Smart Agricultural Technology 7 (2024) 100425

Contents lists available at ScienceDirect

Smart Agricultural Technology

journal homepage: www.journals.elsevier.com/smart-agricultural-technology

The application of machine learning techniques for smart irrigation

systems: A systematic literature review
Abiadi Younes *, Zouhair Elamrani Abou Elassad , Othmane El Meslouhi ,
Dauha Elamrani Abou Elassad , Ed-dahbi Abdel Majid
SARS Research Team, Computer Science Department, ENSAS, Cadi Ayyad University, Morocco

A R T I C L E I N F O A B S T R A C T

Keywords: Smart Irrigation System is a complex concept used to control, monitor and automate the irrigation of yields by
Smart irrigation system integrating artificial intelligence techniques such as Machine Learning strategies. SIS has endorsed various
Machine learning machine learning models. However, there has been no attempt to analyze the empirical evidence on ML models
Systematic literature review
in a systematic way. Moreover, ML based SIS often face problems and raise questions that must be resolved. This
article presents a systematic literature review of ML based SIS; an overview of the literature on ML is designed,
revealing a premier and unbiased survey of the existing empirical research. 55 selected studies published from
2017 to 2023 and nine broadly used ML models were identified. Furthermore, four analysis aspects namely type
of ML technique, estimation accuracy, model comparison, and estimation context have been outlined. The
findings of this review prove the performance capability of the ML techniques endorsed within SIS. The ML
techniques outperform other conventional approaches. However, the application of ML models in SIS is still
limited and more effort is needed to obtain well-formed and generalizable results. To this end, and based on the
outcomes obtained in this work, future guidelines have been provided to practitioners and researchers to grasp
the major contributions and challenges in the state-of-the-art research.

1. Introduction Globally, the main purpose behind smart irrigation systems is that
enhance agricultural productivity at the same time reducing the envi­
The world knows emergency problems, foremost among them being ronmental impact of crops.
the scarcity of fresh water. Around 70 % of freshwater used in agricul­ One of the most used technics in SIS is Machine Learning. It addresses
tural activities such irrigation and the supply nutrient for plant growth. the question of how to build a computer system that improves auto­
The demand for freshwater is escalating with the increasing food de­ matically through experience [4,5]. ML techniques re known for (i) their
mand of a fast-growing population [1]. The optimal solution for this capacity to independently tackle significant nonlinear challenges by
problem is switching from traditional irrigation methods to new irri­ utilizing datasets from various origins, and (ii) their flexibility in inte­
gation systems by including new technologies such internet of thing grating fresh data to enhance the accuracy (Mohammed [6,7]). They are
(IoT), machine learning (ML), wireless sensor network (WSN) giving being used in the context of studying smart irrigation to provide users
birth to the term smart irrigation systems (SIS). This latter allows us to with better recommendations and help constructing powerful ML
improve the irrigation decision-making and monitoring water systems to models.
enhance crop productivity [2]. ML techniques are integrated in many fields, they are used in irri­
SIS aims to prevent both under-irrigation and over-irrigation, gation field to estimate soil moisture content, to estimate the reference
ensuring optimal crop yields. Many types of crops have specific water of evapotranspiration, to improve energy management performance and
requirements that vary throughout their growth stages. Also, SIS Over­ to Predict water table depth. Generally speaking with ML we can provide
comes hurdles soil erosion and crop specific irrigation problem. Efficient decision-making in irrigated agriculture by using meteorological and
Irrigation guaranties a sustainable use of water, this indicates that SIS is soil data, also we can contribute to sustainability efforts by conserving
not trivial task, instead, it composes of several complicated steps [3]. water resources. By precisely delivering water where and when it is

* Corresponding author.
E-mail address: [email protected] (A. Younes).

https://doi.org/10.1016/j.atech.2024.100425
Received 22 October 2023; Received in revised form 20 February 2024; Accepted 28 February 2024
Available online 11 March 2024
2772-3755/© 2024 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
A. Younes et al. Smart Agricultural Technology 7 (2024) 100425

Table 1
Research questions.
RQ# Research questions Motivation

RQ1 Which ML techniques have been Identify the ML techniques

used in the literature for Smart commonly being used in SIS.
Irrigation system?
RQ2 What is the overall performance of Investigate the performance of the
the ML techniques for SIS? ML techniques for SIS.
RQ3 Do ML models outperform non-ML Investigate the performance of the
models? ML techniques over non-ML models
for SIS.
RQ4 Are there any ML techniques that Investigate the performance of the
significantly outperform other ML ML techniques over the other ML
techniques? techniques for SIS.
RQ5 What are the strengths and Determine the information about ML
weaknesses of the ML techniques? techniques.

2. Method

The systematic review planning used in this work has been inspired
from the SLR process suggested by Kitchenham and Charters [10], the
process is detailed in Fig. 1, the important steps of this process:
Research questions identification: we construct the research
questions based on the purpose of SLR.
Search strategy design: we described the search strategy to find out
the important studies to the research question, this step involves both
Fig. 1. Systematic review process. identification of search terms and selection of sources to be searched in
order to identify the initial list.
needed, these systems reduce unnecessary irrigation, mitigate water Study selection criteria: we extract the relevant studies from the
scarcity issues, and promote efficient water management practices in initial list by using the inclusion and exclusion criteria that we will
agriculture. However, the ML discipline does not have a definite clas­ identify in the next section, for each primary study in initial list.
sification scheme for its algorithms, mostly due to the number of para­ Data extraction: we involve the design of data extraction forms to
digms and the uncertainties introduced in the literature ([8,7]b; [9]). collect the required information in order to answer the research
Subsequently, it becomes difficult and confusing to choose an ML al­ questions
gorithm that fits one’s need when developing a smart irrigation Data synthesis: we determined the proper methodologies for syn­
computational model. thesizing the extracted data based on the types of the data and the
In order to gain a comprehensive understanding of the existing research questions the data addressed.
research on the utilization of machine learning (ML) in smart irrigation A review protocol is of critical importance for an SLR (M [11]). To
system, we conducted a systematic literature review (SLR). A SLR serves ensure the rigorousness and repeatability of this SLR and to reduce
to identify potential research gaps in a specific problem area and offers researcher bias as well, we elaborately developed the review protocol by
guidance to both practitioners and researchers interested in pursuing frequently holding group discussion meetings on protocol design. In the
further investigations in that field. The SLR methodology entails next Subsection, we will present the details of the review protocol. At the
accessing relevant studies from electronic databases, synthesizing the end of this section, we will analyze the threats to validity of the review
findings, and presenting them in response to the defined research protocol.
questions. By conducting an SLR, new perspectives emerge, aiding
newcomers in comprehending the current state-of-the-art in the field. 3. Research questions
It is crucial for an SLR to be replicable, meaning that all the steps
taken must be clearly explained, and the results should be transparent The goal of this SLR is to summarize and clarify the empirical evi­
for other researchers to evaluate. Objectivity and transparency are key dence obtained from the studies using the ML techniques for SIS in the
factors for the success of an SLR [10]. As the name suggests, an SLR literature. Toward this goal five research questions (RQs) were defined
necessitates a systematic approach that encompasses all the literature as presented in Table 1. From the selected studies we identified the ML
published thus far. This study provides an exhaustive compilation of the techniques used in Smart irrigation system (RQ1), and then we analyzed
existing literature regarding the use of machine learning for smart irri­ the performance of the ML techniques for SIS (RQ2). This question fo­
gation system. Additionally, we present our empirical findings and cuses on the values of the performance measures for SIS, after that we
address the research questions defined within this review article. For compare the performance of ML techniques with non-ML techniques
this systematic literature review we collected all studies that integrated (RQ3), The aim of this question is to determine whether ML techniques
the important technologies used in the process of SIS. Publications be­ are better than the statistical techniques. (RQ4) we do a comparison
tween 2017 and 2023 were considered, in which the important tech­ between different ML techniques, whose can be synthesized to deter­
nology is machine learning with its different approaches. mine which ML models consistently outperform other ML models. The
The rest of the paper is organized as follows: Section 2 describes the final question (RQ5) determines the strengths and weakness of different
methodology used in this review. Section 3 presents and discusses the ML techniques.
review results. Section 4 provides the limitation of this work and Section
5 provides conclusions and future guidelines obtained from this sys­ 4. Search strategy
tematic review.
The detailed description of the search strategies utilized in this
research consisted of search terms, literature resources and search pro­
cess as explained below:

2
A. Younes et al. Smart Agricultural Technology 7 (2024) 100425

• Search stage 1: A thorough search was launched on the six electronic

database sources and the returned results (papers) were assembled as
sets of prospective papers.
• Search stage 2: The reference lists of all relevant papers were perused
to detect additional relevant papers and then, if any, combine them
with the ones in stage 1.

