the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 27 May 2024
Learning from conceptual models – a study of emergence of cooperation towards resource protection in a social-ecological system
Abstract. Engaging ecological resource users in intervention to protect the resource is challenging for governments due to self-interest of users and uncertainty about intervention consequences. Focusing on a case of forest insect infestations, we addressed questions of resource protection and environmentally responsible behavior promotion with a conceptual model. We coupled a forest infestation model with a social model in which a governing agent applies a mechanism for recognition and promotion of environmentally responsible behavior among several user agents. We ran the coupled model in various scenarios with a Reinforcement Learning algorithm for the governing agent as well as best-case, worst-case, and random baselines. Results showed that a proper recognition policy facilitates emergence of environmentally responsible behavior. However, ecosystem health may deteriorate due to temporal differences between the social and ecological systems. Our work shows it is possible to gain insight about complexities of social-ecological systems with conceptual models through scenario analysis.
- Preprint
(992 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on gmd-2024-57', Anonymous Referee #1, 16 Jul 2024
General comments
The manuscript is very well-structured, presenting a coherent flow from introduction to conclusion. The role of conceptual models in research is effectively discussed and well-situated within the existing literature. The results, discussion and conclusion are well explained and make the gained insights on the social-ecological model clear. The paper provides a relevant contribution to understanding the emergent dynamics in coupled social-ecological systems.
However, there are areas where the manuscript can be improved. The manuscript would benefit from more empirical evidence to justify the choices made in the social model. Furthermore, a clearer and more detailed description of the social model in the methods section would be beneficial for the reader’s understanding. Some modelling details are only mentioned in the discussion section. Including these details in the methods section would provide a comprehensive understanding of the model from the outset and improve the overall clarity of the manuscript.
Specific comments
1. Empirical evidence for modelling choices
- It would be beneficial to include some empirical evidence that granting recognition for promoting environmentally responsible behaviour can be a successful governmental strategy that effectively influences the behaviour of logging companies.
- Other socio-economic factors e.g. environmental awareness of the user agents may be relevant in the considered resource protection problem that are not considered in the presented social model. Therefore, it may be beneficial to explain in the paper why focusing on a simple social model that examines only one aspect (social recognition and norms) is still a powerful tool which helps in understanding the underlying mechanisms.
2. Clearer description of the social model
- Clearly state that “principal agent” and “governing agent” refers to the same entity
- Explain the decision-making process of the user agents, which is based on a simple if-statement without learning capabilities. Clarify how utility is calculated for the user agents, including the calculation of uniqueness and value.
- Describe the economic part of the model in the method section. Explain how the user’s profit is calculated: “In the example of this study, total harvest from the forest comprises market supply, which influences market price and sales quantity, which in turn influence users’ profit and hence individual decisions on harvest in the next time step”
3. Move model description from the discussion section to methods section
Examples:
- Explanation of how the RL algorithm works: “The governing agent’s RL algorithm uses a policy for making decisions, and updates that policy based on the rewards it receives as a result of its decisions”
- Decision-making of user agents
- Real-world context of the ecological model and background information on modeling choices for both the ecological and forestry models.
4. Explain in more detail what happens during preparation time and relate it to the mechanisms of the social model
Discuss why preparation time is beneficial even for random runs where the government is not learning. Can the observation that user agents are more willing to cooperate with the government after a preparation period be explained by the decision-making process of the user agents (uniqueness and value)?
5. A lot of repetition in the results part
Technical corrections
“Figures 5 and 6 show the mean proportions of study area that are covered by healthy and infested forest, as well as the area that is cut, in each time step.” P.11
Only figure 6 shows this, not figure 5.
Citation: https://doi.org/10.5194/gmd-2024-57-RC1 -
AC3: 'Reply on RC1', Saeed Harati-Asl, 31 Jul 2024
These responses were meant to be posted under the reviewer's comments, but were mistakenly posted as an independent message. Please disregard the duplication. Apologies for the inconvenience.
We are thankful to the respectable reviewer for these comments, which helped us improve the manuscript. The revised manuscript will be uploaded upon approval of the discussion by editors. Here are our responses to the reviewer's specific comments:
- On modelling choices
- In this particular context, we did not find works in related literature that address the effect of recognition on emergence of environmentally responsible behavior in logging companies. Our work in this sense is a modest contribution to the field.
- The reviewer raised an important question on the justification for simple models in the study of complex systems. In the case of our social model, our aim was not to produce exact predictions, but rather to gain insight. For us, a simplified and abstract modelling approach was appropriate, as it allowed us to see how model configurations affect its output. This would have been very difficult if we had started our model with multiple degrees of freedom, because complex systems may reach the same state from various paths of change. With the abstract model, we now have an insight about the emergence of our intended behavior. Of course, in reality such emergence may be hindered or hastened by other factors that we have not considered. But still it in insightful to see how the system under study tends to evolve in absence of those other factors.
