# An example of management conflict using GMSE

Agents in GMSE (managers and users) are goal-oriented, and their behaviour is therefore driven to maximise a particular utility of interest such as a target density of resources. For managers and users, this could include animals or trees of conservation interest. For users, it could additionally include a landscape harvest size such as bag size or timber. This model feature allows GMSE to evaluate the actions of agents in the context of their individual objectives, and to therefore quantify the degree to which those objectives are or are not achieved. When the actions of one party clashes with the objectives of another party, the objectives of one might be expressed at the expense of the other, causing conservation conflict (S. M. Redpath et al. 2013). Currently, there is no standard way to measure conservation conflict in a social-ecological system where both the natural resource (e.g., animals, plants, or non-biological resources) and the people (e.g., stakeholders, managers, etc.) are modelled in a single system, and previous modelling approaches have not meaningfully separated agent objectives from agent actions. We suggest that a starting point to developing a useful metric of conservation conflict is to quantify the deviation of an individual’s actions from their objectives (i.e., of actual actions from desired actions), the former of which is restricted by the actions of other individuals. Here we show how GMSE can be used to evaluate the amount of conflict in a simulated social-ecological system under different management options.

To demonstrate how GMSE can be used to understand conflict in social-ecological systems, we build upon the example of resource management in the main text. We consider a protected population of waterfowl that exploits and damages agricultural land and is therefore a source of conservation conflict between those that seek the conservation of waterfowl and those that are concerned with the loss of agriculture (e.g., Fox and Madsen 2017; T. H. Mason et al. 2017; A. I. T. Tulloch, Nicol, and Bunnefeld 2017; Cusack et al. 2018). As in the main text example, the objective of the manager is to keep waterfowl at a target abundance that minimises extinction risks, while the objective of farmers is to maximise agricultural production on their landscape. Here we consider a more complex simulation, with a level of detail that more accurately reflects a scenario that might occur in a real social-ecological system where the manager sets policies that incentivise the user to act in a way that ensures the persistence of the resource. The policies are backed up by a manager budget that can be allocated to set costs of actions and thus incentivise users to perform the desired actions. The user also has a budget to carry out actions, and the cost of the user actions is affected by the manager policies and budget. Our objective is not to model the dynamics of a specific system, but to show how GMSE could be parameterised using demographic estimates from empirical studies. We therefore consider an example population in which such estimates are well-reported and readily available.

We parameterise our model using demographic information from the Taiga Bean Goose (Anser fabalis fabalis), a managed population that is hunted for recreation in Fennoscandinavia (Johnson et al. 2018). Taiga Bean Geese can cause agricultural damage (Johnson et al. 2018), which could potentially lead to conflict between farming and management or conservation objectives.

## Using demographic parameters in simulations

Our goal is not to provide a detailed case study of the Taiga Bean Geese, but rather to demonstrate how such a case study would be possible in GMSE. For simplicity, here we assess conflict using only the gmse function to show how parameter values can be set to provide useful results. Novice R users may prefer to run all of the simulations below using the browser-based GMSE GUI by calling gmse_gui() from the R command line. Alternatively, experienced R users may prefer to simulate by looping time steps through gmse_apply, which allows more flexibility for incorporating custom sub-models and dynamically adjusting parameter values.

Simulations using the default GMSE sub-models described above are run using the gmse function, which offers a range of options for setting parameter values (see Table 1 for some select examples). Output of gmse is an exhaustive list that includes all resources and observations, all stakeholder decisions and actions, and all landscape properties in each time step of the simulation. Results are most easily interpreted visually, so a summary of simulation dynamics is plotted by default (the plot can also be called using the plot_gmse_results function, and summaries of results can be obtained using gmse_summary and gmse_table). An example below shows how simulations are set and interpreted.

