Deep learning-based study on assessment and enhancement strategy for geological disaster emergency evacuation capacity in Changbai Mountain North Scenic Area

Abstract

This study focuses on the northern scenic area of Changbai Mountain, aiming to evaluate the emergency evacuation capacity of the region in the context of geological disasters and to formulate corresponding improvement strategies. Due to the relatively small area of this region, difficulties in data acquisition, and insufficient precision, traditional models for evaluating emergency evacuation capacity are typically applied to urban built environments, with relatively few studies addressing scenic areas. To tackle these issues, this research employs the Real-Enhanced Super-Resolution Generative Adversarial Network (Real-ESRGAN), which successfully resolves the problem of blurriness in remote sensing images and significantly enhances image clarity. Coupled with the Graph Convolutional Network (GCN) model, the study calculates the emergency evacuation time for each raster point, providing a comprehensive assessment of the region’s evacuation capacity. Based on the evaluation results, the study proposes targeted improvement measures for areas identified as having poor emergency evacuation capacity, taking into account the existing infrastructure of the scenic area. By constructing an indicator system encompassing effectiveness, accessibility, and safety, the feasibility of each proposed enhancement strategy is assessed scientifically and rationally. Through these integrated tools and methodologies, this research significantly improves the accuracy of data processing, evaluation, and decision support, showcasing a comprehensive approach to scenic area research that provides critical support for geological disaster management, emergency planning, and the overall safety of the Changbai Mountain scenic area.

Introduction

In recent years, the intensifying effects of global climate change have led to frequent natural disasters, significantly affecting regional economic development and urban safety^1,2. As a result, evaluating emergency evacuation capabilities has become increasingly important. Established in the early twentieth century, modern methods for assessing these capabilities have evolved throughout the century and into the present³. The international community has highlighted the need for effective emergency evacuation plans in response to geological disasters like landslides and mudslides, focusing on evacuation routes, zones, and personnel training. For example, R. Melo et al. (2020) modeled pedestrian evacuation under mudslide scenarios in Portugal, estimating the time to reach at-risk buildings⁴. Similarly, Yin et al. (2022) emphasized resource and traffic path considerations to improve evacuation efficiency⁵, while Tang et al. (2023) planned urban emergency shelters to enhance disaster mitigation⁶. Emerging technologies, such as remote sensing, GIS, and deep learning, are increasingly applied in geohazard management to identify risks and refine evacuation strategies.

Despite significant advancements in this field, certain issues demand attention. Firstly, there is a deficiency in convergent research, and a scarcity of studies integrating the impacts of various emergencies and multiple factors. This limitation impedes a comprehensive understanding of complex disaster scenarios. Second, interdisciplinary research needs further strengthening. Emergency response and evacuation are interdisciplinary fields necessitating a more comprehensive integration of disciplinary knowledge. Finally, a majority of existing studies cover extensive study areas, resulting in outcomes lacking in detail and precision. This limitation hinders an in-depth understanding of specific locations or scenarios, making it challenging to study particular areas or issues in depth. At times, selecting a smaller and more confined study area for in-depth analysis may yield more precise results. Therefore, this paper emphasizes these aspects to address knowledge gaps in the current research area and enhance emergency response and evacuation, ensuring societal stability and safety.

This study focuses on the emergency evacuation capacity of smaller study areas and employs deep learning methods to enhance the comprehensiveness and accuracy of the assessment. We utilized the Real-Enhanced Super-Resolution Generative Adversarial Network (Real-ESRGAN) to address the blurriness of remote sensing images in the smaller study area through super-resolution processing. Following the application of the Graph Convolutional Networks (GCNs), we conducted a pixel-level assessment of the emergency evacuation capacity⁷. This approach enables a more detailed analysis of evacuation capabilities across different regions within the study area, thereby providing a more accurate understanding of the conditions in the area of interest. In the final phase of the research, we proposed a series of improvement measures based on the analytical results, selecting relevant indicators to evaluate the effectiveness, accessibility, and safety of these measures⁸. The objective was to enhance the emergency evacuation capacity of the study area. The results indicate a significant improvement in the evacuation capacity of the scenic area, enabling a more effective response to emergencies and disaster events. These experiences and methodologies provide valuable references for assessing the emergency evacuation capacities of other similar research areas. Figure 1 illustrates the technical steps employed in this study.

Fig. 1.

Flow chart for evaluation of the emergency evacuation capacity of Changbai Mountain North Scenic Area.

Study area

Changbai Mountain, being the highest volcano in China, possesses a steep topography rendering it susceptible to landslides, particularly during periods of intense rainfall, thereby posing a threat to the surrounding area^7,8. Situated in the southeastern mountainous region of Jilin Province, China, the Northern Changbai Mountain Scenic Area spans geographic coordinates of 127°28’ to 128°16’ east longitude and 41°42’ to 42°25’ north latitude. The highest peak, Tianwen Peak, is situated northeast of Tianchi and reaches an altitude of 2,670 m. Composed of grey and light white pumice, this peak forms a "U"-shaped valley with Longmen Peak. The waterfalls of Changbai Mountain serve as the source of the Erdaobai River, and the landscape is characterized by volcanic and tectonic erosion. The valley’s sides are shaped by volcanic cones and their extensions, while the canyon area is a gully formed by the Erdaobai River fault. As shown in Fig. 2. Nevertheless, geological disasters, including Collapse hazards and Debris Flow Hazards, frequently occur in this area, resulting in substantial economic losses to the scenic area and posing risks to the lives of tourists. Hence, there exists an imperative requirement to assess the emergency evacuation capacities within the Changbai Mountain region^9,10.

Fig. 2.

Geographic location of the study area.

