Bureau-specific spatiotemporal carbon footprints and decarbonization pathways of China’s railways

Summary

China’s industrialization and urbanization have driven differentiated railway development across regions, necessitating refined carbon footprint (CF) accounting at the bureau scale to address spatiotemporal heterogeneity beyond national-level analysis. Here we developed a bureau-specific CF model incorporating energy mix, technological efficiency, fleet structure, and transport capacity, paired with driver and efficiency identification models. Results show China’s operational railway emissions dropped 25.5% during 2010–2022 but concentrated in northern bureaus, with Taiyuan (2.78 Mt CO₂-eq), Shenyang (2.75 Mt CO₂-eq), and Zhengzhou (2.52 Mt CO₂-eq) as top emitters in 2022, primarily driven by energy intensity and transport volume variations. Electrification reduced emission intensity (electric locomotives outperforming diesel ones) but relies on low-carbon electricity. Emission efficiency improved alongside declining emissions, with Taiyuan railway bureau ranked first in both total CF and efficiency. These findings emphasize region-specific decarbonization strategies integrating renewable energy, technological upgrading, and operational optimization.

Graphical abstract

Highlights

• •A bureau-specific CF framework was developed for China’s railways
• •In 2022, railway CFs were concentrated in Northern China
• •Coal-based grids weakened electrification benefits in northern bureaus
• •Renewable-rich bureaus gained more from electric traction decarbonization

Applied sciences; energy transportation

Introduction

Climate change mitigation has become a global priority, with energy conservation and emission reduction emerging as critical strategies.¹ The transportation sector, as one of the largest contributors to greenhouse gas (GHG) emissions in China, plays a pivotal role in achieving national decarbonization goals. To advance its emission peaking and neutrality goals, China has adopted clear energy efficiency and emission mitigation objectives, including a 5% reduction in transportation-sector CO₂ intensity by 2025 relative to 2020. Among all transport modes, railway is widely recognized as the most eco-friendly approach, offering substantial mitigation potential. Depending on occupancy rates, railways can reduce carbon emissions by 10%–60% compared with road vehicles and by 46%–73% compared with air transport on a per passenger-km basis.²

Between 2010 and 2022, China’s railway network experienced remarkable growth in both total operating mileage and electrification coverage. In 2010, the national railway system spanned 90,699 km, with electrified lines accounting for approximately 49% of the network. By 2022, the network had further expanded to 154,907 km, and nearly 74% of the entire system was electrified. This infrastructure growth was accompanied by changes in transport volume. The number of passengers transported rose to 3.66 billion passenger journeys in 2019 from 1.68 billion passenger journeys in 2010, but declined sharply to 1.673 billion passenger journeys in 2022 due to the influence of the COVID-19 pandemic. In contrast, freight transport steadily increased from 3.63 billion tons in 2010 to 4.93 billion tons in 2022. The continued growth in total operating mileage (Figure S1) reflects the strategic effort to meet rising freight demand through lower-carbon transport capacity.

As the most electrified subsector of transportation, railways stand to benefit from the growing penetration of renewable energy in China’s power mix. However, the degree of electricity decarbonization varies substantially among regions, leading to marked differences in upstream emission intensities across railway bureaus. Southeastern provinces benefit from higher renewable energy penetration and advanced energy efficiency, resulting in lower electricity-related emissions for railway operations. In contrast, northwestern and northern provinces, where coal remains the dominant electricity source, experience much higher emission intensity in railway traction.^3,4,5 These regional disparities in power mix, energy consumption efficiency, and transportation structure together produce significant heterogeneity in the carbon footprints (CFs) of different railway bureaus. The environmental benefits of railway electrification are highly sensitive to the carbon intensity of regional power grids.¹ Therefore, accurate accounting of railway GHG emissions must integrate both direct operational emissions and indirect upstream emissions arising from regional variations in energy production and utilization efficiencies.

Life cycle assessment (LCA) is a key tool for carbon accounting in the railway system, with several methodological variants developed for different analytical purposes. Globally, LCA methods have been widely applied to assess the environmental impacts of railway systems. Chester and Horvath (2010) conducted a cradle-to-grave LCA of the planned California railway and highlighted that the environmental payback period is highly sensitive to the electricity mix.⁶ In Spain, Bueno et al. (2017) applied a simplified LCA to the Basque Y high-speed rail, finding its operational carbon intensity (32 g CO₂ per passenger-km) is lower than cars (138 g CO₂ per passenger-km) and air transport (127 g CO₂ per passenger-km). However, the line’s annual passenger volume (under 2.5 million) restricts operational emission reduction benefits, which fail to offset its life cycle carbon costs.⁷ Annadanam and Kota (2019) conducted an LCA combining top-down fuel data with bottom-up regional analysis to estimate the Indian national railway emissions during 1970–2015, and found approximately 50% lower CO₂ emissions than road transport.⁸ In Brazil, Soares et al. (2024) applied an operation-phase LCA to the Northeast Corridor railway system, identifying diesel locomotives as the primary emission source for about 83% of emissions.⁹

In China, the rapid expansion of the railway network has spurred a growing interest in evaluating environmental footprints through LCA, with most studies focusing on high-speed railway systems. Yue et al. (2015) carried out a process-based LCA for the Beijing-Shanghai railway line using the Chinese LCA data and found that the operation stage dominates life cycle emissions, contributing around 86% of total GHG emissions.¹⁰ Lin et al. (2019) further applied a hybrid EIO-LCA approach to evaluate the carbon footprint of the same corridor, incorporating indirect upstream emissions. Their results showed that the operation phase contributes roughly 71% of total life cycle CO₂ emissions.¹¹ Another comprehensive study by Wang et al. (2021) developed the Tsinghua-LCA model for this line and confirmed its significantly lower emission intensity compared with road and air transport, with over 90% of emissions originating from the operation phase.⁴ Beyond high-speed railway, Ren et al. (2022) analyzed operational emissions at a regional railway bureau, showing electricity and diesel as dominant sources and emphasizing the influence of regional energy structures.¹² Lu et al. (2023) conducted a comparative LCA of multiple intercity transport modes and found that railway reduces carbon emission intensity by about 76%–82% relative to private cars and 11%–54% compared with diesel buses.¹³ Mao et al. (2024) compared well-to-wheel carbon emission factors between railway and aviation, confirming the former emits only 1/4 to 1/3 as much CO₂ per passenger-km as the latter.¹⁴ Collectively, these studies demonstrate the expanding application of LCA in China’s railway sector, though most remain confined to specific lines or regions rather than providing a comprehensive assessment.

Although the carbon emission accounting standards for railway transportation line provide a standardized framework applicable nationwide, they are inherently macro-oriented and fails to account for regional-scale differences and spatiotemporal variations in railway emissions. As a result, it cannot effectively guide carbon management across different railway bureaus. Current research still faces three main deficiencies: (1) the absence of fine-grained, bureau-level energy emission factors, (2) insufficient exploration of bureau heterogeneity in emission characteristics, and (3) limited understanding of how multiple disparities influence total CF. Addressing these gaps is crucial for establishing a more accurate and regionally differentiated carbon accounting framework for China’s railway system.

