Computers and Electronics in Agriculture 78 (2011) 28–37

Contents lists available at ScienceDirect

Computers and Electronics in Agriculture

journal homepage: www.elsevier .com/locate /compag

Towards automated compliance checking based on a formal representation ofagricultural production standards

Edward Nash a, Jens Wiebensohn a,⇑, Raimo Nikkilä b, Anna Vatsanidou c,d, Spyros Fountas c,d, Ralf Bill a

a Rostock University, Faculty of Agricultural and Environmental Sciences, Chair of Geodesy and Geoinformatics, Justus-von-Liebig-Weg 6, 18059 Rostock, Germanyb Helsinki University of Technology, Department of Automation and Systems Technology, P.O. Box 5500, 02015 TKK, Finlandc University of Thessaly, School of Ag. Science, Fytokou St. N. Ionia Magnisias, 38446 Volos, Greeced Centre for Research and Technology, Thessaly, 1st Industrial Zone, 38500 Volos, Greece

a r t i c l e i n f o a b s t r a c t

Article history:Received 15 September 2010Received in revised form 29 April 2011Accepted 10 May 2011

Keywords:MetadataOntology languageRulesDecision supportKnowledge management

0168-1699/$ - see front matter � 2011 Elsevier B.V. Adoi:10.1016/j.compag.2011.05.009

⇑ Corresponding author.E-mail address: jens.wiebensohn@uni-rostock.de (

Production standards in the form of legal regulations or quality assurance labels are playing an increas-ingly important role in farming. Each farm must therefore gather information on all standards whichapply, which may vary from field-to-field, and ensure that they are respected during operations. Thisinformation may be provided on paper or as electronic documents, by the standards publishers or byadvisors. Together with the need to document compliance, the need to collect and process the require-ments is becoming increasingly burdensome for farmers.

In this paper, two questions are addressed: whether an automation of the compliance checking is pos-sible, in order to assist the farmer by proactively warning against ‘forbidden’ operations, and how the def-inition of the production standard may be formally represented in order to clearly and unambiguouslyinform the farmer as to what is required. This formal representation also forms one of the prerequisitesfor any automated assessment.

As an initial step, a general model of production standards was developed and applied to some commonstandards in European agriculture. Based on this model, separating standards into metadata and a list ofindividual rules (check points), a formal representation was developed and an assessment was made as towhether an automated compliance check was feasible.

� 2011 Elsevier B.V. All rights reserved.

1. Introduction

Production and management standards are becoming increas-ingly important in agriculture: an increasing number of legal regu-lations to ensure food safety and agri-environmental good practiceare binding for all farmers, whilst voluntary standards and labels todemonstrate compliance to stricter requirements and ultimatelygain a higher price for agricultural produce are an important toolfor farmers to market their products or enable them to sell toparticular buyers and markets (Deaton, 2004, Jahn et al., 2005,Fulponi, 2006). Examples of legal regulations are laws affectinguse of fertilisers, plant protection products, seed types, etc. Volun-tary standards may be legally regulated, such as the EU Organicstandard (EC Regulation 834/2007), or may be privately-run indus-try standard such as GlobalGAP (GlobalGAP, 2007). Adherence toparticular standards may be motivated by direct financial benefits,such as subsidy payments, being linked to this, such as is the casewith the European ‘Cross-Compliance’ regulations.

ll rights reserved.

J. Wiebensohn).

Each farm must potentially adhere to a large number of stan-dards, and it is possible, or even likely, that different parts of thefarm must be managed according to different standards, e.g. differ-ent crops being sold to different buyers who stipulate their ownproduction standards, or parts of the farm falling in a specific areasuch as a water catchment protection area where tighter environ-mental regulations are enforced. Additionally, the laws applying tothe farm vary according to the country or federal state, or in somecases even smaller administrative units. Each farm, or even eachfield or even partfield, must therefore be considered a potentiallyunique case in being managed according to a unique constellationof standards. Additionally, standards vary through time as newversions are produced.

Given this large number of different standards in use and theneed for farmers to work with the correct standards, in the correctversions, active support from the farm software during the deci-sion-making process in order to ensure that management decisionssuch as fertilisation and spraying plans conform to the relevantstandards is desirable. The current procedure for assessing compli-ance to standards is typically that the farmer must document thecorrect completion of the procedures and actions as required bythe standard. Additionally, some standards require that certain

E. Nash et al. / Computers and Electronics in Agriculture 78 (2011) 28–37 29

information is presented in a particular form. The documentationis then periodically audited, usually by an external certificationbody, together with an on-site inspection. The farmer must there-fore understand what is required by the standard and take accountof the requirements during the planning and performing of actions– this is frequently done through the use of self-assessments e.g. aschecklists. The farmer must also be able to produce documentationto satisfy the certification body that they have met the require-ments, and so should know in advance what is required, how itshould be presented and integrate collection and management ofrequired data in operational processes. Such a service is offerede.g. for common standards in use in Germany by KKL-Service(2010). This is however not integrated with the farmers’ existingsoftware, where farm data are already held, does not offer the pos-sibility for automated assessment and is restricted to the standardswhich have been analysed and prepared by KKL.

