Hyperspectral imaging

High-resolution spectral methods that have been developed for satellite remote sensing systems are now showing potential for much more detailed evaluation of physical and chemical status of natural products and systems (Goel et al. Reference Goel, Prasher, Landry, Patel, Bonnell, Viau and Miller2003), including food product quality (e.g. protein levels in grains) and environmental markers (e.g. nutrients in organic manures). The advance of high-resolution reflectance spectroscopy has provided a means to detect levels of key molecules in living material with minimal preparation and outside the laboratory. This opens the door on precision estimation of key aspects of crop status relating to quality and environmental impacts (Hatfield et al. Reference Hatfield, Gitelson, Schepers and Walthall2008). For example, wheat production in temperate climates requires nitrogen fertilization in order to optimize production of bread-making wheat: the nitrogen availability must match the demand of the growing grain at the end of the season. The conjunction of a quality premium (for bread-making) with late-season nutrient requirement has the potential to lead to significant risks of excessive fertilization, enhancing the potential for emissions of nitrous oxide and leaching of nitrates. Management regimes that can minimize these risks can be realized through the effective deployment of sensing methods, such as hyperspectral sensing, to give the information required by crop management decision models.

A significant future challenge will be in efficient use of hyperspectral or optimized multi-spectral sensing methods. The complexity of plant tissue structures and the processes of mobilization and reutilization of nutrients within plants mean that simply sensing the levels of key molecules will not be a direct determinant of performance. A combination of sensing with models of transformations and transport will be needed. It will also be important to establish efficient translation of the identification of key hyperspectral bands into functional practical sensors at specific optimal wavebands. Success with a hyperspectral system in the confines of a research study does not give an immediate route to a robust sensor, cheap enough for mass deployment and optimal in its choice of bands for a variety of biological systems.

Trace gas detection and identification of diseases

The use of real-time monitoring of chemical species in production processes has developed through technologies such as the electronic nose. Real-time sensing of gases using electronic nose technology, based on chemometric sensing, has provided initial insights, although it is dependent on correlations that do not provide tight discrimination in all circumstances. Recent studies using real-time mass spectrometry to measure molecular species concentrations directly are suggesting potential routes to animal disease detection, although specificity is currently a challenge (Spooner et al. Reference Spooner, Bessant, Turner, Knobloch and Chambers2009). For plant disease, the release of specific chemicals at inoculation has suggested that there may be scope for monitoring systems based on the detection of key molecules, although the opportunity may be limited to sentinel plants or enclosed systems such as greenhouses (Jansen et al. Reference Jansen, Hofstee, Wildt, Verstappen, Bouwmeester, Posthumus and van Henten2009).

The challenge in taking this forward for the future is at least twofold. First, gas detection is preceded by a phase of gas release and dispersion. Understanding and managing this phase will be critical both to the resolution of the technique and the degree of interference by other factors. The second strand concerns the specificity and sensitivity of the techniques, and this applies broadly across many sensing methodologies. An optimal sensing technique will give high sensitivity even at early stages of the disease progress, such that interventions of whatever type can be triggered rapidly, minimizing the amount of disease that can appear in the population, and potentially allowing lower cost intervention for control. However, high specificity is equally important so that interventions are not triggered unnecessarily. Current sensing technologies demonstrate significant potential (Rumpf et al. Reference Rumpf, Mahlein, Steiner, Oerke, Dehne and Plümer2010), but reproducing this with sufficient sensitivity and specificity in practical situations sets a real challenge for the future.

Biosensors for status monitoring

Biosensors can directly detect active molecules in live systems and new sensors will be capable of determining the levels of molecular species that represent critical states in production processes (Velasco-Garcia & Mottram Reference Velasco-Garcia and Mottram2003). To date, the major global penetration of commercial biosensors has largely been limited to medical use of the glucose sensor for blood sugar to support diabetes management. This is despite the extensive basic and applied science research programme on biosensors. Market pull and breadth of applicability are critical factors (Luong et al. Reference Luong, Male and Glennon2008). However, we can expect significant advances in the first half of the 21st century, as the concept of targeted sensing and management develops. An example for agriculture will be better ways of managing dairy herds by monitoring the levels in milk of hormones associated with pregnancy. The principles are established (Xu et al. Reference Xu, Velasco-Garcia and Mottram2005), and with developments in technologies that give robust biosensors, the approach can be expected to aid productivity and reduce culling, as insemination is optimized and knowledge of pregnancy status is improved. The principles of detecting hormonal indicators of pregnancy status in dairy animals or the presence of food pathogens using biosensors (Nayak et al. Reference Nayak, Kotian, Marathe and Chakravortty2009) have been proven at the laboratory level. The translation into realistic components of biosystems management will be an important target in future.

