Multi-agent modeling to understand the impact of stakeholders decisions in a residential land development project in Southern Alberta

Transcription

1 Multi-agent modeling to understand the impact of stakeholders decisions in a residential land development project in Southern Alberta Michael Kieser University of Calgary mejkiese@ucalgary.ca Dr. Danielle Marceau University of Calgary dmarceau@ucalgary.ca Abstract The objective of this project is to build a multiagent system to simulate the decision making process of stakeholders in a residential land development project and the influence of such decisions on landuse resources. The environment over which decisions are made is Strathmore, Alberta, where competition for land-use resources is increasing as a result of its proximity to the City of Calgary. The stakeholders simulated as agents are the citizens, the town planner and the developer. Interviews were conducted with representatives of each group to gather information about their goals, decision making process and influence. Simulations were performed over 30 years with a one year interval to mimic different land development scenarios. The model generates a series of land-use maps showing the changes in the environment based on the goals and decisions of agents. Conceptual and operational validation is currently done with experts. When fully tested, this model will represent a valuable tool to better understand the complex interactions among several stakeholders involved in the land redesignation process and forecast the cumulative impact of their decisions on the environment. 1. Introduction A large scale land development project requires the planning of the physical and legal changes of the land. The physical changes include the site infrastructure and the environmental protection. In Alberta, the legal changes include the land subdivision, an area structure plan, and the land-use redesignation; the last two working hand in hand. The role of a land-use designation (zoning) on a parcel of land is to identify the legal control on the use and intensity of development permitted on the land. Prior to the development of a parcel of land, the land-use designation must be legally changed to allow for the type of proposed development; i.e. if a residential development is proposed, the land use must be redesignated (changed) to residential. The formal process for creating an area structure plan and a land-use redesignation involves pre-application meetings between the developer, the community and the municipal staff. Open houses are organized where the developer presents the proposed land development and the public has the opportunity to review and comment the proposed changes. The developer can then revise the proposed development before submitting his application to the municipality. The application is circulated to the community and the planner reviews the community comments and makes a recommendation to the municipal council. Finally, a public hearing is held where the municipal council makes its decision on the proposed land development. Several stakeholders are involved in the area structure plan and the land-use redesignation processes including but not limited to: the land owners, the land developers, the citizens, the municipal planners, the utility companies, and the environmental organizations. These stakeholders make decisions by weighing the social need, environmental impact, economic advantages or disadvantages and political support or opposition of a proposed land development. The municipal planner also attempts to fit a proposed land development into future plans that have been developed for the community. Each stakeholder has different opinions on the development of a land parcel and different goals that he attempts to meet. There is a structure of formal and informal communication between the stakeholders, and the outcome of this complex system is a decision which changes the land-use designation of the parcel to be developed. Municipal planners use various tools including forecasting based on present conditions and past tendencies, statistical analyses of census data, and community economic models. They also utilize historical data and past experiences to make decisions that best fit /09 $ IEEE 1