Data sources and methodology

Data sources

Data utilized in this study were sourced from various dependable repositories and platforms. Vector data on building density, emergency response capacity, road density, and medical response capacity were obtained from the Data Harvest website and the Planning Cloud website. Raster data for Gross Domestic Product (GDP) in the year 2022 were acquired from the Data Harvest website. Data on elevation, slope, Fractional Vegetation Cover, and land use/land cover (LULC) were sourced from the Geospatial Data Cloud and the Data Harvest website. Furthermore, vector data on construction roads and rivers were obtained from the Data Harvest website. Information on passenger flow, measured in tens of thousands, was provided by the Management Committee of Changbai Mountain Protected Development Zone in Jilin. Table 1 provides a summary of the sources of indicators and data utilized in this study.

Table 1.

Indicators of emergency evacuation capacity evaluation model for Changbai Mountain North Scenic Area and its data sources.

SI.No	Parameters	Data Types	Source
1	Building Density	Vector data	https://www.databox.store (accessed on 16 January 2023)
2	Emergency rescue capability	Vector data	https://www.guihuayun.com (accessed on 10 January 2023)
3	Road density	Vector data	https://www.gscloud.cn (accessed on 18 November 2022)
4	Medical Rescue Capability	Vector data	https://www.guihuayun.com (accessed on 10 January 2023)
5	GDP	Raster data 2022	https://www.databox.store (accessed on 16 January 2023)
6	Elevation	ASTER GDEM 12.5 m × 12.5 m	https://www.gscloud.cn (accessed on 1 September 2022)
7	Slope	ASTER GDEM 12.5 m × 12.5 m	https://www.gscloud.cn (accessed on 1 September 2022)
8	Fractional Vegetation Cover	Landsat 8 OLI/TIRS	https://www.databox.store (accessed on 3 October 2022)
9	LULC	Landsat 8 OLI/TIRS	https://www.databox.store (accessed on 13 October 2022)
10	Plank road	Vector data	Gaode Map
11	River	Vector data	Gaode Map
12	Passenger flow	10,000 people	Jilin Changbaishan Protected Development Zone Management Committee

Methodology

Real Enhanced Super Resolution Generative Adversarial Network (Real-ESRGAN)

Real-ESRGAN, short for Real-Enhanced Super-Resolution Generative Adversarial Network, is a deep learning model designed for Image Super-Resolution (SR)¹¹. It utilizes a Generative Adversarial Network (GAN) architecture comprising two components: a generator and a discriminator. The generator’s role is to transform low-resolution input images into high-resolution output images, with the discriminator assessing the realism of the generated images¹². Via adversarial training, the generator and discriminator engage in competition, driving the generation of increasingly realistic and detailed high-resolution images. Key innovations of Real-ESRGAN encompass the utilization of real-world, high-resolution images for pre-training and the incorporation of a novel Perceptual Loss function. These enhancements empower Real-ESRGAN to produce high-resolution images that are more realistic and detail-rich across a broad spectrum of domains¹³. It holds value in applications such as image enhancement, medical image processing, satellite remote sensing image analysis, and more¹⁴.

The following are the steps associated with Real-ESRGAN, which include the architectures of the generator and discriminator as well as the mathematical formula expressions for the loss functions involved in the training process.

1. Generator:

The task of the generator is to convert the low-resolution input image x into a high-resolution output image y. Equation (1) represents the output high-resolution image y. The generator may involve multi-layer convolution and upsampling operations.

1

where G denotes the generator network, denotes the parameters of the generator, denotes the hidden layer operation of the generator, and denotes the output layer operation of the generator.

2. Discriminator:

The task of the discriminator is to evaluate the similarity between the generated high-resolution image y and the real high-resolution image z. The discriminator is a simple and efficient method to evaluate the similarity between the generated high-resolution image and the real high-resolution image. Equations (2) and (3) are the scores of the discriminator, which usually involves convolution and pooling operations to evaluate the similarity of the two images.

2

3

where D denotes the discriminator network and denotes the parameters of the discriminator.

3. Loss Functions:

When training Real-ESRGAN, multiple loss functions are usually used to guide the training of generators and discriminators. Equation (4) is Perceptual Loss, which is used to measure the content difference between the generated image and the real image, which can be a pre-trained convolutional neural network. Equation (5) is Adversarial Loss, which is used to encourage the generator to produce more realistic images to deceive the discriminator. Equation (6) is Regularisation Loss, used to control the smoothness of the generated image.

4

5

6

where and are the representations of the images y and z in some feature, and denotes the square of the Euclidean distance between these two feature representations; denotes the gradient of y, and denotes the square of the L2 parameter of the gradient.

Real-ESRGAN is an image super-resolution model that operates through a series of steps, transforming a low-resolution input into a high-resolution output by sequentially processing feature extraction, detail enhancement, upsampling, and authenticity assessment. Initially, the low-resolution image undergoes feature extraction in the generator, capturing foundational textures and shapes. Subsequently, residual-in-residual dense blocks (RRDB) enhance high-frequency details, improving image clarity. The upsampling layer then converts low-resolution feature maps into a high-resolution image. This generated high-resolution output is processed by a discriminator, which extracts local and global features to evaluate image authenticity via an activation function. The discriminator’s feedback guides the generator’s optimization, enabling the model to progressively produce more realistic high-resolution images. Real-ESRGAN model processing is shown in Fig. 3.

Fig. 3.

Real-ESRGAN processing flow.

Graph Convolutional Models (GCNs)

Graph Convolutional Networks (GCNs) find broad applications in the realm of remote sensing image processing¹⁵. Remote sensing images typically encompass a wealth of geographic information and geospatial relationships. GCNs serve as robust tools explicitly crafted to handle such data with intricate correlations¹⁶. GCNs are instrumental in feature classification, enabling the automatic categorization of diverse feature classes (e.g., buildings, vegetation, water bodies, etc.) in remote sensing images. By capturing the geospatial relationships between pixels, GCNs enhance classification accuracy, particularly in areas characterized by complex feature distribution¹⁷. GCNs also possess the potential for data fusion, enabling the integration of remote sensing images with Geographic Information System (GIS) data. This integration enhances the understanding of geospatial data for purposes such as risk assessment, emergency evacuation capacity evaluation, and natural resource management¹⁸.

