Sustainability in agriculture is crucial for environmental conservation and ecosystem resilience. Within this context, companion planting stands out as a key practice, leveraging synergies between plants for enhanced growth and pest control. However, its broader adoption is hindered by large-scale knowledge and data integration challenges. To bring companion planting forward and closer to interested users, we engineered a semantically rich ontology, CoPla. We used several automated techniques to extract knowledge from various sources and capture diferent companion and anti-companion mechanisms.We demonstrate CoPla's versatility through three applications using diferent reasoning mechanisms: identifying plant companionships, evaluating, and optimising garden layouts.
Sustainability stands at the forefront of global concerns, particularly within the realms of
environmental preservation and agricultural innovation.Central to this discourse is the concept
of sustainable gardening, which is paramount for fostering green, resilient ecosystems across
the globe [
The current landscape of companion planting knowledge is dispersed across various platforms, including agricultural manuals, scholarly articles, and digital databases, making it challenging for practitioners to access and synthesise this information efectively. This dispersion presents a significant obstacle to the eficient integration and broader application of companion planting strategies. Due to this, people aiming at planting their own garden often lack an overview of companion planting mechanisms, ultimately limiting their impact on sustainability. In particular, gardeners who are new to companion planting might need help with the following tasks: (1) exploring potential companions and anti-companions of plants they are interested in, (2) understanding the reasons behind (anti-) companion relationships, and (3) analysing and improving garden layouts wrt. companionships.
To address these challenges, we developed a semantically rich ontology called CoPla that delineates the complex relationships among plant species and their interactions with the environment. It is built in two phases: first, a core conceptualisation was created through a knowledge acquisition phase, where knowledge of companion planting and its various mechanisms was acquired from relevant literature. Second, subclasses and class axioms were automatically created using a rich variety of (web) technologies—from SPARQL querying to scraping of PDFs—to extract knowledge from openly available multi-modal sources. The result is a semantically rich large-scale ontology for informed decision-making in companion planting. To illustrate CoPla’s usefulness, we integrated it in a prototypical system to support gardeners addressing the use cases above. In particular, we showcase how diferent reasoning mechanisms for OWL— deductive inference, explanations and repair—can be used together with the ontology towards the three problems mentioned above. The application of CoPla in these scenarios shows that it is manageable and directly applicable in real-world scenarios. Moreover, our method is designed with scalability in mind, allowing for easy adaptation to broader agricultural contexts by simply expanding the underlying knowledge base.
All resources described in this article, which includes the engineered ontology, the code of the engineering process, the backend and frontend, are available in our online repository.1
We divide the related work into (a) companion planting resources, (b) semantic web eforts from other domains, and (c) AI-based applications to help farming and planting.
planting dates back to the 80’s. The book recollects decades of practices to provide disease- and pest-free gardens using mutual restoration conditions. Some attempts to formalise companion planting practises in the form of charts and tables can be found online. A lookup summary for
plant compatibility, for instance, is published online.2 Plant Interactions are described by the IDEP foundation3, which provides worldwide training for Permaculture Design. While being semi-formalised, these sources of information are non-standardised and relatively incomplete.
A collaborative encyclopaedia for plant information, cultivation and association is the
Practical Plants database4. In addition to description of individual plants and their interaction with
other plants, descriptions of plant combinations can be created. This polyculture data collection
is in its early stage. The EPPO pest database5 provides an overview of hosting plants for pests,
and can be used to determine combinations of plants to cultivate—namely, hosts of known pests
should not be combined with plants that can be afected by that particular pest. The US Natural
Resources Conservation Service provides a number of conservation resources for planters,
researchers, and landowners, including the Soil Taxonomy for soil properties6. The Global
Biotic Interactions (Globi) is a large open-access dataset to find symbiotic/opportunistic (e.g.
