Geo-ontology model construction for urban flood disaster: a case study of community scale

doi:10.21203/rs.3.rs-1596515/v1

Urban flooding has become one of the most common natural disasters worldwide due to recent global climate changes. Therefore, disaster simulation and forecasting have become important “non-engineering measures” to manage urban flooding and reduce the effective emergency response time. In this study, an urban flooding-disaster geographic-ontology model (UFD-GOM) was constructed with rainfall, surface, and drainage ontologies as its core, and a Bayesian network model was built after the classes, relationships, and attributes were rationally defined. The Caochangmen Community, Nanjing, China, case-study results showed that the difference between the number of inferred/predicted and actual observed disaster-class grid cells was small. The spatial distribution of the flood-prone monitoring points and the inferred/predicted regional disaster-occurrence intensity were well matched, with the average accuracy being 87.18%, verifying and validating the scientific basis and practicality of the “triple-core” UFD-GOM constructed in this study.

Urban flood disaster

Geographic ontology

Bayesian network

Caochangmen

Nanjing

China

Ontology model

UFD-GOM

Due to recent global climate changes, flooding has become one of the most common natural disasters worldwide (Nigussie et al. 2019; Deepak et al. 2020; Karmegam et al. 2021). Additionally, the conflict between anthropomorphic activities and land use has intensified due to increased urbanization, leading to more frequent urban flash floods that can threaten human life and property (Li et al. 2014). Urban flooding is a natural disaster with strong transients, high frequency, and serious hazards; it has gradually become an important area of research (Zeng et al. 2020). The scientific community needs a method to describe it in a normative manner to more deeply and accurately understand and present it as a geographical event.

Disaster simulation and forecasting have become important “non-engineering measures” to manage urban flooding and reduce the effective emergency response time (Andimuthu et al. 2019). However, since most current studies do not consider urban flooding a unified geographical complex, problems such as fragmentation of scales, scattered methods, and cognitive biases have emerged. Several ontology construction methods using Protégé software, Bayesian networks, and data discovery have been proposed over recent years during the maturation of ontology construction (Cui et al. 2019; Ma et al. 2019). However, most of them are based on macro-static domains that do not fully reflect the underlying cognitive and essential characteristics of the event occurrence processes.

Therefore, this study introduced ontology to the field of urban flooding disasters, designed a construction framework and process for an urban flooding-disaster geographic-ontology model (UFD-GOM), and proposed a visual representation of a UFD-GOM based on web ontology language (OWL) and ontology construction tools. The UFD-GOM developed in this study promotes mutual understanding and communication of urban flood disaster information among different groups and corrects individual cognitive biases arising in understanding its geography, providing a reference value for urban disaster prevention, water resources management, and decision-making support in water circulation.

Location and Research Data

The eastern side of Caochangmen Street, was selected as the study area (Fig. 1). The coordinates of the study area are 32.060°-32.070° N; 118.747°-118.757° E, in Gulou District, Nanjing, Jiangsu Province, China, with an area of approximately 683 644 m². The District has relatively abundant rainfall, with an average annual precipitation of ~1106 mm, and the overall topography is gentle. Because it is in the central business district, the nearby residential communities and schools were densely distributed, and the buildings and hardened road surface areas were saturated; therefore, urban flash-flooding often occurred. An analysis of the drainage network distribution characteristics indicated that the area formed a relatively independent catchment area that was not easily affected by external runoff. Additionally, two sets of road-ponding monitoring and warning devices were distributed in the area, conducive to case-study inference verification.

This case study used data provided by the local government bureau, including flood-prone-point monitoring and rainfall-station observation data from 2016 to 2019; 5 × 5 m digital elevation model, land use, road building, and other basic geographic data; and water conservation facility, and drainage network data. The data were divided into three major categories: hydro-meteorological, basic geographic, and water conservancy and drainage and pre-processed for use in building a case analysis for the UFD-GOM construction. The rainfall observation data from 2016 to 2019 were used as a set of historical case data, and a typical urban flooding geographic scenario that occurred on July 17, 2020, was selected as a reference case study.

