Reshuffling city life: spatial and functional dynamics of urban activity in Tokyo during COVID-19

ABSTRACT

The COVID-19 pandemic, and the measures to curb it have profoundly affected the geography of urban activities in the past years. In this paper, we discuss its effects on urban activity in Tokyo during the first wave of COVID between February and July 2020. Different from other papers, which have analysed general changes in urban activity levels or changes in specific activities, we have focused on changes in activity levels in different types of multifunctional urban activity centres (UAC), allowing us to reveal interactions between UAC types, (combinations of) activities and location within a wider urban system. Our results show how the distribution of urban activity across UAC changed in space and time in reaction to pandemic measures, and relate these dynamics to the spatial patterns of functional specialization of UAC. The existing spatial pattern of UAC allowed urban activities to redistribute spatially, but continue without too much inhibition. Moreover, these changes appeared to be temporary, rather than resulting in irreversible urban transformations. Our analysis thus suggests that Tokyo’s multilayered polynuclear structure appeared to contribute to the city’s pandemic resilience, allowing urban activities to spatially reorganize, without needing to resort to a total lockdown and collapse of urban life.

KEYWORDS:

• Urban activity centres
• COVID-19
• resilience
• Twitter data
• Tokyo

Highlights

• Social media-based method to study the spatial and functional dynamics of city life.

• COVID-driven changes in urban activities and in spatial structure interact.

• Urban activities in Tokyo redistributed across urban activity centres (UAC).

• Changes in activity levels in urban subcentres’ depend on their functional mix.

• Tokyo’s multifunctional polycentrism enhances its disaster resilience.

1. Introduction

It has become a truism to state that COVID has profoundly shaken up everyday life around the world. In cities in particular, not only daily practices, but also their spatio-temporal organization have been unsettled by top-down anti-pandemic measures, as well as the strategies individuals have developed to sustain everyday life under novel conditions. Whereas plenty of studies have mapped out changes in social behaviour and economic activities at the micro- and the macroscale (e.g. Cartenì, Di Francesco, & Martino, 2020; de Oliveira & de Aguiar Arantes, 2020; Fatmi, 2020; Lemieux, Milligan, Schirle, & Skuterud, 2020), few have attempted to understand the underlying changes in the spatial patterns of such activities (Noszczyk, Gorzelany, Kukulska-Kozieł, & Hernik, 2022; Yabe et al., 2020).

Urban spatial patterns have long made up the core of urban studies, from the Chicago School’s concentric and sectoral models of Burgess and Hoyt (Burgess, 1924; Hoyt, 1939), to the more polycentric urban structures described by Harris and Ullman (1945), Mc Millen and Mc Donald (McDonald, 1987; McMillen, 2001; McMillen & McDonald, 1997) and the Los Angeles School (see Davis, 1990). Recently, supported by the availability of new kinds of (big) data, these rather static models of urban spatial structures have given way to more dynamic analyses of a wider range of urban non-residential activities such as work, commerce, culture, leisure, tourism, education … and the multifunctional urban activity centres (UAC) where they tend to concentrate (Wang, Yuan, Wei, Chen, & Wang, 2021; Zhong, Arisona, Huang, Batty, & Schmitt, 2014).

In this paper, we discuss the short- and medium-term changes in urban activity patterns related to the COVID-19 outbreak in Tokyo, Japan. Tokyo is a polycentric metropolis, the result of long term, sometimes unplanned developments interacting with dynamic economic and human flows (Bagan & Yamagata, 2012; Toshiseibi, 2020). This dynamic urban system did not come to a grinding halt in reaction to the first wave of COVID-19 in Japan (early 2020). Rather, the anti-pandemic measures resulted in a drastic and quick reorganization of urban activity centres.

The objective of this paper is to identify such spatio-temporal changes in urban activity in the designated period, and to understand why certain UAC have been affected differently from other. Understanding the spatial pattern of changes can help us assess and improve the resilience of cities with regards to pandemics or other disasters (Acuto, 2020; Banai, 2020; Chen, Guo, Pan, & Zhong, 2021). Following the United Nations Office for Disaster Risk Reduction (UNDRR) , we define the disaster resilience as ‘the ability of a system to resist, absorb, accommodate, adapt to, transform and recover from the effects of a hazard in a timely and efficient manner … ’. As such events are expected to happen more often in the future, the spatial dynamism of urban activity patterns will become a crucial strategic challenge for cities worldwide.

This article makes a key contribution to addressing this challenge, in pioneering a method to study the combined spatial and functional dynamics of city life at the sub-city scale of UAC through the combination of Twitter and POI-data. Whereas earlier papers have either focused on changes in particular activities, or on the overall changes in activity at the city level, this article gauges both the spatial reshuffling of particular types of urban activity, as well as changes in the overall level of activity in different UAC. As areas of intense and concentrated urban activity, UAC are both potential hubs of pandemic contagion, as nodes of economic and social networks. Consequently, understanding their dynamics is crucial to enhance a city’s pandemic resilience.

This article is one of the few studies which have analysed how the spatial and functional reshuffling of city life in its various guises in reaction to COVID-19 can be related to the long-term spatial-functional organization of the city. It spells out important insights in the role of urban spatial structure for urban pandemic resilience.

In the following section, we review the existing literature on how COVID-19 has affected urban activity patterns and indicate how our contribution helps to fill extant gaps in this literature. Second, we discuss the data and methods used. Third, our results reveal how the distribution of urban activity across UAC changed in space and time in reaction to pandemic measures. We also show how these dynamics were related to the patterns of functional specialization of different UAC. Fourth, in our discussion, we offer tentative explanations for the patterns observed and discuss the limitations of our study. In our conclusion, we discuss the potential of our method for further analysis and explain how our study can provide insights for megacity resilience planning.

2. COVID-19-related changes in urban activities

In early 2020, COVID-19 rapidly spread from Wuhan to cities across the world (Lu, Stratton, & Tang, 2020). Preventive government measures to decrease social contact and connectivity strongly affected social and economic activity. Numerous analyses have been produced on how pandemic measures changed urban life, speculating on how urban transport, urban economies, urban spatial structure and urban governance can be improved to make cities more resilient to pandemic crises (Sharifi & Khavarian-Garmsir, 2020). Some studies made a comprehensive analysis of the changes in various types of urban activities. Saha, Barman, and Chouhan (2020) for instance, demonstrated how not only human mobility, but also activities in retail and entertainment facilities, parks and workplaces decreased considerably in India. Most however, have focused on changes in specific types of activities and not in their mutual relationships within an urban system.

