Processes and events in the centre: a dynamic data model for representing spatial change

ABSTRACT Traditional geographic information system models for map representation use superposition of layers to model physical reality, neglecting the integrity of the environment and limiting the ability to express interactions between features in complex phenomenon. This results in limitations regarding dynamic simulation and geographic causality reasoning. In this paper, we extend the framework of the geographic scene by formalizing the relationship between geographic processes and events to construct a dynamic data model: the process-event-centred dynamic data model. The key element of this data model is relationships between processes, events, and states of the natural or man-made phenomenon of interest. The identified relationships can be translated into a network of hierarchical, developmental, and causal graphs and realized in the Neo4j graph database. The implementation in the graph database supports spatio-temporal reasoning in geographic scenes and achieves an organizational framework for simulating spatio-temporal dynamics and complex calculations. The example of a 2019 mega-typhoon process is used to demonstrate the introduced process-event-centred model and its implementation in the graph database. A series of queries to the graph database show the capabilities of the data model for spatial reasoning and dynamic modeling.


Introduction
Inference of causal relationships and effects and complex dynamic simulation in human-environment systems is a requirement for addressing the challenges humanity is presently facing (Bleisch et al. 2014;Goodchild and Glennon 2008). Currently, systems natively supporting integrated spatiotemporal are still missing.
The approach to spatio-temporal modelling taken in this contribution is the development of a data model that is based on a solid conceptual framework called geographic scenes (Lu et al. 2018). Geographic scenes are a holistic representation of phenomena, with particular emphasis of processes, events, and relations between the changes induced by them and further components of the system. The core idea behind studying the spatio-temporal evolution of geography is the necessity to grasp the laws of spatio-temporal changes in the geographic environment (Lü et al. 2019), understand the driving force behind spatio-temporal changes, and then organize these entities and relationships in the geo-environment. Huang et al. (2019) proposed the description of geographic scenes from six aspectstime, space, people, things, events, and phenomena. They observed that scene nestedness and boundary ambiguity are the principal characteristics of geographic scenes and that material and energy exchanges did exist among different geographic scenes. The research object in the geo-environment was the natural environment, and the geo-scene went further to portray the interaction between humans and the environment.
Based on this conceptual foundation, this paper details the dynamic elements of geographic scenesprocesses, events, and statesand the relationships between them. The contribution made regarding the conceptual model of geographic scenes is the explicit modelling of dynamic features included in geographic scenes as this has not been covered by previous work.
The fundamental conception applied is that geographic events and processes are the abstractions of geographic changes at different temporal granularity of the investigation. It also clarifies that geographic events are geographic processes of significance or sudden changes in geographic processes, to propose the spatio-temporal dynamic expression model with the process-event at its core. This conceptualization provides the basis for a logical data model that supports the representation, simulation, and prediction of changes and effects present in geographic scenes.
The logical data model that puts processes, events, and states and their relationships in the foreground suggests the implementation in a graph database. Processes, events, and states are thereby represented as nodes and the relationships are the links between nodes. What is required when representing a real geographic scene, as in the case of the example provided, the geographic scene of a typhoon event, is an organic mechanism for sorting out hierarchical, developmental, and causal relationships in geo-scenes as well as the interrelationships among them to simulate spatio-temporal dynamics and complex computations. The findings about these relationships are translated into the graph database Neo4j by combining graph theory and knowledge graphs.
A main contribution is that we realize spatio-temporal reasoning in geographic scenes to achieve an organizational framework for simulating spatio-temporal dynamics, and complex calculations, by utilizing the Neo4j graph database. The approach using graph databases is an alternative to ontology-based approaches that have been used previously (Cao et al. 2018;Galton 2012;Uschold 2008;Del Mondo et al. 2021).
The remainder of this paper is organized as follows. Section 2 provides a brief description of related work in the field, analyzes the status and limitations of spatio-temporal dynamic data models currently in use, and describes the status of current research on spatio-temporal dynamic models. Section 3 dissects the components and dynamic representation mechanisms of the geographic scene, and analyzes the relationship among geographic events, geographic processes, and geographic states from a geo-scene perspective. Section 4 proposes a spatio-temporal dynamic process representation model with the 'process-event' at the core of the geographic scene, and implements a mapping rule between geo-scene logical model and Neo4j storage. Section 5 constructs a spatio-temporal dynamic model using the Lekima typhoon of 2019 as an example, to demonstrate the superiority of the proposed model for dynamic representation, spatio-temporal reasoning, and complex queries. Section 6 discusses the constraints of the model. Finally, Section 7 summarizes the study and indicates the direction for future research in the domain.

