Three-dimensional virtual representation for the whole process of dam-break floods from a geospatial storytelling perspective

ABSTRACT The objective of disaster scenes is to share location-based risk information to a large audience in an effective and intuitive way. However, current studies on three-dimensional (3D) representation for dam-break floods have the following limitations: (1) they are lacking a reasonable logic to organize the whole process of dam-break floods, (2) they present information in a way that cannot be easily understood by laypersons. Geospatial storytelling helps to create exciting experiences and to explain complex relationships of geospatial phenomena. This article proposes a three-dimensional virtual representation method for the whole process of dam-break floods from a geospatial storytelling perspective. The creation of a storyline and a storytelling-oriented representation of dam-break floods are discussed in detail. Finally, a prototype system based on WebGL is developed to conduct an experiment analysis. The results of the experiment show that the proposed method can effectively support 3D representation of the spatiotemporal process of dam-break floods. Furthermore, the statistical results indicate that the storytelling is useful for assisting participants in understanding the occurrence and development of dam-break floods, and is applicable to the popularization of disaster science for the general public.


Introduction
The occurrence and development of disaster events are not independent of each other, and the successive disasters are linked in time and space Shi et al. 2020). For example, strong earthquakes can not only cause damage to the structures but also generate some secondary geological disasters in mountainous areas (Bergman et al. 2014;. As a typical secondary disaster, the dam-break flood is a geographical process induced by earthquakes, rainstorms, and landslides. It is characterized by sudden occurrence, rapid expansion, high velocity, and severe destruction, which poses a great threat to human lives and property for downstream areas (Y. Li et al. 2013;Xu, Yang, and Zhou 2017;Wang et al. 2021).
It is clear that structural measures (e.g. slope stabilization, flood control dams, etc.) are the most direct ways to mitigate flood hazards, but they are costly and over-reliance on them tends to give the public illusion that they are safe and secure (Meyer, Priest, and Kuhlicke 2012;Islam and Ryan 2015;Vidmar et al. 2019), leading to reduced risk awareness and reluctance to evacuate, which is known as the 'optimistic bias' and 'safe illusion' (Spittal et al. 2005;Day 2011;Evans et al. 2014). Optimistic bias is the belief that people always believe that negative events are less likely to happen to them than to others (W. Li et al. "An augmented representation method" 2021), while the safe illusion is that people are protected by levees or dams are more likely to ignore risks (Kundzewicz et al. 2018).
Non-structural mitigation measures are measures not involving physical construction, which use knowledge and practice to reduce flood risks (Islam and Ryan 2015). As an important component of non-structural mitigation measures, agent-based modeling (ABM) is becoming increasingly popular to address flood-related risk challenges in recent years, which has been applied in flood hydrological modeling, loss assessment, and risk management (Zhuo and Han 2020;Anshuka et al. 2022;Koo et al. 2020). However, most studies have focused on improving the accuracy of hydrological modeling or risk assessment at the technical and algorithmic levels, while neglecting people's awareness of and response to flood risk. From the author's perspective, the results of flood simulation and loss assessment are not only for decision making but also as an important medium of communication between the population at risk (PAR), stakeholders, and authorities (Bilali et al. 2022). In this context, knowledge communication related to floods plays a vital role in the aspect of increasing people's risk awareness (W. Li et al. "An augmented representation method" 2021; Zhuo and Han 2020). Moreover, as a complex geographical phenomenon, dam-break floods have strong spatiotemporal characteristics, and location-based risk communication plays a decisive role in helping the public understand their occurrence and development (Dransch, Rotzoll, and Poser 2010).
