IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 9, NO.

12, DECEMBER 2022 2055

Perspective

DeCASA in AgriVerse: Parallel Agriculture for

Smart Villages in Metaverses
Xiujuan Wang, Mengzhen Kang, Hequan Sun, Philippe de Reffye, Fei-Yue Wang, Fellow, IEEE

Briefing: The demand for food is tremendously increasing will increase by 1.5 °C by 2050, which can dramatically
with the growth of the world population, which necessitates reduce crop yields [2]. This has caused severe problems for
the development of sustainable agriculture under the impact human survival and development. To ensure food security, we
of various factors, such as climate change. To fulfill this must make every effort to prevent climate change and develop
challenge, we are developing Metaverses for agriculture, re- sustainable agriculture.
ferred to as AgriVerse, under our Decentralized Complex With the development of the fourth industrial revolution
Adaptive Systems in Agriculture (DeCASA) project, which is (Industry 4.0), Agriculture 4.0 is coming up, which is more
a digital world of smart villages created alongside the devel- autonomous and intelligent by integrating the emerging tech-
opment of Decentralized Sciences (DeSci) and Decentralized nologies such as the Internet of Things (IoT), robotics, big data
Autonomous Organizations (DAO) for Cyber-Physical-Social and artificial intelligence (AI) into agriculture [3]. However,
Systems (CPSSs). Additionally, we provide the architectures, with the advancement of information and communications
operating modes and major applications of DeCASA in Agri- technologies, the effects of human behavior have been pro-
Verse. For achieving sustainable agriculture, a foundation gressively integrated into agricultural management and control.
model based on ACP theory and federated intelligence is en- Agriculture 4.0 is no longer adequate to address human and
visaged. Finally, we discuss the challenges and opportunities. social factors in systems [4], [5]. As a result, smart agriculture
Keywords: Parallel Agriculture Management and Control, has been converted from Cyber-Physical-System (CPS) to
AgriVerse, Agriculture CPSS, ACP, DAO-Based Platform, Cyber-Physical-Social Systems (CPSSs), where society will
Precision Agriculture. step toward to the fifth industrial revolution (or Industries
5.0). In Industries 5.0, biotechnology, information technology
I. I NTRODUCTION (IT), artificial intelligence (AI) are deeply integrated with the
principles of human centrality, sustainability, and resiliency.
CCORDING to the United Nations report, the global
A human population will increase by approximately 25%
by 2050, reaching nearly 10 billion [1]. As a result, the
This lays the foundation for revolutionizing agriculture by
systematically linking the microworld, where gene expression,
regulation and interaction can be deciphered at the molecular
demand for food will increase massively. Food security, known level, with the macroworld, where crop phenotyping, growth
as availability, access, stability and use of food, is likely to modeling, natural/social plant-growth environment monitor-
deteriorate along with the negative impact of other factors such ing, human intervention and management are carried out
as climate change. For example, the global surface temperature systematically [6]. Specifically, crop phenotyping, i.e., the
process of measuring the quantity, quality, photosynthesis,
Corresponding author: Mengzhen Kang and Fei-Yue Wang.
Citation: X. Wang, M. Kang, H. Sun, P. de Reffye, and F.-Y. Wang, “De- development, architecture, growth or biomass productivity of
CASA in AgriVerse: Parallel Agriculture for Smart Villages in Metaverses,” crop plants, has become feasible in a high-throughput manner
IEEE/CAA J. Autom. Sinica, vol. 9, no. 12, pp.2055–2062, Dec. 2022. with intelligent robotics that are developed with techniques in
X. Wang is with The State Key Laboratory for Management and Control
of Complex Systems, Chinese Academy of Sciences, Beijing 100190, and computer vision, pattern recognition and AI. More generally,
also with the Beijing Engineering Research Center of Intelligent Systems and comprehensive spatial and ground information in agricultural
Technology, Beijing 100190, China (e-mail: [email protected]). production that influences the survival and development of
M. Kang is with The State Key Laboratory for Management and Control of
Complex Systems, Chinese Academy of Sciences, Beijing 100190, and also crops will cover agricultural stages, including preproduction
with the School of Artificial Intelligence, University of Chinese Academy of (scheduling, market and demand analysis, plant optimization,
Sciences, Beijing 100049, China (e-mail: [email protected]). etc.), interproduction (planting task management, environmen-
H. Sun is with Department of Chromosome Biology, Max Planck Institute
for Plant Breeding Research, Carl-von-Linné-Weg 10, Cologne D-50829, and tal control, soil analysis, fertilization, spraying, irrigation,
also with Faculty of Biology, LMU Munich, Großhaderner Str. 2, Planegg- usage of pesticide and herbicide, etc.) and postproduction
Martinsried D-82152, Germany (e-mail: [email protected]). (harvest, storage, processing, transportation and sales, logistics
P. de Reffye is with AMAP, Univ Montpellier, CIRAD, CNRS, INRAE,
IRD, Montpellier F-34398, France (e-mail: [email protected]). scheduling, etc.).
F.-Y. Wang is with The State Key Laboratory for Management and Control With the large agricultural data at the terabyte level, there
of Complex Systems, Chinese Academy of Sciences, Beijing 100190, and also are great challenges for data storage, transfer and sharing.
with the Macao Institute of Systems Engineering, Macau University of Science
and Technology, Macao 999078, China (e-mail: [email protected]). Emerging technologies have shown the potential to address
Digital Object Identifier 10.1109/JAS.2022.106103 these challenges, such as edge computing and Decentralized
2056 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 9, NO. 12, DECEMBER 2022

