The management of agricultural production is undergoing a change, both in philosophy and technology. Until recently, agricultural managers have generally made decisions regarding fields based on average conditions within those fields, with data that was often sparse and qualitative in nature. Soil fertility was determined by compositing soil cores into a single sample that was intended to best describe conditions across a field. Field scouting for crop condition or pest infestations was done at a few locations within the field, and observations often have been more qualitative than quantitative. For the most part, whole fields have been considered to be the basic agricultural production units, and have been managed for the mean condition or, in the case of pest management, managed intensively to overcome variability within that field.

Historically, a desire to improve production efficiency and farm income has stimulated interest in innovative technologies. Advances in technology, as well as other factors such as farm policy have contributed to increases in the size of individual farmsteads and fields within a farmstead. With this larger scale of operation, the potential for the individual to effectively manage variability by observation and experience has declined precipitously. In addition, as individual farm fields increased in size, within-field variability has generally increased. A major feature of today's precision agriculture is that it allows producers to manage previously unmanaged variability as well as the increased variability resulting from increased field size. In other words, precision agriculture will allow several geographic units currently being managed as a single entity (a field) to be addressed as individual decision-making units. Managers will be able to respond to the distinctive agronomic characteristics that exist within the subunits, in contrast to today's approach of addressing the average needs of several units or extreme conditions in parts of the field, such as pest outbreaks in small patches.

The incorporation of information technologies into agricultural production practices began in the mid-1980s and has increased sharply in recent years. While the use of information in agricultural decision making is not new, agriculture is experiencing a vast increase in the amount of information available, and in the timeliness and means by which information can be collected, analyzed, and used to manage inputs and outcomes of agricultural practices. The application of new information technologies in agriculture is known by several terms, including precision agriculture, precision farming, and site-specific management. A variety of definitions have been offered for the concept of integrating information technologies with agronomic practices. Most authors have focused on the ability to obtain data and to vary production inputs on a subfield basis. While this is an important aspect, there are other geographic scales at which information can be obtained and used to facilitate site-specific management. The committee chose to view precision agriculture broadly, adopting the following definition:

A key difference between conventional management and precision agriculture is the application of modern information technologies to provide, process, and analyze multisource data of high spatial and temporal resolution for decision-making and operations in the management of crop production. Advances in the technologies will be an evolutionary process and they will continue to be adapted for agricultural decision making.

Precision agriculture has three components: capture of data at an appropriate scale, interpretation and analysis of that data, and implementation of a management response at an appropriate scale and time. Each particular manageable factor has its own scale of variability. Area-wide management of insects and weather forecasting for crop management decisions are examples of variables that are managed at a scale larger than the individual field. Other factors like soil fertility and pest distributions can vary significantly at the subfield level and over the growing season. Therefore, it is natural and important to perceive precision agriculture in terms of finer spatial or temporal units of decision making.

Advances in information technology and their application in crop production, which are labeled as precision agriculture in this report, are creating the potential for substantial change in management and decision making in agriculture. The word potential in the previous sentence is critically important. The various technologies and practices that will make up tomorrow's precision agriculture are only emerging, being tested and refined, and implemented or rejected

today. This process is further enhanced by the dynamic nature of advances in information technology. A capability that is technically or economically unfeasible can become feasible as a result of a technological innovation occurring well outside the arena of agricultural technology development or agricultural research. Thus the process by which precision agriculture is adopted could be fragmented and discontinuous. Therefore, it is impossible to specify the precise dimensions and characteristics of the precision agriculture of the future.

Precision agriculture could materially affect on-farm decision-making processes that depend on implied knowledge gained by observation and experience. While its precise dimensions continue to evolve, the following features characterize most precision agriculture applications in use or under development:

• Data capture tends to be electronic, automated, and relatively inexpensive.

• Data capture can occur more frequently and in more detail.

• Information, either captured as a part of field operations or purchased externally, can be considered separate input into the production operation. (It is also a feature of integrated pest management and sustainable agriculture concepts.)

• Data interpretation and analysis can be more formal and analytical.

• Scientific decision rules are applicable to actual farming operations.

• Implementation of the response can be more timely and more site specific.

• Performance of alternative management systems can be quantitatively evaluated.

The long time lags between input decision making, application of inputs, and observation of yields in crop production systems make it difficult to evaluate decision-making effectiveness. The chance for misinterpreting results is further heightened when inputs and outcomes are observed rather than measured. The difficulty of learning in such settings is not constrained or unique to farmers. Considerable research has documented that human decision making is more likely to suffer bias and misinterpretation when (1) feedback loops are long between the time the decision is made and the outcome occurs and (2) cause/effect linkages are not simple (Einhorn, 1980; Hogarth and Markridakis, 1981). These two characteristics apply to traditional crop production settings.

The uncertainties associated with the rapid evolution of information technologies and the dynamics of the process of adopting precision agriculture represented a significant challenge in the preparation of this report. However, these same uncertainties provided considerable excitement and a sense of mission for the project. Tomorrow's precision agriculture will be significantly affected by actions in the public and private sectors today.

The focus of this committee, therefore, was not on predicting a single future. Rather, members chose to recognize the uncertainties inherent in the future evolution of precision agriculture and to emphasize possible paths and the implications of those paths. Further, the study recommendations define key actions that

society can undertake to extend the dimensions of precision agriculture where they are deemed most desirable.

Agricultural production systems vary in many ways, including scale of operation, commodities produced, and philosophical approaches to management. Current production systems draw on diverse approaches and knowledge bases. For any approach, information technologies will play an increasingly important role in agricultural production and natural resource management. This impact will be felt directly through the coupling of newly acquired information with recently developed tools for agricultural production, on-demand products and services, and increased access to information and services.

