Local travel behaviour under continuing COVID-19 waves– A proxy for pandemic fatigue?

COVID-19 continues to threaten the world. Relaxing local travel behaviours on preventing the spread of COVID-19, may increase the infection risk in subsequent waves of SARS-CoV-2 transmission. In this study, we analysed changes in the travel behaviour of different population groups (adult, child, student, elderly) during four pandemic waves in Hong Kong before January 2021, by 4-billion second-by-second smartcard records of subway. A significant continuous relaxation in human travel behaviour was observed during the four waves of SARS-CoV-2 transmission. Residents sharply reduced their local travel by 51.9%, 50.1%, 27.6%, and 20.5% from the first to fourth pandemic waves, respectively. The population flow in residential areas, workplaces, schools, shopping areas, amusement areas and border areas, decreased on average by 30.3%, 33.5%, 41.9%, 58.1%, 85.4% and 99.6%, respectively, during the pandemic weeks. We also found that many other cities around the world experienced a similar relaxation trend in local travel behaviour, by comparing traffic congestion data during the pandemic with data from the same period in 2019. The quantitative pandemic fatigue in local travel behaviour could help governments partially predicting personal protective behaviours, and thus to suggest more accurate interventions during subsequent waves, especially for highly infectious virus variants such as Omicron.


Introduction
Coronavirus disease 2019  continues to threaten human life and society. Close contact and longer-range airborne are the main SARS-CoV-2 transmission routes Tang et al., 2021;Zhang et al., 2021a). General non-pharmaceutical interventions (Miao and Zhang, 2022;Zhang et al., 2021b) such as social distancing, city lock-down, school closure, environmental controls, and personal hygiene had been used in many countries. In some cities, complete lockdown or partial lock-downs have been implemented. Pandemic fatigue is defined by the WHO as demotivation to follow recommended protective behaviours, emerging gradually over time and affected by a number of emotions, experiences and perceptions (WHO, 2022). This so-called pandemic fatigue has been observed worldwide in terms of adherence to these protective behaviours (Petherick et al., 2021). Factors such as the need to reduce the economic burden of these behaviours, and time-varying perceived risk adjustment (e.g. following the observed reduced death rate) have been cited as possible explanations. However, the exact mechanism for the occurrence of pandemic fatigue remains unknown. Understanding pandemic fatigue is important, since the efficacy of current vaccines against newer variants is potentially low (Mahase, 2021). The pandemic fatigue phenomenon might also be an important factor when governments consider easing restrictions to live with the virus.
Local travel behaviour is understood to be an indirect indicator that can be used to assess social distancing during a pandemic (Badr et al., 2020;Zhang et al., 2021c). Due to the pandemic, people did indeed reduce their movement in almost all locations (Forsyth et al., 2020;Irawan et al., 2022). Public transport, such as subways, play a critical role in virus transmission, particularly in highly populated cities (Ghosh et al., 2020). In New York City, the subway ridership decreased by 89 % on average between 23 March and 19 April 2020, due to the COVID-19 pandemic (Teixeira and Lopes, 2022). In Seoul, there was a 41 % reduction in subway ridership during late February 2020 (Lee et al., 2020). In some European cities, ridership of public transport decreased by 80 % (Bernhardt, 2020).
Hence, we proceeded to examine changes in local travel behaviours during subsequent pandemic waves in Hong Kong. Hong Kong is one of the most densely populated regions in the world. In 2003, Hong Kong recorded 21.7 % of all confirmed SARS cases worldwide (1755 out of 8906) (WHO, 2003). Major outbreaks of influenza viruses, such as H1N1, H5N1 and H7N9, have also been reported in Hong Kong (Tam, 2022;Wu et al., 2010;Wu et al., 2014). As of 2 September 2020, Hong Kong had reported 12,113 confirmed cases of COVID-19 (HKCHP, 2021). There have been four epidemic waves of COVID-19 since the first case was reported in Hong Kong on January 23, 2020. No city-scale lockdown has ever been implemented in Hong Kong, but local government constantly advised people to exercise social distancing during each of the outbreak waves. It is unknown if people in general relaxed their personal protective behaviour.
