Shelter GO: multiplayer location-based game to promote knowing shelters for emergency evacuation

In the event of a disaster, we evacuate to a shelter. However, if we do not know where to go for refuge, we consume time or end ourselves at an unknown shelter. Furthermore, we cannot decide on a suitable shelter if we do not know its capacities and features. To promote knowing shelters in advance is essential for successful emergency evacuation. A promoting approach is the use of a location-based game. Thus, a prototype of a multiplayer location-based game known as “Shelter GO” was developed to promote players to become acquainted with shelters by adopting a game element similar to Pokémon GO. The game feature allows players to not only earn points by visiting and observing shelters and collecting digital creatures at disaster-related spots but also exchange their collected digital creatures with encountered players. The digital creature exchange entails knowledge exchange. A preliminary comparative experiment demonstrated that Shelter GO can promote knowing shelters and that knowledge exchange may not be frequently done but promote knowing shelters.


Introduction
Severe natural and artificial disasters have occurred globally, causing damage and deaths (CRED, 2023). Although simulation technology has advanced to predict the occurrence and strike of some disasters (e.g., typhoons), we may encounter unpredictable disasters (e.g., earthquakes and terrorist attacks). When disaster strikes or is predicted, we evacuate to a shelter depending on the circumstances. Evacuation is essential for saving lives in the event of a disaster, although it can be difficult at times, and the difficulty increases significantly with unpredictable disasters. For example, when a typhoon is approaching from afar, we can calmly decide which shelter to evacuate to and then begin evacuation.
Contrarily, if an earthquake with a tsunami occurs, we must quickly decide which shelter to evacuate to. If we do not know where to go for shelter, we consume time until we decide or go toward an unknown shelter; in a worst-case scenario, lives may be lost.
Mobile applications have been developed to provide real-time evacuation support to avoid the worst scenario. For example, mobile applications for indoor evacuation can detect a user's current location using radio-frequency identification tags (Chittaro & Nadalutti, 2008) or fiducial markers (Kanangkaew et al., 2023) to navigate the user toward exits or safe zones by presenting location-aware instructions (e.g., arrow and line) based on building models. Other location-aware applications periodically present the shortest and safest evacuation routes from a building's environmental data (e.g., temperature and visibility) while considering user characteristics (e.g., age and health conditions) (Atila et al., 2018;Eksen et al., 2016). Focusing on outdoor evacuation, mobile applications employing a global positioning system (GPS) to detect a user's current location must consider a greater number of evacuation routes across a wider area than those applied in indoor evacuation. Furthermore, the sensors cannot be readily distributed over a wide area to collect the outdoor environmental data available for evacuation. To overcome this difficulty, certain mobile applications for outdoor evacuation adopted a GPS but smartphone-based cloud computing and recommended the best evacuation route while estimating the user's surroundings such as impassable or congested roads (Itoi et al., 2017;Xu et al., 2018). In the event of a disaster, these applications are expected to aid in successful evacuation. However, we may be unable to remain unruffled in the event of a disaster and appropriately utilize such mobile applications. Moreover, a mobile device may become physically unavailable due to breakage and loss. Therefore, for successful emergency evacuation, the shelter locations must be known in advance. (2024) 19:9 et al., 2019), using 3D visualization. Compared with paper-printed hazard maps, digital hazard maps can be easily distributed to modern people who have personal computers, including smartphones.
When an unpredictable disaster occurs, residents can begin evacuation by referring to the marked shelter locations on the map. Although when there is mental instability (e.g., panic), limited perception (e.g., low visibility due to a blackout), or unusable smartphones (e.g., disconnected Internet), residents may be unable to look at the map. In the event of such an unforeseen situation, it is still necessary to have known shelter locations. Even if residents are careful about disasters, to decide on a suitable shelter, they should be aware of not only shelter locations but also other types of shelter information such as capacity and features.
For example, if a large number of residents rush into a small-capacity shelter, the overflowed is forced to abandon the crowded shelter and evacuate to another shelter. If sickly residents are unfamiliar with shelter features (e.g., main equipment), they can go to the nearest shelters that do not possess medical supplies. Some hazard maps show shelter capacities, but most hazard maps do not show enough information for residents to decide on a suitable shelter.