predator-prey, pathogen-host or parasite-host) interactions between biological organisms [
tives exist for the agricultural domain. AgroPortal [
Plant and organism dynamics are also important concepts in biology and ecology. [
3https://www.permaculturenews.org/2010/07/30/companion-planting-guide/ 4https://practicalplants.org/wiki/practical_plants/ 5https://gd.eppo.int/taxon/CUUPE/pests 6https://www.nrcs.usda.gov/resources/guides-and-instructions/soil-taxonomy 7https://foodkg.github.io/ AI-based Planting Applications. A number of tools exist to support smart planting. The Almanac Garden Planner8 is an application to help designing layouts of a garden, and provides insights including planting/harvesting seasons and crop rotation warnings. As a proprietary product, its reusability and interoperability are limited. Elzeard 9 is a crop planning and management tool, designed to replace spreadsheet-based crop management methods. It features a Crop Planification and Production Process Ontology (C3PO) that describes the agricultural production process. The Planting Planner (10 provides more information on soil conditions, with a focus on lfowers. Most of these and other existing applications (GrowIt, Garden Manager, etc.) mainly focus on garden aesthetics rather than farming and sustainability aspects. Similarly, industrial farming and farm managing applications focus on the optimisation of farming processes.
The main purpose of the ontology is to structure knowledge about companion planting and assist
garden planners in identifying potential companion relationships. Our development process
was guided by the METHONTOLOGY [20] methodology because of its focus on knowledge
elicitation from knowledge sources rather than4 domain experts. We started from a list of
requirements, for which we developed a core ontology ofering the main conceptualisation of
the ontology. This core ontology was extended with additional axioms that were generated
through automated data integration from openly available multi-modal sources.
Sources of Knowledge. The ontology creation process was informed by a combination of
books [21, 22, 23], scholarly articles [
To determine how the knowledge should be used and organised we devised an informal list of requirements, taking into account the use cases mentioned in Section 1. The ontology should: R1: include the core concepts for the task of companion planting decision support; R2: describe diferent types of companion and anti-companion relationship; R3: define properties describing various mechanisms of companion planting; R4: describe qualities of optimal and sub-optimal garden configurations; R5: model specific plant species and their metadata (common and scientific names).
9https://en.elzeard.co/ 10https://www.plantingplanner.com/ 11https://https://www.permaculturenews.org/2010/07/30/companion-planting-guide/ 12https://www.permablitz.net/wp-content/uploads/2016/08/Poster_GDN_Com_Plant.pdf Core Classes and Relationships companionWith Fauna Predator
predatorOf
Flora Polinator
Pest polinatedBy companionWith antiCompanionWith repelerOf etraatpeCnrBoypFor attractorOf physicalSupportFor providesNutrientsFor positiveHostingFor
suppressesPestsFor providesPotassiumFor providesWindProtectionFor providesCalciumFor
recruitsPredatorsFor providesShadeFor providesPhosphorusFor
recruitsPolinatorsFor providesNitrogenFor providesWaterFor
To address R1, the core ontology describes classes and relationships that capture the companionship relations (see Figure 1): Core Classes and Relationships. The core ontology consists of classes Flora, Fauna, Predator, Pest, Pollinator. Fauna refers to the animals involved in companion planting and has three core sub-classes based on the role animals play in determining companionship. Derived from [22], the class Pollinator contains all animals that act as pollinator for a given plant. The Pest class encompasses all animals that are considered harmful to the plants. These classes are not disjoint: an animal can take on various roles, based on the role they play in a companionship mechanism. Members of the Predator class represent animals that prey on pests. The Flora class contains the diferent plants that are evaluated for companionship. All lfora and fauna are then modelled as subclasses of these core classes.
The core relationships connecting these classes are eatenBy, pollinatorOf, predatorOf, attractorOf, repellerOf, trapCropFor, companionWith and antiCompanionWith. The companionWith and antiCompanionWith relationships are defined to express the companionship relation between members of the Flora class. To address R3 and R2 and expand upon the specific companionship types, some additional relationships are added that describe the interaction between the core class members. The attractorOf property is defined between Flora and Fauna, expressing that some animals are particularly attracted by a specicfi plant. The eatenBy relationship relates animals of type Pest to the Flora they consume.