A number of scales characterize crop production systems of today. These scales might be viewed as a continuum ranging from individual plants in a field to plant populations, fields, farmsteads, and regions. Others have used this Lewin-Kolb model of hierarchies as an organizational structure to study complex issues such as pesticide regulation and diversity in agroecosystems (Olson et al., 1995). Consider this continuum in the form of a spatial spiral ascending from the subfield to national geographical levels (Figure 1-1). As we move up the spiral we

FIGURE 1-1 Scales in a spiral. A number of scales characterize crop production systems of today. In precision agriculture, an unprecedented amount of spatial and temporal data may become available at the individual plant, farm, and regional scales. At each scale various processes will influence crop production. A goal is to determine an optimal scale for data collection and management response. Communication technologies will provide connecting threads up and down the spatial spiral.

move from individual plants to fields and regions. Fresco (1995) underscores the need to relate phenomena or outcomes to processes occurring at both higher and lower level scales. The goal is to determine an optimal scale at which each process is to be studied, one in which variability is minimal. For example, if plant population is dependent upon small-scale variation in soil physical and chemical properties, then varying seeding rates may require information and hardware capable of rate changes every few centimeters. Such information may reside locally in a nearby computer in a farmhouse. At a wider scale, real-time weather information collected from locally placed weather stations may provide irrigation or area-wide pest management information in a timely manner to improve decision making for a field, farmstead, or county.

Communication technologies will provide connecting threads up and down the spatial spiral. Telephone, high-speed digital lines, and wireless communications are needed to link the various levels together. For example, digital data could be collected by an on-the-go yield monitor in a combine, sent via a wireless cellular link to the operator's home computer, and retrieved via a high-speed Internet connection by an agricultural chemical dealer. The dealer may then add the yield data to a nutrient management analysis and send recommended fertilizer application rates for various subfield units back to the farm operator's computer.

Different scales of assessment are being used to investigate aspects of crop-environment systems. Scale can be considered for both information sources and management actions. Depending on the situation, data from different scales may be combined and used to determine management actions at another scale. For example, a producer deciding what crop and variety to plant (field scale) may consider the available forward contracting prices (national or global scale), the availability of custom field operations (farm scale), and a field map of soil water-holding capacity (subfield scale). With precision agriculture methods, such decisions can be made with more objective data. Some of the uncertainty factors can be reduced with the information technologies of precision agriculture, although the extent to which this will be feasible and of value to the grower is not clear.

Information technologies permit the modern producer to obtain detailed spatially explicit information at the scale of entire farms but with information sufficient for efficiently managing the land at the fine scale. Most of the new precision agriculture technologies can be used to disaggregate information—for example, to characterize soil, yield, nutrients, and water variation within fields—as well as to assemble regional information. Perhaps the ultimate disaggregation would be to look at agricultural fields as a collection of individual plants. The extent to which data are disaggregated or reassembled for different spatial units depends on the nature of the management problem and the resolution of the data gathering techniques. Decision makers will need to consider the spatial heterogeneity of the area being managed and the relative value of the information. (A brief review of

the impact of information technologies on current management decision processes can be found in Chapter 2.)

The potential for individually managing small areas, whose size is determined by local characteristics and crop value, is one of the most enticing aspects of precision agriculture. The ability to repeatedly locate a specific site and measure agronomic characteristics provides an opportunity to optimize management throughout the production area. Subdividing a field into small management units may improve both the economic and environmental sustainability of crop production systems.

The earliest advocates of precision agriculture took the approach that management decisions should be based on soil characteristics, assuming that similar soil series could be managed as homogeneous units. Subsequent research showed that for many soils, nearly the same nutrient variability exists within the mapped soil series as among them (Sadler et al., 1995). Even precise management based on variability of the physical and chemical properties within soil types may or may not be sufficient for optimal management of crop production activities.

As producers try to manage smaller areas, the law of limits comes into play more strongly. For any given site, from year to year, the most limiting factors to crop growth can change from nutrient or moisture availability (deficit or excess), to disease or insect pests, to weather factors. In fact, the limiting factor may change within the growing season as the crop matures and its needs change. For improved decision making, managers must be aware of the limiting factors for each subfield unit and be able to modify management at that scale. The determination of the most limiting factors is currently both difficult and expensive, and these costs are considered by decision makers. All of these concerns point to the need for analytical systems and technologies that can determine the important factors and decision-support systems that can use available data.

Some management factors exhibit a relatively small amount of variability. For example, levels of less mobile soil nutrients (i.e., potassium and phosphorus) may exhibit little variation in crop response within some fields that have received heavy fertilizer applications for many years. These crops may be subject to greater variability from other influences—such as weather, nitrogen, diseases, and insects—particularly if the time frame for assessing the performance of a method is short (i.e., a single growing season). Similarly, technologies that work well for one cropping system or biophysical setting may not work in another. Efficacy testing should be done for a variety of settings and systems and over several growing seasons.

It is unarguable that an individual grower's precision agriculture data has substantial additional value when combined above the subfield level with similar

data from other production operations. Management strategies consistent with our definition of precision agriculture are currently practiced, and new strategies will be developed that address spatial and temporal variability at the scale of the whole field and larger. While this report focuses on subfield level precision agricultural practices, a discussion of two key larger-scale strategies follows.