Location-based game as an alternative to hazard map
A method other than looking at a hazard map is required for residents to effectively locate shelters (i.e., location, capacity, and feature). Residents may visit and observe shelters as a means to know the shelters as perceptual experiences in the real world. These experiences can make knowing more impressive than the monotonous activity of looking at the map.
However, residents may unwillingly accept this method because these experiences entail temporal or physical burdens. Thus, this method should adopt a motivational approach so that residents can visit and observe shelters beyond the burden.
To serve as a motivational approach, digital games have been actively introduced in the field of disaster risk reduction (Solinska-Nowak et al., 2018), with certain digital games focusing on disaster education. For instance, a strategic simulation game has been practically used by high-school students to learn flood protection based on social infrastructure (Tsai et al., 2015). A simple quiz-style game was explored for elementary school students to learn earthquake preparedness on their smartphones (Winarni et al. 2018). Location-based games are especially promising for visiting shelters. For example, Pokémon GO is a globally popular location-based game, and several studies have investigated not only its influences on physical activity (Lee et al., 2021), cognitive performance (Ruiz-Ariza, et al., 2018), and mental health (Watanabe et al., 2017) but also its availability for health education (Finco et al., 2018) and language learning (Wu, 2021).
Besides Pokémon GO, location-based games have been actively applied to education, and location-based educational games are effective in making learning enjoyable and (2024) 19:9 heightening learning efficacy in various learning domains. For example, a location-based educational game for local cultural learning demonstrates context-aware learning contents (e.g., quizzes) by detecting whether students are near historic monuments (Lin et al., 2018).
A narrative-driven location-based augmented reality (AR) game has been developed for players to study English by interacting with nonplayer characters or for obtaining items displayed on a digital map (Lee, 2022). In addition, location-based games have been used in disaster education. For example, location-based mixed-reality games enabling human players and computer agents to interact have been used for first responder training (Ramchurn et al., 2016). A location-based game that provides multimedia content including AR at predefined locations or times depending on a disaster scenario has been used for earthquake and tsunami evacuation training (Mitsuhara et al., 2015;Mitsuhara et al., 2021). Although not a location-based game, a mobile application for tsunami evacuation training has been overlaid on a digital map with a simulated tsunami approach, allowing trainees to run away from tsunamis while seeing the map in a time-trial competition (Yamori & Sugiyama, 2020). From the reported advantages, location-based games are expected to be an alternative to hazard maps in knowing shelters.

Our previous research
Based on this background, we developed a prototype of a single-player location-based game that promotes a player (resident) to visit shelters by introducing a game element similar to that of Pokémon GO . In the game prototype, players visit a shelter to earn a point equal to the capacity of the shelter and then visit a disaster-related spot (e.g., disaster monument and alert sign) to collect (acquire) a digital creature by paying a required point from his/her earned points. The digital creature collection can promote players to visit different shelters while feeling enjoyable and not burdened, and consequently, players can know many shelters. Our preliminary experiment showed that the game prototype helped players to remember more shelter locations and accurate shelter capacities than just looking at a hazard map. However, the game prototype focused on a single-player mode and did not have a multiplayer game element. Pokémon GO introduced multiplayer game elements such as exchanging digital creatures (i.e., Pokémons) for more enjoyable, sustainable gameplay. If a multiplayer mode is implemented in the game prototype, players may be able to know shelters more effectively while interacting with other players.

Research purpose and questions
Our research promotes knowing shelters using a multiplayer game element (multiplayer mode). This is achieved by extending the game prototype to a multiplayer location-based game known as "Shelter GO." Shelter GO promotes knowing shelters by enabling players (2024) 19:9 to exchange their collected digital creatures. The concept behind the multiplayer mode is that digital creature exchange entails knowledge exchange.
Considering the flow from our previous research, we set the following research questions: • RQ1. Which promotes knowing shelters more, the single-player mode or multiplayer mode (Shelter GO)?
• RQ2. How do players exchange knowledge in Shelter GO?
This study is organized as follows. First, a four-phase process for knowing shelters, the game element (digital creature collection), and the game prototype (the single-player mode) are described. Second, as the key game element in Shelter GO, digital creature exchange (knowledge exchange) is described together with the application user interfaces. Third, a preliminary experiment to answer the research questions is reported. Finally, this study is summarized, and future work is presented. This study contains reconstructed concepts and detailed experiment analysis (including partial data correction) as extended parts from .