In this paper, we address these research gaps by developing a refined, bureau-specific carbon accounting framework for China’s national railway system. Unlike conventional national-scale methods, this framework incorporates regional energy mixes, heterogeneity in locomotive technology and spatial differences in upstream emission intensities, using bureau-level panel data compiled from official railway yearbook (Figure 1), including Harbin, Shenyang, Beijing, Taiyuan, Hohhot, Zhengzhou, Wuhan, Xi’an, Jinan, Shanghai, Nanchang, Guangzhou, Nanning, Chengdu, Kunming, Lanzhou, Urumqi, and Qinghai–Xizang Railway Bureau covering the period from 2010 to 2022. These bureaus collectively span a wide geographic range, covering multiple provinces and managing a significant portion of the national railway mileage, which makes them broadly representative for assessing the integrated operational and environmental performance of China’s railway system. The time period is a critical window that encompasses the rapid development of China’s railway sector, enabling the capture of dynamic changes in the operational CFs of railway bureaus and the evaluation of early emission reduction efforts following the announcement of the national carbon neutrality commitment.

Figure 1.

Spatial distribution of China’s 18 railway bureaus overlaid with the 2022 railway network

Each bureau’s administrative jurisdiction is distinguished by a distinct color block with provincial boundaries and national boundaries labeled as geographic references to delineate the scope of each jurisdiction (not strictly aligned with provincial boundaries). The legend provides clear identification for the following elements: provincial boundaries, national boundaries, railways operating at 200 km/h and above, railways operating at 300 km/h and above, operating lines under the jurisdiction of each railway bureau, and boundary points of China State Railway Group Co., Ltd. (hereafter referred to as “China’s Railway Bureau” in the text). The blue arrow indicates north. Scale bar: 500 km (minor ticks = 100 km).

By integrating both diesel and electric locomotive categories and embedding bureau-specific upstream energy mix and efficiency parameters, our framework captures the spatiotemporal heterogeneity of railway emissions with greater precision. The main contributions of this study are 3-fold: (1) we establish a set of fine-grained, bureau-level emission factors, enabling accurate carbon accounting across multiple management jurisdictions; (2) we systematically analyze inter-bureau heterogeneity in operational emission structures, identifying key drivers linked to regional energy composition and locomotive technology; and (3) we evaluate how regional disparities influence carbon emission efficiency by adopting a life cycle perspective that integrates both environmental and economic performance, providing a foundation for efficiency-oriented and region-specific decarbonization strategies.

Results

Energy consumption carbon footprint factors

Nationally, diesel consumption CF factors declined from 0.09800 kg CO₂-eq/MJ in 2010 to 0.09729 kg CO₂-eq/MJ in 2022, due to the efficiency improvements in diesel production. The bureau-scale diesel consumption CF factors exhibited heterogeneity (Figure 2), with generally higher values in the eastern and southern regions and comparatively lower levels in the northern and central regions by 2022. The lowest factor was recorded in the Taiyuan bureau (0.09353 kg CO₂-eq/MJ) and the highest in the Qinghai–Xizang bureau (0.09827 kg CO₂-eq/MJ), as shown in Figure 2. Over 2010–2022, the diesel consumption CF factors remained relatively stable across several bureaus located in Eastern and Southern China, including Beijing (0.09781 kg CO₂-eq/MJ), Jinan (0.09815 kg CO₂-eq/MJ), Shanghai (0.09822 kg CO₂-eq/MJ), Guangzhou (0.09819 kg CO₂-eq/MJ), and Nanning (0.09837 kg CO₂-eq/MJ) bureaus. These values are the multi-year means and indicate limited improvement in diesel production efficiencies. Diesel consumption emission factors for Harbin, Shenyang, Beijing, Taiyuan, Hohhot, Wuhan, Xi’an, Shanghai, Nanchang, Chengdu, Lanzhou, and Urumqi declined. In particular, Taiyuan, Hohhot, and Chengdu bureaus recorded significant reductions by 4.6%, 2.6%, and 2.6%, respectively. These reductions imply more efficient diesel production, especially for the coal-to-liquids (CTL) process improvements in Shanxi and Inner Mongolia, where Taiyuan and Hohhot bureaus are located, respectively.¹⁵ Conversely, in Zhengzhou, Jinan, Guangzhou, Nanning, Kunming, and Qinghai–Xizang bureaus, the diesel consumption CF factor increased slightly by 1.12%, 0.7%, 0.05%, 0.45%, 0.17%, and 0.63%, respectively. This mild upturn can be explained by region-specific differences in diesel production and supply structures. Some of these regions increased the use of CTL diesel, which has a higher cradle-to-gate CF than petroleum-based diesel because coal gasification and liquefaction are highly energy-intensive and feature low conversion efficiency. Meanwhile, relatively backward refining technologies and limited upgrading investments in local small and medium-sized refineries failed to keep pace with national efficiency improvements, resulting in lower production efficiency and higher emission intensities. Also, for bureaus located in plateau or mountainous terrain such as Qinghai–Xizang and Kunming, long-distance fuel transportation, and inadequate storage conditions further elevate upstream emissions from fuel delivery and distribution.

Figure 2.

Carbon footprint factors of diesel and electricity consumption (kg CO₂-eq/MJ) in the operational phase of China’s railway bureaus (2010–2022)

In each chart, pink bars represent the diesel consumption carbon footprint factor and pale blue bars represent the electricity consumption carbon footprint factor.

In comparison, the national electricity consumption CF factors were substantially higher but decreased from 0.2152 kg CO₂-eq/MJ in 2010 to 0.1561 kg CO₂-eq/MJ in 2022, due to electricity generation efficiency improvements and structure optimization. At the bureau scale, most electricity consumption CF factors ranged from 0.15 to 0.27 kg CO₂-eq/MJ, which were approximately two to three times the corresponding diesel consumption CF factors. This indicates that using the direct emission boundary will greatly overestimate the emission reduction benefits brought by electrification. All bureau-scale electricity consumption CF factors exhibited decreases, among which Kunming bureau experienced the largest reduction at 62%, while Wuhan bureau showed the smallest reduction by 3.7%. The multi-year average electricity consumption CF factors for the Wuhan (0.1395 kg CO₂-eq/MJ), Nanning (0.1243 kg CO₂-eq/MJ), Chengdu (0.1082 kg CO₂-eq/MJ), Kunming (0.0621 kg CO₂-eq/MJ), and Qinghai–Xizang (0.0676 kg CO₂-eq/MJ) bureaus were consistently low. This can be attributed to their favorable locations, where low-carbon energy sources such as hydropower, solar, and wind energy were rapidly developed. In contrast, the Taiyuan and Hohhot bureaus consistently exhibited the highest electricity consumption CF factors, with multi-year mean values of 0.2301 kg CO₂-eq/MJ and 0.2262 kg CO₂-eq/MJ, respectively. Their reliance on coal-based electricity, driven by the abundance of coal resources in Northern China, resulted in higher indirect emission intensities from electric traction.

