A Quantitative Assessment Framework for Modelling and Evaluation

Using Representation Learning in Smart Agriculture Ontology

Khadija Meghraoui

1 a

, Teeradaj Racharak

2 b

, Kenza Ait El Kadi

1,3 c

, Saloua Bensiali

4 d

and

Imane Sebari

1,3 e

Unit of Geospatial Technologies for a Smart Decision, Hassan II Institute of Agronomy and Veterinary Medicine, Rabat,

Morocco

School of Information Science, Japan Advanced Institute of Science and Technology, Ishikawa, Japan

School of Geomatics and Surveying Engineering, Hassan II Institute of Agronomy and Veterinary Medicine, Rabat,

Morocco

Department of Applied Statistics and Computer Science, Hassan II Institute of Agronomy and Veterinary Medicine, Rabat,

Morocco

Keywords:

Ontology, Ontology Embedding, Agriculture, Word Embedding, Ontology Evaluation.

Abstract:

Understanding agricultural processes and their interactions can be improved with trustworthy and precise mod-

els. Such modelling boosts various related tasks, making it easier to take informed decisions in the realm of

advanced agriculture. In our study, we present a novel agriculture ontology, primarily focusing on crop produc-

tion. Our ontology captures fundamental domain knowledge concepts and their interconnections, particularly

pertaining to key environmental factors. It encompasses static aspects like soil features, and dynamic ones such

as climatic and thermal traits. In addition, we propose a quantitative framework for evaluating the quality of the

ontology using the embeddings of all the concept names, role names, and individuals based on representation

learning (i.e. OWL2Vec*, RDF2Vec, and Word2Vec) and dimensionality reduction for visualization (i.e. t-

distributed Stochastic Neighbor Embedding). The ﬁndings validate the robustness of OWL2Vec* among other

embedding algorithms in producing precise vector representations of ontology, and also demonstrate that our

ontology has well-deﬁned categorization aspects in conjunction of the embeddings.

1 INTRODUCTION

Historically, ontologies have been a cornerstone for

intelligent agricultural systems in terms of knowledge

modelling (Abbasi et al., 2022). Numerous agricul-

tural ontologies, such as Crop Ontology (CO) (Arnaud

et al., 2012), ARGOVOC (Rajbhandari and Keizer,

2012), Plant Ontology (PO) (Jaiswal et al., 2005), and

AgriOnt (Ngo et al., 2018) have been developed by

experts. Among these, CO stands out as an ontology

designed to represent the vocabulary associated with

various crops traits focusing on plants such as wheat,

soybean, and rice. Developed by the Food and Agri-

culture Organization (FAO), AGROVOC is a compre-

hensive thesaurus that spans various domains within

https://orcid.org/0000-0003-3925-0691

https://orcid.org/0000-0002-8823-2361

https://orcid.org/0000-0001-6483-3704

https://orcid.org/0000-0003-1753-2209

https://orcid.org/0000-0002-6754-8404

agriculture, covering multiple subcategories. One of

its advantages is its multilingual vocabulary, encom-

passing a wide range of concepts and terms. However,

AGROVOC is mainly an expansive vocabulary rather

than a full and a complete ontology designed for direct

applications by users. Indeed, its relational structures

lack clarity and brevity; it is more like a combination

of diﬀerent vocabularies than a singular, and cohe-

sive one. On the other hand, PO acts as a structured

repository detailing plant morphology, anatomy, and

growth stages, integrating particular relationships, es-

pecially the “is-a” and “part-of” links. AgriOnt is a

prime example of a robust agricultural Knowledge-

Base (Kaewboonma et al., 2020). This ontology is

applied in various domains, such as geographical data

and the Internet of Things (IoT), addressing a large

number of practical applications. Yet, it is worth

pointing out that these ontologies omit some funda-

mental factors relevant to crop growth, and the de-

ﬁned concepts might not be user-friendly for small-

1044

Meghraoui, K., Racharak, T., El Kadi, K., Bensiali, S. and Sebari, I.

A Quantitative Assessment Framework for Modelling and Evaluation Using Representation Learning in Smart Agr iculture Ontology.