Autonomous Organizations (DAO) [7], which leverage tech- augmented virtual (AR), mixed reality (MR), and extended
nologies such as blockchains to allocate resources, coordinate reality (XR), cloud computing, edge computing, blockchain,
activities, and make decisions without centralized control or AI and other technology [16]. The difference in technology
third-party intervention [8]. Another challenge is to integrate between AgriVerse and Metaverse lies in the use of various
and analyse data on different dimensions and scales. Models agricultural models regarding the environment, crops and
have been developed on a broad scale from the diagnosis of farmers.
crop disease to yield prediction [9]. However, as a biological In AgriVerse, the real and virtual systems interactively run
entity, crops and their diseases exhibit dynamic behaviors, in parallel, and any agriculture-relevant information can be
and continuous monitoring can be prohibitively expensive and perceived, recorded and analyzed automatically, intelligently
thus impossible to apply. Mechanistic models are able to and systematically to provide optimal solutions or decision-
compensate for empty sites and create links among different making guidelines for the management and control of practical
data. In this regard, extensive studies have been conducted agricultural production. Fig. 1 shows the conceptual frame-
over the last century, including mathematical modeling of the work of DeCASA in AgriVerse based on CPSS. The basis
production and allocation of biomass in relation to the shape, is the interaction between the real and artificial agricultural
structure and ontogenesis of plants. systems (AASs) and their corresponding management centers.
The origin of our agricultural related work can be traced AASs simulate and optimize the dynamic processes related
back to the PhD of de Reffye in 1970s [10], who draw the to agricultural production, the environment, and management
first 2D plant of a coffee tree in Africa. In 1990s, Wang et al. activities around the production chain.
built “virtual or shadow plant” on a computer by monitoring
plants with a camera based on a project granted by the
University of Arizona, Tucson, Arizona and Biosphere 2, Oro
Valley, Arizona. Around the year 2000, the GreenLab model, a
mathematical model describing the development and growth of
plants, was developed through Sino-French cooperation [11].
In 2005, the work of Complex Adaptive Systems for Smart
Agriculture (CASA) characterized by the parallel intelligence
was initiated [12] as the integration of the above works.
Subsequently, at the system level, we proposed a framework,
called DiCASA for distributed CASA for parallel agriculture
[6] following the ACP theory [13] for CPSSs, where ‘A’ refers
to an artificial system, ‘C’ refers to computational experiments
and ‘P’ refers to parallel execution. In recent years, Metaverse
has emerged as a collective virtual space with decentralized
and collaborative features where people will conduct activities
associated with education, sales, entertainment, etc. Along the
Fig. 1. DeCASA in AgriVerse based on CPSS.
lines of Metaverses and parallel intelligence, we are building
Metaverses for agriculture, referred to as DeCASA in Agri-
Verse since 2021 [14], designed along with the development The agricultural operating system (AgOS) is an exten-
of DeSci [7] and DAO for CPSS. With social information sive platform that addresses the management and control of
such as market demand and prices, DeCASA in AgriVerse agricultural facilities and software for various tasks through
can provide decision support for agricultural systems in reality the interaction between the real agricultural system and its
via in-depth computational experiments in the corresponding corresponding AASs. Applications for various purposes, such
virtual system [4]. In short, the development of technologies as climate monitoring and control [17], data analysis and sim-
and theories in biology, computer science, control, communi- ulation, scheme recommendation [4], performance evaluation,
cation, AI and DAO make it possible to improve agriculture and visualization of results, can be integrated into AgOS.
by harmoniously joining the biological, physical and digital For example, information on topography and crop planting
worlds. The challenges, opportunities and future direction of distribution can be displayed with the platform (Fig. 2(a)),
AgriVerse are discussed. the monitored data of the environment and crop growth status
during different growth stages can also be analyzed (Fig.
2(b)), and the 3D architectures of the plant can be displayed
II. DEFINITION OF AGRIVERSE
using Qingyuan (Figs. 2(c) and 2(d)) [18]. In addition, IoT-
AgriVerse is composed of smart villages and farms that are based hardware can be managed (e.g., add, remove, rename)
characterized by automation, digitalization, computerization and controlled (e.g., switch on/off) remotely. Based on the
and intellectualization. The design of the AgriVerse platform analysis of data and forecasts of agricultural commodity prices
is based on the DAO principle [15] and the ACP methodology [19], crop phenology [20] and yield [21], optimized crop
[6], as defined in [14]. Similar to the Metaverse, the common planting schemes [22] will be recommended. Plant growth
technologies used in the AgriVerse also embodies a conver- can be simulated under a variety of environmental conditions
gence of web technologies, the internet, virtual reality (VR), and management operations. It also supports performance
WANG et al.: DECASA IN AGRIVERSE: PARALLEL AGRICULTURE FOR SMART VILLAGES IN METAVERSES 2057