In Eq. (7) is the pixel value of denoised image at coordinate . is the pixel value of original image at coordinate . The summation symbol on the right side of the formula indicates the summation of n neighbourhood pixel values in the image. indicates the averaging of neighbourhood pixel values. The whole meaning of the formula is that the denoised pixel value at coordinate is the average of the pixel values of the original image in the neighbourhood of that coordinate. This averaging operation helps to reduce the noise in the image and improves the quality of the image.

7

In Eq. (8), denotes the pixel value of the radiometric corrected image at coordinates , and is the pixel value of the original image at coordinates . is the value of dark current or dark frame at the same coordinates . It is caused by thermal noise in the camera’s own electronics. is the value of the flat field image at the same coordinates . Flat-field images are used to correct for non-uniformities in the optical system, such as variations in light intensity. The whole equation means that radiometric correction is performed by subtracting the dark level from the original image and dividing by the flat-field image. This process helps to remove noise caused by dark current from the camera sensor and corrects for non-uniformities in the optical system, thus improving the radiometric accuracy of the image.

8

In Eq. (9), is the pixel value of the original image at coordinates .M is a mapping function that accepts as input and generates a new pixel value. C is the pixel value of the image at coordinates obtained by applying the mapping M to . The specific application of Eq. (9) is to achieve feature classification of remotely sensed images by adjusting M.

9

In this study, the shortest path is calculated by Eq. (10), the nearest distance to the plank road is calculated through coordinate mapping, identifying the corresponding point on the plank road. Then, the shortest path from this point to the parking area is computed, with the sum of the distance to the plank road and the shortest distance from the plank road point to the parking area defined as the fitness function fit₁, representing the total path length. Additionally, to enhance path smoothness, a curvature fitness function fit₂ is introduced to minimize the number of turns. To further account for crowd congestion and other impedance factors within the scenic area, a crowding fitness function fit₃ is defined^19,20.

10

where, the initial values of a, b, and c are all 0.5, and they will be adjusted as the model is trained.

In Eq. (11) once the optimal path is determined, the evacuation time is calculated based on the pedestrian average speed.

11

GCNs model processing is shown in Fig. 4.

Fig. 4.

GCNs processing flow.

TOPSIS method

TOPSIS, initially proposed by Yoon et al. in 1995, stands as a classical multi-objective decision analysis method²¹. In this study, the TOPSIS method is incorporated into the GCNs model for evaluating emergency evacuation capability. The positive and negative ideal solutions are determined by normalizing the decision matrix using Eq. (12), Eq. (13), and (14). Subsequently, the similarity measures between each scenario in the decision matrix and the positive and negative ideal solutions are computed using Eq. (15) and (16). Finally, a composite score is calculated using Eq. (17), where a higher value of ​ signifies a superior emergency evacuation capability. Conversely, a lower value indicates a lesser emergency evacuation capability²².

12

13

14

15

16

17

Analysis of indicators and calculation of weights

Analysis of indicators

Effectiveness

Effectiveness refers to the effectiveness of evacuation procedures in achieving the desired goals in an emergency evacuation, such as reducing casualties and shortening evacuation times. Effective evacuation is usually dependent on a number of factors, including the adequacy of the infrastructure, the efficiency of the chain of command, and people’s familiarity with the evacuation plan. In disaster management, the effectiveness of evacuation directly affects the ability of people to evacuate quickly and safely to a safe area²³.

(1) Area of emergency shelters:

The area of emergency shelters determines its capacity and the range of services it can provide²⁴. A sufficiently large site can not only accommodate more people, but can also be stocked with adequate supplies and equipment. Spacious shelters can reduce overcrowding, improve the quality of life and hygiene, and thus enhance the overall emergency response capacity, ensuring that adequate shelter is provided for those affected by a disaster when it occurs.

(2) Area of shelters per capita:

The shelter area per capita is a key indicator for ensuring that evacuees have adequate living space²⁵. Adequate personal space helps to minimize conflict among personnel, increases the sense of privacy, and enhances psychological well-being during prolonged evacuation. Adequate space per person also ensures quality of life and basic comforts during the post-disaster recovery phase.

(3) Capacity:

The capacity of a shelter reflects its ability and efficiency to respond to a disaster²⁶. Shelters that hold more people have a greater capacity for rescue and support and are able to shelter more victims. The larger the capacity, the greater the overall supply and emergency response capability of the shelter, which helps to effectively provide the support needed in the event of a disaster.

(4) Effective service coverage:

Effective service coverage measures the ability of a shelter to provide supplies and support services to the surrounding area²⁷. A shelter with a wide range of service capabilities is able to cover a larger area and ensure that outlying victims are assisted quickly, thus improving the overall efficiency of the emergency response.

(5) Electricity, communications infrastructure:

Electricity and communications infrastructure are fundamental to ensuring the continued operation of shelters²⁸. Electricity supports the proper functioning of medical equipment, lighting and heating systems, while communications facilities are used for command and control and external communication. Both are critical to the smooth running of disaster response and rescue operations.

Accessibility

Accessibility refers to the ability of evacuees to reach safe areas faster and more conveniently during evacuation. This indicator takes into account the rationality of the channel layout and transportation facilities. For example, in the optimization of building evacuation, especially in scenarios facing mixed crowds, the rational arrangement of stairways can significantly shorten the evacuation time, thus enhancing the accessibility of the entire evacuation process. Studies have shown that a reasonable evacuation layout can reduce the evacuation time by 12% to 19%²⁹.

(1) Road accessibility:

Road accessibility is one of the key factors in measuring the accessibility of shelter sites³⁰. A good road network ensures that people can reach shelters quickly and efficiently during emergencies. Smooth roads help to speed up the evacuation process, increase the efficiency of rescue operations and reduce the time people spend on the road. In the event of a disaster, road access is directly related to the ability of shelters to provide safe refuge quickly.