This paper therefore addresses two research questions:

1. Whether an automatation of self-assessment of compliance tostandards is feasible.

2. How standards may be formally represented in such a way thatpersonalised checklists can be easily generated by farmersthemselves and that the individual requirements defined bythe standard are clearly and unambiguously defined such thatthey may potentially be interpreted by a machine.

In the next section, a general model for the composition of anagricultural production and management standard is presented.The analysis of whether compliance-checking may be automated,based on transforming existing standards to this model, follows.The formal encoding for standards is then introduced in three sec-tions corresponding to parts of the general model of standards.Finally, there is some discussion regarding the work presentedhere and what developments may be necessary in order to improvethe definitions of agricultural standards with respect to automatedassessment and to enable their integration into agricultural soft-ware on an ad hoc basis.

2. Structure of agricultural standards

Based on the analysis of representative agricultural productionand management standards, which were previously presented ageneral structural model of an agricultural standard, together withfour criteria which must be met in order to enable automated

Fig. 1. General structural model

compliance checking (Nash et al., 2009a,b). As these form the basisfor the work presented here, they will be reviewed in detail.

An agricultural standard may be considered as being composedof a set of rules together with metadata describing the publisher,the intention of the publisher, the spatial and temporal range ofvalidity, the target audience, procedures in the event of non-compliance, a definition of terms used. Additionally, each rulemay have certain metadata attached to it regarding how compli-ance to that rule is to be assessed, and whether all rules must becomplied with in order for the whole standard to be complied withor whether only a certain percentage of individual rules must bemet. Each rule is effectively a predicate (i.e. a logical statementwhich may be evaluated to true or false), together with a conclu-sion (i.e. compliance or violation of the standard). Rules may beclassified as either an obligation (‘the standard is complied withonly if the farmer does x’),) or a prohibition (‘the standard is notcomplied with if the farmer does y’). Additionally, rules may re-quire that particular actions are documented, whilst not proscrib-ing how they should be performed. Although these may beconsidered as obligations, they are treated separately as they donot directly affect the decision-making related to field operations(e.g. the volume of nitrogen fertiliser to be applied). Individualrules may also be considered as having some metadata such asdescribing which operations they apply to, what data may be usedto assess compliance etc. This model is presented graphically inFig. 1.

Current agricultural standards are not explicitly presented inthe structural form presented in Fig. 1; most legal regulations arepresented as texts, whether paper or electronic, whilst in the bestcase the standard may be presented as a checklist of individualrules (e.g. GlobalGAP). Where the standard is presented as a text,the identification of individual rules may not be straightforward.Any standard can however be converted to the form specified bythis model (Vatsanidou et al., 2009).

3. Determining the potential for automated assessment

In order to enable the automated assessment of each rule, fourprerequisites must be met:

1. The rule must be encoded in a machine-readable form. This maybe hard-coded as algorithms in the software or take the form ofa transfer format (e.g. XML-based) which the software perform-ing the assessment can read.

of an agricultural standard.

30 E. Nash et al. / Computers and Electronics in Agriculture 78 (2011) 28–37

2. The rule, and all terms used in defining it, must be capable ofbeing interpreted by the software. This has two aspects. Firstly,all the concepts used within the definition of the rule (e.g.nouns such as farm, production unit, crop, nitrogen or fertiliser,verbs such as weed, sow or spray or adjectives such as certifiedor organic) can be correctly ‘understood’ by the software inthe context in which they are used in the rule. Secondly, therules must be computable, that is to say that it must be capableof being evaluated by an automaton (e.g. a Turing machine)according to computability theory

3. Each rule must have a discrete outcome which can be deter-mined by a computer. That is to say that compliance to eachrule must be assessed using computational models using digitaldata inputs and producing discrete outcomes (true/false) asopposed to value judgements.

4. The required data inputs for assessment must be available indigital form at the point of assessment. This may be alreadyexisting data held in one or more databases which are accessi-ble by the software, either internally or accessed via web ser-vices, or gathered on-demand by online sensors.

The first of these prerequisites is the basic motivation for thecurrent paper whilst the second one defines the requirements ofa transfer format, namely to enable the transfer of computablerules and the definition of all concepts used within these rules.The former implies use of a transfer format for rules, and the lattera transfer format for ontologies where an ontology is a ‘‘formal,explicit specification of a shared conceptualisation’’ (Gruber,1993) or a definition of terms used and the relationships betweenthem. An ontology may in this case be considered as a formalspecification of a vocabulary.

3.1. Production of checklists

Three common agricultural standards were considered, namelythe EU Cross-Compliance directives, the EU Organic regulationsand the GlobalGAP standards for good agricultural practice. Thesecover mandatory rules for agricultural production Europe-wide,legally-regulated voluntary rules for products which are sold with-in the European Union and a private standard. The methodologyfor preparing the checklists is described in more detail in Nash etal., 2009a,b and Vatsanidou et al., 2009.

3.2. Potential for automating compliance checking

Having introduced the criteria for the assessment of the poten-tial for automated compliance checking and the standards consid-ered here, now there will be presented some preliminary results.