- On clearer description of the social model
- As the reviewer correctly noted, our model’s “governing agent” is the entity that is referred to as “principal” in “principal-agent problem” literature. We understand that a confusion arose because the term “agent” has different meanings in “agent-based modelling” on the one hand, and in “principal-agent problem” literature. We add a clarification on this matter in the revised manuscript.
- In the first draft of the manuscript we referred the readers to our previous work for details on the decision making of user agents. As per the reviewer’s suggestion, we can add these details in the revised manuscript. As the reviewer correctly noted, these decisions are based on simple if-statements. User agents seek uniqueness and value in their actions. To that end, they assess scores for expected uniqueness and value, then they multiply those scores. Numerical multiplication here performs as the logical “and” operator.
Assessment of uniqueness is based on the immediate past. For example, if no other user showed responsible behavior in the previous time step, a full score of uniqueness is assumed for the respective action in the present time step.
Assessment of value is based on the cumulative past. It represents the total number of times the “responsible user” label has been seen in the society. That is the number of times the label has been awarded to agents.
Each user agent has a numerical threshold that it compares with the product of uniqueness and value in each time step. The result of that comparison defines the user agent’s decision to act. The said threshold varies from one user agent to another, but it is constant for each user agent throughout simulation.
- We suggested the possibility of adding an economic model as an avenue for future work. Presently our model does not have an economic component. We modify the wording of our text in the revised manuscript to avoid confusion.
- On moving parts of text from the discussion section to the methods section
- We thank the reviewer for this comment. Our initial intention was to present background information for discussion, but we find the reviewer’s suggestion improves the flow of the text better. We therefore revise the manuscript accordingly.
- On what happens during preparation time in simulations
- During the preparation time, the social model runs alone, before starting the ecological model. This can represent awareness raising campaigns. During this time, the “responsible user” label is introduced to the society, and user agents get a chance to compete for recognition. Through this competition, the “responsible user” label becomes more visible, and therefore its value increases. After the preparation time, the social and ecological models are coupled together, and the social model keeps its memory of the value of the “responsible user” agent. That is why even in our “random” scenarios, where the governing agent does not learn, there is some action by user agents. This shows the importance of the desire for recognition, which creates a strong potential for emergence of environmentally responsible behavior. Such potential, of course, is not optimally used when the governing agent’s decisions are random.
- On repetitions in the results part
- We appreciate the reviewer’s comment. In the results section we had the challenge of presenting multiple scenarios with similar configurations. In the end, we decided to keep the descriptions of scenarios and results long and complete, to avoid the risk of confusion and misinterpretation.
Technical corrections
- We thank the reviewer for this observation. Wrong figure numbers were mentioned in the text (5 and 6 instead of 6 and 7). Corrections are applied in the revised manuscript.
Citation: https://doi.org/10.5194/gmd-2024-57-AC3
-
RC2: 'Comment on gmd-2024-57', Anonymous Referee #2, 25 Jul 2024
General Comments
The manuscript presents a social-ecological model examining the emergence of cooperation between logging company agents and a governance agent in the context of preventing forest insect infestation. It is well-structured and clearly written, with established theories and literature effectively supporting the conceptual validity of the model. Numerical experiments have been conducted to compare different scenarios under various governance settings. However, several areas need improvement to enhance the manuscript. These include the modelling of governance agents, the presentation of the research's novelty, and improvements to the text and figures.
Specific Comments
- Modelling governance agents:
One of the critical challenges in modelling governance agents is capturing their autonomous, adaptive decision-making processes. While reinforcement learning (RL) is a technically feasible approach, it does not capture the essence of real-world policymaking, as policymakers do not have access to data-intensive training opportunities to optimize their strategies. Therefore, it would be beneficial to include a discussion on how the RL algorithm aligns with or diverges from the nature of real-world policymaking behaviours.
- Presenting novelty:
The novelty of the research is a major concern. The model is based on a set of hypothetical arguments, and while it is not expected to produce quantitatively accurate results for real-world policymaking, it should ideally reveal novel trends or patterns within the social-ecological system. The potential novel patterns would also support the use of the RL algorithm. However, the manuscript does not clearly demonstrate that the model has captured new patterns. The effectiveness of "responsible user labels" is a key factor influencing the efficacy of scenarios with policy interventions (RL and random) compared to those without. If the benefit parameters are fixed, the numerical experiments mainly test the RL algorithm against random interventions, which may not yield novel insights. A stronger presentation of the model's novelty is required.
While one innovation of the paper is the coupling of two models, much of the groundwork appears to have been done in the authors’ previous publications. The current model represents a new application of these two models connected through a simple (but effective) communication protocol. Therefore, a major improvement for this manuscript to address might be clearly articulating the novelty of this research.
