Map API-Based Evacuation Route Guidance System for Floods

Abstract

Recently, human casualties and property damage caused by natural disasters have increased worldwide. Among these natural disasters, flood damage is affected by season. Depending on the concentration of precipitation in the summer, heavy rainfall can occur, thus resulting in typhoons, floods, and increased damage. To prevent such damages, the appropriate measures and research are being conducted in response to disasters. When a flash flood occurs, safe evacuation can be realized after detecting the situation and using announcements or laser indicators. However, these route guidance systems are typically used in fire or indoor environments, thus rendering them difficult to access outdoors. Therefore, we herein propose an evacuation route guidance system based on a map API that recognizes flood occurrences in forest areas, recreational forests, and parks. It calculates the route based on the map API and delivers the evacuation route to the nearest shelter to the user; and if there is a second problem on the moving evacuation route and it is difficult to proceed, the user’s current location is identified and the route to the next nearest shelter is provided. This will help you to evacuate safely.

1. Introduction

Among the recent natural disasters, floods have caused an increase in the level of human casualties and financial damage. From 1998 to 2017, among the disasters that occurred worldwide, the most damaging natural disaster is flood, which constituted 43.4% of all disasters. The classification of casualties due to flood damage is 11%, whereas casualties account for 45% of all natural disasters [1,2,3]. The increase in damage from natural disasters from 1998 to 2017 is a harbinger of the greater risks expected in the future. Modern society is becoming larger and more urbanized, so cities are enlarged and forest areas are lost, exposing them to flood damage. Inundation damage in Korea is mainly caused by increased precipitation due to heavy rains in summer and monsoon season. Most of these floods affect large bodies of water or riverbanks that damage surrounding areas [4,5]. There are also many social and structural causes, such as heavy rains. July–September is the rainy season in Korea, during which heavy rains and typhoons pass through the Korean Peninsula. In areas where forests have been lost because of scaling and urbanization, precipitation and flood damage have been increasing significantly. Heavy rain has caused the nearby Han River to overflow and flood the city on account of its topographical structure.

Furthermore, low-pressure areas where the air rises and lowers the pressure in the atmosphere coincide with tsunamis, whereby water enters and returns from the ocean, resulting in coastal inundation [6], as well as the rapid inundation of forests and mountainous areas [7,8]. These phenomena occur intermittently with landslides on the river or the tactical recovery side [9,10]. The human costs of disasters from 2000 to 2019, as published by UNDRR [11], are the highest recorded thus far.

Table 1. Total number of disasters by type: 1980–1999 vs. 2000–2019 [11].

1980–1999 Total number of accidents by type
Drought	Earthquake	Extreme temperature	Flood	Landslide	Mass movement	Storm	Volcanic activity	Wildfire
263	445	130	1380	263	27	1457	84	163
2000–2019 Total number of disasters by type
Drought	Earthquake	Extreme temperature	Flood	Landslide	Mass movement	Storm	Volcanic activity	Wildfire
338	552	432	3254	376	13	2043	102	238
Difference in total number of accidents by type between 1980–1999 and 2000–2019 (increase or decrease)
Drought	Earthquake	Extreme temperature	Flood	Landslide	Mass movement	Storm	Volcanic activity	Wildfire
75	107	302	1874	113	−14	586	18	75

shows the total number of disasters by type over 20 years, i.e., 1980–1999 vs. 2000–2019;

Table 2. Disaster impact: 1980–1999 vs. 2000–2019 [11].

1980–1999	(Unit: 1 million $)
Reported disasters	Total deaths	Total affected	US$ Economic dosses
4212	1.19 T	3.25 T	1.63 T
2000–2019
Reported disasters	Total deaths	Total affected	US$ Economic losses
7348	1.23 M	4.03 B	2.97 T
Change in disaster impact (increase or decrease)
Reported disasters	Total deaths	Total affected	US$ Economic losses
3136	−1.18 T	−3.24 T	1.34

shows the disaster impact in 1980–1999 vs. 2000–2019.