3.2.1. Forms of publication and qualitative evaluationGlobalGAP currently publishes standards in a checklist form,

specifying the individual rules. Many of the terms used are definedin a glossary forming an appendix to the standard. However, therules are still specified using a natural language and so are notunambiguous. Many of the rules fulfil all four prerequisites listedpreviously and so may theoretically be relatively easily automated.An example is rule 6.4.2: ‘Are only biocides and plant protectionproducts used that are officially registered in the country of use,and for use post-harvest on the harvested crop being protected?’(GlobalGap CC, 2007) – assuming the products applied to the cropare recorded in the FMIS, and that a list of all allowed post-harvesttreatments is available, non-compliance may be easily identified.However, this is not true for all rules: 6.5.1 demands ‘Is the riskof contamination by glass or any other physical contaminants pre-vented?’ – apart from being grammatically inelegant, it is hard toenvisage how testing against this rule may be meaningfully

automated (perhaps by continuously recording the position ofthe harvested crop and of all potential contaminants?). Ultimately,what is being asked for by this rule is a production and manage-ment system which minimises this risk rather than regulatingany specific operational action which will be recorded.

The EU Organic standard is not published as a checklist, butrather as a series of clauses in a text document which have beentransformed to the rules presented here using the methodology de-scribed previously. Of these, the assessment is that a smaller pro-portion may be easily automated than for GlobalGAP, but thatthere is also a smaller proportion which may not be automated.Although some rules may be easily automated (e.g. 1.4.3: ‘Haveyou taken care not using mineral nitrogen fertilisers?’, assumingall products applied and their contents are recorded), some rulesrequire a complex analysis – e.g. 1.5.1: ‘Have you implemented aplanned multiannual crop rotation that will maintain or improvethe fertility and biological activity (organic matter) of your soil?’requires not only that the field management plan is recorded butalso that the software can assess whether it will improve the fertil-ity and biological activity. Although not impossible, this is likely arule which makes a reliable testing difficult. 8.2: ‘If the holding ispartly converted does the operator separate the organic land andproducts from those of non-organic?’ may be automatically as-sessed if post-harvest tracking information is available, but other-wise requires an assessment of the management systems in place,which is again hard to automate.

Finally the cross-compliance regulations were considered.Here the rules as defined at the EU level are very general, as def-inition of good agricultural practice, and hence exact rules to befollowed, are implemented by the individual member states.Nevertheless, of the rules which are defined at the EU level(from the checklist prepared by Vatsanidou et al., 2009), thereare again rules which are easier and rules which are harder toautomate. As an example, for 3.1: ‘Are you maintaining soilstructure through appropriate measures, like appropriatemachinery use?’ it is straightforward to record the required data(e.g. which machines are used), but hard to determine whetherthis has helped maintain soil structure. 1.1: ‘Are you protectingsoil through any of the following appropriate measures: mini-mum soil cover, minimum land management reflecting site-spe-cific conditions or retain terraces?’ is, in contrast, relativelystraightforward to automate testing, although defining the re-quired minimum could prove problematic.

3.2.2. Quantitative evaluationNow the quantitative evaluation of the rules from the regula-

tions and standards Crops Base from GlobalGAP, EU Organic Regu-lation, and German Düngeverordnung as an implementationexample of the EU Cross-Compliance directives, namely theNitrates Directive, is presented. For each rule, a categorisationhas been undertaken based on the schema presented in Table 1.Firstly, each rule was categorised as to whether it represents anobligation, a prohibition or documentation for the farmer. Subse-quently, four parameters were assessed relating to the possibilityof formal representation (the first prerequisite), of automated ma-chine interpretation (the second prerequisite), the objectivity ofthe required assessment (the third prerequisite) and to the avail-ability of the required data (the fourth prerequisite).

In the rule category there is the group of Obligation/Prohibitionwith a share of about 85% of all surveyed rules. The remaining15% of rules are identified as Documentation, which are assumedto be not a decidable rule in the sense of this paper. All followingvalues therefore are referring to the first group of Obligation/Prohibition. On the second criteria (possibility of formal representa-tion) it was determined that it should be possible to formulate allObligations and Prohibitions as statements in first-order logic. High

Table 1Categorisation schema for assessing suitability of rules for automated compliance assessment.

Parameter (notes) Values Description

Rule category Obligation A rule mandating some actionProhibition A rule prohibiting some actionDocumentation A rule requiring the collection or presentation of some documentation

Possibility of formal representation Possible The rule may be formally formulated as a statement in first-order logicImpossible The rule cannot be formally formulated as a statement in first-order logic

Interpretability Possible The concepts can be formally defined and the rule is computable based on computing theoryImpossible Either the concepts cannot be formally defined and/or the rule is incomputable based on

computing theoryObjectivity

(assessment was made assuming rules wereadjusted for interpretability where necessary)

Objective Compliance may be assessed objectively with a discrete outcomeSubjective Either compliance may only be subjectively assessed and/or the outcome is not discrete

Availability of digital data Available Data required to assess compliance is widely available in digital formForeseeable Data required to assess compliance is likely to be available in digital form in the foreseeable

futureUnlikely It is unlikely that the data required to assess compliance can be made available in digital form,

or the rule is formulated such that it does not require assessment of data

E. Nash et al. / Computers and Electronics in Agriculture 78 (2011) 28–37 31

percentage values were also determined in the categoriesInterpretability (95%) and Objectivity (97%). The assessment of digi-tal data availability lead to result that most (82%) could becomeavailable somewhere in the future. However, for only 9% of the rulesthe required data was determined to be currently available, and forthe remaining 9% it was determined that it was unlikely that the re-quired data could ever be available digitally. It should be noted thatthe determination of these values is inevitably somewhat subjec-tive, but based on the example rules a potential for automated com-pliance assessment of about 88% of all rules can be assumed, and no

Fig. 2. Determined percentages of rules determined to meet automation criteria a

significant differences between the single standards were observed(see illustration in Fig. 2 and for details in Table 2).