(2) Distance to nearest plank road:

The distance to the nearest plank road is critical to the accessibility of shelters³¹. Shorter distances mean that people can get to shelters more quickly in an emergency, thereby reducing potential hazards and increasing safety. Trestles are often used for walking or evacuation routes, so proximity has a direct impact on the speed and effectiveness of evacuation.

(3) Width of nearest plank road:

The width of the nearest plank road is directly related to the efficiency and safety of evacuation³². A wider trestle can accommodate more people, helping to prevent congestion and increase the speed of evacuation. However, if the width of the trestle is insufficient, it may lead to congestion and obstruction, making evacuation more difficult. Therefore, the width of the trestle is a critical factor that influences the actual effectiveness of shelter accessibility.

Safety

Safety refers to whether the evacuation routes and facilities can effectively protect people from the threat of secondary disasters, such as collapse of buildings after collapse and mudslide, during emergency evacuation. During emergency evacuation, the safety of evacuation can be significantly enhanced by choosing appropriate temporary emergency shelters. By planning ahead and setting up multiple shelters, evacuation time can be reduced and crowd congestion avoided. These shelters are usually equipped with basic emergency supplies and safety protection measures to enhance safety and security during evacuation. Studies have shown that properly laid out shelters can significantly increase safety and reduce casualties during an emergency evacuation³³.

(1) Distance from hazard sources:

The distance from hazard sources is one of the key indicators of shelter safety³⁴. Shelters located away from disaster-prone areas are safer. Shelters located away from disaster-prone areas reduce the potential danger posed by disasters and increase the safety of people in shelters.

(2) Fractional vegetation cover:

Fractional vegetation cover plays an important role in the safety of shelters³⁵. Vegetation can help reduce the risk of natural disasters such as mudslides. Vegetation can also provide protection against the impact of disasters such as storms on buildings. Therefore, a higher level of vegetation cover is often considered a positive factor that contributes to the safety of shelters.

(3) Average slope of shelters:

The slope of a shelter is directly related to the rate of flow of water, earthquakes and other natural hazards³⁶. Moderate slopes may help drainage and reduce the impact of disasters such as flooding, while overly steep slopes may increase the risk of landslides. Therefore, when choosing the location of a shelter, the slope of the surrounding terrain needs to be considered to ensure safety.

(4) Density of buildings around shelters:

The density of buildings also has important implications for the safety of shelters³⁷. High building densities may lead to an increased risk of fire spreading, while in disasters such as earthquakes, high-density buildings are more likely to collapse. Therefore, lower building densities are often considered a positive factor in reducing potential disaster risks.

Calculation of weights

AHP hierarchy analysis method to calculate the subjective weights

AHP Hierarchical Analysis is a systematic approach for multi-criteria decision-making, encompassing five main phases^38,39:

(1) Establish the Hierarchy:

Hierarchize the decision problem and construct a hierarchy consisting of decision level, criterion level, indicator level, and sub-indicator level.

(2) Quantify the Hierarchy:

In this paper, with input from five experts in the disaster research team at Northeast Normal University, each element in the indicator layer and below is ranked. Their relative importance is compared using a number from 1 to 9, where 1 represents equal importance, and 9 represents extreme importance, resulting in a score .

(3) Calculation of Weights:

The weights are computed using the mathematical model of the hierarchical analysis method. The calculation process involves determining the feature vector of each level and the weight of each element.

(4) Consistency Test:

The maximum eigenvalue of the matrix is determined, and the corresponding eigenvectors are obtained. The consistency of the judgment matrix is verified according to Eqs. (18) and (19) in Table 2.

18

19

Table 2.

Random consistency index test (RI) table.

n	1	2	3	4	5	6	7	8	9
RI	0	0	0.58	0.90	1.12	1.24	1.32	1.41	1.45

(5) Comprehensive analysis:

The obtained weights are used for the comprehensive analysis to find the optimal solution or decision.

Improvement of entropy weight method to calculate objective weights

(1) In Eq. (20), construct the original index data matrix. Assume there are m samples for evaluation and n evaluation indicators; the formation of the original indicator data matrix is as follows⁴⁰.

20

where denotes the value of the i-th sample and the j-th evaluation index.

(2)Data processing: In Eq. (21), to mitigate the impact of different levels on the evaluation results, the indicators undergo normalization, and the individual standardized data are calculated using the following formula⁴¹.

21

(3) In Eq. (22), calculate the share of the i-th sample in the total value of the indicator for the j-th indicator as follows.

22

(4) In Eq. (23), calculate the entropy value of the j-th indicator as follows:

23

where constants , k > 0.

(5) In Eq. (24), calculate the improved entropy value as follows:

24

(6) In Eq. (25), Calculate the coefficient of variation of the j-th indicator . The entropy method assigns weights to each indicator based on the degree of difference in the sign value of each indicator to derive the corresponding weight . The larger the value, the greater the importance of the indicator, as calculated using the following formula:

25

(7) In Eq. (26), calculate the objective weights as follows:

26

Calculating the combined weights

In Eq. (27), Employing the multiplicative integration method, the subjective weights are integrated with the objective weights to derive the combined weights. The results of the calculations are shown in the following Table 3.

27

Table 3.

Weight of indicators.

Criterion Layer	Criterion Layer Weights	Indicator Layer	Indicator Layer subjective weights	Indicator Layer objective weights	Indicator Layer combined weights	Contribution
Effectiveness	0.2577	Area of emergency shelters	0.1849	0.1808	0.1832	positive
		Area of shelters per capita	0.2219	0.2532	0.2374	positive
		Capacity	0.3107	0.3291	0.3203	positive
		Effective service coverage	0.1541	0.1292	0.1413	positive
		Electricity, communications infrastructure	0.1284	0.1076	0.1177	positive
Accessibility	0.3093	Road accessibility	0.3860	0.4660	0.4256	positive
		Distance to nearest plank road	0.3216	0.2913	0.3071	negative
		Width to nearest plank road	0.2924	0.2427	0.2673	positive
Safety	0.4330	Distance from the hazard sources	0.4257	0.3219	0.3725	positive
		Fractional vegetation cover	0.2365	0.2683	0.2534	positive
		Average slope of shelters	0.1971	0.2235	0.2112	negative
		Density of buildings around shelters	0.1408	0.1863	0.1629	negative

The results of the calculations are shown in the following Table 3.