The number of climate-related disasters has increased significantly. Table 1 shows the total disaster events by type for the years 1980–1999 vs. 2000–2019. Based on years 2000–2019, among the disasters, flood caused the most damage. Meanwhile, based on Table 2, which shows the disaster impact for years 1980–1999 vs. 2000–2019, the number of casualties arising from disasters is higher in the period 2000–2019. Because these natural disasters are becoming more complex and human casualties are increasing, research [12,13,14,15] is currently being conducted to develop evacuation systems that can minimize flood damage using ICT [16,17,18].

As for existing evacuation systems [19,20,21], researchers are currently attempting to develop evacuation systems using the shortest distance algorithm [22,23,24,25]. However, if more accurate routes and search times can be displayed using a map API implemented based on these algorithms, then human casualties can be reduced by providing more accurate evacuation routes compared with using existing systems. To reduce damage caused by flood, flood routes must be predicted and provided [26,27]. However, many variables must be considered in flood forecasting, and correlating them spatially and temporally is difficult, thus rendering forecasting challenging. For flood prediction, a hydrological model was conducted as a preliminary study based on the testbed data hydrology. The model physically represents the discharge flow according to the characteristics of the sluice gate. Flood forecasting models predict water levels and runoff. In this study, an evacuation system is proposed based on data analysis and a water-level prediction model implemented in a preliminary study [28]. When the results predicted by the flood prediction model are delivered to the evacuation system, it provides a route to the shelter within the user’s GPS range and the shortest distance. Additionally, the shortest distance algorithm [29], level, and meteorological data were used as input data. Based on this, we propose an evacuation route system that predicts flash floods and simultaneously uses a map API to identify the nearest shelter by identifying the user’s current location route. To develop the shelter system, an appropriate testbed was selected through field investigation; and shelter data in the testbed area were created, stored, and utilized as a dataset. In this regard, we propose an evacuation system that creates a route-search environment using a map API and provides users with the shortest route search.

Section 2 presents the research methodology, Section 3 presents the system design of this study, and Section 4 presents the system implementation and considerations. Finally, Section 5 presents the conclusions of the study.

2. Research Methodology

2.1. Training Data

Flooding of valleys and rivers is one of the main causes of flooding due to torrential downpours over short durations. Public data provided by the Ministry of Public Administration and Security [30], which is a national institution, were used for areas affected by flooding. The Ministry of Public Administration and Security in South Korea is responsible for safety and disaster policies, and the public data provided included data from the past 30 years. Heavy rain and floods shift rivers and valleys upstream, midstream, and downstream, thus causing flood damage due to heavy rain to occur downstream [31,32]. Figure 1 shows the amount and frequency of rainfall damage by the administrative district in Korea. The administrative divisions of Korea are as follows: metropolitan government (Seoul), metropolitan cities (Busan, Daegu, Incheon, Gwangju, Daejeon, and Ulsan), provinces (Gyeonggi, Gangwon, Chungbuk, Chungnam, Jeobuk, Jeonam, Gyeongbuk, and Gyeongnam), and a special autonomous province (Jeju).

The level of damage restoration performed in Gyeonggi-do was the second highest among domestic cities, and rainfall frequently occurred in all cities. In addition, Gyeonggi-do surrounds the capital of Korea, which has the highest density; therefore, its population is relatively large. Figure 2 shows the total population and population density by state and province.

2.2. Flood Evacuation Analysis

In August 2022, heavy rain occurred nationwide, primarily in the central region, which resulted in human fatality and property damage. Densely populated large cities are flooded by heavy rains; and apartments, houses, shops, and roads are submerged, thus causing traffic disruptions, power outages, and significant damage. In addition, many disasters occur due to floods caused by bank collapse arising from heavy rains and typhoons as a result of climate change [31,32]. The Ministry of Public Administration and Security of Korea collects and analyzes information related to flood risk areas, weather forecast, and water level, such as observational data, to assess the degree of risk. This evaluation is performed considering the possibility of flooding, the scope of impact, and the degree of damage. This determines which regions or infrastructures are most at risk.

