Object-based Analysis And Knowledge Discovery By Modeling Spatio-temporal Evolution Of Geographical Phenomena

Understanding the spatio-temporal dynamics of our world is one of the eternal topics in geographic research.Geographic phenomena and its spatial pattern are manefastitions of the geographical dynamic processes in space and time.The spatiotemporal data about the geographical dynamic processes contains a wide spectrum of valuable geospatial information and knowledge to be discovered.With the rapid advance in spatio-temporal data acquisition capability and the increasing demand of modern society for high quality spatio-temporal information,the understanding of spatio-temporal structure,evolution,and interactions of geographic phenomena and their dynamic processes across all scales has become one of the most active research areas in geographic information science.With the increasing recognition of the value of geographic knowledge hidden in the spatio-temporal data,there is a pressing need to develop the new conceptual framework and methods that are capable of mining multidimensional,multi-scale,and multi-source spatio-temporal data to discover useful information and knowledge about the spatio-temporal patterns,relationships,and behaviors of dynamic geographic phenomena and processes.Geographic objects are the digital representations of geographic features in the real world.The charecteristics of the geographic features in space and and time can be described by the spatial,thematic,and temporal attributes of geographic objects,and the movement and dynamic behavior of geographic features can be also modeled by tracking the tstates and geometry of geographic objects in space and time.To discover the spatio-temporal information and knowledge about the complex dynamic geospatial proceseses,the object-based spatio-temporal approach to needs a flexible and powerful conceptual framework to represent the internal relations of component geographic objects in order to describe and examine the geographic phenomena and processes at multiple spatial and temporal scales and dimensions.Such a conceptual framework should be also capable of tracking spatio-temporal dynamics of geographic objects and acquiring the advanced knowledge at higher level of semantics with a fine semantic granularity.However,the conventional event-based spatiotemporal data model(ESTDM)is inadequate in defining and representing the semantic and hierarchical relationships between primitive objects and composite objects,which limits the capability in mining and discover information and knowledge about spatial structure and patterns associated with composite objects.From the perspective of spatial representation,the conventional ESTDM method represents the geographic dynamic phenomena as a continuous field of grid cells rather than discrete objects.Therefore,it is unble to capture the properties and the evolution trajectory of the geographic features represented by geographic objects.From the temporal perspectivce,previous studies have explored the statistical methods in identifying the temporal features and incidents and spatio-temporal clustering method in visualizing the spatial distribution of temporal features and incidents.However,little research has been conducted to examine the spatial properties and spatial diffusion of temporal features and incidents or the relationship between the temporal features and incidents and the spatial distribution of the geographic phenomena under study.To address the limitations of the previous studies mentioned above,this research proposes a new object-based conceptual framework and method to discover and analyze information and knowledge hidden in the big spatio-temporal data.The object-based conceptual framework and method is able to represent hierarchical relationships between primitive objects and composite objects,and to track the spatiotemporal evolution of geographic phenomena,and to identify temporal features and their spatial diffusion,therefore providing a solid foundation for various spatio-temporal information queries and knowledge discovery.The main components and contributions of this study are summarized as follows:(1)A novel Relational Attributed Neighborhood Graph(RANG)method has been developed to represent and define the morphological characteristics,spatial relationships and semantic characteristics of the primitive objects,and a numerical algorithm has been proposed to derive the thematic information of composite objects based on their primitive objects.The RANG method has been successfully applied to the derivation of urban land use information from the primitive urban land cover objects.A set of correlated primitive objects can be aggregated to form a higher-level composite geographic object.The automated recognition of the complex composite geographic objects and the inference of the information and knowledge about the composite objects from their constituent primitive objects is a technical challenge.This study proposed a novel data structure,referred to as Relational Attributed Neighborhood Graph(RANG),to represent the properties of primitive objects and their relations.The RANG method allows the thematic information of the composite object to be inferred.The properties and relations encoded in the RANG include various spatial,geometric,thematic properties and topological/spatial relations and thematic relations,which serve as the important information bridge to transfer the first-order(primitive)semantic information to the second-order(composite)semantic information.The structural information extraction process from primitive objects to the composite objects in RANG method is consistent with the cognitive process of the human reasoning.In RANG,the topology of primitive objects is analyzed by constrcuting neighborhood graph.Various features,including centrality,adjacency-event,connectivity,and morphological indicators are examined by a series of algorithms.The RANG method was successfully used to infer the composite urban landuse types in Washington,D.C.from primitive land cover objects with very fine spatial resolution.An overall accuracy of 86.60% has been achieved for the recognition of composite urban landuse types.Compared with the traditional XRAG(eXtended Relational Attributed Graph)data model,the RANG had improvements of about 12 percent.The resulting urban land use map provides the valuable information about the function and structure of different urban districts for urban planners and managers for various appplications.