EN中文

Journal of Geographical Sciences >

2021 , Vol. 31 >Issue 7: 1059 - 1081

DOI: https://doi.org/10.1007/s11442-021-1885-1

Research Articles

Visual method of analyzing COVID-19 case information using spatio-temporal objects with multi-granularity

  • CHEN Yunhai ,
  • JIANG Nan , * ,
  • CAO Yibing ,
  • YANG Zhenkai ,
  • ZHAO Xinke
Expand
  • Institute of Geographic Space Information, Information Engineering University, Zhengzhou 450052, China
* Jiang Nan, PhD and Professor, specialized in electronic map visualization and thematic map data processing. E-mail:

Chen Yunhai (1987-), PhD, specialized in spatial information modeling and visualization. E-mail:

Received date: 2020-10-21

  Accepted date: 2021-04-13

  Online published: 2021-09-25

Supported by

National Key Research and Development Program of China(2016YFB0502300)

Copyright

Copyright reserved © 2021.
Fold

Abstract

Coronavirus disease 2019 (COVID-19) is continuing to spread globally and still poses a great threat to human health. Since its outbreak, it has had catastrophic effects on human society. A visual method of analyzing COVID-19 case information using spatio-temporal objects with multi-granularity is proposed based on the officially provided case information. This analysis reveals the spread of the epidemic, from the perspective of spatio-temporal objects, to provide references for related research and the formulation of epidemic prevention and control measures. The case information is abstracted, descripted, represented, and analyzed in the form of spatio-temporal objects through the construction of spatio-temporal case objects, multi-level visual expressions, and spatial correlation analysis. The rationality of the method is verified through visualization scenarios of case information statistics for China, Henan cases, and cases related to Shulan. The results show that the proposed method is helpful in the research and judgment of the development trend of the epidemic, the discovery of the transmission law, and the spatial traceability of the cases. It has a good portability and good expansion performance, so it can be used for the visual analysis of case information for other regions and can help users quickly discover the potential knowledge this information contains.

Key words： COVID-19; spatio-temporal objects; multi-granularity; case information; visualization; visual analysis; spatial correlation analysis

Cite this article

CHEN Yunhai , JIANG Nan , CAO Yibing , YANG Zhenkai , ZHAO Xinke . Visual method of analyzing COVID-19 case information using spatio-temporal objects with multi-granularity[J]. Journal of Geographical Sciences, 2021 , 31(7) : 1059 -1081 . DOI: 10.1007/s11442-021-1885-1

1 Introduction

Coronavirus disease 2019 (COVID-19), which is mainly transmitted by respiratory droplets or through contact, is a severe acute respiratory distress syndrome (Li et al., 2020) and has a long incubation period and strong infectivity (DCPBChina, 2020). At present, there is no specific medicine to treat COVID-19, which poses a great threat to human health. Since its outbreak, it has had disastrous effects on human society. According to the World Health Organization’s (WHO) statistics, as of February 7, 2021, there were about 105.25 million confirmed cases and about 2.3 million deaths worldwide. Although the epidemic situation in China is basically controlled at present, there is still a growing trend in other countries and regions, which leads to a high risk of overseas import of the virus, and the status of epidemic prevention and control is not optimistic. Therefore, relying on existing public COVID-19 case information (referred to as “case information”) and technical methods, epidemic prevention and control strategies should be formulated to carry out in-depth research on the epidemic situation and to explore the laws and characteristics of the epidemic transmission situation.
Visualization is an important means by which human beings discover the hidden laws of data, and it is also an effective way to transmit complex information. For example, in 1854, the British doctor John Snow discovered the source of the cholera outbreak in London’s Brad Block using a map visualization method and case information (Li et al., 2008). After that, many scholars conducted in-depth research and experiments on map representation, visualization design, and spatial analysis of epidemic information, and a variety of map designs, visualization systems, and analysis tools (Carroll et al., 2014; Dominkovics et al., 2011; Maciejewski et al., 2011; Robinson et al., 2005) were produced, which can intuitively express and analyze the spatio-temporal differences in epidemic case information and help discover the potential epidemic laws and knowledge quickly. Since the COVID-19 global pandemic, researchers have also reported numerous representative studies related to COVID-19 visualization. For example, Mocnik et al. (2020) visualized and analyzed the COVID-19 epidemic situation in China, Europe, and the United States by combining thematic maps, broken line statistical charts, and temporal zone charts. Jiang et al. (2020) proposed an interactive visual analysis method for tracking the epidemic situation using a geographical knowledge map, in which the epidemic was visually analyzed from the macro- and micro-aspects according to the evolving case information. Chen et al. (2020b) used case information to construct a knowledge map of case activities in Zhengzhou, China. In general, researchers have paid more attention to the statistics, modeling, and analysis of macro information such as the number of patients in countries and cities (Booton et al., 2020; Chen et al., 2020a; Liu et al., 2020; Zhang et al., 2020), while few have comprehensively and systematically described the case information from the perspective of the visualization of spatio-temporal objects that presents its contents from different perspectives. Spatio-temporal objects can abstract and describe the real world in a way that makes it understandable for humans (Chen et al., 2020c), describing and depicting the spatio-temporal characteristics of spatial entities from different levels and scales. From the perspective of spatio-temporal objects, everything from the starry sky to cell viruses can be regarded as objects. Spatio-temporal objects are not only characterized by their attributes but also by their spatial forms, compositional structure, spatio-temporal relationships, and spatio-temporal behavior. By looking at the COVID-19 pandemic from this perspective, we will see that thousands of objects interact and influence each other, and they evolve and develop together with time. Using case information to build a vivid and dynamic epidemic object space will help to reduce the burden of understanding the COVID-19 pandemic, to explore effective prevention and control measures, and to improve our ability to cope with the risks posed by the epidemic.
Spatio-temporal objects rely on spatio-temporal data models with an expanding temporal dimension description, which are the data basis for spatial data analysis and visual expression (Yuan and McIntosh, 2003). Wu et al. (2016) reviewed the research status of spatio-temporal data models over the past 50 years and proposed that the connection between time and space should be strengthened to build a spatio-temporal data model that combines multi-spatial scale and multi-temporal granularity. In view of the deficiencies of the traditional spatio-temporal data models, Hua and Zhou (2017) proposed a data model for spatio-temporal objects with multi-granularity. The model has seven features, including multi-granularity, multi-type, multi-form, multi-reference system, multi-correlation, multi-dimensional dynamics, and multi-functional autonomy (Hua and Zhou, 2017), which can fully meet the requirements of COVID-19 pandemic modeling research and spatio-temporal analysis. Hence, in this study, the model is used to carry out object modeling, visual expression, and spatial correlation analysis of case information. This study proposes a new method of case information analysis from the perspective of spatio-temporal objects and systematically describes, expresses, and analyzes case information in order to provide a reference for the expression and analysis of related case information and for the development of prevention and control measures for the related epidemic.