A large number of studies have focused on the effects on mobility and transport. They indicate a global reduction of mobility (Aloi et al., 2020; Bucsky, 2020) both in number of trips as in the distance travelled (de Haas, Faber, & Hamersma, 2020). A shift from collective towards individual modes of transport has been observed in many (Musselwhite, Avineri, & Susilo, 2020; Sung & Monschauer, 2020), although not all cities (e.g. Nguyen & Pojani, 2021 on Hanoi). Whether this resulted in more cycling and walking or motorized transport, seems to depend on available infrastructure, as well as purposes of the trip. In general, recreational cycling and walking increased, whereas commutes were negatively affected by remote work or online education measures (Buehler & Pucher, 2021).

Studies discussing the effects on workplace activity show a strong relation with the spatial distribution of particular economic sectors (Brodeur, Gray, Islam, & Bhuiyan, 2021). Whereas some types of activity (e.g. office work) have seen a shift to remote work, other sectors continued their activity or have witnessed a rise in furlough and lay-offs (Adams-Prassl, Boneva, Golin, & Rauh, 2021). Working and studying from home also affected other activities. Kaufmann, Straganz, and Bork-Hüffer (2020) show how students receiving online education also shifted other activities to a digital environment.

The collapse of international tourism (Abbas, Mubeen, Iorember, Raza, & Mamirkulova, 2021; Gössling, Scott, & Hall, 2020) and the hospitality industry (Gürsoy & Chi, 2020), and the growth of local outdoor recreation also received a lot of attention. In cities in particular, green public spaces and urban forests are increasingly valued (da Schio et al., 2021).

Most studies have analysed the effect of (combined or individual) measures on specific activities of individuals (e.g. Rutynskyi, Kushniruk, 2020; Xu, Wang, Dong, Shen, & Xu, 2020; Iranmanesh & Alpar Atun, 2021), or at aggregated macro-economic levels (Brodeur et al., 2021; Lemieux et al., 2020). Only a handful of studies have analysed the meso-level spatio-temporal changes in urban activity structures related to anti-pandemic measures, often making use social big data, which facilitate dynamic analyses. Combining Earth observation data with mobile phone GPS data, Straka et al. (2021) reveal how activity patterns shifted in American cities during the 2020 spring lockdown period. Using mobile phone geolocation data, Romanillos et al. (2021) have demonstrated how urban activity in the metropolitan area of Madrid (Spain) declined mainly in areas dominated by education, office or leisure functions; whereas activity was largely maintained in areas with more industrial, commercial or health-related facilities or in residential zones. A global comparison of change in night-time activity in cities based on night-light imagery, has not only revealed significant inter-regional differences, but also a centre–periphery dynamic, with a decrease in lighting in inner cities (Xu, Xiu, Li, Liang, & Jiao, 2021). Using STRAVA and Google mobility data, Venter, Barton, Gundersen, Figari, and Nowell (2020) discuss how in Oslo, Norway, recreational activity intensified most in green and remote areas, replacing indoor activities in the city centre.

In line with this growing body of literature, this study will focus on changing activity patterns in Tokyo, and analyse its relation with the existing spatial structure of UAC. Different from earlier articles on meso-level spatio-temporal changes, we explore the usefulness of geolocated tweets and OSM POIs and develop a method to test the correlation between changing activity levels and the spatial distribution of activity types. The latter has also been analysed by Romanillos et al., based on land use maps, but our use of POI from OSM allows a more detailed investigation of activity types and characterization of UACs. This allows us to gain new insights in how the spatial organization of UACs influences how cities are affected by pandemic crises. Additionally, we contribute to the existing literature by merging two aspects – spatial and functional – by analysing not only the redistribution of urban activities across Tokyo due to COVID-19, but also relating it to the specialization of places in certain activities. This approach allowed to reveal that Japan’s lack of local specialization is one of the keys to the city’s resilience.

3. Materials and methods

3.1. Social media data for identification of urban activities

A wide variety of new datasets and methods have become available for the analysis of urban spatial structures and the identification of UAC (Cai, Huang, & Song, 2015; Leslie, 2010; Ma, Osaragi, Oki, & Jiang, 2020). The authors of these works elaborated comprehensive methods based on novel data types including cell phone and social media data. Various analytical tools such as innovative spatio-statistical modelling, night-time imagery, spatial network analysis, Space Syntax which allows to evaluate the centrality of segments of any road network, are applied for the identification of UAC (Cai et al., 2015; Hillier, 2001; Riguelle, Thomas, & Verhetsel, 2007; Veneri, 2013; Zhong et al., 2014).

Social media data are now particularly popular in spatio-temporal analyses of urban structures (Poorthuis, Shelton, & Zook, 2021) having shown to identify concentrations of urban activity (Poorthuis, 2018). Social media activities are typically performed when users encounter something new, navigate through a city or stay in a certain location worthy to be shared with others (Bendler, Brandt, & Neumann, 2015; Kaplan & Haenlein, 2010). At the same time, social media-use related activities can be extremely heterogeneous (Frias-Martinez, Soto, Hohwald, & Frias-Martinez, 2012): tweets or Instagram posts can be made while working in the office, visiting a tourist attraction or shopping in the mall.

In this research, we used Geotagged tweets (tweets with coordinates acquired from the smartphone’s GPS sensor). Such tweets provide spatio-temporally referenced messages with user-created semantic information (Steiger, Westerholt, Resch, & Zipf, 2015), offering insights in the demographic, thematic, behavioural and contextual characteristics of users with a reference to time and space. This kind of data offers a broad spectrum of options for the analysis of human activities in urban settings at a very fine spatio-temporal resolution (Azmandian et al., 2013; Jurdak et al., 2015; Poorthuis, 2018). The fine spatial and temporal resolution of geotagged tweets is crucial to this research since we intend to delineate and study UAC based on clusters of intense human activity, measured at the level of individuals, rather than on aggregated data at the level of pre-defined administrative neighbourhood or district boundaries (Poorthuis, 2018). Secondly, we will study the temporal dynamics of such spatial urban structures on a month-by-month basis. Classic economic or census statistics could not match the high temporal resolution of social media data, which are published almost without time lag.

