The proposed methodology should be generic, representing a large number of agricultural systems in developing countries.

It should take into account the knowledge integration and sharing. The multiplicity of actors – farmers, decision-makers, researchers - pursuing various objectives and interlinked dynamics require the integration of various types of knowledge – from different actors – for a better understanding of the problem to be solved. The methodology to be developed should be able to provide guidance and tools for knowledge sharing and integration from different stakeholders. In order to achieve this specification, we propose using ontology and an integrated modelling approach. Additionally, we advocate the use of participatory approach.

It should be multi-scale. A multi-scale analysis is important to deal with interactions between human and environmental systems. Reynolds et al. (Reynolds et al. 2007) argued that the involvement of multiple stakeholders, with highly differing objectives and perspectives, illustrates the need to pay attention to the multilevel, nested and networked nature of Human-Environment systems. The present document proposes a methodology for identifying the relevant scales in agricultural system modelling.

It should allow for the multi-formal representation of agricultural systems, as the dynamics of these systems depend on the interactions between social, economic, environmental and policy dynamics. The main question that arises is how to integrate socio-economic and environmental dynamics with policy interventions. Representing these various dimensions requires an integrated model of a different nature, which is where the multi-formalism approach comes in. Multi-formalism simulation enables heterogeneous models of the same system, as perceived at different scales, to be integrated (Duboz et al. 2003; Duboz 2004). This increases model reuse and turns models into collaborative tools for scientists from different disciplines.

It should integrate the heterogeneity of farmers: farmers in the agricultural system are heterogeneous. They differ in terms of their cropping system decisions and characteristics. They also do not respond in the same way to policy decisions. Taking account of this heterogeneity enables us to assess the different impacts that farmers have on the agricultural system and to develop policies for different groups of farmers. In this study, we assume that farmers are heterogeneous and propose a conceptual model to represent this.

Firstly, a literature review has been done in order to answer to the following questions (Table 1) about what are the main methodologies in MAS and ABM?

1. How the different studies on MAS and ABM for agricultural system simulation take the different components of agricultural system?

The research has been done in scholar.google and sciencedirect in order to extract the main papers. The most relevant papers published between 2000 and 2025 have been selected for the study.

For the first criteria, 13 papers have been selected taking account their relevance. These papers have been divided in two groups. The first group concerns methodologies for MAS engineering (Table 2) and the second group concerns methodologies for social sciences simulation (Table 3).

Considering the second criterion, we took into account papers that modelled the features that define the agricultural system (Table 3). Antle et al. (2017) identified the following five components: cropping systems, livestock production, socio-economic dimensions and environmental and policy dimensions. Twenty-four relevant papers were selected for this purpose.

Robinson et Dilkina (2004) identified three representational forms of models. The first form of model is the mental model (MM). They are used by humans on a daily basis in fundamental types of learning and decision-making. The second form of model is the communicative model (CM). It is used as a support for communication and data sharing between different persons involved in the modelling process. They are consensual and explicit, allowing different persons to have the same understanding of the representation of the system. The communicative models are commonly used in software engineering (SE) and in knowledge engineering (KE). SE mainly uses conceptual models in UML to describe the object structures (using class diagram) and the dynamics (using state charts, activity diagrams etc.). A conceptual model provides orientation on how the software should meet a need and specification of the behaviour of the system under construction (Dieste et al. 2001). KE mainly uses logics, frame-like languages and, more recently, the ontologies. An ontology is an explicit specification of the conceptualisation of a domain (Gruber 1995). “An ontology defines the basic terms and relations comprising the vocabulary of a topic area as well as the rules for combining terms and relations to define extensions to the vocabulary” (Neches et al. 1991). The third form of model is the simulation model (SM). An SM is an experimental tool allowing to conduct “experiments with this model for the purpose either of understanding the behaviour of the system or of evaluating various strategies (within the limits imposed by a criterion or a set of criteria) for the operation of the system” (Shannon 1998). Concretely, a simulation model is a software programme simulating the behaviour of a system that it is supposed to represent. It can be viewed “as a representational continuum from the most concrete to the most abstract” (Robinson and Dilkina 2018). This classification clearly draws the process from the idea of the system to model (the mental model) up to its concrete realisation (the simulation model) through successive concretisations of the representations (the more or less detailed communicative models).

