Introduction

Over the past few decades, China’s regional transportation infrastructure has undergone a huge expansion1,2. As a result of this tremendous wave of construction, China now has the world’s largest network of high-speed rails (HSRs) and expressways3,4,5,6. The construction of infrastructure has strengthened regional connectivity, which in turn has supported economic growth7,8. China’s determination and actions in its endeavor to advance its transport infrastructure have become a focus of worldwide attention.

Parallel to the infrastructure boom, vehicle ownership in China has surged to overwhelm that of the United States by 2010 to become the world’s largest automotive market9. This has added to the congestion on the roads and its fast-tracking transportation-related carbon emissions10,11. In contrast, rail transport has much lower per-capita carbon emissions and fairer access for different social groups12,13,14, thus it presents a promising alternative. Nevertheless, despite the remarkable growth of the HSR network in China, the spatial patterns of its substitution for road travel remain less clearly known.

Although trips between cities are not a large proportion of all trips, the distances involved are so long that they can still have a significant carbon footprint15,16. In this context, optimizing the inter-city trip modal share structure is essential to ensure sustainable urban development. The interplay of transport modes and inter-city flows goes far beyond environmental impacts to issues of accessibility, connectivity, and economic vitality for both cities and regions17,18,19. It portrays a city that acts as a node in the territory and gives shape to the spatial structure of the regional market20,21,22.

The rapid expansion of China’s infrastructure has also raised questions about equality. As a vast country with diverse geographic landscapes, China faces unique challenges to equitably provide efficient transport to all regions. Traditionally, residents in remote areas have had to take extremely long times to travel to other parts of the country by road23,24. Such travel times can be significantly reduced by HSR and planes, thereby improving the connectivity and accessibility of these regions25,26,27. However, the extent to which benefits from recent infrastructure boom are equitably distributed to different regions is still an open question. This issue of whether changes in modal share occur by way of transportation equitably is not only related to social justice but also to balanced regional development and national cohesion.

Big data analytics has now opened up new opportunities for visualizing and analyzing these complex inter-city flows at an unprecedented scale28,29,30,31,32. The existing research in this vein has so far primarily concentrated on the overall magnitude of such flows and the underlying factors influencing them. However, limited attention has been paid to the modal composition of these inter-city movements. More precisely, little research has been conducted regarding spatial variations of the modal shares of these flows. Moreover, little attention has been given to the multi-year temporal evolution of these modal patterns.

This study aims to bridge this gap by leveraging the integration of multi-source location-based services (LBS) big data to investigate the changing modal patterns of inter-city flows in China. We selected the time period of 2015–2019 as the study timeframe in order to mitigate the potential confounding effects of the COVID-19 pandemic on the results. Specifically, we seek to address the following questions: (1) What are the modal share compositions of inter-city flows across different regions of China? (2) Is road travel in China declining, and to what extent is it being substituted by rail or air travel across different regions? (3) What are the geographic characteristics and equality of the regions exhibiting modal share shifts?

Methods

Data fusion

To obtain a comprehensive understanding of the modal share composition of inter-city flows in China, we employed a data fusion approach that integrated multi-source data. To ensure the accuracy of our dataset, we combined and calibrated data from Amap and Tencent using official statistics from the Ministry of Transport of China33. We selected the research period spanning from 2015 to 2019 to avoid the influence of diverse lockdown policies implemented during the COVID-19 pandemic on inter-city mobility. Our analysis encompasses 367 cities, of which 4 are directly administered municipalities, 333 are prefecture-level, and 30 are county-level directly under provincial jurisdiction. This coverage includes all prefecture-level and above cities nationwide.

Amap, a leading mobile navigation software company in China with a substantial market share, provided data spanning 2018–2019. The Amap dataset consisted of daily road travel records between all cities at the prefectural level and above. These data were derived from Amap’s location-based services (LBS) application, which collects user-authorized location information. Tencent, a prominent internet and smartphone company in China34, contributed data covering the period from 2015 to 2019. Tencent’s daily inter-city records were collected separately for origin and destination cities, with a threshold applied to identify the top 10 flows for each city. The data distinguished between three transportation modes: road, rail, and air travel. As we had obtained data for all cities, we were able to compensate for the missing flow data28.

Our data processing approach involved integrating Amap’s road travel data with Tencent’s rail and air travel data. Subsequently, we calibrated the modal shares using official statistics from the Ministry of Transport of China to ensure consistency with authoritative figures for each transportation mode. The combination of these three data sources leveraged their respective strengths. Amap data was chosen as the source for road travel due to its high accuracy in capturing vehicular movements and its comprehensive nature as a full sample dataset. Conversely, Tencent data was utilized for rail and air travel as it encompassed all travel modes. Official data served as a calibration source, providing authoritative total travel volumes across all transportation modes.

