Chapter 4. Issues which may prevent digital opportunities from being realised

This chapter discusses a number of different issues which may prevent opportunities identified in Chapter 3 from being realised in practice. It discusses challenges to the successful uptake of technologies by policy makers and programme administrators and provides practical guidance in addressing them. Further, it considers new risks which may arise as digital tools are adopted to support policy in agriculture, and provides guidance on steps policy-makers and administrators can take to mitigate such risks.

    

Chapter 3 provided a range of examples of how digital technologies are currently being used to improve policies, and detailed further opportunities for the future. As identified in the conceptual framework (Figure 3.1), a number of different issues may prevent opportunities from being fully realised in practice. This chapter examines different types of issues in turn:

• Practical challenges faced by policy administrators (section 4.1)

• Existing institutional and policy settings may act as an impediment (section 4.2)

• New challenges may occur as digital technologies are increasingly used to support policy in agriculture (sections 4.3 and 4.4).

4.1. Practical challenges for the use of digital technologies by policy makers and administrators

Responses to the OECD Questionnaire show that three challenges are perceived as the most important limiting factors to digital technologies use (Figure 4.1): constrained financial resources, the required substantial change to current workflows, policies or programs, and access to specialised skills required to use “big data”.

Figure 4.1. Organisational challenges in relation to using digital technologies and “Big Data”

Note: N=45, except for EU question, where N=19.

Source: OECD Questionnaire.

Little information is currently available on the costs involved in reconfiguring existing IT systems, retraining staff (or other options such as hiring new staff, or outsourcing). However, some examples are provided via the case studies conducted for this report:

• Case Study 1 (Digital tools for New Zealand’s Our Land and Water National Science Challenge) includes funding data for a range of projects, which include development of digital tools (but are not solely limited to tool development): for example, NZD 3.56 million in funding was provided for the “interoperable modelling” project, which aims to develop a modelling system populated with models which draws on national datasets and which are implemented in an interoperable modelling framework.

• Case Study 2 (ICT-SCAN system for Dutch Agricultural collectives) reports that the Dutch government provided EUR 10 million towards the initial setup cost for the IT system, and ongoing costs are estimates at EUR 1-2 million.

These estimates show that the costs of developing digital tools for policy-making are not insignificant.

Moreover, developing the skillsets and organisational capacity necessary to effectively deploy digital technologies for policies is also likely to be costly, in terms of both agency time and actual costs incurred. There is very little data available on these cost aspects; however, failing to take them into consideration will overstate the net benefits of digitalisation. While many of these costs could be considered “start-up” costs (and therefore diminish in importance over time), there are also likely to be other fixed costs, as well as ongoing variable costs associated with error-checking, testing, maintaining and upgrading digital systems. These costs as well as ongoing skills and management requirements need to be factored into overall budgeting and planning, so that digital systems underpinning policy delivery continue to function well over time.

An additional range of practical challenges for the use of digital technologies in the context of improving agri-environmental indicators (AEIs) were also identified at a joint OECD-Natural Resources Institute of Finland (Luke) workshop in May 2018 (Box 4.1). These are related to the continued inability (despite recent advances) of digital technologies to cost-effectively “fill in” existing information gaps relevant for producing AEIs. They also related to institutional challenges, particularly lack of standardisation and differing regulatory regimes leading to an inability to achieve representativeness or comparability.

Box 4.1. Challenges yet to be overcome: Evidence from OECD-Luke workshop

The OECD, together with the Natural Resources Institute of Finland (Luke) convened a workshop on ‘The use of new technologies for agri-environmental indicators to support effective policy monitoring, evaluation and design’. The workshop provided an overview of cutting-edge developments of the use of digital technology and agricultural “Big Data” to form robust and readily interpretable indicators that can be used for the monitoring, evaluation and design of agri-environmental policies. Apart from identifying a number of promising initiatives, participants identified a number of challenges that have yet to be overcome. Many of these challenges are relevant to use of digital technologies to support better agricultural policies more generally. In particular, participants identified the following:

Technological constraints and data gaps

• For tracking specific indicators like birds or pollinators, remote sensing technologies still need to be combined with in situ monitoring programs which tend to be costly to maintain (see Case Study 9).

• Indicators resulting from a combination of digital technology sources and modelling are highly sensitive to the methods used for processing and modelling the data. No consensus exists across jurisdictions on a standard set of processes and methods.

• Agri-environmental (AE) indicator models (even those using satellite data) are often based on much smaller ecological regions than are the focus of AE policies. So, there is still a need to be able to make justifiable assumptions to interpolate data.

• It is still difficult to analyse environmental performance at the aggregate level of the commodity (in part because aggregate performance is the sum of the performance of many individual actors in a range of specific contexts).

Institutional or regulatory constraints

• Within the EU context, the LPIS/IACS data operate at a level of detail which can currently not be achieved with satellite data in a standardised or automated way at EU level. Moreover, within certain countries where multiple LPIS/IACS operate, LPIS/IACS data are not harmonised. There are also technical problems to keep track of a given field over time using GIS software.

• As much of the data from digital technologies are owned by farmers or private companies, lack of harmonisation of data privacy laws creates imbalanced incentives across jurisdictions for farmers to share their data.

• Data from some technologies, such as sensors are only available from farmers who have access to them. While data from these sources can be highly detailed, its representativeness at the regional or national level is questionable.

• Although digital technologies allow for improved assessments of complex physical processes and improved understanding of risks and scientific uncertainty, it is still very difficult to effectively communicate scientific uncertainty and risk to stakeholders.

• There is a need to adequately prove both the efficiency and uncertainty of new technologies before they can be used for policy, but there is not always agreement or clear processes to achieve this.

• Limits in AEI data still hinders the ability to take a holistic approach to assessing agricultural sustainability and to draw policy recommendations from this assessment.

• Non-alignment of incentives remains between researchers, the private sector and farmers: can new incentives to collaborate be created?

Sources: Presentations from the OECD-Luke Workshop, 29 May to 1 June 2018, Helsinki, Finland.

4.1.1. How are organisations responding to these challenges?

Evidence shows that organisations are also already working to tackle these challenges. The results of the OECD questionnaire (Figure 4.2 and Figure 4.3), show in particular that 80% of respondents indicated that their organisation is undertaking training for existing staff or working with contractors. In contrast, only 46% are hiring new staff; this may reflect that the majority of organisations face financial constraints and difficulty hiring or retaining the specialised skills needed to work with Big Data. These responses suggest that financial constraints and skills mis-matches are both important challenges to address.