Related work
This section introduces previous research on processes and events, a categorization of existing spatio-temporal data models for contrasting the presented approach as well as the conceptual model providing the foundation for the presented dynamic data model.

Processes and events in GIS
A question that was discussed repeatedly is the differentiation between processes and events. The conceptualization of processes and events has been applied in various works for the development of data models for the representation of changes in space and time (Worboys and Michael 2005;Worboys and Hornsby 2004;Yuan and McIntosh 2003). Yuan (2001) argued that events are composed of aggregations of geographic processes in space-time. Claramunt and Thériault (1995) hold the view that events can be modeled as a set of processes and processes can be used to describe loworder spatio-temporal events. According to Galton (Galton 2012, 2015, events are short and occur at specific moments in time, whereas processes are of a more continuous nature with gradual changes. This view is in line with human perception.
In this contribution, an event is seen as a sudden and impactful occurrence. Processes are continuous occurrences that can provide detail of the behaviour of events or exist independently of events. The main difference to other conceptualizations of event and process is the role that events play in the geographic scene model. In geographic scenes, events are people's subjective perceptions, and some occurrences with special significance; they are impactful and cause change. The role of processes in this conceptual model is to provide detail on the development of scene objects being events, things, people and phenomena.

Research on spatio-temporal dynamic data models
Since Langran proposed temporal geographic information systems (GIS) (Langran and Chrisman 1992), there has been an effort to develop GIS data model that possesses capabilities of temporal evolution and dynamic representation (Peuquet 1994). Scholars have proposed many spatio-temporal data models with varying capabilities for geographic analysis and representation. These modelsmostly developed from map representations (overlay), and with different focusescan be classified into three categories: (1) Early Spatio-temporal data models that focus on organization of spatio-temporal structure include: snapshots (Armstrong 1988), base state with amendments model (Langran and Chrisman 1992), space-time cube (Gatalsky, Andrienko, and Andrienko 2004), conceptual approaches (Tryfona, Pfoser, and Hadzilacos 1997), and others. Many of them focus on the description of changes in the state of a single or a class of objects in space and time, weak in depicting the inter-influence relationship and linkage changes among each other (Langran 1992;Worboys and Michael 2005). Parent, Spaccapietra, and Zimanyi (1999) proposed a conceptual model named MADs centred on spatio-temporal objects, relations, and attributes, which emphasized the migration, generation relationships of spatio-temporal objects to represent dynamic change and enables the extension of causality. This work was of great inspiration in expressing dynamics, but MADs were not suitable for organization of complex dynamic changes because of their implicitly expressing the changes.
(2) Dynamic Models based on object-oriented ideology, which focuses on spatio-temporal objects and their associated relations (Raper and Livingstone 1995). These models place equal importance on time, space, and attributes in each spatio-temporal object (Yuan et al. 2010) to changes to geometries invoke changes to object identifiers, limiting abilities to compute geographic complexity (Yuan 2001). Such models are used for representing moving targets such as cabs, and can support massive spatio-temporal data storage and motion object queries (Hornsby and Egenhofer 2002;Liu et al. 2008). Designing for the spatio-temporal processes of a single type of geographic object, and some of them hardly portray the complex spatio-temporal processes of multiple objects, as well as a holistic description of the evolution of a geographic environment.
(3) Models based on geographic event/process involving representational models of geographic events and their state descriptions (Peuquet and Duan 1995;Worboys and Hornsby 2004;Yuan 2001;Allen, Edwards, and Bédard 1995;Campelo and Bennett 2013;Chen et al. 2018). Sengupta and Yan (2004) proposed a hybrid models for spatio-temporal data storage and querying based on changes in data elements. This approach is less fruitful with regards to reasoning and prediction in complex spatial relations for the reason of implicit representation of spatio-temporal topology and relations. Claramunt and Thériault (1995) applied events and processes to connect entities to form lineages networks to discover causal relations. But the versioning mechanisms are not conducive to the retrieval and analysis of big historical data. These types of models often involve cause-and-effect relationships, which are important for geographic knowledge mining and research on the mechanism of interactions. The data model introduced in this paper belongs to this category of models. However, these models present a confusing relationship between geographic events and geographic processes only ignored the role of the object in the surrounding environment and failed to create an overall environmental description for the dynamic process change simulation.
The above models played important role in an era where representation of map topics and manipulation of spatio-temporal data, even the hybrid and formal dual model of field and objects to representation of spatio-temporal dynamic (Hamdani, Thibaud, and Claramunt 2021), but the simulation of complex spatio-temporal dynamics from overall organization still requires more in-depth development.