evaluation of both real and artificial systems. With the ACP (Fig. 4) [25]. The coordination layer organizes the interactions
features, AgOS supports three operating modes for different and actions of the real system and AAAs. According to the
purposes, described below. analysis results obtained through simulations and predictions,
the planting scheme can be given; for example, an environ-
ment control prescription can be set according to the current
environment and plant demand [17]. As a result, AgOS can
send a message to the control facilities in the execution layer
to adjust the installations and thus control the environmental
conditions. The organization layer manages all the resources in
the system, including all the agricultural planting technologies,
the stakeholder’s database, and the typical agricultural scene
tests. In the performance evaluation module, the strategies can
be tested in actual agricultural scenes; then, by interacting with
agricultural experts, the planting technologies stored in the
database can be optimized and updated. The same processes
can also be used in agricultural machinery management and
Fig. 2. System interface of agricultural management and control platform. control. The tasks of machinery, including irrigation, fertiliza-
(a) A topography image of a farm. (b) An analysis display of environment
monitoring. (c) 3D visualization of tomato plant simulated in Qingyuan. (d)
tion and harvesting, can be distributed in real-time.
Plant library based on the GreenLab model.

C. Learning and Training

In this mode, with the help of virtual reality and human-
A. Experimentation and Evaluation computer interaction technologies, users could interact with
In this mode, the various agricultural models supporting different agricultural learning and training modules, which
AASs are used for various simulations and predictions [23], enables farmers and administrators to quickly grasp planting
as described in Fig. 3. AgOS monitors and obtains various technologies and management and control methods during the
data and stores all the information in data centers, including agricultural production process. Trained farmers would be able
climatic data and crop phenotyping information, social and to monitor and evaluate the performance of recommendation
economic information, etc., and those from expert knowledge. systems, performing a “human-in-the-loop” control. In turn,
In the experimental design, computational experiments can be feedback from farmers and agricultural experts will help AgOS
set up to generate various agricultural scenarios using sup- improve and strengthen agricultural management and control
porting models, including climate models, agricultural price intelligence.
models, crop growth models, and farmer behavior models.
As a CPSS, predicting agricultural commodity prices [19] can III. INTELLIGENT DECISION SUPPORT BY
help farmers adjust their harvest time and thus improve their AGRIVERSE
profits by integrating a profit model [22]. For example, crop
phenology and yield can be simulated and predicted under The core of AgriVerse is an accurate decision support for
various experimental conditions using statistical models or agriculture-related planning, management and control, based
process-based models such as ‘GreenLab’ [24]. In the per- on the perception and analysis of various agricultural data.
formance evaluation, the result of computational experiments Greenhouse management and control is a typical scenario for
can be analyzed; emerging properties can appear and bring applying AgOS. We have developed a parallel agricultural
new knowledge which is hard to find in the real world. The system based on ACP theory, JJfarmer, which has been named
results can be displayed in the graphs for numeric data or after “Awakening of Insects (Jing Zhe, one of the 24 solar
in 3D for crop architectures that clearly show the results of terms used in China for agricultural planning)” [4]. Fig. 5
computational experiments. In the decision generation, the shows the architecture of JJfarmer. Information perception
planting scheme recommendation can be given according to includes climate conditions (temperature, humidity and light,
the simulation and prediction results, such as when to plant, etc.) and crop growth status during crop growth, the status of
how many to plant and how to plant during the agricultural greenhouse facilities (on/off), and social information includes
production process. In addition, by interacting with actual the product price (reflecting the balance of product supply and
situations, AgOS could optimize and evaluate the performance demand), growers’ abilities, etc.
of recommended strategies. Evaluation results are stored in Intelligent decision support can provide services cover-
data centers to help farmers and farm administrators make ing the entire pre-inter-post agricultural production process.
decisions. Firstly, suggestions are made according to prior knowledge.
Secondly, decisions need to be updated based on the actual
environment and management during actual agricultural pro-
B. Management and Control duction. For such a complex farming system, ACP theory can
In this mode, three layers are involved, namely, the coor- be applied to descriptive, predictive, and prescriptive learning
dination layer, the execution layer and the organization layer [6]. The objectives of decision support include crop planning,
2058 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 9, NO. 12, DECEMBER 2022

Fig. 3. Agricultural Operations System for Management and Control (AgOS) in Experimentation and Evaluation Mode.

Fig. 4. Agricultural Operations System for Management and Control (AgOS). AASs represents Artificial Agricultural Systems.

general, JJfarmer achieves real-time control in three steps:

1) producing virtual greenhouse data using a greenhouse
climate model; 2) pretraining the greenhouse climate model
according to the produced virtual data and calibrating it using
the historical climate data collected from inside and outside
this greenhouse [17]; 3) transferring the model to the real
greenhouse, or learning the knowledge and experience of
the farmers from the real greenhouse system, and finally
giving the real-time control strategy recommendation. In this
greenhouse, the environmental monitoring facilities measure
air and soil temperature and humidity using low-cost sensors
for smallholders. Control facilities can adjust the size of the
vent opening according to the calculation results based on the
monitored data.
At the modeling stage, the greenhouse environment and crop
growth status can be modeled using the monitored data and
Fig. 5. JJfarmer system for environmental monitoring and control, as well
as production management during the entire process of the pre-inter-post
climate models. The real greenhouse and virtual systems could
agricultural production. operate and be adjusted simultaneously (Fig. 6). In the real
target greenhouse, the observed data are obtained for model
calibration. With greenhouse sensors gathering real-time data,
the climate model of the greenhouse could be refined. Finally,
environmental monitoring and control, facility control, harvest the training model generates a unique solution for each specific
management, contract matching, etc. planting condition, which is stored in the recommendation
Decision support from this system for monitoring and system. The modeling processes are demonstrated in Fig. 6,
controlling the environment of a solar greenhouse has been which consists of two basic modules, the real greenhouse sys-
successfully implemented in Changping District, Beijing. In tem and the virtual greenhouse systems. In a real greenhouse
WANG et al.: DECASA IN AGRIVERSE: PARALLEL AGRICULTURE FOR SMART VILLAGES IN METAVERSES 2059

system, farmers adjust the control system, taking into account To empower AgriVerse and improve its function in agri-
the actual situation and the historical experience of greenhouse cultural management and control in an intelligent and auto-
management systems. Farmer control strategies can be derived matic way, by integrating parallel learning theory [32], [33]
from monitored data [26], which can be the starting point for and federated intelligence [34], we propose an Agricultural
an standalone system. Foundation Model for it, as shown in Fig. 7. It should be noted
that this is very comparable to the Transportation Foundation
Model designed for the transportation system in ‘TransVerse’,
as described in [35].
In its data perception stage, the development of novel
technologies in sensors, remote sensing, cameras and UAVs
provides the efficient infrastructure in this framework for
collecting data [30], including agricultural domain knowledge
and rules, monitored environmental data, crop growth sta-
tus and market prices, etc., as well as data produced from
other production management platforms, from the microworld
where molecular interactions take place to the macroworld
where crop plants grow and enter human society. Data types
include text, imagery, audio and video. Recently, technolo-
gies developed in information and biology, in particular
genome/transcriptome sequencing, phenotyping, and knowl-
edge automation, have become indispensable in the production
Fig. 6. The facilities, system and application scenarios of greenhouse
environment control. and processing of data.
For a better description, agricultural simulation and pre-
dictive models should also take into account the relationship
At the recommendation stage, an appropriate control strat-
between the environment, crops and farmers. Models can be
egy can be recommended for an real greenhouse system
grouped into three categories: data-driven, knowledge-driven,
depending on the current climate and crop growth states.
and data/knowledge-driven. The data-driven models can learn
In addition, the climate control model can also be used to
from data without using any domain knowledge, such as
test the effect of a greenhouse climate control algorithm,
support vector machines, random forests, and artificial neural
which provides feedback to the control system, forming a
networks, which can be used to predict the climate of crop
closed loop. The algorithm sends the control strategy to the
growth or the market price of crops, to identify the pest, and
greenhouse actuators and automatically triggers the execution.
to diagnose the disease of crops. The knowledge-driven models
The greenhouse climate model and control algorithm will be
are derived from domain knowledge, including crop phenology
updated regularly to maintain synchronization with the real
and growth models, which can be used to simulate crop
greenhouse. In parallel control, operations of actual control
growth and development. With the development of AI, IoT and
strategies will be progressively enhanced by comparison, eval-
big data, studies on integrating these two types of modeling
uation, and interaction with virtual control algorithms [27]–
approaches have been conducted to take advantage of both
[29]. With this framework, it is able to control the indoors
knowledge and data-driven models to reduce the demand for
climate in an optimal way and minimize the need for human
the mass of data and improve the training efficiency, leading
intervention and specialized knowledge.
to so-called knowledge- and data-driven models, such as those
used in [21] to predict crop yield and in [36] to incorporate
IV. FOUNDATION MODEL FOR AGRIVERSE human domain knowledge into the neural network model in
With the development of technologies in big data and IoT, semantic segmentation.
etc., a large amount of data has been obtained [30], where AI All the monitored data, domain knowledge, rules and
technology can be used to perform data mining and combine models are crucial to building the Agricultural Foundation
expert experience and more domain knowledge to provide Model underlying AgriVerse. First, the transformer archi-
decision-making support for agriculture [31]. Although it tecture [37] is introduced to integrate deeper bidirectional
becomes possible to collect agricultural data at multiple scales, encoders and to evolve towards larger models and data sets, as
dimensions and modalities even with the help of AI-powered explained in [35]. To train the Agricultural Foundation Model
robots, intelligent computational methods for analyzing such to achieve proper operating processes through the collected
big data are still not sufficient. Given the heterogeneity of data, in the multimodal feature extraction module, the self-
the circumstances, it is difficult to obtain comprehensive, attention mechanism [38], [39] is used, which can capture
accurate and up-to-date agricultural information or knowledge. long-term dependencies with higher performance. Second,
In addition, existing predictive models tend to be overflow and federated intelligence is incorporated to ensure autonomous
cannot be used under different climatic conditions and fields management of the DAO-based agricultural system. Within
with regional, seasonal and cyclical characteristics during agri- this system, agricultural data is securely stored, maintained,
cultural production. These facts restrict the power of AgriVerse updated, shared and exchanged among stakeholders, includ-
to support decisions. ing agricultural input suppliers, farmers, machinery suppliers,
2060 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 9, NO. 12, DECEMBER 2022

