Digital Transformation of the Agricultural Sector: Pathways, Drivers and Policy Implications
Madhu Khanna is a distinguished professor of environmental economics at the Department of Agricultural and Consumer Economics at the University of Illinois at Urbana-Champaign.
Editor-in-charge: Craig Gundersen
Abstract
The digital transformation of agriculture is enabling the collection of vast amounts of geo-referenced information about growing conditions within the field and facilitating the automated implementation of spatially varying input applications. This has the potential to increase production efficiency, reduce overapplication of inputs, lower input waste and pollution, and improve farm profitability. We discuss the pathways by which digital agricultural technologies have the potential to affect crop management, and current trends and patterns of adoption of digital technologies. We provide insights from the technology adoption literature on the factors that can be expected to influence the adoption of emerging digital technologies and the findings from the empirical literature on the determinants of adoption of these types of technologies. We conclude with a discussion of the design of policy incentives to induce the adoption of digital technologies cost-effectively, the challenges in implementing such policies and the opportunity that digital technologies offer for addressing these challenges.
There is increasing use of digital technologies in agriculture, including information technologies, computational decision and analytical tools, and automation of farm machinery. These technologies include a range of component technologies to facilitate digital communication, such as the cloud, sensors, and high-speed networks. They also include several different types of field-based equipment, such as Global Positioning Systems/Global Navigation Satellite Systems (GPS/GNSS), yield monitors, georeferenced soil sampling, remote spectroscopic sensing, and unmanned aerial vehicles (UAVs). GPS/GNSS systems can provide the precise location of fertility levels in a field and combined with drones and yield monitors they allow monitoring and detection of variability in crop health, soil fertility and yields in the field. Recommendations for crop management based on this information can be fed to on-board technologies, such as Variable Rate Technologies (VRT) and other auto-steered and guided equipment, to precisely guide equipment and enable site-specific input placement and application rates while reducing labor and fatigue. These digital technologies enable the collection of vast amounts of geo-referenced information about growing conditions within a field and facilitate automated implementation of spatially varying input applications (van Es and Woodard 2017).
Precision agricultural technologies are not new. Yield monitor, VRT for fertilizer application, and grid soil sampling were commercialized in the mid-1990s and automated guidance was commercialized by the late 1990s (Griffin and Yeager 2018). While precision agricultural technologies have the capability to take in-field variability into account, the recent development of digital technologies is now enabling “smart farming” based not only on location but also on real-time information about weather, climate conditions, soil nutrient needs, and other growing conditions. Additionally, digital technologies have expanded the capacity to aggregate high-resolution data from multiple fields/farms and other sources, including public data, privately held data, and machine and sensor data. With large capacity for cloud storage of data and advanced computational tools for data analytics, it is feasible to aggregate and analyze data on these multiple data layers from many devices and thousands of other farms. The large volume of data that includes both farm-scale information and landscape-level information on weather and other growing conditions can facilitate greater scientific understanding about the agronomics of crop production than has been known so far. This can improve the accuracy with which farmers match the “right seed and the right fertilizer application rate with the right soil at the right time” (Wolfert et al. 2017). Digitally enabled agricultural technologies use electronic information and other digital technologies to gather, process, and analyze spatial and temporal data and combine them with precision agricultural technologies for the purpose of targeted actions (Lowenberg-deBoer and Erickson 2019). We anticipate that these technologies will have the potential to achieve higher production efficiency, lower application of inputs, lower input waste and pollution, and higher farm profitability.
We discuss the pathways by which digital agricultural technologies have the potential to affect crop management in Section 2. Following that we describe the current trends and pattern of adoption of digital technologies in Section 3. Section 4 discusses the existing evidence on the profitability of digital agricultural technologies and their effects on the environment. The focus here is largely on the application of these technologies in developed countries. These technologies are also promising in developing countries with smallholder farmers as described in Birner, Daum, and Pray (2020). We provide insights from the technology adoption literature on the factors that can be expected to influence the adoption of emerging digital technologies in Section 5. Section 6 describes the findings from the empirical literature on the determinants of adoption of these technologies. Although these technologies have the potential to increase farm profits and reduce environmental impacts, farmer self-interest may not be sufficient to induce adoption at the scale needed to improve environmental outcomes. Section 7 discusses the design of policy incentives to induce the adoption of digital technologies cost-effectively, the challenges in implementing such policies and the opportunity that digital technologies offer for addressing these challenges. The paper concludes with a discussion of the directions for future research on the economic, environmental and policy implications of digital agricultural technologies.
Pathways for Impact of Digital Agricultural Technologies on Crop Management
Digital technologies include several components and are not available as an integrated off-the shelf package. Instead, producers have the flexibility to assemble components that address the management issues in their specific production systems and that are compatible with their farming operations. These components vary in the type of investment they require (see Birner, Daum, and Pray 2020 for more details about these components). These investments include capital investments in equipment (e.g., computers, robotic systems, VRT, sensors), service investments that provide information service (e.g., remote sensing, cloud-based decision models) and knowledge and human capital investments (localized actionable knowledge for input application). The benefits and costs of adoption differ across these components and also differ with farm and farmer characteristics (van Es and Woodard 2017).
Some of these components for precision farming, such as yield monitors, GNSS, and VRT, have been available for about three decades, but the agronomic knowledge needed to respond to observed heterogeneity in growing conditions in the field and to develop recommendations for varying input applications has lagged behind the engineering capabilities of technology to manage fields more flexibly (van Es and Woodard 2017). Decisions about variable input application rates were based on “small data” from a farmer's own operation and rules of thumb due to little understanding of how best to respond to spatial variations in soil nutrient availability. Zhang et al. (2002) noted several additional limitations of previous precision technologies that limited their adoption, including ability of farmers to analyze and use the vast amount of data generated by these technologies for management decisions, lack of evidence of the benefits of precision farming, skilled labor intensity and high costs of adoption. Digitally enabled precision technologies have the potential to affect farm management in several ways that are discussed below. A key benefit of several component technologies arises from their potential to increase input-use efficiency and to jointly reduce two main sources of uncertainty that affect farming operations—variations in soil conditions and topography and the weather. These technologies can also affect labor requirements and the type of labor (skilled, unskilled) required as well as the capital equipment requirement and associated capital costs.