In this study, we are concerned by the development of the conceptual model and simulation model. Then, the methodology to be proposed in this study should allow us to develop both a conceptual model and simulation model.

Our general approach is based on agent-based modelling. Our main aim is to propose a unique methodology for agricultural system modelling using agent-based modelling. This methodology takes into account the social, economic, environmental and politic dimensions of agricultural system. In addition, the methodology takes into account the multi-scale representation and agricultural system and the stakeholders involvement.

To support the development of the conceptual model and the simulation model, we use the Unified Modelling Language (UML). “The objective of UML is to provide system architects, software engineers and software developers with tools for analysis, design and implementation of software-based systems as well as for modelling business and similar processes” (OMG 2017). In this study, we use UML for domain analysis, the model design and implementation by making emphasis on stakeholders’ engagement. Specifically, our objective is to use UML as a tool for knowledge sharing and integration between various stakeholders in agricultural system simulation as proposed by Bommel and Müller (2007).

Proposing methodology that allows the design of the conceptual model and multi-agents systems or ABM is not new. These methodologies can be categorised into two (2) sub-groups: the software orientated methodology and social simulations methodologies.

Many methodologies have been proposed in order to deal fully with the MAS development. Amongst them, we have Ingenias (Pavón et al. 2005), Tropos (Bresciani et al. 2004), ADELFE (Bernon et al. n.d.), Message(Evans et al. 2001), Aspects (Cossentino et al. 2007) etc. Each methodology intends to resolve a specific problem of MAS development. For example, ADELFE provides a methodology and a meta-model for the self-adaptative and self-organising systems specification; Mase (DeLoach 2004) is a complete methodology allowing the development of heterogeneous systems; Aspect concerns MAS and Holonic MAS development. The language used by most methodologies is based on the extension of the UML meta-models, such as AUML (Huget et al. 2004), AML (Trencansky and Cervenka 2005), MAS-ML (Da Silva and De Lucena 2003) etc.

These methodologies are not specific to agricultural system modelling and simulation and are difficult for non-computer scientists to use. Most of the time, ABMs for agricultural system simulation are developed by non-computer scientists. Therefore, the methodology used for agricultural system simulation should align closely with the modellers’ perspectives and their conception of the agricultural system.

The second group of methodologies, concern ARDI (Etienne 2009), ORIGAMI (Abrami et al. 2006), ODD (Grimm et al. 2020) and Bommel (Bommel 2009).

ARDI (Etienne 2009) provides a framework for describing the interactions between actors and resources. ARDI is a relevant framework for conceptual model development in the context of knowledge sharing and integration. However, the framework does not take account the simulation phase. Based on the Organisation Centred Multi-Agents System, Abrami et al. (Abrami et al. 2006)proposed the ORIGAMI methodology for modelling the renewables resources management. ORIGAMI is based on AGR meta-model, proposes a set of UML diagrams for describing the different components of a model and finally guidelines for model development and simulation. However, the ORIGAMI meta-model is too general and abstract to be easily used in agricultural system simulation.

Bommel (Bommel 2009) proposed a methodology for ABM of renewable resources management. This methodology can be used in agricultural systems. However, the methodology is specific and does not propose a meta-model to make easier the agricultural system modelling.

As to ODD or ODD+, it provides a standard methodology for ABM modelling and simulation. The objectives of ODD are to make model descriptions more understandable and complete, thereby making ABMs less subject to criticism for being not reproducible (Grimm et al. 2010). Contrary to the ARDI and ORIGAMI, ODD or ODD+ is more and more used in agricultural system simulation (Polhill et al. 2008; Kremmydas et al. 2018; Mora-Herrera et al. 2021). Asssessing papers on Agent Based Modeling for agricultural policy evaluation, Kremmydas et al. (2018) found that the majority of the papers follow the ODD protocol. However, as withother methodologies, ODD is too general to be easily applicable in agricultural system simulation. Using ODD in agricultural system simulation requires an important adaptation.

