Exploring the Spatial Characteristics of Stay Areas in Walking Tours through the Lens of Volunteered GPS Trajectories: A Case Study of the Zhuhai–Macao Metropolitan Area

Abstract

The stay areas in walking tours are the service and management unit of recreational walking in metropolitan areas. The rational characterization of stay areas in walking tours is of great significance for developing local tourism, constructing appropriate public facilities, optimizing the configuration of tourist elements, and improving facility efficiency. The existing research focuses mainly on functional, top-down classifications of tourism, tourist behavior patterns, and route designs, but it has left tourists’ stay areas largely unaddressed. To fill this gap, we propose a new framework for the interpretation of stay areas in walking tours based on GPS trajectory data and accompanying photos uploaded by users. Taking the Zhuhai–Macao metropolitan area as an example, we first captured the stay points and clustered them to the walking tour stay areas using DBSCAN. The characteristics of the stay areas were then collected, and a hierarchical analysis was conducted in terms of spatial features and geotagged photos. The results show that the stay areas can be grouped into six categories displaying obvious differences in spatial distribution, landscape features, and tourist activities. We also found the connections between Zhuhai City and the Macao Special Administrative Region (SAR) to be relatively weak. In conclusion, our results can contribute to tourism planning as well as the further management and allocation of recreational service facilities in the area researched.

1. Introduction

The classification of tourist activities in terms of spatial functionality is of great significance in both theory and practice [1,2]. Essentially, it reflects the diversity of and differences in tourism resources [3], which is helpful in realizing the potential of tourist locations and enhancing hiking areas in terms of local conditions. Classifying specific spaces based on tourist behaviors is an effective method of building a spatial system for tourism services and management [4], which is also conducive to shaping diverse and enjoyable experiences for tourists. This can not only aid in the differentiated development of urban tourism areas, but also facilitate tourists’ choices of travel routes according to their preferences. Therefore, dividing walking tour areas and carrying out spatial cluster analyses is conducive to organizing transportation systems and supplying public facilities based on the needs of tourists [5].

The spatial characteristics of tourism behavior are an important topic in tourism research [6], especially in the context of outdoor leisure activities. Tracking and understanding tourists’ activity patterns is crucial for planning and management [7,8]. At present, tourism planning generally starts from the supply side, with the distribution of tourist resources forming an important basis [9] for classifying tourist areas by means of analyzing the types of tourist resources, tourist industry clusters, or ecological tourism carrying capacity [10]. These methods are less focused on the behavioral choices of tourists and are not suitable for analyzing walking tours in metropolitan areas bottom-up [11]. The primary reason for this is that top-down classification of tourism often lacks a sufficient understanding of tourists’ spatial preferences. By contrast, research based on tourists’ GPS trajectories and geotagged photos may better reflect tourists’ actual views.

Due to the broad geographical scope of metropolitan areas, tourists are scattered on many routes, so it is often difficult to track tourists’ spatial behavior using traditional methods [12]. However, the emergence of VGI apps and platforms enables users to share and obtain GPS data in large quantities, and they are therefore becoming an important data source for tourism research [13,14] as they provide a reliable means of investigating tourists’ activity patterns [15]. The crowdsourced GPS data shared by trekking tourists show obvious advantages in both quantity and quality [16,17]. With GPS trajectory data, a user’s travel mode, movement characteristics, and even stay behaviors can be effectively captured [13,18,19]. Meanwhile, the connections between attractions in variously classified areas formed by GPS trajectories and the geotagged photos attached to those trajectories reflect users’ preferences. Therefore, starting from the demand side, using crowdsourced geoinformation to achieve the bottom-up classification of tourism areas and obtain a comprehensive understanding of tourists’ spatial behavior can effectively support the planning and management of outdoor tourism [20]. The study based on GPS data presents some limitations: (1) The research samples have certain constraints. In most cases, the VGI platform is mainly aimed at specific groups (mountain climbers and other outdoor enthusiasts). It can reflect the walking tour information volunteered by some travel enthusiasts, but it cannot represent all tourists. (2) In dense forest areas, some GPS trajectories may present issues such as positioning errors or upload interruptions, which is a major limiting factor for GPS data. On the whole, the application of GPS trajectory in practical research is justified.