Since Tencent data covered the entire study period, we utilized the overlapping time frame of 2018–2019, where both Amap and Tencent data were available, to supplement the missing road travel data from Amap for the remaining years. Considering that Amap and Tencent data contain equivalent information for the corresponding city pairs, while accounting for variations resulting from sample expansion, we adopted the assumption of a linear relationship between the road travel flow volume derived from Amap and that obtained from Tencent:

$${{VA}}_{{ij}}^{\left(t\right)}={\alpha }_{t}{{VT}}_{{ij}}^{\left(t\right)}+{\beta }_{t}$$
(1)

Here, \({{VA}}_{{ij}}^{(t)}\) and \({{VT}}_{{ij}}^{(t)}\) represents the road travel flow volume between city \(i\) and city \(j\) at time \(t\) derived from Amap and that obtained from Tencent, respectively. After estimating the coefficients \({\alpha }_{t}\) and \({\beta }_{t}\), we used this linear model to calculate the top 10 road travel flows for each city’s departure and destination in the remaining years. Next, we made an assumption that the flows between smaller cities, which were not covered by Tencent data, exhibit a distribution consistent with the patterns observed in the available Amap data. Based on the full-sample Amap data, we estimated the proportion of flow contributed by each city pair to the total flow of each departure city by utilizing data from the existing years with complete sample coverage:

$${{VT}}_{{ij}}^{\left(t\right)}={\gamma }_{{ij}}^{\left(t\right)}\mathop{\sum }\limits_{k=1}^{n}{{VT}}_{{ik}}^{\left(t\right)}$$
(2)

Here, \(n\) is the number of all cities. After estimating each coefficient \({\gamma }_{{ij}}^{(t)}\), we approximated the road travel flows between the remaining city pairs to obtain a complete dataset covering all city pairs. Subsequently, we refined the data by incorporating the total flow volume of various transportation modes, as provided by the Ministry of Transport, for each year. This correction was executed using the following formula:

$${{VC}}_{{ij}}^{\left(m\right)}={V}_{{ij}}^{\left(m\right)}\frac{{T}_{m}}{\displaystyle {\sum }_{i=1}^{n}\displaystyle {\sum }_{j=1}^{n}{V}_{{ij}}^{\left(m\right)}}$$
(3)

Here, \({{VC}}_{{ij}}^{(m)}\) is the refined flow volume from city \(i\) to city \(j\) for transportation mode \(m\); \({V}_{{ij}}^{(m)}\) corresponds to the raw data of the flow volume based on previously processed LBS data; \({T}_{m}\) is the total volume of transportation mode \(m\) based on the official statistics.

Spatial analysis and visualization of flows

To obtain a continuous spatial representation of modal shares, we employed a combination of kernel density estimation and kriging interpolation techniques. This processing transforms the original flow data into simulated volume data distributed across the grid. First, we constructed a spatial dataset of inter-city flows for each transportation mode. Subsequently, for each mode, we applied kernel density estimation to the corresponding flow data. This approach generated a smooth surface for each flow, with the maximum value along the straight line connecting the origin and destination cities, gradually decreasing with distance from the line. The sum of all grid cell values within this surface equals the total flow volume for that specific movement, ensuring that the aggregation of all grid cell values nationwide corresponds to the total flow volume for the respective transportation mode.

Next, for all grid cells with values within the top 99.99 percentile after kernel density estimation, we calculated the modal share percentages for road, rail, and air travel using a raster calculator. This approach ensured that the total flow in the grids reached a certain scale, guaranteeing the reliability of modal share calculations. For grid cells without assigned values, we employed circle neighborhood’s focal statistics to interpolate the modal share percentages. The resulting continuous surface covered the entire study area, accounting for both missing values and the lowest 0.01 percentile values.

We computed degree centrality, betweenness centrality, and closeness centrality for each city in the inter-city transport networks of rail, air, and road modes in 2015 and 2019. Referring to the methods of Hu et al. 31, degree centrality is computed using the weighted network, and closeness centrality and betweenness centrality are computed using an unweighted network constructed by taking the top 20% of edges for the corresponding mode. These metrics were normalized to ensure comparability across modes and years.

In visualizing the modal share patterns, inspired by Toomanian35, we adopted a strategy based on the RGB color space solid cube, encoding the three transportation modes onto the primary colors. This approach allowed for an intuitive representation of the modal composition across different regions, enabling a direct visual assessment of the relative contributions of each transportation mode.