Carbon footprint intensity of locomotives

From 2010 to 2022, the CF intensity of electric locomotives exhibited general downward trends across most railway bureaus, ranging from a 3% reduction in the Shanghai bureau to a 69% reduction in the Urumqi bureau (Figure 3). This trend reflects the combined effects of improvements in electric locomotive efficiency and the decarbonization of the electricity system. The Qinghai–Xizang bureau consistently recorded the lowest CF intensity for electric locomotives (24.79 kg CO₂-eq per 10⁴ t-km in 2022). This is because the region is rich in renewable energy, keeping its electricity carbon footprint factor at an extremely low level from 2010 to 2022. Although the Qinghai–Xizang bureau maintained the lowest CF intensity for electric locomotives in 2022, it was the only bureau to experience an 18% increase over the study period. This is mainly due to the growing energy demand for plateau-specific railway operations, such as onboard oxygen supply, cabin heating and anti-freezing systems. In addition, the complex terrain and high altitude of the region increase the traction energy consumption of electric locomotives, while the relatively low transport workload further limits the improvement of energy efficiency. In contrast, the Xi’an bureau reported the highest electric traction CF intensity (99.44 kg CO₂-eq per 10⁴ t-km in 2022). Except for the Harbin and Qinghai–Xizang railway bureaus, the CF intensity trends of electric locomotives closely mirrored those of total locomotives across all other bureaus, with minimal differences between the two (1.95–10.3 kg CO₂-eq per 10⁴ t-km). This alignment suggests that the carbon footprint of electric locomotives plays a dominant role in shaping overall carbon intensity patterns. Alongside Xi’an, the Urumqi (83.6), Lanzhou (70.2), Shanghai (68.7), and Zhengzhou (68.2 kg CO₂-eq per 10⁴ t-km) bureaus also exhibited high electric locomotive CF intensity. In Xi’an, Urumqi, and Lanzhou, this is largely attributable to mountainous routes and curved tracks, which increase traction energy demand. Additionally, these bureaus exhibit high electricity CF factors as observed for Xi’an (0.191 kg CO₂-eq/MJ), Urumqi (0.187 kg CO₂-eq/MJ), and Zhengzhou (0.183 kg CO₂-eq/MJ), which further contributes to elevated upstream emissions and increased CF intensity. In Shanghai and Zhengzhou, heavy freight operations predominantly rely on electric locomotives, and the energy consumption (see Figure S3) and CF intensity of freight trains are considerably higher than those of high-speed passenger trains, driving up electric locomotive CF intensity in these regions.¹⁶

Figure 3.

Temporal trends in carbon footprint intensity of locomotive operations by railway bureau (2010–2022)

Each graph represents one of the 18 railway bureaus, showing annual CF intensity per unit transport workload (kg CO₂-eq per 10⁴ t-km). Red lines indicate CF intensity for diesel locomotives, blue lines for electric locomotives, and black lines represent the comprehensive CF intensity for all locomotives. The metric reflects the operational carbon footprint intensity across different locomotive types and regions.

From 2010 to 2022, the CF intensities of diesel locomotives in central regions were comparatively higher than those in other areas. This pattern is largely attributed to regional geographic features, which affect the availability and deployment of low-carbon energy sources. Diesel locomotives consistently exhibited much higher CF intensities than electric locomotives. Although diesel fuel generally has a lower carbon emission factor, the substantially lower energy conversion efficiency of diesel locomotives offsets this advantage, resulting in higher overall carbon intensities compared with electric locomotives.^17,18 In 2022, the Lanzhou bureau recorded the highest diesel-powered CF intensity of 911.8 kg CO₂-eq per 10⁴ t-km, while the Harbin bureau showed the lowest of 113.4 kg CO₂-eq per 10⁴ t-km. In the Harbin (10.8%), Shenyang (41.9%), Beijing (9.0%), Hohhot (8.4%), Xi’an (34.1%), Shanghai (55.3%), Nanchang (27.3%), Guangzhou (10.3%), Nanning (41.2%), Chengdu (24.4%), and Qinghai–Xizang (32.0%) bureaus, the CF intensities of diesel locomotives increased steadily over time. A much sharper increase was observed in the Taiyuan (347.3%), Zhengzhou (710.6%), Wuhan (147.2%), Jinan (108.8%), Kunming (117.5%), and Lanzhou (647.0%) bureaus, indicating a rapid deterioration in energy efficiency or operating conditions for diesel locomotives. This is primarily due to the concentration of remaining diesel operations in less efficient routes characterized by challenging terrain, low load factors, aging equipment, and limited technological upgrades. In contrast, the Urumqi bureau experienced a slight reduction (1.8%) in diesel-powered CF intensities.

The CF intensities of electric locomotives decreased steadily, whereas those of diesel locomotives either remained high or increased in specific bureaus. This divergence highlights a deep structural transition in China’s railway system: as electrification expands, residual diesel locomotives are increasingly concentrated in hard-to-decarbonize bureaus, revealing both the progress made and the persistent challenges in achieving nationwide railway decarbonization.

Overall, the national average CF intensities of locomotives declined by 26% during 2010–2022. At the bureau scale, the Qinghai–Xizang bureau achieved the largest average CF intensity reduction by 62.4%, whereas the Shanghai bureau exhibited the smallest decrease by 8.4%. With the continued expansion of electrified railways, the average CF intensities became increasingly influenced by the CF intensities of electric traction and exhibited close trends. In 2022, the Xi’an bureau recorded the highest average CF intensity of 102.7 kg CO₂-eq per 10⁴ t-km, while the Kunming bureau reported the lowest at 31.34 kg CO₂-eq per 10⁴ t-km.

Carbon footprint of railway bureaus

In China, railway operational CFs declined from 41.2 Mt CO₂-eq in 2010 to 30.7 Mt CO₂-eq in 2022. The most significant decline occurred during 2010–2016, coinciding not only with large-scale electrification and the deployment of more energy-efficient locomotives but also with a sharp decline in diesel locomotive activity, which dropped from 1.61 trillion t-km in 2010 to 0.36 trillion in 2022. Meanwhile, the workload of electric locomotives showed an overall upward trend despite some fluctuations, rising from 2.96 trillion t-km in 2010 to 4.26 trillion in 2022 (see Figure S2). These combined changes in traction structure and operational efficiency contributed to the observed CF reduction. After 2017, however, CFs stabilized with minor fluctuations, suggesting that the early benefits of electrification and energy restructuring had largely been realized.

The CFs revealed distinct spatiotemporal heterogeneity among 18 railway bureaus, as shown in Figure 4. Hohhot (57%), Nanning (49%), Kunming (44%), Shenyang (41%), and Nanchang (41%) bureaus exhibited substantial CF declines over time, driven by a marked reduction in diesel locomotive transport volume, a rapid shift toward electric traction and improvements in operational efficiency. By contrast, the Urumqi (38%) and Xi’an (11%) bureaus showed upward trends in carbon footprint, corresponding to increased freight activities in Western China. In Urumqi, the electric locomotive workload expanded more than 700-fold, rising from 0.32 billion t-km in 2010 to 236.29 billion t-km in 2022, while diesel locomotive workload decreased by 106.63 billion t-km. Similarly, Xi’an experienced a substantial increase in electric transport volume, which more than offset the reductions from declining diesel locomotive emissions. Despite improvements in energy efficiency, the rapid growth in electric traction demand led to a net increase in overall operational carbon footprints in these regions. The Qinghai–Xizang bureau maintained stable operational CFs, attributable to its relatively sparse population, less developed economy and challenging geographic conditions. High construction and operational costs, along with harsh weather, further restrict expansion and usage, keeping operational emissions low and stable.¹⁹ In 2022, the CFs were geographically concentrated in Northern China, particularly in the Taiyuan (9.06%), Shenyang (8.94%), and Zhengzhou (8.20%) bureaus, due to their extensive traffic volumes. In contrast, the smallest CFs were observed in the Kunming (0.84%), Qinghai–Xizang (2.02%), and Hohhot (2.19%) bureaus, where the transport demand was comparatively low.

