Remote sensing potential for investigation of maize production : review of literature

Maize is considered globally as the most important agricultural grain which is staple food for many humans and feed to livestock. There is need to enhance productivity through management tools to meet the demand for growing populations. Farmers are likely to be interested in technologies that are beneficial to their operations. A technology that could assist farmers to produce staple foods e.g. maize more efficiently is remote sensing. The paper focuses on reviewing published research that deals with application of remote sensing in maize farming particularly the spectral characteristics of maize leaves, classification and mapping. It further surveys the application of remote sensing in detecting foliar nitrogen deficiency, water stress and disease infestations in maize. Remote sensing can be considered as a fast, non-destructive and relatively cost-effective method to study biophysical and biochemical parameters of vegetation across vast spatial areas. However, selection of appropriate sensors with special attention on their spatial and spectral resolutions as well as processing techniques will validate a success story for remote sensing application in maize production.


Background
Maize has been considered globally as the most important agricultural grain which is staple food in many countries and feed to livestock.It is estimated that by 2050, the demand for maize in developing countries will double, and by 2025 maize will have become the crop with the greatest production globally (FARA, 2009).In 2006 the Abuja Summit on Food Security in Africa identified maize, among other crops, as a strategic commodity for achieving food security and poverty reduction.There was a call to promote maize production on the continent to achieve self-sufficiency by 2015 (AUC, 2006).
Maize production at macro level is limited by climate and soil.The potential areas maize can, therefore, be cultivated are geographically specific to these environmental conditions.At micro level, the determinant or stress factors to maize production would include among other factors water and nutrients deficiencies (nitrogen, phosphorus, potassium, etc.), insect pests, and diseases (Zhao et al., 2003).The proper functioning, growth and eventually yield output of the crop is influenced by these factors.Detection of stress levels to which maize production is subjected is therefore essential for assessing the effects on yield, taking action to mitigate these effects and enhancing production.
Precision agriculture is based on intensive sources of information and attempts to address the site-specific needs with spatially variable application.This involves close monitoring and controlling many aspects of crop production that should aid in identifying proper targets and needs of crops for applying locally varying doses of chemicals.For instance, Mondal et al. (2011) conducted a research to monitor plant nutrient and moisture needs, soil conditions, and plant health (including identification of disease infestation).
Effective crop planning and management requires informed and sound decisions drawn from knowledge about the crops in the field.In order to improve agricultural management, scientist are applying information technology (IT) and satellite-based technology (e.g.global positioning system, remote sensing etc.) to identify, analyze, monitor and manage the spatio-temporal variability of agronomic parameters (e.g.nutrients, diseases, water, etc.) within crop fields.This aids in timely applications of only the required amount of inputs to optimize profitability, sustainability with a minimal impact on the environment (Mondal et al., 2011).There is therefore the assurance of proper resource utilization and management to enhance crop productivity.
The purpose of this paper was to review published research in maize production using remote sensing as major research tool.The review seeks to investigate the potential of using remotely sensed data and analytical techniques to enhance productivity of maize.This is achieved through an understanding of the spectral characteristics of maize leaves for varietal separation, classification and mapping.The ability of these spectral characteristics can be used to monitor the nutrient status and health condition of the maize leaves.The paper concludes with a summary and some recommendations for the application of remote sensing in the production of maize.

Spectral characteristics of maize leaves, classification and mapping
When energy strikes on a surface material, it is either absorbed or reflected back through the electromagnetic spectrum.The visible (400-700nm wavelength) and near infrared (NIR) (700-2500nm wavelength) region of the electromagnetic spectrum is the region at which most agricultural studies carry out measurements.This is because the spectral region includes wavelengths which are sensitive to physiological and biological functions of crops (Lillesand et al., 2008).The spectral characteristics of healthy vegetation are distinctive with low reflectance in blue, high in green, very low in red and very high in the NIR (Chen et al., 2010;Genc et al., 2013).