DOI: 10.5220/0012432900003636

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 16th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2024) - Volume 3, pages 1044-1051

ISBN: 978-989-758-680-4; ISSN: 2184-433X

scale farmers. As a result, its main audience seems

to be researchers. For instance, (Ngo et al., 2020) de-

veloped a knowledge map model based on AgriOnt to

represent and store the insights extracted from datasets

related to crops. By analyzing these ontologies to-

wards their goals, it is evident that there is a signiﬁcant

gap in the agricultural ontology, speciﬁcally focusing

on crop-related aspects. Creating a newly designed

smart agriculture ontology centered on crop yield and

related concepts agrees with (Xie et al., 2008)’s asser-

tion that every smart knowledge system should pos-

sess its own distinct ontology.

In this manuscript, we introduce a novel smart

ontology within the framework of intelligent agricul-

tural systems. We discuss the selection for its diverse

concepts, the data employed for knowledge acquisi-

tion, and the origin of this information. With this re-

spect, an existing deﬁciency is the lack of quantitative

framework for comprehensive and fair judgement on

the modelled ontology in an intrinsic manner. Note

that traditional ontology modelling employs extrinsic

manner with stakeholders on extrinsic tasks.

Nevertheless, there is sill no quantitative meth-

ods for evaluating the quality of ontologies. This

lack could hinder the process of ontology constric-

tion and ontology adoption in areas that lack involve-

ment of domain experts; therefore, having clear un-

derstanding on extrinsic and intrinsic approaches are

necessary and helpful. This work extends the state

of the art of representation learning for ontologies

(Asim et al., 2018) for evaluating our proposed on-

tology on three aspects: (1) the categorization as-

pect, (2) the hierarchical aspect, (3) the relational as-

pect. Indeed, our evaluation process focuses on the

embeddings of the ontological concepts, roles, and

individuals, through the utilization of OWL2Vec*,

RDF2Vec, and Word2Vec. Our intrinsic evaluation is

conducted through multiple steps, including the cal-

culation of cosine similarity metric and visualizing

it using heatmaps for several entities as well as two-

dimensional t-SNE visualization to further ensure the

aforementioned aspects of the modelled ontology.

This paper oﬀers a signiﬁcant contribution by in-

troducing a novel smart agriculture ontology. The de-

veloped ontology focuses on key elements inﬂuencing

crop yield and oﬀers ﬂexibility for further expansion

and integration, aiming for enhanced knowledge pre-

cision. The authors also utilizes the embeddings tech-

nique for learning representations of the agricultural

ontology, taking into account both the Assertional

Box (ABox) and the Terminological Box (TBox) el-

ements of the knowledge base. This represents a pio-

neering application of this technique within the realm

of agricultural ontologies, providing an automated ap-

proach to evaluate the newly formulated ontology,

gauging the integrity of the introduced knowledge-

base via diverse metrics and methods stemming from

the learned embeddings.

2 PRELIMINARIES

2.1 Ontology-Based Knowledge Base

Originally rooted in philosophy, the term “ontology”

was initially used to describe and explain existence

in the universe (Wei et al., 2012). However, given

the rapid advancements in information science, it has

emerged as a pivotal research area in knowledge rep-

resentation. Mathematically, ontologies are expressed

using formal structures and logical principles based on

Description Logics (DLs).

DL-based ontologies have three disjoint sets: con-

cepts, roles, and individuals, and thus form two com-

ponents of ontologies: TBox and ABox. Brieﬂy, a

TBox (or a terminology) is a ﬁnite set of general con-

cept inclusions and role hierarchy axioms, whose syn-

tax is an expression of the form 𝐶 ⊑ 𝐷 and 𝑟 ⊑ 𝑠, re-

spectively, where 𝐶, 𝐷 are concepts and 𝑟, 𝑠 are roles.

An ABox (or assertions) is a ﬁnite set that captures

the relationships of individuals with their concepts

and the relationships between individuals themselves.

Formally, this set contains expressions of the forms

𝐶(𝑎) and 𝑟(𝑎, 𝑏), where 𝑎,𝑏 are individuals.

2.2 Agriculture Ontology

Agricultural ontologies oﬀer farming terms and elu-

cidates their interconnections (Zheng et al., 2012).

These ontologies act as a foundation for subsequent

semantic applications (Wei et al., 2012). The goal is to

encourage the reuse, distribution, analysis, and man-

agement of knowledge in the agricultural domain.