Figure 4.2. Building capacity to work with new digital technologies and "Big Data"

Note: N=48.

Source: OECD Questionnaire.

Figure 4.3. Initiatives to increase use of digital technologies and Big Data for agriculture

Note: N=49. BD = Big data; AE = Agri-environmental.

Source: OECD Questionnaire.

The results also show that a most commonly-used means to foster use of digital technologies and big data for agriculture is to invest in government-maintained sustainable data repositories; 81% of respondents already have such initiatives underway. However, it is unclear to what extent these repositories are open to the public, including farmers. Consistent with the notion that governments should not “crowd out” private sector development, the least-common initiative among respondents was developing analytical tools based on big data for private sector commercial use.

In terms of future initiatives, creation of analytical tools based on big data for governments was the most commonly-anticipated initiatives for the near future (within the next three years), followed by increased collaboration with private sector analytics companies. Around a third of respondents have already developed analytical tools for farmers, and a further 23% were expecting to do so in the next three years. Respondents also indicated they would in future be developing analytical tools based on big data for the general public.

4.2. Institutional and policy settings can limit opportunities for policy from being realised

Throughout the preceding sections, specific pathways for digital technologies to improve existing agri-environmental policies and enable new ones have been explored. It has also been recognised that there are challenges to successfully implementing digital solutions and digitally-enabled policies, some of which have been illustrated via the case studies. This section elaborates on key institutional and policy settings which can limit the potential offered by digital technologies being realised.

4.2.1. Institutional constraints and lock-in

Institutional path-dependencies can act as a disincentive for organisations to change their processes (e.g. administrative processes) to make best use of new technologies. There can be several types of path-dependencies.

First, policy administrators might consider that the cost to be borne for the change in system would be too high. Such costs not only include the cost of setting the new digital system but also the expenditure to train staff, and time needed for them to adjust to new systems, which can vary according to staff skills and flexibility. The questionnaire highlights that a number of organisational path dependencies constrained the adoption of digital technologies and big data (Figure 4.1). These included being hampered by existing IT systems, lack of financial resources, and that substantial changes to organisational workflows, policies or programmes would be needed to make more use of digital technologies.

Second, existing environmental objectives may be specified in terms which reflect pre-existing levels of technological feasibility. For example, specifying air quality or water quality objectives in terms of average levels over a given period may preclude the use of short time-step point-in-time data or continuous monitoring which can provide a higher degree of temporal granularity (Macey, 2013[1]).

Third, many agri-environmental programmes, particularly agri-payment programmes, are designed to make use of reference levels, baselines or thresholds to determine participant eligibility, payment amounts or to establish a set of practices that a participant may receive payment for. Two important examples include:

• cost share payments which use state, regional or national averages for cost elements (e.g. USDA NRCS uses state-based payment schedules for making conservation payments under a range of programmes1);

• use of relatively coarse regional averages as parameters in specifying physical relationships (e.g. nutrient-removal effectiveness of different management practices; emissions factors for livestock or crop types).

Existing modelling apparatus and consensus mechanisms for establishing reference levels or parameters for agri-environmental policies or for policy-relevant research may require reform in order to be able to make use of new, higher resolution data products (demand for which would also increase demand for improved sensor networks). Such reform may be costly, particularly in regulatory settings where processes for establishing reference levels or important parameters may be codified in regulations. However, advances in algorithms and simulation techniques have lowered the cost of building more refined models including processing the data they require.

Fourth, agri-environmental regulations may set procedural requirements that preclude uptake of innovative digital technologies. A key example is regulations specifying monitoring and control procedures which require on-site “boots on the ground” monitoring. Such requirements impede the uptake of earth observation and other remote sensing technologies for monitoring and compliance activities. Further, the use of technology or performance standards and sequencing requirements2 may also limit the realisation of opportunities to improve and expand the use of digitally-supported policy instruments. For example, demand for digital technologies to verify emissions reductions, or demand for online markets for trading environmental credits (e.g. nutrient credits, carbon emissions) may never eventuate if market-based mechanisms as a whole are stymied by technology standards or sequencing requirements (Stephenson and Shabman, 2017[2]).

Fifth, broader regulatory settings may limit individual organisations’ ability to make use of digital technologies. An important example is privacy or confidentiality regulations (see next section). Thirty per cent of respondents to the OECD questionnaire agreed or strongly agreed that they were hampered by existing national or sub-national privacy or confidentiality regulations. While this data indicates that regulatory constraints are not an issue for the majority of responding organisations, nevertheless this challenge occurs for a sizeable minority. Interestingly, in several cases, one respondent from a particular country would strongly agree that national or sub-national regulations were a challenge, whereas other respondents from the same country indicated otherwise. While decisions about privacy regulations are complex and need to account for a broad range of factors, these situations may offer an opportunity for cross-organisational discussion within countries, and for national agencies to better understand exactly when and where privacy regulations are a constraint for government organisations.

Lastly, the influence of technology or data providers may create path dependencies and even the potential for lock-in. For example, some aspects of policy design could be influenced by a small group of technology or data providers, who could significantly benefit if implementation or enforcement relied on broad uptake of certain digital tools by farmers or administrators. While in some cases administrators may wish to work with technology or data providers, for example, to provide customised solutions for a specific context or problem, they should be aware of this potential and take steps to pre-empt this problem from occurring.

4.2.2. A lack of trust can be a roadblock to using digital technologies to reform policies

One of the conditions for effective policy change is to engage stakeholders strategically and build trust (OECD, 2018[3]). This can be especially difficult where different stakeholders have opposing views and non-aligned interests. In a context where policy administrators are considering making use of digital technologies to achieve policy reform, there may be different levels of resistance: for example, some stakeholders may resist the overall objective of reform, while others may accept the overall direction of reform but still have concerns related to the use of digital technologies. Additionally, digitised forms of data gathering (e.g. via UAVs or satellites) or analysis (e.g. via algorithms) may be resisted if they bring about reductions in organisational personnel or funding.