Geographic scene model
Geographers mainly view GIS as a tool for data storage, management, display, mapping, and integration (Miller and Wentz 2003). This usage is owed to the shortage of spatio-temporal data models to support complex geographic analysis, geographic computation, and geographic process simulation.
Integrating previous research foundations of spatial-temporal conceptual models about emphasizing relationships for reasoning, using processes/events to organise spatial change a conceptual model called geographic scene model was constructed with a holistic view of the geographic environment (Lu et al. 2018). The geographic scene model is a combination of the geographic environment and people in a certain spatial region and the carrier and container for the existence, occurrence, as well as development of geographic features, geographic phenomena, and geographic events. The geographic scene can be divided into human scene and natural scene. In a human scene, it is important to emphasize human activism and the role of the person. In a natural scene, the human being is not a necessary element or the human being acts as a receptor. The geographic scene representation framework organically connects different geographic objects/phenomena, and provides beneficial support for GIS to realize complex geographic analysis, geographic knowledge discovery, etc. Huang et al. (2019) described the evolution process and components in the nested structure of the geographic scene and expanded the initially defined spatio-temporal relationships among elements to define numerous relationships (e.g. spatio-temporal, cause-effect, and role-based relationships). The model is abstracted based on human perception of geography; currently, it stays at the stage of conceptual model.
In this paper, we deeply investigate the structure of geographic scenes, explore the scene elements such as geographic events and geographic processes, dissect the complex relationships amongst them, and then construct a geographic scene dynamic expression model which is capable of expressing spatio-temporal dynamics, complex calculations and even spatial reasoning capabilities and practical applications by using a graph database.

Geographic scene conceptual model and dynamic representation
The geographic scene model emphasizes events and processes as the driving force causing changes in the geographic environment, and represents a conceptual model of spatio-temporal data to simultaneously express changes of multiple features with complex connections. The first step in the construction of a spatio-temporal dynamic model is to understand the dynamic structure and the interaction among events, processes, and states in depth. The representation of events, processes, and states is added to the conceptual model of geographic scenes to provide the basis for a logical data model of specific phenomena.