Respond to Spatial Heterogenity in the Field
Conventional methods of farm management have been limited by technology to understand in-field variability and to manage crop production systems accordingly and have relied on uniform rates of application of inputs such as fertilizers and pesticides based on average field and weather conditions and generic crop features. Application rates and timing (for example, for irrigation) are not synchronized with weather dependent crop needs. Additionally, uncertainty about the weather as well as about soil nitrogen levels can lead farmers to apply “a little extra fertilizer” in order to prevent nutritional deficiencies that may limit yield in a year with good weather (Babcock, 1992). Application of inputs at a uniform rate across a field also results in some areas having insufficient nutrients while others having too much, leading to crop productivity being lower than its potential in some areas and discharges of excess nutrients from other areas (Zilberman, Khanna, and Lipper 1997). Brandes et al. (2016) show that even high-yielding fields include areas with low productivity and poor soil quality and that in some areas of a field farmers may lose money by incurring planting and input application costs. These areas that are unprofitable to plant are also likely to coincide with the most environmentally risky areas in the field.
Isolating the effects of subfield heterogeneity in growing conditions on field-level economic and environmental outcomes and determining optimal management responses to this heterogeneity is challenging due to temporal variations in weather, diversity of management practices, and crop genetics. Managing this heterogeneity requires information about site-specific variability in growing conditions, knowledge about how to respond effectively to that variability and technologies that implement spatially varying management decisions.
Digital agricultural technologies are able to gather, process, store, and analyze data from millions of acres on crop genetics (G) planted, environmental (E) conditions, and management (M) practices used and use this together with machine learning algorithms to determine the combinations of GxExM that can enhance productivity and profitability of crop production while reducing adverse environmental impacts. The implementation of this approach requires “big data” from millions of acres of cropland to provide the diversity in G, E, and M to develop agronomic knowledge to respond to subfield heterogeneity. Other digital technologies can reduce the need for manual or mechanical weeding or imprecise chemical applications by using teams of semiautonomous robots equipped with artificial intelligence that can scout for pests and weeds below the plant canopy and target them (McAllister et al. 2019). Digital technologies can thereby unlock the potential of precision technologies to respond to spatial heterogeneity in field conditions.
Address Temporal Variability in Crop Management
Input requirements, harvesting decisions, and other crop management decision choices depend not only on field conditions but also on weather information, soil moisture, and other growing conditions that can vary with time. By combining information from GNSS technologies with precise meteorological information and informing farmers not only how much of the inputs to apply but when to apply them can increase the efficiency of input use. With these technologies, for example, irrigation systems can be controlled remotely and be used to apply water at varying intervals in different parts of the field to improve irrigation efficiency and conserve water while preventing nitrogen loss from leaching. Web-based irrigation scheduling systems use interactive computer models together with high-frequency information from soil moisture sensors placed at various depths in the soil and on weather conditions to synthesize data on soil type, local weather conditions, plant growth stage to guide the timing and quantity of irrigation. This, combined with variable rate irrigation technology, can be used to apply water at different rates based on soil type, soil and plant moisture levels, and weather conditions (Sunding, Rogers, and Bazelon 2016). Similarly, the timing and rate of nitrogen application can be varied more precisely based on information about weather conditions.
Increase Input Productivity
Targeted and site-specific management of agricultural production can be expected to increase yields, improve input use efficiency and reduce input waste. These technologies have been found to increase nitrogen productivity (Khanna 2001) and irrigation water productivity (Khanna and Zilberman 1997; Sunding, Rogers, and Bazelon 2016). By providing timely and precise weather data, they can also reduce the uncertainty about farm management decisions and the variability and risks associated with agricultural production activities. For example, yield monitors can help farmers identify areas in the field where it may be profitable to change practices as well as subfield zones where it may be unprofitable to cultivate crops. Technologies such as automated guidance and automated section control can improve efficiency and lower production costs. Griffin, Shockley, and Mark (2018) note that with GNSS guided equipment the farmer can cover the field area with greater accuracy and speed and will less overlap, thereby reducing the overall efficiency of use of farm equipment so that they can till, plant, spray and harvest more acres in the same amount of time. Large rectangular fields are found to have a larger benefit from guidance systems while smaller, irregularly shaped fields have a larger benefit from section control and row shut-off technologies. GNSS based automatic guidance systems also enable working at night and in bad weather and reduce inputs needed such as fuel, fertilizer, and pesticides. Soil moisture sensors connected to cellular or satellite modems provide high-frequency moisture readings. These can be aggregated with other big data about soil types and weather across millions of acres using cloud-based data centers to inform farmers about irrigation needs and prevent overwatering, thereby reducing costs (Sunding, Rogers, and Bazelon 2016). The effect on input costs of a wide range of precision technologies, based on a survey of farmers in the US, shows that input costs were lower in general with the adoption of precision technologies, though to varying extents (Schimmelpfennig 2016).
Affect Labor and Capital Requirements
These technologies can affect farm finances through impacts on operating costs, overhead costs, and changes in revenue. Adoption can increase capital and overhead costs because these technologies involve purchases of equipment, installation charges, and the time and effort spent learning how to use and maintain these technologies. Adoption of precision technologies can be expected to lead to greater expenditures on machinery and equipment as these technologies are capital intensive. Machinery also has a higher expense base (compared to labor costs) and more potential to influence overhead costs. These costs are usually not recoverable if use is discontinued. These technologies are also highly specialized compared to other capital equipment like land and tractors. While outsourcing technology services to a custom service provider is another option, it also imposes costs. These factors increase the financial risks of adopting digital technologies. The high startup costs with a risk of insufficient return on investment can make it challenging for producers to afford these technologies. The high upfront investments needed can come with the risk of locking a farmer in with a technology provider or vendor as they pay off the debts incurred. Consolidation among technology providers and their growing monopoly power can reduce bargaining power among farmers and expose them to unfair practices and dependency.