There is a large difference in the spectral characteristics between soil and crop, especially at the 'red edge' which is the point where the electromagnetic spectrum changes from visible to NIR (wavelength of approximately 700nm) (Figure 1).This region is used to detect biochemical and biophysical parameters in crops thereby being useful in vegetation studies.The principle is that the majority of the red light is absorbed by the chlorophyll in the canopy while a high proportion of the NIR light is reflected (Ramoelo et al., 2012).The canopy greenness increases, either due to increasing crop density or chlorophyll content.Canopy greenness is therefore related to the percentage of red reflectance absorbed and the percentage NIR reflectance reflected (Lillesand et al., 2008).Therefore, the reflectance spectral techniques are very suitable for providing relevant information on both crop foliar and canopy which could be related to nutrient status and stress factors on the crops (Scotford & Miller, 2005;Ramoelo et al., 2012).
Measurements at the red-edge bands has made possible the estimation of foliar nutrients and chlorophyll concentrations in differing vegetation types at varying growth stages (Huang et al., 2004;Cho and Skidmore, 2006;Darvishzadeh et al., 2008;Mokhele & Ahmed, 2010).Therefore, remote sensing (both multispectral and hyperspectral) can therefore provide an effective means for fast and non-destructive estimation of leaf nitrogen and water status in crop plants through complimentary tools such as regression models (Yao et al., 2010).
Figure 1: Spectral reflectance curves for soil and crop (green vegetation) according to Scotford & Miller, 2005.Spectral unmixing techniques can be used to quantify crop canopy cover within each pixel of an image and have the potential for mapping the variation in crop yield.Each image pixel contains a spectrum of reflectance values for all the wavebands measured.
These spectra can be regarded as the signatures of surface components such as plants or soil, provided that components, referred to as endmembers, covers the whole pixel.Spectra from mixed pixels can be analyzed with linear spectral unmixing, which models each spectrum in a pixel as a linear combination of a finite number of spectrally pure spectra of the endmembers in the image, weighted by their fractional abundances (Yang et al., 2007;2010).Spectral unmixing is an alternative to soft classification for sub-pixel analysis (Lu and Weng, 2004).This is usually very essential in crop identification and classification studies.The classification is performed under either supervised techniques which include maximum likelihood, minimum distance, parallelepiped etc or unsupervised technique.
Adding to traditional unsupervised and supervised classification methods are advanced techniques such as artificial neural networks (ANN), support vector machines (SVM), decision trees (DT) and image segmentation which have been used to classify remote sensing data (Lu & Weng, 2007;Mathur & Foody, 2008).However, these classifiers remain to be evaluated using different types of remote sensing data from diverse crop growing environments.There is also the field-based crop identification which entails field boundary information and results in higher classification accuracy (De Wit & Clevers, 2004).
Remote sensing has played a significant role in crop classification for various purposes.Effective crop classification requires an understanding of the spectral characteristic (foliar and canopy levels) of the particular crop.In order to understand for instance the maize canopy spectral characteristics, a field investigation is targeted when the plant canopy is covered which is usually during six to eight weeks after planting.The light reflectances acquired during such growth stages could apply in differentiation of varieties; classification and mapping of maize varieties.For instance, Yang et al. (2011) used maximum likelihood and SVM classification techniques on SPOT 5 imagery to identify crop types and to estimate crop areas.Different band ratios of multispectral or hyperspectral data and classifications schemes have been applied, depending on geographic area, crop diversity, field size, crop phenology and soil condition (Nellis et al., 2009).
To extract vegetation species from hyperspectral imagery, a set of signature libraries of vegetation are usually required (Xavier et al., 2006).For certain applications, the vegetation libraries for particular vegetation species might be already available.However, for most cases, the spectral signature library is established from data collected with a spectrometer.As such, vegetation mapping using hyperspectral imagery must be well designed to collect synchronous field data for creating imagery signatures (Melgani, 2004).
In most agricultural studies, spectral reflectance values of at least two wavelength bands (on either sides of the 'red edge') are measured to enable the calculation of a ratio.