Disaster knowledge communication is a process of encoding and decoding disaster hazard or risk information. Professionals organize dam-break flood information by coding, and the public decodes to gain mitigation knowledge. However, with traditional media (e.g. cartoon brochures, books, newspapers, etc.) it is difficult to visualize flood hazard information vividly (W. Li et al. "An augmented representation method" 2021). Visual representations of animation or video not only enhance reading and perception, but also increase people's understanding of the dynamic process of disasters. Therefore, the combination of video and location-based risk communication can be an important medium for disaster science education.
Geographic information system (GIS) has been a useful tool for mapping flood risks and emergency management in most parts of the world, and many studies have proved that flood hazard/risk maps are efficient and valuable in disaster knowledge communication (Gaillard and Pangilinan 2010;Meyer, Priest, and Kuhlicke 2012;Ntajal et al. 2017;Z. Li et al. 2018;Henstra, Minano, and Thistlethwaite 2019). In 1998, US Vice-President AI Gore proposed a vision of 'Digital Earth' as a multi-resolution, three-dimensional (3D) representation of the planet, the global features of digital earth offer potential for environmental applications such as climate change, disaster management, air quality management, sustainable development Goodchild et al. 2012;C. Zhang et al. 2015). In terms of disaster management, digital earth allows the planning of collective rescue or mitigation measures after natural disasters and improved understanding of the interactions between human artificial spaces and natural environments Fu et al. 2021). As a core component of digital earth, 3D scenes provide a more intuitive option for understanding dam-break floods, it supports vividly display the details of flood information from an all-around perspective, which leaps forward in risk knowledge communication (Lai et al. 2011;Liao et al. 2017;Macchione et al. 2019;Chen and Lin 2018;W. Li et al. "A Rapid 3D Reproduction System" 2021). However, the current studies on dam-break flood maps or 3D representations have the following limitations that need to be addressed (Hagemeier-Klose and Wagner 2009;Meyer, Priest, and Kuhlicke 2012;Macchione et al. 2019): (1) They lack a reasonable logic structure to organize the whole process of dam-break floods; (2) They present the information in a way that cannot be easily understood by laypersons.
In this context, this article proposes a three-dimensional virtual representation method for dambreak floods from a geospatial storytelling perspective. The authors attempt to combine geospatial storytelling with 3D visualization to present dam-break floods. The creation of storylines and storytelling-oriented representations of dam-break floods are discussed in detail. We aim to tell people a dam-break flood story rather than providing a professional 3D scene, thus enhancing the public's understanding and risk perception of dam-break floods at a deeper level. The remainder of this article is organized as follows: Section 2.1 discusses the basic ideas of geospatial storytelling. Section 2.2 provides an overall framework of the proposed method. Sections 2.3 and 2.4 introduce the storyline. Section 2.5 introduces the storytelling-oriented representation and public cognition. Section 3 gives insight into the development of a prototype system and the experiment analysis. Sections 4 and 5 present the discussion and conclusion, respectively.