Software package Mendeley (https://www.mendeley.com) was used

to store and manage the search results. We identified 55 relevant papers
according to the search process. The detailed search process and the
number of papers identified at each phase are shown in Fig. 2.

5. Study selection

From the first search stage, 1071 prospective studies were realized.
Since many of the candidate papers provide no information useful to
address the research questions raised by this review, further filtering is
needed to identify the relevant papers belong the candidate papers by
using the inclusion and the exclusion criteria (defined below). This is
exactly what study selection aims to do. Specifically, as illustrated in
Fig. 2. Search and selection process. Fig. 2.
We defined the following inclusion and exclusion criteria, which had
been refined through pilot selection. We carried out the study selection
4.1. Search terms
by reading the titles, abstracts, or full text of the papers.
Inclusion criteria:
The following steps were used to generate the search terms [12]:

• Empirical studies using the ML techniques for SIS.

a) Derivation of major terms from the research questions.
• Empirical studies using at least two ML models for SIS.
b) Identification of alternative spellings and synonyms for major terms.
• Empirical studies combining the ML and non-ML techniques.
c) Identification of keywords in relevant papers or books.
• For duplicate publications of the same study, only the most complete
d) Usage of the Boolean OR to incorporate alternative spellings and
and newest one will be included.
synonyms.
• For study that has both conference version and journal version, only
e) Usage of the Boolean AND to link the major terms.
the journal version will be included.
The resulting search terms are described as follows: (“Smart Irriga­
Exclusion criteria:
tion” OR “Precision Irrigation”) AND (“Machine Learning” OR “Artificial
Intelligence”)
• Studies without empirical analysis or results of use of the ML tech­
niques for SIS.
4.2. Literature resources • Studies using the ML techniques in context other than SIS.
• Review studies.
The literature resources we exploited to search for primary studies
contain six electronic databases: Two researchers conducted independent evaluations to test the
aforementioned inclusion and exclusion criteria. Through thorough
Ø IEEE Xplore discussions, they arrived at a mutual decision. In instances of uncer­
Ø ScienceDirect tainty, the complete texts of the studies were reviewed, leading to a final
Ø Web of Science determination regarding their inclusion or exclusion. Furthermore, any
Ø Scopus duplicate studies by a specific author with identical results were elimi­
Ø Google Scholar nated. Ultimately, a total of 55 studies were selected by employing the
Ø ACM Digital Library aforementioned selection criteria.

The search terms established previously were employed to search for 6. Data extraction and data synthesis
journal papers and conference papers in the six electronic databases. The
search was conducted on the first six databases covering title, abstract, For each of the selected studies, we completed a form to extract
and keywords. For Google Scholar, only title was searched since the full pertinent information. The purpose of using the data extraction form
text search will return millions of irrelevant records. We limited the was to identify which research question was addressed by each selected
search to the period from 2017 to 2023, because the most relevant study study. The extracted data included details such as author names, titles,
of the application of ML techniques in SIS domain was launched just in publishing information, independent variables (metrics), and the ML
the early 2017. techniques employed. By analyzing the data extraction cards, we could
determine the specific research questions that each selected study aimed
to answer. These cards served as a means to gather information from the
4.3. search process selected studies.
Two independent researchers collected the necessary information for
SLR necessitates a comprehensive search of all relevant sources. each selected study using the data extraction cards. They compared their
That’s why, we defined the search process and split it into the following findings, and in case of any discrepancies, additional researchers were
two phases (note that the relevant papers are those papers that verify the consulted to resolve any disagreements. The resulting data was then
selection criteria defined in the next section). saved in an Excel file for future use during the data synthesis process.

3
A. Younes et al. Smart Agricultural Technology 7 (2024) 100425

Table 2 Table 3
Publication sources and distribution of the selected studies. Selected studies.
Publication name Type Number Percent ID Authors Title Citation

Agricultural Water Management Journal 10 18 S01 [13] Modeling reference https://doi.org/10.10

Computers and Electronics in Agriculture Journal 6 11 evapotranspiration using extreme 16/j.compag.2017.01.0
Journal of Hydrology Journal 4 7 learning machine and generalized 27
Computers,Materials & Continua Journal 2 4 regression neural network only with
Sensors Journal 2 4 temperature data
Food Quality Journal 2 4 S02 [14] Evaluation of random forests and https://doi.org/10.10
The Crop Journal Journal 1 2 generalized regression 16/j.agwat.2017.08.003
PeerJ Computer Science Journal 1 2 neuralnetworks for daily reference
Science of the Total Environment Journal 2 4 evapotranspiration modelling
Sustainable Energy Technologies and Journal 1 2 S03 [15] An IoT based smart irrigation https://doi.org/10.10
Assessments management system using Machine 16/j.compag.2018.09.0
remote sensing Journal 1 2 learning and open source 40
Heliyon Journal 1 2 technologies
Multimedia Tools and Applications Journal 1 2 S04 [16] Dynamic Neural Network Modelling https://doi.org/
Modern Physics Letters B Journal 1 2 of Soil Moisture Content for 10.3390/s18103408
applied sciences Journal 1 2 Predictive Irrigation Scheduling
Internet of Things Journal 1 2 S05 [17] Reference evapotranspiration https://doi.org/10.10
Intelligent Systems Journal 1 2 estimation and modeling of the 16/j.compag.2018.11.0
Security and Communication Networks Journal 1 2 Punjab Northern India using deep 31
Smart Agricultural Technology Journal 1 2 learning
Energy Reports Journal 1 2 S06 [18] Capability of crop water content for https://doi.org/10.10
Journal of Science Journal 1 2 revealing variability of winter wheat 16/j.scitotenv.2018.0
Agriculture Journal 1 2 grain yield and soil moisture under 3.004
Remote Sensing of Environment Journal 1 2 limited irrigation
Environment, Development and Sustainability Journal 1 2 S07 [19] An efficient employment of internet https://doi.org/10.1
IEEE Access Journal 1 2 of multimedia things in smart and 007/s11042–019–736
International Conference on Computer and Conference 1 2 future agriculture 7–0
Knowledge Engineering (ICCKE) S08 [20] Machine Learning based soil https://doi.org/10.110
International Conference on Data Analytics for Conference 1 2 moisture prediction for Internet of 9/ISPCC48220.201
Business and Industry (ICDABI) Things based Smart Irrigation 9.8988313
IEEE International Conference on Signal Conference 1 2 System
Processing, Computing and Control (ISPCC) S09 [21] A Hadoop based Framework for Soil https://doi.org/10.11
IEEE International Conference on Signal- Conference 1 2 Parameters prediction 09/SITIS.2019.00111
Image Technology & Internet-Based Systems S10 [22] Random forest techniques for spatial https://doi.org/10.10
(SITIS) interpolation of evapotranspiration 16/j.agwat.2019.10590
International Conference on Electronics and Conference 1 2 data from Brazilian’s Northeast 5
Sustainable Communication Systems S11 [23] Machine learning-based irrigation https://doi.org/10.11
(ICESC) control optimization 45/3360322.3360854
International Conference on Artificial Conference 1 2 S12 [24] Estimation of daily potato crop https://doi.org/10.10
Intelligence and Data Engineering (ICAIDE) evapotranspiration using three 16/j.agwat.2019.10
Engineering, Technology & Applied Science Conference 1 2 different machine learning 5875
Research algorithms and four scenarios of
International Conference on Material Conference 1 2 available meteorological Data
Processing and Technology (ICMProTech) S13 [25] Long short-term memory neural https://doi.org/10.114
ICTACT JOURNAL ON SOFT COMPUTING Conference 1 2 network-based multi-level model for 2/S0217984920504187
Total 55 100 smart irrigation
S14 [26] Smart Irrigation IoT Solution using https://doi.org/10.110
Transfer Learning for Neural 9/ICCKE50421.2020
The main objective of data synthesis was to combine and consolidate Networks .9303612
S15 [27] Improve Irrigation Timing Decision https://doi.org/10.
the findings and results from the selected studies to formulate responses
for Agriculture using Real Time Data 1109/ICDABI51230.202
and address the research questions. By aggregating studies that pre­ and Machine Learning 0.9325680
sented similar viewpoints and comparable results, we obtained research S16 [28] IoT based Smart System for https://doi.org/10.110
evidence that contributed to obtaining conclusive answers. Enhanced Irrigation in Agriculture 9/ICESC48915.2020
During the synthesis process, we meticulously scrutinized and eval­ .9156026
S17 [29] Simulating wetting front dimensions https://doi.org/10.10
uated both quantitative and qualitative data. The quantitative data
of drip irrigation systems: Multi 16/j.jhydrol.2020.
encompassed various performance metrics, such as AUC, prediction criteria assessment of soft computing 124792
accuracy, while the qualitative data included the strengths and weak­ models
nesses of the ML methods and the categorization of different ML tech­ S18 [30] Soil water content and actual https://doi.org/10.10
niques. To address the research questions, we employed visualization evapotranspiration predictions using 16/j.agwat.2020.10
regression algorithms and remote 6346
techniques, such as line graphs, box plots, pie charts, and bar charts. sensing data
Additionally, we used tables to summarize and present the results. S19 [31] New approach to estimate daily https://doi.org/10.10
reference evapotranspiration based 16/j.agwat.2020.10
7. Results and discussion on hourly temperature and relative 6113
humidity using machine learning
and deep learning
This section presents the results obtained from selected studies for S20 [32] A Deep Learning Approach for Multi- https://doi.org/
smart irrigation system. First, we present the overview of each selected Depth Soil Water Content Prediction 10.1109/ACCESS.202
study. Second, the answers to the review research questions are pro­ in Summer Maize Growth Period 0.3034984
vided in each sub-section. Then we provide the discussion and inter­ S21 [33] Short term soil moisture forecasts for https://doi.org/10.10
potato crop farming: A machine 16/j.compag.2020.10
pretation of the results in the view of the facts obtained from each learning approach 5902
selected study.
(continued on next page)