Large amounts of spatially referenced data on individual fields are, or soon will be, generated by yield monitors, real-time and remote sensors, on-the-ground sampling and observation by producers and consultants. This site-specific information will have value for use within individual fields in ways discussed in Chapter 2, but will also have value when combined with data on the same variables collected for nearby fields. Seed, chemical, and machinery agribusiness, among others, are assisting growers in data collection and interpretation. In a number of cases, agribusiness is providing financial assistance so growers will share data with the agribusiness itself. Several companies have promoted a concept of data aggregation which permits growers as well as an agribusiness free access to participant's data. Still others have promoted concepts of data collection in which data could be purchased by third parties. Many growers have expressed opposition to any of their data being shared with others. However, most growers do agree that there is economic value in the learning that results from data sharing and that may increase the likelihood of vertical integration of agricultural operations. Though it is unlikely that a commercial interest will freely share information to which they have purchased rights and made further investments, other groups may see benefits from voluntary sharing. Grower clubs such as Practical Farmers of Iowa have been successful models of farmer-directed research in which land grant or private sector consultants act as facilitators in planning and implementing research trials. The idea is for a number of growers to implement similar practices of interest in their farm operation (i.e., row-spacing, herbicide dose and timing, cultivar selection) in statistically sound on-farm experiments (Stroup et al., 1993). In these clubs, data are openly shared to identify desirable practices in local growing areas. Imagine the same grower clubs now sharing spatially referenced data from experiments where growers agree to apply similar agronomic practices. The potential to create locally derived recommendations from locally collected data is a fascinating prospect. In effect, a version of this vision is in practice today with the private crop consultant. By working with numerous growers, the consultant is afforded the opportunity to observe how diverse recommendations can affect crop fitness, yield, and production efficiency in farming enterprises as small as several acres to those that extend over thousands of acres. Such an approach would require growers to openly share data with fellow producers.

There are opportunities to link management decisions at various levels to improve soil and water quality. The National Research Council's report on Soil and Water Quality (National Research Council, 1993) described the inherent links between farming systems and the landscape. Management practices to improve input use efficiency and reduce erosion can improve the quality of the surrounding watershed. The Committee on Long-Range Soil and Water Conservation recommended use of landscape buffer zones that connect farms and fields, provide widespread protection to waterways, and prevent soil degradation. Focusing on the impact of within-field production practices on adjacent ecosystems changes the unit of analysis to the landscape scale for studies on agricultural nonpoint sources of pollution. Landscape analysis considers effects of farming practices on larger areas than a specific crop field. Coordinating information at various levels could enhance protection of the environment. For instance, tracking production practices across a watershed could be useful in targeting areas with soil and water quality problems (National Research Council, 1993).

The appropriate scale for management will vary according to the factor most limiting to productivity. Manageable factors such as soil fertility or weed competition may vary significantly at a subfield level, thus input use can be based on subfield units. However, there may be more utility in managing other factors at a field or farm level. For example, because insects migrate over areas larger than a field, monitoring their movements on a regional scale may be appropriate. Acquiring other regional data also may improve the accuracy of the decision-making process.

Information provided to producers that is regional in nature, can have a direct impact on local management decisions. Evapotranspiration is typically monitored using networks of weather stations that cover large areas. Regional data also interacts with more site-specific data that producers can incorporate into their decision making. The California Irrigation Management Information System (CIMIS) is a computerized crop weather information system that producers can access by modem or the Internet to obtain hourly and daily weather conditions. Producers combine regional evapotranspiration data and local soil-and crop-specific coefficients for their fields to determine the daily water use and water demand of their farms (see Box 1-1).

It is unclear how to appropriately use data collected at different spatial scales together to help make better decisions. There are significant statistical and modeling issues to be addressed. Precision agriculture will greatly increase the amount and perhaps the availability of geographic data snapshots for many cropping fields, which will increase the demand for these analytical techniques.

A fascinating aspect of precision agriculture is that a single technology is not being undertaken to improve a single practice. Instead, across the crop-production sector of the United States, precision agriculture is emerging as the convergence of several technologies with application to several management practices. However, every technology is not necessarily required or applicable for every practice on all crops, and development and enhancement of several of the potentially relevant basic technologies are being driven by forces outside of the agricultural sector. Thus it is difficult to develop a generally accepted view of the dimensions of precision agriculture. Every area of information technology—microelectronics, sensors, computers, telecommunications—is in an evolutionary process of continuous improvement. As these introductions take place, some products will become economically feasible for agricultural applications. In Box 1-2, describing a vision of tomorrow's crop consultant is considered. According to

this vision, while many precision agriculture technologies are available for use, individual producers will assemble those technologies that address given management issues in their particular production systems. The following discussion provides a broad overview of the precision agriculture technologies and practices that are or soon will be available. For more detailed information, the reader may want to access additional literature sources such as Pierce and Sadler (1997); Robert et al. (1995); and Robert et al. (1996).

Research and development of many technologies used in precision agriculture have occurred outside the agricultural community. In the past century, of course, other developments such as the internal combustion engine, electrical power, telephone, and weather satellites produced outside of agriculture have been introduced to the agriculture sector. Precision agriculture technologies such as the global positioning system (GPS), geographic information systems (GIS), and remote sensing have their core constituencies outside agriculture. Crop and

soil sensors operating on farm machinery, variable-rate fertilizer applicators, and yield mapping systems are technologies that have been developed within the agriculture sector by private industry. Other economic sectors have supported the research and development of some of these technologies, which is a financial benefit to agriculture. Precision agriculture involves the integration of these information technologies with agronomic knowledge.

Georeferencing refers to relationships among data based on their geographic locations. This spatial emphasis implies a new way of looking at agricultural information and site variability. Although spatial variability has always been recognized, data comparisons have often been made without specific information on site location, yielding qualitative results. Comparisons of data detailed from specific locations which are obtained from specifications by using various precision agriculture methods will be one of the important new techniques that can improve farm management.

The value of a database for precision agriculture practices increases when the data layers are spatially referenced to each other. Co-registration of data will become critically important as management units get smaller and as more precise field data (location precision from submeter to a few centimeters) become available. It is expected that data referenced to physical location will allow different types of information to be compared and quantitatively analyzed at multiple locations. For example, physical properties of soil core samples collected from a field could be compared with other spatially explicit data available for decision making, such as characteristics of the mapped soil unit and topography, yield monitor data, and irrigation, nutrient, or pesticide applications recorded during variable-rate applications.

