Deciphering Delivery Mobility: A City-Scale, Path-Reconstructed Trajectory Dataset of Instant Delivery Riders PDF Free Download
1 / 18/18
100%
CONFIDENTIAL
COPY OF SUBMISSION FOR PEER REVIEW ONLY
Tracking no: SDATA-25-03822
Deciphering Delivery Mobility: A City-Scale, Path-Reconstructed Trajectory Dataset of Instant Delivery Riders
Authors:Chengbo Zhang (Harbin Institute of Technology Shenzhen), Yonglin Li (Harbin Institute of Technology Shenzhen), and Zuopeng
Xiao (Harbin Institute of Technology Shenzhen)
Abstract:
The rapid expansion of the on-demand economy has profoundly reshaped urban mobility and logistics, yet high-resolution trajectory data
on delivery riders’ consistent movements remains scarce. Here, we present a city-scale, high-resolution spatiotemporal trajectory dataset
of on-demand instant delivery riders in Beijing. This dataset was produced through a path-reconstruction methodology applied to an open
dataset containing delivery order information. Subsequently, detailed and continuous trajectories were reconstructed by simulating cycling
routes via a major online map service to ensure they were realistically aligned. For validation, the reconstructed paths were compared
against ground-tr uth travel metrics, revealing a strong correlation with actual travel patterns. The analysis yielded Pearson correlation
coefficients of 0.92 for route distance and 0.79 for route duration. This high fidelity ensures the dataset's utility for describing delivery
riders’ mobility. This publicly available resource offers unprecedented opportunities for researchers in urban planning, transportation
studies, logistics optimization, and computational social science to investigate rider behavior, model urban freight systems, and develop
more efficient and sustainable city-wide logistics solutions.
Datasets:
Repository
Name Dataset Title Accession Number or DOI URL to data record
Private
reviewer
access
URL/code
figshare
City-Scale, Path-
Reconstructed
Trajectory Dataset of
Instant Delivery Riders
10.6084/m9.figshare.29314796.v1 https://doi.org/10.6084/m9.figshare.29314796.v1
1
Deciphering Delivery Mobility: A City-Scale,
Path-Reconstructed Trajectory Dataset of Instant Delivery
Riders
Chengbo Zhang a,b
zhangcb0027@foxmail.com
Yonglin Li a,b
23S056020@stu.hit.edu.cn
Zuopeng Xiao a,b * (Corresponding Author)
xiaozuopeng@hit.edu.cn
SIDS-C-6-201, Harbin Institute of Technology,University Town,
Nanshan District, Shenzhen, China
aHarbin Institute of Technology Shenzhen, China
bShenzhen Key Laboratory of Urban Planning and Simulation Decision,
Shenzhen, China, China
2
Abstract
The rapid expansion of the on-demand economy has profoundly reshaped urban
mobility and logistics, yet high-resolution trajectory data on delivery riders’ consistent
movements remains scarce. Here, we present a city-scale, high-resolution
spatiotemporal trajectory dataset of on-demand instant delivery riders in Beijing. This
dataset was produced through a path-reconstruction methodology applied to an open
dataset containing delivery order information. Subsequently, detailed and continuous
trajectories were reconstructed by simulating cycling routes via a major online map
service to ensure they were realistically aligned. For validation, the reconstructed
paths were compared against ground-truth travel metrics, revealing a strong
correlation with actual travel patterns. The analysis yielded Pearson correlation
coefficients of 0.92 for route distance and 0.79 for route duration. This high fidelity
ensures the dataset's utility for describing delivery riders’ mobility. This publicly
available resource offers unprecedented opportunities for researchers in urban
planning, transportation studies, logistics optimization, and computational social
science to investigate rider behavior, model urban freight systems, and develop more
efficient and sustainable city-wide logistics solutions.