Although social simulation methodologies are not specific to the agricultural system, they provide a solid foundation for our study. For instance, ARDI enables us to identify the key stakeholders, while ODD provides a foundation for model design. However, the MAASSD methodology is specific to agricultural systems. It considers domain analysis for knowledge extraction and sharing amongst different stakeholders. Finally, unlike existing methodologies, MAASSD proposes a simulation framework and two meta-models specific to agricultural systems.

The methodology is composed by the following steps:

1. Step 1: system requirements aim to understand the objective of the study and to identify the main processes to be taken into account in the modelling process.
2. Step 2: domain analysis aims to understand the main information shared in the system. The main objective is to provide a common understanding of the target system. At this step, we have proposed a meta-model for representing the agricultural system.
3. Step 3: selection of tools: defines how to select appropriate tools according to the system requirements.
4. Step 4: design: the main objective of this step is to propose a methodology to design an ABM for agricultural system simulation.
5. Step 5: implementation: the main objective of this step is to propose a methodology to develop an ABM for agricultural system simulation.

Step 2: Domain analysis

The objective of this step is to identify the main concepts of the target system and the problem to be solved, in order to establish a clear and shared understanding. In terms of the requirements analysis, the domain analysis focuses more on the ‘what’ than the ‘how’ of the system.

This step uses the documentation from existing studies of the target system and existing agricultural system models as inputs to extract relevant concepts. According to the study objective, the main concepts are described by outlining their characteristics and definitions. This can be achieved using a participatory approach to ensure a shared understanding of the problem amongst the different study participants.

In addition, ontologies can also be used at this stage for information extraction. A range of agricultural system ontologies have been developed to date.

For example, in the domain of agricultural domain, Agrovoc (Rajbhandari and Keizer 2012) and Crop-Pest (Rodríguez-García and García-Sánchez 2020) are used for information research and extraction in agricultural domain. Seamless (Athanasiadis et al. 2009) and PLANTS (Consortium 2002) ontologies are used for data sharing amongst the different sub-information systems. AgrODSS (Alharbi et al. 2024), IrrigaOnto (Cornejo et al. 2005), Crop-Pest (Rodríguez-García and García-Sánchez 2020) and AgriNepalData (Pokharel et al. 2014) are developed for information sharing on production techniques and for the improvement of decision making. Beck et al. (2010) used ontology for agricultural system modelling. Based on an ontology-based database and wiki semantic, Belem et al. (2015) proposed an integrated framework for knowledge acquisition, sharing and integration on land use and land management in West Africa.

Once the main concepts of the agricultural domain have been identified, a domain model is developed. This model provides a visual representation of the previously identified and described concepts and their relationships. To achieve this, we propose using a UML class diagram. At this stage, the classes only include attributes, not methods.

We then propose a meta-model that can be extended to develop a specific domain model. The proposed meta-model represents household structuration (Fig. 1), cropping system management strategies (Fig. 2), household occupations (Fig. 3) and, finally, policy intervention (Fig. 4).

Figure 1. 

Meta-model for household structuration.

Figure 2. 

Meta-model for household cropping system strategies’ definition

Figure 3. 

Meta-model for household occupations’ definition.

Figure 4. 

Meta-model for policies’ definition.

Step 4: Model design

The objective of this step is to propose a modelling framework for designing the model. This framework comprises a meta-model (Fig. 5) and a simulation framework (Fig. 6). We recommend using the ODD protocol for the design process.

According to the meta-model, a model is based on interacting households. Each household comprises a set of individuals characterised by their sex and age. Each household belongs to an organisation that provides support through various interventions. Households apply a range of cropping systems that define their cropping management strategies. These strategies are applied at the farm level, which is composed of a set of plots.