Figure 4.

Total operational carbon footprints of 18 railway bureaus from 2010 to 2022

Each subfigure is a radial stacked bar chart presenting the total operational carbon footprint of an individual railway bureau in China from 2010 to 2022, with data displayed in a counterclockwise direction. The pale blue segments represent the carbon footprint from electric locomotives, while the red segments denote that from diesel locomotives.

Structurally, the sources of CFs have undergone fundamental changes from 2010 to 2022. In 2010, diesel traction was the dominant source of CFs in the Harbin (91.7%), Shenyang (57.6%), Shanghai (65.2%), Nanchang (59.1%), Nanning (78.3%), Urumqi (99.5%), and Qinghai–Xizang (98.4%) bureaus, all accounting for 43.0% of national railway CFs. With the steady expansion of electrified lines and retirement of inefficient diesel fleets, the share of diesel-related CFs declined to 17.8% by 2022. Electric traction has increasingly become the main contributor to railway CFs. However, diesel traction still dominated CFs in the Harbin (52.46%) and Qinghai–Xizang (60.46%) bureaus, where the rates of railway electrification remain relatively low. In the Harbin bureau, the harsh climate reduces the operational reliability of electric locomotives and increases the maintenance requirements of power equipment; in the Qinghai–Xizang bureau, the complex terrain and high-altitude conditions weaken the traction performance of electric locomotives and raise construction and operation costs. These natural and technical restraints hinder the progress of electrification in both bureaus, resulting in continued dependence on diesel traction to maintain transport stability and efficiency.

Key drivers for carbon footprint change

Nationally, the energy consumption CF factor and transport proportion reduced CFs by 7.470 and 7.420 Mt CO₂-eq, respectively. In contrast, the energy consumption factor and transport volume contributed to the CF increases by adding 4.016 and 0.354 Mt CO₂-eq, respectively. Despite these national-scale effects, the relative contributions of each driver varied considerably across individual railway bureaus, as shown in Figure 5.

Figure 5.

Operational carbon footprint (CF) changes of 18 railway bureaus in China (2010–2022) and their driver-specific contributions

Red and purple bars represent the operational CF of the 18 railway bureaus in 2010 and 2022, respectively. Blue (ΔCF_ECCFF), yellow (ΔCF_ECF), green (ΔCF_TV), and pink (ΔCF_TP) bars denote CF changes driven by the following factors: (1) energy consumption CF factor; (2) energy consumption; (3) transport volume; and (4) transport proportion (defined as the ratio of electric locomotive workload to diesel locomotive workload).

Energy consumption CF factor (ECCFF) consistently exerted a negative contribution across all 18 railway bureaus, reflecting an overall low-carbon electricity trend, largely driven by a cleaner power mix and railway electrification. It acted as the dominant driver of CF reductions in the Beijing (−59%), Taiyuan (−72%), Zhengzhou (−84%), Jinan (−154%), Guangzhou (−64%), Chengdu (−114%), Kunming (−128%), and Lanzhou (−156%) bureaus. It generated the most substantial CF reductions in the Taiyuan, Shenyang, and Beijing bureaus, with carbon abatement volumes of 0.901, 0.810, and 0.808 Mt CO₂-eq, respectively.

Across most bureaus, energy consumption factor (ECF) played a predominantly positive role in CF increases. The largest CF increases attributable to ECF were observed in the Shanghai (0.772 Mt CO₂-eq), Lanzhou (0.622 Mt CO₂-eq), and Zhengzhou (0.485 Mt CO₂-eq) bureaus, indicating limited improvements in energy efficiency. In contrast, ECF contributed to a decarbonization effect in the Urumqi bureau, reducing emissions by 0.315 Mt CO₂-eq. Furthermore, ECF was the dominant driver of CF changes in the Shanghai bureau (100%), indicating a local deterioration in energy performance.

Transport volume (TV) drove both promoting and restraining effects across different bureaus. The negative CF changes in the Harbin (−19%), Shenyang (−24%), Beijing (−17%), Hohhot (−68%), Zhengzhou (−5%), Shanghai (−65%), Nanchang (−42%), Guangzhou (−42%), and Nanning (−35%) bureaus were caused by it. The largest CF reduction exhibited in the Hohhot bureau (−0.597 Mt CO₂-eq) was due to a transport volume reduction from 188.75 billion t-km to 106.92 billion t-km. Conversely, this driver played a large role in CF increases in the Taiyuan (0.146 Mt CO₂-eq), Wuhan (0.242 Mt CO₂-eq), Xi’an (0.721 Mt CO₂-eq), Jinan (0.259 Mt CO₂-eq), Chengdu (0.0823 Mt CO₂-eq), Kunming (0.119 Mt CO₂-eq), Lanzhou (0.152 Mt CO₂-eq), Urumqi (0.842 Mt CO₂-eq), and Qinghai–Xizang (0.544 Mt CO₂-eq) bureaus. In particular, it contributed the largest CF increases in Urumqi with 0.842 Mt CO₂-eq, reflecting the impact of strong railway activity growth.

In most bureaus, transport proportion (TP) had a CF-reducing effect, signifying a positive structural transition toward low-carbon electric traction. It was the dominant driver of CF reductions in the Harbin (−0.604 Mt CO₂-eq), Shenyang (−0.89 Mt CO₂-eq), Wuhan (−0.591 Mt CO₂-eq), Nanchang (−0.548 Mt CO₂-eq), Nanning (−0.526 Mt CO₂-eq), and Qinghai–Xizang (−0.690 Mt CO₂-eq) bureaus. The largest TP-attributable carbon reductions were observed in Shenyang, Taiyuan (−0.703 Mt CO₂-eq), as well as Qinghai–Xizang, where emissions fell thanks to the aggressive expansion of electrified rail routes. However, the Urumqi bureau was an exception: this driver contributed a 15% increase in CFs (+0.271 Mt CO₂-eq), stemming from the higher energy consumption factors of electric locomotives.

Carbon footprint efficiency measurements

China’s railway operation exhibited a steady upward trend of CF efficiency from 2010 to 2022. However, significant bureau-scale disparities were observed in Figure 6. In 2022, the Taiyuan bureau achieved the highest CF efficiency (1.01). Under the super SBM-DEA model, a score greater than 1 indicates super-efficiency, meaning that the bureau serves as an efficiency benchmark for other railway bureaus. Guangzhou (0.81), Hohhot (0.69), and Xi’an (0.66) bureaus ranked next, and all these top-performing bureaus benefited from sustained electrification upgrades and modernized rail systems. In contrast, the Qinghai–Xizang (0.17), Lanzhou (0.18), and Shenyang (0.18) bureaus exhibited the lowest CF efficiencies. This heterogeneity is due to various factors. For example, prolonged extreme cold in Northeast China (such as the Harbin and Shenyang bureaus) forces engines and heating systems to run continuously during winter, raising energy consumption and reducing equipment performance.^19,20,21 High-altitude routes face a similar challenge. For example, trains operating on the Qinghai-Xizang plateau require supplemental oxygen for passengers, and the onboard oxygen supply system consumes roughly five times the energy of a standard air-conditioning system.^22,23 Local energy endowment further contributes to these differences. The Kunming bureau benefits from abundant hydropower, thereby reducing electricity-related CFs.^24,25 Economic development also plays an important role in CF efficiency. The Shanghai and Guangzhou bureaus are located in highly developed, densely populated cities with advanced electrified rail infrastructure and thus yield higher CF efficiencies than the older industrial cities.^4,19,26,27 Hubs such as the Shanghai, Guangzhou, and Taiyuan bureaus are positioned at the intersection of multiple busy freight corridors that handle bidirectional traffic with high transport demand and efficiency. However, the bureaus in Northeast China predominantly haul unidirectional bulk commodities, leading to high rates of empty return trips.

