key: cord-0826022-0ogqbeig
authors: Yao, Wenbin; Yu, Jinqiang; Yang, Ying; Chen, Nuo; Jin, Sheng; Hu, Youwei; Bai, Congcong
title: Understanding travel behavior adjustment under COVID-19
date: 2022-05-24
journal: nan
DOI: 10.1016/j.commtr.2022.100068
sha: 85ca069485d62466ddaf40958a521629800dbbe6
doc_id: 826022
cord_uid: 0ogqbeig

The outbreak and spreading of the COVID-19 pandemic have had a significant impact on transportation system. By analyzing the impact of the pandemic on the transportation system, the impact of the pandemic on the social economy can be reflected to a certain extent, and the effect of anti-pandemic policy implementation can also be evaluated. In addition, the analysis results are expected to provide support for policy optimization. Currently, most of the relevant studies analyze the impact of the pandemic on the overall transportation system from the macro perspective, while few studies quantitatively analyze the impact of the pandemic on individual spatiotemporal travel behavior. Based on the license plate recognition (LPR) data, this paper analyzes the spatiotemporal travel patterns of travelers in each stage of the pandemic progress, quantifies the change of travelers' spatiotemporal behaviors, and analyzes the adjustment of travelers' behaviors under the influence of the pandemic. There are three different behavior adjustment strategies under the influence of the pandemic, and the behavior adjustment is related to the individual's past travel habits. The paper quantitatively assesses the impact of the COVID-19 pandemic on individual travel behavior. And the method proposed in this paper can be used to quantitatively assess the impact of any long-term emergency on individual micro travel behavior.

1) The category of POI is reclassified to better reflect the land use type. 1 2) Calculate the frequency density (FD) and category ratio (CR) of the reclassified POI data, FD and CR are 2 defined by Eqs. (1) and (2), respectively. 3

3) Use four CRs as a feature vector (CR1, CR2, CR3, CR4) and employ the k-means clustering algorithm to 4 determine the land usage type within the buffer area of each LPR detector in Yiwu. of various categories of POI data; blue denotes residential area and life services, red stands for commercial area and 2 companies, green represents entertainment area, and yellow indicates public services. (c) Distributions of different 3 land usage types; blue, red, and green indicate the first, second, and third types, respectively. 4

Since the COVID-19 outbreak, Zhejiang provincial government has actively responded to the call of the country 6 and put forward a variety of response policies. For instance, on January 23, the Wuhan government announced the 7 closure of the city, and Zhejiang provincial government also announced that it had entered a first-level response state 8 (Chen et al., 2020b; Mei, 2020) . On February 15th, the operation of taxis and online car-hailing has recovered in 9

Yiwu city, Zhejiang Province. On February 21, urban traffic operations resumed, and certain important business 10 centers in the city opened. In addition, the resumption of work and production also started. On March 2, the 11 emergency response status of Zhejiang Province dropped to the second level. On April 13, Zhejiang provincial 12 government announced that middle and high school students in various regions of the province would return to school 13 (Mei, 2020) . Combined with the evolution of pandemic response policies, the pandemic progress can be divided into 14 three stages (Zheng et al., 2020) , and the change of travel behavior in these three stages is analyzed in this study. The 15 first stage is the control stage before the COVID-19 outbreak (hereinafter referred to as the first stage), and this stage 16 is from December 16, 2019, to December 27, 2019. The second stage is the initial stage of resumption of work and 17 production (hereinafter referred to as the second stage), and it includes the period from February 10, 2020, to February 18 21, 2020 . The third stage is the post-COVID-19 era (hereinafter referred to as the third stage), and it lasts from June 19 8 to June 19, 2020. This study analyzes and models the travel behavior based on the workdays' LPR data of the three 20 stages. 21 4 Travel behavior adjustment under COVID-19 22 Since December 2019, the COVID-19 pandemic has been experiencing rapid growth in China first and then has 23 been gradually controlled. In this process, the control measures announced by the Chinese government have played 24 an important role. Affected by the pandemic and the corresponding control measures, travel behavior of motor 25 vehicles has changed to a certain extent. This section will present the changes in a motor vehicle's travel behavior at 26 different stages of the COVID-19 pandemic. 27

Travelers typically travel many times a day, and by linking multiple trips, a trip chain of travelers can be formed. 1

By observing the trip chains of travelers over a certain period of time, frequent travel patterns of travelers can be 2 determined. A frequent travel pattern represents a generalization of a traveler's regular travels, which can reflect the 3 traveler's travel habits in a period of time. 4

