Rapid assessment of crop nitrogen and stress status – in-field assessment of a hand-held near infrared tool

Take home messages

• New hand-held infrared instruments have shown promising results for use in predicting crop nitrogen (N) content of wheat and barley in the field.
• New spectroscopic tools offer improved non-destructive predictions of N compared to normalised difference vegetation index (NDVI) and other available tools.
• Accurate and robust predictive NIR calibrations enable a single calibration to be applicable to both wheat and barley across contrasting environments.
• Significant potential opportunities exist to incorporate infrared technology to manage N inputs.

Background

The use of infrared (IR) technology for prediction of plant nutrient status is a rapid and cheap means of tissue testing. Recent developments in spectroscopic instrumentation has reduced the size and cost of instruments. It has created greater opportunities to consider the use of hand-held equipment for in-situ tissue testing. The use of such devices in the field could provide real-time information and significant benefits to making agronomic decisions. Spectral reflectance techniques using both visible and near infrared wavelengths are quick and easy to apply in the field. They do not require destructive sampling, and therefore, a larger area can be measured, and many more measurements can be taken over the same area.

This research assessed the ability of a hand-held infrared device to predict crop N content and water-soluble carbohydrate (WSC) concentration in the field.

WSCs provide important reserves as a source of carbon for grain-filling, as grain demand frequently exceeds current assimilation.  This trait can then be a measure for drought, heat, frost and/or disease stress tolerance as reserves of WSC are vital during the grain-filling period in southern Australia. By making WSC easier to measure it could be adopted in the plant breeding process as a phenotyping tool to help select for various stress tolerances. For instance, under terminal drought stress which is common in South Australian environments, WSC have been shown to buffer biomass production, grain yield and harvest index (HI), associated with increased water uptake (WU) and water-use efficiency (WUE). Trait-based breeding for genotypes with greater stem storage and remobilization of WSC may result in improved grain-filling and increased yields. This would potentially assist in fast tracking varieties with improved stress tolerances. This measurement would also be useful to grain growers as a crop diagnostic tool to determine the current, real-time ability of the crop to respond to or buffer against frost, heat, or drought stress events.

Plant N content is also a common measurement used by growers in their N management decision making process. Having a real-time, non-destructive measure of N content would improve the ability of grain growers to target specific grain yield and quality parameters for a given season. For example, plant N will relate to the supply of mineralised and fertiliser N and could help growers manage grain protein, through rapid crop assessments for timely applications of N.

Method

Over two contrasting growing seasons a large amount of data was obtained to develop calibrations for the near-IR (NIR) prediction of N content and WSC in wheat and barley. Over 1500 plant samples were taken and analysed for N content and WSC in conjunction with non-destructive field-based NIR spectroscopy (Hand-held FieldSpec®) and laboratory-based NIR spectroscopy (portable MicroNIRTM) in order to create predictive NIR models. Data was collected from a range of field trials in contrasting environments (Roseworthy, Mintaro and Loxton) that contained current commercial varieties of wheat and barley across multiple growth stages, sowing times and N management strategies. Data was also collected from differing soil type and crop row spacings; from 9-inch to 12-inch, to investigate any associated impact of varying ground cover that may influence the field-based NIR readings. This robust data set ensured there was a sufficient range in N content and WSC data to develop an accurate and predictive model suitable to all end users. Field-based NIR readings were obtained using a hand-held FieldSpec® Spectrometer, which recorded spectral reflectance from 350nm - 1100nm (Figure 1).

Figure 1. The FieldSpec® Handheld 2, remote sensing spectroscopy instrument.

Data analysis and interpretation was a crucial part of this research to link NIR spectral data to actual N content and WSC values. Spectral data was analysed using software, The Unscrambler X® (CAMO). This model development software is extremely powerful and was used for Partial Least Square (PLS) regression analysis in creating NIR spectral predictive models.

Results and discussion

N content

The use of the hand-held FieldSpec® NIR device to provide non-destructive predictions of N content showed potential to provide a new method of crop diagnostics for growers. Initial analysis using only one year of data provided some excellent early outcomes. Predictive model regression using only data from 2017 produced an R-value of 0.94 and predictive error of 0.5%. When data from both years was combined the predictive model produced an R-value of 0.9 (Table 1) and a predictive error of 0.64% and residual predictive deviation (RPD) equal to 2.41 (Table 1 and Figure 2). The RPD is defined as the ratio between the standard deviation of the population (SD) and the standard error in cross validation for the NIR predictions. The RPD was used to evaluate the predictive ability of the calibration models that were developed. The higher the value of the RPD, the greater the probability of the model to predict the chemical composition accurately in samples outside the calibration set. An RPD value greater than 5 (range 5 – 6.4) is considered good for quality control, while an RPD value between 3 and 5 is considered good for screening applications.

