Research on the Increase in Commuter Use Immediately After the Opening of LRT Using IC Card Data

Abstract

This study aims to predict the purpose of the use of IC card data in LRT immediately after its opening by means of a questionnaire survey and to understand the changes in the number of commuters to better understand the growth in LRT commuter ridership, which has not been fully clarified in Japan. Furthermore, to assess long-term commuter retention for LRT systems, the analysis revealed the following three points. First, a discriminant analysis based on a national PT survey revealed that commuting and leisure or business activities can be classified with high accuracy. Second, it was found that commuter numbers increased immediately after opening, while the number of leisure or business users decreased in the first few months after opening and then leveled off. Third, the increase in the number of commuters was modeled using a logistic curve, and the annual rate of change in ridership was predicted to be less than 1% in the first three to four years after opening.

1. Introduction

1.1. Background and Objectives

To meet the demand for sustainable cities, light rail transit (LRT) is becoming increasingly popular as a tool to achieve this goal, and by the end of 2021, it was introduced in 403 cities around the world [1]. It has been popularized on a global scale due to its good value in terms of upfront cost when compared to conventional heavy rail metro systems. However, its major disadvantage is its relatively high construction cost when compared to the advent of BRT technologies and other similar new transit types, and it has been criticized for making excessive demand forecasts to cover construction cost.

In particular, it has been pointed out that the strong emphasis on profitability in public transportation in Japan is a factor that has prevented the introduction of LRT [2]. Although 38 cities in Japan are considering the introduction of LRT as of December 2023 [3], the introduction of LRT has not progressed for a long time, since the Toyama Light Rail opened in 2006. In the midst of such issues, the Utsunomiya Light Rail (in this paper, referred to as “Haga Utsunomiya LRT”) opened on 26 August 2023. As the first all-new LRT in Japan in 75 years, including streetcars [4], and given its car-dependent urban structure, it was uncertain whether user numbers would align with forecasts. However, ridership during the first year of operation was 1.2 times higher than initially expected [5]. It is hoped that these results will be used to forecast future LRT demand, thereby facilitating consensus building for the introduction of LRT.

On the other hand, the high ridership of the Haga Utsunomiya LRT may include temporary use, due to its uniqueness factor. Therefore, it is important to evaluate LRT after a certain amount of time has passed and its ridership patterns have stabilized in order to determine how well LRT has taken root. However, external factors such as land use and urban structure along the line are changing over time, making it increasingly difficult to isolate the effect of LRT introduction itself.

Considering both advantages and disadvantages, it is considered effective to conduct the analysis immediately after the opening of the LRT service, focusing on consistent purposes for using public transportation (in this paper referred to as “purpose”) such as commuting to work and school. This approach allows for a more accurate assessment of the effects of LRT introduction by separating transient use due to the novelty of the service from daily use, which is more likely to become established over time. To achieve this, it is essential to obtain data over time from which purpose can be estimated.

Therefore, this study aims to use a questionnaire survey to estimate the purpose of IC card data use immediately after LRT operation began and to understand changes in the number of commuters in order to fully understand the status of LRT commuting and commuting use, which has not been fully analyzed in Japan until now. First, a person trip survey is used to ascertain the frequency of trips by purpose, duration of stay, and the proportion of trips on weekends and holidays. The results are then used to perform linear discriminant analysis to determine purpose. This is then applied to IC card data to identify trends in the number of commuter users. Finally, a logistic regression curve is used to approximate the trend in the number of commuters, and the time when use will become established and the number of commuters at that time are predicted.

1.2. Review of Existing Research

1.2.1. Discrepancies Between Forecasted and Actual Ridership

There are many studies on the discrepancy between demand forecasts and the actual use of new public transportation. Pickrell [6] found a tendency to overestimate demand forecasts and underestimate costs in streetcar implementation projects in the U.S., and that the subsidy system is a reason for this. Flyvbjerg et al. [7] analyzed the discrepancy between demand forecasts and actual ridership for transportation infrastructure projects in 14 countries and found that many rail projects were overestimated and that political factors had a significant impact. Perry [8] found that demand forecasts also tend to be optimistic for BRT projects in the United States. Hoque et al. [9] conducted a time series analysis of the divergence in U.S. projects and found that while the accuracy of forecasts has improved for projects since 2000, this is offset by uncertainty due to the spread of vehicle dispatch services and socioeconomic trends since 2012.