Basic ideas of geospatial storytelling
From the perspective of people's cognition, it is easier to understand and remember information when it is packaged into a well-structured story (Gershon and Page 2001;Segel and Heer 2010;Paelke and Elias 2007;Çöltekin et al. 2020). Therefore, integrating the disaster information of dam-break floods to be transmitted into a story is an effective way for knowledge communication.
There are many terms for storytelling, such as story maps, fictional cartography, narrative atlas, and geospatial storytelling. They are often not explicitly defined in the literature, but this demonstrates a growing interest in the relationship between maps and narratives (Kwan and Ding 2008;Caquard 2013;Roth 2021). In this article, geospatial storytelling is described as the use of the time-space framework to organize and present information, which tells the public stories about time, places, events in a geospatial context.

Time
Time is a fundamental characteristic of geospatial storytelling, which includes points in time and timelines. Points in time represent the story node of important events. The timeline can also be called the storyline, which can be linear, branched, or cyclic.

Place
Unlike literary narratives, geospatial storytelling combines high-precision maps and narrative structures to tell location-based stories. The place can trigger people's associations with the real geographic world and promote their understanding of specific events.

Event
The event is considered as the theme of geospatial storytelling, and the storyline is formed by linking the cause, process, and consequence of events. Geospatial storytelling is made up of some small events that form the nodes of a storyline. Figure 1 shows the overall research framework based on the geospatial storytelling of this article. We abstract the concepts and relationships of dam-break floods and encode them to form a storyline, whose story nodes are instantiated and transcoded to construct a story scene in 3D. Finally, the user gets knowledge of dam-break floods by decoding, which may trigger deeper thinking about flood risk and disaster awareness.

Storyline of dam-break flood scenes
The geospatial perspective explains 'what, where, why, when, how' for a given disaster (Bandrova, Zlatanova, and Konecny 2012). Therefore, to present a complete picture of a given disaster from a geospatial perspective, the virtual 3D scene should include the cause, the process, and the consequence of the disaster. The causes of the disaster explain 'why did this disaster occur?', the process shows 'what, where, and when did this disaster occur?', the consequence reveals 'how about disaster losses?'. This section summarizes the relevant concepts and relationships in the story of dam-break floods from the above three aspects.
2.3.1. Collection of concepts 2.3.1.1 The cause. In this article, we mainly focus on a specific dam-break flood from barrier lakes formed from landslides, so the cause factors are earthquake, intense rainfall, landslide, and barrier lake, some of which are closely linked to each other (W. Liu et al. 2020;X. Li et al. 2021). As shown in Table 1. 2.3.1.2 The process. The spatio-temporal process enables people to better understand the trend development of dam-break floods, which includes occurrence time, location, arrival time, water depth, flow velocity, etc. As shown in Table 2. 2.3.1.3 The consequence. The consequence describes the impact of dam-break floods on disasterbearing bodies, such as affected buildings, affected roads, critical infrastructures, etc. As shown in Table 3.

Collection of relationships
Due to the spatio-temporal nature of dam-break floods, the following three relationships exist between story nodes: semantic relationship, time relationship, and spatial relationship (Du and Gu 2016).

Semantic relationship.
The main semantic relationships that exist in the case of dam-break floods are: (i) Is-a, an object is a subclass of another object; (ii) induce/inducedBy, which means the Table 1. The cause factors and their descriptions.

Number
Factors Description 1 Earthquake Historical earthquakes lead to the loosening of the geological structure of the disaster area, and then a prone landslide is formed under long-term gravity 2 Intense rainfall Intense rainfall increases the possibility of landslide occurrence and development 3 Landslide A large number of deposits formed by landslide block rivers and form a barrier lake 4 Barrier lake The water level in the upper reaches of the river continues to rise, the barrier lake eventually breaks and causes a flood occurrence of one disaster event may lead to another event; (iii) hasPart/isPartof, a specific object is part of a disaster event; (iv) Amplify, where two disaster events simultaneously lead to the occurrence of another event, as shown in Table 4.

Time relationship.
Time relationships are used to describe the sequence of events in dambreak floods, including points in time and time periods (Du and Gu 2016). We summarize time relationships in Table 5: (i) Before, an event occurs before another event; (ii) Meet, an event and another event meet at a point in time, but one event occurs first; (iii) Overlap, an event overlaps with another event for a certain period of time, but one event occurs first; (iv) Contain, the time period of an event contains another event.
2.3.2.3 Spatial relationship. All scene objects involved in dam-break floods have typical geographical properties, such as spatial location, spatial topology. Spatial location (e.g. spatial position, orientation) processes the registration of geographic position and orientation between different scene objects. Spatial topology enables the correct expression of the layout. In this article, we adopt three topological relationships proposed in W.  to achieve seamless integration of the disaster objects in dam-break floods, which include adjacent, disjoint and overlap. These relationships are mainly used to solve problems such as occlusion, suspension and deep burial of models in the 3D scene. The specific time when the dam-break flood occurs, which can be used to help people to identify the pre-and post-disaster 2 Location The geographic location where dam-break floods occurs, thereby communicating location-based disaster information to people 3 Arrival time Evolution time and arrival location of dam-break floods, and evacuation routes can be dynamically planned 4 Water depth Water depth is the depth at which each position is submerged and it is an important indicator for risk assessment 5 Flow velocity The velocity of water flow at different moments is a crucial risk factor that could be valuable to show the accessibility of certain areas

