A full-view scenario model for urban waterlogging response in a big data environment

2.1 The level dimension of waterlogging full-view scenario

In terms of the organizational system of urban disaster response, the hierarchical structure is mostly adopted. Because the goals, routes, and methods of disaster response in different administrative hierarchies are distinctive, the demand of different level decision makers for disaster scenario data is different [12]. For example, to know about the travel risk of residents in urban waterlogging, it is needed for city governors to analyze the main directions of their travel based on their location information. Compared with city governors, residents themselves pay attention to their own travel risk, and they need to combine the waterlogging situation of the destination and different roads to determine whether to travel and the travel routes. At present, the main subject of urban safety management is shifting or has shifted from city managers to community managers or even residents, strengthening the positive role of communities and residents in waterlogging prevention and control work [21]. On the other hand, the waterlogging scenarios at different administrative levels are not independent, but interact with each other, which is reflected as the correlation between the waterlogging scenario data at different levels.

To sum up, this article defines the level of waterlogging scenario as the administrative level of waterlogging disaster response. Combining with the existing hierarchical structure of urban waterlogging response [22], this article divides the waterlogging scenarios into city level, community level, and resident level.

First is the city level scenario. Urban managers pay attention to city level waterlogging response, and the scenario includes all scenario elements related to the overall city waterlogging response and their relationships.

Second is the community level scenario. Community managers pay attention to the waterlogging response of their communities, and the scenario data involved are used to support the waterlogging response decision-making of the communities.

The third one is the resident level scenario. Residents play their active role in waterlogging response, and the scenario data involved are used to guide their own safety behavior.

2.2 The type dimension of waterlogging full-view scenario

Different waterlogging response organizations have different scenario descriptive data and analyze disasters based on the data they have, making these organizations have different perceptions of waterlogging scenarios [23,24]. For example, meteorological departments have remote sensing and meteorological data, and their scenario descriptions focus on terrain and rainfall. Water departments have the data about drainage pipe network and the water level monitoring data of rivers and lakes, and their scenario descriptions focus on the operation of various drainage facilities. Mobile operators have mobile phone signaling data, which can generate the population distribution among different areas. When waterlogging occurs, these organizations have limited data, describing actually some scenario segments from domain perspectives, rather than the full-view scenario [25]. Therefore, when representing scenarios, these data should be integrated to help to recognize scenarios comprehensively, integrally, and systematically.

This article defines scenario type as the data category of certain waterlogging scenario. There are two kinds of common criteria for data classification, that is, data source organization (with this criterion dividing scenario data into meteorological scenario, traffic scenario, hydrological scenario, and others) or storage form (with this criterion dividing scenario data into numerical scenario, graphic and text scenario, video scenario, and others). Among them, the storage form is included in the criterion because in the context of cross-organizational integration of scenario data, the problems of multiple data sources and heterogeneity data are significant, and it is conducive to more effective scenario data management with this criterion [26].

It is worth noting that in the full-view scenario model construction, it is necessary to reduce scenario types according to actual problems. For one, the scenario data of some scenario types are not available for reasons such as data safety and privacy and the organizational coordination failing. On the other hand, if there are too many scenario types, the correlation of scenario elements will be too complex to recognize.

2.3 The granularity dimension of waterlogging full-view scenario

Scenario granularity is the detailed degree of scenario data. Obviously, the higher the detailed degree is, the smaller the granularity is, and the lower the degree of refinement, the larger the granularity [27]. Generally, scenario granularity is divided according to time or space.

The time granularity of waterlogging scenario data is reflected in the time range of scenario description. The scenario granularity is proportional to the time range. For example, for the scenario description of emergency material dispatch, large-granularity scenarios remain unchanged or change slightly in a certain time range, such as the demand resource types and resource amounts of cities [28]. Small-granularity scenarios are more real time, such as real-time road conditions, GPS, etc.

The spatial granularity of waterlogging scenario data is reflected in the spatial range of scenario description. For example, for the scenario description of a city, large-granularity scenarios correspond to larger geographical areas, such as the whole city, districts, and counties. Small-granularity scenarios are targeted at geographical grids, i.e., units with smaller geographical area. Medium-granularity scenarios are defined between the two, which are targeted as areas such as streets and communities.

Large-granularity scenario data can generally be obtained by statistical analysis or simulation with small-granularity scenario data [12]. For example, according to the waterlogging depth and area of each geographic grid in a community, the waterlogging depth of the community is calculated by taking the average. Generally, in the real scenario description, the description effects depend on the time or spatial range corresponding to the smallest granularity scenario, and the relationship between scenarios of different granularities is also clearly described [12,29,30].