These studies reveal that demand forecasts for rail projects tend to be overestimated. However, these studies were based on overall users of new public transportation and were not disaggregated by purpose.

1.2.2. Using Public Transportation IC Cards

Two types of studies that utilize IC card data exist: those that classify users and those that perform time series analysis. The following studies exist as studies that performed classification through unsupervised learning. Briand et al. [10] used a mixed Gaussian model to classify movements that appear to be commuting. Langlois et al. [11] used the k-means method and found that even among high-frequency users, there is a certain amount of usage with no fixed movement patterns. Hosoe et al. [12] used data policing to identify the mobility patterns not only of commuters but also of small-sample users such as the elderly and children.

Some studies exist that use surveys as supervised data to estimate objectives. Alsger et al. [13] created an algorithm to estimate purpose from household travel survey results and found that commuting and returning home purposes can be estimated with high accuracy. Medina et al. [14] used household travel survey results and DBSCAN clustering to estimate commuting purposes from many variables. Kusakabe et al. [15] created a model to estimate purpose with a person trip survey using a naive Bayes classifier and found that commuting to work or school and work or personal matters can be discriminated with high accuracy.

Some studies exist that analyze the impact of changes in transportation services using IC card data. Moylan et al. [16] found that the opening of the Canberra Metro in Canada increased rail ridership, which exceeded the decrease in bus ridership. Nishiuchi et al. [17] analyzed the impact of reduced tram services in Japan and found that the impact was particularly large in city centers. Li et al. [18] developed a means-choice model based on changes in ridership before and after the opening of the railroad and found that cost had the largest impact.

There are also studies that predict future demand based on IC card data. These studies predict the number of public transportation users by considering factors such as season, climate, and events. Meng [19] used the ARIMA model, Nagaraj et al. [20] used the DHSTnet model, Martí et al. [21] used Multi-Agent Simulations, and Du et al. [22] used the LSTM model. These studies focused on predicting the total ridership, and there are no studies that predict users by purpose.

These studies have shown that it is possible to estimate purpose from IC card data, compare changes before and after in transportation services, and make future predictions. However, no study exists that analyzes the change in the number of users by purpose from the estimation of it. This study addresses this phenomenon in the context of LRT.

1.2.3. Post-Opening Surge in Ridership on Newly Introduced Public Transportation Systems

Many studies have clarified the initial surge in ridership and the time when users are established for the ex-post evaluation of projects. Chang et al. [23] defined the initial surge in ridership as a “ramp-up”, examined the reasons for this, and proposed an objective criterion for determining the time when users are established. Flyvbjerg et al. [24] stated that while there are cases where the demand establishment time is set to 3 to 5 years for railway projects, due to data constraints, the only way to evaluate projects is to judge based on data at the time of opening, which is the limit of project evaluation. They also revealed that projects with a lot of tourism demand can expect high usage for several months after opening. Kumar et al. [25] analyzed the ridership of Indian railway projects at the time of opening and found that the strong usage at the time of opening drops sharply after about six months. Shinn et al. [26] used a fixed-effects regression model to eliminate external factors from the increase in ridership at the time of opening of American railway projects and found that the increase in the first two years was highly variable while being statistically significant. As a case study of Japan, there is a World Bank report [27] that investigated the Toyama Light Rail Project, which revealed that the ridership on weekday users remained flat from the first year of operation, but decreased on weekends for several years.

These studies have made it clear that it takes several years for public transport ridership to become established. However, like Section 1.2.1, these studies also focused on overall ridership of new public transportation and did not classify it by purpose.