In (Bhuyan et al., 2021), an agricultural ontology

was introduced for intelligent farming processes us-

ing a lattice framework. They also developed a rule-

based mining algorithm leveraging the features of this

structure. Their knowledge representation spanned

both spatial and temporal dimensions, and they uti-

lized cube data for orderly and sequential informa-

tion representation. Every agricultural location, its as-

sorted attributes, and multiple time markers were con-

ceptualized as a unique triple, which then populated

the lattice structure. From these agriculture-related

triples, they derived association rules to uncover new

relationships between entities. To validate their ap-

proach, a limited dataset comprising ten locations and

six characteristics spanning four time intervals was

A Quantitative Assessment Framework for Modelling and Evaluation Using Representation Learning in Smart Agriculture Ontology

1045

used. However, they did not incorporate real-world

examples and they did not explain how their knowl-

edge model could boost and enhance crop production,

a point they initially highlighted. Moreover, their pre-

sented knowledge can be better described as a knowl-

edge graph rather than a complete ontology.

In (Li et al., 2013), the authors proposed a knowl-

edge representation methodology focusing on crop

cultivation procedures. The study delves into Good

Agricultural Practices (GAP) and the foundational

theories of agricultural ontologies. Within this re-

search, the domain ontology encompasses informa-

tion on soil and agricultural equipment, while the task

ontology focuses on agricultural processes like variety

selection and determining suitable soil types. Using

pepper as their subject for tests, the authors concluded

that their proposed approach eﬀectively oﬀers a struc-

tured knowledge representation, making this speciﬁc

agricultural area more accessible.

In (Zheng et al., 2012), an ontology-driven agri-

cultural management framework was implemented,

consisting of the acquisition, organization, and mining

of the represented knowledge. In their study, informa-

tion about agriculture was derived from a variety of

sources, particularly plants-based food. The authors

also incorporated a data mining approach to provide

users with pertinent content by identifying their re-

quirements through comprehensive data analysis.

3 PROPOSED METHODOLOGY

FOR SMART FARMING

DEVELOPMENT

3.1 Ontology Requirements

Constructing an ontology is a crucial process that typ-

ically requires expertise and insights from specialists.

In our study, we adhered to the approach suggested by

(Xie et al., 2008) who outlined that the development of

an ontology should encompass three primary phases

as follows: (1) constructing a hierarchy tailored to the

speciﬁc domain; (2) outlining the properties and for-

mulating axioms; (3) knowledge gathering, which in-

volves populating values to the ontology.

Our speciﬁc agriculture ontology has been devel-

oped following the detailed steps:

1. Investigation of the primary factors inﬂuencing

agricultural yields.

2. Determination of concepts and the relationships

that may link those concepts.

3. Creation of a knowledge hierarchy based on the

predeﬁned concepts and roles.

4. Development of the ontology taxonomy using an

editor interface.

5. Knowledge acquisition based on the taxonomy.

6. Validating and ensuring the consistency of the on-

tology.

Our ontology comprises four distinct categories of

factors that we believe inﬂuence crop yields. These

factors assist stakeholders in precisely managing their

ﬁelds when they have access to this knowledge. The

ﬁrst category pertains to “Soil”. Soil attributes are

constant factors that have a direct impact on crop pro-

duction. Such characteristics often guide farmers in

making informed decisions for various planting sce-

narios. This can lead to enhanced crop cultivation

under suitable conditions and also aids in mitigating

losses under less favorable agricultural circumstances

(Malik et al., 2021). In our ontology, we have incor-

porated various soil attributes, which are as follows:

1. Soil bdod indicates the Bulk Density of the Dry

soil, and refers to soil compaction. It is determined

by dividing the dehydrated soil by the volume, and

expressed in cg/cm

(de Sousa et al., 2020).

2. Soil cec denotes the Cation Exchange Capacity by

determining the total amount of cations that can

be held by a portion of soil.

3. Soil cfvo represents the Volumetric fraction of

coarse fragments measured in cm

/dm

4. Soil clay identiﬁes the proportion of clay particles

(under the 0.002 mm value) in the ﬁne fraction.

5. Soil nitrogen measures the total amount of the ni-

trogen chemical element in the given soil.

6. Soil phh2o signiﬁes the pH of a fraction of the

soil.

7. Soil sand indicates the proportion of sand parti-

cles (over 0.05 mm) in the ﬁne fraction.

8. Soil silt represents the proportion of silt particles

between 0.002 mm and 0.05 mm in the ﬁne earth.

9. Soil soc labels the Soil Organic Carbon content in

the soil.

10. Soil ocd designates the Organic Carbon Density

of a considered soil.

11. Soil ocs is another soil characteristic which mea-

sures the Organic Carbon Stocks.