3.3. Results of analysis

From this simple quantitative evaluation of the compliance ofthree common standards with the stated automation criteria, it isclear that there is a large potential for farmers to be supportedby use of automated systems which can assess compliance withcommon management standards. However, it is equally clear that

nd availability of digital data required for assessment of compliance to rules.

Table 2Determination of the potential for automated compliance assessment based on rule-by-rule assessment (see Table 1 for definition of terms).

Standard Parameter rule category Formal representation[% rules]

Machine-interpretable[% rules]

Objective[% rules]

Data availability (available/foreseeable/unlikely) [% rules]

Crops base Obligation/Prohibition (n = 98)(Documentation n = 22)

100 95 99 9/81/10

EU organic Obligation/Prohibition (n = 74)(Documentation n = 12)

100 93 96 8/87/5

German DüV Obligation/Prohibition (n = 29)(Documentation n = 3)

100 97 93 7/79/14

Overall Obligation/Prohibition (n = 238)Documentation n = 37)

100 95 97 9/82/9

32 E. Nash et al. / Computers and Electronics in Agriculture 78 (2011) 28–37

it would be nearly impossible to realise this potential immediatelydue to the lack of availability of the required data in digital form.Nonetheless, it is foreseeable that much of the data which is cur-rently available could in future be collected, managed and trans-ferred digitally, thus allowing assessment of compliance to up to90% of the agricultural production rules to be automatically per-formed. The question of how this assessment may be supportedby technical standards and information technology, given the het-erogeneous conditions outlined previously, is therefore a relevantone. The remainder of this paper therefore presents some initialideas in this field, specifically addressing the question as to howthe standards and the individual rules may be structured andencoded in a formal machine-readable way to enable them to beseamlessly transferred between systems, e.g. from rules publishersand administrators to farmers’ software.

4. Formal encoding of agricultural standards

Based on the model previously presented, the basic structure fora formal machine-readable representation of agricultural standardsconsists of three parts; metadata about the standard, a definition ofthe terms used in defining the rules, and the rules themselves to-gether with the rule-specific metadata. The XML schema developedfollows this structure, and its outline is shown in Fig. 3.

4.1. Metadata for agricultural standards

In this section the modelling of the metadata required fordescribing agricultural standards will be described. Metadatamay be defined as a description of the actual contents, and thusthe metadata here is a description of the standard which is

Fig. 3. Top-level structure of XML schema f

necessary to understand its context and to evaluate whether it isrelevant to a particular case. This includes common metadataterms such as publisher, date of publication, etc. as well as special-ist items including spatiotemporal validity and relationships toother standards. There are many existing XML schema for metada-ta, such as Dublin Core (DCMI, 2004) and ISO19115 (ISO, 2003) butnone of them cover fully the details for agricultural standards.Fig. 4 illustrates the elements which are included in the agricul-tural standard metadata. It can be seen that elements from DublinCore (prefix dc) and ISO19136 Geography Markup Language (ISO,2007, prefix gml) are used for standard metadata elements andspatiotemporal representation respectively.

One complication in this case is the representation of the spatialregion to which a standard applies. Two distinct categories can beidentified; the production region (i.e. where the farm is located)and the end product region (i.e. where the farm products willultimately be sold to consumers. The first of these is relevant withrespect to e.g. fertiliser regulations which are valid for all farmswithin a particular country. The second of these is relevant for par-ticular product standards e.g. organic production standards such asthe EU organic regulations are valid for all products sold as organicin the European Union, wherever in the world they were produced.This requires the inclusion of separate meta data elements for eachcase. Within each element, there are multiple ways in which thevalid region may be defined: most common is likely to be the loca-tion name, e.g. the name of the country to which the law applies.Alternatively, the standard may apply to a particular class of re-gion, e.g. nitrate vulnerable zones (NVZ). However, it is also feasi-ble that a geometric region is directly defined using coordinates.These individual region types may also be combined, e.g. to specifythat a standard applies only to nitrate vulnerable zones in Englandand Wales it is necessary to specify the intersection of the feature

or representing agricultural standards.

Fig. 4. Elements for representing metadata for an agricultural standard.

Fig. 5. Structure for representing the valid region for an agricultural standard.