- The method section:
The initial paragraph of the Methods section repeats content from the introduction. Additionally, it may not be necessary to argue for the use of modelling approaches to explore policy impacts on complex social-ecological systems. This paragraph could be shortened or removed to improve the overall fluency of the manuscript.
- Flowchart (Figure 3):
The flowchart in Figure 3 could be improved for clarity. Including "Start" and "End" components would help, and the arrows could be less confusing. Typically, if two arrows originate from the same block, that block should be a conditional block with True/False branches. Even though one model always listens to the other for new messages, organizing the flowchart in a sequential order would better reflect the model's logical flow. Alternatively, using different types of arrows to indicate message direction might be beneficial.
Addressing these points will strengthen the manuscript and better communicate its contributions to the field.
Technical corrections:
In Figure 5, the use of black solid and dashed lines as legends is confusing because they do not appear in the figure. It would be more straightforward to present six legends for the six lines, adding only one more legend than currently used.
Citation: https://doi.org/10.5194/gmd-2024-57-RC2 -
AC2: 'Reply on RC2', Saeed Harati-Asl, 31 Jul 2024
We are grateful to the respectable reviewer for these comments, which helped us better present our work. The revised manuscript will be uploaded upon approval of the discussion by editors. Here are our responses to the reviewer's specific comments:
- On modelling governance agents
- The reviewer correctly noted the difference between real-world policy making and our model. Our model does not aim at producing exact replications or predictions of real world. Rather, the objective of our model is to provide insight about a complex systems problem. As the reviewer correctly mentioned, a constraint of real-world governance situations is lack of sufficient data for policy optimization. Repeated trial-and-error experiments are not possible in real-world situations. Exactly to address this constraint, our approach has been to develop a virtual laboratory where we can run experiments without risk of adverse effects. Such abstraction allowed us to overcome real-world constraints and produce datasets that could be analyzed for optimum policy selection. Then, the advantage of reinforcement learning was accessible, as it allows to associate observed rewards with sequences of past actions. In summary, our abstract reinforcement learning model allowed us to run experiments that were otherwise not possible in the real-world.
- On presenting novelty
- We thank the reviewer for this substantial comment. Before discussing the reviewer’s point, we add as a side note that this work and the previous publications describing our social and ecological models comprise the first author’s doctoral research. Indeed, the studies of our previous publications were carried out exactly in order to make the present work possible.
- The reviewer correctly noted that the “responsible user” label is a key factor influencing simulation outcomes. Here we find it necessary to emphasize the distinction between our social model (previous publication) and the present social-ecological model. In our previous publication, which describes the social model, the decision parameters of the user agents were fixed. Specifically, each user agent’s perception of the cost of requested actions was a fixed value throughout the simulation. In contrast, in the present work, user agents’ perceptions of the cost of requested action depend on the volume of infestation in their allocated forest zones, which varies depending on not only ecosystem dynamics, but also depending on actions of users. As infested areas increase, creating buffer zones around them becomes more costly. Subsequently, user agents become less motivated to participate in the management action of creating buffer zones to stop spread of infestations. This is an added layer of complexity that distinguishes the present social-ecological simulations from the social simulations of our previous publication.
- As the reviewer mentioned, our previous publications addressed technical challenges regarding the social and ecological models of our study. The present study involved technical challenges too, such as the coupling of the models and the analysis and interpretation of results. Nevertheless, we would like to emphasize that the present study additionally addressed the challenge of translating a complex situation into a problem definition, and subsequently developing an approach for that problem. This is especially important in the domain of decision support for sustainable development. Without a well-problem formulation and without an understandable approach, decision making in sustainable development will rely on individuals’ subjective perceptions of subject matters, which are often complex and multifaceted. We demonstrated an exercise of formulation of the problem and an approach of dividing the problem into smaller parts, conquering each part through model development, and finally re-integrating the modelled parts. We hope that our work serves decision making by providing a more formal and less subjective ground for developing and discussing ideas.
Consider for example that the scope of decision making about our social-ecological system is to become wider by considering an additional aspect – market and economic complexity. Arguably, this would be a complex case where intuition does not provide clear insight on how the interaction of system components evolves. Respectively, it would be challenging to make decisions about intervention in such a complex system, primarily due to lack of insight. Our modelling approach simply allows to develop an independent market model of timber supply and demand, and subsequently couple that to the existing social-ecological model by adding an expected revenue term to the cost calculations of user agents.
- On the methods section
- We thank the reviewer for this comment. We find the reviewer’s comments improves the text, and we apply it in the revision of the manuscript.
- On Figure 3
- We thank the reviewer for this comment. We use the reviewer’s suggestions to improve our flowchart in the revision of the manuscript.