Analysis of inequality

Firstly, to quantify the overall inequality in the shifts of transportation mode shares over time, we computed pseudo-Gini coefficients for the annual proportion changes. Given the multiscale nature of space and the uneven spatial distribution of population, we calculated three distinct Gini coefficients: County-administrative-unit-based Gini (C-Gini), Population-weighted Gini (P-Gini), and GDP-weighted Gini (G-Gini). Here, C-Gini was derived by aggregating flow data from grids to the county administrative level, followed by calculating the proportional change estimates at this unit scale. P-Gini accounted for population distribution by weighting each county’s resident population based on census data from China’s Seventh National Population Census. G-Gini was calculated by weighting each county’s GDP. The three coefficients are calculated as follows:

$${Gini}=1+\frac{1}{{\sum }_{i=1}^{n}{w}_{i}}-2\,\cdot \,\frac{{\sum }_{i=1}^{n}\left(\frac{{\sum }_{k=1}^{i}{{x}^{{\prime} }}_{k}{{w}^{{\prime} }}_{k}}{{\sum }_{k=1}^{n}{{x}^{{\prime} }}_{k}{{w}^{{\prime} }}_{k}}\,\cdot \,{{w}^{{\prime} }}_{i}\right)}{{\sum }_{i=1}^{n}{w}_{i}}$$
(4)

where \({x}^{{\prime} }\) and \({w}^{{\prime} }\) are the sorted vectors of variable \(x\) and weight \(w\), and \(n\) is the total number of counties.

Subsequently, we linked these variations in inequality with regional geographical attributes. We incorporated several variables that represent these features: average altitude as a proxy for topography36, per capita GDP as an indicator of economic development levels37, and resident population density to reflect demographic condition38.

Modal share evolution potential index

To design an Inter-city Modal Share Evolution Potential Index, we quantified the potential for a city’s road travel share to decrease in favor of rail and air travel. This index considers the weighted average of the potential decrease in road travel share for each city pair. The calculation process of this index involves all flow data and simulates based on the origin cities. The index is calculated as follows:

$${I}_{i}=\frac{{\sum }_{j=1,j\ne i}^{n}{{VC}}_{{ij}}\cdot \left(\Delta {R}_{{ij}}+\Delta {A}_{{ij}}\right)}{{\sum }_{j=1,j\ne i}^{n}{{VC}}_{{ij}}}$$
(5)

Here, \(n\) is the number of all cities. \({{VC}}_{{ij}}\) is the refined actual flow volume from city \(i\) to city \(j\). \(\Delta {R}_{{ij}}\) and \(\Delta {A}_{{ij}}\) represent the potential increase in rail travel and air travel from city \(i\) to city \(j\) respectively, which are calculated as follows:

$$\Delta {R}_{{ij}}=\max \left(0,\,\hat{R}\left({D}_{{ij}}\right)-{R}_{{ij}}\right)$$
(6)
$$\Delta {A}_{{ij}}=\max \left(0,\,\hat{A}\left({D}_{{ij}}\right)-{A}_{{ij}}\right)$$
(7)

Here, \({R}_{{ij}}\) and \({A}_{{ij}}\) are the current percentage of trips from city \(i\) to city \(j\) that are made by rail and air respectively.\(\,\hat{R}({D}_{{ij}})\) is the fitted rail travel share for distance \({D}_{{ij}}\) in the scenario that both city \(i\) and city \(j\) have HSRs. Similarly,\(\,\hat{A}({D}_{{ij}})\) is the fitted air travel share for distance \({D}_{{ij}}\) in the scenario that both city \(i\) and city \(j\) have airports. Following the results of previous studies39, we assumed a power law function between the volume of trips and the distance after a certain threshold is exceeded. Under this assumption, the relationships between travel modal share and distance are:

$${R}_{1}\left({D}_{{ij}}\right)=a{{D}_{{ij}}}^{-\alpha }+\varepsilon$$
(8)
$${A}_{1}\left({D}_{{ij}}\right)=b{{D}_{{ij}}}^{-\beta }+\varepsilon$$
(9)

Among them, \(a\), \(b\), \(\alpha\) and \(\beta\) are parameters to be estimated based on the available data, and \(\varepsilon\) is a random error term. Specific thresholds are set based on the results of data analysis.