E. Nash et al. / Computers and Electronics in Agriculture 78 (2011) 28–37 33

34 E. Nash et al. / Computers and Electronics in Agriculture 78 (2011) 28–37

class ‘nitrate vulnerable zone’ with the union of the location names‘England’ and ‘Wales’. In order to allow the correct specification ofregions, the structure shown in Fig. 5 was developed. An exampleof the usage of the whole metadata encoding is given as an XMLfragment below:

<ff:AgriStandardMetadata

metadataId=’’urn:agristandards:metadata:uk:

defra:nitrate:2009–04-01’’>

<dc:title>Guidance for Farmers in Nitrate

Vulnerable Zones</dc:title>

<dc:description>The Guidelines for Farmers in

Nitrate Vulnerable Zones (NVZ)

describe what farming operations are allowed

in NVZ in England and

Wales.</dc:description>

<dc:creator>Department for Environment, Food and

Rural Affairs</dc:creator>

<ff:version>2009-04-01</ff:version>

<ff:publishedDate>2009-04-01T00:00:00+

01:00</ff:publishedDate>

<gml:validTime>

<gml:TimePeriod gml:id=’’timePeriod01’’>

<gml:beginPosition>2009-04-01T00:00:00+

01:00</gml:beginPosition>

<gml:endPosition>2010–12-31T00:00:00+

01:00</gml:endPosition>

</gml:TimePeriod>

</gml:validTime>

<ff:productionRegion>

<ff:regionIntersection>

<ff:featureClass>NitrateVulnerableZone

</ff:featureClass>

<ff:regionUnion>

<gml:locationName>England></

gml:locationName>

<gml:locationName>Wales</gml:locationName>

</ff:regionUnion>

</ff:regionIntersection>

</ff:productionRegion>

<ff:implementsStandard>

urn:agristandards:eu:nitrate:1991-12-12

</ff:implementsStandard>

<ff:classification>mandatory legal regulation

</ff:classification>

<ff:originalText>

http://www.opsi.gov.uk/si/si2008/pdf/uksi_

20082349_en.pdf</ff:originalText>

Declaration(Class(duevo:FertiliserWith

SignificantNutrientContent))

EquivalentClasses(

duevo:FertiliserWithSignificant

NutrientContent

ObjectUnionOf(

ObjectAllValuesFrom(duevo:hasPhosphateContent

percentages:PercentGreater0.5)

ObjectAllValuesFrom(duevo:hasNitrogenContent

percentages:PercentGreater1.5)))

SubClassOf(duevo:FertiliserWithSignificant

NutrientContent agrovoc:Fertiliser)

Declaration(Class(duevo:FertiliserWith

SignificantNitrogenContent))

EquivalentClasses(

duevo:FertiliserWithSignificant

NitrogenContent

ObjectAllValuesFrom(duevo:has

NitrogenContent

percentages:PercentGreater1.5))

SubClassOf(

duevo:FertiliserWithSignificant

NitrogenContent

</ff:AgriStandardMetadata>

duevo:FertiliserWithSignificant

NutrientContent)

Declaration(Class(duevo:FertiliserWith

SignificantAvailableNitrogenContent))

EquivalentClasses(

duevo:FertiliserWithSignificantAvailable

NitrogenContent

ObjectAllValuesFrom(

duevo:hasProportionSolubleInCalciumChloride

percentages:PercentGreater10))

SubClassOf(

duevo:FertiliserWithSignificantAvailable

NitrogenContent

duevo:FertiliserWithSignificant

NitrogenContent)

4.2. Ontologies for agricultural standards

One criteria for enabling automated assessment of rules is thatthe terms used in defining the rules are unambiguously defined ina machine-readable form. Also for a reliable manual assessment ofrules, the unambiguous definition of terms is essential in order toavoid differing interpretations between those defining the rulesand those assessing compliance. Ontologies are a tool for definingconcepts and the relationships and differences between them in aformal way, and have particularly risen to prominence as part ofthe semantic web (Berners-Lee et al., 2001). The most commonlanguage for modelling ontologies is the W3C Web OntologyLanguage (McGuiness & van Harmelen, 2004), which also providesan XML-based representation for interchange of ontologies, to-gether with a functional, more human-readable syntax. Ontologies

are widely used e.g. in the biomedical domain where large collec-tions of orthogonal interoperable ontologies are available (e.g. athttp://www.obofoundry.org). There has been some recent interestin ontologies and their role in data exchange in the agricultural do-main (e.g. Sall et al., 2009; Maliappis, 2009, FAO, 2010), but there isno widespread acceptance of ontologies in practical use.

However, many standards for agriculture include a definition ofterms in the form of a glossary or legal definitions at the start of atext. These may with a little effort be converted to a formal ontol-ogy. As an example a definition of a term is presented which isgiven in the German fertiliser regulations (DüVo) and how it maybe represented using OWL. The DüVo defines a subclass of ‘Fertil-iser’, whose use is forbidden under certain conditions, ‘Fertiliserwith significant nutrient content’ is defined as a fertiliser with anutrient content >1.5% total nitrogen or >0.5% phosphate (P2O5).Further subclasses are defined for fertilisers with only significantN content (>1.5%) and significant available nitrogen content(>10% soluble in CaCl). The following presents these definitionsusing the OWL functional syntax, where it can be assumed thatthe general term ‘Fertiliser’ is defined in an Agrovoc ontology andonly terms and attributes specific to the DüVo must be additionallydefined. An XML version of these definitions (and more) is avail-able at http://schema.futurefarm.eu/ontologies/duevo.xml.

