key: cord-0785975-2zrdoup6 authors: Jiang, Peng; Fu, Xiuju; Van Fan, Yee; Klemeš, Jiří Jaromír; Chen, Piao; Ma, Stefan; Zhang, Wanbing title: Spatial-Temporal Potential Exposure Risk Analytics and Urban Sustainability Impacts related to COVID-19 Mitigation: A Perspective from Car Mobility Behaviour date: 2020-08-19 journal: J Clean Prod DOI: 10.1016/j.jclepro.2020.123673 sha: 909731784f3bf23704c998a6678a55e130d5c1ab doc_id: 785975 cord_uid: 2zrdoup6 Coronavirus disease-2019 (COVID-19) poses a significant threat to the population and urban sustainability worldwide. The surge mitigation is complicated and associates many factors, including the pandemic status, policy, socioeconomics and resident behaviours. Modelling and analytics with spatial-temporal big urban data are required to assist the mitigation of the pandemic. This study proposes a novel perspective to analyse the spatial-temporal potential exposure risk of residents by capturing human behaviours based on spatial-temporal car park availability data. Near real-time data from 1,904 residential car parks in Singapore, a classical megacity, are collected to analyse car mobility and its spatial-temporal heat map. The implementation of the circuit breaker, a COVID-19 measure, in Singapore has reduced the mobility and heat (daily frequency of mobility) significantly at about 30.0 %. It contributes to a 44.3 % to 55.4 % reduction in the transportation-related air emissions under two scenarios of travelling distance reductions. Urban sustainability impacts in both environment and economy are discussed. The spatial-temporal potential exposure risk mapping with space-time interactions is further investigated via an extended Bayesian spatial-temporal regression model. The maximal reduction rate of the defined potential exposure risk lowers to 37.6 % by comparison with its peak value. The big data analytics of changes in car mobility behaviour and the resultant potential exposure risks can provide insights to assist in (a) designing a flexible circuit breaker exit strategy, (b) precise management via identifying and tracing hotspots on the mobility heat map, and (c) making timely decisions by fitting curves dynamically in different phases of COVID-19 mitigation. The proposed method has the potential to be used by decision-makers worldwide with available data to make flexible regulations and planning. The outbreak of coronavirus disease-2019 (COVID-19) prompts a series of social, economic 30 and environmental issues. It is yet difficult to project the lasting impacts, such as on the envi-31 ronment system and the public health system, due to the uncertainty over the shape of eco-32 nomic recovery. By 20 May 2020, over 5 M of COVID-19 cases have been confirmed in 216 33 countries (WHO, 2020). COVID-19 might coexist with us for a long time. Even after appar-34 ent elimination, a resurgence could be possible as late as 2024 (Kissler et al., 2020) . Current 35 priority has been given on treatment, vaccine development, containment and mitigation strat-36 egies to reduce the mortality and infection rate. tensity using spatial statistics to make data-driven planning decision for urban development. 60 dress, and location with X and Y coordinates) in Singapore is available from the public data 129 platform (GTA, 2020a). On the data platform, the data of car park availability (GTA, 2020b) 130 of 1,904 residential car parks (Figure 1 ) are updated at a minute level. More than 3 M pieces 131 of data records (e.g. name, time of records, total lots and available lots as in Figure 1 it has been excluded the occasional outliers with total lots fewer than available lots. The data 136 of mobility trend reports are from the Apple COVID-19 data platform (Apple, 2020) . The 137 data on transportation fuel (CSD, 2020) and data on half-hourly electricity system demand 138 (EMA, 2020) Despite a comparatively tiny area of 719.1 km 2 , Singapore has close to 1 M vehicle popula-144 tion (Diao, 2019) . Like the data platform on car park availability, other data platforms also 145 provide relevant data. For example, the Singapore Traffic Watch project (SGTrafficWatch, 146 2020) offers nation-level and region-level traffic conditions, especially the hourly bus obser-147 vation counts and the available taxis for hire in the last 24 h. Since the main objective of this 148 study is to measure spatial-temporal potential exposure risks through near real-time car mo-149 bility changes before and during the circuit breaker, attention is primarily focused on the data 150 coverage in space and time and the data accessibility. This motivates us to highlight the high-151 resolution data on nationwide car park availability up to several months. After checking for public data online in Singapore, this data set is undoubtedly the most suitable one to achieve 181 Potential exposure risk is defined as the potential risk caused by exposing to the active popu-194 lation during disease pandemic. The exposure may cause potential 'exposure risk' (Lai et al., 195 2020) of coronavirus contact or disease infection. Since real data on the behaviour changes of 196 mask-wearing and social distancing are unavailable, the actual exposure risk is challenging to 197 measure. In this study, it has been assumed that a region with higher heat offers higher poten-198 tial exposure risk for civic activities. The exposure risks in space or time mentioned in the 199 following text are all under such a definition. 