However, rather than being an additional source of conflict, digital tools and data may actually be able to foster collaboration and overcome traditional roadblocks created by conflicting views and values. These collaborations help ensure that digital tools are well-designed while at the same ensuring buy-in by all stakeholders. Further, increasing farmer access to agricultural data can in and of itself be a useful agri-environmental policy tool; as farmers increasingly understand the specific environmental impacts which result from their actions, they may become more willing to participate in agri-environmental policies seeking to reduce those impacts.

As shown in Case Study 1 (see Part IV for details), digital tools and data sharing are being successfully used in New Zealand’s Our Land and Water National Science Challenge to help parties with different interests and incentives build consensus. For example, the OVERSEER® nutrient model, which is being enhanced under the Challenge and aligned programmes (e.g. to be made spatially explicit by MitAgator), has been developed using co-innovation and can be scrutinised by all interested parties. It functions as an “authoritative point of truth”, but can be updated with the latest available science and incorporate innovations (e.g. new data sources from new sensor technologies).

4.3. Using digital technologies for policies raises new challenges

Use of digital technologies for agri-environmental policies may raise new challenges which, if not addressed, may limit the actual benefits obtained. In the conceptual framework (refer to Figure 3.1), several types of new challenges caused by use of digital technologies were identified. The following sub-sections consider challenges in the context of agri-environmental policy implementation in relation to:

• The potential for existing actors to respond to new technologies or new incentives in negative ways, or in unforeseen ways which then require a further policy response.

• The potential for new technologies to create new risks which need to be directly addressed, or cause unintended consequences which then require a further policy response.

The challenges discussed here are strongly linked to the specific types of policies used and choices about policy implementation. While broader consultation with other government agencies may be useful to ensure a “joined up” approach and synergies across different policy areas, because of this specificity, they need to be addressed by the organisations responsible for administering agri-environmental policy. Challenges that may need a broader approach are discussed in the following section.

Note that in general, evidence on the extent to which these challenges are actually occurring in practice is scant. Therefore, the following subsections explain the (potential) challenges and in some cases highlight actions governments can take either to help ensure these challenges don’t eventuate, or to address them if they do.

4.3.1. Social impacts and acceptance of increased monitoring

The first challenge is that having better data on negative environmental impacts may increase social tensions between the agriculture community and the rest of society, fostering an “us versus them” mentality, rather than co-operative approaches to improve agricultural sustainability. This is particularly relevant in a non-point source context, which applies for much of the agriculture sector. At present, while there may be estimates of the sector’s aggregate environmental impact (e.g. regarding water quality impairments, GHG emissions, etc.), in many cases it may be difficult to determine the individual impact of a specific farm. If the use of in situ and remote sensing facilitates measurements (or even more reliable estimates) of individual impacts at the farm level, this could be used to label farmers with poorer environmental performance as “polluters” or “resource squanderers” and create a stigma in the community (Myles, Duncan and Brower, 2016[4]). While individual accountability is an important tool for incentivising improved performance, stigmatising individuals (especially in a context where individuals may have been hitherto unaware of their performance and may take time to implement changes) can be destructive and lead to decreased social cohesion and mental health in rural communities (Gregory and Satterfield, 2002[5]).

One important strategy for policy administrators to mitigate such risks is to make use of digitally-enabled result-oriented mechanisms (or performance measurement more generally) to foster identities centred on the concept of stewardship, and to emphasise (where it is the case) that change takes time. Practically, this entails design elements such as:

• Including the objective to foster a stewardship mentality explicitly into policy or programme objectives.

• Measuring or estimating and reporting of individual or collective (depending on approach3) performance to the programme administrator.

• Including a mechanism for broadcasting good performance to programme participants, peers, and potentially to the public in general.

• A well-implemented, graduated compliance or enforcement framework which:

  • encourages participants to self-identify poor performance and to report this to the administrator;

  • provides room for improvement over time (as opposed to more “heavy handed” responses such as immediately rejecting poor performers from participating).

Additionally, allowing policies to be jointly designed with stakeholders (e.g. farmers, environmental groups) can help create partnerships between agricultural and environmental interests rather than entrenching dichotomies. Results-oriented programmes can be particularly useful in this respect as they “create common goals between farmers and conservationists, leading to cooperation between two conflicting groups...result-oriented schemes can [also] communicate the extent to which farmers contribute environmental services to society and, consequently, help to justify financial support to the farming community” (Burton and Schwarz, 2013, p. 632[6]).

While all of these design elements may be possible to achieve in part without the use of digital technologies, several factors suggest that digitally-enabled mechanisms are more likely to work better:

• The use of data generated by digital technologies, particularly from satellite remote sensing and wireless sensor networks, can enable a shared, scientifically-based understanding of resource concerns and results achieved. Further, the emergence of near-real-time data can inform farmers of the state of their environment more easily and more instantaneously, which helps grow their understanding of how their management actions affect the environment.

• The use of models (particularly in programmes where results are modelled rather than measured directly) that are GIS-based and able to take into account a high level of spatial heterogeneity are likely to be more accurate and able to accommodate a wide range of practices when estimating results. Also, use of GIS-based tracking of results enables assessing progress in aggregate, which can be important both for achieving landscape-level goals and for creating a community sense of ownership of those goals.

• The use of computer algorithms to calculate payments based on results could allow for payment structures that pay for multiple environmental benefits and which take into account relationships between different environmental benefits (e.g. relationships between water quality, biodiversity, and greenhouse gas emissions). This could allow payment schemes to minimise non-additional payments and reward farmers who achieve multiple benefits.

• The use of digital platforms for administering results-based schemes can enable simple communication between participants, enabling peer-to-peer learning, and between participants and the broader public, lessening an “us-versus-them” mentality and fostering a stewardship attitude, and between participants and the administrator.