Forall ?fertiliser_application

(violation(DüVo):- And (

?fertiliser_application#agrovoc:

FertiliserApplication

?fertiliser_application[agrovoc:fertiliser->?ft]

?fertiliser_application[agrovoc:

application_area->?applied_to]

?fertiliser_application[agrovoc:date->

?application_date]

?ft#duevo:FertiliserWithSignificant

AvailableNitrogenContentWithoutManure

Or (

And (?applied_to#agrovoc:Cropland

External(ff:during(?application_date

External(ff:time_period(11-01 01-

31)))))

And (?applied_to#agrovoc:Grassland

External(ff:during(?application_date

External(ff:time_period(11-15 01-

31)))))

)))

E. Nash et al. / Computers and Electronics in Agriculture 78 (2011) 28–37 35

5. Formal representation of rules

As stated previously, the automated assessment of complianceto rules requires that the rules are encoded in a machine-readableform. As a practical form for transfer between systems, the eXten-sible Markup Language (XML) is the default choice for machine-readable representations. Although it is debatable whether all rulesmay be automatically assessed with any reliability, and the con-cept of ‘mechanical jurisprudence’ is rejected by experts in the fieldof law and artificial intelligence as fundamentally unrealistic (‘‘lawis more ‘rule-guided’ than ‘rule-governed’’’ – Gardner, 1987). How-ever, this does not mean that the rules may not be formally definedand represented, only that the interpretation may not always bedone automatically (Boer et al., 2007). It is therefore assumed thatthose rules which may be straightforwardly and unambiguouslyinterpreted automatically may be automatically processed, butthat where this is not possible the definition of the rule, togetherwith the relevant data, may be presented to the farmer or advisorin order to manually assess compliance. Furthermore, since theconversion of rules to a formal, logical-mathematical format istime-consuming and the main benefits will first be realised whenthe farm software is capable of reasoning with these rules, andfarm systems are capable of automatically supplying all requireddata, it is proposed that in the initial stages, the formal representa-tion of the rules should be optional, i.e. only the original natural-language version of each rule must be supplied. This considerablylowers the entry barrier to producing standards in a basic machine-readable way.

There have been many proposals for formal representation ofrules in XML, such as RuleML (Hirtle et al., 2006), SWRL (Horrockset al., 2004), WRL (de Bruijn, 2005) and R2ML (REWERSE, 2006).However, none of these have gained broad acceptance. A currentinitiative within the W3C is the creation of a Rules Interchange For-mat (RIF – Boley et al., 2009) which is, at the time of writing, at the‘Candidate Recommendation’ stage. This allows the representationof rules as sentences based on the individual atoms, functions andpredicates which may be identified in their natural languagerepresentation and is expected to become the future standard forrepresentation of rules on the Internet and in XML. As well as anXML-based format, the RIF defines a human-readable ‘presentationsyntax’ based on the Extended Backus-Naur Form for context-freegrammars.

As an example, the German fertiliser law (DüVo) states that it isforbidden to apply fertiliser with significant available nitrogen

Fig. 6. Structure for representation of an individual rule.

content, other than manure, on cropland between 1 Novemberand 31 January and on grassland between 15 November and 31January. This could be formally expressed in the RIF presentationsyntax as follows. Note that the definitions of terms defined inontologies (agrovoc, duevo) are referenced from the RIF rule def-inition.

Additional to the formal definition of the rule using RIF, somemetadata is required for each individual rule both to enable it tobe interpreted correctly and to enable efficient searching and pre-sentation of rules. It is assumed that the most common time forrequiring the detail of a rule is whilst planning an operation whichis likely to be affected by that rule, and therefore each rule may beannotated with a list of operations to which it applies, since thismay not be obvious from the definition of the rule itself, in orderto enable simple filtering. Similarly, the data elements whichmay be required in order to assess compliance to the rule maybe explicitly stated: it is expected that these would be referencedto concepts in the ontologies, as is the case for the specificationof concepts in the body of the rule. Although theoretically a gram-matically-correct natural-language sentence could be recon-structed from the RIF representation, for simplicity and in caseswhere the RIF version is not available, the original text should beincluded, together with a clarifying description to make interpreta-tion easier. The complete structure developed for representing anindividual rule is shown in Fig. 6. Due to the extreme verbosityof the RIF XML syntax, no example of its usage is given here. A fullexample file including usage of the RIF XML may be found at http://schema.futurefarm.eu/agstandard/example.xml.

6. Discussion

The evaluation presented up until now has largely been of a the-oretical nature, which has ignored many of the practical problemsassociated with data management and therefore automating com-pliance testing. The analysis of the feasibility of automating compli-ance assessment showed that conceptually the biggest hurdle is theavailability of digital data, with very little of the required data cur-rently available. However, automated digital data collection isbecoming more widespread, e.g. due to the use of ISOBUS-compliant

36 E. Nash et al. / Computers and Electronics in Agriculture 78 (2011) 28–37

hardware, information-intensive precision farming techniques,wired and wireless sensors, etc. (Sørensen et al., 2010). Farmersare therefore increasingly likely to gather digital data on all aspectsof the farm. This data must however be managed and, if necessary,transferred between components of the system. An improvementin agricultural data management and a standardisation of datatransfer mechanisms is therefore a prerequisite for automated com-pliance assessment. Possibilities for such improvement have beensuggested through use of web services (Murakami et al., 2007; Nashet al., 2009a,b), agricultural data warehouses (Steinberger et al.,2009) and web-based farm management systems (Nikkilä et al.,2010). It is therefore not entirely unrealistic to expect that withinthe foreseeable future, many of the required datasets may be widelyavailable to farmers.