The second category of environmental factors in-

cludes climatic conditions. These dynamic attributes

can directly inﬂuence crop yields. Various climatic

elements have a clear correlation with crop produc-

tion, especially factors related to droughts (Poudel and

Shaw, 2016), water stress (Wang et al., 2018), and so-

lar radiation, which impacts the rate of photosynthetic

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

1046

activity on plant surfaces (Holzman et al., 2018). In

our ontology, we have incorporated the following cli-

matic properties:

1. Climate aet stands for Actual EvapoTranspiration,

representing the genuine evapotranspiration of the

ground cover (Li et al., 2016).

2. Climate def indicates Climatic Water Deﬁcit, and

integrates the impact of rainfall and temperature

(Micheli et al., 2012).

3. Climate pdsi denotes Palmer Drought Severity In-

dex, which is a common marker for detecting dry-

ness (Wang et al., 2022).

4. Climate pet represents Potential Evapo-

Transpiration and denotes the potential volume of

evaporation and transpiration from a vast region

fully covered in vegetation (Li et al., 2016).

5. Climate pr denotes precipitation, a vital compo-

nent of the hydrological cycle (Zhang et al., 2022).

6. Climate ro designates the Runoﬀ phenomenon,

that occurs when the ground can not absorb all the

existing water.

7. Climate srad indicates the Solar RADiation, a vi-

tal factor for planning and establishing agricultural

research (Ikram et al., 2022). Solar radiation en-

compasses the energy and other emissions emitted

by the Sun (Book, 2002).

8. Climate vap stands for VApor Pressure, rep-

resenting the pressure exerted by the vapor

(Schr

oder et al., 2017). It is a method used to mea-

sure humidity.

9. Climate vpd stands for Vapor Pressure Deﬁcit, a

critical factor inﬂuencing the photosynthesis pro-

cess (Yuan et al., 2019), which directly aﬀects crop

production.

10. Climate vs represents wind speed, measured in

m/s (Terraclimate, 2020).

Another signiﬁcant factor impacting crops is tem-

perature, especially the extreme lows and highs.

These temperature variations inﬂuence the metabolic

activities of plants, including cellular respiration

(Sharma et al., 2022), transpiration (Bueno et al.,

2019), and nitrogen ﬁxation (Bytnerowicz et al.,

2022). The ontology we developed incorporates the

temporal dimension via the “year” concept. This is

essential because each agricultural yield is associated

with a speciﬁc year or farming season. Typically, the

target variable of our ontology is related to this tem-

poral aspect, emphasizing the signiﬁcance of the year

class. Agricultural yields can ﬂuctuate yearly based

on various factors, some of which have been previ-

ously highlighted and discussed.

Once the ontology taxonomy has been deﬁned and

ﬁxed, the next step is to populate individuals and

instances from the real world to the ontology. In

this case, the knowledge acquisition has been done

using data from the Consultative Group for Inter-

national Agricultural Research (CGIAR) (CGIAR,

2021). Transformation rules were used to populate the

created classes and properties with data. An example

of these rules is shown as follows:

Individual ∶ @𝐵

Types ∶ 𝐹 𝑖𝑒𝑙𝑑 𝐼𝐷

Facts ∶ ℎ𝑎𝑠𝑐𝑙𝑖𝑚𝑎𝑡𝑒𝑎𝑒𝑡@𝑃

(1)

This aﬀects the values of column P, representing the

climateaet property, to the corresponding individuals

in column B, representing the Field ID.

3.2 Modeled Ontology

Following the requirements discussed in Section 3.1,

our ontology has 29 classes, including 91,672 ax-

ioms, with 76,668 of them being logical. The ontol-

ogy has been populated with approximately 14,950

individuals. Note that our modeled ontology is

available online at https://github.com/realearn-jaist/

evaluation-framework-with-agri-onto.

Figure 1 oﬀers a comprehensive illustration of the

concepts and relationships within the developed on-

tology. Through these classes and properties, the on-

tology elucidates various factors inﬂuencing crop pro-

duction. Figure 2 provides an example of individuals

from multiple classes of the ontology and their rela-

tionships.