Spatial variability and optimized inputs

The precision agriculture concept has been under active development for more than 20 years, but the concepts are still in the process of translation from a vision to practical approaches. The prime challenges can nearly all be associated with the chain from sensors and detection through to management and control (Zhang et al. Reference Zhang, Wang and Wang2002). The basic tenet still holds true – managing inputs uniformly across whole fields has only limited logic as soils and topography vary. The spatial variation of weed infestations also demonstrates that herbicide inputs can benefit from spatially targeted control (Slaughter et al. Reference Slaughter, Giles and Downey2008).

Progress in this area will benefit from new sensing methods. However, it is also critically dependent on the establishment of appropriate mathematical and statistical approaches that can translate uncertain information about the state of soils and crops into reasoned decisions that can be implemented at critical points in the growing season. There will always be distinct limits to the precision possible in these decisions, given the fact that spatial patterns of crop performance can be inverted between wet and dry years, and forecasts of yield in mid-growing season depend on weather prediction – an inexact science.

The use of geostatistics and other mathematical tools is building the basis for interpretation of variation (Lark & Wheeler Reference Lark and Wheeler2003). Data on crop status gathered during one growing season and at harvest will underpin decisions in future seasons. At the centre of the decision-making process will be parsimonious models defining the form and scale of responses to individual interventions, such that the principles of decision support can be convincing to the expert farmer and build confidence in advanced approaches (Oliver et al. Reference Oliver, Robertson and Wong2010). These methods could be readily developed to optimize not just economic productivity but also productivity constrained by well-defined environmental goals.

Of course, precision agriculture will vary enormously across farming systems. Delivering real value will depend on the identification and commercialization of technologies that can have wide application and establish a competitive market. For example, sensors providing information on nutrient content of manure inputs and harvested grain (see above) will provide scope for improved system control (Sinfield et al. Reference Sinfield, Fagerman and Colic2010).

It is important to note that, independent of spatial variation, accurate prediction of optimal fertilizer inputs for crops at field scale is still a significant challenge (Sylvester-Bradley & Kindred Reference Sylvester-Bradley and Kindred2009). The introduction of sensing and modelling systems through precision technology will provide parallel benefits by addressing this problem.

Machines optimized for sustainable production

The developments in agricultural field mechanization through the 20th century were dominated by increases in size and scale, and in vehicle mass. The benefits have been in increased work rates and an economical platform for the latest technological developments, whether in tractors and cultivation control or in harvesters of vegetables or grain. However, a significant adjunct to these developments has been increased risk of soil damage through the ground pressure of these vehicles. The advent of the robot has suggested that there might be a niche for small units as well – ‘service robots’ that can operate autonomously to deliver specific tasks within the field in ways that maximize precision and minimize extraneous damage (Bakker et al. Reference Bakker, van Asselt, Bontsema, Müller and van Straten2010). With 24 h operation, identifying localized problems with weeds or crop diseases, and then either applying local control, particularly for weed eradication, or feeding back to an information base for optimized large-scale interventions in the case of developing disease epidemics, may provide a more timely and controlled approach, often with less recourse to damaging vehicles.

Controlled traffic systems, which restrict the soil damage to accurately defined trackways in the field, have become much more practicable with the availability of global positioning satellites (GPS) and are another way in which machinery systems can be optimized, with production and environmental benefits (Tullberg, Reference Tullbergin press). Integrated approaches to machine-crop systems can be expected to follow from further innovation in these areas.