Spatial cluster analysis is widely considered to be an effective method of examining tourism behavior. Urban-scale human activity tends to follow a heavy-tailed distribution rather than a normal or uniform one, which means that there are far more areas with higher densities of activity than with lower densities [21,22]. As a type of urban activity, urban hiking and recreational activity also follow this distribution feature to a certain extent. Its spatial clustering characteristics therefore make clustering analysis a clear choice for explaining its features. In terms of tourism classification, many scholars have conducted related research through cluster analysis, including measuring tourism competitiveness to classify tourism types at the provincial or national level, that is, at the macro-scale [23], spatially categorizing the agricultural tourism area at the meso-scale [24] and characterizing the layout of tourism facilities at the micro-scale [25]. With regard to tourist behavior research, scholars have attempted to use GPS data for spatiotemporal analysis to generate tourist route selection models [26] to construct new methods of tourist route design [27] or to propose a generative algorithm based on the trekking network [28]. It is noted that the application of GPS trajectory data in travel surveys is increasing in breadth and depth, basically focusing on tourists’ behavior patterns and route design methods, but lacking attention to tourists’ preferences of stay spaces [29]. The use of GPS trajectories and photos combined with cluster analysis of walking tours remains rare.

To fill the gap, we have established a combined analytical framework based on data from geolocation trajectories and accompanying geotagged images, which provides a novel perspective enhancing in-depth understanding and allowing for interpretation of the touristic walking preferences of visitors. This is the first contribution of this paper. As for the second, we conducted an empirical case study of the Zhuhai–Macao metropolitan area to explore the spatial characteristics of tourists’ stay areas. This examination is useful for investigating the rationality of the proposed framework; additionally, the interpreted results are also helpful for assessing the implementations of tourism-related planning and management from the bottom-up perspective.

The remainder of this paper is structured as follows: Section 2 is the methodological portion, providing a detailed introduction to the proposed new analytical framework. Section 3 presents the results and offers an empirical investigation of the proposed framework in the research area, while Section 4 consists mainly of our further discussion and conclusions.

2. Materials and Methods

2.1. Study Area

The study sites are located in Zhuhai City and the Macao SAR (Figure 1). The area has a large number of natural landscape resources; it offers a good ecological environment and rich resources for urban tourism, rendering it attractive to many people seeking hiking tours. The Zhuhai–Macao metropolitan area has a total area of 1769.35 square kilometers whose landscape is mainly composed of mountains and subsidence plains. It is the area with the largest ocean area, the most islands, and the longest coastline in the Pearl River Delta, forming a diverse landscape. The tourism industry in Zhuhai and Macao has reached a certain magnitude, and the number of tourists increased from 69.96 million in 2016 to 85.52 million in 2019. Due to the pandemic, the number of tourists has declined in the past two years, but still, 28.42 million tourists visited the area in 2021 (Zhuhai Culture Tourism and Sports Bureau and the Macao Government Tourism Office), demonstrating that the Zhuhai–Macao metropolitan area is a reliable and representative subject for tourism research.

2.2. Data Sources

The GPS data used in our study come from an outdoor recreational website called ‘2bulu’. This platform contains large numbers of shared trajectories and geotagged photos uploaded by individual users. As of October 2022, from the literature of WOS, SOPUS, and CNKI, 2bulu data has been used in various disciplines such as building science, education, tourism, computer applications, biology, etc. Furthermore, this study confirms that 2bulu’s GPS data cover all travel routes of Qunar (a representative tourism website). In addition, the adopted 2bulu trajectories are tested by tourism planning issued by two governments. Therefore, these data can reliably reflect the situation of walking tours in Zhuhai and the Macao SAR. According to Dangermond and Goodchild (2020), these GPS trajectories can be categorized as typical volunteered geographic information (VGI) data [30], which are gradually becoming regarded as a critical source for future urban analysis. It should also be mentioned that the issue of privacy in the use of trajectory data is avoided here owing to VGI’s openness and the individual nature of its contributions.