Single-player location-based game for successful evacuation
When an unpredictable disaster strikes, we are often forced to evacuate. Therefore, we must prepare for a successful emergency evacuation.

Importance of knowing shelters
In the event of unpredictable disasters, the evacuation process (timeline) is as follows: 1) a disaster occurs; 2) individuals (or people present) recognize the disaster by receiving an alarm or perceiving it from their surroundings; 3) they decide whether to evacuate and which shelter to evacuate to; 4) they begin and continue moving to shelters; and 5) they reach the shelters.
Step 3 occasionally takes time because of dither caused by various factors (e.g., perceived risk, cognitive bias, past disaster experience, and received information), and steps 3 and 4 may be hampered by unexpected situations (e.g., stupor, injury, and blocked road). Regarding the 2011 Tohoku earthquake in Japan, an investigation revealed that 10.1% of residents (surviving respondents) in a coastal area did not know shelter locations, whereas 42.2% were evacuated to undesignated higher places (Makinoshima et al., 2016). Residents in coastal areas should have evacuated to shelters before the arrival of the resulting tsunami (within 30 min in some areas), and many residents may have prioritized immediate survival by reaching undesignated higher places; however, the tsunami caused approximately 14,000 fatalities (approximately 90% of all fatalities). Following the earthquake, many shelters were built in tsunami-anticipated coastal areas, and residents were trained to evacuate to predesignated shelters as part of practical disaster preparation. (2024) 19:9 The more shelters that are built, the more residents can survive by reaching the shelters.
On the one hand, residents should be aware of additional shelters to ensure their survival. This is because all residents cannot always evacuate to their predesignated shelters during unexpected situations (e.g., crowded or collapsed shelters), and some residents must quickly decide on another shelter, as stated in step 4. Therefore, residents must know multiple shelters for their survival. In other words, if this prerequisite is not satisfied, even residents who regularly participate in evacuation training may be unable to quickly decide on a suitable shelter other than predesignated shelters. Despite possessing a smartphone to obtain updated shelter information, residents may not always be able to utilize their smartphones while information and communication infrastructure is disrupted by a disaster.

The process for knowing shelters
Our proposed method aims at disaster preparedness, i.e., the set of actions to follow before the occurrence of an unpredictable disaster, and requires residents to visit and observe the shelters. To promote knowledge, we set the following four-phase process by introducing a game element.

Preview phase
As an introductory phase, a resident (or a resident group) previews a hazard map and gets a sense of shelter locations (e.g., distance and direction) at least, and then decide the shelters to visit. This phase can be skipped if the resident knows the shelter locations or dislikes knowing them in advance.

Walk phase
This phase involves not only walking (moving) but also grasping the shelter environment (e.g., ground condition), which is difficult to recognize by only looking at the hazard map.
To avoid getting lost, the resident can walk while occasionally looking at the hazard map.
Walking may entail a temporal or physical burden and consequently cause residents to be unmotivated to follow the process. In this phase, it is important for the process of sustainability that residents can walk beyond the burden.

Visit and observe phase: VO
In this phase, residents visit (reach) and observe a shelter. The observation is often done visually from the outside because many shelters are released only in the event of a disaster; occasionally, residents may be allowed to enter the shelter or talk with the manager. By observing the real shelter, residents can: • Know the shelter location deeply together with his/her impression. For example, "this shelter is located on a hill, and a nearby tall tree is a landmark." (2024) 19:9 • Know the shelter capacity (shown on the hazard map) deeply by comparing it with his/her presumed capacity (impression). For example, "this shelter looks spacious but actually cannot accommodate many residents inside." • Know the shelter feature that can help determine whether the shelter is suitable for various disaster scenarios. For example, "this shelter looks old and hygienically worrisome and may be unsuitable for long-day evacuation." Each resident has a different impression of the observation. In other words, shelter information acquired by a resident during this phase includes the resident's subjectivity and is regarded as his/her shelter-related knowledge for successful evacuation. In addition, repeated observations can bring widely applicable knowledge (e.g., "tall buildings are often designated as tsunami shelters. In an emergency, we can evacuate to a tall building").
Such subjective shelter information, i.e., acquired knowledge, is not shown on the hazard map that attaches importance to objectivity and invariance.