The simulation framework defines the dynamics of the model. It comprises a set of possible parameters and three phases: initialisation, simulation and output generation.

Figure 5. 

Meta-model of ABM for agricultural system simulation.

Figure 6. 

Simulation framework of the agricultural system.

Model parameters

The analysis of some dynamic models shows the importance of parameters. For example, ALMASS (Topping et al. 2003) integrates an important number of parameters to specify all types of animal. AnimSim (Parrott and Kok 2002) is a generic model for the representation of organisms of different types of species via the specification of appropriate sets of parameter values. In dynamic models, parameters take different forms. The first form concerns the simple parameters used to describe the evolution like, for example, the population growth rate etc. The second forms of parameters concern complex parameters used to initialise models structure and dynamical behaviour. They concern Geographical Information System (GIS) data used to initialise environment configuration, equations and models to simulate some processes depending on the contexts of application and parameters specifying the model structure and behaviour. For example, in Mirana (Aubert et al. 2010), different parameters types are used to initialise model structure and initial state, environment configuration, agent’s behaviour through the definition of their needs (food, money etc.), objectives (production, conservation etc.), planning mechanism, norms and interactions. The use of various parameters types in Mirana provides a high flexibility to simulate a range of systems and scenarios. In this section, we propose a set possible parameters that one can use. These parameters concern the crop and cropping system description, the socio-economic data and finally the biophysical data (Table 5).

Table 5.

Download as

CSV

XLSX

Possible parameters in agricultural model.

Crop and cropping systems parameters
Types	Description
Crop	Identification and characterisation of major cultivated crops
Cropping system	Identification and description of the major cropping system
Crop rotation	For each cropping system, defines the related crop rotation and fallow management
Socio-economic parameters
Input	Description
Household	Describes the household typology in the farming system (farmer type, family size, farm size, number of external workers, off-farm income, in-farm income, livestock size, family budget, number of granary, number of granary, shed and houses, equipment used for the production, cropping system, ethnic group and participation to different organisations)
Household structure	For each household type, describes the structure of household population (by age group, sex repartition, labour provided, money need, off-farm income)
Food sharing	Define the use of different crops in the feeding and selling (% of each crop in feeding and in selling)
Herd	Describes the type and number of animals (beef, sheep and goat) for each household type
Supports	Describes the different kind of support that each household type beneficiate (aid for cash, aid for food, crop insurance, seed provision etc.)
Crop prices	The crop prices in the different periods
Biophysical data
Input	Description
Climate data

Model initialisation

The model initialisation comprises the environment configuration and the population generation.

Environment configuration

The objective is to create a simulation environment. This environment represents the biophysical condition of the agent, specifically its spatial configuration. Spatially, the environment is represented as a set of plots. Each plot is characterised by a soil type, which defines its properties and a land-use and land-cover type. After defining the spatial configuration, the climate conditions are defined spatially.

This phase uses land-cover data, climate data and soil data as input. These data can be collected from the site under study or from public sources.

Population generation

The objective is to generate households and individuals for simulation. The synthetic population can then be used to generate the population. Individuals are assigned to households according to their type. Crop and animal production systems are then assigned to households according to their initial description. Finally, a farm is assigned to each household.

This step is based on household data, crop management data, animal management data and population scenario data. These data can be extracted from surveys conducted at the study site or from public data sources.

Phase 2: Simulation

The model dynamics comprise the population simulation, the livestock production, the crop production and environment dynamics.

Population simulations

The dynamic of demography is characterised by the natality, ageing and mortality of the individual. The natality depends on the population growth rate and it is defined scholastically. An individual can give birth randomly according to the birth rate. The individual age is incremented yearly. According to the change in people’s age, individual changes age group that determines the probability of mortality. The individual death is determined scholastically. A probabilistic function determines the death of an individual depending on the mortality death rate on the individual age group.

Crop production

Household produces crop according on the food and money need, the available land, the labour force and the need in inputs. Then, household defines the food needs and money according to the household size as follows. Based on the money, the food need and labour force, household determines the area to be cultivated for each crop.