Most of the existing studies of human movement use location data from mobile phones for the analysis. This may introduce bias as mobile phone usage does not reflect the use of public transport. Most passengers use their smartcards when taking public transport, and these records provide accurate data of travel behaviour. Hong Kong has an advanced Metro Transit Railway system, and different demographic groups use different types of smart card such as a student card for example. This system enables us to understand any changes in local travel behaviours of these different population groups.
On average, 12.6 million passengers use the public transport system per day in Hong Kong (based on ridership statistics from 2013 to 2019) (HKTD, 2020). Previous research looking at influenza transmission, showed that 4 % of infections occurred in subways (Cooley et al., 2011). Therefore, we used human behaviour connected to subway riding as an index of social distancing and pandemic fatigue. More than 42 % of all Hong Kong commuters use the Mass Transit Railway (MTR) system every day, as measured over a recent three-year period (HKTD, 2020). In this study, we clarified the changes in local travel behaviour via MTR. We collected more than 4 billion smartcard records of MTR use from 1 January 2019 to 31 January 2021. This allows us to understand the impact of the continuous waves of COVID-19 outbreaks on human travel behaviours in Hong Kong, which may in turn reflect personal protective behaviour.

Data sources
The MTR Corporation supplied detailed MTR data covering the period from 1 January 2019 to 31 January 2021. The data includes masked smartcard information (i.e. the ID of the card was masked and processed without the original card ID), entry/exit station, entry/exit time, and card type for each passenger. Four card types were selected for this study: adult (no discount); child (ages 3-11 years); student (ages 12-25 years and enrolled in a primary/middle/high school, university or institution of higher education); and senior (65 years and above) (Zhang et al., 2021c).
Other sources of data we collected for this study were: (1) Reported COVID-19 data between 23 January 2020 and 31 January 2021 in Hong Kong were obtained from the Centre for Health Protection of Hong Kong (HKCHP, 2021). We combined the infection data and MTR data to analyse how human travel behaviour changed due to the pandemic and affected the spread of infection. (2) Reported COVID-19 data from the start of the pandemic to 31 January 2021 for selected cities (Beijing, Hong Kong, London, Los Angeles, Madrid, Moscow, New Delhi, New York, Paris, Singapore, Tokyo, and Wuhan) were collected from different sources (see Appendix A for details). (3) Monthly local public transport data for Hong Kong between January 2019 and January 2021were obtained from the Transport Department of Hong Kong (HKTD, 2020). (4) When comparing human travel behaviours among different cities/countries (Fig. 5, 6B, and 6C), traffic congestion data obtained from TomTom (www.tomtom.com) was used as an index for local travel behaviour (Tanveer et al., 2020).

Data selection and processing
Up to 31 January 2021, Hong Kong experienced four COVID-19 pandemic waves. We selected four peak pandemic weeks, the week in each pandemic wave that had the highest number of recorded daily cases. The four pandemic weeks are Week 1 (26 March to 1 April 2020), Week 2 (24 to 30 July 2020), Week 3 (4 to 10 December 2020), and Week 4 (18 to 24 January 2021). Four non-pandemic weeks, with the same dates in 2019, were used as the control groups. Changes in local travel behaviour could be obtained by comparing a passenger's data during and before the COVID-19 pandemic, and during the different pandemic weeks.
To ensure the accuracy of the results, we removed invalid data based on the following criteria: (1) There has to be card scanning data for both entering and leaving the system for every passenger. (2) A passenger cannot enter and leave the same station. (3) A passenger has to enter and leave the station during MTR operating hours.
