Urban passenger transportation has witnessed a growingly severe carbon emission problem, while the integrated travel mode combining dockless bike-sharing (DBS) and metro offers a promising solution for emission mitigation. However, the synergistic carbon reduction effects of multimodal combinations like “biking + metro” have remained unexplored. This study developed a carbon reduction assessment framework that distinguishes between different travel modes and trip purposes, supported by massive DBS datasets in Shenzhen, China. Specifically, this framework is designed to reveal the differential emission reduction characteristics between the only-biking and the biking-metro-biking mode. The results revealed that carbon reduction hotspots of only-biking mode exhibit a multi-center agglomeration pattern, predominantly distributed in commercial and residential areas. In contrast, the biking-metro-biking mode forms carbon reduction corridors along major roads. Although biking-metro-biking mode accounted for only 3% of total trips, its per-trip contribution to carbon reduction was significantly higher than that of only-biking travel. Moreover, compared to the only-biking model, combined travel offered greater advantages in reducing emissions. Trip purposes also play a significant role in shaping emission reduction patterns: commuting and consumption-related trips are strongly correlated with mixed-use commercial-residential zones, while education- and healthcare-oriented trips generate distinctive emissions reduction hotspots. These findings reveal distinct carbon reduction mechanisms associated with multimodal patterns, providing scientific evidence for formulating targeted low-carbon transportation policies.

Heavy-duty trucks (HDTs) pose a critical decarbonization challenge in China amid growing logistics demand. This study proposes a Well-to-Wheel (WTW) supply-demand coupling framework, based on HDT trajectory data to explore the diesel-to-green hydrogen transition for HDTs in the Pearl River Delta (PRD). By leveraging the Expressway-importance metric to identify emission hotspots, three hierarchical deployment scenarios are designed to address the 41.8% supply shortage for full transition under the current installed capacity. These scenarios achieve WTW carbon reductions of 8.7%, 14%, and 19%, with their hydrogen demand remaining within actual supply. Economic analysis confirms the rationality of the scenarios with annual costs ranging from 145.39 to 329.91 million yuan. This scenario-specific, data-supported method verify the dual rationality of phased green hydrogen HDT deployment and economic cost, providing quantitative basis for corridor prioritization and an actionable decarbonization roadmap. This study offers replicability for offshore wind-rich coastal regions, advancing regional HDT low-carbon transition.

Solar photovoltaic (PV) systems have great potential to supply renewable energy and support the transition of urban energy structures. While integrating PV technology into public facilities is a promising strategy for expanding PV deployment in urban areas, the PV potential of standardized bus stops, which are numerous and widely distributed across cities, remains insufficiently assessed. To address this gap, this study proposes a framework to quantify the PV potential of bus stop roofs by integrating bus stop point-of-interest (POI) data, street view images, and solar trajectory simulations. The research results reveal two key findings. Firstly, the theoretical annual electricity generation of bus stop photovoltaic (BSPV) systems in Guangzhou, China, can reach 5,531.67 MWh, which is equivalent to powering approximately 84,196 LED streetlights for one year. Secondly, BSPV potential exhibits distinct spatio-temporal variability: it follows a unimodal seasonal pattern (peaking in early summer and dropping to its minimum in March), while significant spatial heterogeneity is observed. Specifically, the central urban districts (e.g., Yuexiu, Liwan, Haizhu, and Tianhe) show relatively low BSPV potential due to severe building shading. In contrast, peripheral districts (including Huadu, northern Baiyun, Huangpu, Panyu, and Nansha) demonstrate notably higher BSPV potential. Additionally, areas along the Pearl River also possess considerable PV potential. The study provides a data-driven solution for urban planners to deploy distributed energy systems via public infrastructure. Furthermore, the findings offer scientific guidance for decision-making related to urban sustainable development and low-carbon transport planning.