4
A. Younes et al. Smart Agricultural Technology 7 (2024) 100425

Table 3 (continued ) Table 3 (continued )

ID Authors Title Citation ID Authors Title Citation

S22 [34] Field Data Forecasting Using LSTM https://doi.org/1 irrigation-systemComputers-

and Bi-LSTM Approaches 0.3390/app112411820 Materials-and-Continua
S23 [35] Artificial intelligence methods https://doi.org/10.10 S44 [56] Nation-scale reference https://doi.org/10.10
reliably predict crop 16/j.agwat.2021.10 evapotranspiration estimation by 16/j.jhydrol.2021.
evapotranspiration with different 6968 using deep learning and classical 127207
combinations of meteorological data machine learning models in China
for sugar beet in a semiarid area S45 [57] IoT-Driven Model for Weather and https://doi.org/10.11
S24 [36] Soil color analysis based on a RGB https://doi.org/10.10 Soil Conditions Based on Precision 55/2022/7283975
camera and an artificial neural 16/j.heliyon.2021.e060 Irrigation Using Machine Learning
network towards smart irrigation: A 78 S46 [58] IoT-enabled edge computing model https://doi.org/
pilot study for smart irrigation system 10.1515/ji
S25 [37] Mapping maize crop coefficient Kc https://doi.org/10.10 sys-2022–0046
using random forest algorithm based 16/j.agwat.2021.10690 S47 [59] Estimation of transpiration https://doi.org/10.10
on leaf area index and UAV-based 6 coefficient and aboveground 16/j.cj.2022.08.001
multispectral vegetation indices biomass in maize using time-series
S26 [38] Machine Learning Approach for an https://doi.org/ UAV multispectral imagery
Automatic Irrigation system 10.48084/etasr.3944 S48 [60] Spatiotemporal modeling to predict https://doi.org/10.101
Irrigation System in Southern Jordan soil moisture for sustainable smart 6/j.iot.2022.100671
Valley irrigation
S27 [39] Smart Irrigation System for Urban https://doi.org/ S49 [61] Simulation of daily maize https://doi.org/10.10
Gardening using Logistic Regression 10.1088/1742–6596/ evapotranspiration at different 16/j.jhydrol.2022.
algorithm and Raspberry Pi 2129/1/012044 growth stages using four machine 128947
S28 [40] IoT-IIRS: Internet of Things based https://doi.org/10.771 learning models in semi-humid
intelligent-irrigation 7/peerj-cs.578 regions of northwest China
recommendation S50 [62] Prediction of maize crop coefficient https://doi.org/10.10
S29 [41] Sustainable Irrigation System for https://doi.org/ from UAV multisensor remote 16/j.agwat.2022.1080
Farming Supported by Machine 10.3390/s21093079 sensing 64
Learning and Real-Time Sensor Data S51 [63] Machine Learning Techniques for https://doi.org/10.
S30 [42] An IoT-Based Predictive Analytics https://doi.org/10.1 the Classification of IoT-Enabled 54287/gujsa.1141575
for Estimation of Rainfall for 007/978–981–15–351 Smart Irrigation Data for
Irrigation 4–7_105 Agricultural Purposes
S31 [43] Estimation of Grapevine Crop https://doi.org S52 [64] Intrusion Detection Using Machine https://doi.org/10.11
Coefficient Using a Multispectral /10.3390/rs13132639 Learning for Risk Mitigation in IoT- 55/2022/3955514
Camera on an Unmanned Aerial Enabled Smart Irrigation in Smart
Vehicle Farming
S32 [44] Numerical and artificial intelligence https://doi.org/10.100 S53 [65] Smart irrigation system based on IoT https://doi.org/10.1016
models for predicting the water 7/s10668–021–01453–6 and machine learning /j.egyr.2022.07.088
advance in border irrigation S54 [66] Evaluating soil moisture content https://doi.org/10.10
S33 [45] Estimation of root zone soil moisture https://doi.org/10.101 under maize coverage using UAV 16/j.jhydrol.2023.1290
from ground and remotely sensed 6/j.rse.2021.112434 multimodal data by machine 86
soil information with multisensor learning algorithms
data fusion and automated ML S55 [67] Forecasting vapor pressure deficit https://doi.org/10.10
S34 [46] Fusion of Feature Selection Methods https://doi.org/ for agricultural water management 16/j.agwat.2023.10830
and Regression Algorithms for 10.3390/agricultur using machine learning in semi-arid 2
Predicting the Canopy Water e11010051 environments
Content of Rice Based on
Hyperspectral Data
S35 [47] A low-cost approach for soil https://doi.org/10.10 8. Overview of selected studies
moisture prediction using multi- 16/j.scitotenv.20
sensor data and machine learning 22.155066
algorithm
In this section, we provide the brief description of the selected
S36 [48] Water optimization technique for https://doi.org/10.1016 studies; We identified 55 studies Table 3 which use the ML techniques
precision irrigation system using IoT /j.seta.2022.102307 for SIS. The studies were mostly based on public or open-source data
and machine learning sets. These papers were published during the time period 2017–2023.
S37 [49] MANAGING IRRIGATION NEEDS https://doi.org/10.2
Among them, 46 (83,64%) papers were published in journals, 9 (16,36
BASED ON SMART DECISIONS 1917/ijsc.2022.0353
USING MACHINE LEARNING %) papers appeared in conference proceedings. The publication sources
S38 [50] Irrigation optimization with a deep https://doi.org/10.10 and distribution of the selected studies are shown in Table 2. the most
reinforcement learning model_ Case 16/j.agwat.2022.10 publications were in Agricultural Water Management, Computers and
study on a site in Portugal 7480 Electronics in Agriculture and Journal of Hydrology.
S39 [51] Soil moisture content estimation in https://doi.org/10.10
winter wheat planting area for multi- 16/j.compag.2021.10
The publication sources and distribution of the selected studies are
source sensing data using CNNR 6670 shown inTable 3. The major publications were in Agricultural Water
S40 [52] Application of IoT and Cloud https://doi.org/10.11 Management, Computers and Electronics in Agriculture, and journal of
Computing in Automation of 55/2022/8285969 Hydrology. The impact factor of these three journals is 6.7, 8.3, and 6.4,
Agriculture Irrigation
respectively. These journals are prestigious in agriculture domain, and
S41 [53] Design-of-Machine-Learning-Based- https://doi.org/10.3260
Smart-Irrigation-System-for- 4/cmc.2022.022648 20 (36 %) studies for this review are retrieved from them.
Precision-Agriculture
S42 [54] Estimation of soil moisture content https://doi.org/10.10 9. Publication years
under high maize canopy coverage 16/j.agwat.2022.10
from UAV multimodal data and 7530
machine learning
Fig. 3 introduces the distribution of the studies from the year 2017 to
S43 [55] Iot with evolutionary algorithm- https://doi.org/10.3260 2023. Fig. 3 shows that the number of studies increases year by year
based-deep-learning-for-smart- 4/cmc.2022.021789 from 2017 onwards, which illustrates the evolution of research in the
application of ML techniques for SIS. The number of studies in the years