Figure 6.

Super efficiency scores of operational carbon emission efficiency for China’s 18 railway bureaus (2010–2022)

Each cell in the figure presents the super efficiency score (scaled 0–1.1) of operational carbon emission efficiency for 18 China’s railway bureaus during 2010–2022, which is calculated using the super SBM-DEA model. The color intensity of the cells corresponds to the efficiency level: darker red indicates a higher efficiency score (approaching or even exceeding 1), while darker blue indicates a lower efficiency score (closer to 0).

Furthermore, Operating revenue also helps explain regional differences in efficiency. In 2022, the Taiyuan, Guangzhou, and Xi’an bureaus reported the highest revenues, supported by substantial freight or passenger volumes and favorable economic conditions.^19,28 Taiyuan railway bureau benefits from heavy freight throughput and its strategic position along key corridor routes, while Guangzhou railway bureau capitalizes on dense passenger flows and diversified revenue streams.^19,28 In contrast, the Qinghai–Xizang, Kunming, and Nanchang bureaus recorded the lowest revenues, potentially constraining their capacity to enhance carbon emission efficiency. Labor force size also contributes to efficiency heterogeneity. In 2022, Qinghai–Xizang, Kunming, and Urumqi bureaus reported the smallest employee numbers while Shenyang, Beijing, and Shanghai had the largest. Smaller workforce size may reduce per capita resource consumption and enhance carbon emission efficiency, whereas labor-intensive operations can elevate emissions per unit of output.

Discussion

Our bureau-scale analysis of China’s railway carbon footprint confirms electrification as an effective core measure for transportation decarbonization. Identifying obvious spatial emission heterogeneity also makes national-level decarbonization research more applicable to actual railway operation, which is critical for designing targeted low-carbon policies in China. The 25.5% national reduction in railway operational emissions from 2010 to 2022 indicates electrification can effectively lower emission intensity, which is consistent with global research on the environmental advantages of electric traction and practical evidence for advancing electrification in the transport sector. But clear spatial divergences exist in China’s railway emission levels and decarbonization effects: northern bureaus remain high emitters, while southwestern and southeastern bureaus have achieved more significant emission cuts. The root cause is the unbalanced development of low-carbon power resources across regions. Northern bureaus rely heavily on coal-fired electricity which weakens electrification’s emission reduction effects. By contrast, southwestern and southeastern bureaus leverage abundant renewable energy to maximize the environmental benefits of electric traction. These findings fully show that a one-size-fits-all strategy is not suitable for China’s railway decarbonization and sectoral emission reduction measures must be combined with regional energy transition progress and resource endowments. This also points to the need for differentiated and targeted policy design for China’s railway decarbonization, with measures tailored to the actual conditions of different railway bureaus and regions.

Given that China’s power sector decarbonization is governed by provincial-level power structures, grid dispatch, and central-provincial policy coordination, a natural concern is whether adopting railway bureaus as the spatial unit would mask underlying provincial energy heterogeneity. To examine whether the bureau-level accounting framework (Figure 7) obscures these provincial-level influences, we conducted an additional one-at-a-time sensitivity analysis by reducing province-level electricity consumption factors and corrected diesel upstream production factors by 5%, while holding all other parameters constant. The results, provided in the supplemental information (Figures S4 and S5), show that bureau-scale carbon footprints are particularly sensitive to electricity decarbonization in provinces where railway operating mileage is highly concentrated. For example, the largest sensitivity coefficient was observed for Jinan-Shandong (0.879), followed by Zhengzhou-Henan (0.843), Xi’an-Shaanxi (0.820), Taiyuan-Shanxi (0.814), and Urumqi-Xinjiang (0.793). A 5% reduction in the corresponding provincial electricity factors decreased the associated bureau-scale operational carbon footprints by 71.1, 106.2, 96.2, 113.2, and 96.2 million kg CO₂-eq, respectively. By contrast, the strongest responses to cleaner diesel upstream production were observed for Qinghai–Xizang-Qinghai and Harbin-Heilongjiang, with sensitivity coefficients of only 0.099 and 0.065, corresponding to carbon footprint reductions of 3.08 and 6.26 million kg CO₂-eq under a 5% reduction in corrected diesel upstream production factors. These results indicate that the bureau-level perspective does not ignore provincial energy-system heterogeneity; rather, it captures how province-level decarbonization conditions are transmitted to railway operating organizations through their cross-provincial mileage structure. Therefore, bureau-level assessment provides additional policy value by linking provincial energy transition conditions with railway-sector operational management, complementing rather than replacing provincial-level carbon footprint assessment.

Figure 7.

System boundary for railway operational carbon footprint accounting

System boundary of energy-related carbon footprint accounting for railway locomotive operation. The diesel pathway includes upstream diesel production, represented by coal-to-liquids and crude oil refining, followed by diesel combustion and consumption by diesel locomotives. The electricity pathway includes electricity generation from coal, oil, natural gas, hydropower, nuclear, wind, biomass, solar, and geothermal sources, followed by electricity transmission and consumption by electric locomotives. Greenhouse gas (GHG) emissions are accounted for across the energy production and energy consumption stages associated with railway operation.

Direct carbon emissions from China’s railways peaked in the 1980s. They subsequently declined as electrification expanded and energy efficiency improved.^29,30 Railway electrification is broadly recognized as a crucial strategy for reducing carbon emissions in the transportation sector and advancing toward carbon neutrality.^5,29,31,32 However, our analysis based on a broader system boundary revealed its limited effectiveness when rapidly implemented in regions where the power supply remains coal-dominated. Specifically, electrification delivered limited CF reductions in regions such as Taiyuan (mainly in Shanxi) and Hohhot (mainly in Inner Mongolia) railway bureaus.^33,34 Therefore, accelerating the electricity transition to renewable sources and expanding cross-regional electricity transfers is a matter of higher priority that can ensure all railway bureaus access low-carbon electricity.^5,34,35,36