Results and analysis

Real-ESRGAN processing results

To compare the changes in the images before and after processing by the model, a region in Fig. 5a has been selected. Figure 5b depicts the original unprocessed image, while Fig. 5c displays the result after processing by Real-ESRGAN. The processed Fig. 5c exhibits significantly improved resolution, in contrast to Fig. 5b which suffers from blurring in details. This difference in resolution makes the features on the right side of the image more prominent, aiding in clearer feature classification, and the texture details in the image become clearer and finer. These visual improvements not only assist in analyzing, interpreting, and understanding image content more accurately but also provide more reliable and detailed data to support the research and decision-making process.

Fig. 5.

Study area (a), before Real-ESRGAN processing (b), after Real-ESRGAN processing (c).

This study employs the Real-Enhanced Super-Resolution Generative Adversarial Network (Real-ESRGAN) model to process remote sensing images, significantly enhancing image clarity. Following Real-ESRGAN processing, the resulting images demonstrate notable improvements across multiple quantitative metrics. Specifically, the peak signal-to-noise ratio (PSNR) of the processed images increased from 30.5 dB to 38.7 dB, indicating a substantial reduction in noise interference. Additionally, the structural similarity index (SSIM) improved from 0.75 to 0.92, reflecting a higher level of similarity in structure and details between the reconstructed and original images. The mean squared error (MSE) also decreased significantly, from 45.3 in the original image to 10.8, further indicating enhanced image reconstruction quality. These quantitative results clearly show that the Real-ESRGAN model effectively improves image clarity, establishing a solid foundation for subsequent analysis and applications. The comparison is shown in the Fig. 6.

Fig. 6.

Comparison of PSNR, MSE and SSIM before and after Real-ESRGAN processing.

Evaluation of emergency evacuation capability

Based on long-term monitoring data and information collected in the field, we have determined that the main geological hazards facing the northern scenic area of Changbai Mountain include collapse and debris flow areas. In this paper, we will focus on the assessment of emergency evacuation capacity for these two types of hazards and the study of related enhancement strategies. This aims to ensure that the scenic area can quickly and effectively evacuate tourists when facing potential risks, and minimize possible casualties and property losses.

Evaluation of emergency evacuation capacity for collapse hazards

The collapse hazard is a situation in which land or rock on the ground, hillside, or slope suddenly collapses or crumbles due to a variety of factors. A collapse hazard can range from a small-scale localized event to a large-scale disaster, and it is therefore essential that comprehensive emergency plans, monitoring systems, and evacuation procedures for collapse hazards are in place to minimize the injuries and property damage caused by the hazard.

In Fig. 7, we used the results of existing research to identify areas of high vulnerability to collapse hazards⁴², which are marked in red. The yellow area represents the scenic plank roads, the brown markers indicate the location of the parking lot, and the blue color indicates the river. In view of the possible emergency, tourists were visiting the scenic area and it became crucial to swiftly and orderly evacuate them to the parking lot to ensure their safety. In order to assess the capacity of the scenic area for emergency evacuation, we need to make use of modeling. Simulating the evacuation process will help to determine the best evacuation path and strategy to maximize the evacuation efficiency.

Fig. 7.

Overview of collapse hazards in the study area.

As shown in Fig. 8, we applied the model to fully assess the emergency evacuation capability of the scenic area. We set areas with an evacuation time of 0–5 min as having excellent emergency evacuation capacity, labelled in green; areas within 5–15 min as having medium emergency evacuation capacity, labelled in yellow; and areas within greater than 15 min as having relatively poor emergency evacuation capacity, labelled in red. This visual assessment helps to clearly demonstrate the evacuation efficiency of different zones within the scenic area, allowing scenic area managers to better understand the actions to be taken in an emergency. Particular attention should be paid to the red areas, which may require further improvement of evacuation strategies or additional evacuation routes to ensure that visitors can still be safely evacuated in an emergency.

Fig. 8.

Evaluation of emergency evacuation capacity for collapse hazards in the study area.

Evaluation of emergency evacuation capacity for debris flow hazards

As a natural hazard, debris flow hazards are characterized by the rapid flow of large quantities of sediment, stones, rocks, and water mixed together along mountain slopes or river valleys, usually accompanied by a high degree of destructiveness. Therefore, the establishment of effective early warning systems, monitoring measures, and emergency evacuation plans, along with the improvement of land management and the reduction of environmental damage, are key measures for effectively mitigating debris flow hazards.

In Fig. 9, we identified areas at high risk of debris flow based on the results of existing research, highlighting geographic areas prone to debris flows⁴². Figure 8 illustrates seven debris flow gullies, with gullies 1, 2, and 4 having the most significant degree of debris flow risk. Annually, these three debris flow gullies cause substantial economic losses to the scenic area. Additionally, Gully 8 faces a serious situation, with a considerable economic impact on the scenic area. In contrast, the other debris flow gullies have a relatively average risk and a smaller potential impact on the scenic area. This illustration aids in comprehensively understanding the distribution of debris flow risks and provides clear guidance for developing corresponding risk management and preventive measures.

Fig. 9.

Overview of debris flow hazards in the study area.

Compared to the collapse hazards, debris flow hazards have a broader coverage and more serious potential losses. In Fig. 10, we observe that only a small portion of the area has excellent emergency evacuation capacity, enabling evacuation within 5 min; however, most of the area exhibits relatively poor emergency evacuation capacity, with evacuation times exceeding 15 min. This indicates that debris flow hazards are potentially more dangerous, necessitating heightened attention to associated risks. In comparison with collapse hazards, debris flow hazards may have more severe impacts on scenic area and their surrounding areas. Risk management and emergency response to debris flow hazards should prioritize strengthening preparedness and improving evacuation capacity to ensure the efficient and rapid safety of tourists in the event of a disaster.