1.2.4. Behavior Changes When Light Rail Opens

There are many studies analyzing the impact of LRT introduction on changes in transportation behavior. Cao et al. [28]. analyzed behavioral changes after the introduction of the Hiawatha LRT through a questionnaire survey of residents. The results showed that LRT promotes the use of public transportation and encourages the influx of public transportation users, but there were no significant changes in car ownership. Radzimski [29] et al. analyzed behavioral changes after the extension of LRT in Poland using a similar methodology, finding that while public transportation ridership increased, cars remained the primary mode of transportation. Lee et al. [30] analyzed changes in LRT ridership and car ownership between cities with and without LRT using census data, finding that while LRT ridership increased in cities with LRT, there was no decrease in car ownership, suggesting that the shift was mainly from buses. There is also research on the Toyama Light Rail. Matsuda et al. [31] analyzed behavioral changes through surveys of users and found that the introduction of LRT increased the frequency of outings and decreased the frequency of car use.

These studies analyzed changes in LRT ridership and car ownership resulting from LRT introduction. However, many of the surveys relied on questionnaire surveys and were not conducted at multiple time points, so the speed at which the shift occurred remains unclear.

1.3. Positioning of the Study

Taking these studies into consideration, the novelty of this study can be summarized in the following two points. First, we estimated travel purposes using IC card data collected during the initial introduction phase immediately after the opening of the LRT system. This clarified the ramp-up of LRT use by purpose. As a result, the focus on initial acceptance is processed by purpose, which could not be grasped through conventional means such as a total ridership analysis. Second, it compared actual usage with demand forecasts focusing on commuters, which has been difficult to analyze until now.

In addition, since there has been no all-new LRT in Japan, this study has novelty as a case study on the establishment of LRT ridership in Japan.

2. Materials and Methods

2.1. Target Area

The subject area of this study is Utsunomiya City, Tochigi Prefecture, Japan (Figure 1). Utsunomiya City is located approximately 100 km north of Tokyo, with a population of about 0.5 million and an area of about 400 km².

Utsunomiya City, like many other regional cities in Japan, is reliant on cars. Figure 2 shows Utsunomiya City’s main mode share in each survey year in the national PT (Person Trip Survey). Due to factors such as COVID-19, surveys were conducted at irregular intervals. The share of cars continues to increase, accounting for two-thirds of the total in 2021. On the other hand, public transportation accounts for only 6% of the total. This is thought to be due not only to the spread of automobiles but also due to concerns about contracting COVID-19. In addition, densely inhabited districts (DIDs), which are areas with a population density of 4000 people per square kilometer or more, continued to increase with the spread. As shown in Figure 3, in 1960, the DID was a compact area west of Utsunomiya Station, but it spread to areas far from the station, and the central business district subsequently declined. In 60 years, the city’s population has doubled, but during that time, the DID has expanded fivefold, resulting in a decrease in population density. To solve these problems, the city aims to create a networked compact city based on core transportation systems such as LRT and railways.

2.2. Overview of Haga Utsunomiya LRT

The LRT was introduced as the backbone transportation system to help fully realize a networked compact city within Utsunomiya. The route is shown in Figure 4. It starts from Utsunomiya Station, the hub of the central business district, and proceeds through the urban area that was formed by the 1980s to the Utsunomiya Univ. Yoto Campus stop. A large shopping mall exists here. It then crosses the Kinugawa River, which is a bottleneck for traffic. After passing through industrial parks and newly developed residential areas, it leads to the Haga Takanezawa Industrial Park.

Since the LRT opened in an area where the existing public transportation system was not well developed, the number of services has been particularly greatly improved with its introduction. There are also plans to extend the line to the western part of Utsunomiya Station where the city’s central business district is situated.

Demand forecasts by purpose were also made prior to the introduction of LRT [32].

Table 1. Demand forecast of LRT.