The dataset of geotagged tweets was obtained through the library ‘rtweet’ developed for the R-Studio, the extraction was taking on a daily basis starting two months before the study period and until the end of it since it was the only option given existed subscription. Additionally, a dataset of tweets for February–April of 2017 extracted in the same way (Boratinskii & Tikhotskaya, 2021), is used as a reference sample from corresponding pre-COVID periods. Concerning the ethical aspect of our methodology, we worked with the database of geotagged tweets that may be related to privacy issues. We preserved the anonymity of users by not publishing nor sharing any of their deanonymizing data such as Twitter id, account name, whereas any data potentially containing sensitive information were not extracted for the database at all.

Originally, the Twitter database included more than 2 million of observations. Some users however use tags not to specific (a café or a park) but to general locations (a district or ward of Tokyo) as an alternative to GPS-generated locations. In the latter case, Twitter calculates the coordinates of the tweet based on the geometric centre of that location. To avoid such overly inaccurate geolocations, we removed all tweets tagged to an imprecise general location from the database. Resultantly, only half of the tweets were available for the analysis.

3.2. Identification and typology of UAC in Tokyo

In this article, we reproduced Boratinskii and Tikhotskaya (2021)’s approach to delineate Tokyo’s UAC. Boratinskii and Tikhotskaya (2021) combined two datasets to map leisurely, non-employment-related urban activities: 1 million geotagged tweets made in 2017 and the points of interest (POI) of OpenStreetMap (n.d.) of the same year. Whereas tweets are linked to urban activity of individuals, and sensitive to the organization of temporary events and other short-term dynamics, POI represent the venues and locations where such urban activities can take place. The integration of social media and POI allowed to detect those locations where human activity is most intense over a period of time. We repeated this exercise with tweets and POI from the first half of 2020 that were used afterwards for the analysis of dynamics.

UAC were identified and delimited in three steps. First, a grid of regular hexagons with a radius of about 200 m was constructed over the territory of Tokyo, and the number of geotagged tweets per cell was calculated. The results plotted on the map demonstrated that low-density hexagons in the city centre often had higher values than relatively high-density cells in the periphery. In other words, the overall centrality trend (overconcentration of tweets in the city centre and underrepresentation in the periphery) allows to identify the global city centre but not local maximums of urban activity inside and outside of it. To overcome this issue, spatial autoregression was used. We calculated an approximation model of tweets’ distribution, explained by the distance from the city centre (overall centrality trend) and the values in the neighbouring cells (local trends). The latter was included in the model with an autocorrelation component applying the concept of geographical neighbourhood, meaning that cells with at least one common corner point are neighbouring. The residuals of this model revealed cells where the actual tweets’ count is significantly higher than modelled. Finally, for those hexagons, UAC were delimited by identifying the concentrations of POI. As a result, 146 UAC were identified in Tokyo (Figure 1).

Figure 1. Classes of UAC in Tokyo based on new POI database (2020).

In Boratinskii and Tikhotskaya (2021), UAC were divided into five categories, ranked by the number of POI present. This classification was updated with the more recent database of POI. The result again was very similar (Figure 1 and Table 1). Boratinskii and Tikhotskaya (2021) found that the overall number of POI in UAC correlates with their functional diversification, similar to Christaller’s central place theory. Centres of the first class have almost all functions represented, with a slight dominance of foodservice, leisure and tourism. Second-class centres are their smaller counterparts: they are also multifunctional but their size and uniqueness are lower. The functional spectrum of third-class UAC is more limited: the central ones are mostly office districts, while peripheral ones are large commercial centres catering for local consumption. The fourth- and fifth-class centres specialized in essential services for local people; they accommodate necessary facilities such as police stations, schools, fire stations, pharmacies, post offices, stores and ordinary public food facilities.

Table 1. List of UAC by classes.

First class		Fourth class		Fifth class
A1	Ueno	D1	Narimasu	E1	Itabashi	E46	Kita-Shinagawa
A2	Akihabara	D2	Sakuradai-2	E2	Komagome	E47	Haneda
A3	Ginza	D3	Nogata	E3	Sengoku	E48	Kasai Rinkai Koen
A4	Mita	D4	Nishi-Ogikubo	E4	Yanaka	E49	Shin-Kiba
A5	Shibuya	D5	Takaido	E5	Asakasa Mitsuke	E50	Tatsumi
A6	Shinjuku	D6	Meiji Daigaku – Izumi	E6	Takashimadaira	E51	Toyosu
		D7	Sasazuka	E7	Akatsuka	E52	Tsukuji
Second class		D8	Hatagaya	E8	Tobu Nerima	E53	Nishi Kasai
B1	Ikebukuro	D9	Umegaoka	E9	Kami Itabashi	E54	Toyocho
B2	Meiji Jingumae	D10	Sangenjaya	E10	Oizumi-gajuen	E55	Kiba
B3	Aoyama	D11	Soshigaya	E11	Asahicho	E56	Harucho
B4	Toranomon	D12	Futakotomagawa	E12	Hikarigaoka Park	E57	Funabori
B5	Ogikubo	D13	Jujo	E13	Hikarigaoka	E58	Higashi-Ojima
B6	Asagaya	D14	Oji	E14	Minamioizumi	E59	Koiwa
B7	Koenji	D15	Otsuka	E15	Oizumi	E60	Shin-Koiwa
B8	Nakano	D16	Sugamo	E16	Shakuji Koen	E61	Kameido
		D17	Tabata	E17	Sakuradai	E62	Shibamata
Third class		D18	Nippori	E18	Ekoda	E63	Aoto
C1	Akabane	D19	Ueno Koen	E19	Musashi Seki	E64	Tokyo Skytree
C2	Kita-Senju	D20	Meiji Daigaku	E20	Kamisakuji	E65	Nishikahei
C3	Takadanobaba	D21	Kudanshita	E21	Iogi	E66	Takenotsuka
C4	Higashi-Nakano	D22	Yotsuya-3	E22	Shimoigusa	E67	Daishimae
C5	Okubo	D23	Yotsuya	E23	Mitakadai	E68	Minami-Senju-2
C6	Asakusa	D24	Akasaka	E24	Wada	E69	Minami-Senju
C7	Suidobashi	D25	Tokyo Tower	E25	Sengawa
C8	Iidabashi	D26	Minami-Azabu	E26	Kitami
C9	Marunouchi	D27	Ebisu-2	E27	Seijogakuen-mae
C10	Meiji Jingu Stadium	D28	Nakameguro	E28	Yoyogi Uehara
C11	Roppongi	D29	Meguro	E29	Yoyogi
C12	Daimon	D30	Shinagawa	E30	Yoga
C13	Ebisu	D31	Gotanda	E31	Komazawa 2
C14	Shimokitazawa	D32	Ayase	E32	Komazawa
C15	Edo	D33	Kameari	E33	Idejiri Ohashi
C16	Chitose	D34	Kanamachi	E34	Tamagawa
C17	Jiyugaoka	D35	Keisei Takasago	E35	Ookayama
C18	Kamata	D36	Tateishi	E36	Gakugei Daigaku
		D37	Kinshicho	E37	Chucho 1
		D38	Ryogoku	E38	Musashi Koyama
		D39	Suitengu-mae	E39	Meguro 2
		D40	Nihonbashi	E40	Shimono
		D41	Kasai	E41	Osaki
		D42	Odaiba	E42	Oimachi
		D43	Kokusai Tenjijo Seimon	E43	Zōshiki
		D44	Omori	E44	Keikyu
		D45	Otorii	E45	Aomono-Yokocho