Time-space storyline
The time-space storyline of dam-break floods is constructed based on the logic of cause, process, and consequence of the disaster ( Figure 2). The start point of the storyline is thing, which is the class that represents the set containing all individuals. The storyline usually includes the following aspects. First, the impacts of historical earthquakes cause the loosening of the geological structure in parts of the disaster area, and the subsequent intense rainfall induces a landslide disaster. Second, the landslide created a large number of deposits block the river channel, and form a barrier lake. Third, when the barrier lake breaks, it creates a fast-flowing and destructive dam-break flood, which causes damage to buildings and roads downstream. Moreover, each node in the storyline has its attribute information, e.g. time, location, etc.
2.5. Storytelling-oriented representation and public cognition

Node instantiation and representation
The event nodes and their relationships are statically recorded in the storyline. The instantiation of each node is the basis for constructing a story scene in 3D. Therefore, the authors propose the instantiation approach for typical events in dam-break floods. 2.5.1.1 Earthquake. Earthquake is such an extremely complex geographic process that it is difficult to realistically reproduce it in virtual 3D scenes, so a schematic approach is used to instantiate earthquake, as shown in Equation (1). (1) In the equation, V represents scene viewpoint; P(x, y, z) and R(h, p, r) represent the position and orientation of the camera in the 3D scene respectively; h, p, and r indicate heading, pitch, and roll; σ is a radian constant; t indicates a time state, t0 indicates the state of the camera at the previous moment, and t1 indicates the state of the camera at the current moment. Therefore, varying the heading parameter of the scene camera from t0 to t1 moments enables the 3D scene to be shaken and thus used to schematically present earthquakes.
2.5.1.2 Geological structure. Detailed representation of the geological structure is not possible without rock formation data within the mountain. The schematic representation of geological structure can be performed by using polylines and colors. Moreover, different frequencies of flicker can be used to control the showing and hiding of the information enhancement, as shown in Equation (2).
In the equation, t indicates a time state, t0 and t1 are two different moments; S t0 and S t1 represent a state at two different moments respectively; Dt = t1 − t0 indicates the state maintenance time, which can be used to determine the flicker frequency.
2.5.1.3 Rainfall. Particle systems are an effective approach for simulating realistic rainfall in 3D scenes (Abdillah, Basuki, and Harsono 2020). The structure of the particle system for rain includes position, speed, lifetime, texture, emission rate, size, etc. The properties of these particles undergo three processes of 'generation', 'activity', and 'death' over time, thus achieving the dynamic effect of rainfall.
2.5.1.4 Landslide. Landslide representation focuses on the spatiotemporal process from the landslide body to the deposit, which is constrained by the total volume and landslide boundaries (P. Huang et al. 2020). The total volume of landslide remains constant during a landslide, as shown in Equation (3): 2.5.1.5 Barrier lake. Taking a polygon as the bottom surface, and the schematic effect of the barrier lake rise can be achieved by continuously changing the height attribute polygon and then intersecting with the 3D terrain scene.
2.5.1.6 Flood. The flood simulation results at each moment include information on water depth, flow velocity, arrival time, etc. We take a 2 × 2 cell of four adjacent cells as the basic unit and construct the triangular model by sequentially judging whether the water depth values in the cell are empty (M. Liu et al. 2017). Moreover, the water depth value at each moment and the continuous blue ribbon exhibited one-to-one mapping, as shown in Figure 3.
Finally, we organize these instantiation clips to generate a 'dam-break flood story' along the storyline to improve the people's understanding of the cause, process, and consequence of dambreak floods.
2.5.2. Cognitive evaluation of dam-break flood scenes 2.5.2.1 Experiment design and implementation process. To prove the effectiveness of storytelling in knowledge communication, it is necessary to conduct a cognitive evaluation of dam-break flood scenes. The core idea of cognitive experiment design is shown in Figure 4. The first step is to set some initial conditions, such as the number of participants, preset questions, etc. The second step is to prepare two comparative experiment materials and divide the participants into two groups to observe the above materials separately, after which they are required to answer the preset questions and record the results. The last step is to evaluate and analyze the above results. Table 6. The analysis indices mainly include effectiveness and satisfaction. On the one hand, we use answer accuracy ('true' or 'false') to reflect the effectiveness of disaster information transmission. On the other hand, we use the Likert scale to collect the public direct feedback on effectiveness and satisfaction (a 5-point scale, with 5 being the highest rating, corresponding to '1: Not at all', '2: Not much', '3: Normal', '4: Somewhat', '5: Very'). The Likert scale does not require participants to provide simple and specific yes or no answers, which somewhat reduces the subjectivity of the respondents and makes it easier for them to answer the questions.