These ratios are known as vegetation indices and many have been developed over the years.These indices are usually correlated against field observations of nutrient stress measured at foliar or canopy level.The number of wavelength bands measured will determine the complexity of the data analysis; wherein a small number of bands will translate to a simple data analysis.This could be illustrated in a scenario where in situ measured reflectance values of the red and NIR wavelengths are used to calculate normalized difference vegetation index (NDVI).
However this only holds true until canopy closure when the crop has a leaf area index (LAI) of up to three where LAI is defined as the ratio between total leaf area, one side only, per unit area of ground (Scotford & Miller, 2005).Mirik et al. (2012) also described spectral vegetation indices as mathematical expressions that involve reflectance values from different parts of the electromagnetic spectrum.These expressions are aimed at optimizing information and normalizing measurements made across different environmental conditions.Shanahan et al. (2003) proposed a study evaluating the use of two indices (NDVI and green NDVI (GNDVI)) on a large plot scale.The experiment was conducted on four varieties of irrigated corn treated with five differing levels of nitrogen.Remote measurements were taken with active sensors emitting light in four bands: blue (460nm), green (555nm), red (680nm), and NIR (800nm).The authors concluded that differences in NDVI was significantly impacted by nitrogen and sampling date.Also, increased nitrogen was correlated to increased chlorophyll content but not on a large scale.However, Strachan et al. (2002) recommended that canopy reflectance at red edge position can explain 81% of maize leaf nitrogen variability.

Nitrogen deficiencies in maize
Nitrogen (N) is a biochemical nutrient essential for plant growth.It forms part of many structural, metabolic and genetic compounds.Nitrogen is a critical building block of Chlorophyll which is essential for the process of photosynthesis.Availability of N in the soil usually related to plant growth, photosynthetic capacity and stress (Ustin et al., 1998;Yao et al., 2010).Nitrogen deficiency in a plant will have symptoms on the lower, older leaves first before progressing upward to younger leaves if the condition is not corrected (Sawyer, 2004).
Nitrogen deficiencies interfere with protein synthesis and growth in crops such as maize (Bruns and Abel 2005).Numerous studies have been conducted to proof the direct effect of N to yield levels of maize (Lindquist et al., (2007); Shapiro and Wortmann, 2006;Singh et al., 2003;Abouziena et al., 2007;Savabi et al., 2013).Nitrogen deficiency-induced effects on kernel number could be related to photosynthesis or plant growth at flowering (Andrade et al., 2002).Gitelson et al. (2005) developed a model, based upon field measurements made by means of a hyperspectral radiometer, for nondestructive estimation of chlorophyll in maize and soybean canopies.
Nitrogen deficiency could be detected earlier in crops when visual symptoms of deficiency are less evident with the use of remote sensors (Jackson et al., 1981).This is achieved through its various compounds providing an effective means for monitoring growth status and physiological parameters in crop plants.Its presence in chlorophyll and other cellular structures influences information on spectral reflectance (Yao et al., 2010).
The measurement of chlorophyll content is also important with regard to N management.Hence, a number of studies have utilised remote sensing techniques to determine the status of nitrogen and other essential nutrients in field crops such a maize and wheat as illustrated through this review.Zhao et al. (2003) induced differing levels of nitrogen stress on corn and measured growth parameters, chlorophyll concentration, photosynthetic rates, and reflectance.The study demonstrated that reductions in leaf nitrogen concentrations are greater in plants suffering from inadequate soil N availability.Reduced nitrogen concentrations were correlated with lower rates of stem elongation and leaf area.The authors in 42 days after crop emergence noticed a 60% reduction in chlorophyll a, which caused an increased reflectance near 550 and 710nm.Therefore, reduced chlorophyll concentrations as a result of stress translate in decreased light absorbency and increased reflectance in these wavelength regions.