4 QUANTITATIVE EVALUATION

Developing new ontologies allows for the representa-

tion of real-world concepts and their interconnected

relationships. In our context, constructing an ontology

that delineates these agricultural concepts is invalu-

able. Asserting that this ontology is suitable means

that it can accurately classify the instances within

it. One technique employed for this classiﬁcation in-

volves generating vectors for each instance, a method

commonly referred to as representation learning. By

visualizing these vectors, we can observe the catego-

rization based on our ontology. In contrast to extrinsic

processes that often involves humans, our evaluation

goals here focus on three aspects on the embedding

vectors: (1) the categorization aspect, (2) the hierar-

chical aspect, (3) the relational aspect.

A Quantitative Assessment Framework for Modelling and Evaluation Using Representation Learning in Smart Agriculture Ontology

1047

Figure 1: Detailed concepts and relationships of our created ontology.

Figure 2: Examples of individuals populated in the smart

agriculture ontology.

4.1 Vector-Based Evaluation Process

We implement OWL2Vec* proposed in (Chen et al.,

2021) to learn the embeddings of our ontology, which

takes into account both TBox and ABox. To ensure

the best representation, we also implement Word2Vec

and RDF2Vec as benchmarks, as discussed below.

One popular technique to learn these embed-

dings through artiﬁcial neural networks is Word2Vec

(Karani, 2018). Word2Vec can produce embeddings

using either the Skip Gram or Continuous Bag Of

Words (CBOW) methods. When using CBOW, a core

word is predicted using its adjacent terms. Conversely,

Skip Gram models aim to predict surrounding words

based on a given term (Onishi and Shiina, 2020).

Semantic embedding through OWL2Vec* is di-

vided into two primary stages. Initially, a corpus is

generated from the ontology (Chen et al., 2021). This

corpus is utilized to train the word embedding model.

Within OWL2Vec*, the corpus is divided into three

types of documents: structural, lexical, and a merged

version. Both the structural and lexical documents ex-

plore the ontology’s graph organization, including its

constructors and its lexical attributes and labels. The

merged document is designed to preserve potential

associations between the Internationalized Resource

Identiﬁers (IRIs) and other components. The model

takes an OWL ﬁle as its input and produces vector rep-

resentations for each entity and relationship within the

ontology. Several parameters can be conﬁgured, such

as the embedding dimension, the walker type, the win-

dow size, and the minimum count.

Regarding RDF2Vec, the initial step is to convert

the RDF graph into sentences using graph walks (Ris-

toski et al., 2019). These produced sequences are then

fed into a neural network model. Once trained, this

model can predict sentences and produce a vector rep-

resentation for each graph entity. Various parameters

can be also conﬁgured, such as the embedding dimen-

sion, the window size, the minimum count, and a cho-

sen embedding models (either CBOW or Skip Gram).

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

1048

4.2 Implementation and Analysis

Our implementation utilized OWL2Vec* from the

original repository

. The parameters we adopted are:

‘random’ for the walker, a window size of 5, a mini-

mum count of 1, and an embedding dimension of 100.

We trained the Word2Vec model with an embed-

ding dimension of 100, a window size of 5, and a min-

imum count of 1. These identical parameters were ap-

plied to RDF2Vec after converting the graph structure

into a collection of sentences. We implemented the

three representation learning techniques in the Colab

notebook environment, using Python libraries.

We assessed our developed smart agriculture on-

tology by employing an automated evaluation method,

analyzing the embeddings produced by various tech-

niques. Before applying and using in-depth evalua-

tion methods, and for identifying any potential log-

ical inconsistencies within our ontology, we utilized

automated reasoning veriﬁcation via the HermiT rea-

soner within the Prot

e interface. To ensure a thor-

ough and reliable evaluation of our designed ontology,

we adopted multiple evaluation metrics. Initially, we

assessed similarity by computing the cosine similar-

ity metric. Subsequently, we employed dimension-

ality reduction, where the high-dimensional data or

the resulting embeddings are transformed into low-

dimensional points using t-SNE. The following sub-

sections give a detailed insight into these evaluation

metrics and their results.

Note that our implementation and experimen-

tal results are available online at https://github.com/

realearn-jaist/evaluation-framework-with-agri-onto.

4.2.1 Ontology Evaluation Using Cosine

Similarity Measure

For our ontology, and given the signiﬁcant number of

individuals within each class, we opted to compute

the cosine similarity measure using only ten instances

from every class. Essentially, we determined the co-

sine similarity between every possible pair among

these ten instances and then took the average cosine

value from the resulting matrix as the representative

cosine similarity for that class.