5
A. Younes et al. Smart Agricultural Technology 7 (2024) 100425

Fig. 3. Number of publications per year.

Table 4
The classification of ML techniques and selected studies of each category.
# of
studies
used

Neural networks MLP, RBF, CNN, S01, S02, S04, S05, 33

(NN) BPNN, FFNN, GRNN, S06, S07, S09, S10,
ELM, AFNN, LSTMNN, S11, S12, S13, S14,
DQNN S15, S19, S20, S21,
S22, S24, S26, S29,
S31, S33, S34, S36,
S38, S39, S41, S42, Fig. 4. Distribution of the studies over type of ML technique.
S44, S49, S50, S51,
S53
Support vector S06, S07, S10, S12, 20
machines (SVM) S13, S15, S16, S21,
• Neural networks (NN)
S23, S28, S29, S35, • Support vector machines (SVM)
S37, S40, S41, S43, • Fuzzy & Neuro Fuzzy based (NF)
S49, S51, S52, S53 • Clustering (CL)
Fuzzy & Neuro S09, S10, 2
• Instance Based (IB)
Fuzzy based
(NF) • Decision trees (DT)
Clustering (CL) K-Means S03, S46 2 • Bayesian learners (BL)
Instance Based KNN S12, S16, S23, S30, 10 • Ensemble learners (EL)
(IB) S37, S41, S42, S45, • Evolutionary algorithms (EA)
S53, S54
Decision trees M5P, REPTree, CART, S13, S28, S29, S30, 8
• Miscellaneous (Misc)
(DT) ART S43, S45, S46, S55
Bayesian learners S16, S28, S37, S40, 7 Table 4 introduces the classification of ML techniques for SI used in
(BL) S43, S45, S53 this SLR and the studies that used them
Ensemble learners RF, RSS, GBT, Boosting S02, S05, S07, S08, 31
Among the above listed categories of the ML techniques the most
(EL) S12, S13, S15, S17,
S18, S19, S20, S21, frequently used techniques were from categories NN, SVM and EL
S23, S25, S29, S30, adopted by 72 % of selected studies, as illustrated in Fig. 4. These ML
S31, S33, S34, S35, techniques were used for SIS. The identified ML techniques were used to
S36, S40, S42, S46, evaluate SIS usually in two forms: (1) studies comparing ML techniques
S47, S49, S50, S51,
S52, S54, S55
(2) studies comparing statistical and ML techniques; Fig. 3 is plotted to
Evolutionary GA, GEP S17, S31, S32 3 further outline the distribution of research interest in each publication
algorithms (EA) year. As can be seen, the activity of publications in this field is growing
Miscellaneous WINSRFR, GBRT, S01, S08, S09, S31, 6 at an explosive rate. Note that some studies contain more than one ML
(Misc) XGBR, GL,HG, Arima, S32, S35
technique.
Calibrated-HG
As shown in Fig. 5, for one thing, an apparent publication peak is
shown in 2022; Moreover, compared to other ML techniques, SVM, NN
2018, 2019, 2020, 2021 and 2022 were 3, 6, 8, 12 and 19, respectively and EL seem to have received dominant research attention in many
accounting about 70.9 % of the studies between 2020 and 2022. For year years. It is noteworthy that in neural network studies, both traditional
2023, data is not complete as the search was finished in june 2023 and Artificial Neural Networks (ANNs) and Deep Neural Networks (DNNs)
only few studies were online till then. have been considered for this review. Deep learning methods can be
seen as an improved extension of the classic ANNs consisting of more
10. RQ1: which ml techniques have been used in the literature layers that enable higher degrees of abstraction and enhanced assess­
for SIS? ment from data. An example can be seen in study done by flan et al.[ref]
in which they use Unet architecture for optimizing irrigation system in a
In this section, we outline the ML techniques employed in the small green house for tomato crops.
selected studies gathered during this systematic literature review (SLR).
Through careful analysis of the selected studies, we have categorized the
ML techniques used for SIS, as follows:

6
A. Younes et al. Smart Agricultural Technology 7 (2024) 100425

Fig. 5. Distribution of the studies over publication year.

in SIS, focusing on the performance measures employed by researchers

Table 5
to assess the reliability and effectiveness of the developed ML models in
Performance metrics used in studies.
real-world applications. The utilization of validation methods, such as n-
Performance metrics Studies fold Cross-Validation (n > 1) and Leave-One-Out Cross-Validation
accuracy S05, S13, S15, S16, S26, S27, S28, S29, S34, S37, (LOOCV), has been analyzed, and it has been observed that 22 studies
S40, S41, S43, S45, S46, S48, S51, S52, S53. used k-fold Cross-Validation, while 2 studies used LOOCV.
RMSE S04, S05, S06, S09, S12, S18, S19, S20, S21, S22,
Apart from validation methods, various metrics have been employed
S23, S25, S30, S31, S32, S33, S35, S36, S38, S39,
S42, S44, S47, S50, S53, S54, S55. to evaluate the performance of ML approaches in SIS. These metrics are
R^(2) S03, S04, S05, S06, S08, S09, S10, S11, S12, S18, crucial for comparing and evaluating models developed using diverse
S19, S21, S23, S25, S31, S32, S34, S35, S36, S38, ML and statistical techniques. It is worth noting that a single algorithm
S39, S42, S44, S46, S47, S50, S54. may utilize multiple metrics to describe its performance. Therefore,
MAE S01, S02, S04, S09, S10, S11, S12, S18, S20, S21,
selecting appropriate accuracy metrics for evaluating estimation accu­
S23, S31, S32, S44, S55.
MSE S03, S05, S08, S11, S12, S14, S20, S23, S24, S26, racy is essential. Table 5 provides an overview of the performance
S46, S49. metrics used and the studies in which they were applied. Additionally,
Precission S07, S13, S26, S28, S37, S41, S43, S51. the distribution of studies using each performance metric is illustrated in
Recall S07, S13, S26, S28, S37, S41, S43, S51.
Fig. 6. The most popular metrics found in the selected studies are Ac­
F1-Score S07, S13, S26, S41, S43, S51.
NSE S05, S18, S19, S44.
curacy, RMSE, and R-Squared, with adoption rates of approximately
NS S01, S02, S17. 34.55 %, 49.09 %, and 45.45 %, respectively. Other commonly used
SI S10, S17, S32. metrics include MAE, MSE, Precision, Recall, and F1-Score. Less
Miscellaneous (EVS, AUC, S03, S11, S12, S13, S55. commonly employed measures are categorized under miscellaneous
RRSE, EF, R)
metrics, including EVS, AUC, RRSE, EF and R.
The analysis of evaluation measures reveals their popularity and
11. Rq2: what is the overall performance of the ml techniques introduces other metrics to researchers. However, given the dominance
for SIS? of Accuracy, RMSE, and R-Squared metrics, we have chosen to adopt
them in this review for evaluating the performance of ML models. These
The primary objective of this systematic literature review (SLR) is to metrics are crucial in determining the accuracy of the estimates. To gain
explore the utilization of machine learning (ML) algorithms in SIS. The insight into the distribution of these metric values for ML techniques, we
intention is to offer valuable insights to future researchers and practi­ have utilized box plots in Fig. 5a–c, respectively. Nevertheless, in cases
tioners engaged in SIS studies. To achieve this, an extensive investiga­ where the number of observations for the selected metrics in certain ML
tion has been under taken to understand how ML algorithms are applied categories is small, we have refrained from including them in the figures

Fig. 6. Distribution of the studies over performance metrics.

