Impact and Recovery of Coastal Tourism Amid COVID-19: Tourism Flow Networks in Indonesia

Abstract

This study aims to explore tourism changes in coastal tourism destinations before and during the COVID-19 pandemic from the perspective of regional resilience. A mixed method of a social network and spatial analysis was used to evaluate inbound tourists’ geotagged photos of Indonesia on Flickr from 2018–2022 as metadata. The DBSCAN algorithm and Markov chains were used to comprehensively analyze the hotspot areas and the patterns of tourism movement trajectories amid a complicated recovery. The results demonstrate that: (1) The distribution of geotagged photos before and during the pandemic generally exhibited stage and regional unevenness. The main clusters were Java and the Nusa Tenggara Islands, with the rest displaying a scattered distribution. (2) The tourism flow network was unevenly distributed, and the nodes had obvious core and edge areas. Owing to the crisis, the tourism flow network realized a change in form from network to line and point. (3) Its impact on Indonesian inbound tourism may persist in the short term, and the volatility of national anti-pandemic policies influences the resilience of tourism flow during COVID-19. The dominance of the core nodes highlights the network’s resistance to disruptions due to the prominence of the location of network connections during the pandemic, and marginal nodes reflect the vulnerability to pandemic shocks owing to the hypocentricity of the nodes and the thinness of the connections within and outside the islands. These results provide marketing and promotion policies for the sustainable development of coastal areas.

1. Introduction

COVID-19 has spread worldwide since December 2019. It has been an unforeseen shock, affecting global health, social, and economic systems, and leading to the collapse of countries [1]. This crisis has led to the stagnation of coastal tourism owing to border closures and restrictions on the movement of people [2]. Moreover, the crisis has had a complex impact on tourism flows in the Asia–Pacific international coastal tourism region because of the difference between the “clean-up” model in East Asia and the “co-existence” model in Europe and the US [3]. The recent socioeconomic turmoil in Sri Lanka, an important coastal tourism destination, was also impacted by the outbreak [4]. For consumers, the outbreak and the policy response of local governments is an external stimulus [5]. According to neuromarketing theory, consumers will respond consciously to such stimuli, which influence consumer decisions [6]. In this case, it is critical to explore how the COVID-19 pandemic has affected tourism development in coastal destinations and how to recover from this extreme mobility constraint in the post-pandemic era [7]. Indonesia, the world’s largest archipelagic country, is an important coastal resort region in Southeast Asia, ranked 40th in the 2019 World Tourism Competitiveness Index [8], and welcomes 16.1 million foreign tourists. However, only 3.77 million foreign tourists visited Indonesia in 2020 after the outbreak, a 76.8% decrease year on year [9].

The direct impact of COVID-19 as an infectious disease manifests itself in tourism flows when people are subjected to social distancing regulations, closed borders, and isolation requirements [10]. Tourism flow is the spatial movement between locations generated by tourists during their travels. As the size, scale, and intensity of crowd flow change, tourism flow becomes a crucial indicator of local tourism economic development [11]. Meanwhile, analyses of tourism network structures in tourism flow help clarify the interaction between tourists across regions [12]. Enhancing the resilience of their networks is critical to recovering tourism flows; moreover, it is the most direct factor in judging the tourism development of a destination [13]. Thus, the complexity of tourist movements in tourist destinations during the pandemic suggests the need for a specific study [14].

The initial methods used to study tourism flow included market surveys or statistical data to explore the flow direction of tourism and the spatial distribution pattern of the flow from the perspective of diffusion or transfer. This was then combined with a fuzzy comprehensive evaluation and hierarchical analysis theory, tourist transfer state, and market competition state index [15,16]. Scholars noticed the close correlation between tourism flow and the surrounding environment and explored the coupling degree between tourism flow and network information, destination, urban tourism environment, and regional economy [17,18]. Owing to data limitations, their results could not accurately and comprehensively reflect temporal changes in tourism flow, tourists’ visiting experience, and tourism flow linkage of related regions [19]. Along with the development of the information age and the change in network technology, scholars began to study tourists’ travel photos with geotags posted on the Internet and began to explore tourists’ travel movement trajectories [20,21]. The rapidly growing Internet community has facilitated the emergence of new online media and interactive platforms such as Instagram, Flickr, and Twitter [22]. COVID-19 has increased consumer use of digital media with real-time updates to replace other communication platforms that were limited by the pandemic [23]. The presence of several geotagged photos shared among users on Flickr, containing accurate geolocation information and tourists’ concerns and comments on their travel experiences, is highly applicable to exploring tourism hotspots and has been applied by tourism scholars in spatiotemporal studies of tourism flows [24]. Research based on geotagged photos and other tourist data has focused on inbound tourist behavior, identifying urban hotspots, predicting tourist destinations, and revealing differences between tourist and resident visits [25,26,27,28,29].