	Commute to Work	Commute to School	Business	Leisure	All Purpose
Number of passengers on weekdays	13,357	1305	274	1382	16,318
Number of passengers on weekends	2671	131	82	2764	5648

shows the results of the demand forecast by purpose. It can be seen that commuter ridership accounted for the majority of weekday usage. In addition, when considering the established ridership since the opening, 80% of the demand forecast result was set for the first year and 90% for the second year.

2.3. IC Card Data

The IC card data handled in this study is raw data with one record for each LRT rider, allowing the analysis of frequency.

Table 2. Overview of IC card data.

Target line	Haga Utsunomiya LRT
Data period	26 August 2023–30 November 2024
Target users	All users of transportation IC cards, including Totra
IC card usage rate	Weekdays: 95%, weekends: 88%
Items used in this thesis	Individual ID, boarding/disembarkation time, boarding/getting off station

shows an overview of the IC card data. The data items used in this study were individual IDs, boarding times, alighting times, boarding stops, and alighting stops. The IC card data was collected not only from Totra issued in Utsunomiya City, but also from all types of IC cards. The IC card usage rate for LRT is about 95% on weekdays and about 88% on weekends, so most of the ridership can be ascertained using IC cards. IC card users enjoy many benefits such as greater ease in boarding and alighting, as well as reward programs for discounted fares. Therefore, it is considered that IC card data alone is sufficient to understand the commuters that are the focus of this study.

In this study, the date of operation, origin stop, destination stop, duration of stay, and frequency are defined as follows:

⋅

Date of Operation: The date at the time of the first departure. Even if the date crosses over, it is the same date of operation until the last train.

⋅

Destination Stop: The stop where the individual gets off the train, except for the last use on the individual’s date of operation.

⋅

Duration of stay: The time difference between the individual’s disembarkation time and the next boarding time. However, it is not calculated for the last use on the date of operation. If an individual is transferring from one line to another and gets off at a stop that can be reached only at the transfer destination, the time spent at the transfer destination is used. Note that this study does not cover one-way users who ride only once a day.

⋅

Frequency: The number of days an individual used the destination stop during the month of boarding and the subsequent two months.

2.4. National PT Data

This study used national PT data to determine the frequency and length of stay based on the purpose of taking the LRT and weekday usage. This study used results collected in 2021.

The national PT is a survey conducted once every five years in Japan by the Urban Bureau of the Ministry of Land, Infrastructure, Transport and Tourism (MLIT) and involves a questionnaire on the time, purpose, mode, origin, and destination of people’s travel on a given day on weekdays and holidays.

The survey targets 71 cities across Japan with different population sizes, and 500 samples are used in each city. In addition, a supplementary survey was conducted in the 2021 survey, which included questions on the frequency of visits by purpose. The frequency and duration of stay by purpose were based on bus and rail trips, which are the same types of public transportation as LRT; the subject of this study, and the duration of stay was defined as the difference between the time of getting off at the station or the end of the relevant trip and the time of the next ride on the same mode of transportation. Trips for which the time was not known and subsequent trips that did not involve another rail or bus ride were excluded from the analysis.

2.5. Purpose Prediction Method

Although IC card data can be used to track the use of most users from the time of opening, it has the disadvantage that purpose is unknown. In order to understand the status of commuter retention, which is the goal of this study, it is necessary to determine purpose. Purpose prediction is performed according to the procedure shown in Figure 5.

A linear discriminant analysis is used as the discrimination method. The variables used are frequency, duration of stay, and weekend dummies. Since previous studies have shown that it is possible to discriminate with high accuracy between the two groups of commuting and leisure or business, this study will also discriminate between the above two objectives.

After creating a discriminant function using the results of the national PT as training data, it is applied to the IC card data. Since the last trip for each user is presumed to be a return trip, the purpose of the day’s destination trip is applied. If both commuting and leisure or business are conducted in a single day, the commuting trip is applied to the return trip as the representative purpose. Trips for which the length of stay is unknown or trips that are used only one way are excluded from purpose prediction. The training data used in the linear discriminant analysis was the national PT data for 2021. As mentioned above, this data is likely to include the effects of COVID-19, but since no previous surveys asked about frequency of use, the data is used as is.