After this data screening, more than 203 million valid subway ridership records were obtained for the four pandemic and nonpandemic weeks. To identify regular daily commutes amongst all this data, we defined consecutive round trips between the same two stations to be regular commutes (Zhang et al., 2021d). The effective reproduction number R t was used to reflect the real-time severity of virus transmission. It measures changes in the transmissibility of a disease such as COVID-19, and represents the average number of secondary infections caused by a single infected individual at time t in the partially susceptible population. An epidemic will tend to decline when R t drops below 1 but grow when R t exceeds 1. We estimated the time-varying R t from the incidence of COVID-19 cases and the serial interval distribution using the method proposed by Cori (2020) and Thompson et al. (2019). Confirmed cases of COVID-19 from the Hong Kong Government are classified into six types of cases, which are reclassified into imported or local cases as the incidence of COVID-19. Asymptomatic cases are excluded in this study since we found that asymptomatic cases can influence the peak value of R t for each pandemic wave, even though we impute onset date of asymptomatic cases using a Weibull distribution of symptomatic cases' onset date. The estimated serial interval for COVID-19 was assumed to be a normal distribution with a mean of 7.5 days and a SD of 6.0 days. In addition, the serial interval distribution can also be estimated from the interval-censored exposure data using the Markov chain Monte Carlo algorithm. All analyses were carried out using Epi-Estim version 2.2-3 and R version 4.0.1 (Code is shown in Appendix B).
When assessing changes in local travel behaviours, a pandemic fatigue score (S PF ) was used (Eq. (1)). The higher the score, the more rapid is the onset of pandemic fatigue.
where N p,fw (N p,lw ) and N np,fw (N np,lw ) are the number of passengers in the first (last) pandemic (non-pandemic) week;Δt (month) is the duration between the first and the last week.
In this study, mobility changes in MTR were represented by the difference ratio of subway ridership in a pandemic week, compared to the average ridership in a non-pandemic control week. In addition, because some students/workers go to school/workplace on Saturdays (Zhang et al., 2021c), we focused on travel behaviour on weekdays and Sundays.

Study area
From the statistics, Hong Kong had a population of 7.47 million at the end of 2020 (HKCSD, 2021). The average daily passenger usage of public transport in Hong Kong from January 2019 to May 2021, was 10.6 million ( Figure S1). In 2019 and 2020, 37.2 % and 35.2 % of all commuters used the MTR, respectively. In Hong Kong, there are 95 MTR stations and 11 lines ( Figure S2).

Temporal local travel behaviours
Hong Kong had recorded a total of 10,453 confirmed cases by 31 January 2021. Four COVID-19 pandemic peaks (including both imported and local cases) can be seen before 31 January 2021 (Fig. 1A). The effective reproduction number is shown in Fig. 1A. The average daily numbers of new cases in the four pandemic weeks were 50.7, 128.7, 100.3, and 75.4, respectively.
Local travel behaviour changed a lot due to the pandemic (Fig. 1B). During the non-pandemic weeks, 78.6 %, 3.4 %, 7.8 % and 10.2 % of MTR passengers used adult, child, student and senior cards, respectively. Non-pandemic travel on Sundays for adults, students, and older people was 20.4 %, 32.1 %, and 17.5 %, respectively lower than on weekdays. However, child passengers increased by 21.5 % on Sundays. The average daily number of passengers during the four chosen pandemic weeks decreased by 51.9 %, 50.1 %, 27.6 %, and 20.5 %, respectively compared to ridership during the four non-pandemic weeks. Adults, children, students, and older people on average reduced their MTR travel by 39.7 %, 80.1 %, 72.5 %, and 35.2 % respectively during pandemic weeks.
In this study, we assumed that most of the return trips per day were for daily commutes (e.g., regular movement between school/workplace and residence). All types of passengers took more return MTR trips per day during the pandemic weeks than in the non-pandemic weeks (Fig. 1C). Travel behaviour on weekdays was more regular (2 trips per day) than on Sundays. Adults and older people gradually reduced their rate of regular commutes, while children and students became more regular during the four pandemic waves.
Subway ridership during the pandemic weeks decreased by 41.3 %, 45.5 % and 51.8 % on weekdays, Saturdays and Sundays, respectively (Fig. 1D). Few people have to go to work on Sundays, which means that more than 50 % of MTR users avoided taking the subway to schools or workplaces if it was not mandatory during the pandemic weeks. During the pandemic, passenger numbers during the peak hours on weekdays, Saturdays, and Sundays, decreased by 41.3 %, 57.1 %, and 66.0 %, respectively.