Despite our focus on non-employment activities, we found that UAC largely overlapped with office concentrations, suggesting a strong relation between employment and leisurely activities. Even more clearly, the vast majority of UAC in Tokyo are located around railway stations which is also reflected in Zacharias, Zhang, and Nakajima (2011) and Calimente (2012). The number of POIs in UAC tends to correlate with the passenger turnover of the adjacent stations (Boratinskii & Tikhotskaya, 2021).

3.3. Dynamics of urban activity in UAC

In the next step, we detected changes in the distribution of urban activity across UAC of Tokyo from January to July of 2020 on the basis of tweets.

Spatial units for the analysis of the tweets database – UAC – were prepared in advance; we selected months as temporal units given the size of the dataset and the variability of anti-COVID-19 measures in the period. In case of smaller temporal units, the number of tweets would be insufficient for the adequate representation of activities, while longer periods would not allow to estimate the effect of specific anti-COVID measures. Spatio-temporal changes in tweets were calculated based on the growth rate, which is widely used for demographic and economic comparisons (e.g. De Kroon, Plaisier, Van Groenendael, & Caswell, 1986; Wu, 2007). We adapted its formula to our dataset as follows: (1) $\mathrm{Gr}={\displaystyle \frac{T_m-T_{m-1}}{T_{m-1}}\times 100\text{\%}}$(1) where Gr is the growth rate; T_m is the number of tweets in a territorial unit in the given month, and T_m–1 is the number of tweets there in the previous month. This formula was applied to calculate the changes in urban activity level in UAC between all pairs of neighbouring months (January–February, February–March, and so on) included in the study period. This simple yet powerful method allows to detect and illustrate the ongoing dynamics within pre-defined spatial units, in our case UAC, as single objects that was required for the following functional analysis. The growth indicator revealing spatial distribution of dynamics in connection to the situation in other UAC, such as spatial autoregression or geographically weighted regression would give us rather a view on a whole system, while peculiarities of particular centres might be lost or blurred. The growth rate figures for UAC are demonstrated in a series of maps that were made using GIS software (Figure 2).

Figure 2. The growth rate of tweets’ number between neighbouring months of the study period.

3.4. Functional analysis of the dynamics

Next, we analysed which types of urban activities were most and least affected by COVID-19-induced changes. In the 2020 OSM-POI database, which consists of about 80,000 points divided into 123 classes, we selected the 92 most relevant classes and grouped them into eight groups of functional types as presented in Table 2. Relevant classes are those who potentially generate urban activity and attract people in a concentrated area.

Table 2. Functional types of POI.

Type	Classes	Description	Number of POI
Food	Café, pub, bar, fast food, food court, restaurant, beverages	Venues for eating out	20,669
Tourism	Archaeological, arts centre, artwork, castle, guesthouse, hostel, hotel, monument, museum, theme park, tourist info, travel agent	Unique or related to the tourism, have a high potential to attract non-local people	5217
Open space	Bench, biergarten, camp site, dog park, fountain, garden centre, golf course, park, picnic site, pitch, playground, track, viewpoint, zoo	Objects and venues associated with open space activities	2908
Leisure	Attraction, cinema, library, mall, nightclub, sports centre, stadium, swimming pool, theatre	Ordinary leisure facilities that are used regularly, alternatives may be found elsewhere	925
Daily goods	Bakery, convenience, department store, greengrocer, outdoor shop, pharmacy, supermarket, butcher	Shops specialized in essential goods such as provisions and medicines	8781
Other goods	Beauty shop, bicycle shop, bookshop, car dealership, chemist, clothes, computer shop, dentist, florist, furniture shop, gift shop, hairdresser, jeweller, kiosk, mobile phone shop, optician, shoe shop, sports shop, stationary, toy shop, veterinary, video shop	Shops specialized in non-essential goods visits to which can be postponed if necessary	8969
Public facilities	Bank, community centre, courthouse, doctors, embassy, fire station, graveyard, laundry, police, post office, public building, shelter, town hall, wayside shrine	Services required to maintain the normal functioning of the city, mostly essential	6891
Education	College, kindergarten, nursing home, school, university	Education facilities	2556
Sum	92		56,916

To determine changes in and between activity types, we analyse statistical relationships between the dynamics of urban activity level and the concentration of POI of designated functional types within UAC through the calculation of the Pearson correlation coefficient (PCC). (2) $r_{\mathrm{xy}}={\displaystyle \frac{{\sum }_{i=1}^n(x_i-\overline{x})(y_i-\overline{y})}{\sqrt{{\sum }_{i=1}^n{(x_i-\overline{x})}^2}\sqrt{{\sum }_{i=1}^n{(y_i-\overline{y})}^2}}}$(2) where r is PCC, x_i and y_i are the individual sample points of samples X and Y, and $\overline{x}$ and $\overline{y}$ are their means. In addition to the assumptions of PPC, we verified the applicability of the results by non-parametric Spearman’s rank correlation coefficient as some of the samples had outliers and quasi-normal distribution.

The absolute difference in the number of tweets in each UAC between neighbouring months and between all months and January were used as indicators of changes in urban activity (sample X). The number of POI of all types in UAC was used as an indicator of the concentration of functions (sample Y). Using both comparisons over the entire period, we can track how the relationship between the level of urban activity and the concentration of facilities related to various functions changed (Figure 3). The neighbouring months’ comparison (Figure 3(A)) aims to illustrate the overall dynamics, whereas the comparison with January (Figure 3(B)) highlights the specifics of each month better. Statistically insignificant values (significance level of 0.05) are below 0.11 by modulus in all tables (Figures 3 and 4).