Then, to confirm the validity of the discriminant analysis, the percentages by purpose and by destination stop in the user survey conducted by Utsunomiya City from November to December 2023 are tabulated and compared with the discriminant results of the IC card data. A summary of the survey is shown in

Table 3. Overview of user survey.

Target	User of Haga Utsunomiya LRT (Distributed at stations)
Number of distributions and respondents	Distributions: 6800 Respondents: 1305
Distribution period	29 November 2023–1 December 2023
Response period	29 November 2023–28 December 2023
Items used in this thesis	Purpose, Boarding and drop-off station

.

2.6. Methods for Predicting Future Commuter Ridership

Based on the changes in commuter ridership indicated by the objective forecasting methodology, the projections depict the ridership of commuters three years after opening. This is determined within this timeframe, as this is when Utsunomiya city expects demand to be firmly established [32]. Since the demand forecast predicted that demand would eventually level off and that weekday use of the Toyama Light Rail would remain flat, an S-shaped curve is selected for the forecast based on the assumption that the increase in the ridership would slow down and remain at a certain level. Logistic regression is used for this forecasting model. This is a traditional forecasting method [33] and was used as it has been used in recent studies to approximate increasing and decreasing trends in rail ridership [34]. Furthermore, it has also been utilized to approximate the decrease in public transportation ridership due to new coronavirus infections [35]. Otherwise, under the assumption that the ridership converges to a constant number, the Gumpertz and Von Bertalanffy curves, which are S-shaped curves, are additionally selected for comparison.

Changes in the number of LRT users are expected to include seasonal variations, such as a decrease during the summer vacation period. Therefore, public transportation analyzation based on local bus routes showed almost no difference in the ridership between September 2023 and September 2024. The basis of this lack of change can potentially be tied to seasonal variations. To account for this, the number of LRT users are divided by the bus routes’ user index to reduce the impact of these seasonal variations.

3. Results

3.1. Predicicting the Purpose of IC Card Data

3.1.1. Duration of Stay, Frequency of Use, and Percentage of Use on Weekends in National PT

The following three points were revealed from the composition ratio of time spent by frequency and purpose and the ratio of weekends by purpose in the national PT (Figure 6, Figure 7, Figure 8, Figure 9 and Figure 10). First, commuting to work and school makes up a large proportion of longer and more frequent trips. Second, the proportion of short-duration, low-frequency trips is relatively high for leisure and business. Third, the percentage of trips on weekends is low, apart from personal affairs. These results indicate that commuting to work and school differs from leisure and business trips in terms of the time spent, frequency of trips, and percentage of trips on weekends and holidays.

3.1.2. Linear Discriminant Analysis

Since there are clear differences in the duration of stay and frequency for commuting and leisure activities, it is determined that high-precision classification is possible, and linear discriminant analysis is performed. From the results of the national PT data, a linear discriminant analysis is conducted with commuting as −1 and leisure or business as 1. The results reveal the following three points (

Table 4. Results of LDA.

Section		8.056
Coefficient	Duration of stay (min.)	−0.015
	Frequency (day per month)	−0.326
	Weekends dummy variables	3.094
Model evaluation	Accuracy	0.91
	Recall	0.93
	Precision	0.91
	F-measure	0.92

). First, the coefficients for duration of stay and frequency are negative, while the coefficient for the Saturday and holiday dummy variables are positive. Second, all four indicators for the model are high. These results indicate that the model reflects the results of 3.1 and can discriminate with high accuracy. The results of the graphical representation of the respective target areas are shown in Figure 11.

3.1.3. Comparison of IC Card Data and User Survey

The discriminant analysis function is applied to the IC card data for November 2023, and the results of discriminating by purpose reveal that the percentages accounted for by commuting purposes were are identical (Figure 12).

Note that in the user survey, intrinsic demand for LRT boarding is included in the purpose options and classified as a different purpose from leisure or business. Therefore, it should be noted that some of the uses classified as leisure or business in this study may include intrinsic demand.