Geographic scene elements organization for dynamic representation
The geographic scene is composed of six elements (people, things, events, and phenomena, and processes) that are influenced by one another (Figure 1). Because of the same characteristics, people and things are treated as scene objects for the sake of expression. Geographic scenes with emphasis on people in the scenes reflect the influence of people on nature and society. Geographic things usually refer to the spatial features within the range of the earth's atmosphere, and include objects connected with geology, soil, biology, hydrology, architecture, and atmosphere. Things represent discrete distribution of geographic objects equivalent to features in traditional GIS. Events record sudden changes in the scene which are significant, impactful, identifiable, and discrete. Geographic phenomena describe a continuously and variably distributed field, or a complex of multiple people or things that are interrelated, or a simple geo-scene.
Temporality express scene time through precise time or vague time (Allen 1984). Spatiality includes both spatial location and geometric shape. Spatial location is used to express the place where the scene features occur and develop, and to answer questions related to location using coordinates, landmarks, place names, addresses, and other forms of expression (such as 'Where is Nanjing city?'). Attributes refer to information related to features besides time and space, which provide powerful support for geographic analysis, calculation, and simulation, and can be classified as qualitative and quantitative. Qualitative attributes include names, types, properties, etc. while quantitative attributes include level, codes, quantities, etc. Geographic attributes, physical attributes (density, humidity, boiling point, etc.), chemical attributes (acidity, flammability, etc.), and biological attributes (uniformity, degradability, etc.) are important attributes of geographic scenes and composites. Attributes are mainly used to express the basic characteristics of geographic scene, and answer questions about the characteristics of geographic scene and its elements. Semantics are used to describe the categories and definitions of geographic scene and its element, and to answer questions related to geographic scene features. Semantic information is usually provided in the form of schematic diagrams, classification systems, nomenclature, etc.
The relationships in geographic scenes combine the changes to form a unified whole and play an important role in realizing complex geographic analysis and spatio-temporal reasoning. Traditional GIS data models focus mostly on spatial, temporal, and semantic relationships in describing the association between geographic features. Geographic scenes enhance semantic-related relationships such as the hierarchy, authority, interaction, and attribute related cause-effect relationships, thus answering some of the 'why' and 'how' questions more completely; they further link different geographic elements together, aiding in explaining the mechanisms of change of geographic elements, and greatly enriching the expression of geographic information.
In terms of 'relationships', there are scene-to-scene, scene-to-elements, and element-to-element relationships in geographic scene. Relationships are mainly portrayed as (1) spatial relationships, i.e. topological relationships, orientation relationships, and distance relationships.
(2) Temporal relationships that express the continuous change of geographic scene and its elements over time, and the existence of temporal sequences in the evolution of different geo-scene elements. (3) Attribute relationship which refers to the relationship between the attributes of different geographic scene features, of which the most common attribute relationship is causality. (4) Semantic relations include hierarchical relations, attribution relations, synonymy relations, antonymy relations, and interaction relations. There are four main types of interaction relations: physical, chemical, biological, and human.
The state of a geographic scene is the sum of the temporal, spatiality, attributional, semantic, and relational characteristics of the elements of the scene (people, things, events, and phenomena) at a given instant. The above classification is shown in the right part in Figure 1.
Moreover, the geographic scene depicts the dynamic-static hybrid model with the relationship of scene constituents such as people, things, events, and phenomena which are interrelated and influenced by one another in a geo-process. On the left side of Figure 1, the geographic scene is a hierarchy, and people, things, and phenomena are the participants of geographic events, and are interrelated. A scene contains multiple events, a large scene contains multiple small scenes, but an event is only associated with a scene. In addition, scenes and phenomena are a combination of multiple elements, so scenes can also be a participant of events. Look into the relation between geo-scene and states, the static information of the geographic scene is portrayed by state characteristics such as time, space, attributes, and semantics.
Focusing on the middle part in Figure 1, the dynamic process (geographic process) is used to express dynamic changes of geo-scene and its elements. The construction of a spatio-temporal model is therefore based on a state-based approach. In general, scene/scene objects (things, people, phenomena) in the geographic environment evolve in time and space, and the chain of the changed state during their evolution is presented as a geographic process. Processes are used to organise their changes.