In our study involving 29 classes, we calculated the

cosine similarity measure for each class and then de-

termined an average value. This facilitates a compar-

ison of our OWL2Vec* representation learning tech-

nique with other benchmark methods like Word2Vec

and RDF2Vec. Table 1 and 2 present the average

cosine similarity values for these three techniques.

Speciﬁcally, they show values for individuals within

https://github.com/KRR-Oxford/OWL2Vec-Star

Table 1: Similarity between individuals of the same class.

Embedding method Cosine similarity

OWL2Vec* 0.817

Word2Vec 0.672

RDF2Vec 0.942

Table 2: Similarity between individuals of diﬀerent classes.

Embedding method Cosine similarity

OWL2Vec* 0.655

Word2Vec 0.601

RDF2Vec 0.901

the same class and for those from diﬀerent classes,

termed as negative sampling. For instances belong-

ing to the same class, OWL2Vec* and RDF2Vec re-

ported high similarity values of 0.817 and 0.942, re-

spectively. Conversely, Word2Vec had a lower sim-

ilarity score of 0.672. When assessing individuals

from diﬀerent classes, a decrease in cosine similarity

was observed, with this decline being particularly ev-

ident in OWL2Vec* and Word2Vec.

To better understand the computed cosine simi-

larity, we have visualized the representations using

heatmaps, primarily for individuals within the same

class. As depicted in Figures 3 it becomes evident

that OWL2Vec* eﬀectively distinguished between in-

dividuals from diﬀerent classes while capturing their

inherent similarities. This distinction is further high-

lighted in the diagonal heatmap of Figure 3, as it shows

a darker colour. In contrast, RDF2Vec produced vec-

tors that were remarkably similar despite having dis-

tinct properties. Word2Vec exhibited moderate dif-

ferentiation. Due to the space limitation, we provide

more displayed ﬁgures in our GitHub repository to

support the above explanation.

4.2.2 Ontology Evaluation Using t-SNE

Visualization

Not only similarity-based evaluation, we also per-

formed visualization to see through the characteristics

of modeled ontology. To this aim, we used t-SNE to

assess the ontology we developed, by visualizing the

vector representations derived from the three ontology

embedding techniques.

Figure 4 clearly shows that OWL2Vec* adeptly

clustered similar instances and entities. Due to

the space limitation, we show more detailed images

within our GitHub repository. Indeed, RDF2Vec

failed to distinguish between distinct classes, indi-

cated by the almost singular central cluster. Word2Vec

showed only slight separations with a more dispersed

visual representation.

A Quantitative Assessment Framework for Modelling and Evaluation Using Representation Learning in Smart Agriculture Ontology

1049

Figure 3: Cosine similarity heatmap using OWL2Vec*.

Figure 4: t-SNE visualization for OWL2Vec*.

5 CONCLUSIONS

This paper introduces a novel agricultural ontology

centered on crop production and its primary inﬂuenc-

ing environmental factors. Our representation learn-

ing framework leverages both the ABox and TBox

components of the knowledge base. We benchmarked

our approach against two well-established baselines:

Word2Vec and RDF2Vec. The results conﬁrm that

our ontology learning method oﬀers enhanced vector

representations. We also propose a framework to eval-

uate the developed agricultural ontology using the co-

sine similarity measure among various classes which

further employ the t-SNE visualization method for a

more detailed assessment of our ontology.

In future, we aim to utilize the learned embeddings

for advanced agricultural applications, especially in

predicting crop yields. We plan to extend our intrinsic

assessment process as a general framework for ontol-

ogy modelling (cf. (Alshargi et al., 2018)).

ACKNOWLEDGEMENTS

The authors would like to extend their appreciation

to the anonymous reviewers for their valuable feed-

back. This research was conducted during an intern-

ship at Japan Advanced Institute of Science and Tech-

nology (JAIST) in collaboration with the Department

of Photogrammetry and Cartography of IAV Hassan

II in Morocco. This work was also supported by JSPS

KAKENHI Grant Number JP22K18004.

REFERENCES

Abbasi, R., Martinez, P., and Ahmad, R. (2022). An on-

tology model to represent aquaponics 4.0 system’s

knowledge. Information Processing in Agriculture,

9(4):514–532.

Alshargi, F., Shekarpour, S., Soru, T., and Sheth, A. P.

(2018). Metrics for evaluating quality of embeddings

for ontological concepts.