UFD-GOM Framework

Urban flooding disasters include rainfall, surface, and drainage systems (Fig. 2); each system had a distinct hierarchy and sufficient elements, based on which the UFD-GOM with rainfall, surface and drainage ontologies as its core was built¾as a “triple-core” ontology model. Therefore, it was critical to clarify the system framework before constructing the UFD-GOM. Rainfall ontology can create a response in surface ontology and, in turn, surface ontology can create a response in the drainage system, forming a logical spatial continuity. The hierarchical structure of the three urban flooding ontology scenarios was subdivided to create a framework for constructing the UFD-GOM (Fig. 2.).

Depending on the application scenarios, the ontologies were divided into task ontologies that focused on the application of specific tasks (e.g., emergency disaster relief tasks) and domain ontologies that focused on the analysis of specific domains (e.g., disaster processes) (Zhong et al. 2017). UFD-GOM construction methods differ between ontologies; domain ontology construction methods were used in this study. Additionally, based on previous research (Chang et al. 2004), this study proposed six principles for UFD-GOM construction as follows:

(1) Definitional clarity¾the concept of word selection should be clear, the definition should be objective and unambiguous, and standard terminology should be used whenever possible.

(2) Logical consistency¾the ontology model should be logically smooth, and the meaning of the inference results consistent.

(3) Semantic constraint¾the semantic distance maintained between concepts at each level should be minimal, and information should be conveyed with the least number of words, without semantic overlap.

(4) Geospatiality¾the constructed ontology should be based on a specific geographical location, and the data should have a suitable coordinate system.

(5) Human-geographical integration¾the ontology should be constructed considering the influence of anthropomorphic activities as much as possible.

(6) Extensibility¾suitable space should be reserved for future needs or applications that can be directly extended to existing concepts.

The construction of the UFD-GOM was divided into three main elements: attributes, classes, and relationships, in which the attributes were descriptions of classes that can constrain relationships, and relationships were bridges between connected classes. The key to constructing the UFD-GOM (Fig. 3) was to first define and organize the classes within the geographic scenario. Classes, often referred to as concepts, are the core of an ontology; they refer to a collection of elements with similar properties and are normative and clear descriptions of the knowledge of the domain they cover. Being the top class, urban flooding (Fig. 2) was used as the starting point to define the rainfall, surface, and drainage classes in the next layer, and these, in turn, to define the natural element, human element, rainfall information, and drainage facility classes, in the next layer, and so on. Because urban flooding ontology involves complex anthropomorphic factors, it is necessary to define it using expert evaluation and narrative lists in this field to obtain comprehensive and distinct classes.

Relationships play an important role in urban flooding hazard ontology as links between layers and the elements within layers. The relationships in urban flooding disaster ontology include five primary categories: temporal, spatial, topological, and semantic relationships, and element interrelationships. Temporal relationships refer to the momentary state of the disaster event and the process of dynamic evolution associated with the relationship; spatial relations indicate the spatial geographic relationship between elements within the disaster scene, and together with temporal relationships formed the basis of the relationship descriptions; elemental interrelationships describe the qualitative or quantitative relationships between elements in the disaster scene in time or space (Lv et al.2017), and were the core of the relationship definitions. Topological relationships, also termed geometric relationships, refer to static or dynamic structural relationships such as adjacency, connectivity, and inclusivity between elements in a disaster scene. Semantic relationships represent the relationships between human cognition and the description of the geographic features of a disaster event or scene and are key to realizing the relationship descriptions.

Table 1

Attribute Concept Definitions

Attribute	Data type	Interpretation of attributes
WaterloggingID	long	Event unique identifier
WaterloggingName	string	Event name
RainfallIntensity	double	Rainfall intensity
StartTime	datatime	Rainfall start time
EndTime	datatime	Rainfall end time
Elevation	float	Elevation value
BridgeName	string	Bridge name
AffectedPeopleNumber	long	The number of victims
ReliefWorkersNumber	long	The number of rescuers
PlaceName	string	Location of the event
PlaceLongitude	double	Longitude
PlaceLatitude	double	Latitude
RainwellNumber	long	The number of rainwater well

The attributes are quantitative descriptions of each class within the urban flooding scenario and were divided into three main parts: class name, data type, and semantic information (Table 1) The name of the class is the name of each element, such as BridgeName, RainwellID, and AffectedPeopleNumber; the data type includes long, string, float, double and datatime; and the semantic information is the interpretation of the class. For example, the semantic information for StartTime is the rainfall start-time and can be interpreted in computer language to facilitate human-computer communication.