Although the topic of tourism flows has attracted the attention of scholars, few have investigated coastal areas. COVID-19 has brought an unprecedented negative impact on coastal tourism [30]. Although the coastal tourism industry has become stronger after successive shocks, including the influenza virus, SARS epidemic, and terrorism, the current pandemic has caused an unparalleled decline in business in the global industry compared to previous crises [31]. Although there is considerable literature covering the study of tourism flows and the spatial distribution of regional destinations [32,33,34], few studies have explored the impact of COVID-19 on tourism destinations and recovery in coastal regions. Therefore, we pose the following question: how does COVID-19 affect the distribution and recovery of tourism flows in coastal tourism destinations? Additionally, what is the driving mechanism behind it? This paper uses Indonesia as a case study, exploring the differences in tourism flows before and during COVID-19. The results can help coastal destinations revitalize their local tourism appeal and develop a robust local tourism industry chain system, thus playing an important role in future tourism branding [35].

This study has three main contributions: First, it compares the differences in the spatial distribution of tourism flows before and during COVID-19 and refines the understanding of the patterns of tourism flow network structures in coastal tourist destinations. Second, this study introduces the influential factors of national prevention policies to explore how they have caused COVID-19-induced differences in tourism flows, as well as the driving mechanisms of recovery. This extends the theoretical knowledge of the regional resilience of destination tourism during the pandemic. Third, an optimization approach is proposed to reconstruct the post-pandemic tourism flow network to identify possible paths to recover international coastal tourism for the destination. The results of this study can provide insights based on how tourism flows may recover. With the COVID-19 pandemic nearing its end for most of the world, it will be important to revive the struggling tourism sector to boost stressed economies. Studies such as this one that look at the effect of the pandemic on tourism can pave the way for tourism recovery. The remainder of the study is organized into the following sections: Section 2 presents the reasons for selecting the study area, data sources, and research methodology. Section 3 outlines the study’s findings, Section 4 contains a discussion of the results, and summarizes the main contributions and limitations of the study.

2. Materials and Methods

2.1. Study Area

Indonesia is the largest archipelagic country in the world, consisting of 17,508 islands. It is also known as the “Land of a Thousand Islands”. It comprises 31 provinces and 4 special administrative regions distributed among 6 major islands: Sumatra, Kalimantan, Sulawesi, New Guinea Sumatra, Java, and Nusa Tenggara [36] (Figure 1). Bali is the most popular tourist destination in Indonesia and the world, and in 2015, it was ranked second in the “Top 10 Best Islands in the World” by Travel & Leisure magazine [37]. On 2 March 2020, the first case of COVID-19 was diagnosed in Indonesia, and the Indonesian government issued an entry restriction policy that temporarily banned foreign nationals from entering the country, disrupting the pace of tourism development in Indonesia. In the early days of COVID-19, the government changed its entry policy ad hoc depending on the spread of the pandemic, resulting in an overall decrease in tourist arrivals. In July 2021, when the first case of the Delta variant spread to Indonesia, the government announced an emergency entry restriction policy, and tourist arrivals dropped sharply once again. By August 2021, Indonesia’s daily increase in confirmed cases peaked with the world’s highest single-day increase [38]. In January 2022, the first case of Omicron infection appeared in Indonesia, followed by a third outbreak of an Omicron mutant strain. Instead of banning foreigners from entering the country, as was the case for the previous two outbreaks, the Indonesian government restricted the entry of foreigners to “Produce a negative test 14 days prior to departure with a second negative test upon arrival in Indonesia” [39,40]. (Figure 2).