2 the ideals and vision of the community in terms of future social, economic and environmental sustainability. However, decisions are still quite often made in the face of uncertainty. The rapidly expanding Calgary region in Southern Alberta is a perfect example of such a situation. The central issue addressed here is that planners do not have the ability to forecast the cumulative effect of many individual decisions made by stakeholders on the environment over which they make their decisions. They need a tool that can model how final environmental patterns and trends emerge from the cumulative effect of complex behaviours and interactions of several individual stakeholders who might have conflicting views. Having access to such a tool would enable environmental impact forecasting of current goals, decisions and policies, and would allow stakeholders to perhaps modify their goals and analyze possible future impact before the implementation of their decisions. Multi-agent simulation is a tool that could be used by municipal planners in their decision-making process to forecast the impact of their goals and decisions. Multi-agent systems are simulation models which originate from the field of artificial intelligence [1]. A multi-agent system simulates a community of agents making decisions concerning an environment and taking actions on that environment. Agents have the following properties: they are autonomous, they control their own decisions and actions, they are social, they use a language to negotiate and cooperate with one another; they are reactive, they are able to perceive changes in the environment and react to them; they are also proactive, they have goals and are able to take initiative to achieve them [2]. Agent-based models have been developed by a number of researchers for several purposes including: wildlife management [3][4], natural resource management [5][6], urban simulation [7], spatial planning [8], and land-use and land cover change [9]. These models are built from a bottom up approach, explicitly reproducing the actions of individual agents in attempts to understand how the properties of the system emerge from individuals perceiving and acting upon their environment. They are able to realistically reproduce the non-linear and selforganizing properties of many interacting components creating recognizable dynamic spatial and temporal organizational patterns [1]. 2. Methodology The environment over which the stakeholders make decisions is the Town of Strathmore, located at about 50 km east of Calgary in Alberta, Canada, where competition for land-use resources is increasing as the City of Calgary rapidly grows. The study focuses on the land development project called Strathbury The Environment The environment is represented by a series of rasterbased maps converted from vector data provided by the Town of Strathmore and the Province of Alberta and including: the land-use designation for the Town, the cadastral base that determines the boundaries of the legal parcels and the location of the wetlands. A sequencing attribute for each developable cadastral parcel was defined from the cadastral and land-use map. This sequencing is a function of the land-use designation, the size of the parcel and the distance to the Town center. A spatial resolution of 4 m was chosen to accommodate the narrowest strip of landuse and to minimize the model run-time. A GIS (Geographic information system) database was developed to integrate the relevant information about the study area, to calibrate the model, and to generate the initial state of the environment from which the model simulations begin. One environmental issue that was taken into account is the presence of wetlands in the area targeted for residential development The Stakeholders The stakeholders are those directly involved in the area structure plan and land-use redesignation processes for the Strathbury project in the Town of Strathmore and include the town planner (also representing the town council), the land developer and the citizens. Each of these stakeholders has his own specific goals that he attempts to fulfill. Structured interviews with representatives of each agent type were performed and the information was compiled to determine their goals, how they make decisions, the factors that influence their decisions, and how they communicate with each other. The individuals that were interviewed include: the town planner, a representative from the land development company who has many years of experience and is the developer of the Strathbury project, and two representatives from the community who are highly involved in the land-use redesignation process. A professional engineer who is considered an expert in the field of land development is also collaborating with the project. 2