On the other hand, these technologies could reduce per unit costs of agricultural production. The increase in costs through increased use of inputs (if a need is indicated by mapping) can be more than offset by increase in yields. Simulation models also show that the yield and revenue gains from adopting soil testing and VRT for nitrogen application increase as the average quality of the soil increases and as the variability in soil fertility increases. The implications of adopting these technologies for input costs will depend on the distribution of soil fertility and whether higher expenditures due to higher applications on low-soil-fertility areas of the field are more or less than fully compensated by lower expenditures on more fertile areas since fertilizer applications are bounded from below by zero (Khanna, Isik, and Winter-Nelson 2000).
Adoption of these technologies can be expected to affect demand for labor; these effects will differ for hired labor versus unpaid labor and for skilled labor versus unskilled labor. Spending on technically skilled labor may increase while that on unskilled and unpaid labor may decrease, depending on the extent to which these technologies require specialized information management and field operation skills but save on labor hours. Technologies such as auto-steering and guidance can reduce labor costs and fatigue while increasing accuracy of crop input placement and management. Small, intelligent robots (or agbots) can operate at smaller scales and make targeted input applications, thus reducing labor costs.
The various components of digital agricultural technologies vary in the extent to which they impose upfront costs—including capital cost, human capital costs, and learning costs—and require detailed farm-specific knowledge compared to conventional farming technologies. Incentives for adopting various components can vary with farm size, level of farmer education, availability of enabling infrastructure, and labor availability. Since these technologies are still evolving rapidly there is considerable uncertainty about their benefits. Capital and learning costs can be expected to decrease over time and the gains from implementation can be expected to increase over time due to technological innovation and improvement. Incentives to adopt these technologies are, therefore, likely to grow over time. We now discuss the types of digital technologies that are being adopted and the trend in adoption rates.
Current Trends and Pattern of Adoption of Digital Technologies
Several studies have sought to compile information on adoption of digital and precision technologies across countries over time using varying methods, such as surveys, equipment inventories, percentage of farms, or percentage of land area using a technology and survey of agricultural dealers. These studies differ in the extent to which the areas and samples studied are random or representative of the population. They also differ in technology definitions and questions asked. Among these there are very few studies examining adoption of these technologies in developing countries. As a result, it is only possible to obtain a snapshot of adoption choices across countries at various points in time (for a recent survey of this literature see Lowenberg-deBoer and Erickson 2019).
These data show that adoption of mobile phones and the Internet has spread rapidly even in the developing countries; all regions are converging in access to mobile phone but South Asia and Sub-Saharan Africa are falling behind in Internet access. More than 40% of the global population has Internet access and 70% of the population even in the poorest 20% of developing countries have mobile phones (Deichmann, Goyal, and Mishra 2016). Studies examining the adoption of digital technologies in developing countries find that 80% of farmers in selected Asian countries owned a cell phone and 50% of these farmers were making crop sale arrangements over the phone. There is also increasing use of electronic extension services and social media in developing countries to share planting advice, weather updates, early disaster warnings, and pest outbreaks (Deichmann, Goyal, and Mishra 2016).
In the case of field-based precision technologies, studies show that they are not adopted as a single package; instead farmers decide on the bundle of technologies to adopt. Adoption rates vary across components, crops, and countries. Among the various technologies, rate of adoption of GNSS was highest in the US, UK, Australia, and Denmark, with rates as high as 60% in the US and 77% in Australia. On the other hand, the rate of adoption of VRT ranged from 20%–30% in these countries and was only 7% in Denmark. Barnes et al. (2019) analyzed the rate of adoption of precision technologies, machine guidance and VRT across five countries in the European Union. They also found that adoption rates were substantially higher for machine guidance technologies than for VRT in all countries. Adoption rates of machine guided technologies ranged from about 30% to 50% across Netherlands, Germany, UK, Greece and Belgium. Rates of adoption were higher for larger farms compared to smaller farms. Among other countries, rates of adoption are also high in Argentina and Brazil, particularly of GNSS guidance and GNSS linked soil test mapping. Data on adoption of precision agricultural technologies in Asia and Africa are scarce; the technology is now being offered in these countries and GNSS guidance is being used by large farmers but mainly for mapping of field boundaries.
In the US, there is consistent evidence that adoption rates of various components of precision technologies have grown over time. Findings from national surveys of farmers show that the share of planted acres adopting a precision technology in the US has been growing but the rates of growth vary across technologies and across crops (Schimmelpfennig 2016). Data show that the adoption of diagnostic tools, like yield monitors, and guidance systems has grown faster than that of applicative technologies like VRT. Yield mapping on corn and soybean crops grew from just under 10% of planted acres in 2001 to 30% or higher in 2012. Adoption of guidance systems grew to being used on 45%–50% of acres planted under major field crops. VRT that requires larger capital investment and human capital costs has not shown the consistent increases in adoption that guidance has, but by 2012, it had been adopted on 20% of acres planted under corn, soybeans, and rice. This is, however, higher than rates observed previously; less than 10% of the planted acres under major field crops had adopted VRT for fertilizer, seed and pesticide application by 2002 (Lambert et al. 2004). Other studies find similar findings. Say et al. (2017) report that yield monitors and GPS guidance systems were more widely adopted compared to VRT for fertilizers in the US. Lowenberg-DeBoer and Erickson (2019) report that GNSS guidance was adopted by 59% and yield monitors by 68% of maize farmers while VRT was adopted by 29% of maize farmers. A survey of farmers in New York state suggests that adoption rates for precision agriculture technologies are expected to increase with at least a doubling or tripling of use relative to current adoption rates (but remain less than 100%) in the next five years (van Es et al. 2016). Schimmelpfennig (2016) also reports that precision agricultural technologies were adopted in different combinations. GPS technology and guidance systems were more often adopted alone than in combination with other technologies whereas VRT was adopted more often in combination with GPS and guidance systems than alone. In 2010, only 3.8% of corn farms had adopted all three technologies.