Fig. 2.
Half-hour subway ridership at residential, workplace, school, shopping area, amusement area, and border stations during the week (WKD) and Sundays (SUN) of the four non-pandemic and pandemic weeks (At the workplace and school stations, the data only showed those passengers with regular commutes. At the shopping and amusement area stations, the data only showed those passengers without regular commutes). Because there was a social movement in Hong Kong in non-pandemic Week 3, the data from week 3 is not shown here.

Spatial local travel behaviours
In order to analyse the local travel behaviour in different types of locations, six areas represented by six subway stations were selected: residential (Tin Shui Wai), workplace areas (Quarry Bay), schools (University), shopping areas (Causeway Bay), amusement areas (Disneyand), and borders (Lo Wu) (Fig. 2). Adults reduced their regular commuters to the workplace by 40.3 %, 46.4 %, 23. 6 %, and 12.5 % during the weekdays of the four pandemic weeks. At the school area, the numbers of commuters decreased by 95.4 %, 93.2 %, and 37.9 % during the pandemic Weeks 1, 2, and 3, respectively (Week 4 was not considered here because the University station was closed during that week). On the non-pandemic weekdays, there was an apparent peak in movement in shopping areas between 6 and 7p.m. On pandemic Sundays, adults, children, students and older people reduced their visits to shopping areas by 55.9 %, 77.5 %, 73.2 % and 55.1 %, respectively. The passenger volume to the shopping area decreased by 75.4 %, 69.9 %, 32.5 %, and 45.3 % during the four pandemic Sundays, respectively.
From the pandemic fatigue scores (Table 1), the most rapid fatigue occurred with workers going to workplaces, with a value of 6.5 % per month. This high value showed the social pandemic fatigue with respect to COVID-19 prevention and control for workers. No further strategies, such as work from home, were implemented during the four pandemic waves. The border areas experienced the least pandemic fatigue, which means that the city lock-down was strictly implemented because of the high morbidity of COVID-19 in other territories. During weekends, the rapid pandemic fatigue occurred with children going to schools, which means that more weekend classes gradually returned to normal operation.
People sharply reduced their travel to amusement areas. Children and students reduced their travel to amusement area by 98.9 %, 98.4 %, 96.6 %, and 84.1 % during the four pandemic weeks. At the largest border point in Hong Kong, almost no passengers entered Lo Wu station during the four pandemic weeks. Overall, during the pandemic weeks, the subway ridership in the residential area, workplace, school, shopping area, amusement area and border decreased by 30.3 %, 33.5 %, 41.9 %, 58.1 %, 85.4 % and 99.6 %, respectively.
During non-pandemic weekdays, Tsim Sha Tsui (285,000 passengers per day), Mong Kok (275,000 passengers per day), and Central (252,000 passengers per day) had the highest passenger flow. Mong Kok (215,000 passengers per day), Lo Wu (209,000 passengers per day), and Tsim Sha Tsui (202,000 passengers per day) had the highest passenger volumes during the non-pandemic weekends. A significant correlation was observed between passenger volume before and during the pandemic (p < 0.001), and also between weekdays and weekends (p < 0.001). Hong Kong residents avoided going to public places with high population density during the pandemic (Fig. 3). However, this intention gradually eroded as the gradient of the red line decreasing from the first to the fourth shows. Some stations had greater passenger flow on Sundays than on weekdays such as borders and amusement areas, especially in the non-pandemic period (green indicates where the percentage reduction is below 0).  (Fig. 4).

Relaxation of local travel behaviours
Children and older people gradually returned to peak hour travel during the week, over the four COVID-19 waves (Fig. 5A). Adults and students also relaxed their travel behaviour during the last two waves compared with the first two waves. Children and students exhibited a more relaxed travel behaviour over the weekends of the last two COVID-19 waves.