Real-time machine control for decreased pesticide (herbicide) input

The reduction in acceptability of some chemical interventions because of ecological impacts puts new pressures on intensive farming productivity and quality. For weed control, mechanical interventions are feasible but impractically slow without new control techniques. Real-time image analysis to control the tools has been demonstrated to be viable (Tillett et al. Reference Tillett, Hague, Grundy and Dedousis2008). It draws on key strands in machine vision research, relating to stable discrimination under variable lighting conditions that are intrinsic to outdoor operations, pattern recognition and analysis of key contrasts between weed, crop, soil and debris. Operation at high speed and in wide bouts makes the technique match the requirements of high work rates for low cost production, and also meets the demands of organic systems where the weed pressures can be a critical factor in business sustainability (Bakker et al. Reference Bakker, van Asselt, Bontsema, Müller and van Straten2010). The integration of sensing technologies with sophisticated machine control, linked to knowledge of spatial distribution of weed pressure, will contribute to more sustainable management systems.

It may not be feasible to replace herbicide use entirely with mechanical interventions, although the ability to operate under near full crop cover is greatly aided by precision navigation. However, the withdrawal of approvals for many chemicals suggests that a new paradigm may be upon many farming systems. These concepts may be relevant to insects and disease, although the mobility of these agents is much greater – early detection and finer-scale interventions require further research.

Data integration and data mining – information management systems for decision support and control

As well as the direct interpretation of sensor outputs to determine process control actions, there will also be advances in more general use of information about system performance to guide the management of complex systems. This can be expected to be particularly important in complex, interacting and continually changing systems – examples include optimization (minimization) of pesticide inputs, and food chain management from harvest through shipping to store and consumer. These approaches can also be a major contribution to traceability in the food chain (Kondo Reference Kondo2010). Increased use of widely available sensors can be expected to provide indicative information that can contribute to alarm monitoring or adjustment of setpoints. Much of the interpretation will draw on neural networks or similar optimization approaches that can handle large quantities of data and draw rapid conclusions based on learning approaches.

Models of farming systems, including life-cycle analysis, optimization and decision support

A major challenge is to identify and develop regulatory approaches that not only reflect national and global strategies for environmental protection but also can operate near-optimally in practice. Regulations are often blunt instruments, tailored for ease of use rather than maximal delivery of the required benefit. Life-cycle analysis (LCA) and other tools can provide holistic information on the impacts of production processes and the relative environmental damage associated with various products (Gnansounou et al. Reference Gnansounou, Dauriat, Villegas and Panichelli2009; Meisterling et al. Reference Meisterling, Samaras and Schweizer2009). Farmer optimization in the face of regulatory constraints is generally going to be motivated by maximizing profit under the regulation. The result can be detrimental by steering farm systems into new areas that could be more damaging. Imposing regulatory ceilings on fertilizer or herbicide use can lead to major income reductions if accepted simplistically by the farmer, but crop management and rotation changes can have substantial unexpected environmental consequences. Annetts & Audsley (Reference Annetts and Audsley2002) used a linear programming whole-farm model to demonstrate how this might happen and the scope for more optimal management.

The concept of model-based assessment of the acceptability of farming systems in relation to environmental regulations still needs further study and development. Realistic yet parsimonious models are essential, yet will always be constrained in their validation.

Holistic management systems

The same approaches of modelling and process optimization can be applied to the management of inputs to crops. Excessive use of fertilizers and pesticides in intensive agricultural systems need to be avoided, yet optimal use is dependent on a wide range of factors, including soil type and past and future weather conditions. The development of modelling approaches that can integrate current knowledge of crop and weather with information on likely responses to fertilizers and pesticides have the potential to underpin near-optimal management regimes, oriented to minimizing environmental damage (Parsons et al. Reference Parsons, Benjamin, Clarke, Ginsburg, Mayes, Milne and Wilkinson2009). This is not a simple task. For example, although the optimal level of nitrogen fertilizer can be readily defined experimentally after the cropping season, by analysing outcomes, the ability to predict this optimum at the start of the season is highly problematic, and even in the spring at the time of the main dressings, the prediction accuracy is currently limited. However, the benefits are potentially considerable – losses of nitrates and possibly also of nitrous oxide are low when nitrogen levels are below the optimum for crop production, but increase rapidly when the optimum is exceeded. These complex tools will of course need effective means of communication to the decision maker, requiring effective approaches to decision support and the means to present the advice direct to the end user (Karmakar et al. Reference Karmakar, Laguë, Agnew and Landry2007; Antonopoulou et al. Reference Antonopoulou, Karetsos, Maliappis and Sideridis2009).