Related to the requirement to have the data to assess compli-ance, it is also necessary to have an accurate definition of the rulesto be followed and the regions to which they apply. Since thesemay be different for each farmer, and many rules are defined inrelation to natural features (such as for use of fertilisers nearwatercourses), these must be interpreted in the context of eachfarm. Since the rules change periodically, and as discussed above,must currently often be inferred from the standards, a mechanismis required whereby published changes are automatically detectedby the software. A web-service based system whereby each stan-dards publisher may provide a standardised representation of therules directly to the software would provide such a mechanism.However, this presupposes that the standards are published asindividual rules, that there is a standardised format for these rulesand a standardised web-service interface with which to accessthem. Currently none of these is the case. However, this work illus-trates that such an approach is feasible: the existing standards andregulations may be rewritten as formal rules, which in turn may berepresented in a format suitable for machine-machine communi-cation and even automated interpretation. The technical basis forsuch a system is therefore available: what is more questionableis whether this system will be accepted by the bodies that publishthe agricultural rules. The move from publishing regulations as le-gal texts to publishing as individual rules in a machine-readableformat is a large change in procedure which would require signif-icant political will to make. A practical demonstration of the poten-tial benefits of such a system is therefore important in order toinfluence decision-makers.

As described in the prerequisites for software-based assessmentof rules, the need for the software to be able to ‘understand’ theterms used was stated. If software is to be able to flexibly interpretrules, it is therefore necessary that the vocabulary used is known tothe software. A harmonised vocabulary covering all agriculturalrules and data standards would ensure this, but is probably unre-alistic in practice. The expression of vocabularies using formalontologies, as it has been described in the present paper and theuse of mappings (so-called ‘ontology alignment’) between theseand inference engines is a more realistic solution, albeit one whichwould require significant developments from the current situation.However, there are a number of recent approaches using ontolo-gies in agriculture, and they are already in widespread use in otherfields (notably life sciences, see also http://www.obofoundry.org).A transfer of such techniques to the agricultural sector is thereforefeasible.

7. Conclusion

This paper has presented an analysis of the potential for currentagricultural management and production regulations and stan-dards to be automatically assessed, and an XML-based format fortheir formal representation. This format allows the transfer of

these standards between systems, which enables future ap-proaches to distributing knowledge on agricultural standards.Due to the increasing importance of such standards, and the inher-ent problems in incorporating such knowledge directly intosoftware, such an approach will become increasingly importantin future.

Whilst an automated internal assessment of compliance tostandards is the ultimate aim, and could to a large extent be real-ized using the combination of ontologies and formal presentationof rules described here, even the initial step of making the stan-dards available as a series of individual rules which may be filteredand combined by farmers in order to inform them of requirementsduring decision-making and to enable them to produce individu-ally-tailored checklists for manual compliance assessment shouldprove to be a great benefit to farmers. During the analysis processit had been worked out that the availability of data would be thelargest technical hurdle currently to an automated assessment –the increasing use of sensors and information-intensive techniquesin mainstream agriculture is however likely to ease these problemsin the foreseeable future. However, significant political-organisa-tional issues must also be overcome in order to produce rules inthe required form. A practical proof of the feasibility of such a sys-tem is therefore required in order to make a persuasive argument.

In current and future work, the practical possibility forautomated assessment of compliance to RIF-formulated rules foragriculture is therefore being tested. If this may be realized,together with s distribution system for agricultural standards, itwill effectively enable farmer decision-support software to beself-configuring for each farm or field and to be able to respectlegal and voluntary restrictions on planned operations. This willsignificantly ease the administrative burden on farmers imposedby complying with the wide range of regulations and standardswhich they must currently consider.

Acknowledgements

The research leading to these results has received funding fromthe European Community’s Seventh Framework Programme (FP7/2007-2013) under grant agreement no 212117. Preliminary ver-sions of parts of this paper were published at the JIAC 2009 inWageningen (NL) and the WCCA 2010 in Quebec (CA).

References

Berners-Lee, T., Hendler, J., Lassila, O., 2001. The Semantic Web. Scientific American284 (5), 34–43.

Boer, A., Di Bello, M., van den Berg, K., Gordon, T., Förhécz, A., Vas, R., 2007.Specification of the Legal Knowledge Interchange Format. ESTRELLA Deliverable1.1. Available at: http://www.estrellaproject.org/doc/D1.1-LKIF-Specification.pdf. Accessed 16 February 2010.

Boley, H., Hallmark, G., Kifer, M., Paschke, A., Polleres, A. and D. Reynolds. 2009. RIFCore Dialect. Available at http://www.w3.org/TR/rif-core/. Accessed 16February 2010.

DCMI (Dublin Core Metadata Initiative). 2004. Dublin Core Metadata Element Set,Version 1.1: Reference Description. Available at http://www.dublincore.org/documents/dces. Accessed 15 February 2010.