The Global Positioning System (GPS) is a system of satellites emitting electronic signals that can be received by mobile field instruments sensitive to the transmitting frequency. Positioning is achieved through the use of simultaneously received satellite transmissions from four or more satellites above the horizon. With a constellation of 24 satellites, any location on earth can have four or more satellites in view for 24 hours each day. By referencing the satellite's exact location and the time the signal takes to travel between the transmitter and the receiver, the location of the receiver can be determined by triangulation.

Use of the GPS receiver allows latitude and longitude coordinate information to be associated with data obtained from a specific site on the field. The GPS can also be used to provide navigational guidance, enabling a producer to revisit a spot in the field and check the efficacy of management decisions. The GPS is an

essential field component for most mapping-based precision agriculture and other measurements of field characteristics that would be used to determine product application maps. Even for operations with real-time sensing and control of inputs, GPS positioning will be valuable. If the sensed parameters and application rates are recorded and georeferenced, these data can be included in the management database. Adoption of GPS and other spatial referencing technologies will have a widespread impact on data collection and analysis.

The positions provided by GPS receivers currently are not sufficiently accurate for dynamic real-time precision agriculture uses. Various errors, including those introduced by the U.S. Department of Defense (DOD) for security purposes (selective availability), contribute to the inaccuracy (National Research Council, 1995c). The present system under selective availability has an accuracy of about 100 meters. However, technical solutions are available to improve the positioning accuracy. A technique known as differential correction is widely used to remove the effects of the error sources. Position error is determined by using one or more fixed base stations to compare the calculated position with the station's known location. By combining the error values with the GPS signals, position accuracy can be improved to about 2 meters or less. The augmented positioning is known as differentially corrected GPS (DGPS). These corrections can be made either by software in a postprocessing operation or by hardware for real-time positioning. Most precision agriculture operations require the availability of real-time positioning, necessitating the transmission of the differential correction signals to the GPS receivers in the field. Differential correction procedures are cumbersome, prone to signal loss, or expensive depending on the method used to generate and transmit the differential correction signal.

Commercial applications of georeferencing systems will grow dramatically over the next decade, both in agriculture and other industries. Some commercial businesses offer real-time differential correction services in space-based or land-based networks to their subscribers. Many of these providers are focused on nonagricultural industries and so do not adequately cover rural areas with their signals. The U.S. government provides differential correction signals through Coast Guard beacon signals, but access to these signals is limited to areas near navigable waterways (coastline and rivers). The Russian government continues to operate its GLObal Navigation Satellite System (GLONASS) which could augment basic capabilities of GPS. Since GPS and GLONASS use different time standards and coordinate systems, these differences will need to be corrected by combined receivers (National Research Council, 1995c). Receivers that use techniques such as carrier phase tracking (Real Time Kinematic) offer higher accuracy, but have higher costs.

Several other factors can limit the application of GPS in precision agriculture. Time delays for updating signals may limit the utility of DGPS for on-the-go sensing, particularly for high-speed operations such as aerial applications. System inaccuracies make data collected along a crop row appear to suddenly shift, creating map displays that do not match actual travel paths. Signals can be severely

degraded by moderately inclement weather conditions, foliage and electromagnetic radiation. Position data are not always available at the one-second frequency that is expected, so data gaps are created (data dropouts). Increasing the accuracy and reliability of GPS will increase its acceptance by producers and its utility for geographic referencing in precision agriculture applications.

Digital geographic data that can be stored, analyzed, integrated, and displayed in different representations, form the core of precision agriculture. The software packages used to handle such data, Geographic Information Systems (GIS), are available with a wide range of capabilities and costs, but all are able to graphically display georeferenced data. Although a single data layer (i.e., yield data) can be mapped with the use of less-sophisticated software, more complex relationships (i.e., temporal patterns or multivariate comparisons) are best performed with full-function GIS packages. The data layers derived from combinations of raw data can generate information about spatial variability among factors in crop production. It is expected that adequately co-registered data will be quantitatively analyzed through the use of geostatistical and other procedures.

Available GIS software ranges from simple map display systems to fully functioning systems capable of analyzing and integrating complex spatial databases. Some data can be stored as polygons within which the attributes (i.e., soil types) are considered to be homogeneous. Data can also be stored in a uniform array of grid cells or pixels with homogeneous attributes, which is the format used for remote sensing images and U.S. Geological Survey digital elevation maps. Most fully functioning GIS programs today can be converted between these formats, which has made it easier to combine data from different sources.

Among the most important roles for GIS are the database functions for farm record keeping and for comparing management decisions, yields, pest activity, groundwater quality, and other concerns related to past and current practices. GIS can store farm records of inputs and outputs in a spatial array. For instance, data on crop rotation, tillage, nutrient and pesticide applications, yield, soil type, roads, terraces, or drainage tiles can be stored in a GIS. Data layers can be derived from digital orthophotography. GIS will enhance other components of precision agriculture such as yield monitoring and farm-based research (i.e., crop modeling and efficacy testing) as well as provide better record keeping for producers. Such software has the potential to integrate all types of precision agriculture information, interface with other decision support tools, and output (printed or electronic maps) that can be used in precision applications. A key to realizing the promise of a dynamic GIS will be development of connections between the relational database and the decision support system. A disadvantage of the current generation of geographic information systems is the complexity of the software and the steep learning curve involved in using and interpreting spatial data in a valid and robust

way. The limitation with some commercial software is that spatial relationships among data layers often cannot be rigorously quantified; only visual relationships can be made. This situation is rapidly changing as several vendors are developing fully functional GIS programs intended for use on PCs. This should lead to software and hardware systems that are more user friendly and less expensive. In addition, firms are emerging in the marketplace that can provide GIS services or software tools to growers and field consultants. There is an urgent need to make fully functional GIS easier for nonspecialists to learn and use in order to transfer this technology to the agricultural community.