Figure 3. (A) PCC between the absolute change in tweets’ count and the number of POI in UAC (neighbouring months). (B) PCC between the absolute change in tweets’ count and the number of POI in UAC (January as the basic period).

However, such correlation coefficients may be more dependent on the total number of POI located in UAC than on the functional role of a centre. For instance, the number of restaurants is likely to be higher in a first-class than in a fifth-class UAC irrespective of the role of food there. This bias is especially relevant for the functions with a relatively even spatial distribution (food, public); the results for types with a more complicated geography (e.g. tourism, education) are ‘cleaner’ in this sense; their presence is more frequently related to the specialization of a centre rather than to its scale.

One way to eliminate the effect of this dependence is to use the share of POI belonging to a certain functional group in the total number of POI in a particular UAC as the indicator of functional specialization and the growth rate of tweets’ count as the indicator of changes in urban activity. In contrast to the absolute figures, these allow to take into account the relative importance of functions in each centre, not their concentration, and, therefore, to clarify how dynamics of urban activity are related to these functions in particular. The PCC between these two variables are presented in Figure 4(A,B).

Figure 4. (A) PCC between the growth rate of tweets’ count and the share of functional types of POI in UAC (neighbouring months). (B) PCC between the growth rate of tweets’ count and the share of functional types of POI in UAC (January as the basic period).

4. Results

In this section, we discuss how the distribution of urban activity across UAC changed geographically during the study period by means of maps (Figure 2). We also discuss the relation of these changes with anti-COVID measures and the geographical patterns of UAC specialization. Whereas the former is based on a qualitative comparison with the timeline of the pandemic situation and anti-COVID measures, the latter will be done based on a per month analysis of the correlation between the absolute changes in urban activity and the number of objects of each functional group (Figure 3(A,B)) and the correlation between the growth rate of urban activity and the share of each function in UAC (Figure 4).

For ease of reference, we have included a flow chart (Figure 5) of the main COVID-related measures in Tokyo and Japan, and a graph of the evolution of the pandemic situation in Tokyo on the basis of the number of deaths. Internationally, Tokyo was considered successful in its fight against coronavirus, implementing quick and strong, albeit non-binding measures (Yabe et al., 2020). Its mortality and infection rates (Figure 6) were low in comparison to other megacities like New York and Moscow during the same period (Moscow Mayor, 2020; The New York Times, 2020; Updates on COVID in Tokyo, 2020). This is also reflected in the official death figures that in total did not exceed 340 with the highest number per day at 12 (Figure 7).

Figure 5. Event flowchart of the first wave of COVID-19 in Tokyo (Ameba News, 2020; Bloomberg, 2020; Cohen, 2020; Kyodo News, 2020; NHK, 2021).

Figure 6. Number of COVID-19 positive patients per day (Updates on COVID-19 in Tokyo, 2020).

Figure 7. Number of COVID-19 deaths per day in Tokyo (Deaths by date of death, 2022).

4.1. Overall change

Correlation coefficients in Figure 3(B) have negative values throughout the whole study period. We can thus state that, even though the weather improved, the urban activity level in all of the following months was lower than in January in centres with a high number of POI and higher in UAC with less POI. Figure 3(A), in contrast, contains positive values as well as negative, which indicates that urban activity increased in centres with a large number of POI in certain sub-periods.

Figure 4(A,B) demonstrates that the correlation between the specialization of centres and the growth rate of urban activity is lower than the correlation between the absolute indicators. Another important difference is that in Figure 4, positive and negative figures are mixed in all sub-periods, while it is the case only for July in Figure 3.

4.2. February

The first cases of COVID-19 in Tokyo were registered on January 24 but there was no significant growth in the first two months of the year, and no significant measures were taken by the government in this period. Nonetheless, from January to February, urban activity intensified slightly in most peripheral UAC of Tokyo. However, most of large the UAC in the city centre experienced almost no relative changes. The correlation analysis demonstrated that both absolute and relative activity decline mostly corresponds with leisure and tourism-focused UAC.

4.3. March

On February 27, the Prime Minister asked to close all elementary, junior and high schools in Japan until early April, and plead to shift to remote work and close non-essential businesses (Kyodo News, 2020). The figures started to increase in the second half of March (Figure 6) and more strict measures followed. A lockdown, coercing people to stay at home, is constitutionally forbidden, so the government needed to resort to urgent advises to prevent infections (Steen, 2020).

In March, the further spread of COVID combined with measures implemented at the end of February resulted in the first signs of decentralization. At this stage, the decline of urban activity was most visible around the southern part of the Yamanote Line. The population density in these centres is lower than in surrounding territorial units, while the office density is higher (Figure 1) (e.g. Ginza, (A3), Akihabara (A2), Shibuya (A5), Marunouchi (C9)). However, this does not apply to, for instance, Shinjuku (A6) and Ikebukuro (B1) where urban activity remained stable or even. It is rather complicated to identify a single trend for the periphery. Although the overall level of urban activity outside the Yamanote line was higher than in January, the redistribution of activity between the UAC of Tokyo remained quantitatively insignificant in March.

The correlation analysis showed that the dynamics of urban activity between February and March correlated with the specialization in education and open-space activities and the concentration of leisure facilities in UAC.

4.4. April–May

On March 25, the Governor of Tokyo Yuriko Koike held an emergency press conference urging residents to shift to remote education and work as much as possible on weekdays and refrain from going out at night (Mainichi Japan, 2020). Later in March, the citizens of Tokyo were asked to stay away from Closed (poorly ventilated), Crowded places where Close-contact is unavoidable (e.g. restaurants, nightclubs and bars (NHK, 2021)). Avoiding these san mitsu (Three C’s) became the basic principle in Japan’s struggle against the pandemic.

At the beginning of April, visits from 73 countries were already restricted. On the 7th of that month, a state of emergency was declared in six prefectures including Tokyo for 30 days. Although figures declined from late April onwards, it was extended until May 25 (Bloomberg, 2020). The Governor requested the closure of six types of businesses in the city including entertainment facilities, universities, cram schools, facilities for exercise and play, theatres, facilities for gatherings and exhibitions, commercial facilities, however, there were no official penalties for being opened. Restaurants were asked to limit business hours (NHK, 2021). During the state of emergency, people were free to leave their homes to buy food or essential goods: supermarkets, convenience stores, pharmacies and other essential businesses in Tokyo remained open (Steen, 2020), whereas companies were requested to introduce remote work for as many employees as possible (Martin, 2020). Some public parks were closed and the consumption of alcohol in other open spaces was prohibited. When the state of emergency was lifted, people and businesses were asked to maintain the abovementioned and general sanitary measures optionally.