200 Expressions of daily mobility and heat at a different spatial scale are given as follows: 201 Car park-level mobility m , in Eq. (1) is denoted as the cumulative daily number of total 202 changes in car park available lots: 203 where is the index of residential car parks; is the index of days; is the index of minutes 204 (00:00 to 24:00) in a day; , , denotes the minutely cumulative counts of the changes of car 205 park available lots associating to the car park on day . The definition of mobility can be extended to define the departing/arriving mobility by count-233 ing the cumulative increasing/decreasing number of 'available lots' separately. 234 Car park-level heat h , in Eq. (4) is expressed as the division of the car park-level mobility 235 to the number of total lots of a specific car park: 236 where is the total car park lots of car park . After multiple experiments, outliers of heat 237 values have been observed most likely for those car parks with fewer than 30 total lots. Such 238 car parks are filtered in this study. 239 Region-level heat H , in Eq. (5) is expressed as the division of the region-level mobility to 240 the number of total lots of a specific region: 241 Although the region-level heat could be a robust heat measure by eliminating the interference 242 from the extreme values related to individual car parks, the regions with very few car parks 243 should be filtered for a fair comparison among different regions. This is further illustrated in 244 Section 3.6. 245 In Singapore, to figure out the source of infected cases, the COVID-19 cases are counted by 247 four categories: imported cases from foreign countries (Source #1), community cases from 248 Singapore residents and permanent residents (Source #2), cases from work permit holders 249 who are not living in workers' dorms (Source #3) and cases from work permit holders who 250 are living in workers' dorms (Source #4) (MOH, 2020). This study focuses on the "exposure-251 related cases", including cases in Source #2, cases in Source #3 and some early-time cases in 252 Source #4 before the isolation of this source group. The imported cases and those cases of 253 confined areas are out of the explanation scope of the defined exposure in this study. 254 individual persons. In a macro view, the lag period between population mobility and infected 267 cases should be treated as a variable rather than calculated by patient statistics and reporting 268 dates directly. Such a variable can be optimally determined by curve fitting results. By as-269 suming a six-day lag period as an example, the corresponding matching is shown in Figure 2 . For avoiding overfitting, the degree of the polynomial function is restricted to be less than or 279 equal to 3. 280 of pollutants/emissions. The emission factor applied in this study is based on the passenger 291 transport report (CE Delft, 2020). The well to wheel (i.e. well to tank and tank to wheel in 292 Table 1) The assumptions of the calculation specifically for the case of Singapore include that (a) the 295 main transportation fuel is petrol (CSD, 2020), (b) the car occupancy rate is 1.7 passengers 296 (DBS, 2020), (c) the average distance travelled using car before the circuit breaker is 18.59 297 km (Numbeo, 2020) , and (d) 25.0 % (Scenario 1) to 40.0 % (Scenario 2) reduction in average 298 distance travelled are considered during the circuit breaker (Unacast, 2020) . 299 Bayesian statistical models have advantages to take into account i) uncertainties in data, ii) 303 missing data and iii) uncertainties in parameter estimation (Cressie and Wikle, 2015) . For un-304 derstanding the spatial-temporal relationship better and addressing uncertainties in individual 305 days, an extended Bayesian spatial-temporal model with Poisson regression in Eqs. (8-11) based on the basic spatial-temporal model (Blangiardo and Cameletti, 2015) is built for the 307 spatial-temporal potential exposure risk mapping: 308 , ∼ Poisson78 9, (9) 8 = : ; + = + > + ? + @ + A , where / denotes heat in region and day , which is a continuous variable with a maximal 309 value generally less than 10 for the regional-level car parks. Let , be a rounded value of 310 ./ in region and day , where ., being an integer (order of magnitude, e.g. 10, 100 and 311 1,000), is a coefficient to ensure that , has enough identifiability for different regions. After 312 experiments, it is valid to set . to be greater than or equal to 100. The . is set as 1,000 for the 313 The total mobility is much larger than that during the circuit breaker (Figures 3d, 3e and 3f) . 