The repellerOf property describes that some plants prevent pests from feeding on their target plants. predatorOf expresses the relationship between members of the Predator and the corresponding Pest class they feed on. To reflect the interaction between diferent types of animals that play a role in pollinating certain plants, pollinatorOf is used. Lastly, the trapCropFor relationship expresses a special type of companion planting technique, which utilises some plant to serve as a decoy and attract pests that can be then efectively exterminated. Use-case Specific Classes and Relationships. For R4, we use classes and relationships relating to the use-case of garden planning. This allows individual plants within the garden to be connected with neighbour, companionNeighbour and incompatibleNeighbour relationships, that define the garden configurations. Specific garden configurations are described with the class Garden. Based on those properties, we introduce classes for plants that are wellplaced or badly placed, and gardens that represent good configurations or bad configurations.
Detailed plant classifications, and complex axioms describing them, were extracted from a
variety of multimodal sources, resulting in an ontology with 1405 classes and 5264 axioms,
enabling us to derive (anti-)companionships between a variety of plants, as well as explanations.
Companion Planting Charts. For R2, we retrieved companions and anti-companions
from an openly available planting chart13. The high-level relationships (companionWith
and anticompanionWith) were directly captured in the ontology. For a further understanding
of mechanisms of companion and anticompanion relationships, we extracted companionship
mechanisms from literature [
Globi. The Global Biotic Interactions (GloBI) database [
Tomato subClassOf flowersVisitedBy some Apis Melifera .
These axioms served as the basis for the generation of property chains, such as: visitedBy o eats o parasiteOf SubPropertyOf reqruitsPredatorsFor . 13https://www.permaculturenews.org/2010/07/30/companion-planting-guide/ 14https://www.wikidata.org/wiki/Wikidata:Main_Page 15https://api.globalbioticinteractions.org/ (a) Justification-based explanation service (Protégé) (b) Proof-based explanations (Evee)
CoPla can be used to find (anti-)companionships, analyse and suggest garden configurations, and provide explanations. To illustrate these functionalities, we implemented a simple web-front end (Figure 3). First, let us discuss the three use cases previously introduced.
If we are interested in a specific plant and its (anti-)companions, we can use a subclass query, e.g. in Protégé, using a class expression “companion some PLANT” (or “anti_companion some PLANT”), where PLANT identifies the corresponding plant. For each obtained (anti-)companionship, we can ask for explanations in form of justifications [ 24, 25], using the functionality of Protégé (Figure 2a). All relevant axioms are in the fragment of OWL-EL that is supported by the ELK reasoner[26], allowing us to use the advanced explanation services based on proofs ELK. This includes the protege-proof-explanation plugin [27], Evee [28] and Evonne [29], which gives a clearer explanation on why a companionship holds (see Figure 2b).
We can also analyse specific garden configurations by representing them as knowledge graphs. For this, we use an instance of the class Garden to which we link its plants. The property neighbour models which plants are placed next to each other. To determine whether a plant has a neighbour that is an (anti-)companion, CoPla contains axioms of the following form: hasNeighbour some PLANT and companion some PLANT
SubClassOf companionNeighbour some PLANT hasNeighbour some PLANT and anti_companion some PLANT
SubClassOf incompatibleNeighbour some PLANT
Diferent concepts in the ontology define desirable properties of plants based on their neighbourhood—e.g. having one or more companionNeighbours—and of the garden—having at least 3 well-placed plants. For instance, the class BadGarden is defined as a garden in which at least one plant is placed next to an anti-companion. Through reasoning on the Garden individual, users can now determine which properties their garden possesses—whether it is a bad garden or a particularly well-organised garden—and obtain detailed explanations—e.g. which neighbours are currently anti-companions, and why.
A lesser known reasoning service called ontology repair can be used to not only analyse existing garden configurations, but also suggesting optimal placements of plants. The use case here is that the user specifies what plants they want to have in their garden. Through reasoning, we then determine which plants should be placed next to each other.
Given an OWL knowledge base and some undesired entailment , a repair of that knowledge base is a maximal subset of statements in such that is not entailed anymore, but the properties maximising companionship is preserved. Given a list of plants a naive approach would be to start with a “maximal” knowledge graph in which instances of all of the plants are placed next to each other using the neighbour property, and then use repair to remove those neighbourhood relations that cause the garden to be a BadGarden. The resulting knowledge graph is then a configuration in which no anti-companions are placed next to each other.