7
A. Younes et al. Smart Agricultural Technology 7 (2024) 100425

Fig. 7. Box plots of Accuracy, RMSE and R-Squared (outliers are labeled with associated study IDs).

with 0.78 value. As for RMSE, approximately all the four models (SVM,
Table 6 NN, EL and IB) have their medians around 0.3, With regard to R-
Comparisons of R-Squared between ML models and non-ML models (‘‘+’’ in­
Squared, EL and NN are more efficient (with median around 0.87),
dicates ML model outperforms non-ML model, ‘‘− ’’ indicates non-ML model
followed by SVM (with median around 0.74), then lastly IB with a me­
outperforms ML model; the number following study ID is RMSE improvement in
percentage, the study ID in bold indicates the improvement exceeds 2%). dian around 0.53. In addition to the aforementioned observations, (EL,
NN, SVM) for R-Squared all have their median nearly in the centers of
ML- Non-ML
the boxes, which implies that the values of these models are symmet­
Model Model
rically distributed around the medians.
SVR PLSR MLR naive
NN + S11(50), S39 S06(42.86), NA S04(6.74),
(0.08), S39 S34(3.93), S34 S04(7.95), 12. Rq3: ml models vs. non-ML models
(2.45), S39 (7.27), S34 S04(7.6), S04
(0.82) (4.03), S34 (5.62), S04 The ML models have been compared with four conventional non-ML
(6.97), S34 (10.23), S04
models: multiple linear regression, naïve [36], partial least squares
(8.6) (7.6)
– S39(31.57), S06(10.71), NA NA regression [19] and support vector regression [10]. The details of the
S39(4.56), S34(4.44) comparisons between ML models and non-ML models are provided in
S39(404), (Table 6), One model is alleged to outperform another in experiment if
S50(94.11), the R-Squared value of the first model achieves at least 2 % improve­
S50(25.53)
IB + NA S54(35.29), NA NA
ment comparing to the second one. It was observed that only 18 % of
S54(60.60) studies compared Non-ML and ML models (10 out of 55). among the four
– NA S42(64.70), NA NA non-ML models, PLSR model is the one used most frequently in com­
S42(63.64) parison with the ML models.
EL S47(16.67), S34(5.53), S34 S25(97.44), NA
+
Fig. 8 indicates the overall results of the comparisons between ML
S50(72.72), (8.5), S34 S25(100),
S50(16.95) (8.03), S34 S25(78.57), models and non-ML models, where all the comparisons were in terms of
(5.41), S34 S25 R-Squared metric. The bars above zero line indicate that ML models are
(6.97), S34 (182.61) more accurate, whereas the bars below zero line indicate that non-ML
(8.49), S42 models are more accurate. the majority of the experiments indicate
(39.29), S42
(44.44), S54
that ML models outperform non-ML models.
(28.26), S54 Specifically, Fig. 8 shows that 76 % (35 of 46) of experiment results
(13.21) exhibit the superiority of ML models whereas only 24 % (11 of 46) of
– NA NA NA experiment results exhibit the superiority of non-ML models (note that
SVM NA NA NA NA
+
some studies conduct more than one experiment). The comparison re­
– NA S06(1.72), S06 NA NA
(60) sults show that the EL and NN are the most commonly used techniques in
the literature; they outperform the PLSR model in 100 % and 75 % of
experiments respectively, and these observations are sustained by a
to avoid presenting insignificant data points in a box plot. number of experiments, while PLSR model outperforms SVM model. For
As can be seen in Fig. 7 for accuracy demonstrates higher values of IB, it is hard to determine whether it is more accurate than PLSR model
performance in terms of NN and EL (with median around 0.95), followed or not, because the number of experiments reporting that IB out­
by SVM and DT around 0.87 value, while IB was the least performant performed regression model is equal to that of experiments reporting the

8
A. Younes et al. Smart Agricultural Technology 7 (2024) 100425

Fig. 8. Comparisons of R-Squared between ML models and non-ML models.

opposite results. In addition to PLSR model, other non-ML models have comparison results can be found in Fig. 9. First, SVM and EL are more
also been compared with some ML models; EL outperforms SVR model accurate than NN, and it is supported by most of the experiments con­
while the last one outperforms NN, With regard to the MLR and naive ducting the comparisons. Second, for the case of SVM vs. EL, SVM seems
models, although every model are 4 and 6 experiments, respectively. All more accurate than EL. Third, for the comparisons between other ML
the trials of MLR model are less accurate than EL model, also the trials of models, the number of experiments is relatively small, and some com­
naïve models are less accurate than NN, in general, the ML models parisons are even found to be inconsistent. Therefore, for these ML
outperform the non-ML, However, there is a threat to the validity to this models that are rarely compared with each other, it is difficult to
conclusion as this argument holds only for the three ML models (NN, IB, determine which is more accurate.
EL) vs. PLSR and (NN and EL) vs. SVR, which have been supported by a
sufficient number of experiments; while it is difficult to fully validate 14. Rq5: strengths and weaknesses of ml models
and generalize the conclusions for the other evaluations due the small
number of experiments comparing the performance of ML models with Selecting the appropriate ML models for the SIS contexts can be
the non-ML models. Thus, more number of studies comparing ML addressed by investigating the candidate ML models from their char­
models and other models for SIS should be conducted in order to procure acteristics, which are generally outlined by the strengths and weak­
satisfactory and generalized results. nesses of the ML techniques as recorded by researchers. Hence, the
strengths and weaknesses of the ML techniques supported by more than
13. Rq4: ml models vs. other ml models one study are introduced in this section and the detailed information is
presented in Table 8. The RF technique has been reported to exhibit
As for the comparisons between different ML models, we adopted the good performance in defect prediction problems and is also good in
same analysis scheme as that used in the comparisons between ML handling multiple datasets with varying properties. The SVM has been
models and non-ML models. That is, the comparisons were conducted on applauded for its excellent ability to deal with the redundant features
the same experiments in terms of R-Squared metric; one ML model is and high-dimensionality. NN works reliably with noisy data and has
considered to outperform another in an experiment if it achieves at least been proven to hold strong generalization and learning ability as well as
2 % improvement in estimation accuracy. adaptability.
Fig. 9 shows the overall results of the comparisons between different Summarily, different techniques have different pros and there is not
ML models, together with the corresponding number of supporting ex­ a one size fits all solution to the problem of SI in form of a ML technique.
periments. The bars above the zero-line depict the percentage of datasets It all depends on the domain of application; in a situation where false
in which the ML technique outperforms the other ML techniques. The positives are fatal would favor some techniques in comparison to a sit­
bars below the zero-line were only used when the ML technique did not uation where cost is a factor.
outperform the other ML techniques in any of the dataset. More details
of the comparison results can be found in Table 7. Three significant

9
A. Younes et al. Smart Agricultural Technology 7 (2024) 100425

Fig. 9. Comparisons of R-Squared between ML models.