The state of emergence seems to have initiated the most dramatic decentralization of urban activity. The distribution of urban activity changed drastically from March to April, remaining more or less stable from April to May, until the state of emergency was lifted at the end of the period. The highest decline rates were detected in the largest UAC located in the city centre; the figures were slightly lower in second- and third-class peripheral centres. Relative growth was detected mostly in fourth- and fifth-class centres with an originally low level of activity.

Functionally, the decline of urban activity from March to April was primarily associated with the concentration of food and leisure (Figure 3(A,B)). The latter is also reflected in a relatively high correlation between the negative growth rate of urban activity and the specialization of UAC in food (Figure 4(A)). Interestingly, the decline of urban activity related to the number (Figure 3(A)) of essential stores (daily goods) was as significant as related to the concentration of non-essential ones (other goods) despite the objects of the former group are officially permitted for visits. Moreover, the corresponding correlation coefficients are almost similar throughout the whole study period. Most likely, the reason for it is that in Tokyo, stores of all types tend to be clustered in the same places, usually around railway stations.

Quite a significant inverse correlation was also detected with tourism (and leisure) facilities. In contrast to the abovementioned functions, urban activity intensified in UAC where the relative role (not the absolute number) of shops providing daily goods is higher than average (Figure 4(A,B)). In addition to daily goods, urban activity increased in centres specialized in public facilities (Figure 4(A,B)).

The correlation between the concentration of open spaces and the decline of urban activity decreased to the lowest value among all functions in April (Figure 3(A,B)) and kept this position until the end of the study period. Curiously, positive PCC values of education in Figure 4(A) seemingly contradict the shift to remote education requested at the end of March.

4.5. June

Soon after the state of emergence was lifted, on June 2, the TMG issued the first ‘Tokyo Alert’, saying that there are signs that the infection situation is worsening again (34 new people were infected with the new coronavirus in Tokyo), but it was retracted only nine days later. In general, June seemed to announce a period of policy relaxation, reflected in a reversal of the spatial distribution of tweets: urban activity surged back again in the largest central UAC and decreased in the smallest peripheral ones (except for the furthest from the city centre). This relaxation also reveals itself in the highest PCC values between the growth of tweets and the concentration of public food facilities and ‘unnecessary’ services such as leisure, tourism and other goods (Figure 3(A,B)).

4.6. July

The situation exacerbated rapidly again at the end of June, so the Governor called for avoiding downtown nightlife venues. Yet, the TMG confirmed 224 new infections in Tokyo on July 9, the highest number ever and 33% of them were related exactly to night-time facilities with close-contact entertainment venues (host clubs, cabaret clubs and girls’ bars) (NHK, 2021). Two weeks later, more than 300 new infections urged the governor to ask residents to stay at home for four consecutive holidays if possible (NHK, 2021). Logically, the rapid growth of activity shown in June slowed down or even reversed to a slightly negative trend again in first- and second-class central UAC. In contrast, peripheral second-class centres started to gain activity again in comparison to the previous month. The main belt of urban activity growth shifted from the Yamanote Line to the western semi-periphery again. The correlation between the decline in urban activity and the total number of points (Figure 3(B)) increased compared to June. Except this, the correlation analysis demonstrated almost no correlation of urban activity dynamics between June and July with the functional specificities of UAC. The exception was education that has the highest figure for July in Figures 3 and 4.

5. Discussion

5.1. Overall dynamics and trends

5.1.1. Strong effect of COVID-19 dynamics

A general finding that we can support with our data is a strong effect of the COVID-19 pandemic on everyday life in Tokyo. Even before the first national pandemic measures in the last days of February (Kyodo News, 2020), individual residents seem to have taken some precautions, as in February, although the overall changes in activity were small, a slight peripheralization of activity was already observable. Over the whole period of observation, the change appears, however, to be mainly geographical, as throughout the period, the overall activity level was not severely affected. This includes a general decentralization–recentralization, as well as shifts of activity levels between specific categories of facilities and UACs.

5.1.2. Decentralization–recentralization dynamics

Whereas in February, central UACs along the Yanamote line remain highly used, activities seem to decline in March after the call for remote work. Mostly areas with a low population density, but a high office density seem to be affected (Figure 1). Nightlife areas with no specific relation to office work remain highly visited. This decentralization effect became more salient after the declaration of the state of emergence early April. Both the central location of UAC, as well as the high density of facilities in central UACs seemed to have played a role in the decline of tweets in centrally located UACs. The factor of centrality is definitely related to the nature of the state of emergency: the government asked residents to avoid unnecessary trips (Martin, 2020; Steen, 2020), which strengthened the decline of commuting intensity that was already initiated by remote work and education. In addition, it might have been caused by the san mitsu advice, which stimulated Tokyo citizens to use small peripheral centres close to home instead of larger, more popular commercial centres dense with facilities of their special wards or the capital region. This fact is also reflected in Yabe et al. (2020) who concluded that the length of trips decreased and more people stayed near their homes during March and April.

The number of new COVID cased started to plummet soon after the declaration of the state of emergency, however, some restrictions in Tokyo were lifted only by the very end of May, so both spatial and functional patterns of urban activity did not change between these months: the city centre was still relatively empty in comparison to January, while peripheral centres experienced more visitors than usual irrespective of their size. This is also supported by Denyer (2020) who found that commuting trains – the main mode of transportation from the periphery to the city centre – emptied in April and May, as people opted for smaller peripheral UACs closer to home. As a result of these processes, by the end of May, urban activity in Tokyo was severely decentralized.

In the course of June, after the state of emergency was lifted, we observed an intensification of urban activity, as well as a recentralization. We can relate this to the intensification of commuting and office work taking into account that urban activity did not increase in high-class peripheral UAC that are mostly commercial centres but increased even in small centres located close to the Yamanote Line which tend to have higher concentrations of offices. This recentralization was reversed again slightly in July. Although no new measures were introduced, a new wave of COVID-infections might have stimulated Tokyo residents to avoid central places again. Undoubtedly, it is necessary to take it into account as a separate factor but this dependence made the functional analysis imprecise. It was not clear whether a certain statistical relationship between the concentration of a particular function in UAC and the dynamics of urban activity was in fact dictated by the concentration of POI belonging to this particular type or it was more related to the centrality of its location within Tokyo and its transportation network. This issue was partially tackled by the second part of the correlation analysis when we used specialization rates and direct comparison of the correlation of urban activity dynamics with various functions in all UAC.