2.3. Research Model

To achieve the research purpose, this paper proposes a novel examining framework to analyze tourists’ stay behavior and conducts an empirical test with the Zhuhai–Macao metropolitan area as the research object. Specifically, the entire framework can be divided into four parts (see Figure 2), namely: (1) extracting stay points based on VGI trajectories; (2) identifying high-density stay areas and determining their boundaries; (3) analyzing the spatial characteristics of the stay areas and their landscape features; and (4) classifying stay areas into various types based on their characteristics. These four steps are described in detail below. The subsequent section provides an introduction to each of them.

2.3.1. Identify Stay Points

The first step was to extract stay points from the trajectories. Theoretically, stay points have dual connotations in time and space [31]. Tour stay points can be regarded as individual imprints of activities over a certain period of time within a predefined space. The trajectory consists of a series of coordinate points containing timestamps (see Formula (1) below). This paper defines that if there is a point p_i that satisfies Formula (2), it is considered that the trajectory has a stop at p_i, p_i+1…p_k (k > 1). On this basis, stop behavior can be defined by the point p_s, that is, p_s = (x_s, y_s, t_s), where x_s, y_s, and t_s are the longitude, latitude, and timestamp of the stop point p_s, and the specific value is the average value (Formula (3)) of all trajectory points from p_i to p_k. Combined with the situation of recreational activities that encourage staying, Dr and Tr are set to 250 m and 10 min, respectively, which ensures accuracy for subsequent spatial identification and analysis. Accordingly, this paper obtained 21,253 stop points from 11,418 trajectories in the Zhuhai–Macao metropolitan area as the basic data for the identification of recreational hotspots.

$$\begin{gathered} T=\left\{p_1,p_2, p_3\cdots p_n\right\} \\ {P}_i\left(x_i,y_i,t_i\right) 0<i \le n \end{gathered}$$

(1)

$$\begin{gathered} Dis\left(p_i,p_k\right) \le D_r, Dis\left(p_i, p_k+1\right)>D_r \\ Time\left(p_i,p_k\right) \ge T_r \\ Count\left(p_i,p_k\right)=k-i+1 \ge 4 \end{gathered}$$

(2)

$$\begin{gathered} x_s=\frac{{{\displaystyle \sum }}_i^k{x}_i}{k-i+1} \\ {y}_s=\frac{{{\displaystyle \sum }}_i^k{y}_i}{k-i+1} \\ {t}_s=\frac{{{\displaystyle \sum }}_i^k{t}_i}{k-i+1} \end{gathered}$$

(3)

Theorem 1.

x_i, y_i, and t_i are the longitude, latitude and timestamp of the stop point p_i, respectively; Dis(p_i,p_k) and Time(p_i,p_k) are the spatial and temporal distances of the points p_i and p_k, respectively; D_r represents spatial distance, and T_r describes the temporal distance thresholds. Count(p_i,p_k) is the total number of trajectory points between points p_i and p_k.

2.3.2. Identify Stay Areas

The second step was to identify the stay areas. Stay points were derived from a user’s location and stay time according to the trajectory data, which can provide a strong indicator of an individual’s interest in a certain area [32,33]. Theoretically, due to differences in individual travel preferences [34,35,36], the stays formed by each trajectory are bound to be spatially different from other trajectories. However, when an area attracts a considerable number of tourists and long stays, it becomes a popular leisure hotspot for tourists, that is, a high-density recreational stay area. These few areas are the sites of the most tourist activities, and they are also the focal areas for urban planning, construction, and management, which can contribute greatly to the subjective well-being or satisfaction of tourists [37]. Similarly, for the research area of this paper, the high-density stay areas are of major significance to the study of urban forest tourism and the identification of tourist preferences. The specific analysis and results can be fully used to guide the planning and construction of relevant ecotourism routes.