3 A series of questions were developed and sent to the interviewees prior to the interview. Examples of questions are: Can you describe the initial goals of your organization regarding the land development? Are you pleased with the land-use redesignation process? Is there anything you would like to change about the proposed land development? During the interview process it was discovered that although the questions provided a general understanding of the factors influencing the decision process of the stakeholders, the depth was insufficient; therefore a more unstructured interview ensued. The developer explained the general goals that his company attempts to meet with all proposed land developments, including profit, density, construction cost, development timeline and lot retail value. He also discussed specific details that pertain to the Strathbury development of which wetlands were the most controversial. The Town planner described the Towns zoning bylaws, the municipal development plan, the density goals, the growth plan, the trail network system, the wetland policy, which was recently updated following a public survey of the towns residents, and the planners role as a sounding board to residents concerns and as an advisor to the Town Council. He also presented some details on the Strathbury development project and the issue of wetlands. The citizens presented their general concerns regarding the Towns growth, the typical not in my back yard apprehension, the park system and the preservation of wetlands. The information collected from the interviews was compiled to determine the stakeholders goals and interactions, the factors influencing their decisions, and their decision making process Land developer: goals, decision and influence. The information compiled from the interview with the land developer representative revealed that his general goals are to meet the market demand and follow the town plan, and to increase the number of sellable lots to make profit. From this information, six specific parameters were abstracted in the model: 1) profit, 2) increased residential density, 3) development timeline, 4) park dedication, 5) willingness to move wetlands, and 6) consistency with municipal development plan. Several factors influence the decisions of the land developer about a proposed development: existing land-use, market demand, marketable value of developable lots, construction cost, cost of land parcel, environmental and land-use bylaw regulations, distance to existing infrastructure, distance to town centre, and presence of wetlands within the land parcel. When making the decisions on a proposed development, the developer looks at different development schemes, assesses all the influencing factors and calculates the most suitable and profitable scheme. If he performs his duediligence, his proposed development plan for a parcel of land should be accepted. Typically a development proposal submitted to the town planner contains a report, several plans and possibly other documents. In our model, a sequence of numbers was developed to translate the content of those documents submitted as a proposed development into a tuple. The tuple identifies the proposed cadastral parcel, the proposed density, the timeline for the land development project, the percentage of each land-use type, the depth of residential units, the number of units of each residential type, the wetlands within the development area, and the wetlands proposed to be moved in the proposed development project Citizens: goals, decision and influence. The information compiled from the interviews with the citizen representatives has revealed that they like their town being small and want to maintain its size, they do not want the urban sprawl of Calgary, they like the network of walking trails within their community, and they do not want to disturb any wetlands. From this gathered information, four parameters were extracted for the model: 1) concern with wetland disturbance, and the increase/maintain/decrease of 2) park land, 3) density and 4) building side-yard setback. The last three parameters are all associated to minimizing urban sprawl. The citizens decisions are greatly influenced by the impact on wetlands. They evaluate the development proposal created by the developer in terms of the impact on wetlands and share their positive or negative opinion with the town planner Planner: goals, decision and influence. The information compiled from the interview with the planner representative shows that he wishes to achieve the following general goals: to implement a new community growth strategy (municipal development plan) with transit-oriented design and resulting new developments, and to solve storm and sewer infrastructure problems. The planner representative is acting as the moderator between the land developer and the citizens over the wetland issues; he attempts to reconcile their goals while 3

4 meeting the needs of the growing community. From this information, seven parameters were extracted and included in the model: 1) development approvals per year, 2) weight of citizens opinion, 3) consistency with the towns municipal development plan, 4) concern with wetland disturbance, and the increase/maintain/decrease of 5) park land, 6) density, and 7) building side-yard setback. Several factors influence the decisions of the planner including: the citizens involvement, the existing land use, the potential impact on wetlands, the environmental and land-use bylaw regulations, the municipal development plan, the housing demand, the urban development potential, and the population growth. A key decision that must be made by the planner (town council) is related to the sharing of decision making power with the citizens: the greater is the involvement of the citizens, the more decision making power they are given. The planner must evaluate the development proposal created by the developer in relation to his goal and the existing regulations. Then, he makes a decision that accommodates the opinion of the citizens, the housing demand and the towns municipal development plan, and the right that a land owner, represented by the land developer, has to develop his property. environmental and land-use bylaw regulations, the market demand, and the municipal development plan. Variables include: the population growth, the housing demand, the development potential, the construction cost, and the land market value. Figure 1 shows the flowchart of the model. The initialization of the model is done from the land-use map. The sequencing for the next parcel to be developed has been pre-determined based on the distance to existing infrastructure and to the Town centre. Additional social factors that might influence the decisions of citizens and planners such as the NIMBY (not in my backyard) syndrome, lowincome housing, and senior housing have not been included in the model The Agents The three stakeholder types were modeled as agents. The factors affecting the decision of each agent were stored as agent property parameters. The model allows for the change of behaviour of the stakeholders; this is implemented by allowing certain parameters to change. Each agent type has also a happiness attribute that fluctuates according to the difference between the achievement of its goals and any decisions that are made. The happiness of an agent may impact its goals and future decisions. Agents also have access to a database of past decisions (knowledge base) from which they can draw information to help in making a current decision and to which decisions are appended. Additional factors influencing agents decisions are also quantified in the model and stored as variables or constants. Constants include: the Figure 1. Model Flowchart The developer agent takes into account his goals, the land-use map, the land value, the development cost, the municipal development plan, the housing market, the land-use bylaws, and the wetlands within the parcel and generates and submits a development proposal. The citizen agent evaluates the development proposal in relation with his goals and the wetlands affected by the development, and generate an opinion. Then the planner agent evaluates the development proposal also in relation with his goals, the land-use map, the municipal development plan, the land-use bylaws and environmental regulations, the wetlands affected by the 4