A survey of these dealers in twenty-nine states in the US in 2017 shows that dealers were increasingly offering satellite/aerial imagery, field mapping, and GPS technologies at relatively faster rates over time. The most popular technologies offered by dealers were GPS guidance with autocontrol/autosteer and GPS enabled sprayer boom. Dealers were also offering increasingly offering VRT services with 70% of them offering VRT for single nutrient fertilizer application and about 65%–70% offering VRT for spreading lime, pesticide, and seeds. The survey also shows that while overall use of precision technologies by farmers has grown over the last two decades, it has plateaued in recent years for variable rate technologies (Erickson, Lowenberg-DeBoer, and Bradford 2017).
Effects of Digital Technologies on Profitability and Pollution
The section above described the potential for digital agricultural technologies to improve decision-making about input applications and crop management and thereby increase the productivity of land and input-use efficiency while reducing input waste and pollution. We now present some evidence on the effects of adoption on farm profitability and environmental impacts. It should be noted that most of these studies are examining the earlier generation of precision technologies and there is limited to no publicly available evidence of the profitability of big-data-enabled precision farming. This is because much of the big data being generated at the farm scale is in private hands and not readily available to researchers or policy makers due to concerns about privacy and data ownership. These studies also differ in their consideration of the time period and discount rate used and their inclusion of capital costs, input costs and crop yield while estimating net benefits. They are also case-specific and do not provide generalizable evidence on the conditions under which digital technologies are likely to be profitable or reduce environmental impacts of crop production.
Effect on Profits
The impact of digital technologies on farm profitability will depend on various factors, including the costs of adoption, the gains in yield, and the level of input cost reduction. Evidence from the adoption of current precision technologies indicates that many of them do increase yield, input efficiency and profitability (OECD 2016). Survey data show that adoption of yield monitors, VRT, or GNSS, was associated with higher corn and soybean yields. A review of early studies examining the impact of precision agriculture on profitability found mixed results with only 68% of cases showing an increase in profits (European Parliament 2014). Griffin and Lowenberg-DeBoer (2005) reviewed 234 articles that discussed economic returns from precision agricultural technologies and found that 73% of those focusing on corn reported net benefits from using these technologies; corresponding figures were 100% for soybean, 75% for potato, and 52% for wheat. Adoption of soil sensors and mapping for corn production is associated with positive net benefits 33% of the time. Net benefits are positive 50% of the time for corn farms using GPS guidance and 33% of the farms with yield monitors and VRT. Studies examining the impact of VRT for corn production find mixed results; it was profitable only on corn fields with sufficient variability in fertility. In general, studies find mixed evidence of the effect of VRT for applying a range of inputs, including nitrogen, seeding, pests and irrigation, on profits. Schimmelpfennig (2016) found that the net returns of corn farms that adopted at least one of the precision technologies examined increased by 1% to 2% compared to corn farms that did not use the technology. Net returns are higher with the adoption of mapping technologies with lower capital costs than for VRT with higher capital costs. Others note that it may take several years to realize agronomic and economic benefits from adopting technologies such as yield monitors, grid soil sampling, and variable rate application (Griffin and Miller 2016).
Precision technology dealerships were surveyed about the profitability of the services they offer by Erickson and Widmar (2015). Overall 60% of dealers reported generating a profit from their precision agriculture offerings as compared to 76% that report generating a profit from nonprecision custom application. About 75% of dealers reported generating a profit from VRT application of inputs and 62% reported a profit from soil sampling services. A large majority of dealers found that UAV devices, yield monitor data analysis, and VRT seeding prescriptions were not generating a profit for them. This could affect pricing and willingness to supply various components of precision technology services in the future and have an impact on farmer willingness to adopt them.
Effects on the Environment
Inputs applied for agricultural production are often not taken up entirely by plants; instead only a portion of the applied input is effectively used. The portion of input that is effectively used may vary across fields, farms, and locations due to heterogeneity in soil conditions, production practices, and technology. The effects of digital agricultural technologies on the environment will depend on the component adopted, efficacy of the technology in meeting crop needs and heterogeneous characteristics of the location that determine the impacts of overapplication of inputs. Digital technologies, such as variable rate applicators, can increase the effectiveness with which the applied inputs are taken up by plants, by tailoring the rate and timing of input applications, crop varieties, and management practices to the growing conditions. These technologies can avoid overapplication of inputs and this can reduce the excess amount of applied input available for leaching or run-off (Khanna and Zilberman 1997). Accurate information about the quantity of inputs needed to maximize yields or profits can increase land productivity and input productivity, minimize input waste and reduce adverse impacts on the environment. Muth (2014) shows that areas in the field that are less profitable are also likely to be more environmentally risky and that profitability and environmental performance are not competing. He shows that precision technologies can be used to make field management decisions based on return on investment and that this can increase profitability and environmental performance simultaneously.
By increasing the efficiency of use of natural resources, such as water and land, these technologies have the potential to lead to less pollution per unit input or per unit output. However, since these technologies may increase total resource use and/or yield per acre, their impact on total pollution is ambiguous. Moreover, the effect on pollution can be expected to differ across fields, farms, or spatial locations. Khanna, Isik, and Zilberman (2002) show that efficiency-enhancing technologies may increase or decrease input use and pollution depending on the extent to which these technologies increase the effectiveness of input use, the impact of effective input use on crop yield and the pollution reducing effect of the technology.