At this point, we utilized the DBSCAN algorithm based on density clustering to perform cluster analysis on the obtained stay points. This algorithm was proposed by Ester et al. in 1996; it can filter clusters in a set of data by a predefined value (Distance Threshold) and a minimum number of cluster points (Minpts). In recent years, it has been applied to geospatial issues [38], especially in research that identifies urban hotspots or areas of interest [39,40]. This paper therefore used DBSCAN to accomplish the identification. Specifically, taking into account the actual spatial scale of the study area, we set the Distance Threshold and the Minpts to 400 m and 30, respectively, and successfully identified 61 high-density areas. In addition, in order to clarify the boundaries of the stay areas, we applied the envelope surface algorithm used by Cai et al. (2018) to convert point elements into surface elements, which is convenient for subsequent calculations and visual representation.

2.3.3. Identify Characteristics of Stay Areas

The third step was to analyze the characteristics of the defined high-density stay areas; it was carried out mainly with respect to the two dimensions of spatial characteristics and landscape elements. Spatial characteristics here refer to objective, physical spatial features of the stay areas, such as the size of the area, the degree of its connection with other areas, etc. Considering the characteristics of the study area, we extracted eight main spatial characteristics, namely: ① density of photos, ② density of trajectories, ③ area of stay areas, ④ altitude changes within trajectories, ⑤ speed of trajectories, and three types of trajectories: ⑥ trajectories within a stay area, ⑦ trajectories through a single stay area, and ⑧ trajectories through multiple stay areas. In particular, features ⑥, ⑦, and ⑧ were used to describe the interrelationships between the stay clusters, and their specific definitions can be seen in Figure 3:

Trajectories within a stay area refer to trajectories that are all contained in one stay area; trajectories through one cluster are those trajectories that pass through only one stay cluster and whose length within the stay area is less than 50% of the area’s total length; and trajectories through multiple clusters are defined as trajectories that pass through multiple stay clusters.

Landscape elements are another major dimension that we consider; we do so primarily by parsing the elements in the photos attached to trajectories. It should be noted that since the photos themselves are based on personal preference, this process is one of subjective screening of the landscape; as such, it differs from the objective spatial characteristics, which are objective reflections of landscape characteristics. As for the detailed procedure, an image recognition program (ADE20K) [41] based on MIT open-source code was applied to the analysis of the photos to determine the elements that most strongly affect perceptions of urban spaces [42]. For each image, segmentation algorithms produced a semantic segmentation mask, predicting the semantic category for each pixel in the image. Considering that the given recognition types in the original program were too detailed, we formed nine main semantic categories by merging and deleting some types (the reason for deleting was that some interior elements were obviously irrelevant to this research topic), including: ① vegetation (plants, flowers, grass, trees, and palms), ② water, ③ buildings, ④ sky, ⑤ roads and sidewalks, ⑥ earth (sand, rock, and fields), ⑦ sea, ⑧ mountains and hills, and ⑨ structures (canopies, railings, awnings, poles, etc.).

2.3.4. Hierarchical Classification

Finally, the classifications based on the features of the stay clusters were completed. This step included the reclassification and re-summarization of the existing results so as to facilitate further related analyses and policy formulation. Specifically, the hierarchical classification method can be used for classification of the stay clusters based on the feature analysis results of 17 elements in two dimensions of all the stay clusters calculated in step 3 (see

Table 1. Characteristic of 17 elements.

	Mean	Std	Max	Min
Density of photos (piece/m2)	3.325	7.071	4.768	0.210
Density of trajectories (piece/m2)	1.840	3.499	25.778	0.132
Area of clusters (m2)	1,693,978	2,910,128	16,623,966	22,499
Proportion of trajectories within one cluster	14.4%	15.0%	53.8%	0%
Proportion of trajectories crossing over multiple stay areas	46.7%	30.5%	100.0%	0%
Proportion of trajectories crossing over one stay area	39.0%	26.7%	97.4%	0%
Altitude changes within trajectories (m)	23.272	26.092	123.898	2.139
Average speed of trajectories (m/s)	52.303	23.119	123.834	10.306
Photos: Proportion of waters	2.4%	2.3%	10.4%	0.02%
Photos: Proportion of land/earth	12.4%	9.1%	40.0%	0.3%
Photos: Proportion of plants	30.2%	15.3%	57.4%	3.2%
Photos: Proportion of mountains	3.3%	3.7%	11.5%	0.003%
Photos: Proportion of sea	1.6%	2.5%	9.7%	0%
Photos: Proportion of urban structures	2.5%	1.7%	7.2%	0.2%
Photos: Proportion of sky	20.4%	8.3%	44.5%	3.2%
Photos: Proportion of roads/sidewalks	7.2%	6.3%	33.3%	0.04%
Photos: Proportion of buildings	15.3%	15.2%	52.9%	0.11%