Study area and data description
In this article, the barrier lake in Baige village (98 • 43 ′ 1 ′′ E, 31 • 4 ′ 51 ′′ N), located at the border of Sichuan and Tibet, was selected as the study area for experiment analysis, as shown in Figure 5. On October 10, 2018 and November 3, 2018, there were two large-scale landslides in Baige village, the deposits formed by the landslide blocked the mainstream of Jinshajiang River and formed a barrier lake, which posed a great threat to people's lives and properties in downstream. The original data used in this article were pre-disaster remote sensing images with a resolution of 2m and predisaster DEM with a resolution of 30 m, which were obtained from LocaSpace Viewer, 1 an open and free 3D digital earth. The post-disaster remote sensing images with a resolution of 0.5 m and postdisaster DEM with a resolution of 2 m were obtained by drones. To improve the quality and effect of 3D representation of dam-break floods, we generated a series of tiles based on the post-disaster DEM (2m * 2m) and post-disaster remote sensing images (0.5m * 0.5m), loaded them through the terrainProvider interface and the imageryProvider interface respectively. Based on above mentioned data, we further generated thematic data such as buildings, roads and rivers, the time series data of landslide was generated by P. Huang et al. (2020) through spatial interpolation. The time series data of flood process was simulated by Y. Li et al. (2015) and W. Li et al. "A Rapid 3D Reproduction System" (2021). The coordinate system of all data is GCS_WGS_1984.

Prototype system implementation
Based on WebGL and 3D visualization technology, we developed a prototype system with B/S architecture, which was used to perform three-dimensional virtual representation of dam-break floods. Figure 6 shows the interface of the system. The server side was built with Apache v2.2.34, the browser was Google Chrome 96.0.4664.110. The interface was designed and built base on HTML5, CSS, JavaScript, and Bootstrap.js v3.3.7. The 3D rendering engine used an open-source library Cesium.js v1.45. Cesium is more like a 3D GIS platform, which supports more GIS data formats. As the dambreak flood is a complex geographical phenomenon, we chose the more GIS-based Cesium platform for the visualization.  Figure 5. Study area: the barrier lake in Baige village.
The prototype system was run and tested on Lenovo Legion R9000P2021H. The processor was an AMD Ryzen 7 5800H with Radeon Graphics, with 16 GB memory and NVIDIA GeForce RTX 3060 Laptop GPU 6 GB. Besides the thread for data rendering, we opened a new thread using web worker for pre-loading dam-break flood data and storing. The aim is to improve the running smoothness of the system and the visualization efficiency of the dam-break flood. Therefore, the visualization efficiency of dam-break flood scenes can reach 160FPS.