(2)A spatio-temporal object based representation framework has been established,which progressively organizes spatio-temporal objects to spatio-temporal sequences,and then to spatio-temporal processes.Numerical algorithms and software tool have been developed to generate the spatio-temporal path and spatio-temporal graph,which enables the extraction and visualization of the evolutionary tracks and spatio-temporal relationships of geographic objects.As a proof of concept study,this framework and associated algorithms have been applied for spatiotemporal query and analysis of dynamic drought phenomena.In this study,the conventional event-based spatiotemporal data model has been extended into a new object-and event-based spatio-temporal representation framework on the basis of the “Zone-Sequence-Process-Event” representation.The new representation framework consists of three levels of data structures: spatio-temporal objects,spatio-temporal sequences,and spatio-temporal processes.Numerical strategies for operating spatio-temporal topological relationships and backtracking algorithms have been developed to recognize and form spatio-temporal sequences from the spatio-temporal objects and to identify and form spatio-temporal processes from spatio-temporal sequences.The explicit representation of three levels of structural components allows for for analyzing topological relationships between spatiotemporal objects,spatiotemporal sequences,and spatio-temporal proceses and tracking the evolution of geographic objects in space and time.Particularly,the concepts of “spatiotemporal path” and “spatio-temporal graph” and associated algorithms make it possible to effectively represent and track the evolutionary relationships among geographic objects.The concept of spatiotemporal graph and corresponding computational algporithm was introduced for the first time to represent the spatial and temporal relationships between drought objects,sequences and processes across multiple scales,based on a set of topological and temporal criteria.Custom functions are created using C# with ArcObjects to manage the time dimension in ArcScene such that it can support the visualization of spatio-temporal path.We have applied this method to track,quantify,analyze and visualize the spatial movement and temporal evolution of the drought events occurred in California during 2000-2016.With this method,an in-depth analysis is conducted to examine the occurrence,spatial extent,expansion,reduction,and movement of the severe and extremely severe droughts occurred in California.Compared with existing studies,the new approach is not limited in spatial data handling and relational data tables.And our results also demonstrate that the object-based spatiotemporal representation framework and associated analytical methods have been effective in visualizing,understanding and analyzing the hidden patterns of the drought event.(3)The concepts of “temporal feature objects” and “temporal feature waves” are put forward for the first time to analyze the geographic distribution and spatial diffusion process of the temporal features.A series of automatic algorithms are presented to support the “temporal feature objects” and “temporal feature waves”.The introduction of these new concepts makes it possible to identify and quantify incidents and events from the temporal perspective and to track geographic diffusion of temporal incidents and events.The effectivenesss of this method has been demonstrated through a case study of vegetation dynamics.In this analysis,the temporal features are treated as the basic analytical units of temporal signals for the analysis of the spatiotemporal activities of geographic phenomena.The algorithms have been developed to identify the temporal features and segment the temporal signals according to the temporal features.Inspired by the spatial diffusion theory,we propose the novel concepts of “temporal feature objects” and “temporal feature waves” for the first time.This method begins by using Savizky-Golay smoothing method to shelter the noise in time series data.Then the fine logistic parameters were calculated using a least-squares fitting procedure.The temporal features can be extracted by calculating the curvature variation of logistic curve.The temporal feature objects are thus can be defined as the spatially continuous aggregate of locations where the same or similar temporal features occurred through a Suzuki connected components labeling algorithm.By calculating the spatio-temporal promixity distance between temporal feature objects,an adaptable distance cluster method was proposed to find the temporal feature objects clusters.Based on Delaunay triangular technique and the main skeleton line extraction algorithm,temporal feature waves are delineated from the temporal feature objects clusters.Different from previous studies,the hierarchical “temporal features?temporal feature objects?temporal feature waves” representation framework reflects the evolution of temporal incidents in the geographical space.This original representation framework complements the existing spatio-temporal representation method that emphasize the temporal evolution of the geographical features.We applied this framework and corresponding method to study the phenological dynamics of terrestrial ecosystem of the North Slope in Alaska.Based on the time series NDVI images,four key phenological temporal features have been detected,including green-up,maturity,senescence,and dormancy for every pixel.Then,four types of temporal feature objects are formed based on the spatial proximity of the same type of temporal features.Finally,phenological temporal featurs waves,including green-up wave,maturity wave,senescence wave,and dormancy wave,were derived by grouping temporal feature objects according to temporal proximity and continuity.These temporal feature waves reflect well the spatial diffusion of vegetation phonologic dynamics.Our case study demonstrates that the information and knowledge about spatiotemporal evolution process of vegetation phonology discovered by our method are reliable and geographically and ecologically more detailed and accurate than the previous studies for this region.(4)This study elaborates basic steps for all the proposed algorithms,and designs the corresponding data structure.With the help of the C# programming language and ArcObjects SDK for the Microsoft.NET Framework,a spatio-temporal evolution and visualization prototype system was implemented aiming to make automated processing of the algorithms and applications easier and provide better quality of data visualization on the proposed concepts.