Background & Summary
The rapid expansion of the instant e-commerce and on-demand delivery has
profoundly reshaped urban logistics and transportation systems, introducing a
significant new component of urban mobility: the instant delivery rider1–3. This
global workforce, estimated at over 25 million individuals4–6, powers a market
exceeding $380 billion in 20247. The sector’s explosive growth underscores its
surging impact while raising profound sustainability challenges8–11. A granular
understanding of this massive rider mobility is essential for a range of applications,
including the optimization of instant logistics, the development of sustainable
transport management policies, and the modeling of interactions between delivery
3
traffic and general urban congestion. However, empirical research is persistently
hampered by a lack of access to high-resolution, individual-level, and open trajectory
data. These datasets are typically proprietary, commercially sensitive, and present
significant privacy challenges, resulting in the reliance on data collection methods
like questionnaires or small-scale field observations in existing studies1,2,12,13. While
the increasing availability of general human mobility datasets is widening research
avenues14, a critical scarcity of high-resolution trajectory data specific to delivery
riders persists. This data gap hinders the scientific community's ability to monitor,
model, and develop sustainable management strategies for this transformative
industry. Despite some datasets containing origin-destination (OD) trips by delivery
orders it is available for some datasets containing origin-destination (OD) trips by
delivery orders 15, the rider trajectory cannot be easily constructed.
To accurately reflect instant delivery riders’ mobilities, it is necessary to
incorporate the process of operational delivery order wave, which has been rarely
seen in existing datasets. In practice, riders do not execute simple pickup-to-delivery
trips; they are often assigned multiple orders simultaneously and must make
complex routing decisions to efficiently visit multiple locations within a single tour.
Therefore, merely considering OD pairs of individual delivery order is insufficient to
reconstruct their intricate mobility patterns. The order wave (also called a bundle)
represents a complete, multi-stop trip assigned to a single rider, which can increase
delivery efficiency and improve sudden surge of supply and demand16. For example, a
simplified wave is defined by a sequence of designated tasks, including ActionStep1
(e.g., pick up Order A), ActionStep2 (e.g., pick up Order B), ActionStep3 (e.g., deliver
Order A), and ActionStep4 (e.g., deliver Order B). These action steps are all linked to
a single rider within a continuous period of work, rather than disconnected
OD-pair-level trips. A framework leveraging the wave information is significant for
reflecting real rider behavior, as it moves beyond simulating disconnected
OD-pair-level trips.
Furthermore, existing trajectory generation methods, while valuable, have
4
limitations in capturing the routing process of delivery riders. Agent-based models
and other synthetic data generators often require extensive calibration and may not
reflect real-world operational constraints17–19. Online map services offers a promising
and lightweight approach to simulate paths that conform to real road conditions.
Studies show a strong correspondence between online navigation and travelers'
decisions20–22, and these navigation services serve as indispensable tools for
platform-based delivery work. Therefore, we adopt this approach by sequentially
reconstructing a rider's delivery trajectory. By generating a cycling-optimized route
between each consecutive action step within a delivery wave, we can create a single,
continuous trajectory for the entire tour that is aligned with realistic mobility
patterns.
To this end, we present this city-scale, path-reconstructed trajectory dataset for
on-demand instant delivery riders in Beijing, China23. We generated this novel
dataset by first abstracting wave-based delivery chains from a 28-day record of
anonymized rider tasks, an open dataset from Eleme (one of China's largest delivery
operators) via the Alibaba Tianchi platform. We then applied a path-reconstruction
framework, using a high-precision online map service to generate realistic cycling
trajectories between each consecutive action in a wave. The resulting spatiotemporal
dataset, comprising 79,648 delivery waves completed by 986 riders to deliver
267,529 orders, faithfully represents the complex, multi-stop feature of on-demand
delivery work.
This dataset holds significant potential for various research directions and
applications. Multifacet tasks that this dataset could support are summarized as
follow:
Spatio-temporal mobility patterns specific to delivery riders : The dataset enables
large-scale analysis of the intricate mobility patterns specific to delivery riders.
Researchers can identify and model key delivery corridors, constraints, hotpots,
and route-choice behaviors that are distinct from general urban traffic in a
time-geography framework. This can lead to a more comprehensive
5
understanding of modern urban mobility systems.
Rider workload and labor dynamics: The trajectories provide an objective basis
for studying the labor dynamics of the gig economy. The data can be used to
quantify rider workload, analyze work scheduling strategies, and their relation to
the urban rhythm regularities. These findings can inform the design of more
equitable and effective delivery platforms and labor policies.
Sustainability issues related to instant delivery rider: This dataset provides the
critical evidence to develop data-driven policies for greener, safer, and more
efficient cities. By modeling the environmental footprint and pinpointing safety
risks, we can guide investments that harmonize the explosive growth of instant
delivery with long-term urban sustainability.