Remotely sensed imagery can provide valuable information about in-field N variability in maize as well as variability at canopy level still using the relationship with N content.Maize leaf reflectance (near 550nm wavelength) has a good relationship with leaf N content (Osborne et al., 2002).Martin et al. (2007) found that NDVI increased with maize growth stage during the crop life cycle and a linear relationship with grain yield is best at the V7-V9 maize growth stages.Solari et al. (2008) investigated the potential use of active sensors at a field scale in determining N status in corn.Irrigated plots with uniform soils and fertilization, excluding N, were established at the initial stage and later differing rates and timing of nitrogen applications were administered in order to induce variable growth patterns.The authors discovered that the NDVI was sensitive to differences in N, hybrid, and growth stage.There exist a strong linear relationship between leaf chlorophyll concentration and leaf N concentration, where the greater leaf area and green plant biomass levels result in higher reflectance and higher subsequent NDVI values (Inman et al., 2007).This implies that these variables (leaf area and plant biomass) are directly related to the N content of the plant; hence higher NDVI values indicate higher plant N content (Shaver et al., 2011).

Water stress in maize
The saying that 'water is life' denotes for both flora and fauna.Water is a key determinant in the production of crops with maize inclusive.Accurate water content estimation is required to make decisions on irrigation and also crop yield estimations in agricultural studies (Peñuelas et al., 1993).The water content/status of a plant can be measured from root, stem and leaf material or the whole canopy.Leaf analyses are however, the most important organ for evaluating nutrient and water status of plants in comparison to other tissue types (Suo et al., 2010).The leaf is also mostly responsible for photosynthesis, an essential physiological process in plants.Hence, the health and nutrient status with water status inclusive of the plants can be evaluated from the leaves.
Considerable technological developments have taken place over the years to determine water stress using remote sensing.The basis of detecting water stress relates to the differences in reflectance properties of plants under different water stress levels at certain wavelengths in the NIR portion of the electromagnetic spectrum (Genc et al., 2013).Two As a result of the absorption by oxygen-hydrogen (O-H) bonds in water, its absorption features could be found at approximately 760nm, 970nm, 1200nm, 1450nm, and 1950nm (Li, 2006).The first derivatives of reflectance associated with the slopes of the lines near water-absorption wavelength bands 900nm and 970nm correlates well with leaf water content (Danson et al., 1992).Studies have shown that reflectance spectra of green vegetation in the 900-2500nm region are associated with liquid water absorption and are also weakly affected by other biochemical components absorption.It is possible to investigate the effect of water on nutrient stress (nitrogen) through discrimination based on the visible and NIR reflectance of maize leaves (Christensen et al., 2005).Prior knowledge of water status of plants can increase the ability to discriminate nutrient stress significantly.
The water band index is derived from the ratio of reflectance measured at 900nm and 970nm (Peñuelas et al., 1993).This spectral index has been correlated with ground-based measurements of plant water content at both the leaf and canopy scales.It is, however, more sensitive to leaf water content than the water content of the whole plant.This is advantageous in agricultural applications where leaf water content changes more noticeably in response to drought conditions than the water content of the entire plant foliage (Champagne et al., 2003).Leaf water content can be measured with the spectrometer to determine available water to the plants.Genc et al. (2013) conducted an experiment on corn plants where reflectance measurements were made at the red and NIR portions before and after irrigation application.The results confirmed a decrease and increase in reflectance spectral at respective regions as the water level at field capacity increases.However, when the authors compared results at all four water levels, the reflectance spectra indicated that water stressed corn plants absorbed less light in the visible and more light in the NIR regions of the spectrum than unstressed plants.
A recent significant breakthrough in passive optical remote sensing has been the development of hyperspectral sensors on satellite platforms (such as EO-1 Hyperion) that provide continuous narrow bands and high resolution in the visible and infrared spectral region.Compared with multispectral imagery that only has a dozen of spectral bands, hyperspectral imagery includes hundreds of spectral bands.Hyperspectral sensors are well suited for vegetation studies as reflectance/absorption spectral signatures from individual species as well as more complex mixed-pixel communities can be better differentiated from the much wider spectral bands of its imagery (Yang et al., 2010).