Fig. 7. Future AgriVerse with Agricultural Foundation Model.

financial service providers, consumers, and farm administra- of most rural communities is agriculture, the development of
tors. In AgriVerse, all the processes related to agriculture smart villages must include a roadmap to promote agricultural
are to be virtually achieved, including planning, planting, methods, optimize agricultural human resources, and intro-
processing, packaging, storing, distribution, resale, preparation duce new technologies and processes. As a result, there is
and consumption along the agri-food chain until the food an increasingly diverse and complex demand for intelligent
arrives on our plates in reality [40]. Such an environment can agricultural services that must be addressed. ‘Agricultural
also be a useful sandbox for testing AI techniques to bring Brain’ has been put forward in many provinces of China and
about a responsible AI for agriculture [41]. Therefore, ideally, is regarded as a significant means for the management and
through computational learning from substantial practical data, control of agricultural production as well as a new paradigm
AgriVerse would behave automatically and intelligently by for achieving sustainable rural development.
taking specific information as input and calculating optimal In fact, agriculture has faced many complex challenges, in-
solutions at any stage of the agricultural production process. cluding climate change, environmental pollution, food scarcity
Therefore, we can assist farmers and agricultural sectors make and waste, natural resources constraints, and uncertainties in
agricultural management and control decisions. In addition, the agricultural productivity [44]. Although large-scale research
AgriVerse platform can adjust previous solutions for real-time on intelligent agricultural techniques is underway and several
situations as it is based on parallel learning theory and federal applications are available, the wide utilization is still insuffi-
intelligence. Thus, the AgriVerse platform can be applied to cient, e.g., the identification of pests, disease detection, yield
education, farming training and government management, etc. prediction, and the planning of fertilizer and pesticide use [45].
for example, it is possible to use the AgriVerse to train farmers When it comes to addressing real-world issues and solving
on farming skills or to remotely control robots conducting them through autonomous decisions and predictive solutions,
harvesting tasks. modern precision agriculture is still in its infancy.
DeCASA in AgriVerse offers a solution to this end; how-
V. CHALLENGES AND PERSPECTIVES ever, the construction of AgriVerse is also only in its infancy.
Major technology developments in the past couple of Firstly, since VR and AR information is primarily visual, the
decades, particularly in the areas of telecommunications, com- foundations are not established in the provision of agricultural
puting and AI, focused on the development of urban areas, production, which need the use of all five senses, including
where the concept of smart cities evolved. The level of touch [16]. Secondly, many agricultural management and
computerization in rural areas is far behind, leading to what control systems are at the research stage, which is still difficult
is known as the “urban–rural divide”. to apply in real agricultural production and for different field
To alleviate this issue, it has been recommended to develop areas. Current agricultural models are mainly dedicated to
smart villages to provide a bundle of services which are deliv- specific scenarios that cannot be dynamically adapted and
ered to its residents and businesses in an efficient and effective are difficult to implement in practice. There is a serious
manner [42]. Smart villages are expected to bridge the ur- information imbalance between supply and demand in prac-
ban–rural divide. The smart village can be implemented using tice; for example, farmers face a digital divide in agricultural
modern digital technologies, modern management techniques productivity and economic and social integration. Information
and current social sensibilities [43]. As the economic engine asymmetry (between supply and demand) and lack of knowl-
WANG et al.: DECASA IN AGRIVERSE: PARALLEL AGRICULTURE FOR SMART VILLAGES IN METAVERSES 2061