Arnaud, E., Cooper, L., Shrestha, R., Menda, N., Nelson,

R. T., Matteis, L., Skoﬁc, M., Bastow, R., Jaiswal, P.,

Mueller, L. A., et al. (2012). Towards a reference plant

trait ontology for modeling knowledge of plant traits

and phenotypes. In KEOD, pages 220–225.

Asim, M. N., Wasim, M., Khan, M. U. G., Mahmood,

W., and Abbasi, H. M. (2018). A survey of ontol-

ogy learning techniques and applications. Database,

2018:bay101.

Bhuyan, B. P., Tomar, R., Gupta, M., and Ramdane-Cherif,

A. (2021). An ontological knowledge representation

for smart agriculture. In 2021 IEEE International Con-

ference on Big Data (Big Data), pages 3400–3406.

IEEE.

Book, T. (2002). Solar water heating and the plant engineer.

In Plant Engineer’s Reference Book, pages 42–1. El-

sevier.

Bueno, A., Alfarhan, A., Arand, K., Burghardt, M.,

Deininger, A.-C., Hedrich, R., Leide, J., Seufert,

P., Staiger, S., and Riederer, M. (2019). Eﬀects

of temperature on the cuticular transpiration barrier

of two desert plants with water-spender and water-

saver strategies. Journal of Experimental Botany,

70(5):1613–1625.

Bytnerowicz, T. A., Akana, P. R., Griﬃn, K. L., and Menge,

D. N. (2022). Temperature sensitivity of woody nitro-

gen ﬁxation across species and growing temperatures.

Nature Plants, 8(3):209–216.

CGIAR (2021). https://zindi.africa/competitions/

cgiar-crop-yield-prediction-challenge/data. Ac-

cessed on August 30, 2023.

Chen, J., Hu, P., Jimenez-Ruiz, E., Holter, O. M., Antonyra-

jah, D., and Horrocks, I. (2021). Owl2vec*: Em-

bedding of owl ontologies. Machine Learning,

110(7):1813–1845.

de Sousa, L. M., Poggio, L., Batjes, N. H., Heuvelink, G. B.,

Kempen, B., Riberio, E., and Rossiter, D. (2020). Soil-

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

1050

grids 2.0: producing quality-assessed soil information

for the globe. Soil Discuss, 2020:1–37.

Holzman, M. E., Carmona, F., Rivas, R., and Nicl

os, R.

(2018). Early assessment of crop yield from re-

motely sensed water stress and solar radiation data. IS-

PRS journal of photogrammetry and remote sensing,

145:297–308.

Ikram, R. M. A., Dai, H.-L., Ewees, A. A., Shiri, J., Kisi,

O., and Zounemat-Kermani, M. (2022). Application

of improved version of multi verse optimizer algo-

rithm for modeling solar radiation. Energy Reports,

8:12063–12080.

Jaiswal, P., Avraham, S., Ilic, K., Kellogg, E. A., McCouch,

S., Pujar, A., Reiser, L., Rhee, S. Y., Sachs, M. M.,

Schaeﬀer, M., et al. (2005). Plant ontology (po): a

controlled vocabulary of plant structures and growth

stages. Comparative and functional genomics, 6(7-

8):388–397.

Kaewboonma, N., Chansanam, W., and Buranarach, M.

(2020). Ontology-based big data analysis for orchid

smart farming. LIBRES: Library & In-formation Sci-

ence Research Electronic Journal, 29:292–296.

Karani, D. (2018). Introduction to word embedding and

word2vec. Towards Data Science, 1.

Li, D., Kang, L., Cheng, X., Li, D., Ji, L., Wang, K., and

Chen, Y. (2013). An ontology-based knowledge rep-

resentation and implement method for crop cultiva-

tion standard. Mathematical and computer modelling,

58(3-4):466–473.

Li, S., Kang, S., Zhang, L., Zhang, J., Du, T., Tong, L., and

Ding, R. (2016). Evaluation of six potential evapo-

transpiration models for estimating crop potential and

actual evapotranspiration in arid regions. Journal of

Hydrology, 543:450–461.

Malik, P., Sengupta, S., and Jadon, J. S. (2021). Com-

parative analysis of soil properties to predict fertility

and crop yield using machine learning algorithms. In

2021 11th International Conference on Cloud Com-

puting, Data Science & Engineering (Conﬂuence),

pages 1004–1007. IEEE.

Micheli, E., Flint, L., Flint, A., Weiss, S., and Kennedy, M.

(2012). Downscaling future climate projections to the

watershed scale: A north san francisco bay case study.