). There are two points to note: (1) A hierarchical tree measures the difference between elements by ‘distance’, and there are various calculation methods for distance (such as centroid distance, average distance, etc.). The widely recognized Ward’s distance is used here, that is, Ward’s minimum variance method [43]. (2) We implemented standardization for all elements. This was intended to eliminate the differences in the feature dimensions of different elements (for example, area and trajectory have different units, square meters and meters, respectively); it is also more conducive to distance-based hierarchical classification.

3. Results

3.1. Distribution Characteristics of Trajectories and Photos

The Zhuhai–Macao metropolitan area consists of Zhuhai City and the Macao SAR. Zhuhai City covers three administrative regions, including Doumen District, Jinwan District and Xiangzhou District, and faces the Macao SAR across the sea. The Zhuhai–Macao metropolitan area is rich in mountains and water resources and is adjacent to the South China Sea, with many rivers in the territory. The Xijiang River is divided into five tributaries and enters from the northern part of Doumen District. After converging into three main streams, it runs through the entire territory from north to south. At the same time, a large number of mountains and forests are distributed throughout the central part of Doumen District, the southern part of Jinwan District, and the entire territory of Xiangzhou District, providing pleasant natural scenery.

As shown in Figure 4, the trajectories of tourists are mostly concentrated in the eastern part of the Zhuhai–Macao metropolitan area, that is, the Macao SAR and the central and northern parts of Xiangzhou District. The urban built area of Doumen District and the coastal area of Jinwan District also have relatively high trajectory densities. Specifically, the tourists who only stay in Zhuhai account for 86.8%, and 5.8% of tourists only stay in the Macao SAR. The proportion of tourists passing through these two areas is 7.3%. In general, the urban area of Zhuhai has a strong connection with the North Island of Macao and a weak connection with the South Island. Through the identification of stop points and the analysis of kernel density, it can be seen that the distribution of stop points (Figure 5) is relatively concentrated and they are mainly found throughout the territories of Xiangzhou District and the Macao SAR, as well as in the central mountain area of Doumen District. In comparison, the photo points (Figure 6) are more dispersed and cover a wider area. The density of photo points is higher in Xiangzhou District and the Macao SAR than other districts. Meanwhile, in Doumen District, different from the density of stay points, the photo points are mostly distributed around the central mountain area.

3.2. Spatial Distribution of Stay Areas

As shown in Figure 7, the 61 stay areas obtained through DBSCAN clustering analysis are mainly concentrated in Xiangzhou District and the Macao SAR, while others are distributed in the central part of Doumen District and the southwest corner of Jinwan District. According to the results of hierarchical clustering, we distinguished six types of stay areas and named them Category A, Category B, Category C, Category D, Category E, and Category F.

Specifically, Category A includes four stay area units with 6183 stay points and 2561 photo points. Each stay area has a large coverage area, contains numerous stay points, and shows a relatively concentrated distribution, particularly in the central mountains of Doumen District, the northern mountains of Hengqin, and the northern mountains of Xiangzhou District (Phoenix Cave, Baizu Ridge, etc.).

Category B contains 17 stay area units with 2583 stay points and 1641 photo points. Each stay area covers a medium-sized area with few stay points. The stay area distribution is relatively scattered. In the western area, the stay areas are distributed in the southwestern cape (Feisha Col), the small mountains, parks, and scenic spots; while in the eastern area, the stay areas are mostly concentrated on mountains and in parks. Stay areas of Category B also appear on mountains on Qi’ao Island and Macao.