2 Data

2.1 Case information

The case information used in this paper was mainly obtained via crawling of the epidemic information published by the health commissions (HCHP, 2020; HCLP, 2020; NHCC, 2020; PGJP, 2020; Tencent, 2020). The main contents include the statistical information for confirmed cases in China, the basic information for related confirmed cases in Henan province and Shulan, and the locations of diagnoses and their tracks. The data time windows are 2020.01.13-2020.02.25, 2020.01.17-2020.02.20, and 2020.05.07-2020.05.23, with the time range, number of cases, and number of confirmed patients being 44 days, 1239, and 45, respectively. From the daily changing trend of the newly confirmed cases (Figure 1), within the data time window, the spatial spread of COVID-19 has roughly experienced three stages: the out-of-control period, the effective control period, and the fluctuation danger period, among which the last period mainly represents the epidemic situation throughout China. The large-scale outbreak began in the first half of January 2020, and the number of confirmed cases has increased rapidly since then. Due to multiple measures and efforts, the spread of the epidemic had basically transitioned into the effective control period by the beginning of February, and the daily increase no longer increased sharply and unpredictably, but began to decrease. After late February, the epidemic was basically effectively controlled, with zero growth in many provinces each day. The epidemic prevention and control throughout China has been in a case fluctuation danger period with external input and internal rebound. A small-scale outbreak began in Shulan, Jilin province, on May 7, 2020, which was during the fluctuation danger period. The epidemic situation was effectively controlled after this short period of out-of-control spread, which shows that the epidemic prevention and control mechanisms and anti-epidemic experience had matured at this time, and the epidemic prevention and control abilities had been significantly enhanced.
Figure 1 Trend chart of the daily increase in COVID-19 cases in 2020 (a. Daily increase in China’s COVID-19 cases in 2020; b. Daily increase in Henan’s COVID-19 cases in 2020; c. Daily increase in COVID-19 cases associated with Shulan in 2020)
The case information used was mostly acquired from unstructured text data, which required corresponding data cleaning and structured processing. To accomplish the required data cleaning, Python’s jieba word segmentation tool, the Baidu Map API, and situational semantic analysis method were used to carry out structured processing and contagion relationship reasoning on the case information. The word segmentation tool was used to extract the spatio-temporal information, such as the time and location in the track text. The Baidu Map API was used to obtain the longitude and latitude information corresponding to the place names; and the situational semantic analysis was used for the case spatio-temporal trajectory series generation and the case contagion relationship reasoning. The reasoning results of the infectious relationships of the Shulan cases are shown in Table 1.
Table 1 Reasoning results of the infectious relationships of the Shulan cases
Case Source Case Source Case Source Case Source Case Source
1 / 11 2 21 11 31 22 41 22
2 1 12 8 22 15 32 17 42 40
3 1 13 10 23 11 33 28 43 15
4 1 14 2 24 16 34 31 44 43
5 1 15 10 25 17 35 22 45 43
6 1 16 10 26 4 36 39 / /
7 1 17 10 27 24 37 39 / /
8 1 18 10 28 11 38 4 / /
9 2 19 10 29 17 39 2, 10 / /
10 2 20 13 30 15 40 10 / /

1-45 are the codes of the associated Shulan cases; and a default value of “/” indicates no data

2.2 Prevalence rate

The prevalence rate refers to the frequency of the cases occurring in a certain period of time, which can reflect the impact of the diseases on people’s health to a certain extent. The prevalence rate is proportional to its influence. Its calculation expression is as follows: Prevalence rate = number of cases occurring during observation/population in the same time period. The prevalence rate calculation results for China and for Henan province are shown in Figure 2. As can be seen from Figure 2, the high prevalence rates of COVID-19 in China on February 25, 2020, occurred in central and eastern China, while the prevalence rates in northern, northwestern, and southwestern China were relatively low. On February 20, 2020, the areas with high prevalence rates in Henan were mostly distributed in the southern part of the province, which is close to Hubei province. In addition, as the capital city of Henan province, Zhengzhou had frequent personnel mobility and a relatively high prevalence rate.
Figure 2 Prevalence of COVID-19 in China (a. Prevalence of COVID-19 in China on February 25, 2020; b. Prevalence of COVID-19 in Henan province, China, on February 20, 2020. This is based on the standard map production with map approval number GS (2019) No.1696 on the standard map service website of the Ministry of Natural Resources, China; the base map has not been modified; and the relevant source of the statistical data is the 2019 Statistical Yearbook of China (CD-ROM version).)

2.3 Impact factors of the spread of the epidemic

COVID-19 is highly contagious, and its prevalence rate is affected by many factors, such as local epidemic prevention and control policies, population mobility, medical conditions, and economic strength (Ye et al., 2020). In addition to data collection, in this study, the population, the number of passengers, the number of healthcare personnel, and the gross national product (GDP) were selected as the impact factors of the spread of the epidemic, and the spatial correlation analysis of the prevalence rate of COVID-19 was conducted. In order to facilitate the later attribution detection, each impact factor was sorted according to the magnitude of the statistical values and was divided into seven levels (Level 1: extremely low; Level 2: relatively low; Level 3: low; Level 4: general; Level 5: high; Level 6: relatively high; and Level 7: extremely high) using the natural breakpoint method. Table 2 shows the classification of the relevant levels, and Figure 3 shows the spatial distributions of the epidemic impact factors in China. The higher the level, the darker the color. The treatment of the impact factors of the epidemic in Henan province was the same as that in China. Due to the limitations of the data sources, analysis of the cases associated with Shulan was not conducted at this time.
Figure 3 Spatial distributions of the impact factors of COVID-19 in China (a. Spatial distribution of China’s population in 2019; b. Spatial distribution of China’s number of passengers in 2019; c. Spatial distribution of China’s GDP in 2019; d. Spatial distribution of China’s number of healthcare personnel in 2019. This is based on the standard map production with map approval number GS (2019) No.1696 on the standard map service website of the Ministry of Natural Resources, China, and the base map has not been modified.)
Table 2 Classification of the impact factors of COVID-19 in China
Level Population
(104 people)		 Number of passengers
(108 people)		 Number of healthcare personnel (104 people) GDP
(1012 yuan)
1 < 934.0 < 1.75 < 13.25 < 0.82
2 < 2704.0 < 3.20 < 27.27 < 2.04
2 < 3941.0 < 4.81 < 33.09 < 3.03
4 < 4926.0 < 7.16 < 42.70 < 4.11
5 < 6899.0 < 11.04 < 62.40 < 5.62
6 < 8341.0 < 14.21 < 74.63 < 7.65
7 > 8341.0 > 14.21 > 74.63 > 7.65