Fig. 10.

Evaluation of emergency evacuation capacity for debris flow hazards in the study area.

Validation of GCNs model emergency evacuation capability evaluation results

This study validated the accuracy of the Graph Convolutional Networks (GCNs) model, revealing its good performance in evaluating emergency evacuation capacity in scenic areas. Figure 11 illustrates the trend of loss values and accuracy changes for the GCNs model on the training set. As the training process progressed, the model’s loss values consistently decreased, indicating an enhanced ability of the model to fit the data; simultaneously, the gradual increase in accuracy reflects a significant improvement in the model’s classification ability for the training set. After 200 iterations of training, the accuracy on the test set exceeded 95%, demonstrating that the GCNs model effectively identifies and classifies the emergency evacuation capacity of different regions. In the test set, the loss value dropped to approximately 0.6, with an accuracy of about 80%, indicating that the model maintained a high level of accuracy when faced with unknown data, thus demonstrating strong generalization capabilities.

Fig. 11.

Comparison of PSNR, MSE and SSIM before and after Real-ESRGAN processing.

Emergency evacuation capability enhancement strategies

Improved evacuation routes for collapse hazards

In order to align with the actual situation, our fieldwork found that tourists primarily move in the area of the scenic area plank roads, with relatively few venturing into other places. As depicted in Fig. 12, the model’s emergency evacuation evaluation of the plank roads indicates that nearly 30% of the plank roads are in the red warning section, suggesting that the evacuation of tourists in this area is slow and there is a risk of blockage, potentially leading to stampede accidents. A detailed examination revealed that the connection between the red section and the yellow section is a steep and narrow climbing section, posing serious problems for emergency evacuation. In an emergency, relying solely on this section of the plank roads for evacuation would result in unsatisfactory progress. Therefore, there is an urgent need to improve the emergency evacuation program by selecting a flatter, more spacious, and safer pathway to enhance evacuation efficiency, reduce risks, and ensure the reliable protection of the lives of visitors and staff during emergencies.

Fig. 12.

Overview of the study area plank roads.

Given the actual situation of the northern scenic area of Changbai Mountain, we proposed additional emergency evacuation routes to ensure the safe and orderly evacuation of visitors at the observation deck. The new emergency evacuation green channel in Fig. 13 represents a crucial improvement to enhance the emergency evacuation capacity of the scenic area. According to our fieldwork, this channel has a smoother route and a wider width, enabling more effective evacuation of visitors from the observation decks in the red section of the plank roads. Additionally, it has a larger capacity to accommodate more people, facilitating quicker evacuation in case of an emergency. Although the green corridor is located within the debris flow hazards area, it is considered relatively safe in the absence of severe weather conditions such as heavy rainfall. Implementing this improvement program will significantly enhance the efficiency of emergency evacuation at scenic area.

Fig. 13.

Overview of study area new plank roads.

After assessing the emergency evacuation capacity of the scenic area following the route improvement, it is evident from Fig. 14 that there is a significant increase in the number of areas in the scenic area with excellent and moderate emergency evacuation capacity, meaning that the majority of tourists can complete their emergency evacuation in as little as 15 min. This improvement provides a more reliable safety guarantee for visitors and staff, ensuring a quick and orderly evacuation of the scenic area in case of an emergency. The improved green channel is not only a key initiative in the safety management of the scenic area but also plays an active role in dealing with unpredictable emergencies. Through this comprehensive improvement and assessment, the Scenic Area is better equipped to respond to various emergencies and provide visitors with a safer and more comfortable visiting experience.

Fig. 14.

Evaluation of emergency evacuation capacity of improved routes for collapse hazards in the study area.

Selection of emergency shelters for debris flow hazards

The unpredictability of debris flow hazards makes it challenging to anticipate them in advance, including their scale and speed. Scenic area face significant challenges in emergency evacuation, with the primary focus being on ensuring the safety of visitors. For smaller-scale and slower-flowing debris flow hazards, it is crucial to swiftly evacuate visitors to the parking lot for an organized evacuation. However, in the case of larger-scale and faster-flowing debris flow hazards, increasing the evacuation speed alone is insufficient. A comprehensive emergency evacuation plan must be formulated to address such sudden disasters. A robust emergency evacuation plan should encompass detailed procedures and processes, ensuring that staff can guide visitors out of the risk area in an orderly manner and direct them to a safe location.

The scenic area encounters two primary challenges when addressing debris flow hazards. Initially, addressing the evacuation of tourists on the red section of the plank roads is essential. The substantial number of tourists on the red section of the plank roads, coupled with the relative steepness and narrowness of certain evacuation plank roads, poses a potential risk of stampede during mass evacuation. Secondly, there exists the potential for larger-scale and faster-flowing debris flow hazards in scenic area, specifically in No. 1, No. 2, and No. 4 debris trenches, which could result in the washout of certain evacuation plank roads, thereby contributing to the complexity of emergency evacuation. To address these challenges, there is a pressing need to optimize the emergency evacuation routes and judiciously select emergency shelters within the scenic area. When choosing emergency shelters, locations must be selected in a scientific and rational manner to ensure adequate space and basic safeguards for accommodating visitors, offering support, and conducting rescue operations after the danger has subsided.

Figure 15 illustrates the landscape’s vegetation cover, with denser green indicating higher vegetation density. This is vital for selecting shelters, as areas with dense vegetation typically offer superior shelter and protection. In Fig. 16, this paper introduces a novel research perspective. Given the small size of the selected study area, traditional indicator maps have struggled to depict the landscape comprehensively. Consequently, we collaboratively developed a 3D model of the scenic area, incorporating indicators like elevation, slope, and slope direction. This model can display the geographic layout of the scenic area comprehensively in a 360° view, offering invaluable assistance in our scientific shelter selection process.