15. Limitation performance measures. Third, we assessed the performance of the ML

techniques for smart irrigation system. Fourth, we compared the per­
When interpreting the results of this SLR, some limitations should be formance of models predicted using the ML techniques with the models
captured. Of note, This systematic review considered a number of predicted using non-ML techniques. Then, we analyzed the performance
selected studies to evaluate and assess the performance of various ML of these predicted models using the ML techniques with other ML
models amongst themselves and with Non-ML model. A limitation in this techniques. Finally, we summarized the strengths and weaknesses of the
review is that only 23 out of the 55 selected studies compare Non-ML ML techniques. The main findings obtained from the selected primary
and ML models, Thus, the ML vs non-ML comparison is not definitely studies are:
conclusive. Also, while comparing ML techniques, various experimental
settings are likely to be used by each study, which include validation • (Rq1) The ML techniques were broadly categorized into Decision
methods, feature selection methods and pre-processing methods to Tree, Bayesian Learning, Ensemble Learning, Evolutionary Algo­
remove outliers [68], Furthermore, this review considered only the rithms, Neural Networks, Fuzzy & Neuro Fuzzy based, Clustering,
metrics of Accuracy, RMSE, and R-Squared when evaluating the per­ Instance Based and Miscellaneous. The most frequently used ML
formance of ML models or comparing ML models with other models as techniques for Smart Irrigation system were artificial neural net­
they are among the most important ones and were used by most of the works, Support Vector Machines and Random Forest.
studies. However, considering other performance metrics such Preci­ • (Rq2) The results depict that Accuracy, RMSE, and R-Squared are the
sion, Recall, and F1-Score which were ignored in this review, could be a most commonly used performance measures in the selected studies.
valuable addition to the analysis strategy. finally, this review does not • (Rq3) ML model is more accurate than non-ML model in general,
include any unpublished research studies. We have assumed that all the which is supported by most of the studies. PLSR model is the non-ML
studies are impartial, however if this is not the case then it poses a threat model that is most often compared with ML models.
to this study. • (Rq4) Both SVM and EL are less accurate than NN, which are sup­
ported by most of the studies conducting the comparisons.
16. Conclusion and future guidelines • (Rq5) Every ML technique have different strengths and weaknesses.

In this paper we perform a systematic literature review in order to This review provides recommendations for researchers as well as
analyze and assess the performance of the ML techniques for smart guidelines for practitioner to carry out future research on Smart irriga­
irrigation system (SIS). First, after a thorough analysis by following a tion system using the ML techniques:
systematic series of steps, we identified 55 selected studies
(2017–2023). Second, we summarized the characteristics of the selected • More number of studies for Smart Irrigation system should be carried
studies based on metrics reduction techniques, metrics, and out using the ML techniques in order to obtain generalizable results

10
A. Younes et al. Smart Agricultural Technology 7 (2024) 100425

Table 7 Table 8
Comparisons of R-Squared between different ML models (‘‘+’’ indicates the Strengths and weaknesses of ML techniques.
model given in the row outperforms the model given in the column, ‘‘-’’ indicates Technique Strengths Supporting
the model given in the column outperforms the model given in the row; the name studies
number following study ID is R-Squared improvement in percentage, the study
RF It can handle large data and are consistent S34
ID in bold indicates the improvement exceeds 2 %).
performers. S50
ML ML Fast to train, robust towards parameter S34, S49.
setting. S50, S35.
SVM IB EL DT NF EA
Provide understandable model. S49, S35.
NN + S06 S12 S34(0.1), NA NA NA
Runs efficiently on large data sets.
(2.17), (2.71), S49(0.28),
Helps in identifying most important
S49 S12 S49(1.1),
independent variables.
(0.55), (2.39) S49(1.16),
SVM Good tolerance for high-dimension space and S06, S23
S49 S49(7.66),
redundant features. S06, S10
(1.39), S05(3.13),
Robust in nature. S49, S23
S10 S05(2.05)
It can handle complex functions. S23, S06, S10
(1.26)
It can handle nonlinear problems.
– S06 NA S34(1.53), NA S10 S32
NN Hold strong generalization and learning S05
(9.43), S34 (0.86), (1.68)
ability as well as adaptability. S34, S12
S06 (13.31), S10
Can do fast real-time computation with better S35
(1.18), S34(0.70), (21.88)
computational efficiency. S49
S49 S34(1.33),
Can avoid over-fitting.
(5.32), S50
Capable of dealing with noisy data
S49 (235.29),
KNN Intuitive and easy to understand S12
(6.57), S50
Robust in nature
S49 (46.80),
Technique Weaknesses Supporting
(1.31), S36(2.06),
name studies
S49 S49(4.23),
SVM It has very limited success when applied to S49
(0.13), S49(0.26)
imbalanced data sets.
S10
Choosing an adequate kernel function.
(1.18)
NN Require diversified training data set to train S05, S50
SVM + NA S23 S23(5.19), NA S10 NA
the model effectively. S06, S10
(5.94), S23(3), S23 (0.32)
Cost large computational resource.
S23 (11.64),
(3.22) S23(3.44),
S35
(33.97), CRediT authorship contribution statement
S49(1.05),
S49(7.8) Abiadi Younes: Formal analysis, Funding acquisition, Writing –
– NA NA S49(0.28), NA S10 NA original draft, Writing – review & editing. Zouhair Elamrani Abou
(23.42)
IB + NA NA NA S30 NA
Elassad: Supervision, Validation. Othmane El Meslouhi: Supervision,
(9.57) Validation. Dauha Elamrani Abou Elassad: Resources. Ed-dahbi
– NA NA S42 NA NA NA Abdel Majid: Resources.
(129.41),
S42
(136.36), Declaration of competing interest
S54
(73.53),
S54 The authors declare that they have no known competing financial
(81.81), interests or personal relationships that could have appeared to influence
S23 0.71), the work reported in this paper.
S23(0.21)

Data availability
and gain more proof on the viability of ML models. As there are few
studies that compare the ML techniques with the non-ML techniques, No data was used for the research described in the article.
more studies should compare the performance of the ML techniques
with non-ML techniques.
• There are very few studies that examine the effectiveness of evolu­ References
tionary algorithms such as GP, ACO for Smart Irrigation system. The
[1] H. Zia, A. Rehman, N.R. Harris, S. Fatima, M. Khurram, An experimental
future studies may focus on the predictive accuracy of evolutionary comparison of iot-based and traditional irrigation scheduling on a flood-irrigated
algorithms for Smart irrigation system. subtropical lemon farm, Sensors 21 (12) (2021), https://doi.org/10.3390/
• The ML techniques should be carefully selected based on their s21124175.
[2] Gaiqiang Yang, Lei Liu, Ping Guo, Mo Li, A flexible decision support system for
properties. Before making the decision of selection of a ML tech­
irrigation scheduling in an irrigation district in china, Agric. Water Manage. 179
nique, the characteristics of the ML technique should be understood (2017) 378–389, https://doi.org/10.1016/j.agwat.2016.07.019.
completely by the researcher. [3] Khaled Obaideen, et al., An overview of smart irrigation systems using iot, Energy
• ML models should be adopted in parallel with existing conventional Nexus 7 (2022), https://doi.org/10.1016/j.nexus.2022.100124.
[4] Zouhair Elamrani Abou Elassad, Ameksa Mohammed, Abou Elassad Dauha
models at the early stage in order to unlock the true potential of a Elamrani, Hajar Mousannif, “Efficient fusion decision system for predicting road
given ML technique. crash events: a comparative simulator study for imbalance class handling, Transp.
• The study should clearly specify the parameters values for the Res. Rec. (2023) 03611981231192985, https://doi.org/10.1177/
03611981231192985.
respective ML techniques used so that the framework in the studies [5] Jordan, M.I., and T.M. Mitchell. 2015. “Machine Learning: trends, Perspectives,
can be successfully repeated by the software community. and Prospects.” 349(6245).
[6] Ameksa, Mohammed, Hajar Mousannif, Hassan Al Moatassime, and Zouhair
Elassad. 2021. Crash Prediction Using Ensemble Methods. 10.5220/
0010731200003101.