4.3.2. Dynamic challenges of agri-environmental mechanisms which rely on models

Increased reliance on data and complex modelling software increases the need to be explicit about the limitations of data and models, and how these limitations vary across data sources and modelling efforts. In particular, there is a need to avoid models becoming perceived as “truth machines” by policy makers (Duncan, 2014[7]).4

Also, the increasingly rapid pace of technology innovation creates issues with relying on outdated tools (Duncan, 2014, p. 383[7]). Periodically iterating an entire policy as technologies update may create rigidities or large “step changes” in requirements, which are costly for farmers and which introduce or increase regulatory uncertainty. However, updating requirements in a more piecemeal way (i.e. in line with individual permit cycles, contract terms, land planning cycles, etc.) can introduce inequalities across participants (i.e. some actors are regulated under or participating in the old system, while others are under the new). Therefore, policy makers need to actively consider how to create mechanisms which allow regulatory regimes and voluntary programmes to evolve smoothly with technologies. Environmental markets are a promising tool to support the “piecemeal” approach while mitigating (at least in part) the potential for inequalities. For example, in a regulatory context, actors who are unable to meet updated regulatory requirements on site could be allowed to meet the requirements via purchasing off-site credits.5 Also, “phase in periods”—in which consequences of non-compliance with newly-introduced rules can be gradually ramped up—can be useful to assist participants who were compliant with the old regime to transition towards the new.

Monitoring and modelling should be viewed as complementary.

Often, monitoring and modelling happen as two separate streams of work, and modelling is often described as being needed in the context of incomplete information. This implies that modelling is only needed because of data deficiencies; that is, that monitoring and modelling are substitutes.6

In many cases, data gaps are likely to persist: monitoring of all physical variables of interest is unrealistic, despite advances in sensors, Internet of Things devices (e.g. “smart” agricultural machinery) and remote monitoring technologies which enable much broader physical monitoring at lower cost than previously. Therefore, there will still be a need for models to attempt to “bridge” these gaps.

However, even if all necessary physical measurements could be obtained via monitoring, modelling may still be needed for a variety of functions, such as attributing physical impacts to non-physical drivers (particularly to policy drivers, so that policies can be evaluated), and modelling future scenarios to make ex ante policy assessments and improve planning. Thus, modelling and monitoring should be viewed as complementary: modelling both uses data and allows for analysis in the absence of data.

4.3.3. Policy design elements can be a pull factor for technology adoption on-farm, but there is a risk of exclusion

As discussed in detail in Chapter 3, there is substantial opportunity for policy makers and administrators to make use of digital technologies for better agricultural and agri-environmental policies. While realising such opportunities will obviously result in technology adoption by government organisations (or third parties providing services to these organisations), it may also incentivise adoption of digital technologies on-farm:

• As governments move to increasingly interact with programme participants via digital channels (e.g. requiring applications to be submitted online, making payments using e-banking services, releasing information in digital formats, providing access to online databases, providing technical assistance or extension services via apps or online platforms), use of digital technologies by programme participants (i.e. farmers) is likewise expected to increase.

• Adoption of digital technologies by the public sector may also change the way food system policies are designed, enforced and monitored. This may result in revised or new requirements for tracking and tracing, as well as better management of food safety. Such new requirements may necessitate adoption of technologies by farmers: for example, livestock farmers may be required to adopt RFID tags for all animals, and to record and submit data on animal movements or other aspects (e.g. animal health data) via digital channels. Digitally-enabled traceability schemes (e.g. Blockchain-based traceability systems) may incentivise farmers to adopt sensor technologies for collecting data to be stored in digital databases, and to make increased use of online platforms.

• If policy makers move towards more result-oriented programmes, particularly those which focus on measured results (as opposed to modelled results), this is likely to provide further incentives for farmers to adopt digital technologies on farm, for two reasons. First, farmers will have more flexibility in how they go about improving their environmental performance, and may make greater use of digitally-enabled input-saving practices such as variable rate technologies and highly automated on-farm processes. Second, farmers will be more incentivised to invest in digital technologies and services for measuring improvements in their environmental performance.

Adoption of technology can be costly for farmers. Apart from potentially needing to invest in the technology itself (for example, purchase of precision agriculture machinery, upgrading computer systems, etc.), there may be additional entry costs such as learning costs and adapting production processes. Thus, governments need to carefully consider the potential for adoption costs to produce a net increase in regulatory burden when considering the introduction of new standards or regulations, particularly in cases where farmers are not able to opt-out of participating in regulatory mechanisms (i.e. mandatory regulations).

A related risk is that the production of new digital tools and new knowledge from those tools does not inadvertently produce information asymmetries. This could potentially occur, for example, if only researchers involved in creating new knowledge or tools had access to them. Another potential source of information asymmetries is linkages between large multinational firms and the public sector or academia, which could result in some actors being able to access data or analysis at lower cost than others. Case Study 1 (New Zealand) shows that one way to ensure that address this risk is addressed is to take a co-innovation approach. This way, stakeholders are directly involved and production of new knowledge is readily shared with all stakeholders.

4.4. New challenges which may require a broader approach

In addition to the challenges discussed in the previous section, there are some challenges which are relevant for agri-environmental policy makers, but for which a broader approach may be required. Several reasons may underpin the need for a broader approach:

• First, the solutions to challenges faced by agri-environmental policy makers may be legislative or regulatory solutions that are the remit of other areas of government – key examples here are where solutions relate to privacy laws, competition matters, or consumer protection.

• Second, challenges associated with certain technological solutions that are useful for agri-environmental policy may also arise in other contexts. A key example here is that issues relevant to providing technological solutions to increase access to agricultural microeconomic data for policy-making also arise in relation to increasing access for the development of data-driven services for agriculture.

• Third, technology-related challenges for agri-environmental policy makers may also be faced by other policy makers: an example is that issues relevant for environmental regulation in agriculture may be relevant for other regulators, e.g. animal welfare regulators, economic regulators.

4.4.1. Potential pitfalls of “RegTech” for agriculture

As demonstrated in previous sections, agri-environmental regulators and programme administrators have the opportunity to make increased used of digital technologies in performing their functions. Administrative or regulatory decisions are increasingly based on information provided by digital tools, and the prospect of using machine learning and artificial intelligence to fully automate certain regulatory or administrative processes and decisions is now conceivable (Adams, 2018[8]; Coglianese et al., 2017[9]).

Devolution of decision-making to computers raises several important questions:

• Transparency: how can algorithms be designed so that agri-environmental administrators and regulators, farmers and other relevant stakeholders understand how results and conclusions are obtained?

• Oversight: how can agri-environmental administrators and regulators (who may have little expertise with technology) have confidence that algorithms (which may be designed by technology specialists with little knowledge of agriculture or agricultural policies) are suitable (including suitably accurate) for the purposes they are designed for? How can they determine when such algorithms are “wrong?”