We looked at 10 cities that experienced more than two COVID-19 waves to analyse how human travel behaviour changed over time. Residents in all these cities relaxed their travel behaviour during subsequent waves of SARS-CoV-2 transmission (Fig. 5B). People gradually reverted to normal local travel patterns at an average rate of 12.4 % per 100 days. Interestingly, almost all cities showed a similar increased rate of change in congestion, when compared to the same period in 2019. In all studied cities, Beijing had the lowest travel behaviour relaxation rate (0.09 % per day), while Singapore had the highest, at 0.37 % per day.
The delay from the peak of the COVID-19 outbreak to the valley of population flow using the MTR is not associated with the continuing COVID-19 waves ( Figure S3), but rather with the severity of the outbreak (Fig. 5C). The more serious the outbreak, the shorter the response time to changes in local travel patterns. Most people began limiting their local travel 10.0 days after the most serious day of the outbreak, but this delay was reduced to 7.1 days if the outbreak was the most serious ever in that city (Fig. 5C). Hong Kong had the shortest delay in changes to local travel behaviour as a result of the severity of the outbreak, indicating that Hong Kong residents are the most sensitive to the severity of the COVID-19 transmission.

Discussion
Travel behaviour is probably affected by both the prevalence of the disease, and its severity. Many cities and countries restricted travel during the COVID-19 pandemic (Shakibaei et al., 2020). Local travel may also reflect personal responses to disease transmission. Hong Kong offers a good opportunity for study since local travel was not restricted during the entire pandemic period. In this study, we analysed changes in local travel behaviours during four consecutive COVID-19 outbreaks in Hong Kong, using 4 billion MTR smartcard use records from 1 January 2019 to 31 January 2021 (2019 for the control group).
Residents in all countries reduced their local travel due to the COVID-19 pandemic, and the most prominent reduction in number of trips was in public transport (Munawar et al., 2021;Zang et al., 2022;Hartleb et al., 2022). In Canadian cities, peak rush-hour congestion decreased by 54 % to 75 % (Tian et al., 2021). In Australia, the use of public transport dropped by 80 %. In Chile, Peru, and Argentina, residents reduced their trips by more than 50 % (Andara et al., 2021).  (Jing et al., 2020), which means that students and children had the highest resistance to the virus, while older people had the lowest resistance and the highest morbidity. Therefore, a sharp reduction in local travel by older people would be a very efficient way to prevent and control COVID-19 transmission. Authorities should appeal to older people to reduce their travel during a pandemic.
In addition to population group, the type of destination also influences local travel behaviour. We found that the population flow to workplace, school, shopping, amusement and border areas decreased on average by 33.5 %, 41.9 %, 58.1 %, 85.4 % and 99.6 % respectively, during the pandemic weeks. People reduced their travel to workplaces or schools the least, possibly because most of these trips are mandatory. Fig. 3. Relationship between population flow and reduction rate of passenger volume at 95 MTR stations (there are 95 circles in each colour) during the four pandemic/non-pandemic weekdays/Sunday. The percentage reduction is defined as the difference ratio of the subway ridership number between the pandemic and non-pandemic weeks divided by that for the non-pandemic week or the difference between Sundays and weekdays divided by that for weekdays. A negative value means that the pandemic traffic is greater than non-pandemic traffic. (1: Lok Ma Chau (border); 2: Lo Wu (border); 3: Sha Tin; 4: Disneyland Resort (amusement); 5: Chai Wan; 6: Ocean Park (amusement); 7: Sham Shui Po; 8: Tin Hau; 9: Sunny Bay; 10: Sheung Shui (next to the border).).

Fig. 4.
Daily reported COVID-19 cases and traffic congestion in 12 cities/territories. Many trips to shopping and amusement areas are based on personal choice. Densely-populated indoor environments such as shopping malls, restaurants, bars, and karaoke rooms posed high infection risk because of poor ventilation and frequent close contact (Zhang et al., 2021a;Chen et al., 2020;Gu et al., 2021). Considering these high-risk indoor environments, effective prevention and control policies could include local governments closing subway stations near areas with many restaurants and amusement areas. The sharp reduction in the number of people traveling to these areas in Hong Kong show that residents exhibit good protective behaviour with social distancing. Local travel to shopping areas went from 66.4 % in the first pandemic week to 40.1 % during the third pandemic week, and from 89.0 % to 77.2 % for amusement areas.