Three-dimensional virtual representation of dam-break floods
We have organized the events involved in the various phases of the dam-break floods according to the course of the storyline, the status of the node in the storyline will be further updated as the dambreak flood develops, as shown in the bottom part of Figure 6. The storyline in the prototype system helps users to easily understand the development of the dam-break flood process from a global level.
At the local level, the whole process of dam-break floods was represented in a virtual 3D scene. First, we introduced the story background of dam-break floods. For example, due to the specific topography and climatic conditions, Sichuan has become one of the most serious areas of geological disasters in China, the dynamic diffusion symbol schematically expressed the location of earthquakes in Sichuan, as shown in Figure 7(a). Next, the scene camera flied to Jinshajiang River, the Diexi earthquake, Wenchuan earthquake, and Jiuzhaigou earthquake have continued to affect the geological structure of this area, a prone landslide has been formed under long-term gravity. The shaking of the 3D scene and yellow mesh represent earthquake and geological structure respectively, as shown in Figure 7 Along with a short-term intense rainfall, a large-scale landslide occurred in Baige village at the border of Sichuan and Tibet on October 10, 2018, as shown in Figure 8(a). The spatial interpolation results of pre-and post-disaster DEM and post-disaster image was overlaid to show the impact range of landslide deposits. The deposits blocked the mainstream of the Jinshajiang River and formed a barrier lake. The water level in the upper reaches of the river continued to rise and the storage capacity of the barrier lake reached 183 million m 3 , as shown in Figure 8(b). We combined the water level rise animation with text annotation to represent the formation of the barrier lake. Figure 6. Interface of the prototype system. The prototype system consists of five main parts namely Navigation Bar, Function Menu, Scale, Storyline, and Virtual Globe. The Navigation Bar is used to run the system and load data, the Function Menu includes layer management, frame rate monitoring, compass, zoom, etc., the Scale records the zoom scale of the scene, the Storyline presents the progress of dam-break floods, and the Virtual Globe provides a 3D visualization of the whole process of dam-break floods.
When the dam body breaks, the flood advanced sharply downstream. Four monitoring points were used to dynamically show the arrival time, flow velocity, flow rate, and water depth as the flood passes, as shown in Figure 9. To represent the evolution of dam-break floods realistically, the water depth value was mapped to blue gradient color, which is more in line with the public perception. The darker the color, the greater the water depth. Regarding the representation of affected buildings and roads, the public prefers that they be represented in a simple 'binary' way. Therefore, red indicates danger, green and blue indicate safety, the risk degree criteria is beyond the scope of this article, the output of affected buildings and roads were provided and supported by Zhu et al. (2016). To make it easier for readers to understand the whole process of dam-break floods, we have recorded the 3D representation story of dam-break floods into a video and uploaded it to YouTube. 2

Comparative experiment
3.4.1. Experimental materials and procedure As described in Section 2.5.2, we prepared two comparative experiment materials, one is the storytelling video (85 s), which the screenshot in Figure 10(a) belongs to. The other one is a text report  (206 words) extracted from the recommendation issued by the disaster emergency department of the Sichuan Bureau of Surveying, Mapping, and Geoinformation, 3 as shown in Figure 10(b). This recommendation documented the entire process of the landslide causes, barrier lake formation, and dam-break flooding, the design of the storyline of dam-break flood scenes also referred to this recommendation. Under the principle of information equivalency, the authors think these two materials are trying to convey roughly the same disaster information, the difference is in the way they are presented.
Subsequently, the above comparative materials and preset questions were post in Chinese through a free survey site, 4 and 82 participants participated in this comparative experiment, they were randomly assigned to two groups: group A and group B. The 37 participants in group A were asked to read the text report, the 45 participants in group B were asked to watch the storytelling video and then were required to answer the preset questions after watching the selected material (Table 7). Some questions were based on the work of H. M. Huang, Rauch, and Liaw (2010)