GIS can be used with a spatially distributed process model as the basis for subsequent decisions on precision agricultural practices such as variable-rate applications. Several classes of models should be considered as part of the suite of tools for precision agriculture.

Yield mapping systems record the relative spatial distribution of yield while the crop is being harvested. These systems collect georeferenced data on crop yield and characteristics such as moisture content. The resulting maps can dramatically illustrate the areas of yield variability from either natural processes or agricultural practices. Because yield is a primary factor in most management decisions, precise yield maps are desired to confirm spatial treatment decisions.

Yield monitors have been developed for only a few crops, primarily cereal grains. Reliable monitors for vegetables, fruits, cotton, and other high-value crops are currently under development but are not yet widely available. Yield is more difficult to monitor for fruit or vegetable crops that are harvested manually or repeatedly. Use of machine-mounted yield monitors currently is limited to crops that are mechanically harvested in a single pass, such as potatoes, sugar beets, and processing tomatoes. Other techniques such as remote sensing may provide alternatives to yield monitors. The use of precision agriculture techniques in nongrain crops may be limited by the lack of appropriate yield monitoring systems.

Since 1992, grain yield mapping has been done by using mass flow and moisture sensors to determine grain mass and using GPS receivers to record position. Yield monitors measure wet grain flow, grain moisture, and area harvested to determine moisture-corrected yield per acre. Because the mass-flow measurements are made in the combine's clean-grain conveying system, there is a shift in harvester position between the point where the grain is actually cut and the location of the machine where it is measured. This shift results in dynamic inaccuracies that currently cannot be completely removed by subsequent data processing. Field totals (with recommended calibrations) are considered more accurate than are small subfield yield measurements. Although yield monitors have been promoted widely, further yield monitor refinements are needed to improve their accuracy for precision agriculture applications.

Precision agriculture was pioneered by domestic U.S. industry, beginning with the conception and implementation of Variable-Rate Technology (VRT). VRT applicators spatially vary the application rates of agricultural inputs such as seed, fertilizer, and crop protection chemicals. VRT systems include specialized controllers that vary specific material flow rates, even multiple product rates simultaneously, in response to a desired change in the local application rate (on-the-go). VRT systems can be designed in different ways depending on the products to be applied and the source of the information utilized to specify local rates.

Present commercial VRT systems are either:

1. Map-based, requiring a GPS/DGPS georeferenced location system and a command unit that stores an application plan of the desired application rate for each location within the field, or

2. Sensor-based, which does not require a georeferenced location system, but includes a dynamic command unit that specifies application through real-time analysis of soil and/or crop sensor measurements, for each location within the field as it is encountered.

Historically, VRT methods were introduced by industry during the mid-1980s. Dry nitrogen, phosphorus, and potassium fertilizer application rates were simultaneously varied on commercial spreader applicators based on a predetermined map strategy (developed from earlier data collection such as photographically derived soil maps or grid sampling). Farmer-owned machinery has been equipped with VRT for fertilizer applications requiring a standard liquid blend. In this case, product application rates are based on soil properties measured in real-time. Limited use has been made of sensor-based VRT by commercial applicators to date. Herbicide application responsive to soil organic matter (Gaultney and Shonk, 1988) is the singular exception (McGrath et al., 1990).

Commercially available sensors employed for VRT include those responsive to organic matter, cation exchange capacity (CEC), topsoil depth, soil moisture, soil nitrates, and crop spectral reflectance. Proponents of real-time sensor-based VRT application have observed that soil and crop conditions are more variable than measurements obtained from current map based methodologies. Optimal crop management results are not expected from current GPS/DGPS/GIS methodologies which are limited to one sample and one control change per second. The application of nitrogen fertilizer in response to measurements of side-dress soil nitrates and CEC (Colburn, 1991) and the application of nitrogen fertilizer in response to wheat nitrogen status as detected by spectral reflectance (Stone et al., 1996) are two examples of on-the-go sensing based VRT which do not rely on GPS/DGPS or GIS systems.

Real-time sensors offer some benefits over map-based techniques for VRT. Real-time sensing is a direct and continuous measure of the attribute of interest

thus allowing the user to reduce the amount of unsampled area in a given application. In map-based applications, maps are based on a limited number of samples thus creating the potential for errors in estimating conditions between sample points. An additional uncertainty is associated with GIS due to the temporal disconnection that occurs when samples are mapped at some point in time and a response is made at some later time. In the case of dynamic variables such as soil nitrogen content or pest distributions, significant change in the amount and distribution of the attribute of interest can take place during that time (Sudduth et al., 1997; Wollenhaupt et al., 1997).

Sensor based VRT is employed on Midwest farm equipment to:

• Vary anhydrous ammonia application in response to soil type variations.

• Vary planting population in response to soil CEC and topsoil depth variations.

• Vary herbicide rates in response to soil organic matter variations.

• Vary starter fertilizer in response to soil CEC variance.

• Vary nitrogen fertilizer at side-dress time in response to soil CEC, topsoil depth, and soil nitrate levels.

Map-based VRT is employed in the high-volume commercial (contracted) application of phosphorus and potassium fertilizers and lime using high-flotation applicators. Map-based variable-rate application systems for farm tractor use are widely available for liquid fertilizers, anhydrous ammonia, herbicides, and seeds. Map-based VRT controls for water and fertilizer are also available for center pivot irrigation systems.