Methods
Delivery record datasets
The foundational data for this study was obtained from the "Smart Logistics:
Ele.me Rider Behavior Prediction during the COVID-19 Period" competition, a public
data science challenge hosted on the Alibaba Tianchi platform in 2020
(https://tianchi.aliyun.com/competition/entrance/231777/introduction). This
dataset contains anonymized records of delivery tasks in Beijing, structured into
multiple data sheets including wave information to represents a multi-order tour
assigned to a single rider. All released data were designed to rigorously protect
individual privacy. The competition organizers explicitly state that this data is
desensitized simulated data, meaning all personal identifiers such as real rider
names or contact information were removed or anonymized prior to its release. Our
path-reconstruction process uses only these anonymized location coordinates and
task identifiers. The final, published trajectory dataset contains no real-world rider
identities, ensuring that the privacy of all individuals is preserved in alignment with
best practices for handling mobility data.
6
Path-reconstruction framework
Our path-reconstruction framework is designed to generate a continuous,
high-resolution trajectory for each multi-stop delivery tour by waves (Fig. 1). The
core principle is to simulate the most probable path a rider would take to complete
the sequence of tasks within a given delivery tour, transforming discrete task
locations into a realistic, connected delivery route.
Fig. 1 | Work flow of data processing and delivery route generation.
The process begins with the decomposition of each delivery wave from the
source data. A wave, uniquely identified by its courier_id and wave_index,
represents the fundamental unit of a rider's tour. Typically, a delivery wave
commences at the moment of assignment and concludes upon the final delivery of
the last order within the bundled tasks.
First, for each wave, the series of "assign", "pickup", and "delivery" tasks are
chronologically sorted based on their expect_time timestamps. This sorting
establishes the precise operational sequence of locations a rider must visit. From this
ordered sequence, the framework defines a series of point-to-point routing tasks,
where the destination of task i becomes the origin for the journey to task i+1. To
enable subsequent evaluation of our generated routes, we also recorded the total
duration and grid distance for each wave, as provided in the source data, to serve as
comparable metrics.
7
With the sequence of OD pairs established for a wave, the framework then
systematically queries the Amap routing Application Programming Interface (API) to
simulate the rider's path under realistic street conditions. Each OD pair is submitted
to the API with the query explicitly configured to generate a route optimized for
cycling, ensuring the simulated path accurately reflects the operational mode of
delivery riders.
Finally, the output from the navigation service is synthesized into a standardized
trajectory format. For each path segment, the service returns a high-resolution
polyline, and timestamps are interpolated along this path based on the travel time
estimated by the API. This routing and synthesis process is iteratively applied to all
consecutive tasks in the wave. The resulting path segments are then concatenated in
order, forming a single, complete, and continuous spatiotemporal trajectory that
represents the rider’s entire multi-stop tour.
Data Records
The dataset is formatted in the GeoJSON format, a standard open format for
encoding geographic data structures, and is available on Figshare22. The file contains
a Feature Collection composed of two distinct feature types, which are identified by
the feature_type property: 1) Wave-based delivery routes, represented as LineString
geometries. 2) Action points, the series of discrete stops associated with a route,
represented as Point geometries.
Route Features (feature_type: 'route'): Each line feature represents a complete,
multi-stop delivery tour for a single rider. The properties for route features are
detailed in Table 1.
Table 1: Schema for Route Features
Field Name
Data Type
Description
Route_id
Integer
A unique, continuous identifier for each delivery route across the
entire dataset.
courier_id
Integer
Anonymized identifier for the delivery rider.
8
date
String
(YYYY-MM-DD)
The date on which the delivery route occurred.
no_act
Integer
The total number of actions (stops) within the route.
act_lst
String (List)
A string representation of the chronological list of action types (e.g.,
"['ASSIGN', 'PICKUP', 'DELIVERY']").
r_time_lst
String (List)
A list of real UNIX timestamps for each action, inherited from the
source data.
r_dis_lst
String (List)
A list of estimated real travel distances for each segment of the
route, inherited from the source data.
r_dur_all
Integer
The total actual duration of the route in seconds (time of last
delivery action - time of assignment time).
r_dis_all
Float
The sum of all real travel distance segments in meters.
no_nav
Integer
The number of navigation segments that constitute the route.
nav_dis
Float
The total navigation distance of the route as estimated by the map
service, in meters.
nav_dur
Integer
The total navigation duration of the route as estimated by the map
service, in seconds.
rider_lvl
Integer
The experience level of the rider.
rider_spd
Float
The rider's stated average speed.
max_load
Integer
The maximum load capacity of the rider.
wthr_grd
Integer
The grade of the weather conditions during the route.
feature_type
String
A constant value: 'route'.