335 Car mobility in Figure 3e decreases significantly. Compared to that on 06 April 2020, the re-336 duction rate of the nation-level mobility elevates from 13.4 % (07 April 2020) to 36.4 % (12 337 April 2020) ( Figure S2 ). From 12 April to 28 April 2020, the reduction rate maintains at 338 around 30.0 % ( Figure S2 ). The calculated rates are in line with the results in the same period 339 based on the mobility trends reports (Apple, 2020) . During the circuit breaker, the morning 340 peaks on working days almost disappear, which are dominated slightly by the noon peaks. 341 More details about the departing and arriving car mobility are presented in Figure S3 3.2. The spatial-temporal trend of mobility 351 Figure 4 shows the spatial-temporal trend of region-level mobility. The results cover around 352 three-week before and three-week after the circuit breaker policy. Before 07 April 2020, the 353 region-level mobility is relatively stable. Before the circuit breaker, spatial mobility on 03 354 April and 06 April 2020 are elevated compared to other days. This can be explained as the 355 circuit breaker policy was declared at 4 pm on 03 April 2020. Mobility was therefore elevated on the last two working days (03 April and 06 April 2020). The region-level mobility changes 357 significantly during the first week of the circuit breaker. Compared to that on 06 April 2020, 358 mobility in nearly all regions in this week is reduced. This is in line with the reduction rate of 359 nation-level mobility in Figure S2 . The spatial-temporal mobility is relatively stable in the 360 following two weeks. The variation is significant in the growth-trend of exposure-related cases before and after the 371 dividing line in Figure 5a . When data from two phases are merged, it is difficult to fit the 372 curve via a smoothing function. By testing in the time duration before the circuit breaker, Ta-373 ble 2 compares the sum of squared residuals (SSR) under different lag periods of the cumula-374 tive mobility ranging from 0 to 7. The best-fitting result with the smallest SSR, i.e. 24,083, 375 occurs when the lag period equals 6. The optimal six-day lag period (Table 2 ) and the begin-376 tion of Phase I (from 21 March to 12 April 2020) and Phase II (from 13 April to 04 May 2020) 378 in Figure 5b . Three-order polynomial curves fit both phases well. The fitted curve in Phase I 379 has a convex trend with a more and more steep slope near 12 April 2020. The fitted curve in 380 Phase II has a concave trend with a more and more relaxed slope near 04 May 2020. The bet-381 ter phase recognition and phase division help to understand the disease outbreak trend and the 382 policy effects. More discussion is provided in Section 4. The above evidence demonstrates 383 the potential relationship between the cumulative lagged mobility and cumulative cases. (Table S2 ). The probability distributions of mobility and heat have a similar longtail 437 characteristic, although the longtail of heat is more intuitive than that of mobility. The inter-438 section of both longtail parts corresponds to these hotspots in Figure 7a Table 3 shows the model selection results based on the minimal deviance 487 information criterion (DIC) (Spiegelhalter et al., 2002) , which is introduced in Section 2.2.4. 488 The model with the Type II interaction and a first-order random walk produces the smallest 489 DIC value (i.e. 13,893), which is identified as the best model among the six models. 490 Based on the best model, Figure 10 shows the scaled computational results of temporal expo-494 sure risks and spatial exposure risks. The temporal exposure risks are the summation of base-495 line temporal risks and spatial-temporal interacted risks. In Figure 10a pared to the day before the circuit breaker. Compared to the peak value, the maximal reduc-500 tion rate of potential exposure risk reaches 37.6 %. The spatial exposure risks include intrin-501 sic spatial risks and spatial-temporal interacted risks. Figure 10b shows the scaled spatial ex-502 posure risks for a total of 27 coloured regions. Figure 10b offers an intuitive impression on 503 the potential exposure risks of individual regions and their relative hazards. Such a visualisa-504 tion could be helpful for decision-makers, especially when resources are limited to disease 505 mitigation. The areas with top 5 or top 10 exposure risks can be easily identified. As a 506 supplement to decision support, Figure 10c shows the six-level division results of scaled spa-507 tial risks from Figure 10b . 