Technical corrections
We thank the reviewer for this comment. Indeed, our effort to summarize the legend only eliminated one category in six. We find the reviewer’s suggestion improves the figure, and we apply it in the revision of the manuscript.
Citation: https://doi.org/10.5194/gmd-2024-57-AC2 - AC4: 'Reply on RC2 - figures', Saeed Harati-Asl, 05 Sep 2024
-
AC1: 'Reply on RC1', Saeed Harati-Asl, 31 Jul 2024
We are thankful to the respectable reviewer for these comments, which helped us improve the manuscript. The revised manuscript will be uploaded upon approval of the discussion by editors. Here are our responses to the reviewer's specific comments:
- On modelling choices
- In this particular context, we did not find works in related literature that address the effect of recognition on emergence of environmentally responsible behavior in logging companies. Our work in this sense is a modest contribution to the field.
- The reviewer raised an important question on the justification for simple models in the study of complex systems. In the case of our social model, our aim was not to produce exact predictions, but rather to gain insight. For us, a simplified and abstract modelling approach was appropriate, as it allowed us to see how model configurations affect its output. This would have been very difficult if we had started our model with multiple degrees of freedom, because complex systems may reach the same state from various paths of change. With the abstract model, we now have an insight about the emergence of our intended behavior. Of course, in reality such emergence may be hindered or hastened by other factors that we have not considered. But still it in insightful to see how the system under study tends to evolve in absence of those other factors.
- On clearer description of the social model
- As the reviewer correctly noted, our model’s “governing agent” is the entity that is referred to as “principal” in “principal-agent problem” literature. We understand that a confusion arose because the term “agent” has different meanings in “agent-based modelling” on the one hand, and in “principal-agent problem” literature. We add a clarification on this matter in the revised manuscript.
- In the first draft of the manuscript we referred the readers to our previous work for details on the decision making of user agents. As per the reviewer’s suggestion, we can add these details in the revised manuscript. As the reviewer correctly noted, these decisions are based on simple if-statements. User agents seek uniqueness and value in their actions. To that end, they assess scores for expected uniqueness and value, then they multiply those scores. Numerical multiplication here performs as the logical “and” operator.
Assessment of uniqueness is based on the immediate past. For example, if no other user showed responsible behavior in the previous time step, a full score of uniqueness is assumed for the respective action in the present time step.
Assessment of value is based on the cumulative past. It represents the total number of times the “responsible user” label has been seen in the society. That is the number of times the label has been awarded to agents.
Each user agent has a numerical threshold that it compares with the product of uniqueness and value in each time step. The result of that comparison defines the user agent’s decision to act. The said threshold varies from one user agent to another, but it is constant for each user agent throughout simulation.
- We suggested the possibility of adding an economic model as an avenue for future work. Presently our model does not have an economic component. We modify the wording of our text in the revised manuscript to avoid confusion.
- On moving parts of text from the discussion section to the methods section
- We thank the reviewer for this comment. Our initial intention was to present background information for discussion, but we find the reviewer’s suggestion improves the flow of the text better. We therefore revise the manuscript accordingly.
- On what happens during preparation time in simulations
- During the preparation time, the social model runs alone, before starting the ecological model. This can represent awareness raising campaigns. During this time, the “responsible user” label is introduced to the society, and user agents get a chance to compete for recognition. Through this competition, the “responsible user” label becomes more visible, and therefore its value increases. After the preparation time, the social and ecological models are coupled together, and the social model keeps its memory of the value of the “responsible user” agent. That is why even in our “random” scenarios, where the governing agent does not learn, there is some action by user agents. This shows the importance of the desire for recognition, which creates a strong potential for emergence of environmentally responsible behavior. Such potential, of course, is not optimally used when the governing agent’s decisions are random.
- On repetitions in the results part
- We appreciate the reviewer’s comment. In the results section we had the challenge of presenting multiple scenarios with similar configurations. In the end, we decided to keep the descriptions of scenarios and results long and complete, to avoid the risk of confusion and misinterpretation.
Technical corrections
- We thank the reviewer for this observation. Wrong figure numbers were mentioned in the text (5 and 6 instead of 6 and 7). Corrections are applied in the revised manuscript.
Citation: https://doi.org/10.5194/gmd-2024-57-AC1
Data sets
Datasets for model flipflopSEM Saeed Harati-Asl, Liliana Perez, and Roberto Molowny-Horas https://doi.org/10.17605/OSF.IO/URJQ8
Model code and software
flipflopSEM Saeed Harati-Asl, Liliana Perez, and Roberto Molowny-Horas https://doi.org/10.5281/zenodo.11245520
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 278 | 88 | 131 | 497 | 9 | 9 |
- HTML: 278
- PDF: 88
- XML: 131
- Total: 497
- BibTeX: 9
- EndNote: 9
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1