Point Features (feature_type: 'action_point'): Each point feature represents a
specific task (e.g., ASSIGN, PICKUP, DELIVERY) within a delivery tour. The
properties for point features are detailed in Table 2.
Table 2: Schema for Action Point Features
Field Name
Data Type
Description
Route_id
Integer
The identifier of the route to which this point belongs. Links points
to their parent route.
act_pt_id
String
A unique 7-digit identifier for each action point, formed by
combining the 5-digit Route_id and the 2-digit act_order.
courier_id
Integer
Anonymized identifier for the delivery rider.
date
String
(YYYY-MM-DD)
The date on which the action occurred.
act_time
Integer
The UNIX timestamp for when the action occurred.
act_order
Integer
The chronological sequence number (starting from 1) of the action
within its route.
9
action_type
String
The type of action (e.g., 'ASSIGN', 'PICKUP', 'DELIVERY').
feature_type
String
A constant value: 'action_point'.
Technical Validation
To assess the technical quality of the reconstructed trajectory dataset to reflect
the realistic rider routing, we performed a series of validation analyses. These
validations were designed to (1. quantify the correspondence between our
reconstructed route metrics and the recorded real metrics, and (2. Verify that the
dataset reflects known, real-world urban delivery dynamics.
Correlation with route-level ground-truth records
A primary validation of our path-reconstruction framework involved comparing
the metrics of our synthesized trajectories against the ground-truth spatiotemporal
data recorded for each delivery wave. Specifically, Fig. 2 shows the correlation
between two pairs of variables: 1) the total recorded route distance (r_dis_all) versus
the API-based reconstructed route distance (nav_dis), and 2) the total recorded wave
duration (r_dur_all) versus the API-based reconstructed route duration (nav_dur).
Fig. 2 | Validation of reconstructed trajectory metrics against ground-truth data.
Scatter plots comparing reconstructed (online map API-based) and ground-truth
metrics for each delivery wave on a log-log scale. The solid black line is linear
10
regression.
As illustrated in Fig. 2, the validation reveals a strong positive linear relationship
between our reconstructed data and the ground-truth metrics. The high correlation
coefficients validate the fidelity of our path-reconstruction framework. Specifically,
the correlation for travel distance is exceptionally strong (Pearson's r = 0.92),
indicating that the navigation API provides a robust estimate of the total spatial
routing distance for a multi-stop tour.
The correlation for travel time is also high (Pearson's r = 0.79), confirming that
the sequence of API-generated segments also effectively captures the overall duration
of a delivery wave. The slightly weaker correlation for duration can be clearly
attributed to factors not included in our constant-movement model. Specifically, our
framework does not account for rider waiting time at pickup locations (e.g., waiting
for a meal to be prepared), which can introduce a reduced bias of estimated travel
time to the real-world duration.
Spatiotemporal patterns of delivery trajectories
To better understand the spatiotemporal distribution of the dataset, we
visualized the trajectories from a specific day, February 15th (Fig. 3). The overall
spatial pattern, shown in the main panel of Fig. 3, reveals dense trajectories cover
within Beijing's central urban area, with paths showing a high degree of alignment
with the city's road network structure. Fig. 3 also highlights four individual delivery
waves to illustrate the multi-stop nature of these tours. Each example route connects
the sequence of ASSIGN, PICKUP, and DELIVERY actions, and their corresponding
timelines visualize the distinct timing and duration of each step. Furthermore, to
analyze spatial concentrations, we generated kernel density plots for all pickup and
delivery locations. These plots show clear hotspots concentrated in the urban core,
providing a nuanced illustration of instant delivery adoption patterns across Beijing.
11
Fig. 3 | Spatiotemporal patterns of on-demand delivery in Beijing. Trajectories
from a single day show dense coverage in the urban core, aligned with the street
network (main panel). Insets (1–4) detail individual multi-stop delivery waves,
showing the sequence of delivery actions, with corresponding timelines illustrating
the duration of each step. Kernel density plots (bottom left) reveal the spatial
hotspots for pickup and delivery activities.