2.2. Data Source and Processing

This paper focuses on the acquisition of images and metadata from the photo social platform Flickr. Flickr is a popular photo-sharing website that provides photo uploading and sharing services to a large number of registered users worldwide, and many of these photos include information about the picture’s geographical location (latitude and longitude coordinates). Additionally, Flickr was established in 2004 and has become the primary source of data for tourism photo research due to its large number of images and open access format [41]. We used geotagged photo information posted by non-Indonesian inbound travelers from January 2018–February 2022 on the Flickr platform as a metadata set, including subsidiary information such as photo numbers, photographer, shooting time, longitude, latitude, photo tag, and user information. This dataset only selects photo data where the geographic term belongs to the title. For example, if you search for “Indonesia”, 134,389 photos include Indonesia in the photo title. When the keyword is “Indonesia”, the crawled data are reduced to 4000. When the keyword was set to “country” only, capturing all the data was impossible because of the data density. This study captured data for the first level of administrative district names such as Sumatra (Sumatra province) or Kalimantan (Kalimantan province). Each administrative district and city noun was used as keywords for crawling, totaling 34 first-level administrative districts. After the crawling test, the amount of data for each administrative region was lower than the 4000 limit when the crawling time was set to one month. For example, when the crawl information was “Java” and the crawl time was set to 1 Janurary 2018 to 1 Feburary 2018, there were 3947 photos of place names, which was less than or equal to 4000 pieces of datum, and the data information was both complete and could cover the selected time dimension.

Users cannot open every photo to correct its positioning while uploading photos; moreover, not every photo on the website contains coordinate information. Therefore, the data cleaning was divided into two major parts: with and without coordinates. For photo data cleaning with coordinates, the point element data of photo labels were first exported to obtain vector data layers. The labeled points outside the study area were eliminated by a spatial overlay and superimposed with Indonesian administrative districts. For photo data cleaning without coordinates, the data needed to be cleaned based on other content information within the photos, and the photos needed to have the place name in the data. Other information of this type includes a geographic location description and content elaboration. After the photos with coordinates were labeled, the latitude and longitude were loaded into ArcMap to correspond to longitude by X field and latitude by Y field; geospatializing the photos without coordinates was a prerequisite for entry into the library. The spatialization of photo coordinates was divided into three parts for processing: labeling photo information without coordinates, artificially picking up the coordinates of iconic place names, and using Google coordinates for spatialization (Figure 3).

2.3. Method

2.3.1. DBSCAN Algorithm

DBSCAN is one of the most commonly used algorithms in density clustering. This algorithm’s main idea is to randomly select a core point within the whole dataset and then expand from that point to the surrounding area to connect high-density points within the neighboring range into sections to form various clusters [42]. The two key parameters for algorithm implementation include: (1) $Eps$: for any point $p$ in a given dataset, $Eps$ is the spatial region with $Eps$ as the radius:

$$N_{Eps}\left(P\right)=\left\{q\in D\mid \mathrm{dist}\left(p,q\right)\le Eps\right\}$$

(1)

and (2) Density threshold $MintPts$: if the number of points in the specified $Eps$ of any point $p$ is greater than or equal to the $MintPts$ value, $p$ is the core point of the cluster; if $p$ is not a core point but is within the $Eps$ of a core point, then $p$ is the boundary point of the cluster, and vice versa, $p$ is marked as a noisy point [43]. According to the input parameters, traversing each point in the dataset when the object point satisfies in a particular $Eps$ contains more points than $MintPts$. A cluster with that point as the core object is established, after which the core points are rescanned to determine the core points that are not clustered; the above steps are repeated until there are no new core points in the dataset. Finally, the points that are not in any cluster are labeled as noise points [44].

2.3.2. Markov Chains

A Markov process with a discrete time and state is called a Markov chain, which is a set of series of random variables with Markovian properties [45] that can be written as $\left\{X_n,n=1,2,3\cdots \right\}$. The set of different values used is called the “state space”, the conditional probability of satisfying $P\left\{X_{n+1}=i_{n+1}\mid X_0=i_0,X_1=i_1,\cdots ,X_n=i_n\right\}=P\left\{X_{n+1}=i_{n+1}\mid X_n=i_n\right\}$. Then, $\left\{X_n,=n=1,2,3,\cdots \right\}$ refers to Markov chains [46].