This approach has two limitations: 1) the repair might remove statements other than the neighbour-relation, for instance elements from the TBox such as axioms from the plant ontology, and 2) we not only want to eliminate anti-companionships, but also maximise companionships. We thus use a diferent type of repairs defined as follows: Definition 1. Let , be knowledge bases s.t. ⊆ and , axioms s.t. |= and ̸|= . A repair of |= preserving |= , with fixed , is a knowledge base s.t. 1. ⊆ ⊆ , 2. ̸|= , and 3. |= .
Clearly, such a repair need not always exist. If it does, it can be computed similar to classical repairs by using justifications [ 30]. In order to suggest a garden configuration for a given set of plants, we construct a maximal garden configuration as described above, add all statements except for the neighbourhood relations into the fixed component , and repair the entailment |= garden Type BadGarden preserving |= garden Type NiceGarden, where BadGarden is the Garden class we aim to avoid (e.g. one with neighbouring anticompanions) and NiceGarden is some Garden class that we would like to preserve (e.g. one that has many companionship neighbours). In our implementation, we rank diferent such preferable Garden-classes, and start with the highest ranked class when trying to compute the repair, and then step-wise change to lower-ranked classes until a repair is found, this way computing the best garden configuration that avoid anti-companionships.
Using Protégé for the described use cases requires familiarity with Semantic Web technologies. Therefore, we implemented some of these use cases with a proof-of-concept user interface using JavaScript and Spring Boot [31], using the REST API for communication with the backend. The backend was implemented in Java 17, using the OWL API 5 [32] (5.1.9) for all OWL related services (including explanation services). The tool can be deployed using Docker Compose [33]. All code, including docker files, are available in our repository (see introduction), which also includes instructions on how to run our tool.
The frontend uses Javascript and Node.js. We made use of AnyChart16 to display simple but interactive graphs to the user. The user interface provides three distinct functionalities: (1) exploring companions of selected plants (2) Checking garden configurations (based on a list of plants) and (3) Suggesting Placement of Plants. We describe the first functionality—the others are described in the appendix.
Figure 3 shows the output of the companion exploration after the user presses the “Explore Companions” button. We use a simple network graph with coloured edges. Green signals a companion relationship and red is for an anti-companion relationship. The presented graph is interactive and the user can click on any nodes or edges, as well as drag the nodes around in the graph canvas. The tooltips of the graph provide further information about the plants: the scientific name, a wikidata link to the plant, and an image of the plant retrieved from wikiData. In the future, we plan to additionally implement an explanation functionality when clicking on edges, to explain in more detail why the connected plants are (anti-) companions.
Provenance in the Ontology. The source knowledge on which we based our design may have limitations: A lot of the knowledge regarding companion planting is anecdotal, and the resources describing these relationships are not always supported by scientific claims. Future work could address this issue by indicating and tracking the provenance of the sources for companionship interactions, both to spot contradicting information and to highlight the relationships that are verified by scientific evidence. Additional future work will aim at including additional factors in the ontology, such as environmental factors and soil quality.
Explanations in User interface. We want to integrate further explanation services into our prototype. In the current version, we only show explanations for the analysis of the garden configuration. Those are just displayed in OWL syntax but but not visualised further. In the future, we would also like to provide explanations for companionships. We also plan to integrate the interactive proof visualisations of Evonne [29], which is also based on Javascript, to allow for a more user friendly and interactive explanations.
Further Improving the User Interface. As the name suggests, we engineered a proof of concept, which means that our tool is neither finished nor did we perform any evaluation with users. For future work, we plan to elicit requirements for the user interface with domain experts and hobby gardeners who are unfamiliar with ontologies to go the extra mile and make this tool accessible. This also includes an evaluation of this interface to show its usefulness. Only by making it usable by laymen can we fully overcome the challenge that companion planting is currently facing in terms of heterogeneous and often inaccessible information.
Evaluation with Human Users. We plan to conduct evaluations of the tool with human users in real-world use cases. We will solicit feedback from domain experts and hobby gardeners who are unfamiliar with ontologies to refine our user interface. This user-centered design approach aims to ensure that our tool is accessible and practical, even for those without technical expertise in ontologies. The evaluation will specifically aim to assess the tool’s usability and efectiveness in supporting sustainable gardening practices.