508 (Liu, 2020) . In 549 contrast, the average domestic waste amount in Singapore is elevated by about 3 % during the 550 circuit breaker (Low and Koh, 2020) . Worldwide air emissions have been significantly re-551 duced during lockdown periods. Evidence can be found from cases in China (Bao and Zhang, 552 2020) and Brazil (Dantas et al., 2020) . The calculation in this study indicates that a) the aver-553 age electricity consumption during the circuit breaker has a 6.7 % reduction in Singapore, and 554 b) the transportation-related average air emissions in Singapore are reduced by 44.3 % (Fig-555 ure 6a) and 55.4 % (Figure 6b ) under two scenarios. Greenhouse gas emissions and air pollu-556 tants have to be assessed simultaneously (Fan et al., 2018) across different sectors, not lim-557 ited to transportation. This is important to prevent the shifting of environmental footprints 558 and for a conclusive picture, as the pandemic could have changed the economic and social 559 structure considerately. Although the lasting impact on environmental sustainability still re-560 quires a systematic assessment, economic sustainability is cautiously thought to be the major 561 issue to maintain urban sustainability. During each disease pandemic, public health should be has been severely threatened. Fernandes (2020) estimated that GDP growth ranges from −3.5 % 565 to −6 % in different countries if the shutdown of economic activity lasts for 1.5 months. Ni-566 cola et al. (2020) reviewed the socio-economic impacts of COVID-19 on different sectors, 567 including agriculture, petroleum, manufacturing, education, finance, healthcare, tourism, 568 sports and food. Economic sustainability and its recovery are highly related to lockdown exit 569 strategies, as discussed in Section 4.3. break. Global economic recession due to disease pandemic leads to company bankruptcy and 583 unemployment, which might cause a series of 'butterfly effects' for the whole society, includ-584 ing the adverse effects on disease prevention and control. After implementing more than one-585 month circuit breaker, the disease surge in the community of Singapore has shown an easing 586 trend (Figure 5b) , without considering the cases of confined areas. What is the next step in 587 the near or far away future? Regarding lockdown exit strategies, other than an overall lock-588 down exit strategy and the responsible lockdown exit strategy (Gilbert et al., 2020) (e.g. sys-589 tematic tests, contact tracing and priority rework of people with a low-risk profile), the flexi-590 ble local circuit breaker strategy and the precise management measures might be beneficial 591 for the whole society. In the following, we extend the discussion and offer several sugges-592 tions from the perspective of spatial-temporal potential exposure risks. 593 Germany, where the infection rate is rising just days after Germany eased the nationwide 597 lockdown (BBC, 2020). Conservatively, a flexible local circuit breaker based on big urban 598 data analytics may be an alternative way of lifting lockdown. The spatial exposure-risk distri-599 bution in Figures 9b and 9c can be used to identify crucial regions with high potential expo-600 sure risk for maintaining local circuit breaker. Notably, the region with the highest mobility 601 in Figure 4 is not necessarily the riskiest region measured by heat in Figure 8 It has been already a suitable time to look forward to the post-pandemic period, and it has to 650 be maintained a foresight perspective for the future. For example, what has been seen in Eu-651 rope, the population is still uneasy about using public transport and is predominantly moving 652 to cars. Such social consciousness is being cultivated by the COVID-19 epidemic and even 653 more by the pandemic. What has been a high probability is that urban sustainability could be 654 influenced profoundly, and air emissions might increase worldwide compared to the regular 655 time before COVID-19 pandemic. During and after the disease outbreak, timely decision-656 making and consciousness training deserve more attention and especially planning, even if it 657 may be needed using some likely scenarios as, e.g. the second and following waves. 658 suggested that car park availability data can be regarded as a good proxy for the movement of 662 people in the absence of better data. The results and applicability of the method proposed in 663 this study are still valid despite the limitation. This is the case, especially when the change in 664 car mobility behaviour is a relative concept. 665 Second, although the driving mobility has been observed as a good representation of popula-666 tion mobility in terms of walking, driving and transit ( Figure S1 ), macro population mobility 667 is just a component of the entire daily mobilities in society. The higher heat at some specific 668 car parks and even at some specific regions cannot totally represent micro human interactions. 