The plots are cultivated according to the crop sequence used by the household. At the end of the cropping season, household harvests crop and stores the crop production. Crop production depends on the crop yield and the cultivated area. Finally, a part of the crop production is used as food and the remainder is sold for income generation.

Environment dynamics

The environment dynamics concern the interactions between soil-plant-atmosphere depending on the objective of the study. These dynamics simulate the impacts of human activities on environment and vice-versa. These dynamics can be represented basically or using external model such as STICS (Beaudoin et al. 2023),Century (Parton 1996) etc.

Phase 3: output generation

A range of data can be generated from the simulation. This output data exist at individual level, household level, climate and the population. (Table 6).

Table 6.

Download as

CSV

XLSX

Output data.

Individual_output	Id, age, sex, migration experience,	Provides the output data on each individual simulated in the model.
household_output	Id, type, crop production, income, family size, cultivate d area, Number of migrants by household	Provides the output data on each household simulated in the model.
Climate	Minimal and maximal temperatures, rainfall	Provides the climate data of the simulated year.

Step 5: Model implementation

We propose the use of rapid prototyping to implement the model. Rapid prototyping is the quick development of a prototype model. A prototype model is a simplified version of the final model. It enables the expected behaviour of the model to be defined and discussed. Prototyping reduces development costs and facilitates discussion between users and developers. In addition, it facilitates the verification of specifications (Hilaire et al. 2000). In simulation, prototyping can be used to communicate which actors (modellers, scientists and stakeholders) are involved in model development and to analyse how different scenarios can be taken into account. To achieve this, the modeller can build a prototype for each scenario to identify the relevant variables, before finally building the final model.

Experimentation

Study sites

Our study covers the country of Burkina Faso. Burkina Faso is a country of migration. The migration in Burkina Faso is international and internal. In this study, we are concerned by the inter-provincial migration. The administrative division of Burkina Faso is composed of 45 provinces shared amongst 13 regions.

From climatic point of view, Burkina Faso is characterised by three climatic-zones (Fig. 7): Sahelian, Sudano-Sahelian and Sudanian zones:

• The Sahelian zone in the north with an average annual rainfall lower than 600 mm, a high rate of evapotranspiration as well as high temperatures and a short rainy season (2 to 3 months). It is the zone with the lowest rainfall in the country;
• The Sudano-Sahelian zone in which the rainfall varies between 600 and 900 mm over 4 to 5 months. It comprises the most extensive climatic zone as it extends over all of the central part of the country. The temperatures recorded are generally mid-range (between 20 and 30 °C).
• The Sudanese zone occupies the southern part of the country, where the rainy season lasts from 5 to 6 months with the level of rainfall attaining or even exceeding 1100 mm annually. This area is marked by low temperature ranges (20–25 °C).

Model design

Our modelling approach aims to represent the feedback loop between agricultural dynamics and migration. In this research, we adopted an agent-based modelling approach to represent household migration decision-making in the context of climate change, land management and crop production decision-making and the impact of these factors on migration decisions, as well as the impact of social networks on individuals’ propensity to migrate. The model’s architecture integrates a climate module, a crop production module, a biophysical model and a data management module.

To provide an explicit representation of the social network, we use the SmallWorld approach to depict the relationships between agents in the model (Fig. 8).

The model is multi-scale as it takes into account articulations between farm, household, individual and local levels (Fig. 8). The individual’s decision to migrate depends on the household’s capacity to satisfy its needs. An individual will migrate if their household cannot satisfy their needs (for food and money) due to the climate of the locality (Fig. 9). In return, the individual contributes to household income and the workforce.

Figure 7. 

Climatic zones of Burkina Faso.

Figure 8. 

Conceptual model.

Figure 9. 

Household and individual dynamics.