Empirical evidence on the environmental effect of these technologies is not well known and is often based on experimental data or model predictions. A few studies provide causal implications based on observed data. For example, Khanna (2001) estimates the effects of adopting soil testing and VRT for nitrogen application on yield per unit nitrogen in the Midwestern US. The study finds that gains in nitrogen productivity were higher on relatively lower-quality soils and insignificant on farms with above average soil quality. For farmers that adopted both technologies, adoption of VRT led to the larger increase in nitrogen productivity (33% and 18% on below average and above average soil qualities) compared to the gains due to soil testing (6% and 7%, respectively). Similarly, Rejesus and Hornbaker (1999) show that VRT application of nitrogen and improved timing of fertilizer application can reduce nitrate pollution from corn-soybean rotations in Illinois. Other case studies show that VRT nitrogen application can decrease nitrogen application and nitrous oxide emission (see review in Finger et al. 2019).
Precision technologies can also reduce greenhouse gas emissions. Machine guidance and controlled traffic farming can reduce fuel consumption due to less overlap in farm operations. Guidance systems have been found to cause a reduction in fuel use with the reductions being larger for large-scale fields. They also provide other cobenefits, including reductions in soil compaction, runoff, and erosion as well as reductions in indirect energy consumption by applying inputs such as fertilizer, seeds, and pesticides more efficiently. These technologies have also been shown to reduce pesticide use. Studies show that precision application of herbicides, sensor-based precision control of aphids compare to uniform spraying, and use of VRT for pesticide application can substantively reduce chemical use (Dammer and Adamek 2012; Balafoutis et al. 2017; Kempenaar et al. 2018). Similarly, variable rate irrigation has been found to increase water use efficiency and reduce water use, depending on soil and weather conditions (see review in Finger et al. 2019; Balafoutis et al. 2017).
We now discuss the factors that are likely to affect the decision to adopt these emerging digitally-enabled precision technologies, taking into account the characteristics of these technologies, the mechanisms by which they impact crop management and their potential impact on farm profits and the environment.
Determinants of Adoption of Emerging Digital Agricultural Technologies
There is a large literature analyzing incentives for adoption of new technologies and the economic and noneconomic factors that explain adoption by some farmers and not others (Zilberman et al. 2014; Liu, Bruins, and Heberling 2017; Miao and Khanna 2020). Adoption decisions by economically rational farmers are expected to depend on the costs and benefits of adoption; these differ across farmers and locations and can explain heterogeneity in adoption behavior. The relevant costs and benefits will depend on the objectives of the landowner, including maximizing profits or expected utility or minimizing expected loss at a given point in time or over time taking the dynamics of returns and costs of adoption into consideration relative to the status quo. Since in many cases, new technologies impose upfront costs of equipment or learning and yield returns over time, it is important to assess costs and benefits over time and analyze the present value of the net benefits of adoption. We describe the insights provided by this literature and apply them to discuss the considerations that are likely to affect adoption of emerging digitally-enabled agricultural technologies. The adoption of emerging digitally enabled precision technologies will depend on various factors, including the decision-making criteria used by farmers to make technology choices, the characteristics of these technologies, the heterogeneous characteristics of the farm, and the behavioral preferences of the farmer, as well as the availability of complementary technologies.
Effects of Profitability
The threshold model introduced by David (1975) explains the decision to adopt or not to adopt based on profitability, which depends on farm or farmer characteristics such as farm size, soil quality, risk perceptions, etc. that vary across landowners. It shows that a threshold level of one or more of these characteristics separates the adopters from the nonadopters that are heterogeneous in these characteristics. Economic theory suggests that a risk-neutral farmer will adopt a technology if its net present value is positive; this will depend on the change in revenue, the change in input costs, and the change in fixed costs. This model suggests that the likelihood of adoption increases as the discount rate and initial investment for the farmer become lower, the planning horizon, the price of output (if the technology increases yield), the price of input (if the technology saves input), and any pollution penalty (if the technology reduces pollution) become higher (Khanna, Isik, and Zilberman 2002). An implication of this model is that larger farms are more likely to adopt an indivisible technology and that larger initial investments as well as higher discount rates increase the critical farm size for adoption.
Effects of Risk and Uncertainty
There is a large literature showing that risk considerations affect the technology adoption decision by applying the expected utility model, which assumes that farmers are aware of the riskiness of different technologies and account for it in calculating expected benefits (Just and Zilberman 1983). This model has also been used to analyze adoption choices as an optimal land portfolio management decision rather than a discrete choice of whether to adopt any particular technology. Applications are based on the assumption that farmers value higher profits but associate negative benefits to the riskiness (frequently measured by variance) of those profits. Digital agricultural technologies can reduce the risks and uncertainties of farming by providing precise and geocoded information about production conditions on the field that affect crop management decisions. However, high capital and learning costs of adoption and uncertainty about how to specifically respond to information about growing conditions can also increase the risks of farm operations. Isik and Khanna (2003) apply an expected utility maximizing decision criteria to examine how uncertainty about soil conditions and yield can lead risk-averse farmers to overapply fertilizers even with precision technologies. Uncertainty about the accuracy of precision technologies and risk preferences of farmers can reduce the benefits of adopting these technologies for farmers and the environment.
Additionally, uncertainty about the benefits and/or costs of adopting a technology that involves large sunk costs can affect decisions about the timing of adoption. The real option approach developed by Dixit, Pindyck, and Davis (1994) views adoption as a dynamic investment decision not only about whether or not to adopt a technology but when to adopt; if there is uncertainty in the properties of the technologies or the behavior of prices in the future, there may be gains (option-value) from taking advantage of waiting until uncertainties are clarified.
Technological Factors
Type of Technology
The potential uses of digital agricultural technologies vary across components and this can affect the choice among components and the type of farmers that adopt them. For example, component technologies that are embodied in indivisible equipment (like yield monitors) are likely to be adopted by larger farmers while technologies that are divisible such as grid soil sampling are likely to be scale neutral. Precision technologies that increase input-use efficiency by varying input application rates to meet crop needs are likely to be adopted on fields with higher spatial variability in soil fertility. Technology characteristics such as improving accuracy, reducing input costs, saving time, addressing labor shortages, and improving soil conditions are key reasons for adopting precision technologies (Say et al. 2017).