edge have resulted in soil pollution, unmarketable products, Academy of Sciences-Thailand National Science and Tech-
and economic losses [4]. Furthermore, collecting, processing nology Development Agency Joint Research Program under
and using agricultural productivity data also face a broad range Grant GJHZ2076.
of challenges. For example, farmers are facing data security
and privacy challenges in the information age. In addition, R EFERENCES
data availability and quality issues are often encountered in [1] United Nations Department of Economic and Social Affairs, Population
agricultural information systems [46]. Division, “World population prospects 2022: Summary of results,”
To address these challenges, emerging technologies have Report, 2022.
[2] S. Vermeulen, P. Aggarwal, B. Campbell, E. Davey, E. Grainger-
recently led to exciting innovations in agriculture, offering Jones, and X. Yao, “Climate change, food security and small-scale
unprecedented opportunities to build an intelligent agricul- producers: Analysis of findings of the fifth assessment report (AR5)
tural system. Research in intelligent agriculture is rapidly of the intergovernmental panel on climate change (IPCC),” 2014.
[3] T. Wang, Y. Liu, M. Wang, Q. Fan, H. Tian, X. Qiao, and Y. Li,
developing with advances of deep learning techniques and “Applications of UAS in crop biomass monitoring: A review,” Frontiers
their various variants [45]. The advancement of blockchain, in Plant Science, vol. 12, p. 616689, 2021.
smart contracts, DAOs and Web 3.0 creates a solid data and [4] M. Kang, X.-R. Fan, J. Hua, H. Wang, X. Wang, and F.-Y. Wang, “Man-
aging traditional solar greenhouse with cpss: A just-for-fit philosophy,”
trust foundation for future smart villages characterized by IEEE Transactions on Cybernetics, vol. 48, no. 12, pp. 3371–3380, 2018.
human–machine integration [47] and virtual-real interaction [5] M. Kang, X. Wang, J. Hua, H. Wang, and F.-Y. Wang, “Parallel
[48]. Going forward, it is expected that the above issues can agriculture: intelligent technology toward smart agriculture,” Chinese
Journal of Intelligent Science and Technology, vol. 1, no. 2, pp. 107–
be addressed through the Agricultural Foundation Model for 117, 2019.
AgriVerse. [6] M. Kang and F.-Y. Wang, “From parallel plants to smart plants: Intelli-
The goal of AgriVerse is to achieve food security through gent control and management for plant growth,” IEEE/CAA Journal of
Automatica Sinica, vol. 4, no. 2, pp. 161–166, 2017.
precision agriculture with accurate resource planning and [7] F.-Y. Wang, W. Ding, X. Wang, J. Garibaldi, S. Teng, R. Imre, and
agricultural production in a safe, cost-effective and environ- C. Olaverri-Monreal, “The DAO to DeSci: AI for free, fair, and respon-
mentally friendly way. The development of agriculture is sibility sensitive sciences,” IEEE Intelligent Systems, vol. 37, no. 2, pp.
16–22, 2022.
inseparable from those involved in agriculture and village [8] F.-Y. Wang, “The DAO to MetaControl for MetaSystems in Metaverses:
conditions. The connotation of rural revitalization includes the The system of parallel control systems for knowledge automation and
habitability of the village. It is expected that in the future, control intelligence in CPSS,” IEEE/CAA Journal of Automatica Sinica,
vol. 9, no. 11, pp. 1899–1908, 2022.
on the basis of Web3 and DAO, AgriVerse will transform [9] L. C. Ngugi, M. Abelwahab, and M. Abo-Zahhad, “Recent advances
smart villages into “6S” societies with “6I”, that is, Safe in in image processing techniques for automated leaf pest and disease
the physical world, Secure in the cyber world, Sustainable recognition – a review,” Information Processing in Agriculture, vol. 8,
no. 1, pp. 27–51, 2021.
in the ecological world, Sensitive to individual needs, Serves [10] P. de Reffye, “Modelisation de l’architecture des arbres par des processus
for all, and Smart in all, with cognitive intelligence and stochastiques. simulation spatiale des modeles tropicaux sous l’effet de
parallel intelligence for intelligent science and technology, la pesanteur. application au coffea robusta,” Thesis, 1979.
[11] B.-G. Hu, P. de Reffye, X. Zhao, H. Yan, and M. Kang, “Greenlab: A
crypto intelligence and federated intelligence for intelligent new methodology towards plant functional-structural model-structural
operations and management, and social intelligence and eco- part,” in PMA03: The First International Symposium on Plant Growth
logical intelligence for smart development and sustainability Modeling, Simulation, Visualization and their Applications. Tsinghua
University Press, Springer, 2003, Conference Proceedings, pp. 21–35.
[49]. [12] F.-Y. Wang, “Digital agriculture and parallel intelligence: Towards
complex adaptive systems for smart agriculture,” CASIA, CASKL-CSIS
Tech Report, October 2005.
VI. C ONCLUSION [13] F.-Y. Wang, “On the modeling, analysis, control and management of
In this paper, we present AgriVerse, a DAO-based smart complex systems,” Complex Systems and Complexity Science, vol. 3,
no. 2, pp. 26–34, 2006.
virtual digital world, as a specific implementation of the CPSS. [14] X. Wang, M. Kang, J. Hua, H. Wang, and F.-Y. Wang, “AgriVerse:
The construction of DeCASA in AgriVerse builds on our Application of Metaverses in agriculture based on CPSS,” International
research work on parallel agriculture since the beginning of the Journal of Intelligent Control and Systems, vol. 2, no. 1, pp. 1–6, 2022.
[15] S. Wang, W. Ding, J. Li, Y. Yuan, L. Ouyang, and F.-Y. Wang, “Decen-
21st century, from plant modeling to agricultural management tralized autonomous organizations: Concept, model, and applications,”
and control. The goal of DeCASA is to provide accurate IEEE Transactions on Computational Social Systems, vol. 6, no. 5, pp.
decision support for sustainable agriculture. We presented 870–878, 2019.
[16] K. Bhugaonkar, R. Bhugaonkar, and N. Masne, “The trend of Metaverse
the architectures, operating modes and major applications of and augmented & virtual reality extending to the healthcare system,”
DeCASA in AgriVerse. An Agriculture Foundation Model Cureus, vol. 14, no. 9, p. e29071, 2022.
based on parallel learning and federated intelligence is given [17] Y. Weng, X. Wang, J. Hua, H. Wang, and M. Kang, “Greenhouse envi-
ronment control based on computational experiments,” in Proceedings
as a potential solution to cope with challenging tasks of data of the 2020 3rd International Conference on Big Data and Artificial
analysis and decision support, which paves the way to the Intelligence, ser. HPCCT BDAI ’20. New York, NY, USA: Association
agricultural brain and smart village. for Computing Machinery, 2020, p. 238–245.
[18] J. Hua, M. Kang, and P. de Reffye, “An interactive plant pruning
system based on greenlab model: Implementation and case study,” in
ACKNOWLEDGMENT IEEE International Conference on Computer Science and Automation
Engineering, vol. 4, 2011, Conference Proceedings, pp. 185–188.
This work was supported in part by the National Key [19] Y. Weng, X. Wang, J. Hua, H. Wang, M. Kang, and F.-Y. Wang,
Research and Development Program of China under Grant “Forecasting horticultural products price using arima model and neural
network based on a large-scale data set collected by web crawler,” IEEE
2021ZD0113701, 2021ZD0113704, the National Natural Sci- Transactions on Computational Social Systems, vol. 6, no. 3, pp. 547–
ence Foundation of China under Grant 62076239, Chinese 553, 2019.
2062 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 9, NO. 12, DECEMBER 2022