11
A. Younes et al. Smart Agricultural Technology 7 (2024) 100425

[7] Zouhair Elamrani Abou Elassad, Hajar Mousannif, Hassan Al Moatassime, Class- [32] Jingxin Yu, et al., A deep learning approach for multi-depth soil water content
imbalanced crash prediction based on real-time traffic and weather data: a driving prediction in summer maize growth period, IEEE Access 8 (2020) 199097–199110,
simulator study, Traffic Inj. Prev. (2020) 1–8, https://doi.org/10.1080/ https://doi.org/10.1109/ACCESS.2020.3034984.
15389588.2020.1723794. [33] Amaury Dubois, Fabien Teytaud, Sébastien Verel, Short term soil moisture
[8] Elamrani Abou Elassad, Zouhair, Mohamed Ameksa, Dauha Elamrani Abou forecasts for potato crop farming: a machine learning approach, Comput. Electron.
Elassad, and Hajar Mousannif. 2023a. “Machine Learning Prediction of Weather- Agric. 180 (2021), https://doi.org/10.1016/j.compag.2020.105902.
Induced Road Crash Events for Experienced and Novice Drivers: insights from a [34] Paweena Suebsombut, et al., Field data forecasting using lstm and Bi-Lstm
Driving Simulator Study BT - Business Intelligence.” In eds. Rachid El Ayachi, approaches, Applied Sciences (Switzerland) 11 (24) (2021), https://doi.org/
Mohamed Fakir, and Mohamed Baslam. Cham: Springer Nature, Switzerland, 10.3390/app112411820.
57–73. [35] Sevim Seda Yamaç, Artificial intelligence methods reliably predict crop
[9] Lv, Haoyong, and Hengyao Tang. 2011. “Machine Learning Methods And Their evapotranspiration with different combinations of meteorological data for sugar
Application Research.” 10.1109/IPTC.2011.34. beet in a semiarid area, Agric. Water Manage. 254 (2021), https://doi.org/
[10] Kitchenham, Barbara. 2004. Procedures for Performing Systematic Reviews. 10.1016/j.agwat.2021.106968.
[11] M. Ameksa, et al., Toward Flexible Data Collection of Driving Behaviour, XLIV [36] Ali Al-Naji, Ahmed Bashar Fakhri, Sadik Kamel Gharghan, Javaan Chahl, Soil color
(October) (2020) 7–8, https://doi.org/10.5194/isprs-archives-XLIV-4-W3-2020- analysis based on a RGB camera and an artificial neural network towards smart
33-2020. irrigation: a pilot study, Heliyon 7 (1) (2021), https://doi.org/10.1016/j.
[12] Keele, Staffs. 2007. “Guidelines for Performing Systematic Literature Reviews in heliyon.2021.e06078.
Software Engineering”. [37] Guomin Shao, et al., Mapping maize crop coefficient kc using random forest
[13] Yu Feng, Yong Peng, et al., Modeling reference evapotranspiration using extreme algorithm based on leaf area index and UAV-Based multispectral vegetation
learning machine and generalized regression neural network only with indices, Agric. Water Manage. 252 (2021), https://doi.org/10.1016/j.
temperature data, Comput. Electron. Agric. 136 (2017) 71–78, https://doi.org/ agwat.2021.106906.
10.1016/j.compag.2017.01.027. [38] Blasi, Anas H, Al Karak, Jordan A Mohammad Abbadi, and Jordan Rufaydah Al-
[14] Yu Feng, Ningbo Cui, et al., Evaluation of random forests and generalized Huweimel. 2021. 11 Technology & Applied Science Research Machine Learning
regression neural networks for daily reference evapotranspiration modelling, Approach for an Automatic Irrigation System in Southern Jordan Valley. www.etasr.
Agric. Water Manage. 193 (2017) 163–173, https://doi.org/10.1016/j. com.
agwat.2017.08.003. [39] R. Aminuddin, A.S. Sahrom, M.H.A. Halim, Smart Irrigation System for Urban
[15] Amarendra Goap, Deepak Sharma, A.K. Shukla, C. Rama Krishna, An IoT based Gardening Using Logistic Regression Algorithm and Raspberry Pi, in: Journal of
smart irrigation management system using machine learning and open source Physics: Conference Series, IOP Publishing Ltd, 2021, https://doi.org/10.1088/
technologies, Comput. Electron. Agric. 155 (2018) 41–49, https://doi.org/ 1742-6596/2129/1/012044.
10.1016/j.compag.2018.09.040. [40] Ashutosh Bhoi, et al., IoT-IIRS: internet of things based intelligent-irrigation
[16] Olutobi Adeyemi, et al., Dynamic Neural Network Modelling of Soil Moisture recommendation system using machine learning approach for efficient water
Content for Predictive Irrigation Scheduling, Sensors (Switzerland) 18 (10) (2018), usage, Peer. J. Comput. Sci. 7 (2021) 1–15, https://doi.org/10.7717/PEERJ-
https://doi.org/10.3390/s18103408. CS.578.
[17] Mandeep Kaur Saggi, Sushma Jain, Reference evapotranspiration estimation and [41] André Glória, João Cardoso, Pedro Sebastião, Sustainable irrigation system for
modeling of the punjab northern india using deep learning, Comput. Electron. farming supported by machine learning and real-time sensor data, Sensors 21 (9)
Agric. 156 (2019) 387–398, https://doi.org/10.1016/j.compag.2018.11.031. (2021), https://doi.org/10.3390/s21093079.
[18] Chao Zhang, Jiangui Liu, Jiali Shang, Huanjie Cai, Capability of crop water content [42] H. Shalini, C.V. Aravinda, An IoT-Based Predictive Analytics for Estimation of
for revealing variability of winter wheat grain yield and soil moisture under limited Rainfall for Irrigation. Advances in Intelligent Systems and Computing, Springer,
irrigation, Sci. Total Environ. 631–632 (2018) 677–687, https://doi.org/10.1016/ 2021, pp. 1399–1413, https://doi.org/10.1007/978-981-15-3514-7_105.
j.scitotenv.2018.03.004. [43] Deepak Gautam, Bertram Ostendorf, Vinay Pagay, Estimation of grapevine crop
[19] Shadi AlZu’bi, et al., An efficient employment of internet of multimedia things in coefficient using a multispectral camera on an unmanned aerial vehicle, Remote
smart and future agriculture, Multimed Tools Appl. 78 (20) (2019) 29581–29605, Sens (Basel) 13 (13) (2021), https://doi.org/10.3390/rs13132639.
https://doi.org/10.1007/s11042-019-7367-0. [44] Samad Emamgholizadeh, et al., Numerical and artificial intelligence models for
[20] IEEE Staff. 2019. Machine Learning Based Soil Moisture Prediction for Internet of predicting the water advance in border irrigation, Env. Develop. Sustainabil. 24 (1)
Things Based Smart Irrigation System. IEEE. (2022) 558–575, https://doi.org/10.1007/s10668-021-01453-6.
[21] Asmae El Mezouari, Mehdi Najib, A Hadoop Based Framework for Soil Parameters [45] Ebrahim Babaeian, et al., Estimation of root zone soil moisture from ground and
Prediction, in: Proceedings - 15th International Conference on Signal Image Technology remotely sensed soil information with multisensor data fusion and automated
and Internet Based Systems, Institute of Electrical and Electronics Engineers Inc., machine learning, Remote Sens. Environ. 260 (2021), https://doi.org/10.1016/j.
2019, pp. 681–687, https://doi.org/10.1109/SITIS.2019.00111. SISITS 2019. rse.2021.112434.
[22] da Silva Júnior, José Clodoalves, et al., Random forest techniques for spatial [46] Osama Elsherbiny, Yangyang Fan, Lei Zhou, Zhengjun Qiu, Fusion of feature
interpolation of evapotranspiration data from brazilian’s northeast, Comput. selection methods and regression algorithms for predicting the canopy water
Electron. Agric. 166 (2019), https://doi.org/10.1016/j.compag.2019.105017. content of rice based on hyperspectral data, Agriculture (Switzerland) 11 (1)
[23] Akshay Murthy, et al., Machine Learning-Based Irrigation Control Optimization, in: (2021) 1–21, https://doi.org/10.3390/agriculture11010051.
BuildSys 2019 - Proceedings of the 6th ACM International Conference on Systems [47] Thu Thuy Nguyen, et al., A low-cost approach for soil moisture prediction using
for Energy-Efficient Buildings, Cities, and Transportation, Association for multi-sensor data and machine learning algorithm, Sci. Total Environ. 833 (2022),
Computing Machinery, Inc, 2019, pp. 213–222, https://doi.org/10.1145/ https://doi.org/10.1016/j.scitotenv.2022.155066.
3360322.3360854. [48] D Maria Manuel Vianny, et al., Water optimization technique for precision
[24] Sevim Seda Yamaç, Mladen Todorovic, Estimation of daily potato crop irrigation system using IoT and machine learning, Sustainable Energy Technologies
evapotranspiration using three different machine learning algorithms and four and Assessments 52 (2022), https://doi.org/10.1016/j.seta.2022.102307.
scenarios of available meteorological data, Agric. Water Manage. 228 (2020), [49] R. Bhavani, Thambi Ajoe, Managing irrigation needs based on smart decisions
https://doi.org/10.1016/j.agwat.2019.105875. using machine learning, ICTACT J. Soft Comput. 12 (2) (2022).
[25] Ravneet Kaur Sidhu, Ravinder Kumar, Prashant Singh Rana, Long short-term [50] Khadijeh Alibabaei, et al., Irrigation Optimization with a Deep Reinforcement
memory neural network-based multi-level model for smart irrigation, Mod. Phys. Learning Model: case Study on a Site in Portugal, Agric. Water Manage. 263
Lett. B 34 (36) (2020), https://doi.org/10.1142/S0217984920504187. (2022), https://doi.org/10.1016/j.agwat.2022.107480.
[26] Ali Risheh, Amirmohammad Jalili, Nazerfard Ehsan. Smart Irrigation IoT solution [51] Jiao Guo, et al., Soil moisture content estimation in winter wheat planting area for
using transfer learning for neural networks, IEEE, 2020. multi-source sensing data using CNNR, Comput. Electron. Agric. 193 (2022),
[27] Joao Cardoso, Andre Gloria, Pedro Sebastiao, Improve Irrigation Timing Decision https://doi.org/10.1016/j.compag.2021.106670.
for Agriculture Using Real Time Data and Machine Learning, in: 2020 International [52] Khongdet Phasinam, et al., Application of IoT and cloud computing in automation
Conference on Data Analytics for Business and Industry: Way Towards a of agriculture irrigation, J. Food Qual. 2022 (2022), https://doi.org/10.1155/
Sustainable Economy, ICDABI, Institute of Electrical and Electronics Engineers Inc, 2022/8285969.
2020, p. 2020, https://doi.org/10.1109/ICDABI51230.2020.9325680. [53] Khalil Ibrahim Mohammad Abuzanouneh, et al., Design of Machine Learning Based
[28] BHANU, K. N., MAHADEVASWAMY, H. S., et JASMINE, H. J. IoT based smart Smart Irrigation System for Precision Agriculture, Comput. Mat. Continua 72 (1)
system for enhanced irrigation in agriculture. In : 2020 International Conference on (2022) 109–124, https://doi.org/10.32604/cmc.2022.022648.
Electronics and Sustainable Communication Systems (ICESC). IEEE, 2020. p. 760-765. [54] Minghan Cheng, et al., Estimation of soil moisture content under high maize
[29] Jalal Shiri, et al., Simulating wetting front dimensions of drip irrigation systems: canopy coverage from uav multimodal data and machine learning, Agric. Water
multi criteria assessment of soft computing models, J Hydrol (Amst) 585 (2020), Manage. 264 (2022), https://doi.org/10.1016/j.agwat.2022.107530.
https://doi.org/10.1016/j.jhydrol.2020.124792. [55] P. Suresh, et al., Iot with evolutionary algorithm based deep learning for smart
[30] Roberto Filgueiras, et al., Soil water content and actual evapotranspiration irrigation system, Comput. Mat. Continua 71 (1) (2022) 1713–1728, https://doi.
predictions using regression algorithms and remote sensing data, Agric. Water org/10.32604/cmc.2022.021789.
Manage. 241 (2020), https://doi.org/10.1016/j.agwat.2020.106346. [56] Juan Dong, et al., Nation-scale reference evapotranspiration estimation by using
[31] Lucas Borges Ferreira, Fernando França da Cunha, New approach to estimate daily deep learning and classical machine learning models in china, J. Hydrol (Amst)
reference evapotranspiration based on hourly temperature and relative humidity 604 (2022), https://doi.org/10.1016/j.jhydrol.2021.127207.
using machine learning and deep learning, Agric. Water Manage. 234 (2020), [57] Dushyant Kumar Singh, et al., IoT-driven model for weather and soil conditions
https://doi.org/10.1016/j.agwat.2020.106113. based on precision irrigation using machine learning, Security and Communication
Networks 2022 (2022), https://doi.org/10.1155/2022/7283975.