Precision farming involves a bundle of technologies, such as soil testing, variable rate fertilizer application, and yield monitoring, but not all farmers will adopt all components of the package at once (Khanna 2001). Farmers prefer to customize their adoption decisions to meet their individual needs. Studies show they often prefer to adopt technologies sequentially based on risk considerations, supply constraints, and due to a lack of knowledge about their costs and benefits. Diagnostic tools may be adopted first because they can inform information management decisions about the benefits of precision application for a range of inputs, including fertilizer, herbicides, pesticides, and irrigation. Adoption of yield monitors maybe a necessary first step to build a history of production data that can be overlaid with other variables such as soil types, weather, hybrids and varieties, and other production practices. Applicative technologies may then be adopted more selectively to manage particular management needs. Khanna (2001) found that farmers adopted soil testing for fertilizer requirements and VRT sequentially rather than as a package; many farmers adopted soil testing only to learn whether the spatial variability in soil fertility was large enough to make it beneficial to adopt VRT. They also found that although soil testing was scale neutral, the subsequent adoption of variable-rate fertilizer applicators was more likely by larger, more experienced, and innovative farmers with greater human capital skills.
Information Availability and Learning Costs
New technologies differ in their learning costs and some may involve significant learning and information costs. Individuals tend to learn from the experience of members of the community and adopt the practices used by successful individuals. Given the technical complexity of precision agricultural technologies, farmers are also likely to require advisory and consultancy services that are specialized in data management. Studies indicate that farmers consider agricultural media, informal sources (other farmers), extension, and commercial vendors to be important sources of information (Just and Zilberman 2002) The range of stakeholders involved in providing services, equipment and data analytics can generate competition, challenges to interoperability of devices and equipment and difficulties in obtaining impartial advice. Access to neutral extension services is therefore going to be critical in building trust in the technology, lowering learning costs, and protecting farmer interests. The extent to which these sources can provide information about these technologies may differ across components and influence which components are adopted.
Availability of Complementary Technologies
Enabling technologies are needed to achieve the full utilization of digitally enabled precision technologies (Birner, Daum, and Pray 2020). In addition to computers, access to high-speed internet is critical for wireless data transfer for uploading and downloading data from the field. A dependable, high-speed, and strong cellular connection is essential for adopting precision technologies like auto-steer and VRT and for enabling two-way wireless data transfer between farmers and aggregators. These technologies are increasingly relying on data being transmitted directly to and from equipment cabs via the Internet to enable variable rate prescriptions or to load yield and application data into the cloud. In the absence of proper connectivity farmers need to rely on manual data transfer that is subject to significant delays, potential for loss, and missed opportunities for real-time adjustments of management practices. Internet connectivity and access to computers is not yet universal even in the US.
Other Technology-Related Considerations
As discussed above, efforts at improving the agronomic knowledge about optimal responses to spatial heterogeneity in soil conditions are leading to private sector created products that rely on big data obtained from digital devices deployed on millions of acres to make site-specific recommendations. Farmers purchasing these services need to share their farm data about crop management, input use, yields, and so on with the precision farming service provider. This creates concerns about data ownership, privacy, and confidentiality, and these can create a barrier to adoption in the absence of clarity on these issues. Adoption of digital technologies requires trust in the precision technology service providers to protect data and to provide recommendations that will increase profitability of the farm and reduce its riskiness. It also requires farmers to overcome concerns about data privacy and confidentiality. Landowners that adopt digital technologies are also likely to have less control on their farm operations based on their expertise and previous experience on their farm since these technologies require recommendations for farm management from technology providers based on data from multiple sources and farms.
Farmer Characteristics
Farm Size
Larger farms are more likely to adopt technologies that are either indivisible or have economies of scale. Larger farm size is likely to reduce risk aversion and thus enhance adoption of high-risk, high-return technologies. Larger farm size can also lower per unit costs of inputs, consultants and physical capital and provide the economies of scale needed to make adoption profitable. Existing studies also indicate that the existing soil quality and the extent and pattern of spatial variability in soil quality (proxied by potential crop yields) will influence the magnitude of the gains from adopting precision technologies; specifically the symmetry or skewness of the distribution of nitrogen requirements in the field will influence the portion of the field that is overfertilized and the portion that is underfertilized and the nitrogen fertilizer cost savings with precision application compared to a uniform application based on the average requirement for the field (Isik and Khanna 2003).
Credit Constraints
New technologies typically involve up-front costs and the ability to finance investment is likely to be highly correlated to both a borrower's wealth and capacity to pledge assets as collateral. Availability of credit may be critical to enable adoption of new technologies, particularly by smaller farmers. This is particularly likely to be the case for farmers that are risk averse because investment in new technologies can lead to variability in income over time with low/negative income streams in early years when the investment is made and positive returns in later years (Miao and Khanna 2017).
Behavioral Factors
There is considerable research examining the effects of behavioral factors, such as cognitive, emotional, personal, and social processes that can affect the provision of public goods by decision makers. These studies show that moral and environmental concerns, risk-seeking, and openness to new experiences is associated with higher adoption of sustainable practices. There is also evidence that a farmer's adoption decision is affected by adoption decisions of neighboring farmers and by the opinion of social referents who support adoption. Cognitive factors such as extent of knowledge and competence about new practices and ease of learning also affect the adoption decision (Dessart et al., 2019).
