red band (660±25 nm); NIR: canopy reflectance in the near infra-red band (780±25 nm).

Parametric statistical analyses were performed using the R version 3.3.1 released 21-06-2016. Three major steps were followed: (1) check the normality of response variables, (2) test the significance of differences among treatment means using the least significant difference (LSD) test, and (3) test the significance of correlation coeficients, regression coeficients and the regression models. All variables were submited for analysis of variance (ANOVA) to assess the statistical significance of treatment effects across the four treatments and then means were compared for significant differences. Simple linear regression analyses based on ordinary least square (OLS) estimation were performed to describe the relationship between the independent variables (NDVI readings) and dependent variables (yield and N contents). The level of significance throughout these statistical analyses was set at α = 0.05.

Barley response to N rates

All agronomic variables measured responded positively to different N rates (Table 3). With the exception of the two highest N rates where no significant differences were noticed, FY, GY, FNUP_i, FNUP_h and GNUP significantly increased with increasing N rates, thus delivering enough variability in data to make possible the derivation of N fertilizer algorithms.

Canopy reflectance and N rates

Barley canopy reflectance increased with increasing N levels. With the exception of 5 and 6 WAS where no significant differences were detected between treatments, higher N rates induced higher NDVIs (Figure 1). N fertilization fosters biomass production and enhances leaf greenness. Being an index specific for chlorophyll and vegetation discrimination, NDVI is sensitive to both green and dense canopies (Peñuelas et al., 1997). In canopy reflectance of barley at various N rates under two irrigation regimes, Kleman and Fagerlund (1987) equally noticed that reflectance increased with increasing N rates as a result of biomass and leaf pigment accumulation. Furthermore, when NDVI was regressed against N rates, strong coefficients of determination were observed at all growth stages (Figure 2) except for weeks 5 to 7, therefore reinforcing the assumption that barley NDVIs can be used for indirect assessment of pre-plant N levels between 7 and 15 WAS.

known NDVI weakness of reaching saturation prematurely as compared to other vegetation indices (Figure 1). NDVI saturation usually occurs when at larger leaf area indices rather than increasing linearly, NDVI readings asymptotically reach a constant value (Haboudane et al., 2004). Because NDVI saturation usually occurs under dense canopies (Gu et al., 2013), this suggests that the population density and the two highest N treatments tested in the current experiment induced NDVI near-saturation. Thus, at N rates exceeding 150 kg N ha^-1 and sowing densities above 600,000 plants per hectare, NDVI is expected to reach saturation thereby rendering it a poor estimator of barley biomass and N content.

Over the growing season, NDVI exhibited a parabolic-shape curve irrespective of N rates. For instance, at 150 kg N ha^-1, NDVI steadily increased from 0.35 at one WAS to peak of 0.81 at eight WAS, then decreased to a lower value (0.45) at the ripening stage (GS9). Apart from the declining phase of the NDVI curve for the past week 9, this pattern resembles the N uptake curve and the biomass production curve of barley over time (Lemaire et al., 2008; Whitmore, 1988).  Indeed, under steady N supplies, most cereal crops expand their canopies, accumulate biomass and store N in leaves in the form of Rubisco or Ribulose-1,5-bisphosphate carboxylase/oxygenase, a compound produced during photosynthesis (Millard, 1988). Once maximum leaf area and biomass are attained, the N uptake is restricted to covering deficits resulting from grain filling and limited remobilization from senesced leaves (Jeuffroy and Bouchard, 1999). Thus at anthesis and post-anthesis stages (GS5 or week 11 and beyond), leaf biomass, leaf N content and forage N uptake decline until complete senescence of the canopy occurs. Being an index sensitive to chlorophyll content, biomass and leaf area, NDVI mirrored these physiological processes over the growing season.

Canopy reflectance, yield and N uptake

Barley NDVI correlated well with both N uptake and yield. When NDVI was regressed against FNUP_i at eight WAS, a fairly strong determination coefficient was noticed (R² = 0.63, p < 0.001) confirming the assumption of a linear relationship. The correlation was equally positive for both components of N uptake, namely dry biomass (R² = 0.58, p<0.001) and N concentration (R² = 0.74, p<0.001). Earlier studies on wheat ascertained that NDVI not only correlated with total N uptake but equally with N content (Sembiring et al., 2000; Wright et al., 2004). Just as NDVI is used as estimator of in-season N uptake on wheat the same is achievable on barley. 

It was equally noted that NDVI at GS3 displayed a positive linear correlation with FNUP_h (R² = 0.65, p<0.001) and GNUP (R² = 0.84, p<0.001). Assessments over the growing season of determination coefficient fluctuation of NDVI with FNUP_h and GNUP revealed that from 7 to 15 WAS NDVI could provide an accurate estimation of FNUP_h and GNUP as these coefficients were at least equal to 0.50 (Figure 3). Additionally, upon correlating NDVI with FY and GY, determination coefficients were almost always above 0.50 (Figure 4). It seems that the fairly high correlations of NDVI with end season N uptakes in grain, forage and yields at nearly all sensed growth stages could be due to favorable climatic and biotic conditions. Nevertheless, in a practical sense, only stages in which split N application is possible will be of interest for algorithm development. Therefore, barley NDVI at 8 WAS becomes reliable in predicting barley FNUP_i, FY, FNUP_h, GY and GNUP, and as such, can be used in developing an in-season fertilizer algorithm, especially because side dressing can be performed at that stage.

The low efficiency of N fertilizer threatens farmers’ financial returns but most importantly the environment. Among the existing strategies being developed to address the matter, crop canopy sensing holds an advantageous position, and is being considered a promising tool in precision agriculture. The current study developed two distinct algorithms: one for barley forage production and another for grain production in Queensland using the handheld GreenSeeker sensor. The sensor measured barley canopy reflectance over the growing season and output NDVI values. These values were correlated with yields (grain and forage) and with N uptake (in-season and at harvest). The power of these correlations was high enough to justify the design of in-season N fertilizer algorithms. The developed algorithms were expected to safeguard natural ecosystems through minimization of unused N but also to improve financial returns of barley growers through reductions of N related costs. 

The authors have not declared any conflict of interests.