Visit and collect phase: VC
Repetition of the Walk and VO phases results in knowing many shelters. However, unless the Walk phase is promoted, the repetition will not continue. To promote the Walk phase using a Pokémon GO-like location-based game, we set a new phase known as "visit and collect" (VC)-collect derives from our introduced game element.
In this phase, the residents visit a disaster-related spot and acquire lessons from the spot.
For example, when visiting a monument conveying tsunami lessons together with the past tsunami's height, residents will acquire lessons impressively while comparing it with their height or circumstances.

Game element
Our introduced game element is a digital creature collection and works in the VO and VC phases to promote the Walk phase. The four-phase process including the game element is shown in Figure 1.

Earning point in the VO phase
When visiting a shelter, players earn a point equal to the capacity of the shelter. The earned points can be used to collect digital creatures in the VC phase. Even if players return to a shelter at short intervals, they cannot re-earn the point at the shelter. However, re-earning a point is allowed after a certain time passes. This rule allows players to be promoted to visit many shelters and revisit the previously visited shelters to stabilize acquired knowledge.

Collecting digital creatures in the VC phase
When visiting a disaster-related spot, players can collect a digital creature by paying a required point from their earned points. A different digital creature is assigned to each spot, and the required point depends on the creature (i.e., the spot). This setting, similar to treasure hunting, can promote players to visit many disaster-related spots to collect their unknown creatures.
This digital creature collection promotes the Walk phase, i.e., visiting many shelters and disaster-related spots. Consequently, players can acquire more knowledge (shelter-related knowledge and spot-related lesson).

Game prototype
The game prototype was developed as a Unity3D-based geofencing mobile application (app) that works on a GPS-enabled Android smartphone (or tablet). The app with the single-player mode, which does not require an Internet connection, deals with the Walk, VO, and VC phases. The user interface of the game prototype is shown in Figure 2.

Walk phase
Figure 2-a shows the screenshots presented during the Walk phase. In this phase, the app always displays a digital map, a "gallery" button, and some information such as earned points. The map shows a player's current location, shelters, and disaster-related spots as corresponding icons. When reaching a shelter (invisible rectangle area enclosing the shelter), players can move to the VO phase. When reaching a disaster-related spot, they can move to the VC phase. By tapping the button, players are taken to the gallery where they can see the list of the collected digital creatures. The app compulsorily takes players at random intervals to digital creature encounter events, which are not associated with the location. Although not originally required, this compulsory transfer is prepared to avoid a situation where players cannot efficiently visit disaster-related spots. (2024) 19:9

VC phase
Figure 2-c shows the screenshots presented in the VC phase. In this phase, at the reached spot, the app first displays instructions (e.g., "please look at the disaster alert sign around you"). Following that, the app activates the rear camera of the smartphone and displays a digital creature in an AR mode, which superimposes the digital creature CG onto real-time vision using a markerless AR framework. When players tap right on the moving-around creature on the screen, the app displays a lesson associated with the spot (e.g., "anticipated tsunami height is 3.0 m in this area and 0.8 m above-sea level here. If you stay here after a big earthquake, you will be engulfed by the tsunami"). When the compulsory transfer occurs, the app provides general lessons (e.g., "school buildings are often registered as shelters"). After reading the displayed lesson, players can acquire the creature by paying the required point.

Shelter GO
The game prototype promoted knowing shelters using the single-player digital creature collection while providing simple and player-dependent gameplay. Advantageously, players can start and stop play depending on their circumstances (e.g., free time). Contrarily, the game prototype provides lonely gameplay and is concerned with process sustainability, i.e., how long players keep playing without losing motivation. Our quick idea to remove this concern is to convert the single-player mode to a multiplayer mode.
The single and multiplayer modes have numerous learning-related merits and demerits, and game/instructional designers can suggest a better mode to maximize learning effects in future educational games (Harteveld & Bekebrede, 2011). Regarding Pokémon GO, Laato et al. (2021) reported that players' affection for fictional creatures and social connection with other real players can result in meaningful gameplay. Because encountering other real players may lead to maintaining motivation, the game prototype is extended to a multiplayer mode. The multiplayer location-based game is "Shelter GO."