This study uses the LPR data as a data source; however, the LPR data cannot show where exactly a traveler is 5 at each moment but only provides the spatiotemporal point information captured by the LPR detectors in part of the 6 traveler's trip chain. Therefore, the trip chain is defined as a, a = {P1, P2, P3, …, PK}, where Pi is the information on 7 a traveler obtained at the ith detected point. The specific form of Pi is [ti, xi, yi], where ti, xi, and yi denote the time, 8 longitude, and latitude of a traveler's i-th detected point, respectively; Pi is defined as an item, a is defined as a 9 sequence, and several items without sequence constitute an item set (Saraf et al., 2015) . For commuters, P1 is usually 10 near home, as well as PK, and they are likely to be located at the same intersection. 11

This study aims to mine frequent travel patterns from the travelers' daily trip chains. Frequent travel patterns 12 constitute substrings of a trip chain. To describe frequent travel patterns more clearly and lay a foundation for 13 subsequent frequent travel pattern mining, the following terms are defined (Shou and Di, 2018). 14 Definition 3.1 (Subsequence and super sequence). If for each item in a sequence a', there is the same item in 15 sequence a, and these items are in the same order in two sequences, then a' is called the subsequence of a, and a is 16 called the super sequence of a'. Since the specific form of an item Pi is [ti, xi, yi], if two items have the same time, 17 longitude, and latitude, then they are regarded as the same item. It is easy to prove that even the same traveler can 18 hardly have the same item on different days, which can lead to poor results in the subsequent frequent pattern mining. 19

Therefore, in this study, Eq. (3) is used to judge whether items Pi and Pj are the same. 20 daily sequence a constitutes the sequence database A, and A = {a 1 , a 2 , … , a N }, where N represents the travel days 2 of the traveler. Support is used to measure the frequency of sequence a, which is defined as a ratio of the number of 3 days of the occurrence of sequence a that is a subsequence in the traveler's sequence database to the total number of 4 sequences in the sequence database. 5 it can be inferred that the traveler is likely to be a commuter with a short commute distance. Further, P1 denotes a 21 detected point on the way from home to work, and P2 represents a detected point on the way back home from work. First, the detected point information of a traveler is arranged in chronological order to obtain the traveler trip 3 chain, which represents a sequence. After that, the daily trip chain in the first stage is mined to form a sequence 4 database, which is then used to mine frequent travel patterns by the Prefix-Span algorithm (Han et al., 2001) . The 5 enumeration method can be employed to obtain frequent travel patterns from the sequence database, but this simple 6 and rough method often consumes too much time and thus is not applicable to large data. The Prefix-Span algorithm 7 has been designed to reduce the complexity of frequent pattern mining. See the Appendix for details of how to use 8

Prefix-Span algorithm to mine frequent travel pattern. 9

Although the Prefix-Span algorithm can greatly reduce the complexity of the algorithm that mines frequent 10 travel patterns compared with the enumeration method, in some cases, the results cannot be obtained in an acceptable 11 time. In this study, there are 10-day LPR data in each stage, and minsup is set to 0.6 (Shou and Di, 2018; Saraf et al., 12 2015) . Considering a typical commuter, it indicates that the commuter drives from home to work at the morning peak 13 and returns home at the evening peak; the LPR detectors are densely arranged along the way. The commuter travels 14 every day on weekdays, and the travel trajectory on each day is the same; that is, the spatiotemporal information of 15 the corresponding detected points on each day is the same. The commuter is detected eight times from home to the 16 company in the morning peak and eight times from the company to home in the evening peak; A = {a 1 , a 2 , …, a 10 }, 17 a 1 = a 2 = … = a 10 , where a 1 = {P1, P2, …, P16}. In this case, the total number of frequent travel patterns of the 18 commuter can be calculated as 16 1 + 16 2 + + 16 16 = 65,535. It is worth noting that in this case, it has been assumed 19 J o u r n a l P r e -p r o o f that the spatiotemporal detected points constituting the trip chain of every day are the same; that is, a commuter pass 1 through the same location at the same time every day. However, a commuter is more likely to pass through a similar 2 place at a similar time, which will make the detected points be judged as the same point by Eq. (3) . In this case, the 3 number of frequent travel patterns will be a certain number times that of 65,535. For travelers with a longer trip chain 4 than that mentioned above, the number of frequent travel patterns increases explosively with the increase in the 5 detected point number. 6