5 development, and makes a decision taking the citizens opinion, the housing potential, the housing demand and the population growth into account. A revised decision forces the developer to revise and resubmit his development proposal. A deny decision affects the housing potential and demand. An approved decision affects the housing potential, the wetlands and the land-use map. The model runs in a sequential fashion where the land development proposals and the change of land-use occur sequentially, allowing for one proposed land development to impact the actual land use as well as influence future neighbouring land developments. The user can also modify the land-use bylaws and environmental regulations (Figure 5), the real estate market value and construction costs (Figure 6), and the population and housing properties (Figure 7). The Run-time Variables panel shows the values of the variables as the model runs such as the population statistics and the proposed development (tuple) (Figure 8), the opinion and change of behaviour of the citizen and planner (Figure 9), and the change of behaviour of the developer and the variables affecting his decision (Figure 10) The Model Interface The model has been programmed in Java and simulates agent-environment and agent-agent interactions. Agent-environment interactions occur when agents see the entire environment over which they make decisions. Agent-agent interactions are defined as logical rules and communication between agents that occurs as messages. The model involves three components: 1) a simulation module of the agent goals, interactions and decision making that creates the development scenarios, 2) a simulation module that changes the land use based on development scenarios and neighbouring existing land use according to the land-use bylaws of the town, and 3) the combination of these two modules that takes the agents decisions in a development scenario and applies them to the environment. The model interface contains three panels: 1) Initial Conditions, 2) Run-time Variables and 3) Land-Use Map. The Initial Conditions panel allows the user to modify the initial parameter values that include the planner properties (Figure 2), the citizen properties (Figure 3), and the land developer properties (Figure 4). The properties for the planner, citizen and developer include a pairwise comparison matrix that allows the user to input the intensity of importance of one goal over another. The values in the matrix are then checked for consistency by normalizing the eigenvector by the eigenvalues of the reciprocal matrix. If the consistency ratio is less than a certain value, then the values are said to be consistent; otherwise, the user is asked to check the values he entered. The benefits of using this method over a straight rank weighting are twofold. First, the resultant weights are not only relative to one another, but they also have absolute values. Second, the user only compares two goals at a time rather than subjectively weighting all goals at the same time. Figure 2. Planner properties 5

6 Figure 3. Citizen properties Figure 5. Bylaws and regulations Figure 4. Land developer properties Figure 6. Construction and market properties 6

7 Figure 9. Citizen and planner opinion as well as behaviour change Figure 7. Population and housing properties Figure 10. Developer behaviour change and iterated decision variables The Land-Use Map panel shows the changing land use as the development proposals are approved. Figure 8. Population and proposed development Following the approval of a proposed development scenario, the values that are contained in the tuple define how the land-use environment changes based on land-use change rules. Typical land-use change rules were developed with the assistance of an urban planner. These rules are based on the existing land use on parcels adjacent to the proposed land development parcel, and laws of attraction and repulsion. Examples of these laws are: single-family (R1) housing is attracted to adjacent open/park space, multi-family (R2/3) housing is attracted to adjacent commercial land-use, wetlands (ER: environmental reserve) are surrounded by a park land (MR: municipal reserve) buffer. The information contained in the tuple states the wetlands that are proposed to be moved and the percentage of land use to be developed including commercial, industrial, and each residential type. The proposed land-use change 7

