Spatial Association Analysis Models,Algorithms And Their Geographical Applications Under Local Spatial Effects

"Space correlation" is the core idea expressed by the first law of geography,which lays the foundation for quantitative geospatial analysis.As a deep-level data analysis method,spatial association mining has important implications in ecological and environmental systems,social economy and other fields.Local spatial effects are a special case of spatial association.In geography,local spatial effects reflect the distribution,occurrence,development,and associated patterns of spatial objects(or spatial events)in the real world.During their existence and evolution,they show global conditions due to differences in local environmental factors.In most cases,especially in large-scale and heterogeneous spaces,the widespread existence of local spatial effects makes the results of global-pattern-based number mining often misleading.The result which can reveal "local variation of spatial correlation model" is valuable.Moreover,most of the existing local correlation analysis methods only consider the local features of the variable of interest across the space,and do not consider its evolution process at different time scales,and are not sufficient to provide comprehensive information of spatial object association patterns.Therefore,this paper starts from the spatial correlation analysis,introduces the local spatial effects into mining models and algorithms,and fully summarizes the time dimension analysis methods,constructs a general framework for local correlation and spatial-temporal evolution analysis,and studies the model and algorithm of spatial-temporal correlation analysis,exploring the feature region extraction method of spatial association patterns,interpreting the local variability of spatial association patterns,and designing typical ecosystem and socio-economic applications,empirical and analytical systems.The innovations in this article mainly include the following aspects:First,a general framework(LASEF,Local Association and Spatial-temporal Evolution Framework)for analyzing local associations and their spatio-temporal evolution is proposed.The framework consists of four modules:(a)exploratory analysis module,(b)space-time transition measurement model(c)Time analysis module and(d)Correlation factor analysis module.The four modules reveal spatial autocorrelation and heterogeneity of spatial data through classical spatial statistical methods and their extension methods to further measure and describe the local spatial effects,and make full use of the sampling time series under the premise that they are not based on Pre-segmentation of prior knowledge to study the space-time region can reflect the spatially related nature of geographical objects(events)and their spatial-temporal evolution rules,so as to minimize the omission of information.Second,a local spatial correlation measurement model is proposed.Based on the traditional LISA spatial correlation analysis model and the time dimension,this model proposes a new Spatial-temporal Transition Point(STP)model.The model studies how to analyze and aggregate LISA results over multiple consecutive time periods to find the transition characteristics(including transition positions and types)of the spatial correlation model,avoiding the unreliability of the spatial pattern of the objects identified by the traditional LISA method only at a single time point.Sex,and give the corresponding time-space density distribution indicators,reveal the evolution of the spatio-temporal transition.Again,This paper summarize a CTP(critical time point)method.The method begins by selecting a representative sampling time point as a key time point,and establishes a representation relationship between key time points and common time units.Its purpose is to be able to represent the sampled data in the time unit of common time units,thereby further establishing the correlation between time series and spatial objects(or time),and intuitively analyzing and expressing the space-time evolution process of spatial objects.Then,based on the traditional density-based clustering algorithm DBSCAN,this paper studies a new neighborhood matching clustering algorithm(MDC,Matching domain clustering).MDC tries to find a spatial object(phenomenon)that satisfies a given condition,that is,it reduces the amount of data for the subsequent GWR quantitative analysis operation,and can qualitatively screen the common influence factor to optimize the efficiency of local spatial correlation analysis among multiple variables.Finally,combined with the spatial data structure,spatial indexing technology and algorithm analysis,we construct and support the popular GIS hierarchical data structure.This paper studies the algorithm and implementation of the model and method of mining spatial-temporal association rules in the LASEF framework,and solves the important issues in practical application as the starting point,conducts practical verification and application of spatio-temporal association in different application fields,and solves the practical problems in geography research.The LASEF prototype system was designed to integrate the research framework model and algorithms into the LASEF prototype system.