Result analysis
After receiving participant feedback, the authors first tested the results for normal distribution using the Shapiro-Wilk test, as shown in Table 8. The statistical results show that all the p-values are smaller than 0.05, indicating that none of the above data fit the normal distribution.
The authors adopted mean and standard deviation to analyze participant feedback. The mean reflects the main trends in participant answers, the standard deviation indicates the dispersion of the test results and the stability of the preset questions. Due to the samples did not fit the normal distribution, the Mann-Whitney U test was used to compare differences between group A and group B, u and z indicate statistics, the p-value is the probability, p less than 0.01 is statistically highly significant. As shown in Table 9 and Figure 11 respectively, the p = 0.001, 0.000, 0.000<0.01 of answer accuracy, effectiveness score, and satisfaction score, indicated that there was a significant difference between group A and group B in all three indices.
Median (MD) and interquartile range (IQR) were used to reflect the distribution of participant feedback. In Figure 11, the bold black line and the interval indicate MD and IQR respectively. As shown in Figure 11(a), the mean accuracy of group B (M = 0.70, SD = 0.28) was significantly higher than that of group A (M = 0.47 , SD = 0.29, p = 0.001<0.01), the MD value of group B was greater than 0.6, the IQR was approximately distributed in the interval (0.4, 1.0), which indicated the storytelling video was more effective in conveying disaster information compared to the text report, but also reflected that the cognitive ability of the storytelling video varies widely among different public, which was related to education level and cognitive style.  What are the factors that lead to dam-break floods? Score Q4 Can this approach help you effectively understand the whole process of dam-break floods? Satisfaction Score Q5 Are you satisfied with the representation approach of dam-break floods? As shown in Figure 11(b), the mean effectiveness score of group B (M = 4.53, SD = 0.66) was significantly higher than that of group A (M = 3.24 , SD = 1.30, p = 0.000<0.01), the MD value of group B was 5.0, the IQR was distributed in the interval (4.0, 5.0), the MD value of group A was 3.0, the IQR was distributed in the interval (2.0, 4.0), which indicated that the majority of the public believes that storytelling videos can greatly and effectively improve their understanding of flood hazards.
As shown in Figure 11(c), the mean satisfaction score of group B (M = 4.58, SD = 0.58) was also significantly higher than that of group A (M = 3.46 , SD = 1.28, p = 0.000<0.01), the MD value of group B was 5.0, the IQR was distributed in the interval (4.0, 5.0), the MD value of group A was 3.0, the IQR was distributed in the interval (3.0, 5.0). For the storytelling video, the percentages of participants who were very satisfied, satisfied and normal were 62.2%, 33.3% and 4.4% respectively. For the text report, the percentages of participants who were very satisfied, satisfied, normal, dissatisfied and very dissatisfied were 29.7%, 16.2%, 32.4%, 13.5% and 8.1% respectively. Hence, the text report plays a very important role in emergency decision-making, the storytelling videos are more applicable to the popularization of disaster science for the general public.

Discussion
Based on our experience gained in geospatial storytelling, 3D representation, and risk communication, we underline the advantages, limitations, and future research directions that are associated with the proposed method in this section.

