【作者】 张彬；

【导师】 王海军；

【作者基本信息】 武汉大学 ， 土地资源管理， 2022， 博士

【摘要】 作为20世纪以来人类社会的主要发展进程,城镇化在改变生产生活方式、推动我国社会经济快速发展的同时,也导致资源短缺、环境恶化和城乡发展不平衡等多方面问题。我国为此提出新型城镇化战略,并着手构建新时代国土空间规划体系,这需要充分掌握城镇扩展的演变机制,并精准把握其未来的发展趋势。因此,深入剖析城镇扩展机制,提升城镇扩展模型的模拟和预测能力,能够有效支撑国土空间规划编制工作,对提升其科学性和合理性具有重要意义。元胞自动机(Cellular Automata,CA)因其框架简单、开放且易与其他方法耦合的优势,在城镇扩展模拟研究中应用广泛。CA的转换规则直接影响着其对城镇扩展的模拟和预测性能,主要包括自发式规则和自组织式规则两个方面;前者刻画了城镇用地在空间因素驱动和供需关系影响下的扩展过程,后者表征了城镇用地在局部范围内受已有城镇用地分布的影响而倾向于扩展的过程。转换规则的改进和优化一直是提升城镇扩展CA模拟性能的研究热点和重要内容,然而仍存在以下不足有待解决:(1)城镇扩展是复杂地理系统的演化,兼具非线性和空间异质性特征;然而目前的CA建模研究着重以强大的学习模型挖掘城镇扩展的非线性机制,欠缺对其空间异质性特征的考虑,也尚未实现城镇扩展驱动机制挖掘时非线性和空间异质性的耦合。(2)空间异质性不仅存在于城镇扩展的微观驱动机制中,还因其扩展速率在空间上的差异,存在于城镇扩展的宏观规模层面;然而城镇扩展CA建模时常采用全局规则控制城镇规模分布,缺少对其空间异质性特征的表达。(3)邻域作为描述城镇用地自组织扩展过程的组件,其构建一直基于空间邻域平稳性的假设,认为城镇扩展模拟过程中所有元胞的邻域在空间域内保持一致,这明显与城镇扩展的空间异质性特征相违背。(4)根据空间邻域非平稳性重新定义邻域,使邻域在空间域内差异化分布,涉及邻域尺寸选择的潜在矛盾,即小邻域可较好地约束模拟结果形态,而大邻域可较充分地刻画局部自组织过程;如何在不显著降低效率的情况下解决此矛盾,仍有待探索和研究。针对上述不足,本论文以城镇扩展CA建模中空间异质性的表达为主要目标,以自发式和自组织式城镇扩展过程为研究对象,改进城镇扩展CA转换规则的构建方法并分析其优势,构建了顾及空间异质性特征的城镇扩展CA,以武汉市2000年至2020年城镇扩展历程验证其可靠性及优势,并设定多情景方案以预测武汉市于2035年的城镇用地空间格局,进而为武汉市国土空间规划编制和完善提供相应的对策及建议。本论文的主要内容及研究结论可概括为:(1)顾及空间异质性特征构建刻画自发式城镇扩展的转换规则。本论文基于人工神经网络(Artificial Neural Network,ANN)强大的非线性建模能力,融入地理加权回归(Geographically Weighted Regression,GWR)的空间异质性建模原理,构建“机理-学习”耦合模型,即地理加权人工神经网络(Geographically Weighted Artificial Neural Network,GWANN),实现自发式城镇扩展的非线性驱动机制挖掘过程中空间异质性特征的表达;提出分区数量控制(Partitioned Quantity Control,PQC)规则,依据城镇扩展在子区域的扩展规模/速率差异,表达宏观规模层面城镇扩展的空间异质性特征;结合二者构建了顾及空间异质性特征刻画自发式城镇扩展的转换规则,以此建立CA来模拟武汉市2000年至2020年的城镇扩展进程。结果表明,采用改进的转换规则可以获得更高的模拟精度,形态破碎度和形状复杂度更低的城镇斑块,以及主要斑块优势度更强的斑块分布,显著提升了城镇扩展CA的模拟性能。(2)顾及空间异质性特征构建表征自组织式城镇扩展的转换规则。本论文首先提出了双尺寸邻域(Dual Size Neighborhood,DSN),兼顾小邻域对模拟结果的形态约束功能和大邻域对局部自组织过程的刻画优势,避免了CA邻域尺寸选择中的潜在矛盾,然后进一步设计了尺寸自适应邻域(Size Adaptive Neighborhood,SAN),以元胞所在位置的交通可达性水平决定其邻域大小,基于此构建了顾及空间异质性特征来表征自组织式城镇扩展的CA转换规则,并采用了相应的并行算法计算邻域效应以提升效率。采用改进的自组织式转换规则模拟武汉市2000年至2020年城镇扩展,结果表明其能够显著提升城镇扩展CA的模拟精度,降低斑块破碎度和形态复杂度,提高主要城镇斑块的优势度和斑块之间的连通性。(3)全面解析空间异质的自发式和自组织式转换规则的优势及原因。本论文从方法训练性能、模拟结果的景观形态、模拟精度和模拟效率等多个方面评价和分析了空间异质的自发式和自组织式转换规则在城镇扩展CA建模中的优势及原因。结果表明:GWANN因能兼顾城镇扩展的非线性和空间异质性特征,其驱动机制挖掘性能更为优良且可靠,可以准确地再现城镇扩展的空间分布,获得具有清晰空间细节和规整斑块形态的模拟结果;PQC规则因能精准控制子区域的城镇扩展规模/速率,可以显著提升模拟精度。DSN对邻域功能的分离使其能够避免邻域尺寸选择中的潜在矛盾,提升CA模拟填充式和边缘式城镇扩展的能力;SAN由于顾及了自组织式城镇扩展的空间异质性特征,不仅能够强化DSN的优势,而且可以弥补其在模拟零散斑块扩展方面的不足,提高模拟精度和模拟结果形态的规整性。(4)顾及空间异质性特征的武汉市城镇扩展CA建模及情景模拟。本论文耦合空间异质的自发式和自组织式转换规则,建立了顾及空间异质性特征的城镇扩展CA以模拟和预测武汉市的城镇扩展过程。在国土空间规划的需求指导下设定了多种情景方案,预测了武汉市于2035年的城镇用地空间格局,识别武汉市未来城镇发展过程中的重点关注区域。经分析可知,蔡甸区、江夏区和黄陂区为武汉市未来城镇发展过程中需重点关注的区域,其中蔡甸区和江夏区是落实开发边界限制、加大耕地和生态空间保护的重点区域,黄陂区应着重抑制城镇用地侵占耕地。此外,城镇扩展与规划开发边界的冲突区域主要分布在主城区的边界,在落实规划方案时,应因地制宜地制定开发方略,保障城镇开发边界的底线功能。本论文在理论层面可为城镇扩展CA的转换规则构建提供新方法和新思路,在实践方面可提升城镇扩展CA对城市土地利用演变的模拟和预测能力,有效服务于国土空间规划的编制与完善,满足新时代国土空间规划的任务需求。更多还原

【Abstract】 As the main development process of human society since 20 th century,urbanization reforms the mode of production and life,and significantly promotes the development of socioeconomic.However,it also causes many problems,such as resource shortage,environmental pollution,and urban-rural gap.In order to improve the governance capacity of decision-makers,the China government has proposed the“new-type urbanization” strategy,and starts to build the architecture of territorial spatial planning in new era.The promotion of “new-type urbanization” strategy and the formulation of territorial space planning in new era both require comprehensive grasp of urban development mechanisms and accurate prediction of future urban patterns.Therefore,the detail interpretation of urban growth mechanisms can improve the simulation and prediction performance of urban models and support the territorial spatial planning effectively.Cellular automata(CA)have been widely used in urban growth simulation due to its simple and open framework and easy coupling with other methods.Transition rules directly affect the simulation and prediction performance