Here we could add more reflection on the limitation of our data, and the limitation of our analysis as one that focuses on month-by-month comparisons in one year, but not between years. However, we can rule out seasonality to some extent by using the dataset of geotagged tweets for February–April of 2017. The comparison of March–April sub-periods, which may be considered the most important ones, illustrated that the dynamics of urban activity in 2017 were much less extreme in comparison to 2020 that indirectly highlights the impact of COVID for the study period.

5.2. Specific interactions with spatial structure

5.2.1. Particular functions

UAC’s with particular functions were differentially affected in different stages of the pandemic. From the very start of the pandemic, a decline in activity in tourism-related UACs is observed, corresponding with the early introduction of travel bans from two Chinese provinces (Hubei (1.02) and Zhejiang (12.02); Yabe et al., 2020) and a more global reaction in the tourism sector. The number of incoming tourists more than halved from January to February (JTB, 2021) Moreover, In our typology, tourism includes not only activities directly related to travel but also some non-daily (exceptional) leisure activities, which seems to have been avoided by precautionary domestic tourists early in the pandemic as well.

The gradual abandonment of leisure-related UACs illustrates that after international tourists; also local people were giving up leisure activities. However, some spatial redistribution seemed to have happened as well in reaction to policies. The entertainment districts in the vicinity of Tokyo Station on the east side of the Yamanote Line shortened the business hours of more sophisticated bars/restaurants that serve alcohol according to the instructions of Tokyo. In contrast, there are many entertainment districts in Shinjuku and Ikebukuro with more casual bars/restaurants that did not shorten their business hours. Therefore, office workers who normally enjoyed eating and drinking near Tokyo Station may have moved to Shinjuku/Ikebukuro for these activities. As a result, clusters of corona-infected persons have occurred in the districts west of Shinjuku station.

Only in April, a more general decline in activity in food and leisure-associated UACs was observed. This corresponds well with the fact that the state of emergency implied the closure of leisure businesses and encouraged the shift to delivery services and shortened business hours for restaurants. Moreover, the number of tourists in the country reached its minimum by April (JTB, 2021). However, the abovementioned trends might not only be related to COVID-measures. March is the end of the financial and academic year in Japan and students usually go travelling and shopping after graduation. Many people also change residency in March for that reason, which explains a slight absolute growth in use of stores selling non-essential (other) goods in March that reversed to a significant decline in April. Still, we can be 90% confident that compared to March–April 2017 the negative correlation of the March-to-April urban activity change with the concentration of other goods and tourism was higher in 2020. This indicates that anti-infection measures might indeed have caused a change in dynamics.

After the lifting of the state of emergence, tourism and leisure-related UACs attracted more visitors again in June. The number of tourists, especially international, barely increased (JTB, 2021) – the existence of a positive statistical relationship was rather caused by the intensification of local activity in places such as museums and art centres which we classified as tourism and, of course, in adjacent open spaces of the city centre.

In contrast to tourism and leisure-related spaces, the activity level in UACs focused on daily goods and public services increase during the state of emergency. During the state of emergency, ‘People in Tokyo were free to leave their homes to buy food or essential goods. Supermarkets, convenience stores, pharmacies, and other essential businesses in Tokyo will remain open’ (Steen, 2020). Public services include medical and other services, which are not so easily offered online and had to continue operating. UAC with a high share of daily goods and public facilities are situated in peripheral, residential areas and are mostly visited by local residents. Thus, small generic UAC attracting residents of surrounding neighbourhoods witnessed the strongest growth of urban activity during our study period.

Finally, open spaces revealed contradictory dynamics. In March, a significant decline of urban activity in open spaces contradicts findings in the international literature, which emphasized how people moved recreational activities from indoor to outdoor (e.g. Venter et al., 2020). In the following months, figures moved more in line with expectations. Since open spaces are mostly associated with non-essential, recreational activities, the first reaction to the spread of COVID-19 was to avoid them. With little knowledge on COVID’s infection ways, the risk of activities in different conditions was unclear and residents seemed to have not wanted to take risks. Once open spaces were not mentioned as places to be avoided in the main set of preventive government directives issued in April; they were increasingly seen as the only safe recreational option: walking and jogging were not prohibited (Denyer, 2020). Consequently, urban activity there increased, although not to the pre-COVID level. Moreover, March and April represent a traditionally important period for open space in Tokyo associated with the yearly hanami (cherry blossom festival). People enjoy drinking under the cherry blossom trees in parks on their way home from work, even though COVID-measures prohibited drinking alcohol in parks and other open spaces, and some parks remained closed.

5.2.2. Functional combinations in particular places

Finally, few UACs are completely dominated by one type of activity, and effects of COVID on activity in UACs seem to be related to the particular combination of function in a specific UAC, more than to the dominant or typical function only. Such was the case with education facilities. In March, a decline of urban activity in education-related UAC can confidently be explained by the closure of public schools. However, in later periods, such UACs have witnessed an increase in activity even though schools remained closed. In July as well, education reveals a high positive correlation with urban activity. This might be explained by the fact that the majority of UAC-forming university campuses are located in peripheral areas and are usually surrounded by daily goods shops and open spaces. Even though schools remained closed, these facilities compensated the loss of activity residents of neighbouring blocks started to more intensively use these places for essential and leisure activities when commuting to bigger, more central centres was considered dangerous. In July, campuses reopened shortly for exams, adding to the already intensive use of peripheral centres for daily goods. Hence, it is the functional combination of educational and daily goods facilities that explain these centres’ prominence during the first COVID-outbreak, rather than one specific type of facility.

5.3. Caveats and limitations

The presented analysis is largely based on social media data, which have specific limitations that might have affected our results to some extent. Firstly, as the use of social networks is not universal, social media data struggle with an important sampling bias (Boyd & Crawford, 2012; Steiger et al., 2015), which differs from network to network. Recent studies of the digital divide, in particular in relation to Twitter use, are scarce in Japan (Akiyoshi, 2021). Twitter is the most popular social network in Japan with 35% of the population as active users (Bugajski, 2020; Statista, 2021). However, as in other countries (e.g. Blank, 2017), it does not represent all social groups equally. A prefecture-based spatial analysis of Japan’s digital divide revealed that whereas income levels do not affect Twitter subscriptions, the latter are higher in prefectures where educated young adults prevail (Nishida, Pick, & Sarkar, 2014). Minors and older people are underrepresented compared to the age group from 20 to 60 (Bugajski, 2020). As people of different age and of different educational levels also tend to conduct different activities during the day, this might limit our perspective on the presence of activity centres. Age-specific activities (e.g. health care, education …) may fail to be registered correctly, and our findings might have been influenced stronger by the activity shifts of more highly educated residents.