Let $i,j\in I$ from moment $n$. The system is the time $i$ to $n+1$. The moment $j$ refers to the state that the system is in in a given moment, and the conditional probability is

$$p_{ij}\left(n\right)=P\left\{X_{n+1}=j\mid X_n=i\right\}$$

(2)

for Markov chains $\left\{X_n,n\in T\right\}$ at the moment $n$. For any $i,j\in I$, the transfer probability $p_{ij}\left(n\right)$ of the Markov chains $\left\{X_n,n\in T\right\}$ are independent of the start time $n$. Let $P$ be the one-step transfer probability $p_{ij}$; in that case, the one-step transfer matrix and the spatial state is denoted as $I=\left\{1,2,3,\cdots \right\}$.

$$P\left(i\to j\right)=\left(P_{ij}\right)=\left(\begin{array}{ccc}{P}_{11}& \cdots & P_{1n}\\ ⋮& \ddots & ⋮\\ P_{m1}& \cdots & P_{mn}\end{array}\right)$$

(3)

And satisfies

$$\{\begin{array}{c}0\le P_{mn}\le 1\left(m,n=1,2\dots n\right)\\ {\displaystyle \sum _{x=1}^x}{P}_{mn}=1\left(m,n=1,2\dots n\right)\end{array}$$

(4)

3. Results

3.1. The Spatial Distribution of Tourism Flows

Using a correlation analysis between the number of photos and inbound tourism in Indonesia for each month from January 2018 to February 2022, the Pearson’s correlation value between the number of photos and the amount of inbound tourism was 0.911. Therefore, these values were positively correlated with inbound tourism moving in tandem with the number of photos.

Based on ArcGIS visualization, the geotagged photos before and during the COVID-19 pandemic were displayed as points on the map to obtain the spatial distribution of tourist flow (Figure 4); each red dot represents a geotagged photo. The distribution of geotagged photos before and during the pandemic generally showed that Java and the Nusa Tenggara Islands were the main clusters, and the rest of the islands were sporadically distributed. In Java and Nusa Tenggara, there were 16,897 and 6358 geotagged photos before and during the COVID-19, respectively, accounting for 83.75% of the total number of photos on average, while the remaining four islands had more scattered geotagged photos. Therefore, both Nusa Tenggara and Java, in the southern part of Indonesia, were first-class islands and constituted the main body of inbound tourist flow to Indonesia. This reflects the leading position of these two island cities in the behavioral preferences and mental images of inbound tourists. The number of geotagged photos of Sumatra and New Guinea islands accounted for 19.17% and 27.96% of the total data before and during the COVID-19, respectively, indicating that these two islands complementarily carry the inbound tourism flow to Indonesia and belong to the second rank of islands. In contrast, Sulawesi and Kalimantan only accounted for 2.64% and 1.78% of the total number of photos before and during the COVID-19 pandemic, respectively, and they belonged to the third class of islands.

3.2. Construction of Tourism Flow Networks

3.2.1. Tourist Focus Node Identification

The DBSCAN clustering algorithm clusters the nodes, and MATLAB software is used to import the DBSCAN algorithm code. To achieve the optimal clustering effect, different neighborhood radii and density thresholds were selected for the nodes in the administrative regions of different islands. The clusters that were close and had the same connotation were merged according to the internal structural characteristics of Indonesian islands; finally, 89 clusters were retained as the node hot zone in this study (Figure 5).

Prior to COVID-19, all nodal spatial hotspots were included, of which 34 were in Java, the largest number and a diffuse stretch, accounting for 37.7% of the overall number. The Nusa Tenggara archipelago included 28 nodal trajectory tourism hotspots, with the highest concentration in the Bali province. Moreover, during COVID-19, 69 single-node trajectory spatial hotspots were identified. Among them, 23 were in Java, which was more dispersed than the pre-pandemic distribution, accounting for 33.3% overall. The Nusa Tenggara Islands included 22 single-node trajectory tourism hotspots, consistent with the pre-pandemic period, concentrated in the Bali province. Sumatra included eight single-node trajectory tourism hotspots, and Kalimantan, New Guinea, and Sulawesi had the same number of single-node trajectory tourism hotspots as they had prior to the pandemic. However, the neighboring hotspots within the islands were farther apart and less connected.

3.2.2. Classification of Tourist Focus Nodes

The tourism flow was organized into network matrix data, and according to the inter-regional tourism flow network matrix data, the original network matrix was data processed and visualized using the Gephi social network tool (Figure 6). To assess the role and position of each node in the flow at different times, all nodes were categorized into four classes according to the degree of centrality of each node in the network [47]: (1) Core nodes (Denoted by N_I): degree centrality value greater than 15, in a link in a dominant position, the degree of entry and exit located in a high position, strong concentration and radiation ability, the most concentrated node hot zone, and the key link to other nodes to establish the trajectory tourism passenger flow between the hub. (2) Secondary core nodes (Denoted by N_II): degree centrality value between 10 and 15, includes many links with other nodes. (3) Important nodes (Denoted by N_III): degree centrality values between 4 and 9, no significant connection advantages or excessive interactions with other nodes. (4) Edge nodes (Denoted by N_IV): degree centrality values less than 3. 87 (49 such single-node trajectories appeared before and during COVID-19), sorted according to the node centrality index. A total of 217 and 74 multi-node tourism flows were found before and during COVID-19 and were ranked by node degree centrality.