Fig. 15.

Fractional vegetation cover in the study area.

Fig. 16.

3D model of the study area.

The selection of emergency shelters must, first and foremost, adhere to stringent safety standards. Locations with dense vegetation and higher ground can effectively mitigate the threats posed by debris flow hazards and rolling rocks from high mountains, thus maximizing the safety of visitors’ lives in the event of a disaster. Secondly, we will prioritize locations situated at a considerable distance from potentially hazardous areas to ensure the relative safety of the sites. Additionally, considering the imperative for a swift response in emergencies, we will also opt for locations with convenient transport links to facilitate timely rescue operations by rescue personnel. Such comprehensive considerations will contribute to enhancing the safety of the scenic area and ensuring the protection of visitors.

After a detailed analysis of the vegetation cover in the scenic area depicted in Fig. 15 and the model presented in Fig. 16, we have identified areas 1 to 5 within the scenic vicinity as emergency shelters to fulfill the three aforementioned requirements. Figure 17 shows the five pre-selected emergency shelters. However, it is imperative to acknowledge that these five selected shelters may have distinct potential issues. Consequently, we have conducted a thorough evaluation of these five shelters, considering three key indicators: effectiveness, accessibility, and safety. Using this scientific evaluation method, we can fulfill emergency evacuation needs while optimizing resource utilization. This ensures that the selected shelters cater to the specific requirements of the scenic area and can effectively respond to potential emergencies.

Fig. 17.

Study area emergency shelter siting.

(This map was generated using ArcGIS 10.8 ESRI https://www.esri.com/en-us/home).

As shown in Fig. 18, regarding the assessment of effectiveness, Shelter 2 and 3 outperformed the other selected sites. Shelter 2 encompasses 2,419 m², and Shelter 3 covers 2,351 m². Their larger footprints allow for the accommodation of more visitors and provide a wider effective range. Additionally, they have enhanced emergency response capacity by implementing more sophisticated measures related to power and communications. This observation indicates that Shelter 2 and 3 are not only superior in size but also more robust in terms of infrastructure and communications. Consequently, they are more effective in providing safe shelter services in the event of a disaster.

Fig. 18.

Comprehensive evaluation of emergency shelters in the study area (a), effectiveness evaluation (b), accessibility evaluation (c), safety evaluation (d).

Regarding accessibility assessment, Shelter 1 and 2 stand out significantly because of their proximity to the parking lot. The roads where these two shelters are situated are relatively smooth, offering visitors more convenient access and effectively enhancing the speed of evacuation. Their proximity to the parking lot has been designed to facilitate visitors’ easy access in case of an emergency, simplifying the evacuation process in the event of a disaster. This design not only considers practical accessibility in shelter selection but also maximizes the convenience of the parking lot as an access route. Consequently, Shelter 1 and 2 play an active and pivotal role in accelerating the evacuation speed of visitors.

Concerning safety, Shelter 2 and 5 are more dependable as they are situated in relatively secure areas. Their surroundings boast dense vegetation, and the terrain is steeper, decreasing the likelihood of being threatened by natural disasters like mudslides; consequently, they are relatively unaffected. Being situated away from hazardous areas, these locations offer a safer shelter option for tourists. The dense vegetation and steep terrain collectively constitute a natural protective barrier, effectively mitigating the potential threat of natural disasters to the shelter sites. Consequently, Shelter Sites 2 and 5 are geographically more favourable concerning visitor safety.

Considering all the assessments collectively, it is concluded that Shelter 2 is an optimal choice for emergency sheltering as it excels in all aspects. Nevertheless, opting for Shelter 2, followed by Shelter 1 and 3, would lead to overlapping service coverage, resource wastage, and heightened uncertainty. Consequently, the selection of one among Shelter 4 and 5 is necessary. Based on the field study, Shelter 5 is situated at the highest point of the scenic area, which is notably superior to Shelter 4 concerning safety. Despite Shelter 4 being comparable to Shelter 5 in other aspects, opting for Shelter 5 serves to maximize the scope of evacuation services and offers better protection to visitors due to its safety advantages.

This scientifically sound shelter selection program thoroughly considers the inter-relationships among various shelters, skillfully avoiding resource duplication and potential uncertainty escalation. Decision-making relies on a comprehensive assessment to guarantee that, in the face of sudden disasters, the scenic area can offer emergency shelter services to the utmost extent, effectively ensuring the safety of visitors and staff. The development of this program not only offers practical guidance for current emergency management but also establishes a robust foundation for future disaster response efforts.

Predicting the accurate scale and flow rate of debris flow hazards is typically challenging. In the case of smaller scale and slower flow hazards of debris flow, scenic area should proactively organize swift evacuation of tourists to parking lots. Regarding larger-scale and faster-flowing debris flow hazards, scenic area can implement targeted measures for certain tourists based on monitoring conditions. This involves evacuating them using the newly added emergency evacuation plank roads, while directing other tourists to proceed to the emergency shelter, awaiting rescue actions. Through refining the scenic area’s emergency evacuation plan for debris flow hazards, we observe from Fig. 19 that the area with excellent emergency evacuation capacity is significantly expanded, while the range of medium emergency evacuation capacity is also more extensive. Concurrently, the area with poor emergency evacuation capability is markedly reduced. This series of enhancements, encompassing the improvement of the scenic area evacuation plank roads and the addition of new emergency shelters, substantially bolstered the scenic area’s ability to cope with debris flow hazards.

Fig. 19.

Evaluation of improved emergency evacuation capacity of debris flow hazards shelters in the study area.