Multiplayer mode toward visiting more shelters
It is rational from an efficiency perspective for players (residents) to only visit shelters and disaster-related spots in their local community (i.e., living area). The single-player mode was designed to work in four phases, assuming that individual players walk around their local community. However, players should visit more shelters within and outside their local community because where there are disaster strikes is unpredictable. This ideal process, which may entail long-distance walking, is anticipated to increase the walking-caused burden. Thus, a new game element special to the multiplayer mode is required to promote people to visit shelters outside their local community.

Involving other players
It is a waste that knowledge for successful evacuation only has been contained inside individual players. Therefore, knowledge should be shared among players. For example, a sickly player's knowledge of shelter features may be useful for not only homogeneous players but also for other diverse players, e.g., "old shelters have no air conditioner and should be avoided in summer or winter." The multiplayer mode involves other players in the four-phase process, assuming that players desire encounters with other players and will be motivated to walk around regardless of whether they are inside or outside their local community for more encounters.
Furthermore, the encounter becomes the basis for a new game element. To make knowing shelters more enjoyable and effective, the multiplayer mode helps encounter other players during the Walk phase. A player can do the following in multiplayer mode: (2024) 19:9 • Maintain motivation by interacting face-to-face with encountered players.
• Acquire (receive) knowledge from encountered players-knowledge sharing.
Knowledge sharing imposes knowledge exchange (i.e., transfer and receive) between players. If focusing heavily on knowledge sharing (i.e., encounter with other players), players may be unmotivated to visit shelters. However, the encounter does not always happen because of factors such as player population, density, and personality.
Consequently, players must acquire sharable knowledge in advance during the VO and VC phases to make the encounter attractive and meaningful, and they can be motivated to visit shelters.

Game element: exchanging collected digital creatures
Players can collect digital creatures in various local communities regardless of whether they are playing in single or multiplayer modes. To exceed the digital creature collection's enjoyment, we implemented a game element that makes knowledge sharing enjoyable. The game element we envision is digital creature exchange, which includes knowledge exchange. Encountered players can collect unique digital creatures and agree to exchange each player's digital creature. Players are expected to not only collect many digital creatures by visiting shelters and disaster-related spots but also collect unique digital creatures from other players via digital creature exchange. In other words, players are expected to acquire many pieces of knowledge by not only visiting shelters and disasterrelated spots but also receiving from other players. The multiplayer mode and its game elements are shown in Figure 3. The knowledge exchange is not done by direct interaction (e.g., conversation) between players but is done via the app. Therefore, the app imposes the following duties on the players.

Register knowledge
When acquiring a digital creature at a disaster-related spot, a player must register his/her knowledge of the creature. This registration is known as knowledge externalization, which helps stabilize knowledge. Registering shelter-related knowledge is most expected, but any type of knowledge useful for successful evacuation is accepted (e.g., "do not forget to bring your indispensable medicines when evacuating"). The registered knowledge will be a candidate to exchange via the app.

Transfer knowledge
When exchanging a digital creature with an encountered player, he/she must transfer the registered knowledge together with the creature. After exchanging the knowledge (digital creature), the player can no longer see the registered knowledge on the app. This rule may impose a burden on players, but it can motivate them to surely remember knowledge while assuming that their collected digital creatures will leave. Figure 4 shows the configuration of the Shelter GO app that works together with a server for the multiplayer mode. The Shelter GO app allows players to switch to single-player (i.e., without an Internet connection) or multiplayer modes (i.e., with an Internet connection). For the single-player mode, the app has basic modules (e.g., phase detection, phase transaction, and point management) and internal data storage (e.g., shelter, disaster- related spot, and point). Shelter data comprises name, latitude, longitude, and capacity (i.e., point). Disaster-related spot data comprise a latitude, longitude, digital creature, lesson, instructional message, and image. For the multiplayer mode, the app includes modules (e.g., knowledge registration and transfer) and external data storage required for knowledge exchange. The external data storage, where a data interface module (implemented by PHP) in a server system makes the transaction, has a comma-separated values (CSV) file for each exchange that records temporary data, such as player ID, player name, creature ID, and registered knowledge.