Furthermore, considering the potential impact of terrain on travel39, we also assumed the relationships between travel modal share and distance that take into account the influence of terrain:

$${R}_{2}\left({D}_{{ij}}\right)=a{{D}_{{ij}}}^{-{\alpha }_{1}}{{H}_{i}}^{-{\alpha }_{2}}{{H}_{j}}^{-{\alpha }_{3}}+\varepsilon$$
(10)
$${A}_{2}\left({D}_{{ij}}\right)=b{{D}_{{ij}}}^{-{\beta }_{1}}{{H}_{i}}^{-{\beta }_{2}}{{H}_{j}}^{-{\beta }_{3}}+\varepsilon$$
(11)

Among them, \({\alpha }_{1}\), \({\alpha }_{2}\), \({\alpha }_{3}\), \({\beta }_{1}\), \({\beta }_{2}\) and \({\beta }_{3}\) are parameters to be estimated based on the available data. \({H}_{i}\) and \({H}_{j}\) are the average altitudes of city \(i\) and city \(j\).

Results

Modal share evolution of inter-city flows

By visualizing the vast inter-city flows based on data fusion, it is observed that there is striking spatial imbalance in China’s inter-city traffic distribution (Fig. 1a). The distribution was heavily concentrated in the southeast of the Hu Line (a population density demarcation proposed by geographer Huanyong Hu in 1935, running from Heihe in the northeast to Tengchong in the southwest, with over 94% of China’s population living southeast of it on less than half of the land area40), especially in the diamond-shaped area formed by Beijing, Shanghai, Guangzhou, and Chengdu, which verified the conclusion of the latest big data studies28. From 2015 to 2019, there was little change in the general trend regarding the spatial pattern for inter-city flows.

Fig. 1: Modal share of inter-city flows from 2015 to 2019.
figure 1

a Visualization of the volume of OD flows dominated by different modes using RGB color channel blending. b Gird-level modal share based on kernel density estimation for different modes using RGB color space cube. The three primary colors correspond to the situations where the estimated values of the three modes of transportation on the grid are 100%. c Total volume and share of different modes across the country.

Full size image

We used an RGB color channel blending method to visualize the distribution of the transportation modes. In this visualization, all flow data is displayed by transportation mode. The region in general has a high preponderance of the green color, exhibiting pivotal role of railways in inter-city transportations in China. In the most active diamond-shaped region, key mega city regions such as Beijing-Tianjin-Hebei, Yangtze River Delta, and Pearl River Delta have indicated road dominance over the years. The rise of blue curves year after year reflects the remarkable increase of long-distance air travel, specifically very prominent in coastal areas to the east.

For a more granular analysis of the evolution of transportation modes, we applied a kernel density estimation method to allocate traffic flows to a 1 km grid (Supplementary Figs. 1–4). From the results, we see that inside the diamond-shaped area, the eastern half had more activities of travel. But over the years, the spatial distribution of traffic flows tended to be balanced, and variation was not dramatic. We then applied this method to all transportation modes, calculated the modal share in each grid, and mapped it into RGB color space (Fig. 1b). The result was stunning.

The first is the national-scale transition from road-dominated to rail-dominated regions. At the national scale, the percentage share of road travel reduced from 84.5% to 75.1% in the five years (Fig. 1c), but the spatial changes have been more marked. Indeed, looking at 2015, only a few sparsely populated areas of the diamond region and parts of northwest China were rail-dominated. By 2019, although the percentage share of rail travel only increased from 13.2% to 19.1% (Fig. 1c), most of the diamond region had become rail-dominant, with the road domination only remaining inside some mega city regions. In city pairs with an annual travel volume of over 50,000 people, the share of road travel as the main mode of transportation dropped from 37.26% to 23.73%. Correspondingly, rail-dominated trips increased from 37.95% to 54.27%.

Second, road transport prevailed in peripheral regions, such as Xinjiang in the northwest, Yunnan in the southwest, and Heilongjiang in the northeast. These have been relative blind spots in terms of railway development, leaving HSR itself within the framework that connects the provincial capitals with the main cities in China, while setting intra-provincial connectivity sparse.

Finally, air transport dominated in few places and few times. At the national level, it increased its modal share from 2.3% to 3.8% (Fig. 1c). The exception is the diagonal edge of the diamond region between the Yangtze River Delta and the Pearl River Delta. Unlike Beijing and Shanghai, which are connected by a direct HSR link41, these two highly developed mega city regions have no direct HSR link. However, in the recent past, with the steady economic boom in China, the business travel demand in both directions has increased significantly, and airlines have started playing a major role in this corridor.

Spatial dynamics of road travel substitution

To delve deeper into the spatial characteristics of road travel substitution, we analyzed the changes in modal share between the initial year of our study, 2015, and the final year, 2019. This was done at both the grid level (Fig. 2a) and the provincial administrative level (Fig. 2b).