Relationship among geographic processes, events, and states
The definitions of geographic states, events, processes, and phenomena in the geographic scene model are given in Appendix 1.1. Both geographic events and geographic processes can model geographic change within a finite period. Events are bounded and discrete from one another, but processes within events are continuous.
First, the inherently nested events can be composed of temporally non-intersecting discrete subevents of lower granularity. For example, a geohazard event may include the occurrence of an earthquake followed by a landslide or mudslide. The geohazard is a geographic event, while the earthquake and landslide are its subevent. The symbol ';' is the concatenator operation. E is the main event, E1, E2, E3 refers to the subevent, Second, processes can be divided into different temporally non-intersecting sub-processes according to semantics. For example, according to the strength of the typhoon, typhoon process is divided into four sub-processes (birth, development, keep, and vanish). Big process P can be an aggregation of its the sub-process P1, P2, P3, as the equation, If a detailed view of an event is considered, with changes to the granularity of investigation, an event can also be regarded as a process that has certain phases of its own. For example, a riverbank flood event E, observed at a more fine-grained temporal granularity, in which a process existed is the gradually increasing of water level with time until the water level is higher than the riverbank. Thus, A process P of the gradual increase of the water level can be detailed to portray this riverbank flood event E. Assuming that Detail is the operator, the relationship upward can be expressed as: Thus, the difference between events and processes is related to the temporal granularity of the investigation.
Every geographic event must have a geographic process or a series of geographic processes corresponding to it. Moreover, the process is not certainly an existing geographic event. The occurrence of a geographic event initializes, terminates, and changes the state of the people, things and phenomenon in the scene and causes changes in geographic processes and events. Geographic processes can maintain or change the geographic state, while the relationship among geographic processes entails mutual influence or independent parallelism.
The relationship between events, processes, and states is shown in Figure 2. The cause-effect relationship describes the relationship between geographic scene elements that cause and are caused. The changes caused by geographic events are the creators of direct causality in the geographic scene, and among the driving forces that helped realize the spatio-temporal reasoning behind geographic scenes in this study.
4. Spatio-temporal dynamic model with 'process-event' as the core 4.1. The logical data model for geographic scene elements The idea of building a dynamic model based on the conceptual framework of geographic scenes and the definitions of events, processes, and states from above, is to formulate geographic processes using sequences of interconnected states. For reaching this objective, two tasks are important: (1) Sorting out the skeleton of relationships involved in the dynamic geographic processes in the geographic scene; (2) Building an event-process-centred state sequence model to represent spatio-temporal changes in the geographic scene. In task 1, the sorting out of the skeleton of relationships involves (a) establishing the geographic scene hierarchy using the relationships between events, processes, and states. (b) Identifying the developmental relationships among states, and among processes. (c) Using the inclusion relationships of scene-events/processes, processes-states, states-objects (include people and things), and so on, to organize spatio-temporal evolution changes.
The skeleton relationships for the geographic evolution process are organized as shown in Figure 3. The inclusion relationship in the vertical direction is indicated by the blue lines. The temporal evolutionary relationship in the horizontal direction and the bidirectional arrow connections indicate the interrelationship between states. They are represented by red and black lines, respectively.
Inclusion relations include processes containing sub-processes or state sequences, state chains containing states, and processes containing states. The development relation is the evolution of the process in time, that expresses the progression among sub-processes or among states, and are expressed as sequential relations in time (Precede and Next). The interrelationship mainly contains causality, subordination, interaction, and spatial relationship. Interrelationship established between states and states, or within scene objects (established facts independent of time), same as the semantic linkages in the knowledge graph. The dynamic data model constructed for the geographic scene should not only pay attention to the portrayal of both spatiotemporal relationships at specific moments and the expression of spatial relationships at different times in the same space but also depict the relationships of spatio-temporal processes at different time and space.
It is important to divide large processes into sub-processes to simplify complex processes. In general, processes are divided according to different stages within the life cycle. When sub-processes are used to describe the lifecycle of an event, the timespan of sub-processes should not overlap.
In the geographic scene model, a known event type is registered in a specific state receptor with a trigger condition. When the state changes and the trigger condition is reached (quantitative to qualitative change of the process), an instantaneous event burst occurs. The occurrence of an event triggers a change in the geographic state of some geographic scene features and associated geographic processes. The initiator of an event can be a thing, an event, a phenomenon, or even a scene, and the same is for those affected. When the event acts on the scene or geographic phenomenon, its essence is to work on the people and things of the scene, so the relationship between the event to the people and things in the scene should be portrayed. For states, scenes, phenomena, people, and things all have states, and it is also necessary to establish the connection between states and each scene element. Considering the relationship between processes and states, and the internal structural components of the geographic scene model, the event-driven model of geographic processes as shown in Figure 4 is obtained.
The relationships that exist in geographic space are not limited to the above-mentioned relationships. To achieve reasoning, the properties of these relationships were provided in Appendix 1.2.

Mapping rule between geographic scene elements and neo4j node-edge
To implement the storage of geographic scenes in Neo4j, the types, instances, scene elements, properties, and various relationships involved in geographic scenes need to be converted to a node-edge graph database storage form. In this part, we proposed a mapping rule to realized scene elements and relationships store in the Neo4j database.
(1) Storage strategies for scene elements Geographic scenes, peoples, and things are stored as nodes in the graph database, and their attributes as graph node attributes. Phenomena are decomposed into multiple people and objects with connections, storing them separately in node and using relational edges to tie them together into a 'phenomenon' type node. Things are also stored as node; it can be a hierarchy to organize different categories of things.
(2) Storage for type and instance Many classes of types and instances are involved in a geographic scene such as geo-scene type and instance, process type and instance, and event type and instance. The instances are stored as nodes, and the types are stored as properties with 'type-of' as property name, and every kind of type owns a specific value domain.
(3) Storage for process and event The process class consists of multiple sub-processes, each of which represents the development process of a different scene object. Geographic events and processes are stored as nodes, and each event is associated with a geographic process used to detail itself. The attributes of the events and processes are stored as node attributes.
(4) Storage for relationship Relationships are represented in the form of both directed edges and graph nodes. The node approach requires us to define the domain and range of the subject and object to link the relationship. In the nodal approach, explicit expression of the relationship is stored, which is conducive to the extraction of information relating to the subject-object relationship in the geographic scene. The relationships 'Next', 'Type of', and 'Include By' between states S1 and S2 shown in Figure 5 are stored using directed edges, and the 'Mutual' (interaction) and 'R' (Disjoint With) relationships between scene types are stored in the form of relationship nodes. Properties of a relationship are often used to represent constraints in the geographic scene (e.g. absolute numerical constraints, and transitivity or symmetry of the relationship). (5) Storage for states Geographic states cannot exist in isolation and must be attached to scene elements. The states are stored as nodes in the graph database (S1, as shown in Figure 5), and each state is connected to the scene element instance to which it belongs to via a directed edge 'Include By'. There is a temporal sequential relationship between different states of the same object, stored as 'Precede' and 'Next' directed edges in Neo4j. Similarly, state attributes are usually stored as graph node attributes.
The scene, event, process, state, and phenomenon are stored in neo4j with the label as 'Scene', 'Event', 'Process', 'State', 'Phenomenon' respectively, and things and people labeled by the specific name relative to itself. The attributes of nodes/relationships, like the fields of RDB, depend on the objects we describe.