To solve the problem that the Prefix-Span algorithm cannot obtain results within an acceptable time for travelers 7 with a too-long trip chain, sequence compression is used for such travelers. The following criterion is adopted to 8 judge whether a trip chain is too long; if the trip chain meets the following criterion, the sequence compression 9 method will be used. 10

1) The trip chain length of a traveler is longer than or equal to 10 in more than or equal to (minsup * 10) days. 11

2) If the first criterion is not met, then it is still needed to judge whether the average daily trip chain length is 12 longer than or equal to 10. 13 The sequence compression algorithm is given in Algorithm 1. 14 15

Input: a sequence database A={a 1 ,a 2 ,…,a 5 } Output: a sequence database A'={a 1' ,a 2' ,…,a 5' } Algorithm steps:

1. For each sequence a i in A, compress items in a i ; if several items are detected in the same hour, then keep only the one with the largest support and delete others.

Step 1, judge whether it is still needed to perform sequence compression; if so, for a i whose length is longer than or equal to 10, keep the first five items and the last four items; otherwise, return result A'. 

In Section 4.1, frequent travel patterns at various stages have been extracted. Frequent travel patterns reflect the 5 regular travel behavior of travelers at a particular stage. Affected by the pandemic, the travel behaviors of travelers 6 can change to a certain extent. Aiming to measure the change in travel behavior in different stages of the pandemic, 7

it is necessary to define the similarity of frequent travel patterns. This study adopts the similarity measurement 8 method of frequent travel patterns proposed by Shou and Di (2018) , and a few concepts need to be introduced before 9

defining the similarity of frequent travel patterns. 10 with a length of one are deleted from the frequent travel pattern set because they contain little information, and the 18 frequent travel patterns with a length longer than one are retained. As for frequent travel patterns with a length of 19 two, they might represent the travel behavior of a typical commuter, with a frequent item in the morning peak and a 1 frequent item evening peak. 2

Eq. sup port( ) sup port( )

The described similarity measurement procedure of frequent travel pattern sets can be used to measure the 8 similarity of frequent travel pattern sets between travelers in different stages of the pandemic, i.e., to explore the 9 change in the regular travel behaviors of travelers in different stages of the pandemic. 10

In addition to measuring the change in travel behavior in each stage, it is also desirable to analyze its regularity 11 quantitatively in each stage. The regularity of travel behavior of a traveler I can be calculated by Eq. 

To analyze possible behavior adjustments of travelers due to the pandemic, cluster analysis is performed. Taking  22 the first stage as a control group, vehicles with frequent travel patterns in the first stage are extracted, and a total of k-means algorithm (MacQueen, 1967) is used to cluster travelers into several different categories. The k-means 1 clustering is a method of vector quantization, which divides n observations into k clusters such that each observation 2 belongs to the cluster with the nearest mean. In practice, the k-means algorithm is very fast, and currently, it is one 3 of the fastest clustering algorithms (Pedregosa et al., 2011). However, to successfully implement the k-means 4 algorithm, it is necessary to specify the number of clusters k in advance. In this study, the number of clusters varies 5 from two to 18, and the Silhouette coefficient (Rousseeuw, 1987) is used to evaluate the effect of the clustering 6 algorithm. It is necessary to calculate the mean intra-cluster distance (a(i)) and the mean nearest-cluster distance 7 (b(i)) of a point pi first, and then the Silhouette coefficient of the point pi can be calculated by Eq. (7). 8

Finally, the Silhouette coefficients of all samples are averaged to obtain the Silhouette coefficient of the whole 10 dataset using the clustering algorithm. The range of the Silhouette coefficient is [-1, 1]; the closer the value of the 11

Silhouette coefficient is to one, the better the clustering effect is; if the value of the Silhouette is less than zero, it 12