Fig. 2: Modal share change of inter-city flows from 2015 to 2019.
figure 2

a Grid-level spatial distribution of percentage changes in road, rail, and air travel. b Differences in the percentage changes in road, rail, and air travel among provincial administrative units.

Full size image

We find that one clearly dominant phenomenon is the fall of the proportion of road travel. This decrease, in general, was sharper in the southern regions. Mountainous or hilly provinces, involving Fujian and Jiangxi in the southeast, as well as Gansu and Qinghai in the northwest, experienced key declines in road travel. Most often, it is because of the accessibility of improved alternative infrastructure that shifted road travel to other modes of transport. Hainan, the only island province in mainland China, also represented a major drop in road travel. In contrast, some of the economically advanced eastern provinces, like Jiangsu, Shandong, and Guangdong, had smaller declines in overall road travel. These provinces contain mega city regions with dense urban networks and strong economic links between neighboring cities that include substantial inter-city commuting42. As the infrastructure in these regions is already very well developed, there is a little chance of further decline in road travel.

Rail travel share rose correspondingly in most regions. The rapid expansion of China’s railway network during this period brought about by the wave of HSR construction can be a direct reason for the increase in modal share. During this period, 687 new railway stations were opened across the country and the railway network mileage increased by 20,239.9 km (Supplementary Fig. 5). As shown in the bivariate spatial correlation results, the increase in the number of railway stations and the increase in railway density both exhibit high-high and low-low clustering with modal share change in typical areas. A high decline in road travel mainly in mountainous and hilly provinces, was matched by a high increase in rail travel, therefore, railways have more than compensated for the loss of road travel by filling the gap. An interesting exception is Beijing, which already had the highest railway share among provincial administrative units but still saw a significant increase. That can be explained by the significance of Beijing as a key railway hub in China’s HSR network.

The changes in air travel underwent an evident east-west division. As shown in Supplementary Fig. 6, 44 new airports were opened across the country during the study period. Although the distribution of these airports is relatively even, improving the overall coverage of air services, the increase in air travel share in the eastern region is more prominent. In the case of eastern regions, the share of air travel increased substantially. In western regions, it either remained unchanged or decreased. The provincial scale indicates that Hainan had the highest increase in the share of air travel. As an island province, although regular ferry services capable of carrying cars and trains connecting Hainan and Guangdong make road and rail travel across the strait possible, Hainan still largely relies on flights to have closer connections with the rest of China. In 2018, Hainan became the only province-wide free trade zone in China43, making this province more economically and touristically attractive, increasing its connectivity with other regions, mainly by air. At the other end, provinces like Henan and Hebei, situated in central China, have recorded the lowest growth rate in air travel. Since these places are located in the central part of China, their average distance to other major cities in the country is relatively short; therefore, air travel demand is depressed. In addition, these provinces have low airport densities, which further constrains future prospects of air travel growth.

Through network analysis conducted under different transportation modes, the changes in the nodal status of cities within these networks can be observed (Supplementary Figs. 7–9). Although the three modes—road, rail, and air—exhibit similarities in the importance of network nodes, with cities located southeast of the Hu Line generally holding higher centrality than others, there are notable distinctions in the prominence of central cities across these modes. Specifically, the centrality of major cities becomes increasingly pronounced from road to rail to air, with the air network featuring the most prominent central cities and the road network showing relatively less advantage for its central cities. Over a five-year period, the centrality rankings of cities in central and western regions have shown a general improvement across all three modes. However, the enhancement in centrality for these regions is particularly significant in the rail and air networks. This observation suggests that the rise in the status of central and western cities within the rail and air networks has contributed more substantially to the substitution of road travel by rail and air travel, highlighting the growing role of these modes in regional connectivity.

To provide a comprehensive picture of these spatial dynamics, we further decomposed directional and distance-related change for various modes of transportation. From the point of view of travel directionality and scale (Fig. 3a), the majority of road and rail travel was east-west oriented. More importantly, the volume of road travel was significant along the northwest-southeast axis. In contrast, air travel revealed a very different pattern of scale and proportion, mainly along the axis from northeast to southwest. Detailed examinations of the directions of inter-city travels for key cities, as illustrated in Supplementary Figs. 10–13, suggest that this pattern is likely driven by strong air travel links among major cities like Shanghai and those of the Pearl River Delta.

Fig. 3: Spatial patterns of inter-city flows of different travel modes.
figure 3

a Direction distribution of road, rail, and air travel in 2019. b Modal share changes in different directions of road, rail, and air travel from 2015 to 2019. c Distance-dependent changes in the volume of road, rail, and air travel from 2015 to 2019.