Bayesian Reasoning Model

Ontology inference is an application extension of ontology and an important means of verifying the constructed ontology (Li et al. 2014). The current inference methods for implementing ontologies primarily include neural networks, Bayesian networks, word graph matching, and concept similarity. Many researchers have favored Bayesian networks as a flexible probabilistic graphical model with rigorous mathematical expression logic and causal inference (Bruce et al. 2019; Nam et al. 2020). It has been favored by many researchers (Liu et al. 2014; Shi et al. 2020). The introduction of Bayesian networks to the construction of UFD-GOMs enables the formal representation and application of ontologies and contains three modules: variable definition, structure learning, and parameter learning.

Variable definition refers to the assignment of values to relevant concepts and causal logic in the UFD-GOM as a means of inferring uncertain variables; therefore, the range of values of the variables should be determined. The variables in the UFD-GOM were plotted as a directed graph, in which the likely influencing factors were screened as node variables. Structure learning is the core of the Bayesian network UFD-GOM construction; its main purpose is to determine the topology, which directly reflects the dependency or independence among variables, and facilitate screening and error correction. Parameter learning refers to calculating the posterior probability distribution of parameters based on their prior probability distribution and data sample sets. Because Bayesian networks tend to deal with discrete probability data, the data should first be discretized to calculate the conditional distribution of each node. Based on the Bayesian network topology structure, the parameters were learned by training the sample dataset, and the conditional probability distribution among the variables was obtained using the maximum likelihood estimation algorithm to determine the accuracy of the inference. The specific calculation formula is as follows.

Suppose the sample set D= {\({x}_{1},{x}_{2},{x}_{3}, \dots ,{x}_{n}\)} ,and the sampl, s were independent of each other, so:

\(P\left({A}_{i}\right)=\sum P\left(\text{B}\right)P\left({A}_{i}\|B\right)\)	(1)
\(P({A}_{1},{A}_{2},{A}_{3}, \dots ,{A}_{n})={\prod }_{i=1}^{n}P\left({A}_{i}\right\|B)\)			(2)
We obtain the Bayesian formula:

\(P\left({A}_{i}|B\right)=\frac{P\left({A}_{i}\right)P\left(B|{A}_{i}\right)}{\sum _{j=1}^{n}P\left({A}_{j}\right)P\left(B|{A}_{j}\right)}\)

(3)

where \(P\left(B\right)\)denotes the prior probability of event B, that is, the normalization constant; \(P\left({A}_{i}\right)\)denotes the nodal probability of event A, which is independent of event B; \(P({A}_{1},{A}_{2},{A}_{3}, \dots ,{A}_{n})\)denotes the joint probability of event A. \(P\left({A}_{i}|B\right)\)denotes the conditional probability of A after the known occurrence of B, that is, the value of B and is called the posterior probability of A; \(P\left(B|{A}_{j}\right)\)denotes the conditional probability of B after the known occurrence of A.

UFD-GOM Visualization

The visualization of urban flooding requires the abstraction of elements in the disaster scene according to the spatial geometric characteristics of geographic entities. The elements are divided into three types: point, line, and surface. For example, rainwater wells can be abstracted as point elements, drainage networks as line elements, and artificial greenery as surface elements. In OWL language, the spatial geometric types of geographic elements can be described by defining PrimitiveGeometry and PointGeometry, LineGeometry, and SurfaceGeometry to represent the point, line, and surface geometry element classes, respectively; the three types are interrelated. Additionally, as urban flooding involves topology, distance, and orientation relationships, they should also be visualized and displayed.

The embedding of UFD-GOMs is the process of populating the ontology with specific instances that can be determined using classification mapping of geographic entities (Fig. 4). The instances were constructed by first determining the class to which they belong and then adding instances of that class to complement and complete the information on the attributes and constraints that make up the class. For example, the location name, longitude, and latitude instances where flooding occurred were defined for the surface class. The corresponding attributes and constraints were added according to the characteristics of the flood-prone locations.