Case study and analysis
To verify the usability of the 'event-process-state' spatio-temporal dynamic model constructed in this study, the case of a typhoon with complex changes was simulated and analyzed from the perspective of the geographic scene model. in dynamic representation model for geographic. R/Mutual is relationship between graphic scene class C E1 and geo-scene class C E2 in node method, domain and range (Mutual) is same as concept in RDF, I E1 is an instance of C E1 and I E2 is an instance of C E2.
Considering the coastal area of China as a geographic scene, a total of 492 typhoon events and 631 landfall events have occurred in the scene since 1949. In this study, Lekima (No. 1909), the largest typhoon that hit the area in 2019, was selected as a case to study; the Lekima typhoon trajectory consisted of 160 typhoon monitoring points collected by the China Weather Typhoon Network (http://typhoon.weather.com.cn/), as shown in Figure 6. 5.1. Modeling typhoon Lekima with 'event-process' as the core The model expression process was divided into three steps: first, a 'scene-process-state' model was established for the Lekima typhoon process to simulate the typhoon life cycle dynamically. Then, the correlation between the typhoon warning process for Lekima and the typhoon warning announcement process was integrated to establish the interconnection between two different processes. Finally, we built a geographic scene with typhoon disaster information and the typhoon process to facilitate the integration, analysis, and reasoning of multiple pieces of information related to typhoon events.
The time range of the geographic scene involving the typhoon was from 14:00 on 4 August 2019 to 14:00 on 13 August 2019, and the spatial range was from 0 to 60°N and 70 to 150°E. To portray the dynamics and real-time impact of typhoons in the geographic scene, we extracted typhoon Lekima as a discrete typhoon object. Setting one hour as the time interval, the characteristic states of the typhoon were identified as wind circle, direction, central pressure, latitude, longitude, speed, and time; the spatial and temporal changes of the typhoon were simulated using the state sequence. In terms of geographic scene elements, elements of the typhoon Lekima scene involved human victims, things(disaster), events, and processes. Things include transportation, communication, and electric power facilities, farmland, houses, and enterprises, natural disasters, as well as other Figure 6. Typhoon Lekimaintensity and trajectory. disasters objects affected by the typhoon. Geographic scene events consisted of the 'typhoon Lekima event', typhoon landfall events and other events related to the invasion of typhoon Lekima. Geographic processes included the 'typhoon Lekima process', the 'Zhejiang typhoon warning process', and the typhoon invasion process of the city of Taizhou, etc. The model expression process is detailed below: (1) Structure of Typhoon Lekima events and processes 'Typhoon Lekima event' is described in detail by the 'typhoon Lekima Process', which contains four sub-processes (i.e. birth, development, keep, and vanish), each sub-process being connected to the temporally adjacent typhoon states. Based on the typhoon trajectory data, 160 typhoon states were identified in typhoon Lekima, which were numbered S0-S159 in chronological order. Typhoon events were associated with typhoon processes by 'depicted by' edges, and processes and sub-processes, and states were associated with the 'include by' relationship edges, as shown in Figure 7.
Typhoon landfall was registered as a 'Land fall' event type in typhoon state through relationship edge associations; In Lekima typhoon event, at the moment of S80 state the typhoon 'Typhoon landfall' event was triggered for the first time, and S122 the typhoon 'Load fall' event was triggered the second time.
(2) Lekima typhoon real-time warning process Typhoon warnings take the multi-temporal typhoon state as a basis and issue warnings after assessing the influence of the typhoon on the disaster-affected area. Zhejiang Province issued three typhoon warnings about Lekima namely the Level III, Level II, and Level I warnings. The Lekima typhoon geographic scene considers the three typhoon warnings as three states of the warning process. With warning time as a reference, the interrelationship between the warning status and the Figure 7. Typhoon Lekimaschematic diagram of the dynamic evolution model. typhoon status for a particular moment can be constructed, providing a link between the typhoon status and the warning status (S p1 , S p2 , S p3 ), facilitating the organization of information for subsequent typhoon studies. (