In modern society, as cities develop, urbanization intensifies in many areas; these cities become larger and forests are diminished, thus resulting in a high risk of flood damage. Korea is actively conducting research pertaining to flood damage while considering climate change and have implemented various countermeasures.

Table 3. Current status of damage and restoration cost for the past 10 years [31,32].

Division	2012	2013	2014	2015	2016	2017	2018	2019	2020	2021	Sum	Average
damage	1,089,210	172,137	180,019	31,862	288,862	187,302	141,284	216,226	1,318,17	66,053	3,691,132	369,113.2
recovery cost	2,053,176	386,559	507,065	38,122	590,607	499,672	443,270	1,348,759	4,161,548	297,322	10,326,100	1,032,610

shows the state of loss and restoration costs over the past 10 years, and

Table 4. Status of casualties by cause for the past 10 years [31,32].

Division	2012	2013	2014	2015	2016	2017	2018	2019	2020	2021	Sum	Average
heavy rain	2	4	2		1	7	2		44	3	65	6.5
typhoon	14				6		2	18	2		42	4.2
typhoon, heavy rain							1				1	0.1
heavy snow
heat wave							48	30	29	39	146	36.5
sum	16	4	2		7	7	53	48	75	42	254	25.4

The current status of casualties for the rainy season of 2019–2021 was prepared based on the statistics on the causes of death from the National Statistical Office [31].

shows the state of human damage over the past 10 years.

Statistics compiled by the Ministry of Public Administration and Security show the status of damage, restoration costs, and casualties over the past 10 years (2012–2021) [31]. Property damage amounted to KRW 3691.3 billion, accounting for most of the total damage. Although these studies are conducted annually, heavy rain and flood damage during the monsoon season increase annually. The course of a flash flood depends on the intensity, duration, and local characteristics of the heavy rainfall. Urban flooding is directly caused by heavy rainfall during the monsoon season. Water can enter a house, such as a semi-basement, during the rainy season, resulting in loss of life and property damage. However, in forested areas, large amounts of rain are absorbed by soil. The absorbed rainfall is not visible to the naked eye until the groundwater coincides with the underground surfaces. As this is not direct urban flooding, the flooding in forested areas is delayed; however, the delay does not imply that there is no damage.

To solve this problem, the government is attempting to establish flood control measures to safely evaluate nearby people during flooding. Therefore, in this study, we propose a route-search system using an API map for flash floods, based on the flood discrimination model used in a previous study.

In the water level prediction model [33] conducted as a preliminary study, the water levels resulting from rain at the upper and lower valley locations in the testbed area were overturned, and the occurrence of floods was predicted. The user’s current location inside the testbed was identified and guided. The water level prediction model provides information on the location of a flash flood, displays an evacuation route from the user’s location to a nearby shelter on a map, and transmits this information to a smart device. Accordingly, the user checks the provided evacuation route and proceeds safely.

2.3. Testbed Comparison and Selection

In testbed selection, the criteria used were the number of evacuation routes, existence of a valley, and flood damage in the testbed (valley or river). In each testbed region, including Gangwon-do, there were no locations satisfying any of the three conditions. Therefore, a testbed area that was a valley with a sufficient number of evacuation routes was selected. Figure 3 shows the location chosen as the testbed area. Figure 4a shows a blank map of Yeoju-si, the study area; and Figure 4b shows a blank map of Seoul and the surrounding Gyeonggi-do in Korea.

2.4. Map API Comparison and Selection

A map API was used to retrieve routes for pedestrians. Google Maps and OpenStreetMap are available in Korea. However, they were excluded because they could not accurately visualize the pedestrian route search. Therefore, based on a comparative analysis of Naver Map, Kakao Map, and T-Map, the T-Map API was selected as the map API for this study. Naver Map supports a walking route search in mobile applications, but is not supported in the computer version; therefore, a walking route search was not provided in the testbed area. Kakao Map identifies the most accurate route, but cannot be used outside the function provided by the API. T-Map does not output as accurately as Kakao Map, but supports the walking route search function API and can be used for development. Hence, the T-Map API was used in this study.