[20] J. Hua, X. Wang, H. Wang, X.-R. Fan, and M. Kang, “Prediction of [44] N. Trendov, S. Varas, and M. Zeng, Digital Technologies in Agriculture
crop phenology — a component of parallel agriculture management,” in and Rural Areas Status Report. FAO, 2018.
Chinese Automation Congress (CAC), 2017, Conference Proceedings, [45] T. Ayoub Shaikh, T. Rasool, and F. Rasheed Lone, “Towards leveraging
pp. 7704–7708. the role of machine learning and artificial intelligence in precision agri-
[21] X.-R. Fan, M. Kang, E. Heuvelink, P. de Reffye, and B.-G. Hu, culture and smart farming,” Computers and Electronics in Agriculture,
“A knowledge-and-data-driven modeling approach for simulating plant vol. 198, p. 107119, 2022.
growth: A case study on tomato growth,” Ecological Modelling, vol. [46] X. Yang, L. Shu, J. Chen, M. A. Ferrag, J. Wu, E. Nurellari, and
312, pp. 363–373, 2015. K. Huang, “A survey on smart agriculture: Development modes, tech-
[22] M. Fan, M. Kang, X. Wang, J. Hua, C. He, and F.-Y. Wang, “Parallel nologies, and security and privacy challenges,” IEEE/CAA Journal of
crop planning based on price forecast,” International Journal of Intelli- Automatica Sinica, vol. 8, no. 2, pp. 273–302, 2021.
gent Systems, vol. 37, no. 8, pp. 4772–4793, 2022. [47] Z. Shen, A. Elibol, and N. Y. Chong, “Understanding nonverbal commu-
[23] X. Wang, J. Hua, H. Wang, M. Kang, and F.-Y. Wang, “CASA: Complex nication cues of human personality traits in human-robot interaction,”
adaptation system for agriculture based on computational experiments,” IEEE/CAA Journal of Automatica Sinica, vol. 7, no. 6, pp. 1465–1477,
Journal of Intelligent Science and Technology, vol. 1, no. 1, pp. 14–20, 2020.
2021. [48] L. Li, X. Wang, K. Wang, and Y. Lin, “Parallel testing of vehicle
[24] P. de Reffye, B.-G. Hu, M. Kang, V. Letort, and M. Jaeger, “Two decades intelligence via virtual-real interaction,” Science Robotics, vol. 4, no. 28,
of research with the greenlab model in agronomy,” vol. 127, no. 3, pp. p. eaaw4106., 2019.
281–295, 2021. [49] F.-Y. Wang, “Parallel intelligence in Metaverses: Welcome to Hanoi!”
[25] M. Kang, H. Wang, J. Hua, X. Wang, and F.-Y. Wang, “DeCASA: De- IEEE Intelligent Systems, vol. 37, no. 1, pp. 16–20, 2022.
centralized intelligent management and control for agriculture,” Journal
of Intelligent Science and Technology, vol. 2, no. 2, pp. 30–35, 2022.
[26] X. Zhao, H. Wang, X. Wang, U. Lewlomphaisarl, J. Hua, and M. Kang,
“Learning greenhouse climate control policy from monitored data,”
Chinese Automation Congrass (CAC), p. In Press, 2022. Xiujuan Wang received her Ph.D. degree in applied mathematics from
[27] Q. Wei, H. Li, and F.-Y. Wang, “Parallel control for continuous-time Ecole Central Paris, France, and in soil resource use from China Agricultural
linear systems: A case study,” IEEE/CAA Journal of Automatica Sinica, University, Beijing, China, in 2011. She is currently an Associate Professor
vol. 7, no. 4, pp. 919–928, 2020. with the State Key Laboratory of Management and Control for Complex
[28] G. Xiong, X. Dong, H. Lu, and D. Shen, “Research progress of parallel Systems, Chinese Academy of Sciences, Beijing. Her current research interests
control and management,” IEEE/CAA Journal of Automatica Sinica, include plant growth modeling and parallel agriculture.
vol. 7, no. 2, pp. 355–367, 2020.
[29] J. Lu, Q. Wei, and F.-Y. Wang, “Parallel control for optimal tracking
via adaptive dynamic programming,” IEEE/CAA Journal of Automatica
Sinica, vol. 7, no. 6, pp. 1662–1674, 2020.
[30] A. Kamilaris, A. Kartakoullis, and F. X. Prenafeta-Boldú, “A review
Mengzhen Kang (M’16) received her Ph.D. degree in pattern recognition
on the practice of big data analysis in agriculture,” Computers and
and intelligent systems from the Institution of Automation, Chinese Academy
Electronics in Agriculture, vol. 143, pp. 23–37, 2017.
of Sciences, Beijing, China, in 2003. She is an Associate Professor with the
[31] O. Friha, M. A. Ferrag, L. Shu, L. Maglaras, and X. Wang, “Internet of
State Key Laboratory of Management and Control for Complex Systems,
things for the future of smart agriculture: A comprehensive survey of
Chinese Academy of Sciences. Her current research interests include parallel
