Publications

This study aimed to explore the impacts of COVID-19 on car and bus usage and their relationships with land use and land price. Large-scale trip data of car and bus usage in Daejeon, South Korea, were tested. We made a trip-chain-level data set to analyze travel behavior based on activity-based travel volumes. Hexagonal cells were used to capture geographical explanatory variables, and a mixed-effect regression model was adopted to determine the impacts of COVID-19. The modeling outcomes demonstrated behavioral differences associated with using cars and buses amid the pandemic. People responded to the pandemic by reducing their trips more intensively during the daytime and weekends. Moreover, they avoided crowded or shared spaces by reducing bus trips and trips toward commercial areas. In terms of social equity, trips of people living in wealthier areas decreased more than those of people living in lower-priced areas, especially trips by buses. The findings contribute to the previous literature by adding a fundamental reference for the different impacts of pandemics on two universal transportation modes.

Understanding different mechanisms in trip changes depending on transportation modes due to COVID-19 pandemic is the key to providing practical insights for healthy communities. This study aimed to investigate the impact of the COVID-19 pandemic on individual-level travel behavior in Daejeon Metropolitan City, South Korea. Using smart card and private vehicle records, we explored different travel behaviors exhibited while using buses and private vehicles. An individual's travel behavior was represented in trip frequency, trip regularity, and trip distance and was compared weekly for about three months, including the initial period of pandemic. A significant decrease in trip frequency during non-peak hours on weekdays and during weekends indicates that people reduced non-mandatory trips more than commuter trips. This was also verified in that, as the number of infection cases increased, trip regularity with 24-hour intervals intensified. People maintained the size of their activity boundaries but reduced their daily trip distances. The interesting point is that private vehicle usage increased for shorter trip distances while bus usage dropped regardless of the ranges of trip distances under the pandemic. The findings provide evidence of possible inequality issues in transportation during the pandemic and can help make precautionary policies for future pandemics.

This study examined how inter-city car travel behavior changed due to the COVID-19 pandemic over the three pandemic waves in South Korea. We used daily freeway traffic volume data from 2019 and 2020 to represent inter-city car travel behavior and compiled a dataset that includes information on the COVID-19 pandemic, weather conditions, holidays, and socio-economic factors. A conditional inference tree model was employed to classify and understand changes in inter-city car travel during the pandemic. The modeling result demonstrates that inter-city travel-related responses to the pandemic varied over time but had few variations across regions. The main findings can be summarized as follows: i) inter-city car travelers responded more sensitively to the number of national confirmed cases than the number of regional confirmed cases; ii) the threshold of travelers' sensitivity to the number of confirmed cases rose as the pandemic continued; iii) inter-city travel behavior exhibited a more resilient recovery as the pandemic prolonged; and iv) travelers tend to resume their inter-city car travels before social distancing rules were actually eased. The findings in this study provide useful insights to suggest a transportation-related policy strategy for effectively managing possible future pandemics at the national level.

The regularity of trips has been a fundamental assumption for estimating or forecasting travel demand, which is an essential part of the analysis for traffic operation and planning. For this type of analysis, a survey method has been traditionally used to understand personal trip behavior in detail. However, surveys have limitations in collecting accurate traffic data on a large scale because they depend purely on participants’ memories. Recently, as the development of ICT facilitates collection of various traffic information, research using this new information has drawn attention, which discovers human mobility pattern on a wide range. In this study, a month of trajectory data collected from on-board transponders equipped in the vehicles in Daegu metropolitan city, South Korea was used to unveil the individual trip regularity in trip chain level. We applied dynamic time warping and the inter-spike interval algorithms to examine these data and to quantitatively measure spatial and temporal regularity separately. The outcomes showed that i) the degree of trip regularity can be properly estimated using the indices, ii) spatial and temporal regularities are correlated - drivers who made trips at regular times also used similar paths in space across days, and iii) commuters and non-commuters have different distributions of regularity scores - commuters made more regular trips. This finding is intriguing because the trip regularity can prove the predictability of human mobility. In addition, if regularity indices are used to measure historically collected trip behavior, this method can provide an alternative way of estimating or forecasting travel demand at the individual trip level. 