indicates that the clustering effect is not good, and many points are classified incorrectly. The clustering performance 13 change with the number of clusters is shown in Fig. 4 . Fig. 4 , when the number of clusters is three, the clustering performance is the best, and the 1 Silhouette coefficient reaches 0.573. The average values of the five features of each cluster are given in Table 1 number of travelers, respectively. The similarity of frequent travel patterns of vehicles between the first and second 7 stages and between the first and third stages of Cluster 0 is low. This result indicates that travelers from Cluster 0 did 8 not recover their frequent travel patterns after the pandemic, even in the post-COVID-19 era. In Cluster 1, the 9 similarity of frequent travel patterns between the first and second stages is also low, but the similarity of frequent 10 travel patterns between the first and third stages is high; thus, travelers have recovered the pre-pandemic frequent 11 travel patterns in the post-COVID-19 era. The similarity of frequent travel patterns between the first and second 12 stages and between the first and third stages of Cluster 2 is high, indicating that travelers from this cluster have 13 recovered their pre-pandemic frequent travel patterns at the initial stage of resumption of work and production. 14 5 Results and discussions After extracting the first and last detected times by the above-given method, draw the cumulative probability 10 distribution curves of the first and last detected times of each stage for each cluster. The obtained distribution curves 11 of the three clusters are shown in Fig. 6a-6c . The average length of frequent travel patterns of vehicles in Cluster 1 is almost zero in the second stage, and it 1 increases to 2.898 in the third stage, which is similar to the value in the first stage. For vehicles in this cluster, the 2 first-and last-detected time distributions in the first stage are almost the same as those in the third stage. In the second 3 stage, the first detected time is delayed, and the last detected time is earlier compared to those in the first stage. About 4 80% of vehicles in the first and third stages travel before 9:00, and about 80% of the vehicles finish their trips before 5 19 :00, indicating the commuting travel patterns. As shown in Figs. 6d-6f, travelers in Cluster 0 have obvious morning and evening peak travel patterns in the 19 first stage; in the second stage, the daily travel frequency is very low; whereas, in the third stage, the morning and 20 evening peak travel patterns are recovered, but the travel frequency is still lower than in the first stage. As for Cluster 21 1, the hourly detection frequency distributions in the first and third stages are almost the same; the travel frequency 22 at the evening peak in the third stage is even greater than that in the first stage, while the second stage's travel 23 frequency throughout the day is small. Lastly, for vehicles in Cluster 2, the hourly detection frequency distributions 24

in the three stages are very similar, showing obvious morning and evening peak travel patterns. 25

Aiming to analyze the travel behavior of each cluster further, the number of vehicles in transit and the average with different license plate numbers detected by the LPR detectors on the road network during the day. Meanwhile, 1 the average travel intensity (ATI) of a cluster is calculated by Eq. (9). 2

where Ddi is the total detected times of vehicles in a cluster detected by the i-th detector during the day. 4

The number of vehicles in transit and the average travel intensity of each cluster are calculated on each day, 5 and the results are shown in Fig. 6g-6i.  6 According to Fig. 6g-6i, 

Faced with the pandemic, travelers have adjusted their travel behaviors differently. These adjustments have been 9 characterized by either homogeneity or heterogeneity. Heterogeneity indicates that, under the pandemic, specific 10 reactions of travelers have not been the same. Heterogeneity originates from the individual differences between 11 travelers, which are defined by differences in travelers' personal characteristics such as behavioral habits and external 12 characteristics such as occupations. In contrast, homogeneity indicates that reactions of travelers in a cluster on the 13 pandemic have been similar regarding certain indicators. 14 As explained above, after cluster analysis, travelers have been divided into three clusters, corresponding to three to the pandemic, several features of travel behavior are extracted, and the random forest model is used for the analysis. 2

Aiming to analyze the affecting factors of the travel behavior pattern adjusting of different groups in response 4 to the pandemic, the travel behaviors of travelers before the pandemic are analyzed. The spatiotemporal travel 5 behavior characteristics of the travelers are extracted. To describe the travelers' travel behaviors from the temporal 6 perspective, a day is divided into six time periods: before dawn (00:00-4:00), early morning (4:00-6:00), morning 7 The features of the travelers' spatiotemporal travel characteristics are extracted by the above-presented method, 18

and rug 1 , which has been defined in Section 4.2, is also used as a feature. A total of 29 features are used to describe 19 the spatiotemporal travel behaviors of travelers before the pandemic. The dependent variable is the label of three 20 clusters obtained by the clustering method, which are marked as 0, 1, 2, respectively. The labels and meanings of the 21 independent and dependent variables are given in Table 2.  22 23 Table 2 The labels and meanings of the independent and dependent variable. shown in Fig. 8 . The three graphs in the same column represent the changes in the relative probability that a vehicle 5 is predicted to belong to Clusters 0, 1, and 2 as the target feature value changes. In the dependence plots, where the 6 y-axis represents a change in the model prediction probability compared to the leftmost value, the coordinate values 7 all start from zero, and the shaded blue areas represent the confidence intervals. 8 J o u r n a l P r e -p r o o f to resume their pre-pandemic frequent travel patterns as early as possible under COVID-19. This conclusion is 1 consistent with the fact that the more regular a travel behavior of a vehicle in the first stage is, the more likely it is to 2 resume the pre-pandemic frequent travel patterns as soon as possible. 3