Tourism flows were classified into single-node and multi-node patterns based on the number of nodes. In the single-node pattern, the tourist visits the tourist destination and refers to a tour within a node [48]. Among the major islands in Indonesia, Java and Nusa Tenggara had the highest number of tourists, and constitute several nodal tourism hotspots, forming a patchwork of clusters with continuous spatial stretches that are central among the islands. Tourists generally choose their destinations to satisfy cost minimization [49]. Half of the inbound tourists who choose the single-node pattern will travel along a direct mode of transportation and then “drop in” to a node within Indonesia, ignoring other nodes. In the multi-node pattern, the tourist destination includes touring among two or more nodes in this study. Our research classifies multi-node patterns into 51 categories (including 10 categories for two-node patterns, 12 for three-node patterns, 8 for four-node patterns, 5 for five-node patterns, and 12 categories for more than five nodes) (

Table 1. Classification of tourism flow patterns.

Number of Nodes	Type and Proportion of Tourism Flow before COVID-19								Total
1	NI	10.98%	NII	2.29%	NIII	3.44%	NIV	1.39%	18.1%
2	NINI	17.10%	NINII	5.80%	NINIII	3.41%	NINIV	2.20%
	NIINII	2.22%	NIINIII	1.17%	NIINIV	1.98%	NIIINIII	1.22%
	NIIINIV	1.73%	NIVNIV	1.49%					38.32%
3	NINI	3.22%	NINII	3.17%	NINIII	2.46%	NINIV	2.24%
	NIINII	2.24%	NIINIII	1.71%	NIINIV	1.24%	NIIINIV	1.73%
	NINIINIII	1.71%	NINIIINIV	1.73%	NINIINIV	0.73%	NIINIIINIV	0.24%	22.42%
4	NINI	2.98%	NINII	1.49%	NINIII	1.46%	NINIV	1.24%
	NIINIII	0.73%	NINIINIII	1.22%	NINIINIV	1.22%	NIINIIINIV	0.73%	11.07%
5	NINII	2.24%	NIINIII	1.24%	NINIINIII	1.22%	NINIIINIV	1.73%
	NINIINIIINIV	0.79%							7.22%
Number of Nodes	Type and Proportion of Tourism Flow during COVID-19
1	NI	38.84%	NII	20.13%	NIII	11.38%	NIV	3.06%	73.41%
2	NINI	2.91%	NINII	2.34%	NINIII	2.25%	NINIV	1.38%
	NIINIII	1.28%	NIINIV	0.78%	NIIINIV	0.44%	NIVNIV	0.34%	11.72%
3	NINII	1.98%	NINIII	1.48%	NINIV	1.03%	NIINIII	0.78%
	NIIINIV	0.78%							6.05%
4	NINI	1.74%	NINII	0.78%	NINIII	0.78%	NIINII	0.56%
	NINIINIII	1.13%							4.99%
5	NINIII	0.24%	NINIIINIV	0.24%					0.48%

).

The tourism flows of inbound tourists in Indonesia with multi-node patterns are visualized according to the Markov chain transition probability. The 89 nodes are divided into different intervals with different results, plotted with line segments of different colors and widths. In the two-node model, tourist routes are organized in a “straight chain” (Figure 7), where tourists had a range of activities among the islands before the pandemic, with high-frequency interactions among all core nodes and a simple hierarchical linear model of other nodes and a schedule of 3–4 days. The core nodes between Java and the Nusa Tenggara Islands had a high probability of tourist movement. The tourist flow is obvious and constitutes part of the local network, among which Bali, Yogyakarta, Jakarta, and other N_I nodes had the most frequent flow and a high probability of core node transfer.

In contrast, the link organization during COVID-19 was “radial”, with tourists taking Bali as the core and spreading to the port cities of different islands that serve as windows connecting the islands to the outside during the pandemic and are the distribution centers of tourist routes for each island. There were few flows for type N_II, N_III, and N_IV nodes; moreover, there were other non-port city streams, and the flows were only concentrated on Bali in the Nusa Tenggara Islands. The flows between N_I nodes and N_II, N_III, and N_IV nodes before and during the COVID-19 pandemic followed a dispersion pattern from higher to lower levels, and the emission-like diffusion and post-pandemic radiation were dominated by low-level peripheral nodes with a low probability of node transfer across islands during the pandemic and reduced closeness of connection.