Because of the additional capital and maintenance expense for high volume, pneumatic or liquid material control systems in high-flotation VRT, application costs are higher than for conventional floater application technology. Floater VRT application of granular fertilizers is typically $2 to $3 per acre higher than non-VRT applications.

Costs for upgrading tractor-mounted application controllers to add VRT capability are often nominal. Upgrading a controller to allow for automated adjustment of application rates is a minor technical departure, representing only a software/hardware interface. However, the producer must also have a computer that manages GIS data and sends rate change commands to the controller, and a GPS/DGPS receiver. Such a system can be assembled by more technologically sophisticated producers. In other cases, a VRT system may be more complex and costly, incorporating multiple chemical injection hardware and GIS/GPS/DGPS systems as an integrated, dealer-installed unit. Regardless of the type of VRT system utilized by a grower, implementation of a map-based VRT system requires full consideration of all related costs, including data acquisition, the GIS and GPS/DGPS to create and execute application maps, and the often time-consuming intellectual capital investment in learning how to successfully use all components of the technology.

The cost of obtaining and interpreting soil test information on which to base floater or tractor-based application rates is a limiting factor in the site specificity of map-based VRT. Soil samples normally are acquired at a rate of one sample per 2.5 acres to reduce costs for collection and analysis. In an Illinois test, fertilizer requirements based on 2 grid sizes were compared to uniform application rates. With a grid size of 0.156 acre, recommended fertilizer rates decreased dramatically resulting in a fertilizer savings of $18.00 per acre compared to $0.25 per acre savings with a 2.5 acre sized grid. The cost to collect the samples on the more detailed grid, however, far exceeded any savings in fertilizer costs (Illinois Agri-News, 1996). One key to improving the efficacy of map-based VRT is the development of additional cost-saving, higher sampling density sensor methodologies.

Basic research is needed to investigate soil and crop processes applicable to development of ground-based sensing systems. Sensors offer the opportunity to automate collection of soil, crop, and pest data at a level of intensity not economically feasible with manual sampling and laboratory methods. Fields are highly heterogeneous. Increased sampling will result in accurate characterization of within-field variability. Improvements to VRT and crop modeling are expected to advance rapidly with a higher spatial density of measured soil and crop parameters. Sensors are needed that are fast, efficient, and can assess factors important to crop production.

Moran et al. (1996) concluded that the information from ground-based sensors is needed for soil organic matter, soil moisture, cation exchange capacity, nitrate nitrogen, compaction, soil texture, salinity level, weed detection, and crop residue coverage. These parameters as well as soil pH, and availability of phosphorus and potassium cannot be ascertained by remote-sensing technology. Moreover, the use of real-time ground-based sensors provides the grower control over timing of data acquisition not possible with satellite or aircraft sensing techniques.

Sensors have been developed or are under way to measure soil and crop conditions including soil organic matter, soil moisture content, electrical conductivity, soil nutrient level, and crop and weed reflectance (Sudduth et al., 1997). Continuous, real-time electrochemical soil chemical constituent sensors are currently available for nitrate measurement and are dedicated to specific application in corn side-dress applications. A real-time acoustic soil texture sensor and a real-time soil compaction tester are also under development at Purdue University (Liu et al., 1993; Morgan and Ess, 1996).

Some important real-time indexes may be determined by their relationships to other variables rather than by direct determination. Soil conductivity is appropriate for concurrent real-time assays of salinity, soil moisture, organic matter, cation exchange capacity, soil type and soil texture. Recently, this work was extended

to non-saline soil methods in combination with electrochemical constituent sensing which separates components of direct contact conductivity (Colburn, 1997). Conductivity component analysis is employed for georeferenced data gathering and analysis by several commercial companies as well as for VRT in midwest crops. Apparent soil conductivity using electromagnetic methods is an indicator of clay content, depth to claypan, soil water content, hydraulic characteristics, productivity (Kitchen et al., 1996), and as a promising substitute for yield monitoring (Jaynes et al., 1995).

For immobile constituents (i.e., phosphorus and potassium), industry has not yet chosen to introduce real-time sensors. In some cases, phosphorus and potassium levels in the corn belt states where VRT was first used, are very high, and field availability has been found to exceed producer needs for the current crop year and the near future. In other regions, such as western states, lower availability of immobile nutrients is common. For these nutrients, discontinuous nutrient sensor mapping methods have the potential for gathering and analyzing soil samples in separate field operations. Three systems are under development by government and academia which automatically extract and analyze soil samples for phosphorus, potassium, and nitrates (Adsett and Zoerb, 1991; Birrell, 1995; Morgan and Ess, 1996).

There exists the potential for a vast increase in the timeliness and amount of information if additional means of data collection and analysis become available. Sensors will play an important role in supporting technology for precise applications of nutrients, pesticides, and other inputs. Only a few commercial sensors are available today. Efforts continue by both private companies and the public sector to develop real-time sensors for additional agricultural indexes. Basic research in the sensors arena is fundamental to an improved understanding of the variations in site-specific crop production in a wide variety of regional production systems.

Remote sensing—the acquisition of information from remote locations such as an airplane or satellite—is a potentially important source of data for precision agriculture. In the long term, remote sensing could provide numerous forms of information, both spatially and temporally. However, improvements are needed in the analytical products and delivery systems if remote sensing is to meet its promise for precision agriculture.

For more than 30 years remote sensing has been envisioned as a valuable source of information for crop management. The pioneering research of Colwell (1956) showed that infrared aerial photography could be used to detect loss of vigor of wheat and other small grains resulting from disease. Although much research and development was directed at large-area crop inventory applications of satellite data in the 1970s (MacDonald and Hall, 1980), much less attention

has been directed at crop management applications. Satellite data have not had spatial resolution, temporal frequency, and delivery times sufficient for the needs of production agriculture. In addition, supporting technologies and infrastructure have not been available. Nevertheless, the understanding of crop spectral and radiometric relationships gained from past research is relevant to crop management applications (Bauer, 1985).