Full size image

The modal share change for each travel mode also varied at different directions (Fig. 3b). A noticeable drop in road travel share, often beyond 15%, in the northeast-southwest axis corresponded to increases in rail and air travel shares. While an increase in rail travel was somewhat ubiquitous, the increase in air travel share was much more centered along this single direction. This trend is partly explained by the previously noted strong ties between Shanghai and the Pearl River Delta. Another contributor is Hainan, which showed the highest increase in air travel share among provinces, primarily connecting with eastern and northeastern China, further contributing to this directional trend.

The shares of different modes over different travel distances also show interesting shifts (Fig. 3c). For road travel, there was relatively constant power-law decay. In contrast, rail travel displayed segmented power-law decay; the decline in the first segment is slower and then increases. Already visible in this figure is that the air travel pattern is different: it first increases and then drops. Notably, the crossover points—the distances where one travel mode becomes more preponderant over another—changed between 2015 and 2019. For instance, in 2015, road travel dominated for distances up to approximately 1000 kilometers. By 2019, this crossover point had come down to about 500 kilometers, indicating rail mode dominated the 500–1000 kilometer range instead.

The second equally important change is in the road-air modes. In 2015, this cross-over was seen at about 1200 km. Beyond this distance, travel volumes by both modes were similar. In 2019, this tipping point had shifted to around 1000 km, suggestive of the fact that for distances greater than 1000 km, air had become the mode dominant over roads. These changes highlight increasing importance through rail travel for medium distances and the increasing reliance on air for longer-distance travel.

Regional inequality of modal restructuring

One critical aspect of modal restructuring in inter-city transportation in China is the spatial equality of these changes. To explore this, we calculated the Gini coefficients based on county-administrative units (C-Gini), population (P-Gini), and GDP (G-Gini) (Fig. 4), using the previously processed volume distribution at the grid level. We conducted 1000 Bootstrap replications to calculate the differences in three types of Gini coefficients. The results indicate that the differences among these three Gini coefficients are statistically significant at the 95% confidence level. The C-Gini analysis revealed that air travel changes were the most balanced, with a Gini coefficient of 0.0289, while rail travel was the most unequal, with a Gini coefficient of 0.0866. However, when considering population and GDP, the results shifted. The P-Gini and G-Gini for rail travel were 0.0665 and 0.0587, respectively, indicating a decrease. This suggests that despite the inherent inequalities, the rise in rail travel share was more pronounced in cities and regions with larger populations and higher GDP. This trend likely reflects a prioritization of HSR infrastructure development in densely populated and economically active areas. Conversely, for air travel, the P-Gini and G-Gini were slightly higher than the C-Gini, at 0.0291 and 0.0301, respectively. This indicates that the increase in air travel share was more distributed among relatively remote and less densely populated regions. Such findings highlight a distinct divergence in how different transportation modes contribute to regional connectivity.

Fig. 4: C-Gini, P-Gini, and G-Gini, along with their corresponding Lorenz curves, for changes in modal share of inter-city transportation modes.
figure 4

The subfigures in the first, second, and third rows represent the changes in the share of road trips, rail trips, and air trips, respectively. The subfigures in the first, third and third columns represent the Gini coefficients based on county-administrative units, population and GDP, respectively.

Full size image

Corresponding infrastructure construction is a direct influencing factor of the increased shares of rail and air travel. At the prefecture level, we examined the dates of new opening of airports and HSRs and their correspondence with changes in modal share (Fig. 5a). Cities without airports had to rely on other cities’ airports. Their changes in air travel share were up-and-down deviations from 0. Cities that already had an open airport prior to 2005 varied quite significantly in their changes in air travel share, with some cases of very strong growth and others of decline. This variability underscores differentiation among established airport cities over time. Cities that have opened an airport between 2005 and 2009 had more growth in air travel compared to cities opening up an airport in the period of time between 2010 and 2014. Thus, among cities that opened an airport in this period, earlier opening dates mostly corresponded with larger increases in air travel, suggesting that time is needed for new airports to make a big difference in people’s travel behaviors. The cities with no HSR connections had limited changes in the rail travel share. In most cases, cities with HSR opened before 2015 had overall increases in rail travel. The other cities which opened during the study period witnessed an increase in rail travel shares. Especially for all the cities which opened their HSR services in 2015, there was an increase in rail travel shares.