Implementation
Location-based games potentially bring danger because of the outdoor gameplay with the fixation on the smartphone screen. For example, Faccio and McConnell (2018) reported that over 286 Pokémon GO-caused traffic accidents, including two deaths, occurred in a US county for 148 days after its release. We must avoid letting a player get injured or die in a disaster while using the Shelter GO app to protect lives from disasters. Therefore, we implemented a safety-net function that prohibits gameplay during bad weather. Although only applicable in Japan, this function collects open weather data from the Japan meteorological agency at short intervals and inactivates the app when a weather warning (e.g., heavy rain) is issued in the municipality where each player is currently located.

Exchange knowledge
Currently, the app only allows two players to exchange knowledge (digital creatures), which occurs when two players meet face-to-face in the Walk phase and agree on the exchange. Figure 6 shows the screenshots presented when exchanging knowledge. A player who recognizes a nearby player in the Walk phase can speak to the player and suggest a digital creature exchange. If the suggestion is accepted, both players tap the "creature exchange" button to display the menu (Figure 6-a). One player taps the "host mode" button and the app displays a quick response (QR) code (Figure 6-b). The other player taps the "guest mode" button and then scans the QR code using the rear camera of a smartphone. After a successful scan, the app displays another QR code for the host player to scan. When both scans are successful, the app displays a list of digital creatures (Figure 6

Preliminary experiment
To answer the research questions, we conducted a preliminary comparative experiment between the single and multiplayer modes. We focused on the tsunami evacuation because participants resided in a coastal area where a tsunami can cause devastating damage. Thus, participants require to be aware of tsunami shelters.

Participant
We recruited participants by inviting acquainted undergraduate students of Tokushima University to participate without reward. The participants were ten undergraduate students who were in the same course (computer science course) and knew each other. A prequestionnaire revealed that they had little interest and knowledge about tsunami evacuation and shelters. They were divided into the following groups at random: • Group A (N = 5): The participants simultaneously played an experimental Shelter GO that excluded the knowledge exchange function, i.e., the single-player mode with knowledge registration to digital creatures.
• Group B (N = 5): On a different day from Group A, the participants simultaneously played Shelter GO, allowing them to register and exchange knowledge. They freely decided whether to exchange knowledge.

Procedure
After the students were instructed on how to use their assigned Shelter GO app, we asked them to learn about the evacuation to tsunami shelters in an experimental area using the app installed on their smartphones. We did not impose a time limit, although they were informed that it would take approximately 1 h to visit the tsunami shelters.
Participants in each group started playing the game from a different location simultaneously. During the gameplay, the compulsory transfer to the VC phase, point re-earning by shelter revisit, and special digital creature acquisition based on a player's activeness was deactivated. The Shelter GO app recorded participant behavior such as visiting shelters, registering knowledge, and exchanging knowledge.
Following the gameplay, each participant completed a surprise test and a postquestionnaire. The surprise test was a recall test asking about the shelter locations and capacities. To show whether they remembered the shelter locations, they marked them on a blank map of the experimental area with no designated tsunami shelters. Furthermore, they drew the shelter capacities on a map to show whether they remembered the capacities.
The questionnaire required them to itemize all their acquired knowledge from the gameplay, and they were also asked about the effects of the gameplay.

Area
The experimental area, which was a familiar area for participants and covered a 700 m 2 of Tokushima University campus and its surrounding area, had sixteen tsunami shelters (buildings that met safety standards) and eight tsunami-related spots (e.g., alert signs against tsunami inundation). The capacities of the tsunami shelters ranged from 26 to 1,736.

Results
Because of the small number of participants, statistical tests were not applied to all the obtained data. The data of each participant are discussed below.