Jackson (1984) described the potential for remote sensing in crop management, and stressed that it is critical to provide frequent coverage, rapid data delivery, spatial resolution of 5 to 20 meters, and integration with agronomic and

meteorological data into expert systems. These points were reiterated by Moran et al. (1997) in a recent review of the potential of remote sensing to acquire information for identifying and analyzing site-soil spatial and temporal variability within fields.

In the past 10 years there have been rapid advances in acquiring and processing multispectral imagery with multispectral video by using digital cameras from aircraft. This approach has the flexibility of aerial photography acquisition and the advantage of digital multispectral imagery (Moran et al., 1996; Pearson et al., 1994). Although most planning and effort are going into the development of satellite systems, aircraft-acquired imagery may continue to be needed when extremely high resolution imagery is required. Aircraft platforms also provide an opportunity for developing and testing new sensors (i.e., thermal infrared and hyperspectral sensors) for future satellite systems.

A sequence of remotely sensed images over time can provide information about crop growth and the spatial variation within fields. Detailed spatially distributed multitemporal information, in visual form, is not readily obtainable from conventional crop management systems or from site-specific crop management methods. Remotely sensed images (i.e., color infrared aerial photographs or multispectral images acquired from satellites or airplanes) show spatial and spectral variation resulting from soil and crop characteristics. These images show the state or condition of fields when the images were acquired. One of the most useful aspects of remote sensing is its ability to generate images showing the spatial variation in fields caused by natural and cultural factors. This information is not limited by sampling interval or geostatistical interpolations (Moran et al., 1997). Images acquired at different times during a season can be used to determine changes such as growth rates and condition. These data, in turn, can be compared with data from previous years and may be helpful in predicting yield.

Commercial interest is growing in the potential of remote sensing to contribute to site-specific crop management, particularly as precision agriculture techniques are being developed and the possibility of routine, frequent acquisition of remote sensing data by satellites seems likely. Several earth-observing satellites are scheduled for launch over the next decade by governments and private industries. By 2005, 40 or more land observation satellites are expected be available (Stoney, 1996). Many of these satellites will acquire imagery with spatial resolutions ranging from 1-3 meters for panchromatic images to 3-15 meters for multispectral imagery. Others will have resolutions of 10-30 meters but with additional spectral bands, including thermal infrared on LANDSAT-7. Still other systems will collect radar data at varying resolutions. These sensors have promise for many types of measurements beyond identifying crop type, including monitoring crop stresses and condition, soil properties, and moisture. A major research challenge is the development of robust image analysis methods for agriculture, and a major educational need is training satellite data providers to meet agriculture needs.

A broad range of spatially explicit crop response models is needed to evaluate the efficacy of precision agriculture methods and provide the basis for precise recommendations. Many models for predicting how crops respond to climate, nutrients, water, light, and other conditions already exist, yet most of these do not include a spatial component appropriate to precision agriculture applications (Sadler and Russell, 1997). GIS can provide the means to run the model continuously across an extensive area using data that reflect continually varying conditions. Time series and other temporal analyses can aid in predicting final crop yield. Current models may be extended to account for spatial effects, such as edge effects along field boundaries. In the ecological and biometeorological literature, however, several spatially explicit models have been developed to predict hourly, daily, and annual rates of evapotranspiration and photosynthesis, and several spatially distributed hydrologic models predict surface and subsurface flows. Mesoscale climate models can resolve cells as small as 5 to 10 kilometers for predicting weather conditions.

Pests are not dispersed evenly throughout the environment. To the extent that the factors influencing their spatial distribution are understood, their dispersion and potential for damage can be modeled. GIS can be used for spatially variable data for these factors. As with crop response models, a distinct pest model can be run continuously across a landscape, using GIS to input data to the model and display results (loosely coupled model), or a spatially explicit model can be created within the GIS software (tightly coupled model). GIS can provide the basis for multiscalar effects, for example, incorporating results of a regional pest pressure model into a system for generating within-field recommendations based on locally variable conditions.

A crop growth model could be used as a decision aid for determining different yields based on varying plant populations, which could help a producer decide when to plant or replant areas within a field based on plant population data and risk factors for various soil types. Having to make a decision to replant a field that is in a questionable condition is perhaps the hardest decision a producer faces. Any information to aid such decisions and reduce risk would be valuable.

In many crop production areas, landscape factors can cause dramatic variations in yield. Landscape elements affect many properties relevant to plant growth, including soil texture, soil organic matter, and temperature. Landscape morphology affects soil moisture available to crops by its influence on drainage and catchment area. Soil surveys typically do not have sufficient resolution to capture this variability in enough detail to support precision recommendations; even field-based sampling on a regular grid may miss relevant soil-landscape features. Stratifying sampling density on the basis of landscape features may be more cost effective and informative than a simple grid. GIS allow users to create and manage digital elevation or digital terrain models created by photogrammetric methods

(analysis of stereo pairs of aerial photographs) with new techniques using interoferometric radar or by continuous three-dimensional coordinate measurements with in-field equipment. Precise recommendations can be made to the extent that the relationships are understood between soil properties and surface morphology (i.e., slope, slope length, aspect, curvature, landscape position, catchment area, and drainage) derived from digital elevation or digital terrain models.

Crop models do not offer a panacea for problem solving; they are limited in their ability to simulate various parts of a biological system. Most of the crop and pest models available or developed to date were not designed to be used for managing spatial and temporal variation. It is not clear whether a predictive model, an explanatory model, or a hybrid approach will be more appropriate for precision agriculture. Alternatively, data mining and other techniques may be used to extract valuable information from large amounts of stored data. However, crop modeling is currently an important tool for gaining a theoretical understanding of a crop production system.