2.5. Indoor and Outdoor Evacuation Time Analysis

Evacuation strategies should be improved by comparing indoor and outdoor evacuation times. It is necessary to explore the factors affecting the evacuation time and the results of previous studies. The testbed area must be identified to provide evacuation times and routes. To provide an indoor evacuation route, only the structure of the building needs to be identified; however, to provide an outdoor evacuation route, it is necessary to determine the topography. As the proposed system was designed for outdoor evacuation, a field survey was conducted in this study to determine the topography of the testbed area. The outdoor evacuation time of the testbed was analyzed and converted into JSON format through pre-processing. Based on the model, the result is analyzed, and the value is transmitted to the manager system; in addition, if it falls within the range that a flood will occur, the user’s smartphone identifies the current location with a danger warning and provides the distance to the nearest shelter. Moreover, if secondary damage due to flooding occurs on the moving route, the current location of the user is re-identified, the evacuation route is re-searched excluding the existing shelter, and the evacuation route is again provided to the user.

3. Design and Requirements Analysis

Because location information must be acquired during natural disasters or floods, we searched for the location after obtaining prior consent. Figure 5 shows the configuration of the proposed system. The shortest-distance calculation method using the T-Map API was designed and implemented to search for an evacuation route to the nearest shelter. The T-Map API was used to print the T-Map API. A comparative analysis of the API and related information is described later herein. The data stored in the database contain user location and recreation forest information. In addition, by creating a spring controller on the web to handle web requests, the model receives GPS data and converts them into a GPX file representing the terrain data integrated into roads, buildings, and ecosystems. Therefore, the services provided by this system provide information regarding recreational forests and evacuation routes.

The figure shows the process to inform users of a flash flood or disaster situation in advance, obtain location information, and receive location information through a GPS. The user’s GPS data are received and viewed by the monitoring system and stored in a database. Existing databases create and store the terrain and building data. Therefore, based on the data stored in the database, the location is identified and prepared for unexpected situations. If an emergency does not occur, then the user’s location is identified and an emergency evacuation route is searched to provide the user with a flood risk warning and an evacuation route to the nearest shelter. Figure 6 shows a flow diagram of the system.

3.1. Testbed Area Graph Design and Dataset Creation

The shortest path search requires nodes and edges; thus, a graph of the testbed area was created. The testbed area was first selected based on the walking complexity, followed by flood damage, and then deep and long valley areas. The testbed area was the Yeojubo River, which was near a park and comprised a bridge over a river; therefore, the water depth of the river was large, and the river was at a high risk of flood damage. Therefore, shelters in the testbed area were selected and displayed as nodes, and a tree-like graph with edges was constructed. Figure 7 shows a graph of the testbed area.

Figure 7 shows the form of a tree structure created by connecting the nodes of the evacuation route. The number of each node does not have much meaning, but it is shown to indicate that it is the number of nodes. Buildings were created in a single layer to form a dataset with shelters. The created dataset contained basic information regarding the shelter, latitude, and longitude, which were output on the map. The latitude, longitude, and location name of each location were saved, generated as a single dataset, and saved as a .csv file. Additionally, based on the testbed area, the model was designed using terrain data and a shelter dataset. Figure 8 shows the generated dataset, and Figure 9 shows the dataset in the testbed area.

3.2. Flash Flood Scenario

We propose a shortest-path detection system based on the JSON data of the testbed path. Scenarios in which users were routed were simulated. The current location was visualized when a user was detected. Two methods were used to provide a path; i.e., evacuation routes were searched during flooding and the current location was identified using GPS signals only when no flooding occurred. Meanwhile, the evacuation route was identified by considering the shortest path from the user’s current location to the nearest shelter and the path that allows the user to escape the testbed area from his/her current location.

Table 5. Evacuation scenario in case of flash flood.