The relevant statistical data source is the 2019 Statistical Yearbook of China (CD-ROM version), in which Hong Kong and Macao have no statistical indicators of passengers, and the statistics used were the total number of people entering and leaving the region at the end of the year

3 Methods

3.1 Data model of the spatio-temporal objects with multi-granularity

The data model of the spatio-temporal objects with multi-granularity is the basic component describing the digital computer world, which aims to abstract the real world into an analog world composed of dynamic entities with different granularities, in order to establish a digital world corresponding to the real world in the information space (Hua and Zhou, 2017). Spatio-temporal objects with multi-granularity have multiple characteristics, including the spatio-temporal reference, spatial location, composition structure, spatial form, association relationship, life cycle, and behavioral ability. Their formal representation is shown in Figure 4 (Hua and Zhou, 2017). We organized the case information into spatio-temporal objects with multi-granularity, which offered a good data foundation and running environment for case information expression and analysis.
Figure 4 Formal representation of the spatio-temporal objects with multi-granularity

3.2 Spatio-temporal object modeling of cases

Spatio-temporal object modeling is the basis of case information visualization, which aims to organize the case information into spatio-temporal objects with multi-granularity. Referring to the class template technology (Zhang et al., 2018) and the rules and constraint principles of the spatio-temporal objects with multi-granularity, the modeling can be divided into four parts: basic information modeling, contagion relationship modeling, dynamic feature modeling, and objectification processing.
(1) Basic information modeling
By establishing rules and constraints, the basic object information is organized into an object feature set, namely, OBInfo = {OID, SRS, TRS, Position, Attributes, SpatialForm, DateTime}, where OBInfo is the basic object information, OID is the unique identification of the object, SRS is the spatial reference set, TRS is the temporal reference set, Position is the initial position of the object, Attributes is the object’s attribute feature set, SpatialForm is the initial spatial form of the object, and DateTime is the object’s life cycle. Common modeling rules include the formal description of the spatio-temporal objects with multi-granularity (Hua and Zhou, 2017), unique OID index constraints, object feature conversion rules, and other constraints, such as the spatio-temporal reference mapping relationship, the life cycle length, and the attribute’s value range.
(2) Contagion relationship modeling
The contagion relationship records the information about the infection’s source, infected object, and infection time, which can be expressed and managed throughout the life cycle in an object-oriented way. The case infection has a one-to-many relationship and can form a collection of relationships. If “[]” represents the set and “{}” represents the relationship, the formal description of the case infection relationship is R = [{SO1, DO1, Relation Type, StartTime, EndTime}, {SO1, DO2, RelationType, StartTime, EndTime}, …, {SO1, DOn, RelationType, StartTime, EndTime}], where SO1 is the infection source of the object, DOn is the infected object, RelationType is the relationship category, StartTime is the relationship’s creation time, and EndTime is the relationship’s end time. As is shown in Figure 5, the change in the case infection chain with time can be further expressed through the infection relationship. The infection relationship is created, stored, and managed using the spatio-temporal series snapshot model (Worboys, 1994).
Figure 5 Modeling process of the suspected infection relationship of the cases
(3) Dynamic feature modeling
The dynamic features of the cases include the spatio-temporal trajectory, dynamic attributes, and status information. These types of information change rapidly, and the adoption of the series snapshot model may cause large data redundancy, which affects the efficiency of accessing the case information. Therefore, this information needs to be stored and managed separately. The formal description of the dynamic feature is DOA = [{OID, TableName, DateTime1, Field2, …, Fieldn}, {OID, TableName, DateTime2, Field2, …, Fieldn}, …, {OID, TableName, DateTimem, Field1, Field2, …, Fieldn}], where DOA is the dynamic feature set of the case, OID is the unique identification of the object, TableName is the storage location of the object’s dynamic feature information, Date Time1-m is the dynamic feature time stamp, and Field1-n is the formal description of the dynamic feature. As is shown in Figure 6, the spatio-temporal trajectory process and the status process can be visually characterized via dynamic feature modeling, and the other dynamic feature spatio-temporal processes are also modeled using this method.
Figure 6 Example of the process of spatio-temporal modeling of the dynamic characteristics of the cases
(4) Objectification processing
In essence, the objectification processing of case information is conducted to organize the basic case information, infectious relationships, and dynamic features into spatio-temporal objects with multi-granularity, i.e., the instantiation of the cases, under the constraints of the time-space domain and the class template (Zhang et al., 2018). Commonly used objectification methods include single interactive objectification and batch objectification. Among them, the monomer method has a strong flexibility and a friendly interface, but it has a low efficiency and a relatively high probability of human error. The objectification efficiency of the batch method is high, and the accuracy of the results is only related to the input parameters. The disadvantage is that if there is an input error or a network abnormality in the process, the objectification process will easily fail and the objectification process will need to be restarted. In this paper, based on the advantages and disadvantages of the two methods, the batch + single interaction method was adopted for the objectification and the single interaction method is mainly used to deal with the failure of the batch objectification. Finally, the objectification results are stored in the spatio-temporal object database or are generated in the corresponding object exchange format files (Figure 7). The spatio-temporal object database mainly provides dynamic storage of and access to the spatio-temporal objects with multi-granularity, while the exchange format files are mainly used for the sharing and dissemination of the objects’ data. Spatio-temporal objects with multi-granularity are stored in a mashup database, in which the master database (Postgresql) mainly stores the basic information about the cases, and the slave databases (Neo4j, Geomesa, Mongodb, and HDFS) store the case infection relationship, the dynamic features, the spatial morphology, and the data files, respectively. The exchange format file is a file representation method for spatio-temporal objects with multi-granularity, and its formats include XML, JSON, and binary.
Figure 7 Object-oriented processing flow of case information