However, the interpretation of this hyperspectral data can be complicated by the inter-relationships between wavelength variables but many statistical techniques have been utilised to analyse such data.For example, neural networks; partial least-squares analysis; fuzzy logic; principle component analysis and stepwise multiple linear regression have all been used (Xie et al., 2008).
Maize growth rate can be used as a good predictor of kernel number when nitrogen and water supply are considered as variables.This relationship was proven by modifying maize growth rate by N and/or water supply to that similarly obtained when plant growth was changed by variations in plant density and incident radiation.The effect of reducing N availability was similar to the effect of reducing water availability (Andrade et al., 2002).
Thus, the effect of water deficiencies, nitrogen stress, plant density, and incident radiation on maize kernel set can be predicted through a relationship between growth rate and kernel number.This is explained by two aspects: the correlation between growth rate at flowering to growth of reproductive structures, and also that early seed development and kernel set in maize is dependent on a continued supply of assimilates from concurrent photosynthesis (Zinselmeier et al., 2000).
A few water indices have been developed to study crop stress which include the water band index (WBI) (Peñuelas et al., 1993), shortwave infrared water stress index (SIWSI) (Fensholt & Sandholt, 2003) and normalized difference water index (NDWI) (Gao, 1995;Serrano et al., 2000).Recent studies have focused on combining the blue, green, red with blue and the NIR wavelengths in indices to estimate vegetation water content (Table 1) (Genc et al., 2013).Therefore, the use of remote sensing is particularly and practically suitable for assessing water stress and implementing appropriate management strategies because it presents unique advantages of repeatability, accuracy and cost-effectiveness over ground-based survey methodologies for water stress detection.
Table 1: Spectral indices and some wavelength bands used to detect water stress in maize

Disease detection
Detection and identification of plant diseases and planning effective control measures are important to sustain crop production.Maize production is challenged by disease attacks ranging from insect to fungi.Plant diseases do not only affect yields but also increase the cost of production through treatment procedures on the crops.Some of the common diseases on the maize crop include but are not limited to grey leaf spot, common rust, northern corn leaf blight, phaeosphaeria leaf spot, maize streak disease, stem rot diseases and others.
When crops are infested by a disease, the biochemical constituent of the plant changes and so would be the spectral signatures.This is the reason why remote sensing could be used to monitor the infestation of diseases in field crops.Applications of remote sensing in field crops, for rapid detection of pest damage or disease, also include the use of handheld optical devices (Sudbrink et al., 2003;Moshou et al., 2004;Xu et al., 2007;Mirik et al., 2007) and airborne sensors (Sudbrink et al., 2003).
The health of the leaf can also be monitored wherein changes in leaf chlorophyll content provide an indicator of maximum photosynthetic capacity, leaf development and/or stress (Si et al., 2012).Assessment of photosynthetic functioning is one of the most important bases for the diagnosis and prediction of plant growth and subsequent yield estimation. Williams et al. (2012) used the potential of the hyperspectral NIR imaging to evaluate fungal contamination in maize kernels.Using the principal component analysis (PCA), bad pixels as well as shading of acquired absorbance images were removed before further analysis.They concluded that the methodologies used were able to detect infection, the degree of infection and increase of infection over time.Therefore, remote sensing could assist in early detection of disease infestation in maize.
2012) that could be utilised in estimating crop productivity.For instance, the LAI is used to quantify canopy structure, crop growth and hence predict primary productivity.
High correlations are found between vegetation indices and green biomass in studies done at field level (Groten, 1993).This is related to the crop type and yields but requires (GNDVI), developed by Gitelson et al. (1996), derived from images acquired during midgrain filling were the most highly correlated with grain yield (Table 1).Therefore GNDVI could be used to produce relative yield maps depicting spatial variability in fields, offering a potentially attractive alternative to use of a combine yield monitor.
While most studies on yield prediction with remotely sensed data apply multispectral data due to its availability, hyperspectral images could also be utilised.For instance, Uno et al. (2005) statistically analysed hyperspectral images with an artificial neural network (ANN) and vegetation indices to develop models to predict yields for maize in Canada.