Scenario
User detection	I agree with the location information to the user.
Situation detection	Identifying the location based on user GPS data
	1. If a flood occurs
	- Identify the evacuation route.
	2. If no flood occurs
	- Identify the location based on user GPS data.
Identifying the evacuation route	Shortest path provided
	1. Provides the shortest route from the user’s current location to the nearest shelter
	2. Provides the shortest exit route from the testbed area based on the user’s current location
	If an unexpected situation occurs in the movement route, then the route is rediscovered at the current location and provided.

lists the possible flood scenarios.

3.3. Flood Prediction Model [32]

In this study, a predictive model was designed and implemented as a preliminary study, and a system was designed based on it. Three structures were used for model design. In [None, 1], the three models were composed of [None, 2] using the input data and training data for 20 h; and S2 and S3 were composed of [None, 5]. Figure 10 shows the configuration of the proposed predictive model.

3.4. Database Design

The data stored in the database were as follows: recreational forest information and the location and name of a shelter in a recreational forest. The evacuation route was stored and used in the form of JSON by calculating the evacuation route in the case of a flash flood after identifying the current location using GPS signals. Additionally, recreational forest and terrain data were stored in the database. It shows a 1/n relationship with ID, Shp, and GPX data, as shown in Figure 11.

4. Implementation and Considerations

In this study, an evacuation-route search system was implemented based on a map API. A suitable T-Map API was selected based on a comparative analysis of the map API in advance. A map was printed using the selected map API, markers were created, and the shortest route was identified based on route calculations. The route to the searched shelter was provided to the user, and the system development environment for outputting the route is shown in

Table 6. Development environment.

Type	Composition
OS	Window 11
CPU	Intel i7-9700
GPU	Nvidia Geforce RTX 3060
RAM	16 GB × 2
Tool	eGovFrame
Language	Python 3.9, JAVA

.

4.1. Implementation of Map Output and Evacuation System

Information regarding the Korean coordinate system was prepared using seven variables of the Bursa–Wolf model of the “National Coordinate Conversion Factor” implemented by the National Geographic Information Institute (No. 2002-433) in December 2002 for ellipsoid transformation, among which EPSG:5178 (which is a coordinate system) was selected and used. In the KATEC series, the UTM-K (Bessel) coordinate system representing the entire Korean Peninsula can be used as a new address map. By selecting EPSG5178 as the coordinate system, the map and data were visualized, and the testbed area was printed on the map. The aforementioned public data were classified into four types of terrain data that were selected as testbeds. Instead of a single-file format, it was saved in a shape file in an extended format and included DBF, SHP, and SHX. Data were saved in the SHP format, which is a shape file, and a building layer was created and output. The map API optimized for walking uses the T-Map API by providing an evacuation route for walking to accommodate the characteristics of the recreational forest selected as a testbed. Meanwhile, the route was searched using the route calculation method, which was expressed in JSON form and stored in the database. Figure 12 shows the map output obtained using the map API, and Figure 13 shows the starting point of the shelter.

Figure 14 shows a map linking the flood prediction models. In a preliminary study, a flood prediction model was implemented and a risk warning was transmitted to the system by selecting a flood risk range. Using the transmitted data, the system output a flood-risk warning on a map and displayed it on the output data. The corresponding terrain data were output, the coordinate system was selected, and the flood range was output on the map via flood prediction. When this range reached an arbitrary value in the water level data, it was displayed on the map as the first warning step, and then the locations of nearby users were identified.

The user’s GPS data were only monitored and not stored data in the system during normal times, whereas they were obtained, stored, and used during flash floods. The user’s location was identified using the acquired data, and an evacuation route to the shelter was provided for each location, or the existing route was stopped when an unexpected situation occurred on the moving route. The user’s current location was identified, and existing shelters were limited to searching for the nearest shelter. The building of the testbed was selected as a shelter and then saved as a dataset. The saved shelter dataset is displayed on the map as a marker, and the risk ranges predicted based on the model, route search, and provision are shown in Figure 15.