Principal findings and potential applications
As a crucial component of disaster management, risk communication has shifted from the pure safety science to an interdisciplinary culture that involves all stakeholders, which aims to communicate relevant information about disasters and help the people at risk be prepared for an emergency (Macchione et al. 2019;Dransch, Rotzoll, and Poser 2010). In this article, taking dambreak floods as an example, we innovatively propose a new approach to disaster information dissemination and risk communication by combining geospatial storytelling and 3D representation. Geospatial storytelling can effectively express the spatiotemporal relationships in the story, and its use to organize the concepts and relationships involved in the dam-break floods is helpful to people's understanding and perception of dam-break floods from a macro level. Besides, the literature showed that vivid representation of disaster information plays a crucial role in knowledge sharing and risk communication, because vivid information may improve memorability and the construction of mental representations (DiBiase et al. 1992;Dransch, Rotzoll, and Poser 2010). 3D representation can transform complex 2D simulations into 3D scenes and provide a vivid and immediate understanding of the whole process of dam-break floods (Costabile et al. 2021). The combination of the two aims to tell people a dam-break flood story rather than providing a professional 3D scene, thus enhancing public perception and risk communication efficiency.
To the author's knowledge, the principal findings of this article have two potential applications. One application is disaster education for the general public. Pre-disaster education is the focus of disaster education, the story scenes with the advantage of intuitiveness and easy understanding could be used for pre-disaster education to disseminate the knowledge about dam-break floods, thus enhancing people's awareness of floods and strengthening risk perception. One of the characteristics of story scenes is that they can convey location-based disaster information to the public, so they will be one of the best carriers for showing evacuation plans, rescue routes, and temporarily assembling points to the public in during-disaster education. Disasters have not only immediate mental and physical harm, but also have long-term consequences for the public. The story scenes could also provide the affected people with a deeper awareness of dam-break floods and improve their prevention capabilities, and have the potential to make disaster prevention a good habit of life. Another application is for disaster information broadcast. The current way of disaster information broadcast whether on SMS, TV, or Social Media is rely excessively on text, pictures, and videos, this kind of broadcast without geographical location is difficult for the public to understand what really happened. The combination of geospatial storytelling and 3D representation allows the public to observe the whole process of dam-break floods from an all-around perspective.

Limitations and future research directions
For additional inspiration and information to consider, the limitations of and future research directions of this study are discussed below.
In the creation of the storyline, the authors considered only some of the causes, processes, and consequences of the dam-break floods. However, as described by Gaagai, Boudoukha, and Benaabidate (2020), Gaagai et al. (2022), W. Li et al. (2020), and Psomiadis et al. (2021), the causes and consequences of dam-break floods are more than the factors summarized in this article, but this does not mean other affected elements are not important (e.g. extreme storm, affected land, affected ecology, etc.). Therefore, we are considering more factors related to dam-break floods and further expanding the storyline and enriching the 3D scenes. From the perspective of cartography, it is important to note that too much content in disaster scenes will lead to information overload and insufficient risk perception of users, so how to balance the cognitive efficiency and the information density in 3D scenes is also a research question worthy of our in-depth study.
In the cognitive evaluation, the evaluation method is somewhat single and we need to further design a more reasonable evaluation mechanism under the comprehensiveness and normalization principles, the analytic hierarchy process and fuzzy comprehensive evaluation can be introduced. Moreover, combining eye tracking with virtual reality (VR), augmented reality (AR) and mixed reality (MR) technologies for cognitive evaluation allow for a more in-depth evaluation of user feedback on flood hazards (Chen and Lin 2018; G. Dong et al. 2021).
In terms of promotion of the method, we actually addressed a specific dam-break flood from barrier lakes formed from landslides, but the idea of combing geospatial storytelling and 3D representation proposed in this article is widely applicable to other disasters, such as floods in general, earthquakes, and even more complex chains of disasters. Furthermore, this intuitive 3D representation approach with logic also has potential advantages in evacuation simulation and risk assessment (Y. Psomiadis et al. 2021).
In addition, the authors will try to create a storyline-based visualization of disasters in an automatic manner, a topic worthy of investigation is the automatic generation of the storyline underlying graphs, this would be the subject of future research using Statistical Relational Learning (SRL) following the ideas of Dehbi et al. (2017). Besides, the authors will continue to refine the storytelling system of dam-break floods, with the aim of developing a participatory and interactive 3D application of disasters.

Conclusion
The use of storytelling to describe complex geographical processes can improve the efficiency of spatiotemporal information communication and is more in line with public demand (Roth 2021). In this article, we proposed a three-dimensional virtual representation method of dambreak floods from a storytelling perspective, which can provide an intuitive and realistic dambreak scene in 3D for the public. We first abstracted the concepts and relationships involved in the whole process of dam-break floods and formed a time-space storyline. Then, each story node was instantiated and a 3D scene of the dam-break flood was constructed using a combination of static and dynamic representation. In addition, a comparative cognition experiment was conducted on the storytelling video and the text report, the results show that the storytelling and 3D