Argument Default Description
time_max 100 Maximum time steps in simulation
land_dim_1 100 Width of the landscape (horizontal cells)
land_dim_2 100 Height of the landscape (vertical cells)
res_movement 20 Distance (cells) a resource can move in any direction (for movement rules, see res_move_type)
remove_pr 0 Density-independent probability of resource mortality during a time step
lambda 0.3 Poisson rate parameter for resource offspring number produced during a time step
agent_view 10 How far managers can see on the landscape for resource counting when observe_type = 0
res_birth_K 100000 Carrying capacity applied to the number of resources added during a time step
res_death_K 2000 Carrying capacity applied to the number of resources removed during a time step
res_move_type 1 Type of resource movement (default is up to res_movement cells in any direction)
observe_type 0 Type of resource observation (default is density-based; i.e, counting a subset on the landscape)
fixed_mark 50 For mark-recapture observation (observe_type = 1), number of marked resources
fixed_recapt 150 For mark-recapture observation (observe_type = 1), number of recaptured resources
times_observe 1 For density-based observation (observe_type = 0), landscape subsets viewed during observation
res_consume 0.5 Pr. of a landscape cell’s value reduced by the presence of a resource in a time step
max_ages 5 The maximum number of time steps a resource can persist before it is removed
minimum_cost 10 The minimum cost of a user performing any action
user_budget 1000 A user’s budget per time step for performing any number of actions
manager_budget 1000 A manager’s budget per time step for setting policy
manage_target 1000 The manager’s target resource abundance
RESOURCE_ini 1000 The initial abundance of resources
scaring FALSE Resource scaring (moves a resource to a random landscape cell) is a policy option
culling TRUE Resource culling (removes a resource entirely) is a policy option
castration FALSE Resource castration (sets a resource’s lambda to zero) is a policy option
feeding FALSE Resource feeding (increases a resource’s lambda) is a policy option
help_offspring FALSE Resource helping (increases a resource’s offspring number) is a policy option
tend_crops FALSE Users can increase landscape cell values
tend_crop_yld 0.2 Proportional increase per landscape cell from tend_crops action
kill_crops FALSE Users can decrease landscape cell values to zero
stakeholders 4 Number of users in the simulation
land_ownership FALSE Users own land and increase utility indirectly from landscape instead of resource use
manage_freq 1 Frequency (in time steps) with which managers revise and enact policy
public_land 0 Proportion of land that is public (un-owned by users) if land_ownership = TRUE

Table 1: Select parameter values for initialising generalised management strategy evaluation simulations

Where available, we use estimated demographic parameter values from Johnson et al. (2018) and AEWA (2016). Where GMSE parameter values are not available, we use reasonable values or GMSE defaults. To make model inferences for real case studies, we strongly recommend replicating simulations and simulating across a range of parameter values when empirical estimates are unavailable, as social-ecological dynamics might be sensitive to these unknown values.

Johnson et al. (2018) recently estimated key demographic parameters of the Taiga Bean Geese from the Central Management Unit, which includes geese that breed in “Northern most Sweden, Northern Norway, Northern and Central Finland, and adjacent North-western parts of Russia, wintering mostly in Southern Sweden and South-east Denmark” (AEWA 2016). They estimated goose survival under ideal conditions to be ca 0.878; this can be interpreted in our model by setting mortality to remove_pr = 1 - 0.878. Similarly, Johnson et al. (2018) estimated mean reproductive rate and carrying capacity to be 0.55 and 93870, respectively, so we set lambda = 0.275 (for simplicity, we simply use half the mean reproductive rate; GMSE does not currently distinguish female and male individuals) and res_death_K = 93870. The global abundance of Taiga Bean Geese in 2009 was ca 63000 (Fox et al. 2010), with ca 35000 in the Central Management Unit (AEWA 2016), which we can take as a starting abundance for our simulations (RESOURCE_ini = 35000). The International Single Species Action Plan has a target population size of ca 70000 in the Central Management Unit (AEWA 2016), which we can use as a management target (manage_target = 70000). We simulate social-ecological dynamics over 30 time steps, which could be interpreted as years.

The code below calls gmse using the empirically derived parameters for Taiga Bean Geese described above. We also set manager_budget = 10000 and user_budget = 10000. Further, we consider the case of a region in which farmland makes up 60% of all land, with 40% of land being ‘public’ (public_land = 0.4; which might be interpreted as any land in which stakeholders are not, or cannot be, invested in goose presence), and divide the farmland amongst 80 individual farmers (stakeholders = 80; land_ownership = TRUE). Landscape size is set to default 100 by 100 cells, so each farmer effectively owns 75 cells, which might be interpreted as hectares of land (for instructions on how to more precisely control landscape ownership, see the advanced GMSE options using gmse_apply). Because we need both density-dependent (res_death_K = 93870) and density-independent (remove_pr = 0.122) sources of mortality, we set res_death_type = 3. We assume that a single goose decreases agricultural production on a cell by 2% per time step (res_consume = 0.02). We further assume that the population is very well-monitored, with observers counting goose numbers on each cell of the landscape in every timestep (observe_type = 3) with the ability to observe one landscape cell in every direction (agent_view = 1). All other parameter values are set to GMSE defaults.