Deaton, B.J., 2004. A theoretical framework for examining the role of third-partycertifiers. Food Control 15 (8), 615–619.

De Bruijn, J. (ed). 2005. Web Rule Language (WRL). Available at http://www.w3.org/Submission/WRL/. Accessed 16 February 2010.

FAO (Food and Agriculture Organization of the United Nations). 2010. Agrovoc:from a thesaurus to an ontology. Available at http://aims.fao.org/agrovoc.Accessed 26 March 2010.

Fulponi, L., 2006. Private voluntary standards in the food system: The perspective ofmajor food retailers in OECD countries. Food Policy 31 (1), 1–13.

Gardner, A., 1987. An artificial intelligence approach to legal reasoning. MIT Press.GlobalGAP CC, 2007. Control Points and Compliance Criteria. Integrated Farm

Assurance. Combinable Crops. English Version V3.0-2 Sep07. Available at http://www.globalgap.org/cms/upload/The_Standard/IFA/English/CPCC/GG_EG_IFA_CPCC_CC_ENG_V3_0_2_Sep07.pdf. Accessed 16. February 2010.

Gruber, T., 1993. A translation approach to portable ontology specifications.Knowledge Acquisition 5 (2), 199–220.

E. Nash et al. / Computers and Electronics in Agriculture 78 (2011) 28–37 37

Hirtle, D., Boley, H., Grosof, B., Kifer, M., Sintek, M., Tabet, S. and G. Wagner. 2006.Schema Specification of RuleML 0.91. Available at http://ruleml.org/0.91/.Accessed 16 February 2010.

Horrocks, I., Patel-Schneider, P., Boley, H., Tabet, S., Grosof, B. and M. Dean. 2004.SWRL: A Semantic Web Rule Language Combining OWL and RuleML. Availableat http://www.w3.org/Submission/SWRL/. Accessed 16 February 2010.

ISO (International Organisation for Standardisation). 2003. Geographic Information– Metadata. ISO19115:2003.

ISO (International Organisation for Standardisation). 2007. Geographic information– Geography Markup Language (GML). ISO19136:2007.

Jahn, G., Schramm, M., Spiller, A., 2005. The reliability of certification: Quality labelsas a consumer policy tool. Journal of Consumer Policy 28 (1), 3–73.

KKL-Service. 2010. Kriterien-Kompendium Landwirtschaft (Criteria Compendiumfor Agriculture). Available at:www.kkl-service.info. Accessed 15 February 2010.

Maliappis, M., 2009. Applying an agricultural ontology to web-based applications.International Journal of Metadata, Semantics and Ontologies 4 (1/2), 133–140.

McGuiness, D.L. and F. van Harmelen (eds.). 2004. OWL Web Ontology Language.Available at http://www.w3.org/TR/owl-features/. Accessed 16 February 2010.

Murakami, E., Saraiva, A.M., Ribeiro Jr., L.C.M., Cugnasca, C.E., Hirakawa, A.R., Correa,P.L.P., 2007. An infrastructure for the development of distributed service-oriented information systems for precision agriculture. Computers andElectronics in Agriculture 58 (1), 37–48.

Nash, E., Korduan, P., Bill, R., 2009a. Applications of open geospatial web services inprecision agriculture: a review. Precision agriculture 10 (6), 546–560.

Nash, E., Vatsanidou, A., Fountas, S., 2009b. Can compliance to crop productionstandards be automatically assessed? In: van Henten, E.J., Goense, D., Lokhorst,C. (Eds.), Precision agriculture ’09, Proceedings of the 7th European Conferenceon Precision Agriculture, Wageningen, the Netherlands, 6-8 July 2009.Wageningen Academic Publishers, pp. 809–906.

Nikkilä, R., Seilonen, I., Koskinen, K., 2010. Software architecture for farmmanagement information systems in precision agriculture. Computers andElectronics in Agriculture 70 (2), 328–336.

REWERSE. 2006. R2ML – The REWERSE I1 Rule Markup Language. Available athttp://rewerse.net/I1/oxygen.informatik.tu-cottbus.de/rewerse-i1/@q=r2ml.htm.Accessed 16 February 2010.

Sall, O., Lo, M., Gandon, F., Niang, C., Diop, I., 2009. Using XML data integration andontology reuse to share agricultural data. Internation Journal of Metadata,Semantics and Ontologies 4 (1/2), 93–105.

Sørensen, C.G., Pesonen, L., Fountas, S., Suomi, P., Bochtis, D., Bildsøe, P., Pedersen,S.M., 2010. A user-centric approach for information modelling in arablefarming. Computers and Electronics in Agriculture 73 (2010), 44–55.

Steinberger, G., Rothmund, M., Auernhammer, H., 2009. Mobile farm equipment as adata source in an agricultural service architecture. Computers and Electronics inAgriculture 65 (2), 238–246.

Vatsanidou, A., Fountas, S., Aggelopoulou, K., Gemtos, F., Strobl, K., Charvat, K.,Pedersen, S.M., Sørensen, C.G., Nash, E. and Blackmore, S., 2009. Compliance tostandards specifications. FutureFarm Project Deliverable 2.1.1. Available athttp://www.futurefarm.eu/node/159. Accessed 15 February 2010.