3) Typhoon Disaster Information Organization in Taizhou
Typhoon incursions were natural disasters that caused big damage to Taizhou in terms of transportation, communication, electricity, agriculture, and business. The Taizhou invasion event was outlined by the typhoon invasion process, which consists of a sequence of typhoon states semantically ranging from landfall in Taizhou to the departure of the wind circle from the Taizhou district (S80-S94); a nested 'event-process-state' pattern was formed to represent this dynamic process. The typhoon Taizhou invasion evoked typhoon disasters that included transient phenomena involving traffic, communication failures, power failures, agricultural losses, business losses, and sub-types of natural disasters, including human casualties. Each disaster sub-type included many disaster points consisting of other attributes such as the location of the disaster, its time of occurrence, etc.
In Neo4j, the 'typhoon disaster' node was the parent node of each sub typhoon disaster, and the association was established through the 'is_part_of' relationship edge. Disaster points instances nodes are added by creating next-level nodes of the corresponding disaster type, and overview information node instances for different disaster types can directly be created.
The typhoon Lekima process, typhoon warning information, and Taizhou disaster situation were organized based on the spatio-temporal dynamic model. Figure 7 shows the schematic diagram. The geographic scene features and the relationships between features are represented by different colours, with details provided in the legend. Figure 8 presents the results in Neo4j Browser.
For complex dynamic data, such as massive trajectories, we can pre-create CSV data structure and realize the automatic import of data into graph relations under different scale data update methods in neo4j by programming. Meanwhile, it can be supplemented with semi-automatic method to create specific relationships. The construction progress for the practical data: The first step is to specify which dynamic occurrences need to be simulated, then construct the main events, processes, and states to form the hierarchy structure of geo-scene; The second step is to consider the temporal development relationship between the state and state, the sub-process and sub-process; The third step is the interrelationship between states and states, things (people) and things. Of course, the causality and the predefinition of events in it need special attention. The procedure for typhoon dynamic model construction can be referred to a published paper (He et al. 2021).

Typhoon model support for spatial inference and complex query capabilities
Based on the ability of the NeoSemantics plugin in the Neo4j database to support inference on RDF (Resource Description Framework) models (Huang et al. 2020), we subsequently demonstrated the ability for spatial inference and complex queries of the typhoon evolution data model that resulted from this study.
The model can provide answers to the following propositions, demonstrating the dynamic representation and spatial reasoning capability of our model. return diaster.Name as Name,subdisater.Name as Catelogy,subdisater as Detial (Table 1) Q2: What was the landfall information for typhoon Lekima?
Cypher Language: Match (cat:Disaster{Name:Waterlogging"}) CALL n10s.inference.nodesInCategory(cat,{inCatRel:"give_rise_to",subCatRel:"is_part_of"}) yield node return labels(node) as Type,node.Name (Table 4) The query results showed that the typhoon dynamic information is organized into a geographic scene by the framework we constructed in this study. To know the damages information caused by typhoon Lekima, like the description by Q1, the method presented in the study can obtain the result just by using one query statement. This capability is implemented by virtue of the developmental relationships between states, and thus, the model in this paper can trace the changes of things. Using the ability to recursion in terms of hierarchical relationships in geographic scenes and continuous temporal relationships between states, Q2 demonstrates the ability to quickly retrieve dynamic information from the hierarchical structure by inference. Its input parameters are the process name and the type of registered event. Q3 shows the correlations between the different processes. Q4 demonstrates the ability to reason in typhoon dynamic models. Further, complex expression, retrieval, and reasoning of the typhoon dynamics and associated information can be realized by combining the information stored in the Neo4j database and the RDF reasoning tool NeoSemantics.
Compared to relational databases, our method avoids the establishment of complex multitable and multi-table relationships. Compared to the object-oriented database, the graph database is application-independent, keeping the data separate from the classes and objects of the application. The complex computational complexity degree of traversal algorithm in the graph database is O(1), which is based on Index Free Adjacency. And it is not affected by the increase of graph data volume. The graph-based model is suitable for retrieval and complex computation of big data. Since Neo4j databases also have much built-in functionality for working with graphs, for example, shortest path, depth traversal algorithm and other machine learning algorithms, our model is more flexible and extensible than the usual model for query, retrieval, and knowledge discovery.
The capabilities and efficiency of the ESTDM data model have been proven in supporting both spatial and temporal queries (Peuquet and Duan 1995;Pelekis et al. 2004). The Comparison analysis