3.3 Multi-level visualization

Spatio-temporal objects with multi-granularity have seven multi-features (Hua and Zhou, 2017), and their granularity characteristics, morphological characteristics, and association characteristics can be visually displayed using the multi-level visual expression method. Based on the objectification of the case information, the visual representation of the case information is divided into three levels: the macro-level, meso-level, and micro-level. The macro-level mainly analyzes the development trend of the epidemic situation and its impact on the social economy. The meso-level mainly analyzes the flow of people and the spread of the epidemic situation. The micro-level mainly analyzes the spatial traceability of the cases and the risk assessment of the epidemic situation. The concrete representation methods for the multi-level visualization include a three-dimensional (3D) dynamic thematic map and spatio-temporal process simulation (Table 3).
Table 3 Multi-level visual expressions of the case information
Expression level Expression method
Macro-level Spatio-temporal process simulation (attribute), three-dimensional dynamic thematic map
Meso-level Spatio-temporal process simulation (spatial form, trajectory, attribute)
Micro-level Spatio-temporal process simulation (spatial form, trajectory, attribute, infection link), dynamic thermodynamic map
(1) Three-dimensional dynamic thematic map
A 3D dynamic thematic map dynamically represents the attributes of the objects in the form of maps, realizes the automatic update of the map information through real-time access to the case information, and intuitively simulates the spatio-temporal changes in the thematic case information. Unlike conventional maps, 3D thematic maps based on spatio-temporal objects with multi-granularity can realize stepless scaling and rotation of scenes and effectively avoid mutual occlusion and glands between objects. Moreover, in the life cycle of the object, the update speed of the map information can be changed by back tracing and jumping, and the features of objects can be displayed in all directions in all periods. The thematic map making methods used in this paper include the equivalent area method and the zoning statistical chart method.
(2) Spatio-temporal process simulation
Spatio-temporal process simulation is mainly used to express the features of cases with continuous changes, for example, the number of newly confirmed cases each day and the movement track, spatial form, health status, and infection chain of the confirmed cases. Among them, the case trajectory describes the process of the spatial movement before the case is confirmed, and the spatial morphology expresses the process of the spatial morphological changes, such as before the case is confirmed, when the case is suspected, when the case is confirmed, and when the person is cured or dies. The different spatio-temporal processes are independent of each other, and the types and quantities of the spatio-temporal processes that can be added to the same object are not limited. In the life cycle of the object, the dynamic features of the object can be displayed in all directions, and all of the spatio-temporal processes can be traced back and jumped in the simulated space and time domains.

3.4 Spatial correlation analysis

Spatial correlation is a universal characteristic of spatio-temporal objects. The closer the spatial distance, the higher the correlation between objects. COVID-19 is mainly transmitted through close contact with infected people or the virus. If there are infection cases, there must be a spatial correlation between the infected people and infection sources. Therefore, spatial correlation analysis of the epidemic related objects is helpful in excavating the latent transmission law of the epidemic.
The spatial correlation can generally be measured by corresponding indicators, such as global spatial autocorrelation statistical indicators and local spatial autocorrelation statistical indicators. The commonly used statistical indexes of global spatial autocorrelation are the Moran’s I (Moran, 1950), Geary’s C (Geary, 1954), Getis-Ord General G (Getis and Ord, 2010), and Ripley’s K (Liu et al., 2020) indexes. In addition, semi-variograms are used in geo-statistics (Matheron, 1963). Local autocorrelation statistical indexes include the Getis-Ord Gi* (Getis and Ord, 2010), LISA (Anselin, 1995), and Satscan (Kulldorff, 1997) indexes. This work mainly analyzes the case information from three aspects: global spatial correlation analysis, local correlation analysis, and spatial impact factor analysis. The spatial correlation analysis methods used include the Moran’s I, LISA, and geographic detector methods (Wang et al., 2010).
(1) Moran’s I
The Moran’s I is mainly used to measure the relationship between the characteristic attribute values of the spatio-temporal objects as a whole, and the value range is [-1, 1]. If the exponent is positive, the attribute values of the spatio-temporal objects are positively correlated, and the objects attributes have spatial aggregation. If the exponent is negative, there is a negative correlation, and the distribution of the objects’ attributes is relatively scattered. An exponent of zero means irrelevance, and the attributes are randomly distributed. The Moran’s I is calculated as follows (Moran, 1948):
$I=\frac{n\mathop{\sum }_{i=1}^{n}\mathop{\sum }_{j=1}^{n}{{w}_{i,j}}\left( {{x}_{i}}-\bar{x} \right)\left( {{x}_{j}}-\bar{x} \right)}{\mathop{\sum }_{i=1}^{n}\mathop{\sum }_{j=1}^{n}{{w}_{i,j}}\mathop{\sum }_{i=1}^{n}{{\left( {{x}_{i}}-\bar{x} \right)}^{2}}}$
where xi is the prevalence rate of spatio-temporal object i; xj is the prevalence rate of spatio-temporal object j; $\bar{x}$ is the average prevalence rate of all of the spatio-temporal objects in the region; wi,j is the spatial weight, which can be obtained from the spatial weight matrix; and n is the number of spatio-temporal objects in the region.
(2) LISA
The LISA method, also known as the Local Moran’s I, is mainly used to describe the degree of the spatial aggregation of the object’s attribute values with a certain degree of similarity. There are usually four spatial aggregation modes, namely, high-high aggregation (high values are surrounded by high values), high-low aggregation (low values are surrounded by high values), low-high aggregation (high values are surrounded by low values), and low-low aggregation (low values are surrounded by low values). Its calculation equation is (Anselin, 1995):
${{I}_{i}}=\frac{\left( {{x}_{i}}-\bar{x} \right)\mathop{\sum }_{j=1}^{n}{{w}_{i,j}}\left( {{x}_{j}}-\bar{x} \right)}{\mathop{\sum }_{j=1}^{n}{{\left( {{x}_{i}}-\bar{x} \right)}^{2}}}$
where the meanings of xi, xj, i, j, $\bar{x}$, wi,j and n are the same as in Eq. (1).
(3) Geographic detectors
Geographic detectors are a mechanism attribution analysis method (Wang et al., 2010), which is mainly used to detect the spatial differentiation of the impact factors of numerical changes in the attributes of the spatio-temporal objects. The core idea of geographic detectors is to assume that an independent variable has an important influence on the dependent variable. Then, the independent variable and the dependent variable should have similar spatial distributions (Wang and Xu, 2017). There are four types of geographic detectors: interactive detectors, ecological detectors, factor detectors, and risk detectors. The impact factors of the spread of an epidemic can be identified from different angles. The interactive detector is used to detect whether the different factors act independent or interactive. The risk detector is used to detect the influences of the different levels of the impact factors on the epidemic. The factor detector is used to detect the interpretation degrees of the different impact factors. The expression for the geographic detectors model is (Wang et al., 2010):
$q=1-\frac{\mathop{\sum }_{h=1}^{L}{{N}_{h}}\sigma _{h}^{2}}{N{{\sigma }^{2}}}$
where N is the prevalence rate of COVID-19, and ${{\sigma }^{2}}$ is the variance of the prevalence rate. If each prevalence impact factor is divided into L layers for expression, h is equal to 1, 2, …, L. Nh and ${{\sigma }^{2}}$ are the prevalence rate and the variance of the impact factor layer, respectively. The range of q is [0, 1]. The larger the q value, the stronger the explanatory power and the greater the influence of the impact factors on the prevalence rate, and vice versa. If the q value is 1, the influencing factors completely control the spatial distribution of the prevalence rate. If the q value is 0, there is no correlation between the two.