• Responsibility, right to challenge and access to remedies: who is responsible if algorithms make the “wrong” decision? For example, if a farmer participating in an agri-environmental scheme is denied payment due to a flaw in a payment algorithm, is the farmer able to challenge this decision? What process is there to “right the wrong”?

Potential pitfalls await if these questions are not considered and answered satisfactorily. If transparency is not achieved, farmers and other stakeholders are unlikely to have confidence in decision-making processes, which may lead to unwillingness to participate in policy mechanisms (particularly voluntary programmes) or to costly challenges to regulatory or administrative regimes. If design and use of algorithms lacks suitable oversight, there is potential that algorithms may not be suitable for their intended uses. If regulators and administrators do not take responsibility when algorithms arrive at the “wrong” decision, they may suffer reputational damage and risk legal action. Moreover, farmers should not face additional costs in the event that algorithms fail.

Such considerations are not specific to agriculture. In fact, use of advanced technology by regulators—referred to as “RegTech”—first arose in the financial sector, in the aftermath of the 2008 financial crisis (Arner, Barberis and Buckey, 2016[10]). Regulators and administrators in the agriculture sector have the opportunity to learn from their peers in other sectors, and should adopt best practices for use of algorithms to support regulatory and administrative decision-making.

Technological progress and regulatory remit

A related challenge is the temptation, real or perceived, for agri-environmental regulators to expand their regulatory authority according to what the latest technology is able to measure (sometimes referred to as “regulatory role creep”). The potential for this to occur in the agriculture sector is the result of various trends including:

• Agricultural databases are becoming increasingly interlinked, which creates the ability to use data for purposes for which they were not originally intended, including regulatory purposes such as developing farm profiling (Directorate-General for Parliamentary Research Services (European Parliament), 2018[11]).

• Advances in remote sensing technologies (satellites, aerial-borne sensors) has vastly increased regulators’ and administrators’ ability to gather data on farmers without involving farmers themselves. This gives rise to the possibility that farmers may be monitored without being aware of it and that farmers may (whether correctly or not) perceive that they have insufficient (or no) opportunity to dispute regulatory or administrative decisions based on such data.

The solution to this challenge is not to preclude agri-environmental regulators from making use of digital technologies and agricultural data to improve their performance. Neither should it be taken as given that existing regulatory frameworks, which may in part be shaped by pre-existing technologies and data availability, should remain unchanged as technologies develop. Rather, it is recommended that agri-environmental regulators and administrators:

• Implement transparent processes to enable scrutiny of how regulators and administrators are using agricultural data and new technologies.

• Implement clear and participatory processes for considering how regulatory and administrative frameworks should evolve with technologies, and for vetting technologies for their suitability for use in regulatory contexts.

However, it is recognised that in some cases, the extent to which agri-environmental regulators and administrators have complete jurisdiction to implement these processes may be unclear; implementation of these recommendations may require a cross-government effort.

A further challenge is the impulse for policy administrators to move to limit policy coverage to only those farm practices which can be easily monitored using specific technologies (e.g. remote sensing). An example would be a policy which limits payments for agri-environmental practices or results which can be monitored using satellite-based remote sensing. While consideration of administrative transaction costs is a fundamental component of designing cost-effective policies, and monitoring via remote sensing appears likely to contribute to large reductions in administrative costs (see case studies in sections 3.2.5 and 3.3.1), cost-effectiveness still needs to be assessed holistically.

It is important to recognise that issues about regulatory remit and use of technology to enable regulation is unlikely to be specific to agri-environmental policy. Other regulators likely face similar issues, both within and beyond the agriculture sector. Regulators may be able to learn from each other about how to best integrate digital tools into their overall approach.

4.4.2. Access to farm-level agricultural data for policy-making

As discussed above, there are opportunities for policy makers to make better use of agricultural data to design and deliver better policies, whether by implementing better spatially-targeted policies, results-based mechanisms, new monitoring and compliance approaches, etc. To deliver such data-based policies, policy administrators and related researchers would likely require improved access to agricultural data, including the ability to link different datasets. This linkage may need to occur at the farmer, farm or field level in order to evaluate policy microeconomic and environmental impacts (Jones et al., 2017[12]; Petsakos and Jayet, 2010[13]).7 Further, data may need to be shared across borders, for example, to facilitate comparative policy analysis and to underpin national, regional, and global policy-making (Legg and Blandford, 2019[14]; Carletto, Jolliffe and Banerjee, 2015[15]).

Data confidentiality requirements are often cited in the literature as a limiting factor for using micro-level agricultural and agri-environmental for policy delivery and related analysis (Martínez-Blanco et al., 2014[16]) (Tukker and Dietzenbacher, 2013[17]) (VanderZaag et al., 2013[18]). Access issues are particularly prevalent where different government (or even non-government) entities have responsibility for different aspects of the agri-environmental policy cycle (or across different policies): for example, agricultural agencies responsible for administering programmes may be unable to share farm-level administrative data with environmental regulators; agencies collecting data on rainfall, soils and water quality may be unable to link their records with data on farm decision-making, output, and profits.

Limitations on accessing agricultural data (whether for policy-making or other uses) are generally of long-standing and have been crucially important for establishing trust between farmers and government data-collection organisations. For example, agricultural censuses and surveys conducted by or on behalf of government agencies, which have long been a key source of such data, generally contain strict confidentially requirements in their enabling legislation, which limit the ability of agencies to combine data from different sources or share it with policy researchers.8 While these mechanisms are aimed at protecting farmers’ interests, they have the consequence of limiting the potential for farm-level data to be used (and re-used) for policy-making and implementation. In addition, administrative data,9 usually gathered and held by government agencies, is an important source of information relevant for policy-making. However, access to administrative data is often even more limited than access to farm level survey or census data.

Options to improve access to agricultural data held by public organisations to improve policy

In theory, one solution to improve access to agricultural data for policy-making, administration and related research is to reduce or modify confidentiality obligations, for example by developing data-sharing agreements between different government organisations and related researchers, or by publishing de-identified farm-level data. However, these options may be unpalatable or unworkable in practice, or may not significantly improve the usefulness of data for policy. An important issue is the question of how to provide access to farm-level data that has geographic attributes that are meaningful for research and policy, without allowing identification of the farmer’s identity or precise location of the farm.