Constrains
This model is limited to the representation of the change process of geographic objects using the state sequence method in the scene. In a dynamic model of a geographical scene, time, space, relationships, and objects are an inseparable whole encapsulated in the state. It is clear from the structure of the scene that the basic unit of construction of the scene turns out to be the state. In the dynamic process of development, people, and things are members of states. Even if people and things can appear separately, they represent facts that are already well-established (not change over time). When adding or deleting geographic states, relationships, geometries, attributes, semantic information, and other characteristics associated with the state will be added/removed along with state, to maintain the consistency, accuracy, and wholeness of the information inside the model. This differs from traditional conceptual models, and it is a pre-defined constraint for dynamic models to ensure that the data is correct.
The ability of the geographic scene to represent spatio-temporal dynamics depends on the granularity of the abstraction of the event and process. Because time as a dimension always permeates the objects and relationships in the model, inevitably increasing management complexity and time consumption. Exactly because of the time, the reasoning and complex calculations become more efficient.

Summary and conclusions
This study analyzed geographic events, geographic processes, and their interrelationships in-depth from the perspective of geographic scene. It deduced a process-event-centred model for geographic dynamic expression; and proposed a geographic event/process driven view of spatio-temporal dynamic evolution. This study adopts a 'whole to local' thinking (i.e. scene-process-state hierarchy) to tie the dynamic changes in the geographic scene in a holistic manner to enhance the computational capability of GIS to represents the complex dynamic changes. Adopting Neo4j to store the scenes, we formulated storage rules for scene elements and scene relationships in the geographic scenes to meet the demands of the expression for geographic changes. Based on the causal, hierarchical, and developmental relationships constructed by the geographic scene model, a graph query approach via NeoSemantics tool is used to provide complex query retrieval and spatial reasoning services for spatio-temporal processes. The ability of our proposed model in expressing spatio-temporal processes was demonstrated by considering the Lekima typhoon as an example, and inference capability of the data model was realized using Cypher language. The granularity of the abstraction in the geo-scene of the real world is different according to the actual requirements. With the storage and analysis ability of ten billion nodes and relationships, and graph models as a direct data source for machine learning and AI algorithms, the scalability of the geographic scene model is big, can be competent for the storage and analysis and prediction of big geo-scene data.
The organization method of this study for simulating spatio-temporal dynamics is portable and can be used for effective simulation and analysis calculations and even inference predictions for other geographical changes. Considering the environmental context, and the driving factors of geospatial dynamic processes, the spatio-temporal dynamic model that we propose not only represents the events and processes that have occurred but also provides opportunities for deep knowledge mining and spatio-temporal dynamic information retrieval and inference.
In terms of dynamic change expression and spatial reasoning in our model, there is still room for improvement in relationship expression and mining. First, the ability for inference based on semantic relations is mostly limited to direct causality, requiring further improvement in inference related to indirect causality implicit in the geo-scene. In terms of model analysis, a more comprehensive topological expression between geo-states will be established in our future work. In the next step, we also intend to use the data mining techniquescombining it with established referring modelto further explore the spatial reasoning of the geographic process model in different big data application scenarios.

Data and software availability
The typhoon dynamic model data was stored in Neo4j 4.1.0 which was constructed by manipulating Cypher language in visual studio 2017 with C#, and the spatial data is developed and organized by ArcObject 10.8 SDK for.net framework. The computational environment used during this research included a Dell WPS-15-9560 laptop with a Windows 10 operating system, 16 Gigabyte RAM and a 500G hard disk. The Lekima Typhoon information can be obtained from http://typhoon.weather.com.cn/gis/typhoon_p.shtml (last accesses 15.04.2021). The trajectories JSON was obtained from http://d1.weather.com.cn/typhoon/typhoon_data/2019/1909.json?callback=getData&_= 1618482419300 (last accessed 10.02.2021), then transformed into lima.csv by python code. Lekima Typhoon warning in Zhejiang Province information was extracted from the website (http://www.cma.gov.cn/2011xzt/2019zt/2019tf/ 20190808/index_3304.html (last accessed 10.02.2021)). The data mentioned (lima.csv, process.csv and warning.csv) and code used is available in a GitHub repository: https://github.com/heyf2018/DynamicModeOfTyphoon/tree/ main/DynamicModel_Typhoon.