8 then affects the future land use of adjacent parcels and the impact on wetlands can affect the goals, happiness and decision making of the agents Temporal Boundary A 30 year future simulation time period has been defined as the temporal boundary, with a one year incremental step; however the number of developments approved within the one year increment can be varied by the planner agent. The 30 year temporal boundary lies within the future plans of the Towns municipal development plan and appears as a reasonable length of time over which to run simulations. 3. Results Several different development scenarios were performed using the model to visualize the impact on the development of Strathmore. These scenarios include: increasing and decreasing residential density, allowing and forbidding any impact on wetlands, increasing the multiplier (by which recreated wetlands must be enlarged or monetarily compensated) when moving wetland due to a loss of natural habitat, and decreasing the average household size. Figure 13. Planner opinion Figure 14. Planner decision Figure 15 shows the existing undeveloped parcel with a UR (urban reserve) land-use designation containing eleven wetlands. Figure 16 illustrates the same parcel with a change of land-use R1&R2X (residential), ER (environmental reserve), and MR (municipal reserve) and the consolidation (moving) of eight wetlands that has been obtained after one year of simulation. The following figures display the results of one development scenario over one land parcel illustrating respectively a development proposal (Figure 12), the opinion of the planner when evaluating his goals (Figure 13), and the decision of the planner (Figure 14). Figure 12. Proposed development Figure 15. Existing parcel undeveloped 8

9 5. References [1] Marceau, D.J., What can be learned from multiagent systems?, In: Monitoring, Simulation and Management of Visitor Landscapes, R. Gimblett, ed., University of Arizona Press, 2008, pp [2] Wooldridge, M. J., Reasoning about Rational Agents, The MIT Press, Cambridge, MA, Figure 16. Approved parcel with land-use change 4. Conclusion The model output is a series of raster-based landuse maps showing the change of the environment based on the goals and decisions made by the agents. The results generated by the model have yet to be validated using two approaches: conceptual validation that consists in verifying if the underlying assumptions are appropriate for the purpose of the modeling exercise, and operational validation that will determine the extent to which the model produces outcomes that match the real system under investigation [10]. Meetings with the stakeholders will be organized where they will be shown the results and asked if they accurately portray the intentions of the actual decisions made on the land development project. Providing the model is generating realistic results, the stakeholders will be invited to suggest different development scenarios to forecast the possible future town growth based on the carry-over of agent goals. This will allow further testing of the robustness and usefulness of the model in a range of conditions. This model could potentially lead the stakeholders to achieve a beneficial outcome in the best interest of everyone when facing the challenge of new urban development. By proving the value of a simulation model having few stakeholders and a small geographical extent, the model could in the future be enhanced to include a larger number of stakeholders having more complex interactions and expanded to cover a much larger and complex region. There are many regions in the Province of Alberta and elsewhere that would benefit from such a model. [3] Anwar, M., C. Jeannert, L. Parrott, and D.J. Marceau, Conceptualization and implementation of a multi-agent model to simulate whale-watching activities in the St. Lawrence estuary in Quebec, Canada, Environmental Modelling and Software, 22, 2007, pp [4] F. Bousquet, C. Le Page, I. Bakam & A. Takforyan, Multiagent simulations of hunting wild meat in a village in eastern Cameroon, Ecological Modelling, 138(1-3), 2001, pp [5] M. Janssen, B. Walker, J. Langridge & N. Abel, An adaptive agent model for analysing co-evolution of management and policies in a complex rangeland system, Ecological Modelling, 131(2-3), 2000, pp [6] S. Feuillette, F. Bousquet & P. Le Goulven, Sinuse: A multi-agent model to negotiate water demand management on a free access water table, Environmental Modelling and Software, 18, 2003, pp [7] M. Batty, Agent-based pedestrian modeling, Environment and Planning B, 28(3), 2001, pp [8] A. Ligtenberg, A. Bregt & R. van Lammeren, Multi-actor-based land use modelling: spatial planning using agents, Landscape and Urban Planning, 56(1-2), 2001, pp [9] M. Monticino, M. Acevedo, B. Callicott, T. Cogdill and C. Lindquist, Coupled human and natural systems: A multi-agent-based approach, Environmental Modelling & Software, 22(5), 2007, pp [10] M.A. Louie and K.M. Carley, Balancing the criticisms: Validating multi-agent models of social systems, Simulation Modelling Practice and Theory, 16, 2008, pp