Further, unilateral attempts to lessen public sector obligations to preserve privacy or confidentiality have the potential to result in erosion of public trust in these agencies. Thus, any decision to fundamentally change such obligations (whether for policy purposes or more generally) will require discussion and agreement between governments, farmers, researchers, the private sector and NGOs about important questions of data ownership and access to data—these questions are discussed in section 4.1. Government organisations may in fact have limited ability to lessen legislated confidentiality guarantees, especially in relation to existing datasets; therefore an open data approach for agricultural micro data may not be an achievable or desirable end goal.

However, where governments wish to improve access to agricultural data while maintaining confidentiality, there may be solutions which policy administrators can take to avoid the confidentiality-accessibility dilemma altogether, including:

• Technological solutions, such as encryption, “confidential computing” and other gatekeeper technologies, which permit greater use of farm-level data for policy purposes while using technology to preserve confidentiality (see Section 2.1.3 for an overview of these technologies).

• Administrative or institutional solutions, such as creating research collaborations whose aim is to improve access to and use of agricultural micro data specifically for research and policy analysis (for example, the OECD Farm Level Analysis Network—see Box 4.2.), and providing differential access based on data sensitivity. Policy-makers can also consider use of new data collection methods which do not require direct participation from farmers. However, policy makers need to take into consideration how this might impact on the existing trusted relationships with farmers, in both positive and negative ways.

• Incentive-based solutions, including:

  • policies which use farmers’ preferences to maintain anonymity as an incentive mechanism to encourage improved environmental performance through collaborative, landscape-scale initiatives and “trust-based” environmental regulations (Lange and Gouldson, 2010[19]) (Case Study 7, Box 4.3);

  • policies which provide voluntary incentives for farmers to provide data for public research and analysis (e.g. making payments or providing services such as benchmarking or advice in return for provision of data for policy purposes).

It should be noted that solutions could have elements of all of the above; they are not mutually exclusive. Moreover, the choice of solutions raises fundamental questions about how best to balance different considerations, including fostering trust between data providers and users and how to balance maintaining confidentiality or privacy with increasing access to derive greater benefits from agricultural data. While these questions may be crystallised in debates about the use of government-held agricultural data to improve agri-environmental policy, they form part of the broader debate about data governance and digitalisation.

Box 4.2. OECD Farm Level Analysis Network

The Farm Level Analysis Network (FLAN) was created in 2008 under the auspices of the OECD. It includes experts from government-related institutions, and other agricultural economics research institutes involved in the collection or analysis of micro-level data and interested in collaboration. Membership is voluntary and a representative coverage of OECD countries is sought. The OECD acts as convenor and contact between network members and delegates to OECD meetings.

Network members and the OECD share the common goal of improving the quality and relevance of policy analysis applied to the agricultural sector through the use of micro-level data, recognising the increasing need for good micro data and related analytical tools to support improved policy decision making.

The main objective of the network is therefore to support OECD policy analysis through the use of micro-data and sub-national information. The network contributes to OECD projects by providing micro-level data on a consistent basis across a number of countries, thus facilitating access to data needed for micro-level analysis. From the projects adopted in the programme of work of the OECD Committee for Agriculture, the network identifies issues that would benefit from a micro-level approach, identifies data sources and suggest innovative and adapted approaches.

Another objective of the network is to share experiences and to demonstrate how micro-level analysis can be used for policy analysis. This is achieved through communication of relevant analysis and discussion of data and analytical issues. As part of this objective, the network draws the attention of delegates to emerging policy issues, where micro-level approaches could be particularly rewarding, with a view to contributing to reflections on the programme of work in the longer term.

Source: Adapted from https://www.oecd.org/agriculture/farm-level-analysis-network/.

Box 4.3. Case Study 7: Data transparency regulations enabling Californian water quality collectives

This case study provides an example of how data regulations and coalition-based water quality monitoring regimes can be used to underpin collective governance mechanisms to address nonpoint source environmental impacts from agriculture.

California agriculture is extremely diverse, producing more than 400 commodities and spanning a wide array of growing conditions from northern to southern California. However, water discharges from agricultural operations can affect water quality by transporting pollutants from cultivated fields into surface waters. Groundwater bodies have also suffered pesticide, nitrate, and salt contamination. To prevent agricultural discharges from impairing receiving waters, the Californian Irrigated Lands Regulatory Program (ILRP) regulates nonpoint source discharges from irrigated agricultural lands. This is done by issuing waste discharge requirements (WDRs) or conditional waivers of WDRs (Conditional Waivers) to growers or groups of growers called Coalitions.

The California State Water Resources Control Board’s (State Water Board) Policy for the Implementation and Enforcement of the Nonpoint Source Pollution Control Program1 (Nonpoint Source Policy) directs that any nonpoint source program (such as the ILRP) incorporate monitoring and reporting. Programs must “include sufficient feedback mechanisms so that the [regional water board], dischargers, and the public can determine whether the program is achieving its stated purpose(s), or whether additional or different [management practices] or other actions are required.”

This requirement to undertake monitoring of agricultural runoff and receiving water bodies and reporting constitutes an effort to reduce information gaps about the quality of these waters, as well as the impact of agriculture on water quality. This data is crucial for the California Water Boards to achieve their mission. However, these requirements are controversial to the agricultural community because they are costly to comply with and result in lessening of information asymmetries that producers may have incentive to maintain. Therefore, the challenge for California Water Boards is to balance “the need for transparency and measurable benchmarks” and maintaining acceptable regulatory outcomes with ensuring regulatory burden is minimised and respecting “the need for the agricultural community to protect trade secrets and other sensitive information” (State Water Board, 2018[20]). This challenge is not unique to this context; it arises from the characteristics of agricultural production, which uses inputs (e.g. fertiliser and pesticides) and commercially sensitive information (e.g. fertiliser application regimes) to produce valuable outputs, but which also produce environmental externalities that are costly to address.

The Water Boards have devised monitoring and reporting regimes which aim to provide data for the required “sufficient feedback mechanisms”, while minimising regulatory burden and risks for producers related to information disclosure. An example is the regime of the Central Valley Water Board (one of nine regional water quality control boards),2 which comprises:

• The use of water quality coalitions to act as intermediaries between growers and the regulator;

• Data transparency requirements which incentivise growers to participate in the coalitions;

• A representative approach to water quality monitoring;

• Mandated and voluntary use of digital tools, including e-reporting and publicly-accessible data repositories, to minimise costs of data collection and reporting requirements.

