基于多代理模型的城市土地利用博弈模拟

Multi-agent based modeling of land-use decision making process in a democratic setting

以美国达拉斯北部地区为例,通过建立基于多代理的人类系统和自然系统的耦合互动模型,研究不同土地管理策略的有效性。自然系统包含土地覆盖变化、流域水文动力学和野生动物栖息地生态系统。人类系统包括土地利益相关者(地主、屋主、开发商、政府)的影响土地利用变化价值观。系统强调自然对土地利用决策的反馈机制,不同类型代理相对空间位置的重要作用,以及同一类别中不同类型代理在特定情形下的相互转化。研究进行了两类模拟,一是模拟初始化的若干情景,包括不同类型地主和户主的分布、不同类型政府以及经济模型的假定;二是模拟主动式土地政策的实施效果。模拟结果表明,当政府购置保留地时考虑土地拥有者的价值观,较之单纯基于机会或生态因素更能导致有效的增长管理策略。

Residential development is a major driving force in the dynamics of urban land use and ecosystems.The type and rate of land development depend on the complex interaction among the stakeholders and their responses to the environmental consequences after a development decision is made.This paper introduces a framework of a coupled human and natural system and uses a multi-agent model to simulate the complex interactions among land stakeholders with respect to the decision making process for local land development.The human system was built on a multi-agent model,with each class of agents representing different types of landowners,homeowners,municipal government,and commercial developers,respectively.The decision making process of each agent class was modeled using an agent-specific multi-attribute utility function.The complex interactions among different classes of agents as well as among the agents of the same class were simulated with a given time lag in sequence.The natural system was built on a cellular automata platform,where the land cover transition rules were governed by a landscape model（MOSAIC） and a patch model（FACET）.The environmental quality indices to be used as feedback to the human system were generated from a hydrological model and a habitat model.In each lifecycle of simulation,human decisions on land development were passed to the natural system,which in turn generated environmental indices to be considered by the concerned agents（such as homeowners and the government） in the land-related decision making in the next cycle.Using a fast-urbanizing region in northern Texas as the study area,the model was run to produce simulations with a 25-year time span.Preliminary results demonstrated the ability of the model in simulating real dynamic situations at the qualitative level.It revealed a cyclic trend of interactions among agents,which was also observed in the real situation,with landowners and homeowners being the most active agent types.In addition,this study tested several land management strategies and revealed that considering landowner values when targeting available open space for preservation may lead to more effective growth management strategies than solely purchasing land based on opportunity or ecological factors.Extensive efforts are also required when applying this modeling framework to different socioeconomic and cultural settings,such as China.