2.4 The time dimension of waterlogging full-view scenario

Scenario often changes with time. On the one hand, the status of waterlogging varies at different time nodes. For example, the water depth of waterlogging is largely affected by rainstorm intensity, which has high time dependency. On the other hand, emergency departments adopt various responses for reducing disaster effects, e.g., emergency rescue and waterlogging risk communication, which also change the waterlogging scenarios [8].

To sum up, an effective scenario description should present the scenarios at different time for timely response. In this article, scenario time is proposed as a dimension of waterlogging full-view scenario, which can be defined as the time stamp of a scenario.

3.1 Structuring scenario elements

The premise of representing and recognizing scenarios is to clarify the elements of the situation [32]. Based on the existing studies [10,11], the scenario elements are defined as the disaster-related things in waterlogging disaster system, including hazards (i.e., rainstorm), disaster-bearing bodies (e.g., roads, communities, and residents), and environment (e.g., temperature and geology). Setting the domain of waterlogging scenario elements as E, which includes n scenario elements, the domain E can be expressed as follows.

where e _i and e _j represent scenario elements. There are differences in scenario elements of different scenario levels, types, and granularities. This article describes scenario elements from their features and relationships, and constructs the formal representation of scenario elements as follows.

where t indicates time stamp. The feature values of waterlogging scenarios and the relationships between the features often change with time. S _i is the feature set of waterlogging scenario element e _i . R _i is the correlation set of the features of scenario element e _i . For the feature set S _i , s is an element in it, which represents a feature of scenario element e _i . Feature s is described with its data value type d _s (e.g., number and text), its value o _s , and its value of evolution function or rule f _s (e.g., how waterlogging depth changes with time). For the correlation set R _i , r is an element in it, which represents a correlation relationship between the features. The relationship r is described with input feature set S r I , output feature set S r O , and the conversion relationship f r , which can be expressed as S r O = f r ( S r I ) .

3.2 Multi-level scenario nesting

In the field of computer science, when the attribute of a kind of data is the data with the same data structure, we call this data a nesting data [33]. Similarly, the nesting structure of waterlogging scenarios is formed when the features of the scenarios at a specific level are correlated with those at other levels. Based on existing literature and field research [12,24], there are four typical scenario nesting structures defined in this article, namely inheritance nesting, feedback nesting, aggregation nesting, and selection nesting.

1. Inheritance nesting: For a scenario element feature, if its value is correlated with the feature value(s) of one or more upper-level scenario elements, this nesting structure is called inheritance nesting, as shown in Figure 2(a). Inheritance nesting is mostly embodied in environmental analysis and risk analysis. For example, a community inherits scenario data such as waterlogging area and maximum waterlogging depth of each waterlogging point from the city, and identifies potential travel risks for the community residents. For the current level feature s h l t and upper-level features s 1 l w , s 2 l w , … , s k l w , if the inheritance nesting relationships exist, then the scenario nesting structure can be represented as s 1 l w ⊕ s 2 l w ⊕ … ⊕ s k l w → s h l t , s h l t ∈ S m l t , s 1 l w , s 2 l w , … , s k l w ∈ S n l w , where S m l t and S n l w indicate the feature set of scenario element e m l t at level s l t and the feature set of scenario element e n l w at level s l w , respectively.

2. Feedback nesting: For a scenario element feature, if its value is correlated with the feature value(s) of one or more lower-level scenario elements, this nesting structure is called feedback nesting, as shown in Figure 2(b). Feedback nesting appears when lower administrative units report key information for upper level scenario analysis. For example, residents can report waterlogging risk cases (e.g., road collapse and sinkholes) to water departments for timely response. For the current level feature s h l w and lower-level features s 1 l t , s 2 l t , . . . , s k l t , if the feedback nesting relationships exist, then the scenario nesting structure can be represented as s 1 l t ∨ s 2 l t ∨ … ∨ s k l t → s h l w , s h l w ∈ S n l w , s 1 l t , s 2 l t , … , s k l t ∈ S m l t .

3. Aggregation nesting: For a scenario element feature, if its value is correlated with the feature values of multiple lower-level scenario elements, the objective feature value is obtained by the cluster analysis of the lower-level scenario features, and the nesting structure is called aggregation nesting, as shown in Figure 2(c). Aggregation nesting mostly reflects the management of upper administrative units to lower ones. For example, urban managers cluster the rescue capabilities of communities and then rationally allocate rescue resources according to the severity of waterlogging. Another example is that community managers indirectly analyze the potential travel directions of the residents in their jurisdiction according to their work or study places. For the current-level feature s h l w and lower-level features s 1 l t , s 2 l t , . . . , s k l t , if the aggregation nesting relationships exist, then the scenario nesting structure can be represented as s 1 l t ⊗ s 2 l t ⊗ … ⊗ s k l t → s h l w , s h l w ∈ S n l w , s 1 l t , s 2 l t , … , s k l t ∈ S m l t .