Additionally, the construction of a UFD-GOM requires the use of developmental tools. After a comprehensive analysis and comparison of the current major ontology development tools, we selected the feature-rich Protégé software for ontology development and expression. Protégé offers intuitive graphical visualization and the advantages of ease of operation and support for modular design and OWL language (Daniel et al. 2007). Based on these attributes, a three-core ontology model for rainfall, surface, and drainage was constructed with urban flooding as the top-level class. The subclasses, class hierarchies, internal relationships and their properties were defined separately using Protégé and corresponding instances were added to each subclass (Fig. 5.).

Reasoning and Verification

A Bayesian network architecture based on UFD-GOM was built to estimate the sample data set using variable definition, structure and parameter learning. It was also used to calculate and predict the probability of urban flooding occurrence during that period, to infer the probability distribution of disaster occurrence, and classify the level of occurrence intensity as low, medium, or high according to the predicted probability of urban flooding intensity. The predicted probability was less than 30% for low-level intensity, 30%-60% for medium-level intensity, and greater than 60% for high-level intensity. After several repetitions of the experiment, the results of one of them were randomly selected (Fig. 6.).

The left panel shows the results obtained using UFD-GOM Bayesian inference prediction in this study, and the right panel shows the actual observation results, revealing that the predicted and actual results were remarkably similar. For an in-depth analysis of the inference results, the study area was divided into 5 × 5 m grids with a combined total of 19 103 grid cells that were used to validate the differences between the inference and actual results (Table 2). The inference results did not differ significantly from the actual results; the average accuracy was 87.18%, validating the effectiveness of the UFD-GOM inference model constructed in this study.

Table 2

Differences between UFD-GOM experimental and actual results
Level	Low	Medium	High
Experimental results	14121	4749	233
Actual results	13073	5822	208
Accuracy (%)	91.99	81.57	87.99

Urban stormwater flooding disasters involve many disciplines, including hydrology, meteorology, ecology, and disaster management. However, these disciplines are all within the scope of research geography and satisfy the regular features of geography. If the conditions and parameters of urban flooding can be rationally summarized, then the propensity of another similar spatial region to flood formation can be inferred or predicted from similar geographical scenarios; corroborating the third law of geography (Zhu et al. 2018).

The UFD-GOM constructed in this study was based on a trinuclear model consisting of rainfall, surface, and drainage ontologies. The classes, relationships, and attributes were defined, and the construction and formal representation of the UFD-GOM was realized using OWL language and ontology construction tools. A Bayesian network model was built to realize the UFD-GOM inference, and the results showed that the difference between the inference prediction and the actual observed number of disaster-level grid cells was sufficiently small. The average accuracy of the inference results in the Nanjing, Caochangmen case-study reached 87.18%. The spatial distribution of the flood-prone monitoring points and the regional disaster occurrence intensity predicted by inference were well matched, verifying and validating the scientific basis and practicality of the UFD-GOM proposed in this study.

UFD-GOM：urban flooding-disaster geographic-ontology model; OWL: Web Ontology Language

Availability of data and materials

The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.

Competing interests

The authors declare that they have no competing interests.

Funding

The study was supported by National Natural Science Foundation of China (Nos: 42071364 and 41871299), and the Jiangsu Province Graduate Research and Practice Innovation program (Nos: KYCX21_1351)

Authors' contributions

ZZ was responsible for most of the work on this paper, including data collection, data processing, analysis of results and writing of the manuscript. ZG and QY were responsible for experimental design and field comparison of results. SZ was responsible for review and validation of results. All authors read and approved the final manuscript.

Acknowledgements

The completion of this study was indebted to Dr. Qiqi Yang, Dr. Shu-Liang Zhang and Ms. Zhi-Yan Gong for their support and guidance, and to Dr. Qiang Dai for his valuable advice. On the other hand, the authors would like to thank Editage (www.editage.cn) for English language editing.

Authors' details

¹Key Laboratory of Virtual Geographic Environment for the Ministry of Education, Nanjing Normal University, Nanjing, 210023, China. ²Jiangsu Center for Collaborative Innovation in Geographical Information Resource Development and Application, Nanjing, 210023, China