669 The other micro factors, such as the destination (e.g. crowded) and the person who has met 670 with, are having a considerate impact on the infection risk. However, it is an open and chal-671 lenging task to measure the entire daily mobilities at both macro and micro levels. It needs to 672 integrate more subsystems for such a task, as stated in the future work agenda. 673 The strength of the proposed method is that there is a massive high-resolution spatial-674 temporal data with usually every-minute updating available for big urban data analytics and 675 modelling before and during the circuit breaker. Buckee et al. (2020) appealed to the society 676 that aggregated mobility data with real-time information have been urgently needed to fight 677 the COVID-19 pandemic. This study provides a novel perspective and puts a small step for-678 ward on this topic. 679 This study has provided a novel perspective to analyse the spatial-temporal potential expo-681 sure risk of COVID-19 by capturing human behaviours based on high-resolution data of car 682 park availability. A testing ground, Singapore, which is threatened by a second coronavirus 683 wave from the beginning of April 2020, has illustrated the analytical procedure and demon-684 strated its applicability. For a nation or an area with available spatial-temporal data on mo-685 bility, the proposed method offers a possibility for precise urban management (i.e. the hotspot 686 identification and temporal crowd division) and timely decision-making related to COVID-19 687 mitigation. Regarding urban sustainability, although there is an apparent improvement in en-688 vironmental performance, arises from the lockdown measures, e.g. the pollution reduction by 689 transportation and industrial sector, the long-term impact of environmental sustainability re-690 quires to be assessed further. The post effect, especially dealing with medical waste disposal the lockdown or circuit breaker period, the main challenges lie in economic sustainability and 693 its recovery, as the mitigation measure makes many social contacts and production activities 694 pause citywide. The spatial-temporal potential exposure risk analytics in this study offers in-695 telligent decision support to plan a flexible local circuit breaker strategy, by which the deci-696 sion-makers might be guided to gradually reopen partial regions of the city or country for 697 economic recovery. 698 The quantitative results derived from the Singapore case, being obviously of crucial im-700 portance, are summarised as follows: 701 a) The reduction rate of mobility reaches 36.4 % in the first week during the circuit breaker. 702 It maintains at around 30.0 % in the following two weeks. Compared to regular times, the 703 morning peaks on working days during the circuit breaker almost disappear, which are 704 dominated slightly by the noon peaks. 705 b) Three-order polynomial (i.e. sub-exponential) functions well fit the curves between cu-706 mulative lagged mobility and cumulative cases (exposure-related cases) for both two 707 phases. The six-day lag setting produces the smallest sum of squared residuals than other 708 lag settings. 709 c) The 16 th day after implementing the circuit breaker policy is observed as the turning point 710 for the exposure-related COVID-19 cases, by which the cumulative exposure-related cas-711 es tend to present a gradually decreasing trend. 712 d) The longtail characteristic is observed for the probability distributions of both nationwide 713 mobility (skewness = 0.99) and nationwide heat (skewness = 3.86), indicating that, re-714 garding the hotspot identification, decision-makers should focus on the intersection of 715 both longtail parts, rather than those active car parks with high heat only. 716 e) The transportation-related average air emissions are estimated to be reduced by 44.3 % 717 (Scenario 1) to 55.4 % (Scenario 2) during the circuit breaker, indicating that the circuit 718 breaker not only keeps residents safe but also contributes to environmental health from 719 the transportation pollution perspective. This would be more interesting by simultaneous-720 ly considering that the disease pandemic triggers some companies, e.g. Twitter, to let 721 some employees work from home 'forever' if they choose (Fung, 2020). 722 g) The maximal reduction rate of potential exposure risk reaches 37.6 % by comparing with 726 its peak value. Fluctuations and uncertainties along the time horizon have been observed 727 for the heat and potential exposure risks, implying the spatial-temporal interactions 728 among different regions. 729 At least two research directions deserve investigations in future. 731 It is worthwhile to integrate more subsystems under big urban data, e.g. transportation, 732 environment, electricity, express delivery and social media, into a comprehensive 733 platform to guide decision-making better in a system-of-systems manner under the 734 premise of privacy protection. 