Companion planting is a key practice in sustainable agriculture and a way for laymen to contribute with their own garden. However, the information on companion planting is scattered and often inaccessible, which calls for integration and publication in a more digestible format. Our contributions to this challenge are three-fold: (1) we engineered the semantically rich companion planting ontology, integrating information from various sources on plants and their efects on each other; (2) we leveraged various reasoning paradigms to infer additional information as well as to check plant constellations against certain garden criteria; (3) we build a simple yet useful decision support system which makes use of the ontology and reasoning abilities for users unfamiliar with Semantic Web technologies. Through our contributions, we showcase the usefulness of these technologies outside of research bounds and also bring them closer to everyday users. ontologies, Trends in ecology & evolution 23 (2008) 159–168. [17] L. Penev, M. Dimitrova, V. Senderov, G. Zhelezov, T. Georgiev, P. Stoev, K. Simov, Openbiodiv: a knowledge graph for literature-extracted linked open data in biodiversity science, Publications 7 (2019) 38. [18] S. Babalou, E. Kleinsteuber, B. El Haouni, F. Zander, D. S. Costa, J. Kattge, B. KönigRies, iKNOW-a knowledge graph management platform for the biodiversity domain, in: proceedings of ISWC 2022, Springer, 2022. [19] N. Le Guillarme, W. Thuiller, A practical approach to constructing a knowledge graph for soil ecological research, European Journal of Soil Biology 117 (2023) 103497. doi:https: //doi.org/10.1016/j.ejsobi.2023.103497. [20] M. Fernández-López, A. Gómez-Pérez, N. Juristo, Methontology: From ontological art towards ontological engineering, in: Proceedings of the Ontological Engineering AAAI-97 Spring Symposium Series, AAAI, 1997. URL: https://oa.upm.es/5484/. [21] H. Philbrick, R. Gregg, Companion Plants and How to Use Them, Devin-Adair Publishers, 1966. [22] A. Greer, Companion Planting for the Kitchen Gardener, Skyhorse Publishing, 2014. [23] M. Jefreys, A Beginners Guide to Companion Planting: Companion Gardening with
Flowers, Herbs & Vegetables, 2013. [24] S. Schlobach, R. Cornet, et al., Non-standard reasoning services for the debugging of description logic terminologies, in: Ijcai, volume 3, 2003, pp. 355–362. [25] M. Horridge, Justification based explanation in ontologies, Ph.D. thesis, University of
Manchester UK, 2011. [26] Y. Kazakov, M. Krötzsch, F. Simancik, The incredible ELK — from polynomial procedures to eficient reasoning with ℰ ℒ ontologies, J. Autom. Reason. 53 (2014) 1–61. doi:10.1007/ s10817-013-9296-3. [27] Y. Kazakov, P. Klinov, A. Stupnikov, Towards reusable explanation services in Protégé, in: Description Logic Workshop, volume 1879 of CEUR Workshop Proceedings, CEUR-WS.org, 2017. [28] C. Alrabbaa, S. Borgwardt, T. Friese, P. Koopmann, J. Méndez, A. Popovič, On the eve of true explainability for owl ontologies: Description logic proofs with evee and evonne, Proc. DL 22 (2022). [29] J. Méndez, C. Alrabbaa, P. Koopmann, R. Langner, F. Baader, R. Dachselt, Evonne: A visual tool for explaining reasoning with OWL ontologies and supporting interactive debugging, Computer Graphics Forum (2023). doi:https://doi.org/10.1111/cgf.14730. [30] B. Parsia, E. Sirin, A. Kalyanpur, Debugging OWL ontologies, in: A. Ellis, T. Hagino (Eds.),
WWW-05, ACM, 2005, pp. 633–640. [31] C. Walls, Spring Boot in action, Simon and Schuster, 2015. [32] M. Horridge, S. Bechhofer, The OWL API: A Java API for OWL ontologies, Semantic web 2 (2011) 11–21. [33] D. Merkel, Docker: lightweight linux containers for consistent development and deployment, Linux Journal 2014 (2014) 2.