However, more days a vehicle travels in the afternoon (12:00-17:00), the more likely it is to resume its pre-4 pandemic frequent travel patterns in the post-COVID-19 era or never recover its frequent travel pattern, and less 5 likely it is to resume it in the second stage. Perhaps this is because travelers who like to travel in the afternoon are 6 likely not commuters. 7

Besides, the spatial information on a traveler's origin and destination of the first trip and the types of the 8 surrounding land usage also have a certain impact on the traveler's behavior adjustment, but the impact is smaller 9 than those of the temporal-perspective features. The travel behavior adjustments of travelers under the COVID-19 pandemic can be divided into three categories. 18

The first category refers to travelers whose frequent travel patterns have greatly changed from the first stage under 19 the COVID-19 pandemic; their trips reduced significantly in the second and third stages, and even in the post-20 COVID-19 era, their frequent travel patterns did not recover. The second category refers to travelers whose frequent 21 travel patterns also have significantly changed compared to those in the first stage, but in the post-COVID-19 era, 22 their frequent travel patterns have been recovered. The third category refers to travelers that resumed their frequent 23 travel patterns in the initial stage of resumption of work and production. The proportions of the first, second, and 24 third categories are 48.70%, 41.97%, and 9.32%, respectively. 25

The first category of vehicles (i.e., travelers) almost did not have frequent travel patterns in the second and third 26 stages. For a traveler of this category, the time to go out which means the duration between the first travel and the 27 last travel in a day shortens in the second and third stages. The first-category travelers have obvious morning and 28 evening peak travel patterns in the first stage, while in the second stage, their daily travel frequency is very low, and 1 morning and evening peak travel patterns are not obvious. In the third stage, the morning and evening peak travel 2 patterns are recovered, but the travel frequency is still relatively low. Compared with the first stage, the number of 3 vehicles in transit and the average travel intensity in the third stage decreased by 30.1% and 7.6%, respectively. 4

For the second-category vehicles, the travel pattern in the third stage is the same as that in the first stage, and 5 the travel intensity in the third stage even exceeds that in the first stage. About 80% of the vehicles in the first and 6 third stages start traveling before 9:00, and about 80% of the vehicles finish their travel before 19:00, indicating the 7 commuting travel patterns. The behavior adjusting patterns have a certain relationship with the pre-pandemic travel behavior of travelers. 18

A traveler who has more traveling days in the morning (6:00-12:00) and evening (17:00-21:00) and a stronger travel 19 regularity is more likely to resume his/her pre-pandemic frequent travel patterns as soon as possible. A traveler who 20 always travels in the afternoon (12:00-17:00) may not recover his/her pre-pandemic frequent travel patterns or 21 recover frequent travel patterns until post-COVID-19 era. In addition, the spatial information on a traveler's origin 22

and destination and types of the surrounding land usage also have certain impacts on the traveler's travel behavior 23 adjustment, but their impacts are smaller than those of the temporal-perspective features. 24

This paper uses the LPR data as a basis to analyze the spatiotemporal behavior adjustment and its influencing 25 factors under the COVID-19 pandemic profoundly. However, due to the data dimension limitation, the range of 26 travelers analyzed in this study is relatively small, and there is no in-depth analysis and discussion for travelers who 27 travel without using motor vehicles. In addition, the analysis of influencing factors is not comprehensive enough, and the trip's purpose, traveler's occupation, and other factors are not considered. In the future research, multi-source data 1 from mobile phone signaling, smart card, and bicycle sharing will be used to analyze the characteristics of urban 2 residents' travels in all modes, and urban planning information and questionnaire survey data will be combined to 3 expand the data dimension to conduct a more in-depth analysis of travel behaviors and their influencing factors, 4 which can help to understand the impact of COVID-19 on transportation system and travel behavior more deeply. The pseudo-code of the Prefix-Span algorithm is given as Algorithm A1. 1. Scan A|α once; find frequent items whose support values are larger than minsup; append them to α to form a frequent travel pattern α'; output α'. The license plate recognition data used in this research was provided by Yiwu City Brain, and the point of interest 2 data used in this research was crawled from the application programming interface (API) of AMAP. The code and 3 data sample of this research can be found at https://github.com/RobinYaoWenbin/Understanding-travel-behavior-4 adjustment-under-COVID- 19. 5 Declaration of competing interest 6 The authors declare that they have no known competing financial interests or personal relationships that could have 7 appeared to influence the work reported in this paper. 8