Fig. 5: Relationship between inter-city modal share and other factors.
figure 5

a Violin plots illustrating the relationship between modal share change and the opening timings of high-speed rail and airports. b Scatter plots illustrating the relationships between modal share change and population density, per capita GDP, and average altitude, with non-parametric smoothing using generalized additive modeling (GAM) and 90% confidence intervals.

Full size image

Finally, we mapped the relationship of modal restructuring to main geographical factors at a county scale level in Fig. 5b. In counties with higher population densities, the growth in rail travel was less pronounced, likely due to the early establishment of mature HSR networks and more stable travel patterns. However, in regions with the lowest population densities, changes in both rail and road travel shares were also minimal, probably because these areas have not yet been covered by HSR or airports. The relationship between per capita GDP and modal restructuring was non-linear. Counties with moderate per capita GDP witnessed the largest increases in the rail and air travel mode shares. The impact of elevation on modal restructuring was more complex. In areas of the very lowest elevations, both rail and air travel shares increased significantly. At around 1000 meters, the growth in both modes moderated. In areas between 2000 and 3000 meters, rail travel share increases were substantial, but changes in air travel were small. In regions at higher elevations, shares of rail travel increased less or air travel even decreased. Areas like this often already had assembled airfields, and the railway enlargement supplied a substitute for long distance traveling.

Potential of further modal share evolution

Based on the relationship between modal share and distance in the presence of well-developed HSR and air infrastructures, we measured the modal share evolution potential index for each prefecture, which reflects the potential for further declines in the share of road travel in each prefecture. The index is simulated using cities as the origins of the flows. When the effects of terrain are not taken into account (Fig. 6a), an overall pattern is that areas southeast of the Hu Line have lower potential indices, while areas northwest of the line have higher indices. Areas including Xizang (Tibet), Qinghai, Xinjiang and eastern Inner Mongolia have the highest potential indexes, but it is clearly uneconomical to make large-scale transportation infrastructure investments in such sparsely populated areas.

Fig. 6: Prefecture-level inter-city modal share evolution potential index.
figure 6

a Scenario assuming that the influence of topography is not considered. b Scenario under the assumption of being influenced by topography.

Full size image

More remarkably, the coastal areas of Jiangsu and Shandong in eastern China also have high potential indices. In fact, the HSR link between Beijing and Shanghai does not pass through these areas, but rather through traditional corridors further inland. With the Beijing-Shanghai HSR saturated with capacity, a planned second Beijing-Shanghai corridor closer to the coast is on the agenda and is likely to fully exploit these potentials. A similar situation exists in the westernmost parts of Shandong province, which are along the Beijing-Kowloon Railway. Unlike the Beijing-Shanghai corridor to the east and the Beijing-Guangzhou corridor to the west, the Beijing-Kowloon corridor has a much shorter history of completion, and no HSR lines have been built yet. This reflects a certain path dependency in the construction of transportation infrastructure, which may put some late-developing regions in a relatively unfavorable position.

When considering the role of terrain in the potential index (Fig. 6b), it is noticeable that many parts of western China do not have as high a potential, especially the Tibetan Plateau and the Hexi Corridor. On the contrary, the potential of parts of the region between Beijing and Shanghai is much more prominent. This area is highly populated and travels heavily, and if this outstanding potential can be fully realized, it will bring great benefits to the overall transportation carbon emission reduction and transportation efficiency improvement.

Discussion

The patterns of inter-city travel modes in China have undergone substantial shifts, driven by infrastructural developments and evolving travel preferences. The contribution of our study is to unpack the changes in transportation modes over the years prior to the COVID-19 pandemic on fine spatial units. Our analysis across three key dimensions—modal share evolution, spatial dynamics, and regional inequality—provides a comprehensive understanding of these changes.

A key finding is that the process of substitution of road travel modes manifested itself more spatially than in aggregate. In the past decade, China’s HSR lines has begun to expand deeper into the interior44. All of China’s provincial capitals have been covered by HSR, except for Lhasa, a city located on the Tibetan Plateau and far from other provinces. The characteristics of this spatial transformation align closely with policy directives. For instance, China’s 13th Five-Year Plan for the development of a modern integrated transportation system (2016–2020) emphasizes the expansion of network coverage. The plan aims for HSR to serve over 80% of the urban population in cities with more than one million residents. This highlights a strategic shift towards enhancing HSR coverage in previously underserved inland areas. As infrastructure expands remarkably across these regions, many inland areas are transitioning from road to rail as their primary mode of inter-city transportation. However, the overall share of rail travel in the country was still much lower than that of road travel, due to the fact that the coastal mega city regions were still dominated by road travel, which occupy a large amount of population and industry in small areas45. If road travel is expected to be substituted in these areas, it may be necessary to consider alternatives to HSR that are better suited to short distances, such as inter-city rail transits46. Through collaboration between local governments, this type of inter-city infrastructure is actually emerging in China’s mega city regions47.