Behavior and recall test
The mean gameplay durations of Groups A and B were 43.4 min (37.9-49.5 min) and 52.6 min (43.2-67.3 min), respectively. The quantitative data obtained in the experiment are presented in Table 1. The mean values of Groups A and B for the number of visited shelters (Num_VS) were 9.6 (7-12) and 9.2 (6-13), respectively. The mean values of Groups A and B for the number of visited tsunami-related spots (Num_VTRS) were 5.4 (4-7) and 5.6 (3-8), respectively.
A recalled shelter location can be correctly recalled if a participant correctly marked the shelter-building correspondence on a blank map. All participants recalled locations only from their visited shelters; they did not answer the recall test by guess. The mean values of Groups A and B for the accuracy rate for all shelter locations (AccRate_ASL) were 43.7% (31.2%-68.7%) and 33.7% (18.7%-56.2%), respectively. The mean values of Groups A and B for the accuracy rate for visited shelter locations (AccRate_VSL) were 66.5% (41.6%-100%) and 59.3% (33.3%-100%), respectively. The mean values of Groups A and B for the number of capacity-recalled shelters (Num_CRS) were 3.6 (1-9) and 5.4 (3-9), respectively. All participants recalled capacities only from their visited shelters. For the accuracy of recalled shelter capacities (Acc_RSC), we calculated the mean absolute value of the difference between the correct and recalled capacities, and the mean values of Groups A and B were 86.9 (36.0-217.5) and 70.9 (47.0-141.6), respectively. Because these (2024) 19:9  Figure 7 shows the knowledge exchange among Group B participants using the labels of the registered knowledge shown in Table 2.  I recommend the gym (on campus) because it is large. [S] The second tsunami wave is higher than the first one. [D] The university gym has a big capacity.  There are more buildings designated as shelters than I guessed.
There are not many shelters outside the university, and the capacities are less than 100.

AK2
There are many shelters other than public facilities such as the university.
An apartment my friend lives is assigned as a shelter.
More shelters exist in this area than I imagined.
I was surprised that the university has big capacity shelters.
The K-building's capacity exceeds 1,000.

AK3
The above-sea level around the university is low.
I realized that I do not know about disaster risk reduction.
The above-sea level is 3 m around the university.

AK4
The Tsunami inundation depth is 3 m around the university.
A few shelters are gathered in narrow areas. AK5 A map does not show which entrance we can enter shelters from. AK6 Shelter locations are easy to find on a map but difficult to recognize the first time I visited.
Tokushima University has multiple shelters.
A three-meter-high tsunami may come to the Suketobashi area, and the Abestate building is a shelter.
The university has many facilities probably available for shelters.
There are small-capacity shelters around the university.
The university gym has a big capacity.

AK2
There are many shelters that I do not know.
The Suketobashi area had fewer shelters.
A signboard on an electricity pole often shows above-sea level.
Even distant from the coastline, many places will be damaged by tsunamis.
The capacities of the condominium or apartments are smaller than I expected. AK3 Signboards show the above-sea levels.
Tokushima University can accept 1,800 people per shelter.
I should not forget the above-sea level and tsunami inundations. AK: acquired knowledge, PA: participants in Group A, PB: participants in Group B (2024) 19:9  However, for Q4, the mean value in Group B was 1.2 points higher than that in Group A.
The mean values of Q5-Q7, which were only given to Group B, were higher than 4.0.

Discussion
Knowing shelters is considered acquiring shelter-related knowledge. To answer the two research questions, we focus mainly on participants' recalled shelter location and capacity, as well as registered and acquired knowledge.

RQ1. Which promotes knowing shelters more, the single-player mode or multiplayer mode (Shelter GO)?
First, we focus on shelter location and capacity that can be evaluated as quantitative data.
Although Group B played the game approximately 10 min longer than Group A, participants in both groups visited a similar number of shelters (Num_VS) and tsunamirelated spots (Num_VTRS), indicating that Group A participants visited shelters and tsunami-related spots more efficiently than Group B participants. The longer gameplay time of Group B participants resulted from the time spent on knowledge exchange because knowledge exchange entailed face-to-face interaction and additional app operations. In the four-phase process, visiting shelters is the prerequisite for knowing shelters. If deciding only from Num_VS and Num_TRS (i.e., frequencies of VO and VC phases) whether to promote knowing shelters, we assumed that both single-player (Group A) and multiplayer modes (Group B) were effective and had no remarkable difference. (2024) 19:9 In the recall test, Group A participants remembered more shelter locations than Group B participants (10.0 points higher in AccRate_ASL and 6.9 points higher in AccRate_VSL). Under the experimental conditions (i.e., five participants in a 700 m 2 area within an average of 52.6 min), encountering another player and exchanging knowledge was rare.
We found one case where knowledge exchange may have successfully affected knowing shelters (knowledge acquisition). On the prequestionnaire, PB2 itemized two pieces of knowledge that had already been acquired before the experiment: • My apartment was assigned as a tsunami shelter. During the earthquake, I evacuated to a higher floor of my apartment.
• The tsunami may reach fast. Thus, I evacuate quickly.
On the postquestionnaire, PB2 itemized three pieces of knowledge acquired using the gameplay: Answer to RQ2: Knowledge exchange may not be done regularly, but it does promote knowing shelters. Exchanged knowledge not only allows a player to receive knowledge from other players but also motivates the player to visit shelters that were not initially intended to visit.