Effects of the COVID-19 lockdown on urban mobility: empirical 14 evidence from the city of Santander (Spain)

Assessing the impact of reduced travel on exportation dynamics 16 of novel coronavirus infection (COVID-19)

Appraising the impact of air transport on the environment: Lessons 18 from the COVID-19 pandemic

Mobility network 20 models of COVID-19 explain inequities and inform reopening

Modeling and interpreting the COVID-19 22 intervention strategy of China: A human mobility view

Understanding the modifiable areal unit 4 problem in dockless bike sharing usage and exploring the interactive effects of built environment factors

Spatial heterogeneity in distance decay of using bike sharing: An 7 empirical large-scale analysis in Shanghai

Borderline-SMOTE: a new over-sampling method in imbalanced data 10 sets learning

Prefixspan: Mining sequential 12 patterns efficiently by prefix-projected pattern growth

Comparison of the Flow Rate and Speed of Vehicles on a 15

Representative Road Section before and after the Implementation of Measures in Connection with COVID-19

Hub airport slot Re-allocation and subsidy policy to speed up air traffic 18 recovery amid COVID-19 pandemic---case on the Chinese airline market

Estimating and projecting air passenger 21 traffic during the COVID-19 coronavirus outbreak and its socio-economic impact

Measuring travel behavior in Houston, Texas with mobility data during the 1 2020 COVID-19 outbreak

Assessing air quality changes in large cities during COVID-19 lockdowns: The impacts of traffic-free urban 4 conditions in Almaty

Analysis of mobility data to build contact networks for

The relationship between 8 trends in COVID-19 prevalence and traffic levels in South Korea

High-resolution assessment of environmental benefits 11 of dockless bike-sharing systems based on transaction data

How did micro-mobility change in response to covid-13 19 pandemic? a case study based on spatial-temporal-semantic analytics

Incorporating Multi-source Big Geo-data to

Sense Spatial Heterogeneity Patterns in an Urban Space

Queue length estimation for signalized intersections using

License Plate Recognition data

Grouped travel time estimation in signalized arterials using point-to-21 point detectors

Policy style, consistency and the effectiveness of the policy mix in China's fight against COVID-19

Quantifying the impact of COVID-19

Determining rural areas vulnerable to illegal dumping 15 using GIS techniques. Case study

Spatiotemporal dynamics in demography-sensitive disease 17 transmission: COVID-19 spread in NY as a case study

Impact on city bus transit 19 services of the COVID-19 lockdown and return to the new Normal: The case of A Coruña

Spatially explicit models for exploring COVID-22 19 lockdown strategies

Analysis of road traffic pattern changes due to activity 24 restrictions during COVID-19 pandemic in Chennai

Prefixspan algorithm for finding sequential pattern with various 3 constraints

Spatiotemporal influence of land use and household properties on 5 automobile travel demand

Similarity analysis of frequent sequential activity pattern mining

Amplified ozone pollution in cities during the COVID-19

COVID-19: epidemiology, evolution, and cross-disciplinary perspectives

Covid-19, Lockdowns and Motor Vehicle Collisions: Empirical Evidence from Greece

Travel Time Estimation Method for Urban Road Based on Traffic 15

Trade characteristics, competition patterns 17 and COVID-19 related shock propagation in the global solar photovoltaic cell trade

Transmission and control pressure analysis of the COVID-20 19 pandemic situation using multisource spatio-temporal big data

A preliminary assessment of the impact of COVID-19 on environment -A case study 22 of China

Understanding vehicles commuting pattern based on license plate 10 recognition data

Analysis of key commuting routes based on 12 spatiotemporal trip chain

Time-aware point-of-interest recommendation

The effect of human mobility and control measures 17 on traffic safety during COVID-19 pandemic

The Fall and Rise of the Taxi Industry in the COVID-19 Pandemic: A Case 19

Available at SSRN 3674241

The authors declare that there are no conflicts of interest in this paper.We declare that we have no financial and personal relationships with other people or organizations that can inappropriately influence our work, there is no professional or other personal interest of any nature or kind in any product, service and/or company that could be construed as influencing the position presented in, or the review of the manuscript.J o u r n a l P r e -p r o o f