Moreover, regional differences in energy structure, resource endowment, and economic development drive disparities in carbon footprints and decarbonization potential,^33,34,37,38 region-specific policy responses are necessary. For hard-to-electrify regions (challenging terrain or harsh climates), alternative solutions should be taken including the development of cleaner diesel. For example, developing biomass diesel derived from abundant agricultural residues such as straw in Eastern China could simultaneously address rural pollution from straw burning.^39,40 In addition, although low-emission diesel is used in extremely cold regions, the harsh environment leads to increased fuel consumption due to reduced engine efficiency, additional energy demands for heating and anti-freezing systems, and higher traction resistance from icy tracks and dense cold air. These factors collectively offset the emission reduction benefits of low-carbon fuels, resulting in higher overall carbon emissions for diesel locomotives in extremely cold areas compared to mild regions.^40,41 A phased transition approach is recommended to address this issue. Hybrid locomotives can serve as an intermediate solution while new energy locomotives are gradually promoted.^42,43 For non-electrified railways, focus should be placed on the research and development of fuel cell vehicles, as well as the advancement of battery technologies.^44,45 As hydrogen production costs decline, the viability of hydrogen-powered locomotives is growing.^46,47,48 Therefore, prioritizing the promotion of hydrogen fuel locomotives in these areas may be a more feasible decarbonization pathway than grid-connected electrification alone.^41,49,50 This pathway, however, requires the deployment of core infrastructure including hydrogen refueling facilities, compatible maintenance services, and safety monitoring systems.^39,49,51

Furthermore, our research highlights substantial variations in carbon intensity among railway bureaus, largely due to differences in energy utilization efficiency. Bureaus with high energy consumption factors, such as Qinghai-Xizang railway bureau (with high diesel consumption emission factor) should prioritize investments in oil refining technologies and adopt energy-saving technologies. Improving operational management is another critical aspect of energy conservation, including integrated planning of passenger and freight services, optimizing transport scheduling, and leveraging digital tools and advanced locomotive technologies to enhance efficiency and reduce emissions per unit of transport.⁵²

Beyond traditional emission accounting, integrating an input-output efficiency framework based on DEA to conduct a comprehensive efficiency assessment yields broader policy implications. For railway bureaus with large workforce but relatively low revenue, such as Harbin and Shenyang railway bureau, policy support should prioritize two core strategies. One focus is optimizing human resource allocation and upgrading service quality to directly enhance operational efficiency and revenue generation. Another key direction is accelerating the adoption of automation technologies and artificial intelligence to reduce manual labor in routine tasks and further boost operational efficiency. Meanwhile, providing financial support and technical upgrades is essential as a policy backup to facilitate these improvements.

Last but not least, a critical prerequisite for achieving refined railway carbon emission accounting and management is addressing the lag in railway energy consumption informatization. Policy efforts should focus on constructing modern railway information infrastructure and an integrated information platform. This infrastructure will enable systematic collection, real-time monitoring and efficient integration of energy consumption data, laying a solid foundation for accurate carbon accounting and science-driven emission reduction management.^53,54,55

Limitations of the study

In this study, we assume that as diesel transportation is expensive and inconvenient, the diesel consumed by each railway bureau is locally produced. This study also has certain analytical limitations that provide directions for further improvement. First, we did not distinguish between passenger and freight transport, as well as conventional electrified railways and high-speed rail systems in the decomposition of emission drivers, thus ignoring the structural heterogeneity within the railway sector. Second, key operational factors such as load factors and line-specific energy efficiency are not separately analyzed in the current driver framework.

From a qualitative perspective, these unaddressed structural and operational heterogeneities across railway bureaus substantially influence the interpretation of estimated carbon emission patterns and driver decomposition results. Structurally, different railway modes present distinct emission performance: Electrified railways exhibit much lower carbon intensity than diesel traction lines when electricity is low-carbon and systems are efficient.^43,45 High-speed rail outperforms conventional passenger rail in emission efficiency but is sensitive to passenger load factors,^56,57 and heavy-haul freight railways achieve notable low-emission advantages through scale economies.⁵⁸ In addition, passenger and freight transport have incomparable emission benchmarks without unified functional units, further creating inherent structural disparities across regions. Operationally, these mode differences also lead to divergent driving mechanisms of emission reduction. Electrified and high-speed rail rely heavily on grid decarbonization to cut emission factors, while freight and heavy-haul lines depend more on energy efficiency improvement via operational optimization and technical upgrades.

In the Chinese context, such structural heterogeneity is particularly pronounced. Continuous investment in railway infrastructure has greatly improved total transport capacity. However, the development of passenger and freight services shows a clear divergence. High-speed rail has expanded rapidly, while the growth of freight transport has been relatively slow. In 2024, railways accounted for 25% of China’s total interregional passenger traffic, making them the second-largest passenger transport mode. Meanwhile, railways made up 9% of national total freight volume, remaining the largest mode for bulk cargo transport, but its growth rate was lower than that of road freight.

These unaddressed differences in transport structure and locomotive types will lead to clear biases in the LMDI decomposition results. First, failing to distinguish passenger and freight transport leads to unreliable estimates of the energy consumption factor. Passenger and freight services differ greatly in energy use per unit output and overall emission intensity. Aggregating these two types together reduces the observed influence of energy efficiency improvements. The real contribution of the energy consumption factor is therefore underestimated. Second, not separating high-speed rail from conventional electric rail distorts the understanding of structural changes within passenger services. High-speed rail uses more energy per passenger km than conventional electric rail. Mixing these two modes together weakens the model’s ability to capture real efficiency differences. The effect of structural shifts among passenger rail services is also underestimated. Third, omitting the special features of heavy-haul freight transport further reduces the accuracy of decomposition results. Heavy-haul freight achieves much higher energy efficiency through large scale operations. Without recognizing this mode, the model cannot capture real efficiency gains related to scale optimization. The energy consumption factor for freight related activities is thus underestimated.

Therefore, the real major drivers of emission changes including energy efficiency differences between passenger and freight, structural shifts between high-speed and conventional rail, and efficiency improvements in heavy-haul freight cannot be clearly identified from the decomposition results. In the future, passenger demand on major routes is likely to become saturated. Freight volume is expected to keep growing as more goods are shifted from road to rail and as industrial demand rises. This growing share of freight could strongly increase total carbon emissions even if transport efficiency improves.

Furthermore, operational factors such as empty haulage and low load factors make emission intensity higher and affect decomposition results. For example, low seat occupancy on some high-speed lines can significantly increase emissions per passenger. High empty rates in freight services also waste energy and raise emission levels. These internal features, including different emission performances of various transport modes, unbalanced growth between passenger and freight services, and differences in operational efficiency, all play an important role in shaping real emission performance and interpreting decomposition results.

Without detailed classification, the aggregated framework restricts an in-depth understanding of inter-bureau emission disparities and complicates result interpretation, which further restricts the formulation of more targeted decarbonization policies. All these limitations could be effectively addressed in follow-up research if relevant data become available, and such refinements will further improve the explanatory power and policy relevance of the analytical framework.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Junjie Li (jjli1@bjtu.edu.cn).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •All data reported in this paper will be shared by the lead contact upon request.
• •This paper does not report original code.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

We sincerely thank the National Key Research and Development Program of China (grant no. 2023YFC3709500), the Siyuan Postdoctoral Research Initiation Fund (grant no. YB25XKBH00020) and the National Natural Science Foundation of China (grant nos. 42303085 and 42475181).