In Category C, there are nine stay area units with 511 stay points and 23 photo points. Its stay areas cover a small area with a dispersed distribution, mainly in the northern part of Xiangzhou District. They include various communities and activity centers (Jinjing Haoyuan Community, the gymnasium, Student Activity Center, etc.). The other two stay areas are distributed in rural areas.

Category D consists of 13 stay area units with a total of 1562 stay points and 4586 photo points. Most of the medium-sized stay areas are found in Xiangzhou District, which contains the popular Changlong International Ocean Resort, the northern urban area of Macao, and the Venetian Casino in the southern part. The remaining smaller stay areas are scattered in the western urban area of Zhuhai City.

Category E has 11 stay area units with 1654 stay points and 2459 photo points. The stay areas, displaying medium coverage, are mainly distributed in various parks in the central part of Xiangzhou District (including Jingshan Park, Mingting Park, etc.), wetland ecological parks, and cultural monuments (Xianren Pavilion, Guanyin Temple, etc.) on Qi’ao Island. Other stay areas appear in farmsteads and parks in Xiangzhou District.

There are seven stay area units in the last group, Category F, which contains 869 stay points and 1389 photo points. These stay areas cover a medium area, and most of them are distributed in the Hengqin and Macao areas, namely, in parks and wetland areas related to water features (including Mangzhou Wetland, Shipaiwan Country Park in Macao, etc.).

3.3. Spatial Characteristics of Stay Areas

Through the hierarchical classification analysis of 17 elements (eight types of trajectory features and nine types of photo semantics in Figure 7), the six types of stay areas present different characteristics (Figure 6), and the typical landscapes of each are shown in Figure 8 below.

Category A shows stay areas covering a large area in terms of trajectory features. The tourists who visit Doumen District are more inclined to stay in the same stay area, and the altitude changes of the tourists’ trajectories are significant. Semantically, the photos taken by tourists in Category A mostly show earth elements, vegetation elements, and mountain elements. This category can be viewed as an extensive, densely wooded mountain trekking area that attracts visitors interested in long-distance trekking.

In Category B, most tourists travel in multiple stay areas, while a small number of tourists pass through only one stay area; the altitudes of the tourists’ trajectories vary greatly. As for the semantic segmentation, earth elements and vegetation elements appear most frequently, with mountains as the third most frequently occurring element. A small number of photos contain sea elements and sky elements. Based on the above features, this category can be called a small-scale mountain hiking area. Most such spaces are located near urban construction areas with a relatively wide field of vision; in some stay areas, the ocean and other bodies of water are visible, attracting tourists for short- and medium-distance hiking activities.

With regard to the trajectory characteristics of Category C, the density of trajectories passing through stay areas is relatively large. Most tourists tend to pass through only one stay area, and a minority tends to move across multiple stay areas. Significant altitude changes appear in a small number of trajectories; moreover, a few tourists walked at widely fluctuating speeds. In terms of image recognition, there are few photo points in this are, whose features are atypical. Category C is a village and community recreation area, and the scenery primarily represents a rural community style with little urban construction and some natural elements.

Figure 8. Example landscapes from categories A–F.

Figure 8. Example landscapes from categories A–F.

In terms of trajectory characteristics, Category D contains two stay areas with many photo points. Visitors stay in the same stay area or pass through only one stay area. The altitude changes within visitors’ trajectories are gentle, and the visitors’ walking speeds vary greatly. Photos in these stay areas are recognizable by means of three elements: specifically, building elements account for the largest proportion, followed by road and sidewalk elements. This category is considered an urban tourism pedestrian area and includes famous tourist attractions and the city center, providing tourists with spaces for play and shopping.

Unlike those in Category D, most tourists in Category E tend to move through multiple stay areas. Similar to Category D, the altitude changes of the visitor trajectories in Category E are gentle, and the speed changes within visitors’ walking trajectories are considerable. Semantically, sky elements appear most frequently, followed by structure elements and road and sidewalk elements. Water elements and building elements are also present in some photos, and land elements appear the least frequently. This category can be regarded as a natural waterscape area which includes various wetland parks and park spaces with views of shallow water.