   In theory, clustering should be carried out for the feature values of all the underlying scenario elements involved. However, in reality, it is often impossible to obtain the feature values of all the scenario elements, or the actual problem does not need to cluster all feature values. At this time, the corresponding rules of scenario element selection should be formulated. For example, some communities do not have their own information platforms, and it is difficult to share the rescue capability data about their communities. However, for communities with similar basic parameters, such as the number of residents and building area, rescue capability is often similar, so only some representative communities need to be selected for clustering.

4. Selection nesting: In the selection nesting structure, for a scenario element feature, there are many ways of value assignment, each of which corresponds to a nesting structure (either inheritance nesting or aggregation nesting), and it is needed to select one of them to determine the value of objective scenario feature according to management requirements, as shown in Figure 2(d). Selection nesting can appear at any scenario level, including city, community, or resident level. For example, when analyzing the travel direction of residents, urban managers can not only get the data with the location data of social media platforms (resident level) through data clustering but also do that by clustering the travel direction data of the communities (community level).

3.3 Modeling of full-view waterlogging scenario

Based on scenario nesting, the relationship among waterlogging scenario elements at different levels can be established. Combining with the correlation relationship of scenarios of different granularities and those of different types (identified in advance), the full-view waterlogging scenario model can be constructed. The form of the full-view model is as follows.

where SL = { s l k | k = 1 , 2 , … , n l } indicates scenario level set, and n l is the number of levels. R SL = { R t . m SL | t ≠ m } indicates the nesting relationship set of scenario elements at different levels, t , m ∈ { 1 , 2 , … , n l } . SS = { s s k | k = 1 , 2 , … , n s } indicates scenario type set, and n s is the number of types. R SS = { R t . m SS | t ≠ m } indicates the nesting relationship set of scenario elements of different types, t , m ∈ { 1 , 2 , … , n s } . SG = { s g k | k = 1 , 2 , … , n g } indicates scenario granularity set, and n g is the number of granularities. R SG = { R t . m SG | t ≠ m } indicates the nesting relationship set of scenario elements of different types, t , m ∈ { 1 , 2 , … , n g } . ST = { s t k | k = 1 , 2 , … , n t } indicates scenario time set, and n t is the number of time nodes.

This article focuses on the correlation relationships of waterlogging scenario elements at different levels, and the correlations of scenario elements of different granularities and those of different types at the same scenario level are supplemented to the relationship set of the level. Then the full-view model is simplified as follows:

For ∀ s l ˜ t , s l ˜ m ∈ SL ˜ , t ≠ m , t , m ∈ { city , community , resident } ,

where R ˜ k t , m is the nesting relationship set when nesting the scenarios of level l _t to the scenario element e ˜ k l m of level e ˜ k l m , and | E ˜ l m | indicates the number of scenario elements at level l _m .

This equation illustrates that there are n features of scenario element e ˜ k l m have nesting relationships with the scenario element features at level t.

5.1 Research objects and data source

From June 30 to July 6, 2016, Wuhan suffered from heavy rainstorm and waterlogging. There were 187 sections of roads waterlogged in the whole city. East Lake High-tech Development Zone was a serious affected area. As many as 26 sections were impassable due to waterlogging. The south section of Guanggu Avenue almost ran into a river. The severe waterlogging situation aroused great attention of the management committee and widespread concern of the society. Focusing on the maximum rainfall time on July 6 (i.e., 8 a.m. when the hourly rainfall reaches 61.3 mm) as the time node, this article chooses representative research objects from three waterlogging response levels (city, community, and resident), and describes in detail the construction of full-view scenario and its specific generation process. The selected research objects of each level are shown in Figure 3.

Figure 3

The selected research objects of three levels.

Figure 4

The scenario generation result at city level (risk of travel destination). (a) The risk of different regions, (b) people distribution in the following several hours.

5.2 Scenario element representation of different levels

Considering the information demands of decision makers at different levels for recognizing waterlogging scenarios, the scenario types and scenario granularities are reasonably selected to construct waterlogging scenarios at three levels, that is, city, community, and resident. The elements and their features of waterlogging scenarios at all levels are listed in Table 1. There are correlations between the features of scenario elements at the same level. For example, the discharge of each geographic grid is determined by the diameter of the drainage pipe and the elevation. The specific correlations are shown in Table 2. It is noted that not all scenario feature values can be obtained directly, such as the travel direction of the residents at the city level and the number of waterlogging grids at the community level. The assignment of these values depend on the feature values of waterlogging scenario elements at other levels, and it is needed to build scenario nesting relationships to obtain the values.