## Simulating goose management

Below, we first only allow culling as a policy option and plot the dynamics of the social-ecological system from a single simulation. Next, we run the same simulation but also allow scaring as a policy option; we then use the model to make inferences regarding how scaring as a management option might affect goose population dynamics, agricultural production, and conservation conflict in the system. We emphasise that the simulations below are intended only to demonstrate one use of GMSE on a species of conservation interest, not to make recommendations for management of Taiga Bean Geese.

sim_1 <- gmse(manager_budget = 10000, user_budget = 10000, res_death_K = 93870,
manage_target = 70000, RESOURCE_ini = 35000, plotting = FALSE,
stakeholders = 80, land_ownership = TRUE, public_land = 0.4,
scaring = FALSE, lambda = 0.275, remove_pr = 0.122, time_max = 30,
res_death_type = 3, res_consume = 0.02, res_birth_K = 200000,
observe_type = 3, agent_view = 1, converge_crit = 0.01,
ga_mingen = 200);

The results of the above simulation are plotted in Figure 1 below.

plot_gmse_results(res = sim_1$resource, obs = sim_1$observation,
land = sim_1$land, agents = sim_1$agents, paras = sim_1$paras, ACTION = sim_1$action, COST = sim_1$cost); Figure 1 shows the dynamics of goose abundance and agricultural yield, along with how managers react to change in abundance and farmers react to manager policy. In the case of the simulation above, managers quickly set a policy of high cost for culling, which leads to a rise in the goose population and a decrease in crop yield for farmers. After roughly 20 time steps, the goose population rises above the manager target, at which point the manager becomes more permissive of culling and the cost of culling for users therefore declines. In response, users begin to cull geese on their land, and the goose population begins to stabilise around the target of 7000 total geese. We can investigate the conflict between management policy and farmers more directly using the plot_gmse_effort function (Figure 2). plot_gmse_effort(sim_1$agents, sim_1$paras, ACTION = sim_1$action,
COST = sim_1$cost); Black lines in Figure 2 indicate how permissive a manager is toward a particular action on a scale of 0 to 100, while coloured lines indicate how much effort farmers expend on a given action. When black lines are far below coloured lines, we can (cautiously) interpret this as a conflict between management of the goose population and farmer’s interest in agricultural production. These time periods represent instances in which the manager is not permissive of a particular action (in this case culling), but farmers continue to expend effort to do the action anyway. In the case of the above simulation of potential conflict between farmers and goose conservation, conflict is highest before time step 20, where the manager is not permissive of culling because the population is below the manager’s target. Once the goose population has increased above the manager’s target, conflict decreases because the desired culling is permitted by managers to keep the population at a target abundance. It is worth noting that, despite conflict as we define it decreasing, agricultural damage is still relatively high after the target goose population size is achieved (Figure 1). Hence, on a broader scale, conflict might persist around the appropriate target population size rather than what actions are permitted for farmers; currently, this potential aspect of conflict is not modelled, but future versions of GMSE may attempt to incorporate such additional complexity in conflict scenarios. We can model the consequences for goose population dynamics, agricultural production, and conservation conflict when scaring is a policy option available to the manager. The code below runs a simulation identical to the one just discussed, but with a scaring option included using the argument scaring = TRUE. sim_2 <- gmse(manager_budget = 10000, user_budget = 10000, res_death_K = 93870, manage_target = 70000, RESOURCE_ini = 35000, plotting = FALSE, stakeholders = 80, land_ownership = TRUE, public_land = 0.4, scaring = TRUE, lambda = 0.275, remove_pr = 0.122, time_max = 30, res_death_type = 3, res_consume = 0.02, res_birth_K = 200000, observe_type = 3, agent_view = 1, converge_crit = 0.01, ga_mingen = 200); The results are plotted in Figure 3. plot_gmse_results(res = sim_2$resource, obs = sim_2$observation, land = sim_2$land, agents = sim_2$agents, paras = sim_2$paras,
ACTION = sim_2$action, COST = sim_2$cost);

When scaring is introduced to an otherwise identical simulation (compare Figure 3 to Figure 1), the goose population increases as before, but it achieves the manager’s target population size and stabilises 2-3 time steps earlier. The reason for this earlier stabilisation is due to the change in farmer’s actions as a consequence of the introduced scaring policy. At the start of the simulation, managers adjust policy by quickly increasing the cost of culling and decreasing the cost of scaring. In response, farmers turn to scaring rather than culling to remove geese from their land cells (Figure 3). This is in contrast to the simulation in which scaring was not an option, and farmers simply culled as much as possible despite the high costs (Figure 1). After the population has risen to slightly above the manager’s target, the cost of culling again decreases, with the manager balancing the incentivisation of culling and scaring. The consequence of scaring as an available policy also reveals some potentially unexpected outcomes, such as increased variance in among-farmer agricultural production, which arises as geese are scared from one area of the landscape to another.