Also, the spatial reference of tweets may be inaccurate due to the effect of the built environment, atmospheric radiation, and mobile device characteristics. In urban areas, however, these technical positioning errors rarely exceed 10 m (Bettinger, 2019; Blunck, Kjærgaard, & Toftegaard, 2011), while in this study, we aggregate at a larger scale, so this spatial inaccuracy does not really affect our analysis.

A second caveat is related to our periodization. The periodization of our data was done at the moment of data collection. By using monthly data, we were able to conduct analyses on a relatively fine-grained time scale. However, the structure of our database did not fully correspond with the periodization of relevant anti-COVID measures. Similarly, the time frame over which our data were collected (January–July 2020) captured the changes happening during the first wave of the COVID-pandemic, but did not allow us to rule out the effect of yearly social rhythms (e.g. annual festivities, academic year, seasons, …) by a month-by-month comparison with a pre-COVID year. We have tried to make up for this by a comparison with the dynamics during the short period (February–April) of 2017 as well as a qualitative assessment of the impact of such annual events in the discussion. Yet evidently, a full statistical correction based on an analysis of the same period in 2019 would have been better and should make up the object of a follow-up analysis.

Finally, our analysis has focused on spatial dynamics of activity between mostly multifunctional urban activity centres. This methodological choice is definitely warranted, and part of the originality of this analysis compared to studies focusing on specific activities. We have been able to show how such multifunctional places with a combination of specific facilities, and with a specific location in relation to the city centre and public transport lines, affected where people concentrated at different moments in time during the first pandemic wave. Yet we also encountered, during our analysis, strong effects of COVID on the use of specific types of facilities which we haven’t studied systematically with the data at hand, but, as other studies have shown before, can be equally relevant.

6. Conclusion

Using real-time data collection during the first months of the COVID-19 pandemic, this paper contributes to an understanding of how city life reacts to disasters. Different from other studies, this paper did not focus on changes in general activity levels or changes in particular types of activities, but on changes in activity levels in multifunctional UACs. This offers us a broader perspective on the wider impact of the pandemic on city life in general, and allows us to identify spatial as well as functional reshuffling between and in particular UAC.

The dynamism of urban activity in Tokyo from January to July of the first pandemic year, 2020, was identified by means of a multistage analysis combining spatial modelling, statistical techniques and even qualitative methods. We explored how the spatial distribution of urban activity changed and how it was related to the implementation of anti-COVID measures. During the state of emergency in Tokyo from April to May, the strongest decentralization of activity was observed. As the Tokyo Government made great effort to minimize human mobility and reduce unnecessary activities, commuting, as well as leisure, public food and non-essential shopping-related activities declined. Simultaneously however, more essential activities were reshuffled towards the periphery, as revealed by the growing importance of small, but mostly multifunctional peripheral centres.

The rapid spatial redistribution of urban activities during the initial, most uncertain period of the pandemic, can be interpreted as indicative of Tokyo’s disaster resilience, embedded in its polycentrism. Japanese cities have generally been planned with regard to earthquake and flooding-resilience. This paper reveals how the urban structure of Tokyo also guarantees a degree of pandemic resilience, as its spatial flexibility allowed urban life to continue and adapt to the pandemic with minimal costs (see also Yabe, Rao, & Ukkusuri, 2021). The spatial structure existing in Tokyo appeared to be very stable, yet also malleable: UAC did not collapse or disappear completely due to COVID-19; instead, urban activity was redistributed among them.

Even during the state of emergency and the strongest decentralization, the largest central UAC survived and continued their functioning at reduced capacities. This seems to be a consequence of the polynuclearity of Tokyo’s urban structure. Since there is no single CBD in the Japanese capital but rather a constellation of multifunctional subcentres along the Yamanote Line, the negative effect of anti-COVID measures was distributed evenly among them, as well as among more peripheral centres. Simultaneously, such decentralization of activities allowed for a rapid reduction of contact density, a major factor in the spread of COVID (Verma, Yabe, & Ukkusuri, 2021).

In addition, our research demonstrated that the spatial dynamics of urban activities are differentiated according to their functional typology, and the specific targets of anti-COVID measures. UAC with a dominance of essential functions or with safer open spaces were affected less than centres with a high concentration of businesses providing non-essential services. The former are mostly located in residential areas where the overall level of activity was relatively low in the normal pre-pandemic situation. Some of these centres are commercial streets near university campuses or railway stations with adjacent shopping and leisure centres in the middle of densely built housing neighbourhoods. The multifunctionality of these centres continued to attract considerable user activity, even when some of their functions (such as commuting or education) declined. This allowed everyday life to continue more or less unaffected, albeit in different spatio-temporal arrangements.

Whereas this research presents us with important preliminary conclusions on urban pandemic resilience, our findings also contain some recommendations for further research. First, an expansion of the study period would give more insights in the differential effect of the different waves of infection and policy measures. The strongest infection wave in Tokyo ended only in October 2022, and insights in its effects are missing.

Secondly, the suggested relationship between pandemic resilience and urban spatial activity structure could only be tested thoroughly in a comparative analysis of COVID-related dynamics in cities with different spatial structures and similar pandemic policies. In future research, we hope a comparison of Tokyo’s experiences with those of other Japanese cities with different spatial structures would be possible. Such interurban comparisons of pandemic reactions are rather scarce so far, not in the least because such endeavour would struggle even more with issues of data validity if based on social media information. The use of mobile spatial statistics might be a (albeit costly) way to solve this issue.

Thirdly, our analysis does not only show how cities can be planned in order to enhance pandemic resilience. It also provides insights in the potential effects of spatially targeted pandemic crisis measures. This opens up opportunities for modelling and scenario-building, which would be extremely useful for policy development aiming to enhance a city’s pandemic resilience and prepare for future pandemic situations. Such modelling, however, will require a multilevel analysis distinguishing the effects on activities in specific types of facilities more clearly from their combined effects in multifunctional UACs.

Disclosure statement

No potential conflict of interest was reported by the author(s).