3.5 Visualization design

In case information visualization, adopting C/S architecture, the VS2010 programming environment, the C++ programming language, the QT4.8.6 x64 image interface framework, the OSGEARTH three-dimensional digital earth framework, and the Google Protocol Buffer network transmission sequence structured data format have been used to load the basic scene, schedule and display the spatio-temporal object, express the dynamic feature information, and analyze the case information. The visualization design of case information includes three parts: function design, view design, and expression flow design.
(1) Function design
In case information visualization, the function design includes constructing the basic scene, managing and scheduling the objects, displaying the basic object information, and expressing the dynamic features. The basic scene is constructed using the OSGEARTH digital earth framework. The object management and scheduling is mainly achieved using object information loading, management and global scheduling, and by effectively interact with the spatio-temporal object database. The basic object information is displayed using the case attribute information from multi-dimensions. The dynamic features are expressed through the expression of the dynamic features of the case information in the form of spatio-temporal processes. Finally, the backtracking and jumping of the dynamic features of the spatio-temporal processes are achieved via interactive means.
(2) View design
The view design involved in the case information visualization is shown in Figure 8. Related views include: the case management and interaction view, the basic information display view, the scene display and interaction view, and the time management and backtracking view. The case management and interaction view mainly achieves the interactive loading, management, and list display of the objects, and the display content includes the objects’ spatio-temporal domain, class template, and tree list. The basic information display view mainly displays the multi-dimensional basic feature information of the objects, such as the case’s spatio-temporal reference, spatial position, attribute features, and spatial form. The scene display and interactive view is the main view in the case information visualization, which is mainly from the geographical perspective. The spatial form, spatial position, spatial relationship, and spatio-temporal dynamic process of the case are intuitively displayed, and the interaction with and control of the scene are realized by operating the mouse, such as the step-less scaling and rotation of the scene, object selection and querying, and the right mouse button function menu. The time management and backtracking view mainly realizes the control of the system’s simulation time through interactions, such as fast forward and backward, acceleration and deceleration, start and stop, and fast setting.
Figure 8 View design involved in the case information visualization
(3) Expression flow design
As is shown in Figure 9, the visual expression of the case information can be divided into two parts: data loading and visual expression. The case information is stored in the spatio-temporal object database, and the data loading process requires users to login to the system to obtain their corresponding access rights and to obtain the objects list index information. Based on the object list, users can load objects’ information as needed through interactive operations and can store the loaded objects in the memory object manager, which provides support for global scheduling and dynamic update of objects’ information in visual scenes. In addition, the case information can also be loaded by reading the exchange format files of the spatio-temporal objects with multi-granularity, but this process lacks flexibility, which is not conducive to user interaction. Visual expression can directly call the content in the object manager to achieve the display of basic case information, dynamic feature expression, scene display and update, entire life cycle management, and spatio-temporal process backtracking and control.
Figure 9 Flow of the visualized expression of case information

4 Multi-level visual analysis of case information

4.1 Macro-level visual analysis

Taking the statistical information for the cases in China as an example, the macro-level visual analysis of case information was carried out, and some of the visual effects are shown in Figure 10. The main view shows the existing confirmed, cured, dead, and cumulative confirmed cases in the various provinces and cities in the form of a 3D dynamic thematic map. The system simulation time step is a day, which is synchronized with the data acquisition cycle. The case-related feature information in the basic information view is updated synchronously with the change in the system simulation time. In addition, through interaction with the time management and backtracking view, the backtracking and analysis of the spatio-temporal processes of the statistical case information in China can be realized. As can be seen from Figure 10, except for Hubei, there were more cases in Henan, Hunan, Guangdong, and Zhejiang, which were greatly affected by the epidemic situation, and they had a high-risk epidemic situation. The spatio-temporal evolution of the spatial influence of the epidemic situation generally followed the first law of geography, i.e., with Hubei as the center, and the influence scope and influence gradually decreasing from the center to the periphery, and extending in the direction of the convenient transportation, such as toward Guangdong, Zhejiang, and Beijing.
Figure 10 Visual analysis of case information at the macro-level

4.2 Meso-level visual analysis

Taking the cases in Henan province as an example, meso-level visual analysis of the case information was carried out, and some of the visual effects are shown in Figure 11. The spatio-temporal processes of the cases are displayed in the main view. The visualization of the trajectory spatio-temporal process is assisted by the addition of trajectory lines for identification, which can intuitively display the historical trajectories of the cases. The hierarchical relationships between the different objects in the view are controlled by the viewpoint height constraint. The view at the high viewpoint mainly shows the trajectory movement and the spatial form change of the objects, and the view at the low viewpoint mainly shows the attribute changes of the objects. The initial simulation time step of the system is seconds, which can be adjusted by the user through the time management and backtracking view according to the actual needs. As can be seen from Figure 11, most of the cases in Henan are related to a history of tourism or residence in Hubei province, accounting for 60% according to the incomplete statistics. Before January 23, the cases in Henan province were mainly imported from outside, but the subsequent new cases were mainly due to local infection. As the number of infected people and the deterioration of the health status of cases increased, the spatial behavior of the cases gradually moved toward designated hospitals, and the number of hospital admissions increased daily.
Figure 11 Visual analysis of the case information at the meso-level. (a. Visual analysis of cases in Henan province on January 22, 2020; b. Local amplification effect of visual analysis results of Henan cases on January 25, 2020; c. Local amplification effect of visual analysis results of Henan cases on February 2, 2020)

4.3 Micro-level visual analysis

Taking the cases related to Shulan as an example, the micro-level visual analysis of the case information was carried out. Some of the visual effects are shown in Figure 12, and the results of the suspected infectious relationships are shown in Table 1. The main view of the system shows the spatio-temporal processes of the cases and the dynamic heat map of the permanent location before the cases were diagnosed, in which the 3D dynamic directed arc in the infectious link represents the infectious relationship and the connection node represents the cases. The dynamic heat map uses red, yellow, and green to express the degree of the permanent location of the cases. The red areas are severely susceptible areas, the yellow areas are moderately susceptible areas, and the green areas are mildly susceptible areas. The basic information view adds a list of infectious relationships to the original view, and users can control whether it is visible or not through interactions to assist in case feature information analysis. As can be seen from Figure 12, Jilin had the largest number of confirmed cases, accounting for 56%, making it the most affected in the entire epidemic situation related to Shulan. Based on the interpretation of the dynamic heat map, the high-risk areas (highly susceptible areas) in Jilin mainly included the Sihe Garden Community, the Linjiang Morning Market, Xinli Village, and the Neighborhood Community, which are the key areas for prevention and control. The suspected infectious link between cases is complete, and the source of the infection cases is clear (except for case 1), so this method can intuitively reproduce the spatio-temporal process of the spread of the epidemic and it can provide a more intuitive reference for related epidemic research.
Figure 12 Visual analysis of the case information at the micro-level