Figure 16 shows the evacuation route in JSON format, and Figure 17 shows the final result of the proposed system. The map on the system’s administration website provides a visualization of the current situation, allowing one to determine whether the correct route is to be provided to the user.

4.2. Review

In modern society, damages caused by floods are increasing every year as forests disappear and extensive modernization is in progress. Flash floods can cause deaths and property damage. In this paper, we propose a route-search system for flash floods.

The data were obtained through a testbed survey. The requirements of the test district were selected based on the presence or absence of an evacuation route, sufficient walking distance to provide various traffic lines; and valleys, streams, and rivers as auxiliary routes. Based on the walking path search system, the T-Map API was selected and used as an API optimized for walking.

A map was printed using SHP of the testbed area, public data of the corresponding administrative district, and T-Map API. A building that can be evacuated was selected as a shelter on the printed map, a dataset was created and saved, and markers were displayed on the map. The flood identification model used is also shown on the map. Based on the disaster scenario, it determines the user’s current location in the event of a flood and provides a route beyond the flood area to the nearest shelter.

The latitude and longitude values appearing in the JSON file of the route provided in this study were the same as those of the actual GPS. Therefore, the search time proposed in this study for floods affects the search for the nearest shelter in the evacuation route system and depends on whether the evacuation route is provided. However, an evacuation route was provided by the proposed system, which prevented secondary accidents.

5. Conclusions

Recently, disasters such as earthquakes, floods, and landslides occurring worldwide have increased human fatalities and property damage. In fact, 43.4% of all disasters between 1998 and 2017 were caused by flooding, which includes the case in Korea, a country that is experiencing rapid changes in terms of climate emergencies and crises. Korea is affected by severe flood damage caused by typhoons in the summer or during the rainy season from July to September. Flooding can be classified into four main categories: urban, coastal, fluvial, and flash. Evacuation areas, such as flooding areas in cities, coasts, and rivers, are wide; and movement is not restricted therein. However, flash floods occur as landslides in narrow valleys, rivers, and steep slopes in forested mountainous areas. These flash floods increase the land surface area, as rain from the monsoon season or typhoons absorbs into the surrounding soil. In this paper, we propose a path-detection system based on the shortest distance for flash flooding. Before designing and implementing the proposed system, we conducted a site survey to determine the testbed selection criteria. The first was an area with many evacuation routes, and the second was the location of the area. Owing to the location of the river and surrounding park, we predicted that the damage to the area would be considerable during flooding. Therefore, identifying a test center that satisfied these requirements based on field trips to various regions of the country, including Gangwon-do, was challenging. Hence, we selected Yeojubo as the test center. In addition, by searching for a route while walking, the shortest shelter from the user’s current location to the destination shelter was identified, and the shortest evacuation route was selected.

The maps were printed using a map API, and the real-time location was verified after obtaining the user’s consent. The map contained information regarding the available evacuation shelters during a disaster, which was based on a shelter dataset and was created as a marker to confirm the location. A flash flood risk warning was delivered to this system in the implemented flood prediction model. The delivered flood hazard path was specified on a map to indicate the extent of the flood. This system determines the user’s current location in the same manner as the danger warning system. The system transmits and receives GPS data, stores them, and accesses a database. The database stores topographic data and shelter datasets and uses them to search for the nearest shelters from the current location. The shortest distance to the evacuation route was identified and output to the system. Simultaneously, a safe route to the shelter was provided by providing information to the user. Notably, the system was designed such that the nearest shelter or evacuation route was outside the flood risk range.

For example, when the first shelter in the user’s current location is outside the flood range, if the existing route is used, then damage may occur due to flooding, in which case the system performs a search by excluding the first shelter. In addition, when the flood area spreads while it propagates along the provided route and a problem occurs in the movement route, then the existing shelter and evacuation route are canceled; and the user’s current location is identified and searched. Accordingly, a system was implemented to prevent secondary accidents by providing shelters and evacuation routes. In future studies, we plan to add more routes to reduce the evacuation route search time.