We can use the plot_gmse_effort function as before to investigate how the inclusion of scaring as a policy option might affect conservation conflict. Conflict results when scaring is included are plotted in Figure 4.

plot_gmse_effort(sim_2$agents, sim_2$paras, ACTION = sim_2$action, COST = sim_2$cost);

Unlike the case in which culling was the only policy option (compare Figure 4 to Figure 2), the effort that farmers expended on a given action did not rise as highly above the manager’s permissiveness of the action. Hence, under the conditions of this model, the inclusion of scaring as a policy option has reduced conservation conflict in the social-ecological system. We again emphasise that the simulations presented here only serve as an example for how GMSE could be used as a tool for simulating social-ecological systems and understanding the potential for conflict; it is not intended to inform policy in Taiga Bean Geese or any other specific system.

## Conclusions and future development

The GMSE function gmse and its graphical user interface counterpart gmse_gui offer wide a suite of options for parameterising simulations to fit empirical case studies of conservation interest using default GMSE natural resource, observation, manager, and user sub-models. Future development of these sub-models might usefully incorporate additional details relevant to specific case studies, such population structure, multiple natural resource and user types, or different manager policy and user action possibilities. Requests for new features and contributions to GMSE can be made through GitHub. Additionally, where entirely different types of sub-models are required, the gmse_apply function can be used to more flexibly simulate social-ecological systems. In SI4, we show an example of this using the same Taiga Bean Geese case study that we did here.

AEWA. 2016. “International single species action plan for the conservation of the Taiga Bean Goose (Anser fabalis fabalis).” http://www.unep-aewa.org/sites/default/files/document/aewa{\_}egmp{\_}inf{\_}2{\_}issap{\_}tbg.pdf.

Cusack, Jeremy J, A Bradley Duthie, Sarobidy Rakotonarivo, Rocío A Pozo, Tom H E Mason, Johan Månsson, Lovisa Nilsson, et al. 2018. “Time series analysis reveals synchrony and asynchrony between conflict management effort and increasing large grazing bird populations in northern Europe.” Conservation Letters, e12450. doi:10.1111/conl.12450.

Fox, Anthony D, and Jesper Madsen. 2017. “Threatened species to super-abundance: The unexpected international implications of successful goose conservation.” Ambio 46 (s2). Springer Netherlands: 179–87. doi:10.1007/s13280-016-0878-2.

Fox, Anthony D, Bart S Ebbinge, Carl Mitchell, Thomas Heinicke, Tomas Aarvak, Kendrew Colhoun, Preben Clausen, et al. 2010. “Current estimates of goose population sizes in western Europe, a gap analysis and an assessment of trends.” Ornis Svecica 20 (3-4): 115–27. doi:10.1111/j.1365-2486.2005.00941.x.

Johnson, Fred A, Mikko Alhainen, Anthony D Fox, Jesper Madsen, and Matthieu Guillemain. 2018. “Making do with less: Must sparse data preclude informed harvest strategies for European waterbirds.” Ecological Applications 28 (2): 427–41. doi:10.1002/eap.1659.

Mason, Tom H.E, Aidan Keane, Stephen M Redpath, and Nils Bunnefeld. 2017. “The changing environment of conservation conflict: geese and farming in Scotland.” Journal of Applied Ecology, 1–12. doi:10.1111/1365-2664.12969.

Redpath, Steve M, Juliette Young, Anna Evely, William M Adams, William J Sutherland, Andrew Whitehouse, Arjun Amar, et al. 2013. “Understanding and managing conservation conflicts.” Trends in Ecology & Evolution 28 (2): 100–109. doi:10.1016/j.tree.2012.08.021.

Tulloch, Ayesha I T, Sam Nicol, and Nils Bunnefeld. 2017. “Quantifying the expected value of uncertain management choices for over-abundant Greylag Geese.” Biological Conservation 214. Elsevier: 147–55. doi:10.1016/j.biocon.2017.08.013.