Recent review of monitoring and reporting regime

In February 2018, the State Water Board amended and updated the WDR for growers within the Eastern San Joaquin River Watershed (within the Central Valley region) that are “Members” of a Third-Party Group. These amendments were the result of an extensive consultation process that commenced in February 2016. At the heart of the review is the broad question whether the existing regime strikes the appropriate balance between providing sufficient data to evaluate the ILRP and ensuring that the burden of monitoring regime for growers satisfies the test of bearing a reasonable relationship to the need for and benefit of monitoring. In theory, various institutional, legal or technological factors could contribute to a decision to change the existing regime, for example:

• Evaluation of existing data provided by monitoring may lead to the conclusion that the existing monitoring regime is;

• Changes in the cost of the monitoring regime due to technological innovation could reduce the regulatory burden of monitoring for growers, leading to a re-balancing of monitoring requirements;

• Methodological innovations could lead to a change in the monitoring approach towards using new and improved methods;

• Evaluation of the existing third party-based mechanism may reveal unintended consequences which need to be addressed.

Methodological innovation was perhaps the most important factor underpinning changes. In particular, the Order introduces a new indicator for monitoring potential nitrate impacts from agriculture: the AR metric—an indicator of the amount of nitrogen in the soil that could potentially reach groundwater as nitrate (see Part IV for details). This metric is considered scientifically robust and less prone to misinterpretation; both key factors underpinning the decision to require de-identified field-level reporting of AR data. In response to concerns expressed by some stakeholders that the existing monitoring regime is inadequate, the State Water Board also directed several revisions to data reporting requirements, in particular:

• to require more granular, anonymous field-level reporting of growers’ land management practices and nitrogen application (related to the AR metric) to the Central Valley Water Board.

• to expand the requirements currently imposed only on Members in high vulnerability groundwater areas on all Members, with some limited exceptions.

Despite concerns raised by some stakeholders, the State Water Board continued to support the representative monitoring approach, considering monitoring farm discharge points as “impractical, prohibitively costly, and often ineffective method for compliance determination”. Thus, despite suggestions in the relevant literature that the cost of wireless water quality sensor networks has declined sufficiently in recent years to make monitoring water quality on-farm a potentially feasible option, at least in this context this does not yet appear to be the case.

The State Water Board also continues to support the third party (coalition-based) approach. However, it recognises that “concerns with privacy and protection of proprietary information may create strong incentives in support of a framework where the third party retains most information on farm-level management practice and water quality performance rather than submitting that information to the regional water board and, by extension, making it available to the public” (State Water Board, 2018, p. 21[20]). This finding suggests several possible unintended or undesirable consequences of supporting the third party mechanism. First, this support could be seen as legitimising the view that growers have some kind of “right” to confidentiality. Second, the third party may encounter a conflict of interest in that, on the one hand, it needs to report “sufficient” detail to the regulator (which may include farm-level data and even potentially data which identifies individuals), but on the other hand, its members favour reporting of aggregated data only. While the State Water Board has been careful to clarify that it does not recognise any right to privacy in relation to field level data, grower submissions during the consultation process cited an expectation of confidentiality for growers participating in coalitions (Agricultural Council of California et al., 2017[21]), and thus the regulator needs to be continually attentive to these issues and ensure that there is appropriate regulatory oversight of the third party.

Notes

← 1. http://www.waterboards.ca.gov/water_issues/programs/nps/docs/plans_policies/nps_iepolicy.pdf, accessed August 2018, AR 36138-36157.

← 2. The State Water Board works with the regional water boards and sets state-wide standards and policies.

Source: Case Study 7, Part IV.

References

[8] Adams, W. (2018), “Conservation by Algorithm”, Oryx, Vol. 52/01, pp. 1-2, https://doi.org/10.1017/s0030605317001764.

[21] Agricultural Council of California, I. et al. (2017), Comment Letter - Second Draft Waste Discharge Requirements General Order No. R5-2012-0116 for Growers Within the Eastern San Joaquin River Watershed that are Members of the Third-Party Group – SWRCB/OCC Files A-2239(a)-(c), https://www.waterboards.ca.gov/public_notices/comments/a2239ac/comments20171205/tricia_geringer.pdf (accessed on 29 August 2018).

[10] Arner, D., J. Barberis and R. Buckey (2016), “FinTech, RegTech, and the Reconceptualization of Financial Regulation”, Northwestern Journal of International Law & Business, Vol. 37/3, p. 371, https://heinonline.org/HOL/Page?handle=hein.journals/nwjilb37&id=383&div=&collection= (accessed on 18 September 2018).

[6] Burton, R. and G. Schwarz (2013), Result-oriented agri-environmental schemes in Europe and their potential for promoting behavioural change, https://doi.org/10.1016/j.landusepol.2012.05.002.

[15] Carletto, G., D. Jolliffe and R. Banerjee (2015), From Tragedy to Renaissance: Improving Agricultural Data for Better Policies, The World Bank, https://doi.org/10.1596/1813-9450-7150.

[9] Coglianese, C. et al. (2017), “Regulating by Robot: Administrative Decision Making in the Machine-Learning Era”, http://ssrn.com/abstract=2928293.

[11] Directorate-General for Parliamentary Research Services (European Parliament) (2018), Precision agriculture in Europe - Legal, social and ethical considerations, https://doi.org/10.2861/278.

[7] Duncan, R. (2014), “Regulating agricultural land use to manage water quality: The challenges for science and policy in enforcing limits on non-point source pollution in New Zealand”, Land Use Policy, https://doi.org/10.1016/j.landusepol.2014.06.003.

[5] Gregory, R. and T. Satterfield (2002), “Beyond perception: The experience of risk and stigma in community contexts”, Risk Analysis, https://doi.org/10.1111/0272-4332.00017.

[23] Hand, D. (2018), “Statistical challenges of administrative and transaction data”, Journal of the Royal Statistical Society: Series A (Statistics in Society), Vol. 181/3, pp. 555-605, https://doi.org/10.1111/rssa.12315.