Prior to COVID-19, the tourism flow in the three-node pattern was in the form of a “ring”, which is a kind of transit core class node added to the two-node pattern and is a circular flow based on two nodes. Moreover, the transfer probability in this mode was higher than in other modes. Compared to the two-node pattern, three nodes were chosen as destinations. Covering different classes of paths, the tourist had a wider choice of nodes within the islands, and the path trajectory was more spatially complex and diffuse but still spread outward with the type of N_I nodes in Java and Nusa Tenggara Islands as the center. The three-node path behavior pattern during the epidemic was in the form of a “closed-loop” ring, with most of the travelers’ path choices shifting within Java and Nusa Tenggara, with only two islands as the scope of activities for the three destinations related to the travel restriction policy between the islands (Figure 8).

Before COVID-19, the share of passenger flow in the combination of four-to-five-node patterns was insignificant compared to that of two and three nodes, and this node pattern showed the same circular pattern. However, at this time, the flow was less restricted by spatial distance when the number of destinations increased, and the combination of different levels of nodes across the islands was more diverse, with 7–10 days of deeper travel. The “bowl” shape of the main tourism flow was more obvious, and the hot nodes of the two popular islands in the south carried a large amount of passenger flow, with the core nodes flowing with each other as the main axis between paths. Therefore, due to the pandemic, tourists were guided by “spatial proximity”, and inter-travel flows only occurred between nearby nodes (Figure 9 and Figure 10).

In general, the pattern of tourists traveling in Indonesia was dominated by single-node patterns; multi-node patterns with hierarchical linear routes in the form of rings or straight chains were mostly limited to vertical flows within islands. There were few horizontal flows between islands with large spatial distance spans, almost all of which are spread by radiation from Java and the Nusa Tenggara Islands to other islands; the primacy of these two islands, as the first level islands, is consistent with the findings. Meanwhile, tourists are dependent on and tend to be consistent in their path selection, concentrating on certain core nodes with high visibility and forming traditional tourist routes.

3.3. Tourism Flow Network Structure

Tourism flows between nodes essentially show the movement of tourists between high-ranking hotspots, and the network of inter-island core node flows mapped for high-ranking nodes was based on the role identification of Indonesian core nodes and sub-core nodes. The network pattern of tourism flows changed due to the pandemic. A total of 22 hot zones in five islands appeared in the core node network structure in the Indonesian region before the pandemic, which was composed of 217 and 74 flows before and during COVID-19 (Figure 11).

Before COVID-19, Bali and Denpasar were the two core areas in the Nusa Tenggara archipelago. In contrast with the circular flow pattern within the islands, Bali was used as the hub of distribution and spread to other nodes. Java Island had Yogyakarta, Jakarta, and Ambalava as the three core areas, with Yogyakarta and Jakarta as special zones with a higher degree of concentration and dispersion. Sumatra Island was the second-ranked island; Batu and Padang were the core nodes within the island, and Batu, Batam, and Medan were the secondary core nodes within the island. Medan was a node far from other cities, and the flow pattern of a single node pattern was more obvious; other nodes showed a multi-node pattern. There were no core nodes on New Guinea and the Sulawesi islands, and the only port cities that circulated with the outside as secondary core nodes were the Four Kings Islands and Menado, respectively. In the interlinking of the five islands, the islands were connected by airport routes and maritime corridors, and the attraction of rich tourism resources within the nodes created a linear network pattern covering the whole country.

During COVID-19, the network showed a point or intra-regional distribution pattern, with only Bali in Nusa Tenggara and Jakarta and Yogyakarta in Java as the core nodes for inbound tourists in Indonesia, flowing to Batu, the second core node in Sumatra. In contrast, the flow within the first-rank islands took the core node as a hub for dispersal to other nodes for radiation evacuation. Although the traffic was in a restricted state, the network shows an intra-regional clustering pattern. In favor of national inbound international airports and ports, Jakarta’s role as the capital city depended more on these nodes for inbound tourism during the pandemic than as a tourist resource destination. Consequently, the pandemic did not affect these nodes with higher accessibility to transportation routes but made the distribution role of such nodes stronger. The interconnection between these distribution center nodes was the main route for inbound tourism to Indonesia during the later stages of the pandemic. However, nodes in the third-ranked islands, such as the Siwang Islands and Gyanja, were rich in tourism resources but had low accessibility; although they were important nodes in the network before the pandemic, their tourism levels were disrupted by the pandemic, thus demonstrating obvious gap characteristics in the network at this stage (Figure 12).