Consequently, an RPD of 2.41 (slightly lower than 3) indicates that the model is potentially suitable for screening applications of N content and robust enough to be used to make management decisions. This model includes wheat, barley, varied growth stages, contrasting crop canopy cover, size, and architecture. All data was found to be suitable to be included together. This was a very good result, as it indicates that the one predictive model can be use in all situations and environments rather than having a specific model for each scenario.

Table 1. Summary of all data and model outputs (2016 & 2017) included in the total nitrogen and water-soluble carbohydrates NIR calibration models (MicroNIRTM – lab NIR, ground samples and FieldSpec® – field-based NIR).

Figure 2. PLS regression calibration plots of NIR predicted total N content verses actual total N content using field-based FieldSpec® (350nm-1100nm) on crop canopies of wheat and barley at various growth stages across multiple locations during 2016 and 2017.

The ability to predict N content using the whole spectra of data with such devices as the hand-held FieldSpec® is a significant improvement on the current NDVI sensors available. Instead of using only 2-3 specific wavelengths as used in calculating NDVI, the method used in this study uses every wavelength from 350nm-1100nm. This provides much more information relating to the chemical composition of the crop canopy compared to just how green it is in the visible spectra as per NDVI sensors.

Sample preparation of dry, uniformly ground plant tissue samples (as prepared for wet chemistry) improved the accuracy of the model for total N. The R-value was 0.96 with a predictive error of 0.35% (Table 1 and Figure 3). This method however, is destructive but its prediction accuracy is acceptable for laboratory-based standards and provides a good option for a fast, high through-put method for prepared samples in a laboratory situation. Results from this test confirmed that a single calibration model is applicable across wheat and barley.

Figure 3. PLS regression calibration plots of NIR predicted total N content verses actual total N content using lab-based MicroNIRTM (900nm-1700nm) on dried, ground plant samples of wheat and barley at various growth stages across multiple locations during 2016 and 2017.

WSC content

The WSC predictive model did not have the same level of accuracy as N content. The R-value for WSC using the FieldSpec® hand held device was 0.82 with a predictive error of 39 g/kg (Table 1 and Figure 4). This increased error resulted in a lower RPD of 1.78, which is below the screening threshold. This result would not enable any reliable use of this device to create plant stress related indices in its use as a measure of stored assimilates in the plant. Accuracy can be marginally increased if the data set becomes more specific to a particular environment (or trial) provided there is a sufficient range in data values to create the calibration regression, however it’s not as robust as the predictive model for N content. The current regression may, however, still have potential for use as a selection/screening tool for plant breeders, as it may be sufficient to categorise varieties or treatments into high or low ability to store assimilates.

Figure 4. PLS regression calibration plots of NIR predicted WSC verses actual WSC using field-based FieldSpec (350nm-1100nm) on crop canopies of wheat and barley at various growth stages across multiple locations during 2016 and 2017.

Conclusion

The purpose of this research was to improve the ease of measurement of the plant trait components; WSC and N content. This study investigated the use of portable NIR devices to provide non-destructive, real-time measurements in the field. The hand-held FieldSpec® spectrometer was used to create field-based calibrations for N content and WSC. A laboratory-based MicroNIRTM spectrometer was also used for comparison on laboratory prepared samples.

The N content calibration regression developed for the field-based NIR sensor was less accurate than the corresponding model developed in the laboratory. Despite this however, the calibration model could be used to estimate N from a screening level of accuracy with an error of 0.6%. This level of accuracy could be used to screen a large number of plots or map the variation in N tissue concentration across a paddock.

Although not equivalent to a laboratory diagnostic accuracy level, the NIR predictions could be used to comfortably distinguish nutritional zones within the paddock for improved management of N fertilisers. The ability to have measurements conducted in-field in a matter of seconds, enables many more measurements to be taken, and therefore, provides much more information across the entire paddock rather than targeting a single test in specific zones, as currently practiced with tissue testing.

WSC prediction via the NIR model was much more variable and further research would have to be undertaken to develop a more accurate NIR calibration model. The current NIR predictive ability using the developed model in this project is only capable of providing comparisons of high and low levels of WSC and cannot be used as an accurate diagnostic tool for measuring WSC.

Acknowledgements

The research undertaken as part of this project is made possible by the significant contributions of South Australian growers through both trial cooperation and the support of SAGIT, the author would like to thank them for their continued support. This research was conducted at the University of Adelaide.

Contact details

Michael Zerner
Landmark Pfitzner & Kleinig,
3 Gunn Street, Eudunda SA 5374
0439 802 600
michael.zerner@bigpond.com