of CA models,which mainly consist of spontaneous rules and self-organized rules.The former characterizes the spontaneous urban growth driven by spatial drivers and supply/demand relationships,while the latter represents the self-organized urban growth driven by the local interaction influences from existing urban land.The improvement and optimization of transition rules have always been the focus of urban CA modeling.However,there are still some issues to be solved: 1)Urban growth is the evolution of a complex geographical system,which has both nonlinearity and spatial heterogeneity.However,current CA modeling focuses on mining the nonlinear urban growth mechanisms with powerful learning models,lacks the consideration of spatial heterogeneity.Moreover,the combination of nonlinearity and spatial heterogeneity has not been realized in the interpretation of urban growth mechanisms.2)Since the urban growth rate differs in spatial domain,the spatial heterogeneity also exists in the macroscopic quantity of urban growth,but urban CA modeling usually adopts global rules to control the allocation of newly grown urban cells without considering the spatial heterogeneity in urban growth quantity.3)As the main component to characterize self-organized rules,neighborhood is defined based on the hypothesis of “spatial neighborhood stationarity”,its size is homogeneous for all the cells in the urban growth simulation,which obviously violates the spatial heterogeneity of urban growth.4)Dispensing the constraints of the “spatial neighborhood stationarity” hypothesis means to separately define the neighborhood for each cell.This leads to the conflict in the selection of neighborhood size,namely,small neighborhood is good at constraining the landscape of simulated urban patches,while large neighborhood has better ability to characterize local interactions.It needs further research and exploration to solve this contradiction without reducing simulation efficiency.In order to improve the simulation performance of urban CA,this paper incorporates the spatial heterogeneity into urban CA modeling,and builds the CA with the consideration of spatial heterogeneity(SH-CA).Taking the urban growth of Wuhan during 2000–2020 as an example,the reliability and advantages of the SH-CA have been validated.With several scenarios for future urban growth,the SH-CA is used to predict the urban land distribution of Wuhan in 2035,and the results are adopted to support its territorial spatial planning.The contents and conclusions of this research are generalized as follows.(1)The transition rules that can characterize the spatial heterogeneity of spontaneous urban growth have been established.A “physical-learning” hybrid model,namely,geographically weighted artificial neural network(GWANN),has been proposed to interpret the spatial heterogeneity of spontaneous urban growth when mining its nonlinear mechanisms.The GWANN is built by incorporating the locally weighted principle of geographically weighted regression(GWR)into the nonlinear modeling process of artificial neural network(ANN).The partitioned quantity control(PQC)rule is constructed to characterize the spatial heterogeneity of macroscopic urban growth quantity according to the spatially differed growth rate.The transition rules that better consider the spatial heterogeneity of spontaneous urban growth are established by coupling the urban growth mechanisms mined by GWANN and the PQC rule,and are used to simulate the urban growth of Wuhan during2000–2020.The results show that the optimized transition rules increase the simulation accuracy of urban CA and the dominance of main urban patches,and reduce the fragmentation and shape complexity of simulated urban patches,which significantly improves the simulation performance of urban CA.