3.4. Tourism Resilience Driving Mechanism Analysis

As discussed in Section 3, policies were important tools for strengthening regional resilience during COVID-19 and significantly influenced inbound tourism flows during the outbreak. During COVID-19, most ports of entry were virtually closed, creating a gaping area in the network of tourist pathways. In March 2020, the Indonesian Ministry of Foreign Affairs issued the “Regulation on the Prohibition of Temporary Foreign entry into the territory of the Republic of Indonesia”, which suspended visa waivers and visas on arrival for foreigners from all countries. Government agencies’ strict entry restrictions and pandemic prevention measures resulted in almost no leisure travelers during this period. As the pandemic improved and Indonesia relaxed entry controls, tourism achieved a brief recovery, creating a small regional network of tourist flow trajectories. However, the invasion of the Delta mutated strain in 2021 forced inbound tourism operations to halt again in August when the Indonesian government issued a restrictive entry policy in Regulation No. 27. It only reopened to tourists two months later. In January 2022, the emergence of the Omicron mutant strain upset this balance, and the government issued the circular on the Adjustment of Requirements for Remote Prevention and Control to prevent the importation of infected cases and to temporarily ban foreigners from entering the country before gradually easing the restrictions. In the context of the pandemic, changes in entry policies showed volatility and recurrence, affecting the tourism flow network, thus demonstrating an overall change from cross-regional to intra-regional tourism flows. Changes in national policies affected traffic accessibility and were particularly evident during COVID-19. When the internal pandemic status stabilized, entry management restrictions were gradually relaxed, and more ports of entry were opened to allow tourism to slowly recover. However, when virus mutations increased the difficulty of pandemic prevention and control, the government again adopted strict entry management measures.

Before COVID-19, the interconnections between the five islands relied on airport corridors, as well as the rich tourism resources within each node and high-quality infrastructure under local economic development, forming a national linear network pattern. During COVID-19, owing to the influence of the national pandemic prevention policies and transportation accessibility constraints, the network of tourist pathways was distributed in a point-like or intra-regional pattern. This study constructs the influence mechanism and enhancement of regional network resilience during the pandemic (Figure 13). The driving mechanism of the network pattern influence factor demonstrates that the marginal nodes reflect the vulnerability that was evident during the pandemic impact owing to the low centrality of the nodes and the thin connection between the internal and external islands, while the core nodes dominate because of the prominent network connection position, highlighting the core nodes by the failure of other nodes. The anti-disturbance motive reflects the level of regional resilience of the network. In contrast, the resilience of vulnerable edge nodes to the risk and the reduction in the possibility of dominant core node failure enhances the overall centrality of nodes. This enhances regional proximity links, thus enriching cross-regional connectivity and securing each tourism node.

4. Discussion and Conclusions

4.1. Findings

Employing Indonesian inbound tourists as the study sample, this study uses photo information data on the Flickr social platform and explores the changes in the spatial distribution of tourism flow in Indonesia before and during COVID-19. Moreover, it refines the behavioral patterns of tourism flow and summarizes the mechanisms driving the resilience of tourism flow networks, with the following main findings:

(1) Regarding spatial distribution, photographs of geographic markers before and during COVID-19 showed stage and regional unevenness. Java and Nusa Tenggara were the main concentrations, while the rest of the islands showed sporadic distribution.

(2) COVID-19 influenced the travel flow behavior pattern, showing obvious scale differences in space. Multi-node travel flow patterns dominated pre-COVID-19 and display link pattern variability according to the number of nodes, while single-node travel flow patterns dominated during COVID-19. There was an unevenness within the travel flow network, and the nodes had obvious core and edge areas. The crisis of COVID-19 has caused a simplification of the overall paradigm shift in travel flows from networks to lines and nodes.

(3) The spatial hierarchical spreading of tourism flows in Indonesia is obvious, and the performance before and during the epidemic shows a descending hierarchical spread from high-level tourism areas or tourism center cities to low-level tourism areas or secondary tourism center cities in sequence.