The particular day of the week (weekday/weekend) also influences local travel behaviour. In Spain, residents reduced their local travel by 86 % on weekends, but only by 65 % on weekdays (Saladie et al., 2020). Our study showed that Hong Kong residents on average, also reduced their local travel via public transport more on Sundays (51.8 %) than on weekdays (41.3 %) during the pandemic weeks, because few people are mandated to go to work/school on Sundays. Residents should control their travel behaviour during both weekdays and weekends. More schools and workplaces should encourage online courses and work-at-home during a serious pandemic, and local governments could monitor the effectiveness of such interventions using smartcard type data.
Due to the consecutive waves of COVID-19, people gradually relaxed their personal protective behaviours. Pandemic fatigue is an expected and natural response to prolonged public health crises (WHO, 2003). Between 1.5 % and 5.5 % fewer residents followed the physical distancing policies over the four pandemic waves in Hong Kong (Du et al., 2021), however, our study showed that the relaxation in local travel behaviour by subway was much greater. Some studies found that ridership on the subway system slowly increased after the outbreak (Lee et al., 2020;Park, 2020), but no studies have shown how travel behaviour relaxed over the consecutive COVID-19 waves. Based on the subway use data, our study also showed a pandemic fatigue for Hong Kong residents. They reduced their subway use by 51.9 % during the first pandemic week, but only reduced it by 20.5 % during the fourth pandemic week. Although the vaccination rate in many countries was high, and existing COVID-19 vaccines appear to be effective against SARS-CoV-2 (Chagla, 2021), the efficacy of current vaccines against newer variants is potentially low (Mahase, 2021). Relaxed local travel behaviour reflected a relaxation in personal protective behaviour, and on roads) in 10 cities worldwide (the points represent the rate of change of congestion in 10 cities during different pandemic weeks); (C) Relationship between the delay from the outbreak peak day to the day when traffic congestion was at its lowest, and the percent of new cases in a pandemic week to the number of new cases for the most severe pandemic week. (In Figures 6B and 6C, grey dotted lines show the change rate for traffic congestion in each of the cities; green and red dotted lines show the trend lines for the fastest and slowest change rate; Moscow and Wuhan were excluded due to the non-existence of multiple waves). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) this trend may increase the potential risk of infection in new waves of virus variants. Local travel behaviours were also strongly influenced by non-pharmaceutical interventions such as stay-at-home orders, regional lockdowns, and travel restrictions (Beaute and Spiteri, 2020;Schlosser et al., 2020;Bian et al., 2021;Hensher et al., 2021), but many of these interventions were gradually removed because of the negative economic impact. As a variant virus, the basic reproduction number of omicron was estimated to be several times greater than that of the Delta variant (Davido et al., 2022), pandemic fatigue lead to a high infection risk when a virus variant with high infectivity broke out in Hong Kong. Local governments should take adequate countermeasures quantitatively that take into consideration the inclination toward pandemic fatigue, especially for those areas with high pandemic fatigue scores such as workplaces and schools.
Pandemic fatigue occurred not only in local travel behaviours, but also in most personal protective behaviours. Americans avoiding small gatherings with family and friends fell from 71 % in May, to 45 % in September 2020 (Meichtry et al., 2020). College students in the Philippines are experiencing physical and psychological discomfort and are beginning to resist the lockdown after a long period of lockdown (Labrague and Ballad, 2020). The elderly cooperated with the quarantine measures for their own protection, while the younger generation, who have not experienced health risks, have gradually stopped complying with the quarantine measures, and report quarantine fatigue (Franzen and Woehner, 2021). Pandemic fatigue is strongly associated with an understanding of the dangers of COVID-19, and it was more common in young people and males during the pandemic (MacIntyre et al., 2021). In China, we found that university students relaxed their protective behaviours of hand-washing and mask-wearing after vaccination, but healthcare workers did not change their personal protective behaviours. In Australia, UK, and USA, avoiding public areas (80.4 %), hand hygiene (76.4 %), mask wearing (71.8 %), and social distancing (67.6 %) were the most common measures adopted during the COVID-19 outbreak, however, following these guidelines became less common from March to July 2020 (MacIntyre et al., 2021). The rate of decline in protective behaviour decelerated over time, with small rebounds seen in later months (Petherick et al., 2021).