In addition, the percentage of commuting by destination stop reveal that the IC card data and the user survey are highly correlated (Figure 13).

These results indicate that the accuracy of IC card data in discriminating commuting to work and school is high and reflects actual use.

3.2. Changes in Commuting

3.2.1. Changes in Ridership by Purpose

Ridership by purpose was calculated by multiplying the ratio of use by purpose determined in the previous section by total IC card ridership. IC card users include one-way users, which were excluded from the purpose classification. As of September 2024, these users accounted for approximately 13% of the total, and it is possible that their purpose composition differs from that of round-trip users whose purposes could be identified. However, for the purposes of this study, it is assumed that the purpose composition remains unchanged. The evaluation months for the ridership were September 2023 and July 2024. September 2023 was the first month of full-scale operations. July 2024 was the final month of the first year of operations. Since data for the entire second year is not available, the evaluation period will be limited to the first year.

This result is compared with the demand forecast. Since no forecast was made at the beginning of the service focusing on commuting purposes, 80% of the forecasted value was used for the first year and 90% for the second year, as was the case with the forecast of the overall ridership.

The following three points became clear from the weekday results (Figure 14). First, the overall ridership has been increasing except during the winter vacation period from December to January and the summer vacation period in August, reaching approximately 13,000 IC card users alone in March 2024, which is the forecasted value for the first year, and 130% of the forecasted value as of July 2024. Second, the number of commuters is increasing, but as of July 2024, it was 94% of the projected value. Third, the number of leisure or business users greatly exceeded the forecast immediately after the opening, reaching 389% of the forecast in September 2023, but then declined slightly. Nevertheless, by July 2024, they were 316% of the forecast.

The results for Saturdays and holidays reveal the following three points (Figure 15). First, the overall ridership, which had been on a downward trend immediately after the opening, leveled off within a few months, and by July 2024, the ridership was 189% of the first-year projection. Second, the commuter ridership has been slightly increasing, but as of July 2024, it was 69% of the forecasted value. Third, leisure or business users account for most of the total ridership, at 308% of the projected value as of July 2024.

3.2.2. Predicting Commuter Ridership

In this section, based on the growth trend of commuters shown in the previous section, we forecast the amount of commuter ridership in the third year after opening, which was predicted to be the period when demand was expected to be established. Figure 16 shows the index of LRT commuter ridership, excluding the effects of seasonal fluctuations in buses.

Comparison of the logistic regression curves and demand forecasts revealed the following five points (Figure 17 and

Table 5. Parameters and fit indices.

	Logstic	Gompertz	Von Bertalanffy
Formula	$\mathrm{y}={\displaystyle \frac{L}{1+e^{-k(x-t0)}}}$	$\mathrm{y}=\mathrm{L}·\mathrm{e}\mathrm{x}\mathrm{p}(-k·e^{rx})$	$\mathrm{y}=\mathrm{L}{\left(1-e^{k\left(x-t0\right)}\right)}^3$
Parameter	L: 12,702 k: 0.233 t0: −0.479	L: 12,955 k: 0.672 r: 0.184	L: 13,128 k: 0.168 t0: −9.47
95% confidence interval	L: (9835, 15,570) k: (0.012, 0.455) t0: (−2.363, 1.405)	L: (9429, 16,561) k: (0.424, 0.920) r: (−0.022, 0.390)	L: (9216, 17,040) k: (−0.034, 0.369) t0: (−19.912, 0.965)
R2	0.845	0.847	0.848
RMSE	733	727	726
MAE	603	593	590
AIC	69.06	68.86	68.80
BIC	70.75	70.56	70.50

). First, the R² value was 0.845, confirming the good fit of the model. All of the other goodness-of-fit indicators were also favorable. Second, the convergence value was approximately 2.0 times the ridership in September 2023. Third, while the demand reached the forecasted value up to the second year, it was about 14% below the forecasted value during the third year of establishment. Fourth, the projected increase in ridership from the third to the fourth year was less than 1%, which is almost in line with the city’s forecast for the period when demand is established. Fifth, other curves were also analyzed to confirm that R², goodness of fit, and convergence values are comparable.