Furthermore, we validated the dataset by examining its internal temporal
patterns. First, we analyzed the daily distribution of order pickups across the 28-day
period with differentiate with weekdays and weekends (Fig. 4). The data reveals
both daily fluctuations and a general upward trend in delivery activity throughout the
month. Second, we analyzed the average hourly distribution of the three main action
types, separating weekdays from weekends (Fig. 5). The results for both periods
reveal a distinct bimodal distribution, with pronounced peaks corresponding to the
characteristic lunch (11:00–13:00) and dinner (17:00–19:00) rushes of on-demand
food delivery. Within these rushes, the temporal ordering follows a logical sequence:
12
ASSIGN actions peak with PICKUP actions, and followed by a slightly delayed
DELIVERY peak. Notably, the fluctuations in activity are more moderate compared
to the sharper, more concentrated peaks observed on weekdays.
Fig. 4 | Daily distribution of pickup actions with weekend days are highlighted in
orange. The plot illustrates daily fluctuations and a general increase in delivery order
over the month in the dataset.
Fig. 5 | Average hourly distribution of delivery actions on weekdays versus weekends.
The plots show the average number of ASSIGN, PICKUP, and DELIVERY actions per
hour, aggregated over 20 weekdays (left) and 8 weekend days (right).
In addition, we analyzed the hourly distribution of rider capacity and workload,
separating weekdays from weekends to identify distinct temporal patterns. Fig. 6
shows the average number of active riders throughout the day. On both weekdays
13
and weekends, rider availability forms clear peaks in labor participation during lunch
and dinner hours. The comparison between the panels reveals differences in
workforce availability and scheduling patterns between weekdays and weekends,
Fig. 6 | Average hourly distribution of active riders. The average number of unique
riders actively working per hour, aggregated across 20 weekdays (blue, left) and 8
weekend days (red, right). Both plots show a sustained high level of rider supply
throughout daytime and evening hours.
Fig. 7 shows the hourly distribution of orders deliverd per wave by a rider, shows
the average order in a bundle is x and the peak accuring at 11 am at week days is 5
orders per wave. On weekends, the fluction of workload is also moderated. The
difference further supporting the dataset's utility for studying the dynamics of the
urban gig economy
Fig. 7 | Average orders per active rider per hour. This figure illustrates rider
14
workload, measured as the average number of orders handled by an active rider in a
wave per hour. The weekday panel (green, left) shows distinct peaks corresponding
to lunch and dinner. The weekend panel (yellow, right) exhibits a more sustained
plateau of high rider workload.
Usage Notes
The dataset is provided in GeoJSON format, ensuring compatibility with
standard Geographic Information System (GIS) software. For programmatic analysis,
we recommend using Python with the geopandas library to load the data as
GeoDataFrames, which preserves both the geometric LineString objects and their
associated metadata. Geometric operations, such as simplifying trajectories or
calculating path lengths, can be performed using the shapely library.
The complete dataset is publicly available under the Creative Commons
Attribution 4.0 International (CC BY 4.0) license and can be downloaded from the
Figshare repository (https://doi.org/10.6084/m9.figshare.29314796.v1). To facilitate
data exploration, the repository also includes an animated visualization of the
February 15th trajectories, created with kepler.gl (Fig. 8)
When using this dataset, it is important to consider two key technical aspects.
First, the foundational data is from February 2020, and users should account for the
potentially anomalous urban mobility patterns of that specific period in any
comparative or longitudinal analysis. Second, the scope of each trajectory is limited
to a single delivery wave; it represents the path taken during active order fulfillment
and does not include a rider's travel before the wave assignment or after the final
delivery.
15
Fig. 8 | Interactive data visualization with kepler.gl. Screenshot of the web-based,
animated visualization showing all delivery trajectories from February 15th.
Code availability
The Python code for processing original data and generating routes by Amap
API is available on GitHub (https://github.com/Nicholas0027/ODIDMobTraj).
Reference
1. Oviedo-Trespalacios, O., Rubie, E. & Haworth, N. Risky business: Comparing the riding
behaviours of food delivery and private bicycle riders. Accident Analysis & Prevention 177,
106820 (2022).
2. Aguilera, A., Dablanc, L. & Rallet, A. Digital Work and Urban Delivery: Profile, Activity and
Mobility Practices of On-Demand Food Delivery Couriers in Paris (France). Information 13,
433 (2022).