5.3 Construction of full-view model

Referring to the big data analysis modes of existing literature [25,26], combining with the field investigation of East Lake High-tech Development Zone, we extract the correlations among the features of different levels of scenarios, establish the nesting patterns of cross-level waterlogging scenarios, and complete the construction of the full-view waterlogging scenario model. Some instances of the nesting patterns are shown in Table 3.

5.4 Iterative generation of full-view scenario

According to the multi-level nesting correlations of waterlogging scenarios in the last section, following the iteration algorithm of full-view scenario generation given in Section 3.4, the value assignment of waterlogging scenario features at different levels can be realized. The feature values of scenario elements at city level are exported to the big data platform of East Lake High-tech Development Zone, and the feature values of the scenario elements at community level are exported to the information system of Jiufeng New Area Community. The feature values of the scenario elements at resident level are transferred to the mobile terminals. Finally, the full-view waterlogging scenario for the three level decision makers is formed. To visualize the results of full-view scenario generation, the full-view scenario is expressed through the geographic information platform. The granularity of waterlogging scenario expression is adjusted reasonably and the final results are shown in Figures 4–8.

Figure 5

The scenario generation result at city level (risk of travel routes).

Figure 6

The scenario generation result at community level with translations (real-time waterlogging risk cases).

Figure 7

The scenario generation result at community level (waterlogging situation and travel destination risk). (a) Waterlogging situation of the community, (b) travel destination risk of the community residents.

Figure 8

The scenario generation result at resident level.

5.5 Result analysis and validation

With the help of virtual geographic information system provided by ArcGIS, the use case integrates waterlogging scenario data of different types and granularities, constructs a full-view waterlogging model for three levels, that is, city (Wuhan East Lake High-tech Development Zone), community (Jiufeng New Area Community), and resident, and realizes the instantiation of the full-view model with the iteration algorithm of full-view scenario generation. This section analyzes the results of full-view scenario generation.

1. The scenario generation result at city level: We analyze the result in terms of the risk of travel destination and the risk of travel routes. To analyze the travel destination risk, the risk of different regions shown in Figure 4(a) and the people distribution in the following several hours shown in Figure 4(b) are generated. Comparing the two figures, we can find that some places have a large number of gathered people and serious waterlogging (such as Guanggu Square, Luoyu Road, Central-South University for Nationalities, Guanggu Avenue, and surrounding areas) and thus exist higher risk. It is necessary to give these information back to the public in time and suggest that they adjust their travel plans. To analyze the travel route risk, the traffic flow data of each road are generated. Compared with the risk of different regions shown in Figure 5, we can find that some roads have heavy traffic and serious waterlogging (such as the Luoyu Road, the part of National Avenue close to the Luoyu Road, the part of National Avenue close to the Third City Loop, the part of Gaoxin Avenue close to the Third City Loop, and the part of Guanggu Avenue close to the intersection with Luoyu Road). For these roads, the traffic control should be strengthened. Those city-level results are highly consistent with the governmental disaster survey reports.

2. The scenario generation result at community level: At the beginning of 2018, the Wuhan Government built a city-wide event report platform that is called “Wuhan Micro Neighborhood.” A sample page of the platform is shown in Figure 6. Through this platform, the residents of Jiufeng New Area Community can report real-time waterlogging risk cases (e.g., road collapse and sinkholes) to the community for timely response. According to the survey in this community, residents are really helpful in city waterlogging scenario recognition and emergency response. This real-time case report mechanism has also greatly reduced the demands for emergency workers. Up to now, this platform has accumulated more than 12,000 case records, involving secondary disasters, serious water inundation, infrastructure damage, and trapped residents. In terms of waterlogging situation, the generation result is shown in Figure 7(a). We can find that there are three intersections with serious waterlogging, and two of them have brought about some road sections being blocked. This result has been proved to be consistent with the historical situation through practical surveys. To analyze travel risk of residents, the generation result is shown in Figure 7(b). We can find that travel directions are mainly around Guanggu Square and Xiyuan Park, but there is serious waterlogging in the two places. Therefore, it is necessary to communicate the potential risks to the residents in time. Although this result is hard to be validated, the risk analysis method has been highly praised by the community managers, and relevant measures have been adopted to carry out the destination risk analysis and communication.

3. The scenario generation result at resident level: This article considers the residents who have a car, need to travel to work, and need to drive to destinations far from the community as an example to analyze the scenario generation result. According to the demographic data and historical cases of Jiufeng New Area Community, the number of this kind of residents is the largest, and the analysis can be highly representative. According to the result of scenario generation shown in Figure 8, if such residents drive according to the shortest route (route 1), the waterlogging along the route is serious and the risk relatively pretty high. Compared with route 1, although route 2 is longer, the waterlogging is not serious, and the mobile terminals will recommend residents to choose this route for travel. This route planning method has also been adopted by the Jiufeng New Area Community, and is expected to be integrated to the “Wuhan Micro Neighborhood” platform.