Author contributions

Conceptualization, X.W. and J.L.; data curation, X.W.; formal analysis, X.W.; funding acquisition, Y.Y. and L.P.; investigation, X.W. and J.L.; methodology, X.W. and J.L.; project administration, J.L. and L.P.; resources, J.L.; software, J.L. and L.P.; supervision, J.L. and L.P.; validation, X.W. and J.L.; visualization, X.W.; writing – original draft, X.W.; writing – review and editing, X.W. and J.L.

Declaration of interests

The authors declare no competing interests.

Declaration of generative AI and AI-assisted technologies in the writing process

During the preparation of this work, the authors used Doubao AI for the proofreading and linguistic polishing. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the article.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Deposited data
China’s railway bureau-scale operational data	China Railway Yearbook (2011–2023)	https://www.shujuku.org/china-railway-yearbook.html
China’s province-level electricity generation volume, energy structure and inter-provincial electricity transmission volume	China Electric Power Industry Yearbook (2011–2023)	https://www.shujuku.org/china-electric-power-yearbook.html
China’s province-level calorific values of refined oil and fossil fuels	China Energy Statistical Yearbook (2011–2023)	https://www.shujuku.org/china-energy-statistical-yearbook.html
default diesel production CF factor	GHG S3.3 China’s Gasoline and Diesel Upstream Emission Factors (2024)	http://lingtan.chinacsrmap.org/wp-content/uploads/2025/03/gzgqyhcysypfyz.pdf
diesel combustion carbon emission factor data	2006 IPCC Guidelines for National Greenhouse Gas Inventories	https://www.ipcc.ch/report/2006-ipcc-guidelines-for-national-greenhouse-gas-inventories/
Software and algorithms
SimaPro 10.2	SimaPro	https://simapro.com/
IPCC GWP 100a	IPCC	https://www.ipcc.ch/assessment-report/ar6/
Other
Ecoinvent database V3.11	Ecoinvent	https://ecoinvent.org/

Experimental model and study participant details

Omitted as our study does not involve biological models, experimental organisms, human participants, plants, microbe strains, cell lines, or primary cell cultures. Sex- or gender-based analyses are therefore not applicable to this computational railway carbon-footprint study.

Method details

Carbon footprint accounting boundary

Our analysis focuses on the operational stage of railway bureaus, given that operational energy CF contributed 71%–86% to the life cycle GHG emissions in the railway sector,^10,11 which makes it the primary driver of climate impact and a stage of prioritizing the deployment of mitigation strategies. The system boundary is defined to include only carbon emissions directly related to locomotive operation, which are categorized into diesel and electric locomotives, as shown in Figure 7 For electric locomotives, the boundary includes the life cycle of electricity from cradle to consumption and incorporates nine types of power generation sources for refined modeling. Non-traction electricity consumption, such as office and station lighting, is excluded from this study. For diesel locomotives, the range encompasses upstream indirect emissions from fossil resources extraction and diesel production, as well as direct emissions from diesel combustion during locomotive operation. In addition to the conventional petroleum refining, we included China’s unique coal-to-oil route to better match the actual emissions. Emissions from diesel transportation and storage are excluded due to limited data availability at the railway bureau level and their negligible contribution to the total life cycle emissions.^9,59

Each railway bureaus typically operate across multiple provinces, and each with distinct energy production structures. To address the misalignments between railway bureau boundaries and provincial administrative divisions, we develop an operational-mileage-weighting method to correct province-specific CF factors to the railway bureau scale. The spatial boundary transcends traditional provincial administrative divisions and is anchored to the actual jurisdiction of each railway bureau, thereby ensuring the final factors accurately reflect the true energy mix and efficiency of the regions where the bureaus operate. We also apply the Logarithmic Mean Divisia Index (LMDI) method to decompose changes in energy consumption and carbon emissions into their key driving factors. In addition, the Slack-Based Measure Data Envelopment Analysis (SBM-DEA) model is employed to identify inefficiencies and provide a comprehensive evaluation of the operational and environmental performance of each railway bureau.

The railway bureau-scale operational data were compiled from China Railway Yearbooks (2011–2023). Province-level electricity generation, energy structure, and inter-provincial electricity transmission data were obtained from China Electric Power Industry Yearbooks (2011–2023), and province-level calorific values of refined oil and fossil fuels were obtained from China Energy Statistical Yearbooks (2011–2023). Default diesel production CF factors and diesel combustion carbon emission factors were obtained from the sources listed in the key resources table. Life cycle characterization was performed using SimaPro 10.2 with the IPCC GWP 100a method and Ecoinvent database V3.11.

Carbon footprint model

In the railway operation stage, CF is jointly influenced by transport scale, locomotive structure, energy efficiency, and energy CF factor, and can be expressed as a multiplicative function of these four factors, as shown in Equation 1:

$$\begin{array}{c}{\text{CF}}_{ij}=\sum _k{\text{CF}}_{ijk}=\sum _k{\text{TV}}_{ij}\times {\text{TP}}_{ijk}\times {\text{ECF}}_{ijk}\times {\text{ECCFF}}_{ijk}\text{,}\end{array}$$ (Equation 1)

where i denotes the railway bureau; j denotes the year; k denotes the locomotive type; CF is the operational CF; TV is the transport volume; TP is the transport proportion; ECF is the energy consumption factor; and ECCFF is the energy consumption CF factor.

The energy consumption CF factors are constructed through three steps: province-level correction based on energy production efficiency, railway bureau-level mileage weighting, and bureau-level CF factors accounting, as shown in Equations 2, 3, 4, and 5:

$$\begin{array}{c}{\mathrm{W}}_{ijkp}=\frac{{\mathrm{D}}_{ijkp}}{\sum _p{\mathrm{D}}_{ijkp}}\text{,}\end{array}$$ (Equation 2)

$$\begin{array}{c}{\text{ECC}\text{FF}}_{ijk}=\sum _p{\text{ECCFF}}_{jkp}\times {\mathrm{W}}_{ijkp}\text{,}\end{array}$$ (Equation 3)

where p denotes the province; W is the operational mileage weight; and D is the operational mileage.

The province-level electricity consumption CF factor is calculated by a comprehensive approach that we previously proposed, which integrates multiple efficiency-specific power sources into an electricity node-flow model to track consumption-side spatiotemporal changes.⁶⁰

The province-level diesel consumption CF factor includes both upstream refining factor and combustion factor, with the former corrected based on provincial diesel production efficiency, as shown in Equations 4 and 5.

$$\begin{array}{c}{\text{DPE}}_{jp}=\frac{{RQ}_{jp}}{{CQ}_{jp}}\text{,}\end{array}$$ (Equation 4)

$$\begin{array}{c}{\text{ECCFF}}_{jkp}^{k=\text{diesel}}=\text{DPCFF}\times \frac{{\text{DPE}}_{jp}}{\text{DPE}}+\text{DCCEF,}\end{array}$$ (Equation 5)

where DPE is the diesel production efficiency; RQ and CQ are the calorific values of refined oil and fossil sources, respectively; DPCFF is the default diesel production CF factor; and DCCEF is the diesel combustion carbon emission factor.