Discussions

1. In terms of plank roads, this study proposes the addition of a new boardwalk approximately 300 m long and 3 m wide, while retaining the existing infrastructure, as illustrated in Fig. 13. Under optimal conditions, this new plank road can pass up to 350 people per minute, playing a crucial role in reducing evacuation time for visitors and significantly decreasing the risk of secondary incidents such as crowding and stampedes. With increasing visitor numbers, the existing evacuation routes may become insufficient during peak periods, resulting in evacuation delays and heightened potential hazards. To further enhance the emergency evacuation capacity of the scenic area, it is recommended to optimize the visitor pathway. Specific measures include adding clear route signage to help visitors quickly locate the nearest evacuation routes, as well as implementing a closed-loop route design to enable one-way movement, thereby effectively minimizing congestion caused by cross flows. This design not only significantly improves evacuation efficiency but also enhances the visitor experience by reducing discomfort associated with crowding.

2. This study recommends adding two new emergency shelters, as depicted in Fig. 18. Shelter 2 spans approximately 2,419 m², and Shelter 5 covers about 700 m², with a combined capacity to accommodate 1,560 individuals, providing essential emergency refuge within the scenic area in the event of sudden disasters. These shelters are particularly valuable in scenarios such as mudslides, where they can rapidly host large numbers of tourists and facilitate orderly emergency evacuation. Additionally, they offer safe temporary shelter for those unable to evacuate immediately during potential emergencies. The locations of the new shelters have been strategically chosen, situated on elevated ground with dense vegetation to mitigate risks from natural hazards like mudslides. This carefully selected positioning not only enhances visitor safety but also improves the shelters’ practical protection and reliability in disaster situations. The construction of these new shelters not only strengthens the security experience for tourists but also fortifies the scenic area’s emergency management framework, ensuring that rapid and orderly rescue and evacuation are achievable when natural disasters occur.

3. Regarding visitor capacity, this study estimates that the scenic area could initially accommodate approximately 7,300 visitors based on an infrastructure area of around 14,762 m² and an average required space of 2 m² per person. Considering the unique characteristics of the site, it is recommended that visitor numbers be limited to 5,000 to ensure safety and enhance comfort. This recommendation is not a legal restriction but is made with an emphasis on safety and comfort. Following infrastructure improvements, the total area has increased to approximately 17,761 m². These upgrades may include additional rest areas, optimized pathway designs, and other amenities. Based on these improvements, the scenic area can now reasonably accommodate around 8,880 visitors, effectively enhancing its capacity. A closed-loop route design enables one-way visitor flow, facilitating more effective crowd management and reducing congestion. To maintain optimal conditions, it is recommended that visitor numbers be kept below 7,000, prioritizing both safety and comfort. This flow management strategy not only improves the visitor experience but also strengthens provisions for safe evacuation.

Conclusions

This study focuses on evaluating and enhancing the emergency evacuation capacity of the northern scenic area of Changbai Mountain. The primary innovations of the study are outlined below:

1. Given the limitations in image resolution commonly encountered in small-area studies, this study employs the Real-Enhanced Super-Resolution Generative Adversarial Network (Real-ESRGAN) technique to significantly enhance image clarity. This approach overcomes the resolution constraints typical in small-area analyses, providing more precise data for the scenic area compared to traditional evaluation methods. The innovative application of Real-ESRGAN in this study effectively addresses the challenge of low-resolution data acquisition in the area, laying the groundwork for more accurate data collection methods in future research within the Changbai Mountain region.

2. This study integrates the Real-ESRGAN model with the GCNs model, allowing Real-ESRGAN to enhance image data resolution and facilitate seamless data exchange between the two models. The enhanced image data output can be directly recognized by the GCNs model, significantly accelerating the matrix construction speed of GCNs and improving the efficiency of emergency evacuation time calculations. This approach makes a valuable contribution to the assessment of emergency evacuation capacity in the scenic area.

3. This study proposes a series of improvement measures tailored to the characteristics of the scenic area, aiming to comprehensively enhance its emergency evacuation capacity, reduce safety risks, and ensure visitor safety. By analyzing the specific conditions of the scenic area’s plank roads, the study identifies shortcomings and proposes new evacuation routes. For the newly emergency shelters, an emergency evacuation capacity assessment system was developed based on effectiveness, accessibility, and safety, providing a scientific and rational evaluation of each shelter’s actual functionality. The two shelters with the largest service areas and most comprehensive features were selected. Additionally, using elevation, slope, and aspect data, a 3D model of the scenic area was constructed in Plotly, enabling a 360° panoramic view of the area’s geographic profile and offering a more in-depth perspective for study. These innovative methods not only consider the unique conditions of the scenic area but also integrate advanced scientific and technological approaches, providing practical improvement solutions for scenic area management.

The approach and methodology of this study provide new perspectives and techniques for study in regions similar to the Changbai Mountain North Scenic Area. In terms of data processing, the use of the Real-ESRGAN model significantly improved data accuracy, enabling more precise and detailed analysis. The GCNs model was then employed to conduct an in-depth evaluation of the scenic area, refining the assessment of its emergency evacuation capacity. Additionally, this study developed a scientific set of evaluation indicators for shelters, providing a useful reference for similar study. However, there are still limitations in this study. The scenic area experiences significant seasonal variations, particularly with heavy rainfall in the summer and snowmelt in the spring, which were not fully considered in this research. Future studies will place greater emphasis on the impact of these seasonal factors to enhance the comprehensiveness and applicability of the evaluation.

Author contributions

Conceptualization, E.Z.; Data curation, E.Z. G.L. and J.Z.(Jiale Zhu); Formal analysis, E.Z J.Z.(Jiale Zhu) and J.Y.; Methodology, E.Z. and J.Z.(Jiquan Zhang); Writing—Original draft, E.Z.; Funding acquisition, Y.Z.; Writing—Review and Editing, Y.Z. and J.Z.(Jiquan Zhang). All authors reviewed the manuscript.All authors have read and agreed to the published version of the manuscript.

Funding

Research on Chain Effect of Geological Hazards and Fine Risk Early Warning Technology in the Northern Scenic Area of Changbai Mountain (12556SF010483432).

Data availability

The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.