Decision support systems (DSS) are used in agriculture for tactical, strategic, and policy-level decision support. Because producers are continually faced with making tactical decisions, such tools are becoming increasingly useful on the farm. However, few DSS are in general use by agricultural producers today, in part due to difficulty in use and limited information provided—from their point of view. They have been used to aid in decisions that are complicated by large amounts of information and data. A simple conceptual diagram of a DSS is shown in Figure 1-2 (Petersen et al., 1993). Data collected by a consultant, obtained through a weather forecasting service, or acquired through a sensing operation are analyzed and linked with appropriate decision rules that identify actions to assist in producer decision making.

DSS rules are not developed to make a single recommendation but rather to provide decision makers with choices; decision support systems should be seen as sources of valuable tactical information. As is the case for crop modeling and current management recommendations, DSS have been developed for whole fields, and subfield variation has been largely ignored. Although subfield tactical decisions have been practiced by producers for many years (i.e., rouging, spot-spraying or rope-wicking residual weeds, or spot-treating chinch bugs in sorghum), most management practices are implemented for whole fields.

The relationship between the scale of an operation and the resolution and variability of sample data used in a DSS is important. To demonstrate this point, consider the appropriateness of using DSS in two sites with widely differing characteristics. The variation in the assessed attribute used in the DSS is high at one site and low at the other. The DSS may be adequate for wholefield decisions at the site with low variability but not appropriate for the site with high variability.

FIGURE 1-2 Conceptual diagram of a decision support system. Tracing the steps in the figure, information can be viewed as flowing from the environment via instrumented or human sensors as data to a database. The information as data is analyzed and manipulated for storage or transmission to a user as part of a decision process. The information processed for a decision results in an action to be executed within the environment. After the action is carried out, the environment is again monitored to begin a new cycle of information flow. Thus, information flows to and from the environment in an endless loop that begins with sensing and ends with action. A DSS integrates expert knowledge, management models, and timely data to assist producers with daily operational and long-range strategic decisions.

SOURCE: Petersen et al., 1993. Reprinted with permission; copyright 1993, Agronomy Society of America, Crop Science Society of America, and Soil Science Society of America.

The site with high variability may require a DSS in which other attributes are assessed or the whole field is subdivided to overcome the variation. Assessing the relationship between attribute variation and DSS performance has been largely ignored in relation to pest management and only superficially addressed regarding soil fertility.

Similarly, decision support systems do not address the problem of spatial heterogeneity. This is true for weed management DSS such as HERB, WeedSOFT, and PC-Plant Protection, and for insect and disease management programs; irrigation and crop selection programs are all whole-field based. Researchers recently combined weed management DSS with spatial weed infestation

maps to determine the value of spatial information in pest management. In these simulations, pest density at individual locations in fields was used for the infestation level input to the DSS. Lindquist et al. (in press) found that a treatment map based on spatial information (800 observations) was a great improvement over use of the mean field population. The simulation indicated that, on average, herbicide use would be reduced by 30 percent to 40 percent with such an approach. Christensen et al. (1996) also found herbicide reductions of 30 percent to 40 percent when they mapped weed populations in several cereal grains in Denmark. In each case spatial data were used to run an economic-threshold-based DSS.

Although such simulations show that subfield management could lead to significant changes in management practices, numerous questions remain unanswered. First, the issue of risk of improper decisions is a real concern to consultants and producers. DSS have only recently begun to be used for many large acreage crops. Their slow adoption has partly resulted from concern over risk of nonperformance. Consultants are providing a service to a client and are concerned that the client be pleased with the outcome of their service, and the producer is concerned about the real agronomic impact of uncontrolled pests and the social implications of infested fields. Another concern is that the long-term effect of spatial management on infestation level and distribution is largely unknown. Seed production by uncontrolled plants and egg or cyst production by insects and nematodes may result in infestations growing or in spatial orientations changing in ways that make GIS maps less valuable. Such concerns require studies to assess these longer-term impacts on precision agriculture.

There is also the question of the extent to which a knowledge base exists for subfield decisions. For example, relatively little is known about the suitability of crop cultivars for specific soil types or cultivar-fertility-pesticide interactions. Little is known about the interactions between agronomic practices and their environment at the subfield scale. A solid knowledge base will become more important as a foundation for more information-intensive practices. Additionally, as the complexity of databases in DSS grows, the inputs needed to initiate these applications will also grow. For example, two years ago, the University of Nebraska released a weed management DSS that required little information on soil type. In the most recent release, the user can determine the potential risk to ground and surface water contamination from pesticide use, but the user must be familiar with the specific soil type in that field. Also program developers will be challenged to make these decision aids easy to use. In the example, county soil maps are being incorporated in the new version of WeedSOFT; the user will find the field on the county map and click on the location and the DSS will do the rest.

To develop the needed database, researchers will need to approach parameterization used to aid decision making in a new way. Rather than restricting data collection to a handful of research station field trials, researchers will have to find a way to use producers' fields as laboratories. Harnessing spatially referenced

data collected on individual farmsteads makes it possible to set parameters for data sets within localized areas. Such an approach would allow DSS to incorporate local parameters, which has not been possible due to the cost of parameterization and of programming expertise. It is likely that future development and maintenance of decision support systems will be accomplished through land grant, Agricultural Research Service, consultant, producer, and other information service provider consortia.

Information technologies have the potential to provide considerable amounts of useful information for decision making in precision agriculture. A suite of tools will be used to assess and manage agronomic factors important to crop production. For these new tools to function properly, however, they will need to be user friendly for producers and consultants. Information technologies will produce enormous data sets on crops and their interactions with their environment. The challenge remains as to how to convert these data into useful suggestions to aid in the decision-making process for the producer.