5 Spatial correlation analysis of case information

5.1 Global spatial correlation analysis

Using ArcGIS’s spatial statistical tool (Moran’s I), the prevalence rates of the provincial administrative units in China and the municipal administrative units in Henan were analyzed via global spatial autocorrelation. The calculation results are shown in Table 4. As can be seen from Table 4, the P values corresponding to all of the Moran’s I are greater than 0.1, which indicate a lack of statistical significance. This shows that there is no global spatial correlation between the spatial distributions of the prevalence rates in China and Henan.
Table 4 Results of Global Morin’s I for the prevention of COVID-19 in China
Region Value Date
2020-01-20 2020-01-25 2020-01-30 2020-02-05 2020-02-10 2020-02-15 2020-02-20
China Moran’s I -0.06 -0.02 -0.03 -0.04 -0.05 -0.05 -0.05
Z value -1.02 0.19 -0.01 -0.34 -0.48 -0.67 -0.70
P value 0.31 0.85 1.00 0.73 0.64 0.50 0.48
Henan province Moran’s I 0.01 -0.05 0.02 0.14 0.11 0.12 0.11
Z value 0.74 0.03 0.41 1.22 1.12 1.18 1.17
P value 0.46 0.98 0.68 0.22 0.26 0.24 0.24

The statistical inferences in the table are based on the 999 random permutations proposed by Anselin (1995)

5.2 Local spatial correlation analysis

Global spatial autocorrelation analysis can only detect whether the prevalence rate aggregates in space, but it cannot determine where the aggregation occurs. Local spatial autocorrelation analysis was performed using the spatial statistical tool (Anselin Local Moran’s I) in ArcGIS, and its cluster map (LISA) is shown in Figure 13. As can be seen from Figure 13, Hubei, Henan, Hunan, Chongqing, Xinjiang, and Tibet had local correlations at the provincial level. Hubei was a high-low value gathering area, and the prevalence rate was significantly higher than those of the other regions within the data window time range, which is thefocus of epidemic prevention and control. Henan, Hunan, and Chongqing were low-high value gathering areas, and there was an abnormal situation in which a high value area included a low value, which needs to be focused on during epidemic prevention and control. Xinjiang and Tibet were low-low aggregation areas, and the prevalence rate was low within the data time window. At the municipal level, Zhumadian, Zhengzhou, and Luoyang had local correlations, but there was no local correlation in other areas. Zhumadian was a high-high value aggregation area, and the overall prevalence rate was on the high side. Zhengzhou was a high-low value area, and there was an abnormal situation in which the low value area included high values. Luoyang was a low-low value area, and the overall prevalence rate was low.
Figure 13 Local spatial clustering characteristics of the prevalence of COVID-19 in China (a. Local spatial clustering characteristics of the prevalence of COVID-19 in China on February 25, 2020; b. Local spatial clustering characteristics of the prevalence of COVID-19 in Henan province, China, on February 20, 2020. This is based on the standard map production with map approval number GS (2019) No.1696 on the standard map service website of the Ministry of Natural Resources, China, and the base map has not been modified.)

5.3 Results of the geographical detectors

Spatial differentiation detection of the factors influencing the spread of the epidemic was carried out using the geographic detector method, and the results of the correlation calculations and analysis are as follows.
(1) Interactive detection results
The interactive detector mainly analyzes whether there is an interaction between the impact factors and the spread of the epidemic. The detection results (Table 5) show that the spread of the epidemic was the result of interactions between various factors. After the interaction of any two influencing factors, the influence of the factors shows bidirectional or nonlinear enhancement. For the epidemic situation in China, the interaction between the number of healthcare personnel and the number of passengers had the strongest influence, reaching 0.99, indicating that population mobility and medical conditions have significant influences on the prevalence rate. For the epidemic situation in Henan, the interactions between the number of healthcare personnel, the GDP, and the number of passengers had the strongest influence, exceeding 0.9, indicating that economic strength, population mobility, and medical conditions have important impacts on the spread of the epidemic situation.
Table 5 Results for the factor detector and interaction detector
Region Impact factors GDP Number of passengers Number of healthcare personnel Population
China GDP 0.119942
Number of passengers 0.335308↑ 0.143713
Number of healthcare personnel 0.505342↑ 0.34431↑↑ 0.238725
Population 0.999561↑ 0.510934↑ 0.34469↑↑ 0.233914
Henan
province		 GDP 0.272791
Number of passengers 0.560397↑↑ 0.387768
Number of healthcare personnel 0.905541↑ 0.952972↑ 0.230124
Population 0.762325↑ 0.751544↑ 0.394333↑↑ 0.211114

“↑” indicates nonlinear enhancement of factor interaction; and “↑↑” indicates bilinear enhancement

(2) Factor detection results
The factor detector mainly detects the influence of each influencing factor on the spread of the epidemic. The detection results are shown in the diagonal values of the two lower triangular matrices in Table 5. According to the spatial differentiation interpretation ability from high to low, the order is as follows: (1) in China, number of healthcare personnel > population > the number of passengers > GDP; and (2) in Henan, the number of passengers > GDP > the number of healthcare personnel > population. For the prevalence rate of COVID-19 in China, the number of healthcare personnel had the greatest influence (0.24), which indicates that good medical support ability and timely follow-up of supporting medical care can effectively alleviate the spatial spread of COVID-19. The influence of the population was second only to that of the number of healthcare personnel, and it also had a strong control effect on the spread of the epidemic. Effective personnel control is necessary for the prevention and control of epidemic situations. The influence of the number of passengers was relatively low, which shows that inter-provincial population mobility was only the key link in the spread of the epidemic, and it was not a direct cause of the rising prevalence rates in the various provinces, cities, and regions. The influence of the GDP was small, which indicates that the correlation between the spread of the epidemic and the economic strength of each province was low. In Henan, the number of passengers had the greatest influence (0.39) on the prevalence rate and had the strongest controlling effect on the spread of the epidemic, making it the most impact factor. In the prevention and control of the epidemic situation in Henan, first the flow of personnel should be controlled. The influence of the GDP was second. Although the correlation between the economic strength of each province and the spread of the epidemic was low, its influence on the epidemic situation in Henan increased, indicating that the economic strengths of the various cities in Henan aggravated the development of the epidemic situation in some aspects, such as population mobility. The number of healthcare personnel and the population also had significant impacts on the epidemic situation. Improving relevant medical facilities and carrying out reasonable and effective personnel control would contribute to the effective control of the epidemic situation in Henan.
(3) Risk detection results
The risk detector mainly detects the differences in the effects of the different impact factors on the prevalence rates, and the detection results are shown in Figure 14. As can be seen from Figure 14, as the value and level of the impact factors increased, the prevalence rate generally increased in a fluctuating manner. Affected by the extreme epidemic situation in Hubei, the high-risk levels of the impact factors on the prevalence rate in China tended toward the factor level corresponding to the relevant statistics in Hubei, which once again confirms that the province was the pole of the epidemic situation in China. For the epidemic situation in Henan, the high-risk levels of the population, the number of passengers, GDP, and the number of healthcare personnel were 4, 4, 5, and 3, respectively; and the corresponding values were 5.81 million people, 63.36 million people, 69,400 people, and 319.849 billion yuan, respectively. Generally speaking, the prevalence rate after Level 3 of an impact factor is significantly higher than values corresponding to Levels 1 and 2. This indicates that in epidemic prevention and control, it is necessary to effectively control subordinate personnel, control population mobility, and rationally allocate medical resources during the epidemic period.
Figure 14 The results of the risk detector
The results of the ecological detection reveal that there was no significant difference in the influences of the impacting factors on the spatial distribution of the prevalence rate.