3. Krier, C., Dablanc, L., Aguiléra, A. & Louvet, N. Sharing within the gig economy: The use of
shared e-bikes by on-demand platform-based instant meal delivery workers in Paris. Case
Studies on Transport Policy 10, 2280–2289 (2022).
4. Kumar, N. How Many Uber Drivers Are There In 2025 (Stats). DemandSage
https://www.demandsage.com/uber-statistics/ (2025).
5. How Many People Use DoorDash in 2025? [New Data]. Backlinko
https://backlinko.com/doordash-users (2021).
6. Dong, C. Meituan paid $11.3 billion to millions of delivery riders in 2023, averaging less than
$0.52 per order · TechNode. TechNode
16
http://technode.com/2024/09/19/meituan-paid-11-3-billion-to-millions-of-delivery-riders-i
n-2023-averaging-less-than-0-52-per-order/ (2024).
7. Online Food Delivery Services Market | Industry Report, 2030.
https://www.grandviewresearch.com/industry-analysis/online-food-delivery-services-market
.
8. Zhong, Y. et al. Carbon emissions from urban takeaway delivery in China. npj Urban Sustain
4, 39 (2024).
9. Lord, C. et al. The sustainability of the gig economy food delivery system (Deliveroo,
UberEATS and Just-Eat): Histories and futures of rebound, lock-in and path dependency.
International Journal of Sustainable Transportation 17, 490–502 (2023).
10. Matsuyuki, M. et al. Assessment of the sustainability of online food delivery from the
perspective of CO2 emissions and transport intensity: A case study in Jakarta. Transportation
Research Interdisciplinary Perspectives 27, 101200 (2024).
11. Zhang, Y. et al. Urban food delivery services as extreme heat adaptation. Nat Cities 1–10
(2025) doi:10.1038/s44284-024-00172-z.
12. Vecchio, G., Tiznado-Aitken, I., Albornoz, C. & Tironi, M. Delivery workers and the interplay
of digital and mobility (in)justice. Digital Geography and Society 3, 100036 (2022).
13. Yu, D., Zhang, J. & Yun, G. Delivery riders’ safety and delivery efficiency in on-demand food
delivery industry: The moderating role of monitoring algorithms. Research in Transportation
Business & Management 55, 101143 (2024).
14. Yabe, T. et al. YJMob100K: City-scale and longitudinal dataset of anonymized human
mobility trajectories. Scientific Data 11, 397 (2024).
15. Qiu, Y., Ding, J., Wang, M., Hu, L. & Zhang, F. Understanding the urban life pattern of young
people from delivery data. Comput.Urban Sci. 1, 28 (2021).
16. Zhu, L. et al. Order Fulfillment Cycle Time Estimation for On-Demand Food Delivery. in
Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery &
Data Mining 2571–2580 (Association for Computing Machinery, New York, NY, USA, 2020).
doi:10.1145/3394486.3403307.
17. Zare, P., Leao, S., Gudes, O. & Pettit, C. A simple agent-based model for planning for bicycling:
Simulation of bicyclists’ movements in urban environments. Computers, Environment and
Urban Systems 108, 102059 (2024).
18. Cai, Y., Ong, G. P. & Meng, Q. Understanding bike-sharing as a commute mode in Singapore:
An agent-based simulation approach. Transportation Research Part D: Transport and
Environment 122, 103859 (2023).
19. Kashiyama, T., Pang, Y., Sekimoto, Y. & Yabe, T. Pseudo-PFLOW: Development of nationwide
synthetic open dataset for people movement based on limited travel survey and open
statistical data. Preprint at https://doi.org/10.48550/arXiv.2205.00657 (2022).
17
20. Toledo, T. et al. Intercity truck route choices incorporating toll road alternatives using
enhanced GPS data. Transportmetrica A: Transport Science 16, 654–675 (2020).
21. Bandeira, J. M. et al. Exploring multiple eco-routing guidance strategies in a commuting
corridor. International Journal of Sustainable Transportation 12, 53–65 (2018).
22. Sampaio, C., Coelho, M. C., Macedo, E. & Bandeira, J. M. Emissions based tolls – Impacts on
the total emissions of an intercity corridor. Transportation Research Part D: Transport and
Environment 101, 103093 (2021).
23. City-Scale, Path-Reconstructed Trajectory Dataset of Instant Delivery Riders. figshare
https://doi.org/10.6084/m9.figshare.29314796.v1 (2025).