735 (ii) Since the exposure risk defined in this study has a 'relative' and 'potential' concept, 736 there is not necessarily a strong causality between spatial potential exposure risks and 737 regional infection rates. While the possible relationship between them still deserves to 738 be investigated with regional infection data available in future. Managing infectious medical waste during the COVID-19 759 pandemic Does lockdown reduce air pollution? Evidence from 44 cities in 762 northern China Coronavirus: Germany infection rate rises as lockdown eases Singapore's urban data affirms the compliance with the Circuit Breaker 766 measures Spatial and spatio-temporal Bayesian models with R Gren-771 fell Aggregated mobility data could help fight COVID-774 19 COVID-19 outbreak: Migration, effects on society, global 778 environment and prevention Lockdown for CoViD-2019 in Milan: What are the effects on air quality? Sci-781 ence of The Total Environment Statistics for spatio-temporal data Motor vehicle population by type of fuel used We need to be alert': Scientists fear second coronavirus wave as China's 788 lockdowns ease The impact of COVID-19 partial lockdown on the air quality of the city of Rio de Janeiro Reinventing the wheel: the future of automotive Towards sustainable urban transport in Singapore: Policy instruments and 797 mobility trends Population trends 2019. Department of Statistics (DOS) Historical electricity system demand for every half hour period (MW) A review on air emissions assessment: 806 Transportation Minimising energy consumption 808 and environmental burden of freight transport using a novel graphical decision-809 making tool Impact of non-pharmaceutical interventions (NPIs) to reduce COVID-19 mor-812 tality and healthcare demand. COVID-19 Reports, Faculty of Medicine, Imperial Col-813 lege Economic Effects of Coronavirus Outbreak (COVID-19) on the World 815 Air pollution clears in some cities in South-east Asia during lockdowns, 817 but not all, study finds. The Straits Times Global shutdown: Visualising commuter activity in the world's cities 824 Min-894 istry of Trade and Industry (MTI) Releases/2020/05/MTI-Downgrades-2020-GDP-Growth-Forecast-to--7_0-to--4_0 Public health recommendations for community-related exposure Neighbourhood alcohol environment and in-903 jury risk: A spatial analysis of pedestrian injury in Baltimore City The socio-economic implications of the coronavirus and COVID-19 pandemic: 907 A review Traffic in Singapore Spatio-temporal analysis of association between in-911 cidence of malaria and environmental predictors of malaria transmission in Nigeria Investigation of three clusters of COVID-19 in Singapore: implications for surveil-915 lance and response measures Real-time forecasts of the COVID-19 epidemic in China from Infectious Disease Modelling Approximate Bayesian inference for latent Gaussian 926 models by using integrated nested Laplace approximations Innovations in the use of data facilitating insurance as a 929 resilience mechanism for coastal flood risk Effectiveness of the measures to flat-932 ten the epidemic curve of COVID-19. The case of Spain. Science of The Total Envi-933 ronment Realtime Singapore traffic watch Transmission potential and severi-937 ty of COVID-19 in South Korea Geospatial analysis of building 940 structures in megacity Dhaka: The use of spatial statistics for promoting data-driven 941 decision-making Bayesian measures 943 of model complexity and fit Singapore economy could be headed for its worst-ever contraction this 946 year. The Straits Times Real-time 953 monitoring the transmission potential of COVID-19 in Singapore Covid-19 Social Distancing Scoreboard -Unacast Record global carbon dioxide concentrations despite COVID-19 cri-959 sis Using social and behavioural science to support COVID-19 pandemic response. Na-963 ture Human Behavior From mitigation to containment of the COVID-19 pan-965 demic: Putting the SARS-CoV-2 genie back in the bottle Response to COVID-19 in Taiwan: Big data ana-968 lytics, new technology, and proactive testing Coronavirus disease (COVID-2019) situation reports. World Health Organiza-970 tion (WHO) Im-973 proving risk management by using the spatial interaction relationship of heavy metals 974 and PAHs in urban soil Spatial-temporal economic analysis of 976 modern sustainable sanitation in rural China: Resource-oriented system Evaluating temporal factors in combined interventions of workforce shift 980 and school closure for mitigating the spread of influenza Spatiotemporal analysis of influenza in China, Declaration of interests ☒ 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. ☐The authors declare the following financial interests/personal relationships which may be considered as potential competing interests The authors declare no competing interests.