Andimuthu R, Kandasamy P, Mudgal BV, Jeganathan A, Balu A, Sankar G (2019) Performance of urban storm drainage network under changing climate scenarios: Flood mitigation in Indian coastal city. Scientific Reports 9:1-10. https://doi.org/10.1038/s41598-019-43859-3
Bruce GM, Trent DP (2019) Advances in Bayesian network modelling: Integration of modelling technologies. Environmental modelling & software 111: 386-393. https://doi.org/10.1016/j.envsoft.2018.09.016
Chang C (2004) Construction and Conversion of Ontology in Agricultural Information Management. Chinese Academy of Agricultural Sciences, Beijing
Cui K, Jiang Y, Li Y, Pfoser D (2019) A vocabulary recommendation method for spatiotemporal data discovery based on Bayesian network and ontologies. Big Earth Data 3: 220-231. https://doi.org/10.1080/20964471.2019.1652431
Daniel LR, Natalya FN, Mark A, Musen MA (2007) Protege: a tool for managing and using terminology in radiology applications. Journal of Digital Imaging 20: 34-46. https://doi.org/10.1007/s10278-007-9065-0
Deepak S, Rajan G, Jairaj PG (2020) Geospatial approach for assessment of vulnerability to flood in local self governments. Geoenvironmental Disaster 7: 1-19. https://doi.org/10.1186/s40677-020-00172-w
Karmegam D, Ramamoorthy S, Mappillairaju B (2021) Near real time flood inundation mapping using social media data as an information source: a case study of 2015 Chennai flood. Geoenvironmental Disasters 8: 1-11. https://doi.org/10.1186/s40677-021-00195-x
Li B (2017) Research on Geo-ontology construction on oriented to Geographical event. Wuhan University, Wuhan
Li MJ, Wang M, Shi PJ (2014) Temporal-spatial distribution of rainstorm-flood disastersin Hunan, China and its affecting factors. Journal of Beijing Normal University (Natural Science) 50: 429-434.
Liu ZJ, Guo SL, Li TY, Hong XJ (2014) Comparative study of Bayesian probabilistic flood forecasting models. Journal of Hydraulic Engineering 45: 1019-1028. https://doi.org/10.13243/j.cnki.slxb.2014.09.002
Lv GN, Yuan LW, Yu ZY (2017) Surveying and Mapping Geographical lnformation from the Perspective of Geography. Acta Geodaetica et Cartographica Sinica 46: 1549-1556. https://doi.org/10.11947/j.AGCS.2017.20170338
Ma MM, Chen CH (2019) Construction method on traffic geography ontology based on Protégé. Beijing Surveying and Mapping 33: 1566-1570. https://doi.org/10.19580/j.cnki.1007-3000.2019.12.030
Nam K, Wang F (2020) An extreme rainfall-induced landslide susceptibility assessment using autoencoder combined with random forest in Shimane Prefecture, Japan. Geoenvironmental Disasters 2020 7: 1-16. https://doi.org/10.1186/s40677-020-0143-7
Nigussie TA, Altunkaynak A (2019) Modeling the effect of urbanization on flood risk in Ayamama Watershed, Istanbul, Turkey, using the MIKE 21 FM model. Natural Hazards 99: 1031-1047. https://doi.org/10.1007/s11069-019-03794-y
Shi HY, Luo GP, Zheng HW, Chen C B, Bai J, Liu T (2020) Water use analysis of Syr Darya river basin:Based on “Water-Energy-Food-Ecology” nexus and Bayesian network. Acta Geographica Sinica 75: 1036-1052. https://doi.org/10.11821/dlxb202005011
Zeng J, Wang QW, Guo HS (2020) Knowledge map analysis and progress review of international research on flood disaster risk. Journal of Catastrophology 35: 127-135. https://doi.org/10.3969/j.issn.1000-811X.2020.02.024
Zhong S, Fang Z, Zhu M, Huang, Q (2017) A Geo-Ontology-based approach to decision-making in emergency management of meteorological disasters. Natural hazards 89: 531-554. https://doi.org/10.1007/s11069-017-2979-z
Zhu AX, Lu G, Liu J, Qin C.Z, Zhou C (2018) Spatial prediction based on Third Law of Geography. Annals of GIS 24: 225-240. https://doi.org/10.1080/19475683.2018.1534890

No competing interests reported.

Geo-ontology model construction for urban flood disaster: a case study of community scale

Status:

Version 1

Abstract

Figures

Introduction

Materials And Methods

Location and Research Data

UFD-GOM Framework

Results And Discussion

UFD-GOM Visualization

Reasoning and Verification

Conclusion

Abbreviations

Declarations

References

Additional Declarations

Status:

Version 1