(2)The transition rules that represent the spatial heterogeneity of self-organized urban growth have been constructed.Firstly,the dual size neighborhood(DSN)that couples a large sub-neighborhood and a small sub-neighborhood has been proposed to avoid the potential contradiction in the selection of neighborhood size.Secondly,the transition rules to describe the spatial heterogeneity of self-organized urban growth are constructed by the size adaptive neighborhood(SAN),of which the size of large sub-neighborhood is determined by the local accessibility of cell locations.Furthermore,the parallel algorithm is used to improve the simulation efficiency of urban CA.With the optimized self-organized rules applied to simulate the urban growth of Wuhan during 2000–2020,the simulation accuracy,the dominance of main urban patches,and the connectivity between urban patches can be significantly increased,and the fragmentation and shape complexity of simulated urban patches can be markedly reduced.(3)The advantages of spatially heterogeneous spontaneous and self-organized rules and their mechanisms have been comprehensively interpreted.This study evaluates and analyzes the advantages of optimized transition rules from several aspects,including training performance,landscape of simulated urban patches,simulation accuracy and simulation efficiency.The results show that the GWANN has excellent and reliable ability to mine the driving mechanisms of spontaneous urban growth to reproduce its spatial distribution with clear spatial details and regular patch shapes,because it can couple spatial heterogeneity and nonlinearity to explain the spontaneous urban growth.The PQC rule can significantly improve the simulation accuracy of urban CA due to its precise control of the urban growth quantity in sub-regions.Since the DSN structure separates the two core functions of neighborhood,the potential contradiction in the selection of neighborhood size can be avoided,and the performance of CA to simulate infilling and edge-expansion urban growth can be improved.The SAN further strengthens the advantages of DSN,remedies its deficiency in simulating the growth of scattered urban patches,and improves simulation accuracy and the shape regularity of simulated urban patches,because it considers the spatial heterogeneity of self-organized urban growth.(4)Urban CA modeling and scenario prediction of Wuhan with the consideration of spatial heterogeneity.This paper combines the spatially heterogeneous spontaneous rules and self-organized rules to build the SH-CA for the simulation and prediction of urban growth in Wuhan.Several scenarios have been designed to satisfy different demands of the territorial spatial planning,and the spatial patterns of urban land in 2035 are predicted to identify the key-supervision region in the urban development.The results show that Caidian,Jiangxia,and Huangpi are key-supervision regions for the sustainable development of Wuhan.Among them,Caidian and Jiangxia are the key areas to implement the constraint of urban development boundary and strengthen the protection of cultivated land and ecological space,and Huangpi is the key area to prevent the occupation of cultivated land by urban growth.Moreover,there are many conflict areas of urban growth and urban development boundary distributed along the boundaries of the main urban area.It is crucial for sustainable development to formulate development strategies according to local situations to ensure the bottom-line constraint of urban development boundary.This paper can provide new methods and ideas for the development of CA modeling in theory,and improve the simulation and prediction performance of urban CA models in practice.The results and conclusions of this study can effectively support the formulation and improvement of the territorial spatial planning in new era and satisfy its requirements.更多还原