Differences in the power-law decay with distance of different transportation modes are ultimately presented as their advantages at different distances. Road travel and rail travel are considered to have fierce competition in short to medium distances48. We find that the range of distances in which rail travel has an advantage has expanded. The low value of this range is decreasing further to squeeze the range where road travel is dominant. In fact, another important objective outlined in the 13th Five-Year Plan is to achieve integrated and efficient connectivity. The plan stipulates that core cities within mega city regions and surrounding node cities should be accessible within 1–2 h. Achieving this goal will necessitate the replacement of road travel with HSR to compress travel times. We propose that future efforts to enhance the advantages of rail travel should focus on two key directions. First, to elevate the curve, it is essential to integrate multi-modal rail systems in densely populated areas. Second, efforts should be made to flatten the power-law decay curve of rail travel, making it more competitive than air travel over longer distances. This may require technological advancements, such as ultra-high-speed low-vacuum tube maglev systems. However, due to the uncertainties surrounding the transition from experimental to practical application of these new technologies, the primary focus for enhancing rail travel in the near future may still remain on densely populated mega city regions.

The appraisal of the potential for modal shift can precipitate a contemplation of the balance between efficiency and equity49. The development campaign of China’s western regions, initiated in the late 20th century, has yet to bridge the significant gaps in both infrastructure and population size between the western frontier areas and the eastern regions. Our research indicates that there is still a great potential to change the inter-city transportation mode structure in the ethnic minority areas of China’s borderlands. Further transportation infrastructure investment in these areas may be economically inefficient, but may be beneficial in promoting regional equality, highlighting the existing equality-efficiency tradeoff. Also, despite the small volume of trips, travel distances in these remote areas are generally long, so greater turnover still implies scope for carbon optimization. We recommend that the infrastructure development in western China focus on major transportation corridors, aligning with the Belt and Road Initiative, rather than adopting a scattergun approach to investment. This targeted strategy could yield more meaningful improvements in transportation accessibility and equality in these underserved areas.

In eastern and central China, certain areas remain ensnared by the neglect ingrained in China’s railway skeleton, a legacy of decades past. The entrenched transportation corridors have perpetuated a path-dependent development, steering new infrastructures such as HSR lines along similar routes, thereby exacerbating the Matthew Effect—whereby the already well-connected become more so, and the underprivileged fall further behind50. The layout of the ten vertical and ten horizontal corridors emphasized in the 13th and 14th Five-Year Plans continues to reflect a paradigm based on existing transportation networks. We advocate for a paradigm shift in future planning, urging greater attention towards the regions we have identified as possessing a high Potential Index. This paradigm should be given priority consideration in the eastern and central regions, rather than existing transport hubs or administrative center cities. This approach not only promises to enhance regional integration and economic vitality but also advances social justice by bridging the gap between the haves and have-nots in terms of access to efficient and reliable transportation services.

Although China’s inter-city road travel has been in a transition to other modes, it does not automatically mean that the transition is positive in the sense of carbon footprint reduction. On a per-passenger-kilometer basis, air travel is usually one of the most carbon-intensive modes of transportation51. Also, from a climate justice perspective, aviation is one that disproportionately serves the wealthier segment of the population52. In addition, while HSR is generally regarded as a low-carbon choice, it may unintentionally create additional demand for travel by improving the connectivity of airports, thereby potentially increasing carbon emissions53. The other uncertainty is the promotion of new energy vehicles, which include electric and hybrid electric vehicles. The pre-eminence of these vehicular alternatives has the potential to recalibrate preferences for travel distances on roads, concurrently altering energy use and emissions patterns. Despite such uncertainties, rail travel’s greater capacity and lower per capita energy consumption represent a compelling reason to enhance its modal share. Rail travel, and HSR in particular, is a scalable and energy-efficient solution to inter-city travel that, when properly integrated with emerging transportation systems, can play a leading role in sustainable development.

In conclusion, our investigation provides a comprehensive analysis of the transformation of inter-city transportation modal structures in China. It elucidates both the ongoing changes and potential opportunities. Our analysis emphasizes the multifaceted nature of transportation modal evolution, highlighting the interplay between technological advancements, behavioral shifts, and infrastructure development. This comprehensive perspective is crucial for navigating the complexities of China’s rapidly changing urban and inter-regional dynamics and fostering a more sustainable future for its burgeoning population.