This quantitative measure of pandemic fatigue could be used by local governments to impose more informed interventions in the face of additional pandemic waves. For example, they could determine how many workers need to work at home, how many students need to study online, and what percentage of passengers need to wear masks when using public transport. In addition, without considering pandemic fatigue, interventions in different cities/countries could not be adopted because of different human behaviours. Through quantitatively comparing pandemic fatigue in different cities/countries, interventions in other territories could be adopted to Hong Kong, and help assess the effectiveness of each intervention.
Pandemic fatigue has been studied by many researchers. Hong Kong has implemented strict interventions to curb COVID-19 pandemic since January 2020, especially in four pandemic waves. Around 6.60 % more people worry about being infected and 3.77 % of people avoid social gatherings when 100 new confirmed cases was reported (Du et al., 2021). The fourth pandemic would lead to 14 % of the infected less if there is no pandemic fatigue (Du et al., 2021). A survey with more than 500 participants showed that one third of people relaxed their personal protection after roughly-one year since the COVID-19 occurred (Haktanir et al., 2022). During the pandemic, many countries put forward different strict interventions and achieved great progress. However, pandemic fatigue ultimately became the key factor for the following outbreak (Ala'a et al., 2021). Therefore, relieving pandemic fatigue plays an important role in preventing the spread of SARS-CoV-2, especially during the late pandemic period. Governments should pay more attention to the biopsychosocial nature of human beings when making interventions to prevent and control infectious disease transmission. If governments want to control infectious disease transmission effectively, they need to keep monitoring human behaviors on interventions (e.g. mask wearing, social distancing) and emphasize the importance of interventions when pandemic fatigue was observed. Especially when a new outbreak is coming, the government should remind all residents paying more attention to their personal protection than last time.
There were some limitations in this study. The accuracy and reliability of the results would be influenced by following situations: (1) all Hong Kong residents were divided into four groups: adult, child, student and senior. However, some seniors, children, and students use nondiscount cards (e.g. adult), which may introduce some error; (2) these four pandemic weeks and the non-pandemic weeks may not be completely representative of the pandemic and non-pandemic periods; (3) that one week was considered to be the length of each wave may also introduce some errors. Due to different cultures, human travel behaviours, and urban public transports design, the results may not represent the characteristics of other cities. However, an accurate result which is suitable for other cities could be generated using their local smartcard data. Pandemic fatigue is influenced by many factors, such as local policies (Table S3) for COVID-19 prevention and control, and the number of mild cases and asymptomatic cases, which cannot be completely excluded. These all could introduce some errors. Finally, although the reduction in subway ridership due to the social movement in October 2019 had riding almost disappeared by December 2019, a little error may exist when calculating pandemic fatigue for the Week 3 (4 to 10 December 2019) control group.

Conclusion
Adults, children, students, and older people on average reduced their subway travel by 39.7 %, 80.1 %, 72.5 %, and 35.2 % respectively, during the four pandemic weeks. The population reduced their travel to the amusement area (85.4 %) and border area (99.6 %) the most, and to residential (30.3 %) and workplace (33.5 %) the least during four pandemic weeks. People gradually reverted to their normal travel behaviour at an average rate of 0.124 % per day. Relaxed human travel behaviour reflected their protective behaviour during the COVID-19 pandemic, which very likely resulted in an increase in the infection risk during subsequent waves of SARS-CoV-2 transmission, especially for the more infectious variants such as the delta variant.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability
The authors do not have permission to share data.