4. Discussion

This study used a questionnaire survey to estimate the purpose of IC card data immediately after the LRT began operation and to understand changes in the commuters, thereby gaining an understanding of the status of LRT commuting purpose and use, which has not been clarified in Japan until now. The following three points became clear.

First, the results of the national PT survey revealed that commuting and leisure or business activities can be classified with a high degree of accuracy. This result is similar to that of Kusakabe et al. [15]. For a more detailed classification than this, it is considered necessary to understand the type of commuter pass and land use.

Second, changes in overall ridership and ridership by purpose indicate that overall ridership exceeded the forecast on weekdays, but commuter ridership was lower than what was forecasted, which was then compensated for by the number of leisure or business users exceeding the forecast immediately after the opening. However, the commuter ridership gradually increased and approached the forecast. In the study by Chang et al. [23], the initial increase in ridership was attributed to the time it takes for people to understand the new service and the time it takes to change their travel behavior. This is thought to be the reason why it takes longer for commuters to become established users. Flyvbjerg et al. [24] revealed that routes with many tourist passengers tend to see faster adoption. In the case of Utsunomiya, however, many commuters are present, and it was possible to clarify this by appropriately identifying the purpose of travel. Compared to the Toyama Light Rail [27], Japan’s first LRT system, the weekday ridership of the Toyama Light Rail has remained stable since its opening. This is thought to be because the line existed as a conventional railway before the LRT opened, and residents were quick to understand it. When introducing a new LRT, there is a high possibility that the ramp-up period will be long, like for the Utsunomiya LRT. Research by Cao et al. [28], Radzimski et al. [29], and Lee [30], which focused on the shift in transportation modes, showed that the increase in light rail users was mainly due to a shift from bus users, while the shift from car users was slow. In Utsunomiya, bus routes were reorganized in conjunction with the introduction of LRT, so it is thought that bus users accounted for the majority of initial users.

Third, future projections based on logistic regression curves revealed that commuter ridership may not reach the projected values in the third year, when demand is expected to settle in. Nevertheless, the total ridership is expected to exceed the forecast because the number of leisure or business users is much higher than forecasted. In addition, since the service levels of LRT, such as frequency and travel time, are below the conditions at the time of the demand forecast, it is considered that service improvements can bring the ridership closer to the forecasted demand value. Previous studies such as Meng’s study [19] have used ARIMA models to remove seasonal variations. While this study does not employ time series models like ARIMA, seasonal fluctuations were accounted for by dividing LRT ridership by the bus routes’ user index. This approach has become possible to remove seasonal variations from IC card data covering only about one year and achieve a high degree of accuracy in fitting.

5. Conclusions

The above results suggest that ridership will be established at different times depending on purpose. In future project evaluations, clearly defining the anticipated purposes of use in advance will enable us to predict ramp-up speed. For routes expected to be used for commuting attendance, paying attention to the slow ramp-up will enable us to appropriately address criticism regarding the low number of initial users.

While this study focused on commuters to forecast changes in ridership and future retention, a similar analysis for leisure or business use would make it possible to forecast the retention of overall ridership. However, since leisure or business use includes the use of LRT for its intrinsic demand at the time of its opening, it is necessary to categorize this type of use and understand the changes for each type of use. In addition, it will be possible to verify the predictions of this study by updating the data in the future. In doing so, it will be necessary to eliminate external factors such as population growth along the rail line and the development of employee sites. The national PT data was used to perform a discriminant analysis of railway and bus usage, but it is thought that a more detailed analysis could be performed by conducting a direct survey of LRT users to consider the unique usage patterns of LRT. This would also make it possible to eliminate differences in the frequency of usage due to COVID-19, which had a significant impact as of 2021. For one-way users whose purpose could not be determined in this study, it may be possible to classify them using factors such as travel time.