San Francisco Estuary and Watershed Science, 10(4).

Ngo, Q. H., Kechadi, T., and Le-Khac, N.-A. (2020).

Oak: ontology-based knowledge map model for digi-

tal agriculture. In Future Data and Security Engineer-

ing: 7th International Conference, FDSE 2020, Quy

Nhon, Vietnam, November 25–27, 2020, Proceedings

7, pages 245–259. Springer.

Ngo, Q. H., Le-Khac, N.-A., and Kechadi, T. (2018). Ontol-

ogy based approach for precision agriculture. In Multi-

disciplinary Trends in Artiﬁcial Intelligence: 12th In-

ternational Conference, MIWAI 2018, Hanoi, Viet-

nam, November 18–20, 2018, Proceedings 12, pages

175–186. Springer.

Onishi, T. and Shiina, H. (2020). Distributed representation

computation using cbow model and skip–gram model.

In 2020 9th International Congress on Advanced Ap-

plied Informatics (IIAI-AAI), pages 845–846. IEEE.

Poudel, S. and Shaw, R. (2016). The relationships between

climate variability and crop yield in a mountainous en-

vironment: A case study in lamjung district, nepal.

Climate, 4(1):13.

Rajbhandari, S. and Keizer, J. (2012). The agrovoc concept

scheme–a walkthrough. Journal of Integrative Agri-

culture, 11(5):694–699.

Ristoski, P., Rosati, J., Di Noia, T., De Leone, R., and Paul-

heim, H. (2019). Rdf2vec: Rdf graph embeddings and

their applications. Semantic Web, 10(4):721–752.

Schr

oder, M., Lockhoﬀ, M., Shi, L., August, T., Bennartz,

R., Borbas, E., Brogniez, H., Calbet, X., Crewell, S.,

Eikenberg, S., et al. (2017). Gewex water vapor as-

sessment (g-vap), wcrp report 16/2017, world climate

research programme (wcrp): Geneva, switzerland, 216

pp.

Sharma, N., Thakur, M., Suryakumar, P., Mukherjee, P.,

Raza, A., Prakash, C. S., and Anand, A. (2022).

‘breathing out’under heat stress—respiratory control

of crop yield under high temperature. Agronomy,

12(4):806.

Terraclimate (2020). https://www.climatologylab.org. Ac-

cessed on September 23, 2023.

Wang, J., Vanga, S. K., Saxena, R., Orsat, V., and Raghavan,

V. (2018). Eﬀect of climate change on the yield of

cereal crops: A review. Climate, 6(2):41.

Wang, Z., Yang, Y., Zhang, C., Guo, H., and Hou, Y. (2022).

Historical and future palmer drought severity index

with improved hydrological modeling. Journal of Hy-

drology, 610:127941.

Wei, Y.-y., Wang, R.-j., Hu, Y.-m., and Xue, W. (2012).

From web resources to agricultural ontology: a

method for semi-automatic construction. Journal of

Integrative Agriculture, 11(5):775–783.

Xie, N., Wang, W., and Yang, Y. (2008). Ontology-based

agricultural knowledge acquisition and application. In

Computer And Computing Technologies In Agricul-

ture, Volume I: First IFIP TC 12 International Con-

ference on Computer and Computing Technologies in

Agriculture (CCTA 2007), Wuyishan, China, August

18-20, 2007 1, pages 349–357. Springer.

Yuan, W., Zheng, Y., Piao, S., Ciais, P., Lombardozzi, D.,

Wang, Y., Ryu, Y., Chen, G., Dong, W., Hu, Z., et al.

(2019). Increased atmospheric vapor pressure deﬁcit

reduces global vegetation growth. Science advances,

5(8):eaax1396.

Zhang, Y., Wu, C., Yeh, P. J.-F., Li, J., Hu, B. X., Feng,

P., and Jun, C. (2022). Evaluation and comparison of

precipitation estimates and hydrologic utility of chirps,

trmm 3b42 v7 and persiann-cdr products in various cli-

mate regimes. Atmospheric Research, 265:105881.

Zheng, Y.-L., He, Q.-Y., Ping, Q., and Ze, L. (2012). Con-

struction of the ontology-based agricultural knowl-

edge management system. Journal of Integrative Agri-

culture, 11(5):700–709.

A Quantitative Assessment Framework for Modelling and Evaluation Using Representation Learning in Smart Agriculture Ontology

1051