Implementation

The model implementation is based on the GAMA platform (Grignard et al. 2013). GAMA is an open-source modelling and simulation environment for creating spatially explicit agent-based simulations. It is widely used in several domains, including urban mobility, climate change adaptation and epidemiology. GAMA is based on GAML, a language specifically designed for modelling and simulation. It enables users to build integrated models incorporating different modelling paradigms, explore and calibrate their parameter space and run virtual experiments with powerful visualisation capabilities. GAMA also integrates functionalities that facilitate the creation of social networks, the import of GIS data and the multi-scale representation of systems.

Simulations Configuration

All the simulations have focused on the Province of Boulkiemdé. It lies between 12°14'59″N, 2°22'03″W. Located in the central part of Burkina Faso, it is bordered by the Provinces of Passoré to the north, Kourwéogo to the north-east, Kadiogo to the east, Bazèga to the south-east, Ziro and Sissili to the south and Sanguié to the west. It covers an area of 4,269 km². The provincial capital, Koudougou, is 100 km from Ouagadougou, the capital of Burkina Faso. Finally, the Province of Boulkiemdé is situated in the Soudano-Sahelian climatic zone.

The simulations involved a population of 400 individuals, shared amongst 40 households. These households are divided into four categories (Table 7): small farmers with lower potential (type 1), small farmers with higher potential (type 2), large farmers with low potential (type 3) and large farmers with higher potential (type 4). The objective of the simulations was to assess the impact of climate shocks on migration and the resilience of different household groups under different climate scenarios. Household data were extracted from the Agriculture Ministry Database. These data concern household type, number, size, cultivated area, most important crops, income and livestock size.

We have identified three climate scenarios, which are based on the likelihood of a climate shock occurring. The first is the wet scenario. We assume that there is a 30% probability of a drought season occurring. The second scenario is moderate, with a 50% probability of a drought season. The third scenario is a drought scenario, with a 70% probability of a drought.

In the model, crop yield depends on season quality, which influences assets and migration decisions.

Each scenario was simulated for 45 years and the results were compared.

Table 7.

Download as

CSV

XLSX

Description of simulated households.

Description	Household type	Household size	Farm size	TLU	Crop rotation
small farmers with lower potential	type1	8	4	6.8	Maize. Milet. Sorghum. Sorghum
small farmers with higher potential	type2	8	4	4.5	Maize. Milet. Sorghum. Sorghum
large farmers with low potential	type3	15	6	4.9	Maize. Sorghum. Sorghum. Sorghum
large farmers with higher potential	type4	16	10	6.2	Maize. Sorghum. Sorghum. Sorghum

The results showed that climate seriously impacts migration decisions. The most significant migration trends were observed in Scenario 3, where drought was a key factor (Fig. 10). The population is therefore likely to migrate when climate conditions are unfavourable. In Scenario 3, the probability of a drought is highest. Under these conditions, households cannot meet their food and income needs and consequently use migration as an adaptation strategy.

Figure 10. 

Migration trends in different climate scenarios.

To assess the strategies of different groups of households under different climate scenarios, we compared the number of migrants from each household type. The results showed that, in all scenarios, types 3 and 4 have the highest number of migrants (see Figs 11, 12), while type 1 had the lowest. The simulations showed that types 3 and 4 were the most vulnerable to climate shock and were most likely to migrate. Additionally, the results showed that type 2 households are the most resilient to climate shock.

Figure 11. 

Impacts of climate scenario 1 on migration of different categories of households.

Figure 12. 

Impacts of climate scenario 3 on migration of different categories of households.

MAASSD methodology pays a particular attention to the stakeholders’ engagement as the stakeholders are engaged in all stages of the methodology. That ensures a smooth model development process: “while the participants may determine the questions that the model should answer and may supply key model parameters and processes, the structure of the model must be scientifically sound and defensible” (Voinov et al. 2016).

Although we propose a methodology for ABM for agricultural system simulation, this study does not take account the scenarios development, simulation and validation phases. In addition, this study did not implement a tool for model development and simulation.

The author has declared that no competing interests exist.

No ethical statement was reported.

AI was used to correct the english.

No funding was reported.

The author solely contributed to this work.

All of the data that support the findings of this study are available in the main text.