(4) During COVID-19, tourism flow resilience was influenced by the volatility of national pandemic prevention policies. Changes in national policies affected traffic accessibility, demonstrating that the marginal nodes reflect the vulnerability to pandemic shocks owing to the hypocentricity of the nodes and the thinness of the connections within and outside the islands. In contrast, the dominance of the core nodes highlighted the resistance to disruptions owing to the prominence of the location of network connections, reflecting the regional resilience level of the tourism flow network during the epidemic.

4.2. Suggestions

The constant mutation of the SARS-CoV-2 virus caused many uncertainties in the tourism industry, but it also provided new models and development opportunities for its reconfiguration. At present, affected by the pandemic, in the overall dynamics of passenger flow, the spatial solidification of tourist behavior patterns is obvious; moreover, the core nodes with strong agglomeration weaken the status of other nodes, the single convergence of behavior patterns, and the low level of inter-island interaction. To accelerate the reconstruction of the post-pandemic trajectory network, this study proposes different modes of optimization. Indonesian tourists’ current tourism flow patterns are somewhat flawed, with a clear spatial solidification and concentration in one hotspot area leading to a weakening of other areas and an uneven tourism flow network. Moreover, they are mostly limited to the core and sub-core regions within. This study puts forth relevant suggestions to construct a new network pattern: (1) First, accelerating the flow between nodes, improving the centrality of all nodes for the single-node tourism flow model, promoting the nodes’ differentiation, and enhancing the node region’s internal connections are crucial. For the direct transportation mode, the tourism behavior in a node, through the change to “base type”, relies on convenient transportation to join more neighboring nodes, prolonging the stay in the node. (2) For the two-node tourism flow mode, by appropriately increasing the number of nodes and interesting themes in the series, two nodes can be developed into more than three connotative closed-loop spatial trajectory patterns to create a high-quality tourism product tour line to realize the transformation from “simple line” to “ring or chain”. (3) For a multi-node tourism flow mode, appropriate contact with the tourism resources of other nearby islands, from the first-class islands to the tourism resources of the second- and third-class islands, can help form a multi-level tourism line and cooperation to create tourism products so that most tourists perceive the “surrounding radiation” type of depth in their tour. Simultaneously, we should strengthen the prevention of risks and ensure the safety of the nodes to realize the change from “chain or ring” to “radiation”.

In summary, the research framework herein is feasible. It can comprehensively capture the tourism flow changes and recovery in coastal areas from individual nodes to the whole network before and during the pandemic. From a practical perspective, the findings align with the rising trend of coastal tourism in the international market. They can provide scientific informational support for worldwide tourism companies to help them expand overseas tourism markets, develop tourism routes, and formulate reasonable marketing strategies in the context of the pandemic. These findings can also offer helpful guidance and instructions for tourism planning and scientific development in coastal destinations. Furthermore, they can guide Indonesian tourism destinations to improve their management level and develop more experience-worthy tourism products in the future. Ultimately, they will promote the sustainable development of tourism in coastal areas and prevent the occurrence of crisis events.

4.3. Limitations, and Further Research

This study has some limitations. First, exploring the impact and influence of the pandemic on the tourism industry is a long-term project. This article is only available for raw data through February 2022. COVID-19 is an ongoing crisis, and subsequent studies should focus on the ongoing impact and recovery of the pandemic situation, national pandemic prevention policies, and tourism recovery measures on the tourism flow network in Indonesia. Second, owing to the influence of uploaders’ own habits, most users did not label detailed specific demographic characteristics such as gender, travel mode, education, income, or occupation. Although this study collected as many samples as possible, the travel behavior reflected by them could not fully reveal the spatial movement patterns of all tourist groups.

In a follow-up study, data sources should be enriched, and data should be collected and aggregated using multiple methods, such as with a combination of offline and online data, to draw more accurate conclusions. The textual information in the crawled data is rich in semantic and emotional information, and future research will focus on the mining of emotional information to comprehensively explore certain correlations between tourists’ emotional preferences and tourism flow and improve the accuracy and comprehensiveness of the data.

This paper intends to expand the resilience mechanism model in the future by capturing the hierarchy of regional resilience cognition with multi-level analysis method and multi-dimensional analysis as follows:

(1) Regional resilience involves patterns, actions, and policies at different geographical scales, such as global, national, regional, and local scales.

(2) Multi-dimensional analysis is used to analyze the pattern of association among various elements in four dimensions: regional economic resilience, regional social resilience, regional environmental resilience, and regional institutional resilience, as well as their spatial and temporal changes.