Risk and time preferences of farmers also affect technology choices. Risk-averse farmers are likely to be willing to pay high premiums to adopt technologies that reduce risk for example, due to weather or disease (Zilberman et al. 2014). Previous studies show that the effect of behavioral preferences on adoption behavior depends on the farm, technology, and input characteristics. Isik and Khanna (2003; 2002) show that spatial variability within the field and the risk-increasing/decreasing nature of the variable input can mitigate the extent to which risk aversion, uncertainty, and farm size can influence input use and technology choice. They show that the gains from adoption of VRT under uncertainty and risk aversion are lower than those under certainty and the reduction in input applications and pollution is lower than under certainty.
Farmers that have high discount rates may be less willing to adopt a technology with high upfront costs and whose benefits may be realized over a long horizon (Khanna, Louviere, and Yang 2017). These discount rates combined with uncertainty about market prices, expectations about declining technology costs, or improving performance can create incentives to delay investment in precision technologies, particularly on components that have relatively high fixed costs. This is particularly the case on land parcels with low soil quality and low variability in soil quality, where the benefits of adopting these technologies are relatively small (Khanna, Isik, and Winter-Nelson 2000; Isik, Khanna, and Winter-Nelson 2001).
Socioeconomic Characteristics
Farmer demographic and socioeconomic characteristics are indicator of heterogeneity among adopters that may affect the economic gains and costs of adoption. Following the threshold model of adoption, these characteristics can influence the dynamics of the diffusion process. Key farmer characteristics that can affect adoption include human capital or education level, ownership, age, wealth, farm size, and attitudes towards environmental stewardship and so on. To the extent that digital technologies require specialized skills, their adoption may be more likely by better-educated farmers. The literature on adoption of various technologies—computers, new seed varieties, machinery, and better management systems—shows that more educated farmers are early adopters. On the other hand, some innovations, such as pesticides and management consulting, are human capital augmenting technologies and are more likely to be adopted by human capital-challenged or less educated farmers. Wealthier farmers may be more likely to adopt capital-intensive technologies since they face less credit and other financial constraints and they may also be less averse to risk. Age is another factor that has been shown to affect adoption but with mixed effects. Younger farmers may be more willing to invest in learning about new technologies and have a longer planning horizon for using the technology. On the other hand, older farmers may be more likely to adopt technologies with shorter repayment periods or those that may reduce effort. Adoption decisions are also likely by farmers that are less risk-averse and have smaller discount rates (Khanna, Louviere, and Yang 2017; Liu, Bruins, and Heberling 2017)
Agricultural and Environmental Policies
Environmental policies can create an incentive for adopting digital agricultural technologies by rewarding farmers for providing environmental services. However, the effectiveness of these policies depends on how well they are targeted to the source of the environmental problem. There are a number of conservation programs that provide financial, educational, and technical assistance for adopting best management practices on working lands in the US (see Shortle 2017). The incentives provided are uniform across farmers and locations. These incentives are not well targeted to maximize environmental benefits at least cost. These efforts are likely to be inefficient and costly because they do not recognize the differences in environmental impacts of the same set of practices due to differences in location, topography, weather, and soil conditions. Requiring all areas in a watershed to adopt the same practices is not an efficient strategy because not all areas contribute equally to the environmental outcome or have the same cost of abatement and because the incentives needed may differ with farmer-specific behavioral characteristics.
Technology adoption decisions are also affected by other agricultural policies, such as remove extra space.
subsidized crop insurance programs, absence of full-cost pricing of inputs, such as irrigation water or pesticides and fertilizer. Woodard et al. (2012) show that crop insurance can reduce incentives for adopting skip-row technology even though it is yield increasing.
Empirical Findings on Incentives for Adopting Digital Technologies
Studies show that incentives for adoption vary with the heterogeneous characteristics of farmers and across the different components of digital technologies. In general, adoption of precision agriculture is more likely by farmers that have larger farms and are younger, better educated, more innovative, and less risk averse (Khanna 2001; Isik and Khanna 2003; Schimmelpfennig and Ebel 2011;Torrez et al. 2016; Barnes et al. 2019). Schimmelpfennig (2016) reports that only 12% of farms less than 600 acres in size adopted each of the three main precision technologies (GPS/yield mapping, guidance system and VRT). For farms above 1700 acres in size the adoption rates of these technologies were 50%, 40%, and 23%, respectively. Size of the farm had a bigger impact on adoption of VRT than on the adoption of GPS or soil testing). Similarly, innovativeness of farmers and human capital variables were more important in influencing adoption of sophisticated technologies like VRT compared to soil testing. Diagnostic technologies, such as soil testing, were also likely to be more frequently adopted than VRT because of their potential to inform farmers about in-field variability and the benefits of varying input applications spatially (Khanna 2001).
Farmer characteristics have also been found to influence adoption decisions. Household income is also an important factor in inducing adoption because it creates the capacity to accommodate longer payback periods and in particular to adopt technologies with high entry/learning costs. Positive perceptions about the profitability of precision agricultural technologies were also found to affect adoption decisions (Barnes et al. 2019). Closer proximity to professional dealers that provide precision technology services also had a significant impact on the likelihood of adoption (Khanna 2001). Additionally, farm financial variables, such as primary occupation of the farm operator, legal structure of the operation, cost of hired labor, and level of farm financial leverage can also affect the profitability of various precision technologies on the farm (Schimmelpfennig 2016). Slow rates of adoption of VRT, in particular, are also likely due to the high costs of soil testing, laboratory analysis, and implementing map-based VRT systems; lack of clarity on decision rules to apply VRT; and uncertainty about the impact of VRT on farm profits (Lowenberg-deBoer and Erickson 2019). Hudson and Hite (2005) elicited farmer willingness to pay for a package of precision technologies and found that it was significantly lower than cost of adopting the technology, necessitating a 60% government subsidy to induce adoption, on average. Other determinants of the willingness to pay for the technological package included soil characteristic variability and soil quality, as well as how well the technology integrates into current farming practices.