As city-level modal splits are outcomes of city functions, it is essential to understand whether and how city attributes affect modal splits to derive a modal shift toward low-emission travel modes and sustainable mobility in cities. This study elucidates this relationship between modal splits and city attributes in 46 cities worldwide, proposing a two-step data mining framework. First, using the K-Means method, we classify cities into private-vehicle-, public-transit-, and bicycle-dominant groups based on their modal splits. Second, we categorize city attributes into environmental, socio-demographic, and transportation planning factors and quantify their interlocked impacts on cities' modal splits via the decision tree method. We observe that the socio-demographic factor has the highest impact on determining the cities' modal splits. In addition, high population density and employment rate are positively associated with low-emission travel modes. High gasoline tax and low public transit and taxi fares often make people reconsider possessing private vehicles. On the other hand, extreme weather conditions (e.g., hot temperatures) can prevent bicycle usage. Our contribution expands the impact of introduced city planning and policies for modal shifts toward a real-world paradigm and we present implications of the proposed framework in developing practical modal shift strategies.

Limited and inaccurate information on parking availability often leads to inconvenience and unintended congestion from cruising-for-parking. To avoid this, it is essential to provide a forecast for parking availability based on spatio-temporal characteristics for parking demand. In this context, parking lots were clustered with similar temporal demand patterns and the clusters were analyzed in a spatial context using building use. Temporal Fusion Transformer–a deep learning model for time-series prediction–was applied to the parking data sets. This method allowed multi-horizon predictions for the availability of multiple parking lots with a single model. Actual data for 41 off-street parking lots in the Seoul metropolitan area in South Korea were used and three clusters were identified: leisure-, residence-, business-related parking lots. To predict parking availability in the short term, weekly, monthly, and yearly parking demand patterns, route search request, and cluster value were reflected. The model showed the performance with 7.65% of SMAPE. The provision of accurate forecast on parking availability can help service operators efficiently manage parking demand, and allow drivers to make better trip decisions. 

Observations from empirical data in the roadway network showed that the relation between averages of network flow versus density often exhibit hysteresis and bifurcation phenomena, which may obscure the reproducibility of a well-defined macroscopic fundamental diagram (MFD). In this paper, we analyzed large-scale trip data from passenger vehicles in an urban network of South Korea and evaluated the shapes of MFDs over many days. It was found that MFDs have two distinctive, reproducible forms: a well-defined, unique relation on weekends and a bifurcation in high-density regime on weekdays. With regard to the bifurcation, we observed higher network flows in the morning and lower network flows in the evening for the same average network density. This implies that the same set of drivers in the network collectively formed two different trip patterns. Hence, we evaluated possible factors that may have effects on the bifurcation phenomenon. In view of this, four factors – heterogeneity, trip completion rate, detouring ratio and commute trips – were analyzed in this study. The findings showed that travelers’ detours could be a key factor for the occurrence of bifurcation because, by detouring travelers improve neither their own travel times nor network-wide travel times, and thereby degrade the network production.

In the context of shared connected autonomous vehicles (SCAVs), the relocation of idle vehicles is a crucial issue for the operation of autonomous mobility-on-demand (AMoD) services. Unlike traditional human-chauffeured taxis, AMoD operations are fully controllable by central systems and not affected by unpredictable human driver behavior. To address the spatial-temporal imbalance between supply and demand and optimize the level of service while minimizing agency costs, we propose a real-time AMoD relocation model. However, vehicle-specific control every second for large fleet sizes may cause computational burdens for the control center. To overcome this, we present a bi-level framework that decomposes the original system-level problem into an inter-zonal relocation problem for the entire service area and an intra-zonal relocation problem for each zone. This reduces the decision space to periodic inter- and intra-zonal relocation of idle vehicles. Using real-world taxi operation data from Daejeon City, Korea, we demonstrate the proposed method via agent-based simulations, assuming that SCAVs replace existing taxis. The results show that the method can significantly reduce the total generalized cost for both users and the agency. Through a sensitivity analysis, we investigate how the performance varies depending on the zone size, inter- and intra-zonal relocation interval, and demand uncertainty and discuss the observed tradeoff. The intended contribution is twofold: first, we propose a novel computationally feasible method that can efficiently operate AMoD systems in real time; second, we provide a closed-form analytical formulation that can help decision-makers explicitly understand the relationship between the cost components and the decision factors.