Finally, in Category F, most tourist trajectories reveal a preference to stay in the same stay area, and the altitudes of some trajectories vary greatly. Semantically, water elements, sea elements, structure elements, and sky elements appear most frequently. This category can be perceived as a fitness park area with wide views comprising many kinds of parks with wide views and pleasant landscapes, attracting tourists for recreation and sports.

4. Discussions and Conclusions

Walking tours are a green and sustainable approach to tourism that rely on scenic local resources and integrate multiple objectives such as leisure, exercise, and social networking. For quite some time, the Internet has been the main information medium and an interactive tool for walking tours, whose market scale is constantly expanding. This indicates great potential for development. The present paper provides a new analytical framework for the study of walking tours using data from volunteered GPS trajectories. High-quality urban walking spaces help to attract tourists and enhance recreation and subjective well-being [44]. The clustering of stay points for hiking tours reflects the spatial distribution of tourist hotspots, which is the result of tourists voting in a bottom-up manner. Compared with top-down spatial divisions based on the distribution of tourism resources, the scale of the tourism space unit obtained by clustering is more closely matched with actual tourist activities and demands, which provides a basis for regional tourism planning, differentiated services, and management in accordance with local conditions as well as the optimization of the allocation and efficiency of facilities and resources. From the perspective of tourism management, the division and identification of characteristics of hiking stay areas can provide the basis for tourism service management and security facility layouts. For the planning and design of hiking spaces, the characteristics of the hot spots in stay areas can provide a reference for planning and design. Different hiking tourism areas have their own geographical environments, natural resource endowments, and spatial characteristics, and they attract different groups of tourists with their particular travel habits, hiking modes, and, consequently, tourism service demands. Once the characteristics of a stay area have been accurately identified, hiking service and management resources can be more accurately matched to spatiotemporal changes in tourism service demands, thus allowing for the achievement of more efficient allocation.

Based on the identification of stop points, semantic image segmentation, and cluster analysis, we have constructed a research model focusing on crowd preferences and analyzed the preferred stay areas of tourists in the Zhuhai–Macao metropolitan area. Our research shows two findings. First, tourists visiting the Zhuhai–Macao metropolitan area tend to stay in urban spaces with the six types of spatial characteristics outlined above: the extensive, densely wooded mountain trekking area, the small-scale mountain hiking area, the village or community recreation area, the urban tourism pedestrian area, the wetland landscape area, and the fitness park area with wide views. It can be seen that in addition to tourists’ enthusiasm for popular urban attractions, pleasant natural features—including urban forests, mountain scenery, waterscapes, visible sky, and wide views—are also hugely attractive to tourists. The construction of urban natural landscapes is therefore conducive to improving the image of a city [45], building a city brand, attracting more tourists [46], and creating economic value [47]. Although urban forests are an important component of urban tourism systems [48] and both have been extensively studied in the literature, they are often examined separately, with limited research on urban forests from a tourism perspective [45]. Based on this, this study explores the impact of urban natural landscapes on tourists’ behavior and tourist hotspots, which is beneficial to the optimization of urban forest tourism management. Secondly, although Zhuhai and the Macao SAR have a prominent advantage of geographical proximity, the walking tour trajectories between them are far fewer than expected. On the contrary, there are far more walking tour trajectories within each area; the border management regimes greatly weaken tourism links between neighboring cities. Currently, Macao’s economy is constrained by a single dominant industry: gambling [49,50]. Various industries related to Zhuhai (including tourism) can bring more development opportunities to the Macao SAR. Therefore, improving the border management system can be beneficial to the Macao SAR’S tourism economy, and this finding will also be incorporated into local governance.

With the development of network information and GPS technology, the content of travel GPS trajectory information and photos shared on the Internet is constantly becoming more abundant and comprehensive. The amount of available data will continue to grow, and the information collected from tourists will thus be increasingly accurate. For practical planning applications, the walking tour characteristics and spatial preferences of tourists obtained by the method established in this paper can provide effective early guidance for tourism-related planning. With the positioning errors or upload interruptions of GPS trajectories [44] being further reduced, this method can be combined with field research to provide relatively judicious delineations for planning and implementation.