Estimating the nitrogen status of crops using a digital camera

Abstract

Spectral reflectance of cereal canopies measured with a digital camera correlated closely with N status during the vegetative and early stem elongation phases. Photos of the crops taken at a height of about 1 m were processed to indicate the canopy cover (CC) on a 0–1 scale representing the proportion of pixels that met a criterion of greenness, $(1+L)\frac{Green-\mathrm{Re}d}{Green+\mathrm{Re}d+L}>0\text{,}$ where Green and Red are reflectance values of the green and red bands of the camera's colour detector and L is a soil base line. The values of CC measured at one site were closely correlated with values of leaf area index (LAI), above-ground biomass and N content for two wheat cultivars with different growth habits. The functions obtained by calibration were validated on cereal crops at other sites and with different cultivars. The estimate of N content was more precise than the other crop properties, with a root mean square error of 16% of the range of values. Similar values of CC were obtained with three different brands of digital cameras. The CC values were closely correlated (R² > 0.86) with the normalised difference vegetation indices (NDVI) from the commercial GreenSeeker^® and Yara N-Sensor, which detect the red and near infra-red reflectances of the crop canopy. We conclude that estimates obtained from a digital camera or specialised sensors provide equivalent information about crop-N supply to a crop. However information is needed about crop-N demand as well as N supply in order to estimate fertilizer-N requirement. The use of a handheld computer with an inbuilt digital camera offers a method of combining N supply and demand data to estimate fertilizer-N requirement.

Introduction

The increasing cost of nitrogen (N) fertilizer and environmental degradation caused by excess N are reasons for farmers to increase nitrogen-use efficiency (NUE), defined as the additional crop yield per unit of applied N. Information about the N status of soil and crop is important for farmers and advisers in making decisions about N fertilizer application. When N fertilizer is applied at or before the time of sowing, the only source of information is from soil tests. For many wheat crops some or all of the N fertilizer is applied during the tillering or stem elongation phases, when the responses of yield and grain protein per unit of applied N are comparable with responses to N applied at sowing (Angus and Fischer, 1991). NUE usually decreases with fertilizer applied later than these stages.

Measuring the N status of above-ground crop tissue is cheaper and easier than soil tests. Plant tests are based on the concentration or amount of above-ground N (Batten et al., 1991) or nitrate (Papastylianou et al., 1982) in all or part of the shoots, chlorophyll content (Peng et al., 1996), shoot density (Angus et al., 1989) and colour charts (Witt et al., 2004). There is a critical difference between methods based on N concentration and those based on N amount. Concentration obviously depends on the amount of N as well as the amount of biomass, so N taken up early in crop development leads to increased growth but lower N concentration than a similar amount of N taken up later. Accordingly, the most successful tests of crop-N status are those that measure, or are correlated with, the mass of N per unit ground area (Angus et al., 1989, Flowers et al., 2003). These tests rely on manual sampling and some require laboratory procedures, so there is normally a delay before the results are available to the farmer. Data on plant N status may be more reliable than soil data because roots explore more soil than is normally sampled by coring (Angus et al., 1989, Webster and Oliver, 2007), particularly if the root system is well developed. For this reason measurements on plants with >3 leaves are generally more reliable than measurements on younger plants. Because of the need to measure advanced plants, there is a compromise between the delay needed for reliable plant tests and application early enough to ensure high NUE. There are also inevitable delays during transport and distribution of N fertilizer, so it is important for the results from plant tests to be available quickly.

Crop spectral reflectance is well correlated with crop growth, so has the potential to provide information about N status (Raun et al., 2008). The first generation of multispectral sensors measured four relatively wide bands of the visible and infra-red spectra. Ratios of the reflectance in these bands were shown to be related to crop productivity, apparently because they distinguish between vegetation and bare soil (Rouse et al., 1974). We are unaware of published evidence that they are suitable for estimating productivity after full canopy cover (CC), defined as the proportion of land area covered by foliage when viewed from above. The more recent hyperspectral images containing hundreds of narrow bands have the potential to detect N concentration of the foliage as well as CC (Haboudane et al., 2004). Hyperspectral information may be useful after full CC but the information about N concentration needs to be combined with information about crop biomass in order to indicate the mass of N per unit area.

Airborne and satellite-mounted multispectral sensors have the potential to deliver information for N management over a large region. Limitations of these platforms for commercial use on individual fields are the high cost of images from aircraft, the infrequency of satellite overpasses, risk of images being obscured by clouds and delays between image capture and availability of usable data. Proximal sensors mounted on tractors or implements provide similar information to sensors on aircraft and satellites and overcome some of the logistical limitations of remote sensing. The GreenSeeker^® (Raun et al., 2001; http://www.ntechindustries.com/RT100-handheld.html), measures reflectance in red and NIR regions of the spectrum (Fig. 1). It is mounted either on a wand for hand operation, or as multiple units mounted on a toolbar to sense reflectance on-the-go and actuate individual fertilizer-spreading units. The Yara N-Sensor (Reusch et al., 2002, Mayfield and Trengove, 2009; http://www.yara.com/en/sustaining_growth/business_impact/responsible_fertiliser/nitrogen/n_sensor.html) estimates a biomass index from algorithms based on red and near infra-red bands which can differ between applications (S. Reusch, pers. commun., 2008) so are not shown in Fig. 1. When mounted at a height of ∼3 m it senses the spectral reflectance of ∼4 m² of crop in four areas beside the tractor or implement on which it is mounted. The N-sensor software compares the values to a previously set reference and provides an on-the-go signal to a variable-rate controller to adjust the rate of N fertilizer applied across the full swath of a distributor to compensate for the low N areas.

Gitelson et al. (2002) suggested that the intensity of green and red reflectance could be used to as an alternative to red and infra-red reflectance to measure canopy properties. Domestic digital cameras measure the intensity of reflectance in the red, green and blue bands so they have promise as an inexpensive alternative to multispectral sensors for measuring crop-N status. Leaf reflectance is greater in the green than in the red parts of the spectrum (Fig. 1) so we investigated the ratio of the intensity of green to red reflectance in relation to canopy properties.

Information about crop-N status, however obtained, is not necessarily useful in determining the optimum N fertilizer for crops. The N status of a young crop provides an estimate of N supply from the soil and from any previously applied fertilizer. The optimum N fertilizer strategy depends on the balance of N supply and demand (Angus, 2001). Crop demand for N depends on the potential yield and protein content, which are affected by environmental, management and genetic factors such as water status, root disease and yield potential. Budgets of N supply and demand can be used to prescribe N fertilizer for a single location and new methods are needed to match N supply and demand across different zones in a field. This paper explores the possibility of acquiring inexpensive data about N supply from a digital camera combined with a handheld computer as a precursor for calculating optimum N fertilizer from the balance between N supply and crop-N demand.

Images from domestic cameras, which consist of millions of pixels, are slower to process than the single record from the Greenseeker^® and N-Sensor. Processing images from a domestic camera with current software is too slow for on-the-go assessment of crop status. Faster software will be needed for handheld computers with inbuilt cameras to be able to process image data fast enough for such applications. On the other hand, processing data from some specialised sensing equipment can be time consuming, so the comparison of all the images was made with the simplest settings of the equipment. The similarity of images obtained from different cameras is unknown, so an additional goal is to compare the results of several models of camera.