Carbon footprint decomposition model

To quantify the contribution of each driver to CF changes during 2010–2022, the LMDI method was adopted. The CF change is expressed by ΔCF and can be decomposed into the following factors, as shown in Equations 6, 7, 8, 9, and 10:

$$\begin{array}{c}\Delta {\text{CF}}_i=\Delta {\text{CF}}_i^{\text{TV}}+\Delta {\text{CF}}_i^{\text{TP}}+\Delta {\text{CF}}_i^{\text{ECF}}+\Delta {\text{CF}}_i^{\text{ECCFF}}\text{,}\end{array}$$ (Equation 6)

$$\begin{array}{c}\Delta {\text{CF}}_i^{\text{TV}}=\sum _k\frac{{\text{CF}}_{ijk}^{j=2022}-{\text{CF}}_{ijk}^{j=2010}}{\ln \left({\text{CF}}_{ijk}^{j=2022}\right)-\ln \left({\text{CF}}_{ijk}^{j=2010}\right)}\times \ln \left(\frac{{\text{TV}}_{ij}^{j=2022}}{{\text{TV}}_{ij}^{j=2010}}\right)\text{,}\end{array}$$ (Equation 7)

$$\begin{array}{c}\Delta {\text{CF}}_i^{\text{TP}}=\sum _k\frac{{\text{CF}}_{ijk}^{j=2022}-{\text{CF}}_{ijk}^{j=2010}}{\ln \left({\text{CF}}_{ijk}^{j=2022}\right)-\ln \left({\text{CF}}_{ijk}^{j=2010}\right)}\times \ln \left(\frac{{\text{TP}}_{ijk}^{j=2022}}{{\text{TP}}_{ijk}^{j=2010}}\right)\text{,}\end{array}$$ (Equation 8)

$$\begin{array}{c}\Delta {\text{CF}}_i^{\text{ECF}}=\sum _k\frac{{\text{CF}}_{ijk}^{j=2022}-{\text{CF}}_{ijk}^{j=2010}}{\ln \left({\text{CF}}_{ijk}^{j=2022}\right)-\ln \left({\text{CF}}_{ijk}^{j=2010}\right)}\times \ln \left(\frac{{\text{ECF}}_{ijk}^{j=2022}}{{\text{ECF}}_{ijk}^{j=2010}}\right)\text{,}\end{array}$$ (Equation 9)

$$\begin{array}{c}\Delta {\text{CF}}_i^{\text{ECCFF}}=\sum _k\frac{{\text{CF}}_{ijk}^{j=2022}-{\text{CF}}_{ijk}^{j=2010}}{\ln \left({\text{CF}}_{ijk}^{j=2022}\right)-\ln \left({\text{CF}}_{ijk}^{j=2010}\right)}\times \ln \left(\frac{{\text{ECCFF}}_{ijk}^{j=2022}}{{\text{ECCFF}}_{ijk}^{j=2010}}\right)\text{,}\end{array}$$ (Equation 10)

where ΔCF^TV, ΔCF^TP, ΔCF^ECF, and ΔCF^ECCFF are the CF changes caused by transport volume, transport proportion (electric locomotive workload/diesel locomotive workload), energy consumption factor, and energy consumption CF factor changes, respectively.

Carbon footprint efficiency model

In this model, the decision-making units (DMUs) are defined as the 18 railway bureaus. The following formulations (Equations 11, 12, 13, 14, 15, and 16) are derived from Tone et al.^61,62 where the Slack-Based Measure model is extended to handle undesirable outputs under variable returns to scale. For each DMU, we define the input and output vectors, as shown in Equations 11, 12, 13, and 14:

$$\begin{array}{c}{x}_{ij}={\left({\text{LAB}}_{ij},{\text{EC}}_{ij},{\text{CAP}}_{ij}\right)}^{\mathrm{T}}\text{,}\end{array}$$ (Equation 11)

$$\begin{array}{c}{\text{EC}}_{ij}=\sum _k{\text{TV}}_{ij}\times {\text{TP}}_{ijk}\times {\text{ECF}}_{ijk}\text{,}\end{array}$$ (Equation 12)

$$\begin{array}{c}{y}_{ij}={\left({\text{REV}}_{ij}\right)}^{\mathrm{T}}\text{,}\end{array}$$ (Equation 13)

$$\begin{array}{c}{b}_{ij}={\left({\text{CF}}_{ij}\right)}^{\mathrm{T}}\text{,}\end{array}$$ (Equation 14)

where x, y, and b are the input vector, desirable output vector, and undesirable output vector, respectively; LAB is the employee number; EC is the energy consumption; CAP is the capital input; and REV is the revenue. Due to insufficient data availability, we use the operational mileage to substitute the capital input. This assumption is a consensus for the engineering field, such as the power sector, where capital investment is replaced by the installed capacity of power generation.^63,64

With the variable-returns-to-scale assumption, the production possibility set excluding the DMU being evaluated is defined as Equation 15:

$$\begin{array}{c}{\mathrm{P}}_{\left(ij\right)}=\left\{\left(x_{ij},y_{ij},b_{ij}\right)|x_{ij}\ge {\mathrm{X}}_{\left(ij\right)}·\lambda ,y_{ij}\le {\mathrm{Y}}_{\left(ij\right)}·\lambda ,b_{ij}\ge {\mathrm{B}}_{\left(ij\right)}·\lambda ,{\mathrm{e}}^{\mathrm{T}}·\lambda =1,\lambda \ge 0\right\}\text{,}\end{array}$$ (Equation 15)

where P represents all feasible input-output combinations producible by convex combinations of other railway bureaus; X, Y, and B are the input, desirable output, and undesirable output matrices of all DMUs’ inputs and outputs excluding the targeted one; and λ represents the intensity vector.

For each DMU, the super-efficiency score, d∗ is obtained by minimizing the weighted ratio d of the average proportional input expansion to the average proportional output contraction, as shown in Equation 16:

$$\begin{array}{c}{{\mathrm{d}}_{ij}^{\ast }=\min \mathrm{d}}_{ij}=\frac{\frac{1}{3}\times \left(\frac{{\widetilde{\text{LAB}}}_{ij}}{{\text{LAB}}_{ij}}+\frac{{\widetilde{\text{EC}}}_{ij}}{{\text{EC}}_{ij}}+\frac{{\widetilde{\text{CAP}}}_{ij}}{{\text{CAP}}_{ij}}\right)}{\frac{1}{2}\times \left(\frac{{\widetilde{\text{REV}}}_{ij}}{{\text{REV}}_{ij}}+\frac{{\widetilde{\text{CF}}}_{ij}}{{\text{CF}}_{ij}}\right)}\text{,}\end{array}$$ (Equation 16)

If d∗ = 1, the bureau achieves best-practice efficiency relative to all other bureaus. If d∗>1, the bureau is super-efficient and outperforms other bureaus. If d∗<1, the bureau is inefficient, indicating redundant input use or insufficient desirable output generation given its CF.

Quantification and statistical analysis

This study used deterministic, model-based quantification rather than inferential statistical testing. The analytical scope comprised n = 18 railway bureaus, and annual observations cover 2010–2022. Operational carbon footprints were calculated using the life cycle accounting model described in method details; LMDI was used to decompose carbon-footprint changes; and SBM-DEA was used to evaluate efficiency using the 18 railway bureaus as decision-making units. Life cycle characterization used the software, method, and database listed in the key resources table, including SimaPro 10.2, IPCC GWP 100a, and Ecoinvent database V3.11. No statistical hypothesis tests were performed; therefore p values, confidence intervals, randomization, blinding, and measures of center or dispersion are not applicable.