Possible improvement
Shelter GO originally aims to promote knowing shelters, i.e., acquiring shelter-related knowledge. However, in the preliminary experiment, participants in both groups acquired not only shelter-related knowledge but also evacuation and disaster knowledge. Tsunamirelated knowledge occupied approximately 30% and 26% of their registered and acquired knowledge, respectively. This is because knowledge registration was performed during the VC phase, i.e., tsunami-related spots in the experiment. Focusing on the original goal, we should change the opportunity for knowledge registration from the VC phase to the VO phase. A player must register shelter-related knowledge to knowledge stored in the app when observing a shelter rather than when collecting a digital creature; from the knowledge storage, the player selects knowledge to be registered to a digital creature. If we are focusing on acquiring wider knowledge, we should allow players to register knowledge during the VO and VC phases. In addition, knowledge registration should allow the uploading of multimedia data (e.g., picture, voice, and video) because text-based knowledge registration can be difficult to represent shelter location and features.
In the experiment, participants in Group B exchanged knowledge at least once, but we cannot ignore the fact that they were acquaintances who tried exchanging knowledge whenever they encountered it. When Shelter GO is released to the public in the future, players may have difficulty speaking to strangers, even if they recognized them as other players. To promote knowledge exchange, the Shelter GO app should encourage speaking to stranger players. For example, the app is expected to display knowledge-exchangeable players on a digital map while being careful about privacy issues. (2024) 19:9

Limitations
The most significant limitation was that the small number of participants was demographically restricted (i.e., ten acquaintance undergraduate students). The obtained data could not be analyzed using statistical methods. Furthermore, participants were not examined and homogenized in terms of their characteristics (e.g., physical condition, perceptual trait, and motivation) between the groups. The small number of Num_RK may indicate that participants have not been accustomed to itemizing their acquired knowledge.
Thus, the findings may not be versatile or reliable. Five participants in Group B may have been difficult to be encountered within approximately 1 h, despite the small experimental area (700 m 2 ). If the five participants were not acquaintances, they may not be encountered.
If long-time gameplay was allowed in a wider experimental area, participants may have visited and known more shelters. In other words, the gameplay time was insufficient to collect many digital creatures, and participants may have abandoned the game before they found enjoyment in the digital creature collection. Participants in Group B may not have felt motivated to acquire a unique digital creature through digital creature exchange due to the small number of their collected digital creatures.
Considering these limitations, we must carefully apply the findings to the improvement and practical use of Shelter GO.

Conclusion
This study described Shelter GO, a multiplayer location-based game, which promotes players to know shelters (i.e., shelter location, capacity, and feature) according to the fourphase process, including the Preview, Walk, VO, and VC phases. The Shelter GO enables players to not only earn points in the VO phase and collect digital creatures in the VC phase but also exchange their collected digital creatures with encountered players. Digital creature exchange involves knowledge exchange, which entails knowledge registration of their collected digital creatures. The preliminary experiment implied that Shelter GO can promote knowing shelters, and knowledge exchange may not be frequently done but promote knowing shelters. The small-scale experiment had some limitations and was insufficient to clarify the effect of Shelter GO. Therefore, we must conduct large-scale experiments involving a diverse range of participants. In addition, we will improve the Shelter GO app while refining the visual design (e.g., more attractive digital creatures) to make it more appealing to the public.