6 Discussion and conclusions

6.1 Discussion

In this paper, from the perspective of spatio-temporal objects, the case information was organized into spatio-temporal objects with multi-granularity, and the spatial spread of the epidemic situation was mined through multi-level visual analysis of the interactions between the objects and their spatial influences. Based on the exploratory experiment on case information visualization and spatio-temporal analysis, the case information visual analysis method based on spatio-temporal objects achieved certain visualization effects and superficial analysis conclusions, but shortcomings remain that require further consideration and research.
(1) The spatio-temporal accuracy of the research data sources is limited, and the processing of the unstructured case information is relatively extensive. The case information used in this paper was obtained from the activity track of COVID-19 confirmed patients provided by various health and health commissions and Tencent. The data types are complicated. In addition, the statistical standards and information integrity vary from place to place, so it is difficult to analyze the case information semantically. When and where the cases originated or the activity track before the diagnosis includes a certain subjectivity and uncertainty. For example, in the process of epidemiological investigations, people may not remember clearly or deliberately conceal their travel information, and thus, it is difficult for the epidemiological data to accurately reflect the spatio-temporal trajectory and susceptible areas. In the process of constructing the spatio-temporal objects, there is not much multi-party verification of the effective evaluation of the accuracy of the case information, and the research results are uncertain. The unstructured text track data used in this paper were mainly processed through human-computer interactions, and the processing results are subjective and inefficient. For the big data of case information, the construction of spatio-temporal objects will involve a huge workload, and it is necessary to seek new automatic construction methods. In addition, due to the limitations of the data collection methods, the prevalence rate has not been accurately evaluated and calculated through the relationship between the infected population and the exposed population, and the calculation method used in this paper is not representative in terms of accuracy. The impact factors of the epidemic situation have not been confirmed through systematic literature review and expert scoring, and there is subjectivity involved in the factor selection. In the future, we plan to focus on the data collection and evaluation, seek an automated method of constructing the spatio-temporal objects through in-depth learning or machine learning methods, reasonably select spatial impact factors, and ensure that the research results have scientific significance and application value.
(2) The mechanisms of the spatial interactions between and spatial influences of the spatio-temporal objects are still unclear. The spread of the epidemic situation is the result of the interactions between and spatial influences of the epidemic related objects. Although the data model of the spatio-temporal objects with multi-granularity can model the behavior ability and cognitive ability of the spatio-temporal objects, mechanisms of the interactions between and spatial influence of the epidemic spatio-temporal objects are abstract and complex, and understanding them is difficult, and thus, they are still unclear. In this paper, behavioral ability modeling of the epidemic related objects was not carried out, and there is still much work to be done to explain the epidemic and to discover its spread law by using the interactions between and spatial influences of the objects. In the future, we plan to deeply analyze the mechanism of epidemic spread, the spatial behavior model of the spatio-temporal objects, the trigger mechanism, and its visualization method.
(3) The visual expression and spatio-temporal analysis functions were not integrated, and the spatial analysis ability of the objects is weak. Spatio-temporal objects not only have spatial behavior ability, but they also have the cognitive ability of independent judgment. Spatio-temporal analysis falls within the category of the self-cognitive objects. By endowing the object with the behavior ability of spatio-temporal analysis, the object can independently detect, analyze, and visualize its own spatial relationship and influence. Because the mechanism of the epidemic situation was not studied clearly, the visualization of the spatio-temporal objects and the spatial analysis in this paper are basically separate. According to the specific situation of the epidemic mechanism research and spatio-temporal object behavior ability research, we can promote research on the spatio-temporal analysis ability of epidemic spatio-temporal objects in an orderly manner, gradually improve the cognitive ability of the entities themselves, and finally, realize the independent description, independent judgment, independent learning, and independent update of epidemic spatio-temporal objects, providing an intelligent and robust analysis platform for epidemic-related research.

6.2 Conclusions

COVID-19 pandemic research has attracted increasing amounts of attention. Effective analysis and expression of its case information can reveal its spatial transmission law and provide a powerful theoretical reference and technical support for epidemic prevention and control. Multi-level, multi-dimensional visual expression of case information is helpful in determining the transmission path, pattern, and trend of the epidemic situation. Based on the official case information, this paper proposes a visual analysis method for COVID-19 case information using spatio-temporal objects with multi-granularity. Through the construction of spatio-temporal case objects, multi-level visual expression, and spatial correlation analysis, the case information can be abstracted, described, expressed, and analyzed in the form of spatio-temporal objects. First, the case information was organized into spatio-temporal objects with multi-granularity. Second, according to the multi-level visualization expression method, the 3D dynamic visualization was designed and implemented and the impact factors of the epidemic spread were selected from the aspects of spatial correlation and differentiation of the spatio-temporal objects. Finally, the rationality of this method was verified using three visualization scenarios. The method proposed in this paper has a good portability and expansion performance, and it can be used for the visual analysis of case information in other regions and to help users quickly discover the potential transmission law. Moreover, this method provides a reference for the formulation of relevant epidemic prevention and control measures.

References

Publishing order | Descend order by publishing year | Descend order by cited within
[1]
Anselin L, 1995. Local indicators of spatial association: LISA. Geographical Analysis, 27(2):93-115.