12
A. Younes et al. Smart Agricultural Technology 7 (2024) 100425

[58] S. Premkumar, A.N. Sigappi, IoT-enabled edge computing model for smart Science Part A: Engineering and Innovation 9 (4) (2022) 378–391, https://doi.org/
irrigation system, J. Intell. Syst. 31 (1) (2022) 632–650, https://doi.org/10.1515/ 10.54287/gujsa.1141575.
jisys-2022-0046. [64] Abhishek Raghuvanshi, et al., Intrusion detection using machine learning for risk
[59] Estimation of transpiration coefficient and aboveground biomass in maize using mitigation in IoT-Enabled smart irrigation in smart farming, J. Food Qual. 2022
time-series UAV multispectral imagery, Crop Journal 10 (5) (2022), https://doi. (2022), https://doi.org/10.1155/2022/3955514.
org/10.1016/j.cj.2022.08.001, 1376–85. [65] Youness Tace, et al., Smart irrigation system based on IOT and machine learning,
[60] Varun Yarehalli Chandrappa, Biplob Ray, Nanjappa Ashwatha, Pramod Shrestha, Energy Reports 8 (2022) 1025–1036, https://doi.org/10.1016/j.egyr.2022.07.088.
Spatiotemporal modeling to predict soil moisture for sustainable smart irrigation, [66] Yu Zhang, et al., Evaluating soil moisture content under maize coverage using UAV
Internet of Things (Netherlands) (2023) 21, https://doi.org/10.1016/j. multimodal data by machine learning algorithms, J. Hydrol (Amst) 617 (2023),
iot.2022.100671. https://doi.org/10.1016/j.jhydrol.2023.129086.
[61] Zongjun Wu, et al., Simulation of daily maize evapotranspiration at different [67] Ahmed Elbeltagi, et al., Forecasting vapor pressure deficit for agricultural water
growth stages using four machine learning models in semi-humid regions of management using machine learning in semi-arid environments, Agric. Water
northwest China, J. Hydrol (Amst) 617 (2023), https://doi.org/10.1016/j. Manage. 283 (2023), https://doi.org/10.1016/j.agwat.2023.108302.
jhydrol.2022.128947. [68] Zouhair Elamrani Abou Elassad, Hajar Mousannif, Hassan Al Moatassime,
[62] Prediction of maize crop coefficient from UAV multisensor remote sensing using Aimad Karkouch, The application of machine learning techniques for driving
machine learning methods, Agric. Water Manage. 276 (2023), https://doi.org/ behavior analysis: a conceptual framework and a systematic literature review, Eng.
10.1016/j.agwat.2022.108064. Appl. Artif. Intell. 87 (2020), https://doi.org/10.1016/j.engappai.2019.103312.
[63] Aamo IORLİAM, et al., Machine Learning Techniques for the Classification of IoT-
Enabled Smart Irrigation Data for Agricultural Purposes, Gazi University Journal of

13