emerging technologies,” IEEE/CAA Journal of Automatica Sinica, vol. 8,
agriculture and computational plants.
no. 4, pp. 718–752, 2021.
[32] L. Li, Y. Lin, N. Zheng, and F.-Y. Wang, “Parallel learning: a perspective
and a framework,” IEEE/CAA Journal of Automatica Sinica, vol. 4, no. 3,
pp. 389–395, 2017.
[33] L. Li, Y. Lin, D. Cao, N. Zheng, and F.-Y. Wang, “Parallel learning-a
new framework for machine learning,” Acta Automatica Sinica, vol. 45, Hequan Sun received his Ph.D. degree in computer engineering from
no. 1, pp. 1–8, 2017. Nanyang Technological University (Singapore) in 2014. Currently, he is
[34] F.-Y. Wang, R. Qin, J. Li, X. Wang, H. Qi, X. Jia, and B. Hu, “Federated working as a bioinformatician at the Max Planck Institute for Plant Breeding
management: Toward federated services and federated security in fed- Research (Cologne, Germany) and Ludwig Maximilian University of Munich
erated ecology,” IEEE Transactions on Computational Social Systems, (Germany). His research focuses on crop genomics and, more specifically,
vol. 8, no. 6, pp. 1283–1290, 2021. developing computational methods for deciphering crop biology in the context
[35] C. Zhao, Y. Lv, J. Jin, Y. Tian, J. Wang, and F.-Y. Wang, “DeCAST in of next/third generation sequencing and genetics.
transverse for parallel intelligent transportation systems and smart cities:
Three decades and beyonds,” IEEE Intelligent Transportation Systems
Magazine, pp. 2–13, 2022.
[36] J. Mei and H. Zhao, “Incorporating human domain knowledge in 3-D
LiDAR-based semantic segmentation,” IEEE Transactions on Intelligent Philippe de Reffye received his Ph.D. degree in agricultural engineering from
Vehicles, vol. 5, no. 2, pp. 2379–8858, 2020. Université Paris-Sud/Orsay - Paris/France, in 1979. He was a Senior Professor
[37] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, and L. Jones, “Atten- with the Plant Modeling Unit: AMAP Montpellier, CIRAD, France. He was
tion is all you need,” International Conference on Neural Information Head of the GreenLab project in LIAMA, CASIA (1998-2010). He was
Processing Systems (NIPS), pp. 5998–6008, 2017. awarded the outstanding foreign expert by the Chinese government in 2006
[38] P. Liu, Y. Zhou, D. Peng, and D. Wu, “Global-attention-based neural and the Outstanding Contribution Award of Chinese and French Scientists by
networks for vision language intelligence,” IEEE/CAA Journal of Auto- the French government in 2017. His current research interests include plant
matica Sinica, vol. 8, no. 7, pp. 1243–1252, 2021. growth modeling and visualization.
[39] Y. Zhang, B. Xu, and T. Zhao, “Convolutional multi-head self-attention
on memory for aspect sentiment classification,” IEEE/CAA Journal of
Automatica Sinica, vol. 7, no. 4, pp. 1038–1044, 2020.
[40] M. C. Rillig, A. Lehmann, J. Lehmann, T. Camenzind, and C. Rauh,
“Soil biodiversity effects from field to fork,” Trends in Plant Science, Fei-Yue Wang (S’87–M’89–SM’94–F’03) received his Ph.D. degree in
vol. 23, no. 1, pp. 17–24, 2018. computer and systems engineering from Rensselaer Polytechnic Institute,
[41] T. Asaf, D. Medha, K. Brian, A. Shahar, and H. Seán, “Responsible Troy, NY, USA, in 1990. His current research focuses on methods and appli-
artificial intelligence in agriculture requires systemic understanding of cations for parallel systems, social computing, and knowledge automation. He
risks and externalities,” Nature Machine Intelligence, vol. 4, no. 2, 2022. received IEEE ITS Outstanding Application and Research Awards in 2009,
[42] N. Viswanadham and S. Kameshwaran, Smart Villages and Cities. 2011 and 2015, and the IEEE SMC Norbert Wiener Award in 2014.In 2021, he
World Scientific Publishing Co. Pte. Ltd., 2013, pp. 175–192. became the IFAC Pavel J. Nowacki Distinguished Lecturer. He is a fellow of
[43] V. I. Lakshmanan, A. Chockalingam, V. Kumar Murty, and S. Kalyana- the International Council on Systems Engineering; International Federation of
sundaram, Smart Villages: Bridging the Global Urban-Rural Divide. Automatic Control (IFAC); American Society of Mechanical Engineers; and
Springer Nature Switzerland AG, 2020. American Association for the Advancement of Science.