DOI

[2]
Booton R D, MacGregor L, Vass L et al., 2020. Estimating the COVID-19 epidemic trajectory and hospital capacity requirements in South West England: A mathematical modelling framework. http://dx.doi.org/10.1101/2020.06.10.20084715.

[3]
Carroll L N, Au A P, Detwiler L T et al., 2014. Visualization and analytics tools for infectious disease epidemiology: A systematic review. Journal of Biomedical Informatics, 51:287-298.

DOI PMID

[4]
Chen S, Zhang C, Ren P, 2020a. Preliminary study on exploring the trajectory of patients with COVID-19 by data mining algorithms. Chinese Journal of Medical Science Research Management, 33:E005. (in Chinese)

[5]
Chen X, Liu J, Xu L et al., 2020b. Construction of the COVID-19 epidemic cases activity knowledge graph: A case study of Zhengzhou City. Geomatics and Information Science of Wuhan University, 45(6):816-825. (in Chinese)

[6]
Chen Y, Jiang N, Zi L, 2020c. The cognition and representation of spatio-temporal behaviors of wetland based on spatio-temporal objects of multi-granularity. Bulletin of Surveying and Mapping, 2020(2):147-150. (in Chinese)

[7]
Disease Control and Prevention Bureau of China (DCPBC), 2020. COVID-19 prevention and control plan (sixth edition). https://china.huanqiu.com/article/9CaKrnKpMVo. (in Chinese)

[8]
Dominkovics P, Granell C, Pérez-Navarro A et al., 2011. Development of spatial density maps based on geoprocessing web services: Application to tuberculosis incidence in Barcelona, Spain. International Journal of Health Geographics, 10(1):62.

DOI

[9]
Geary R C, 1954. The contiguity ratio and statistical mapping. The Incorporated Statistician, 5(3):115-146.

DOI

[10]
Getis A, Ord J K, 2010. The analysis of spatial association by use of distance statistics. In: Anselin L, Rey S J (eds.). Berlin: Springer, 127-145.

[11]
Health Commission of Henan Province (HCHP), 2020. Make every effort to prevent and control New Coronavirus pneumonia epidemic situation. http://wsjkw.henan.gov.cn/zt/2020/778686/index.html. (in Chinese)

[12]
Health Commission of Liaoning Province (HCLP), 2020. Make every effort to prevent and control New Coronavirus pneumonia epidemic situation. http://wsjk.ln.gov.cn/wst_zdzt/xxgzbd/. (in Chinese)

[13]
Hua Y, Zhou C, 2017. Description frame of data model of multi-granularity spatio-temporal object for pan-spatial information system. Journal of Geo-Information Science, 19(9):1142-1149. (in Chinese)

[14]
Jiang B, You X, Li K et al., 2020. Interactive visual analysis of COVID-19 epidemic situation using geographic knowledge graph. Geomatics and Information Science of Wuhan University, 45(6):836-845. (in Chinese)

[15]
Kulldorff M, 1997. A spatial scan statistic. Communications in Statistics: Theory and Methods, 26(6):1481-1496.

DOI

[16]
Li G, Wang J, Xu T et al., 2020. Spatio-temporal evolution process and integrated measures for prevention and control of COVID-19 epidemic in China. Acta Geographica Sinica, 75(11):2475-2489. (in Chinese)

[17]
Li L M, Wang Y H, Lv J, 2008. Epidemiology: History and challenges. Chinese Journal of Disease Control & Prevention, 12(4):304-308. (in Chinese)

[18]
Liu M, Ning J, Du Y et al., 2020. Modelling the evolution trajectory of COVID-19 in Wuhan, China: Experience and suggestions. Public Health, 183:76-80.

DOI PMID

[19]
Maciejewski R, Livengood P, Rudolph S et al., 2011. A pandemic influenza modeling and visualization tool. Journal of Visual Languages and Computing, 22(4):268-278.

DOI PMID

[20]
Matheron G, 1963. Principles of geostatistics. Economic Geology, 58(8):1246-1266.

DOI

[21]
Mocnik F B, Raposo P, Feringa W et al., 2020. Epidemics and pandemics in maps: The case of COVID-19. Journal of Maps, 16(1):144-152.

DOI

[22]
Moran P A P, 1948. The interpretation of statistical maps. Journal of the Royal Statistical Society: Series B (Methodological), 10(2):243-251.

DOI

[23]
Moran P A P, 1950. Notes on continuous stochastic phenomena. Biometrika, 37(1/2):17-23.

DOI

[24]
National Health Commission of China (NHCC), 2020. Notes on continuous stochastic phenomena. http://www.nhc.gov.cn/xcs/xxgzbd/gzbd_index.shtml. (in Chinese)

[25]
People’s Government of Jilin Province (PGJP), 2020. Fight against COVID-19: Jilin is in action. http://www.jl.gov.cn/szfzt/jlzxd/. (in Chinese)

[26]
Robinson A C, Chen J, Lengerich E J et al., 2005. Combining usability techniques to design geovisualization tools for epidemiology. Cartography and Geographic Information Science, 32(4):243-255.

PMID

[27]
Tencent, 2020. Activity track of confirmed patients with COVID-19. https://xw.qq.com/act/fytrace. (in Chinese)

[28]
Wang J, Xu C, 2017. Geodetector: Principle and prospective. Acta Geographica Sinica, 72(1):116-134. (in Chinese)

[29]
Wang J F, Li X H, Christakos G et al., 2010. Geographical detectors-based health risk assessment and its application in the neural tube defects study of the Heshun Region, China. International Journal of Geographical Information Science, 24(1):107-127.

DOI

[30]
Worboys M F, 1994. A unified model for spatial and temporal information. The Computer Journal, 37(1):26-34.

DOI

[31]
Wu Q, Sun M, Cui L, 2016. A survey of the spatio-temporal data model. Advances in Earth Science, 31(10):1001-1011. (in Chinese)

[32]
Ye Y, Wang C, Zhang H et al., 2020. Spatio-temporal analysis of COVID-19 epidemic risk in Guangdong province based on population migration. Acta Geographica Sinica, 75(11):2521-2534. (in Chinese)

[33]
Yuan M, McIntosh J, 2003. GIS representation for visualizing and mining geographic dynamics. Transactions in GIS, 2(3):3-10.

DOI

[34]
Zhang J, Hua Y, Li X, 2018. The basic content and methods of multi-granularity spatio-temporal object modeling. Geomatics World, 25(2):12-16. (in Chinese)

[35]
Zhang X, Lin H, Zhu C et al., 2020. Spatiotemporal characteristics and dynamic process analysis of epidemic diagnosis time of COVID-19 at early stage in Chinese mainland. Geomatics and Information Science of Wuhan University, 45(6):791-797. (in Chinese)

Options
Outlines

/

〈 〉