Adoption of digital technologies is currently voluntary and is being promoted by the private sector as being in the self-interest of farmers. A significant share of farmers are yet to adopt these technologies for various reasons, including uncertainty about how to implement these technologies profitably, high costs, insufficient technical support, limited broadband/ mobile connectivity and the unproven nature of these technologies (van Es et al. 2016). Some of these technologies involve high fixed costs of adoption, large learning costs, and transactions costs that limit incentives for adoption. Farmers have also several other concerns about data-enabled precision farming that are likely to constrain their adoption. These include the need to give up production control by handing over their decision-making role to a “black box” that provides recommendations based on machine learning and computer algorithms. Concerns about data privacy and security issues may also influence adoption of different combinations of precision technologies. Agricultural technology companies are offering platforms for gathering and storing data from multiple sources and although these data may be protected, they are still linked to individual farm GPS coordinates and subsequent use of the data is out of farmers’ control. Additionally, precision technology providers note several factors that limit demand for their products and services by farmers. Precision agricultural technology dealers surveyed by Erickson, Lowenberg-DeBoer, and Bradford (2017) report that farm income pressures and time requirements for interpreting and making decision with precision agricultural information were key factors limiting use of precision services. Key barriers to growth perceived by over 50% of the dealers included the inability to charge fees that made it profitable for them to offer the services, rapid changes in equipment technology, incompatibilities across equipment, and finding trained employees.
Designing Policy Incentives to Induce Adoption of Digital Technologies
Adoption of digital technologies that have the potential to provide environmental benefits can be accelerated by policy incentives that reward farmers for adopting those technologies. The environmental benefits of adopting these technologies will depend on location relative to environmentally sensitive resources, soil quality, topology, groundwater availability, and other features of the land. Costs of adoption will also differ by location, farm, and farmer characteristics. Policy incentives for adoption should be targeted to locations where they can provide the largest environmental benefits at least cost (Khanna, Isik, and Winter-Nelson 2000).
First best approaches to targeting incentives to the source of the externality (for example, a tax on nitrate runoff) have been difficult to implement due to the absence of information and data needed to identify the sources of pollution and to measure their contribution to the pollution generated. Determining environmental impacts of on-farm production decisions involves linking production practices to onsite pollution generation and then linking the latter to off-site pollution loadings in water bodies.
Digital technologies and site-specific data on farm-level management decisions, soil and weather for millions of acres coupled with analytical and model-based tools can be used to more accurately determine the environmental consequences of farm-level management practices in a region. Pollution generated at a given time can be quantified and linked to management practices on specific fields after controlling for natural conditions at that time using predictive modeling based on information about farm practices (input applications, crop rotations, and tillage practices), crop varieties planted, soil and topographical conditions, and distance from water bodies. Since pollution generated by a field/farmer cannot be directly measured, the rewards and penalties for generating it still need to be based on practices adopted. However, with digital technologies, payments/penalties for adopting a practice can vary depending on the performance outcome to which they lead. This capability can facilitate implementation of policies that are performance-based rather than practice-based or input-use based.
Digital technologies together with computational tools and data analytics enable detailed record-keeping about input application rates, timing and methods used by a farmer. This lowers the cost of gathering information about agricultural management practices used by farmers.1 This together with biogeochemical and hydrological models and other environmental models can link farm management decisions with environmental outcomes, such as soil carbon sequestration, nitrous oxide emissions, nutrient loss, and run-off. By providing site-specific information about production decisions, environmental conditions, crop varieties, and yields, it can enable analysts to quantify the environmental impacts of agricultural production activities. Information on management practices, crop genetics, weather, and environmental conditions can be combined with process models to predict environmental outcomes. By enabling a data- and science-based linkage between activities and environmental outcomes, digital technologies can convert nonpoint pollution into point source pollution. These technologies offer the potential to track food production from farm to table and to provide credible certification for food supply chains that are healthy and sustainable.
Conclusions
Technologies for precision farming are rapidly being transformed with the emerging application of digital technology, big data, and artificial intelligence. These technologies offer the potential to increase crop yields, enhance productivity of input use, and reduce loss of nutrients and other inputs that can cause environmental pollution. The impact of these technologies on farm profits and the environment are likely to vary across locations, across components adopted, and with farm and farmer characteristics. Studies using historical data on adoption of earlier generations of precision technologies provides some case-specific insights on the type of considerations and factors that influence adoption decisions. However, there is limited systematic and generalizable evidence on the conditions under which these technologies lead to increased profits for farmers and reduced environmental impacts.
Much of the farm-scale adoption of the emerging digital technologies is being induced by private sector technology service provides and data being generated by these technologies are in the private sector and not readily available to researchers and policy makers due to concerns about confidentiality and data ownership. As a result, independent verification of the private and public benefits of these technologies is not feasible. Instead researchers need to apply other approaches, such as randomized control trials, to compare outcomes for adopters and nonadopters of a technology, choice experiments to ascertain the technology attributes likely to induce adoption, and contingent valuation methods to determine farmer willingness to pay to adopt these emerging technologies. Future research applying advances in behavioral economics is needed to examine the role that cognitive factors, peer pressure, demonstrations, trust in the information provider, and other considerations are expected to influence adoption decisions.
Policy incentives are likely to be required to induce the adoption of technologies that improve environmental outcomes. To be cost effective, these incentives need to be performance based and site specific rather than uniform across farmers and locations. Digital technologies have the potential to improve the ability to track the environmental outcomes of farm production decisions and to be used to design policies that can reward farmers for improving their environmental performance. By enabling detailed record keeping, including records of application of inputs and other management practices, these technologies can be used to verify compliance with environmental protection programs. Digital technologies are transforming the way that agricultural production decisions are made; significant research and development is needed to determine how these technologies may be used not just to provide private benefits but also for the public good.
ACKNOWLEDGMENTS
The author acknowledges support from the World Bank Project “Digital Acceleration of Agricultural Transformation” Agriculture Global Practice, World Bank, Washington DC, for this research.
REFERENCES
- 1 Benami and Carter (2020) offer other examples where digital technologies can lower information costs and lead to greater efficiency in rural credit markets.