After using the PCA to reduce the number of input variables for analysis, a greater prediction accuracy (20% validation) was obtained with an ANN model than with either of the three conventional empirical models based on NDVI, SR or photochemical reflectance index (PRI).
The use of remote sensing to estimate biological crop yields is being explored in many countries such as the United States, China and India, and likely will become the keystone of agricultural statistics in the future (Zhao et al., 2007).The fact that crop productivity vary greatly across climatic regions since it depends on agroclimatic conditions, the application of remote sensing in this field would be necessary.The variability of these conditions warrants models to be developed based on the conditions of different areas where the crops are planted.Moreover, there is room to improve on methodologies and principles already developed in the creation of new models.

Summary and recommendations
According to this review there is great potential for remote sensing to be used in investigating maize growth process in order to enhance production of this global crop.This is backed by the availability of products with higher spatial, spectral and temporal resolution becoming more efficient, affordable and cost-effective.Its success in biophysical vegetation parameters identification and monitoring through the various growth stages of maize has been studied even though not uniformly across the globe.
Spectral reflectance measurements can provide a basis for variable biophysical vegetation parameters.There is a good relationship between these parameters when the spectral reflectance is measured at the leaf scale.Research in the past has been undertaken to estimate foliar nitrogen concentrations in experimental fields using portable spectrometers, with promising results.However when measured at the canopy level the relationship is further complicated for various parameters (LAI, wet and dry biomass etc.) and therefore further research are needed.
Sensor readings were also found to be more associated with chlorophyll content during vegetative growth phases than during reproductive phases (Solari et al., 2008).
Taking advantage of improvements in sensor characteristics and processing techniques, the use of remotely sensed data for yield predictions for maize is gaining grounds.This is contributing substantially in a more accurate description of within-field crop yield variability which is a great concern in precision agriculture.However, it is still a challenge to develop accurate operational maize yield-estimating models.
The challenge of water deficiency monitoring over a large spatial area has been overcome through the use of remote sensing.It is practically suitable for assessing water stress and implementing appropriate management strategies because it presents unique advantages of repeatability, accuracy, and cost-effectiveness over the ground-based surveys for water stress detection.
One of the potential applications of remote sensing technique in agriculture is the detection of plant disease on extensive areas before the symptoms clearly appear on the plant leaves.This is advantageous because remote sensing detects biophysical changes before physiological changes are visible.The challenge of disease infestation on maize production warrants more investigation as not much was found during this literature survey.Early detection and delineation of maize infested areas especially in some of the high productive areas that are prone to diseases (e.g.grey leaf spot) using hyperspectral remotely sensed data could be attempted.Therefore, spectroscopy analysis could be considered as an efficient technique for non-destructive, rapid, and accurate measurement which is widely applied in agricultural fields for crop discrimination, monitoring of nutritional status as well as diseases (Sankaran et al., 2010).
spectral regions have been found useful for detecting water status in plants; one characterised by high reflectance caused by reflections and scattering of light in the spongy mesophyll layer (NIR 0.7 -1.3µm) and the other characterised by strong water absorption (mid infrared (MIR) 1.3 -3.0µm).The first one is based on the turgor pressure in the leaf tissues while the second is directly related to leaf water content.The reflectance spectra of water stressed plants absorb less light in the visible and more light in the NIR regions of the spectrum than plants not experiencing water stress.
wave infrared); NIR (near infrared); R (red); G (green); B (Blue) ground truthing and actual yield measurements in selected fields (pixels) that cover the full range of observed vegetation indices such as NDVI values.Viña et al. (2004) using visible atmospherically resistant spectral indices documented a capability for detecting changes in corn due to biomass accumulation, changes induced by the appearance and development of reproductive structures, and the onset of senescence.Shanahan et al. (2001)  used remotely sensed imagery to compare different vegetation indices as a means of assessing canopy variation and its resultant impact on corn (Zea mays L.) grain yield.Results showed that green normalized difference vegetation index