[12] Jones, J. et al. (2017), “Brief history of agricultural systems modeling”, Agricultural Systems, Vol. 155, pp. 240-254, https://doi.org/10.1016/J.AGSY.2016.05.014.

[19] Lange, B. and A. Gouldson (2010), “Trust-based environmental regulation”, Science of the Total Environment, Vol. 408, pp. 5235–5243, https://doi.org/10.1016/j.scitotenv.2010.07.052.

[14] Legg, W. and D. Blandford (2019), “The Role of International Organizations in Agricultural Policy Analysis”, in World Scientific Series in Grand Public Policy Challenges of the 21st Century, Global Challenges for Future Food and Agricultural Policies, World Scientific, https://doi.org/10.1142/9789813235403_0016.

[22] Lubulwa, M. et al. (2010), Statistical integration in designing Australian farm surveys, ABARE-BRS Conference Paper 10.13, Speke Resort, Kampala, Uganda, http://www.fao.org/fileadmin/templates/ess/documents/meetings_and_workshops/ICAS5/PDF/ICASV_5.2_099_Paper_Lubulwa.pdf (accessed on 21 August 2018).

[1] Macey, G. (2013), “The Architecture of Ignorance”, Utah L. Rev, Vol. 21/1627.

[16] Martínez-Blanco, J. et al. (2014), “Application challenges for the social Life Cycle Assessment of fertilizers within life cycle sustainability assessment”, Journal of Cleaner Production, Vol. 69, pp. 34-48, https://doi.org/10.1016/J.JCLEPRO.2014.01.044.

[4] Myles, F., R. Duncan and A. Brower (2016), “Measuring to manage: Reconfiguring people–water relations through water measurement standards and technologies in New Zealand”, Environment and Planning C: Government and Policy, https://doi.org/10.1177/0263774X15614676.

[3] OECD (2018), “Navigating pathways to reform water policies in agriculture”, [COM/TAD/CA/ENV/EPOC(2018)16/FINAL], Paris.

[24] OECD (n.d.), Short-Term Economic Statistics (STES) Administrative Data: Two Frameworks of Papers, http://www.oecd.org/sdd/short-termeconomicstatisticsstesadministrativedatatwoframeworksofpapers.htm (accessed on 21 August 2018).

[13] Petsakos, A. and P. Jayet (2010), Evaluating the efficiency of a N-input tax under different policy scenarios at different scales, http://ageconsearch.umn.edu/bitstream/109397/2/Petsakos_Jayet.pdf (accessed on 19 April 2018).

[20] State Water Board (2018), Order WQ 2018-0002, State of California State Water Resources Control Board, http://www.swrcb.ca.gov/water_issues/programs/agriculture/docs/ILRP_expert_panel_final_report.pdf (accessed on 29 August 2018).

[2] Stephenson, K. and L. Shabman (2017), “Where Did the Agricultural Nonpoint Source Trades Go? Lessons from Virginia Water Quality Trading Programs”, Journal of the American Water Resources Association, https://doi.org/10.1111/1752-1688.12565.

[17] Tukker, A. and E. Dietzenbacher (2013), “Global Multiregional Input-Output Frameworks: An Introduction and Outlook”, Economic Systems Research, Vol. 25/1, pp. 1-19, https://doi.org/10.1080/09535314.2012.761179.

[18] VanderZaag, A. et al. (2013), “Towards and inventory of methane emissions from manure management that is responsive to changes on Canadian farms”, Environmental Resource Letters, Vol. 8, p. 035008, https://doi.org/doi:10.1088/1748-9326/8/3/035008.

[25] Williams, R. et al. (2013), A critical examination of the role of overseer® in modelling nitrate losses from arable crops, http://flrc.massey.ac.nz/workshops/13/Manuscripts/Paper_Williams_2013.pdf.

Notes

← 1.  See https://www.nrcs.usda.gov/wps/portal/nrcs/detail/national/programs/financial/?cid=nrcseprd1328426, accessed September 2018.

← 2. “Sequencing requirements” refers to hybrid mechanisms in which, in order to participate in an economic instrument, certain other requirements need to be satisfied first. These other requirements could be technology standards, requirements to first exhaust options for on-site mitigation, requirements to purchase first from certain “pools” of credits before accessing others, etc.

← 3. An example would be to provide farmers with their own monitoring device and track the collective achievement of a group rather than individual performance, unless specific group targets are not met. Case Study 7 provides such an example.

← 4. An example of using a complex model for agri-environmental policy is the use of the OVERSEER® model by regional authorities across New Zealand in developing plans to manage water quality. OVERSEER® is a computer model originally designed to assist farmers and their advisors with on-farm nutrient use, for estimating nitrate losses from individual pastoral farms. See Williams et al. (2013[25]).

← 5. See also section 4.4.1. This paper does not assume that iterative policy updates are necessarily desirable. In addition, Stephenson and Shabman (2017[2]) point out that combining regulatory requirements with environmental markets may create lacklustre demand for environmental credits if regulatory requirements take a sequencing approach in which buyers may only enter the market after having satisfied certain technological requirements.

← 6. In particular, discussions of the use of modelling to support water quality policies for agriculture often centre on the notion that nonpoint sources (including agriculture) are sources for which it is not possible or prohibitively costly to measure and attribute emissions to particular sources (farms).

← 7. A range of other factors also contribute to the inability to link datasets, including: the absence of common linking variables (which enable record matching) (Lubulwa et al., 2010[22]); high costs or lack of resources or expertise needed to perform the linkages (Hand, 2018[23]); and lack of interoperability between datasets (e.g. different definitions with no rule to “translate” definitions in one dataset to match up with another) (Hand, 2018[23]).

← 8. The same is also often true of institutional policies governing the collection and use of farm-level data for research, and of contracts and other agreements governing the collection and use of farm-level data by the private sector; while this section focusses on the case of public agencies, much of the discussion is relevant to these other contexts.

← 9. OECD (n.d.[24])defines “administrative data” to have the following features:

• the agent that supplies the data to the statistical agency and the unit to which the data relate are usually different: in contrast to most statistical surveys;

• the data were originally collected for a definite non-statistical purpose that might affect the treatment of the source unit;

• complete coverage of the target population is the aim;

• control of the methods by which the administrative data are collected and processed rests with the administrative agency.

Chapter 4. Issues which may prevent digital opportunities from being realised