Crowdsourced individual trajectory data have become a valuable resource for examining environment-activity relationships at the streetscape scale. Such analyses critically depend on two key geo-processing decisions: (1) the trajectory assignment method (Hidden Markov Models [HMM] vs. buffer-based approaches), and (2) the spatial delineation of built environment variables. While it is intuitively understood that methodological choices can influence results, systematic evaluations of their combined effects remain limited. This study addresses the gap through a comparative analysis of different combinations of assignment methods and spatial ranges, using walking and cycling trajectory data from the historic urban core of Guangzhou, China. The findings reveal: (1) Assignment methods significantly affect both the statistical and spatial properties of trajectory allocation. The HMM approach produces finer representations of walking and cycling activity, while buffer-based methods capture broader trends due to the lack of probabilistic decision-making. This also explains why cycling data are more sensitive to assignment choices than walking data. (2) In combination with spatial range, assignment methods jointly influence both linear and non-linear correlation patterns between the built environment and activity. These effects are amplified in non-linear models compared to linear ones. These findings carry important methodological implications, highlighting previously hidden uncertainties embedded in common analytical workflows. The study also extends the discussion of the Modifiable Areal Unit Problem (MAUP) to trajectorybased streetscape research, underscoring the need for careful spatial decision-making in studies of active mobility.

Poverty is a persistent global challenge. However, relying solely on socioeconomic indicators or low-resolution nighttime light data hinders the precision of poverty identification. Moreover, the spatial characteristics of buildings, as key indicators of socioeconomic development, remain under-investigated for their potential in identifying impoverished areas. To bridge these gaps, this study combined high-resolution (10-m) SDGSAT-1 nighttime light with three-dimensional (3D) building indicators to enable precise poverty identification. The XGBoost algorithm was employed to assess the Multidimensional Poverty Index (MPI) for counties within Jiangxi Province, China. A comparison was also made with conventional NPP-VIIRS nighttime light (500-m resolution). The findings indicate that the incorporation of 3D building indicators into SDGSAT-1 data substantially improves model performance, providing a more robust poverty estimation than single-source datasets. Specifically, the combined model achieved R2 values of 0.8986 and 0.8540 for SDGSAT-1 and NPP-VIIRS, respectively. This represents an information gain of 0.1160 for the high-resolution SDGSAT-1 data compared to its baseline value of 0.7826. Additionally, SHAP analysis elucidates that "mean building height", "sky view factor", and "standard deviation of pixel light values" are the predominant drivers of the model. In conclusion, combining 3D building indicators with fine-grained nighttime light data enables a higher precision in identifying poverty across county-level units. This approach presents a novel methodology for the timely monitoring of poverty across large regions. These findings provide a robust foundation for optimizing resource allocation and informing sustainable poverty alleviation strategies.

Hydrogen-fueled heavy-duty trucks (HDTs) hold great decarbonization potential but suffer from insufficient refueling infrastructure. This study integrates 173,481 real-world HDT trajectories into the multi-capacitated NMC-FRLM model, enabling one-step joint optimization of siting and capacity via endogenous cost weights (3.72:1.93:1 for Class A [4000 kg/d], B [2000 kg/d], C [1000 kg/d]). Validated in the Pearl River Delta across six penetration scenarios (0.065%-1.5% hydrogen HDT share), the model clusters 774,556 refueling strategies into 199 corridors and selects 4-44 HRSs from 398 candidates. It achieves 85-100% utilization (Class A: 99.55%), meeting 7.9-176 t/d hydrogen demand across scenarios. Cost-weight sensitivity analysis shows small HRSs should not be prioritized; large/medium HRS deployment follows threshold rules: Class A for core corridors when A/B cost ratio >2:1, and B for suburban/secondary corridors when <2:1. Outperforming OD-based methods, this approach provides targeted policy support for hydrogen HDT transition.

Amid accelerating urbanization and climate warming, urban humid-heat environments pose growing threats to human health and thermal comfort. Human-centered humid-heat metrics are critical for assessing the urban thermal environment and human comfort. The use of mobile sensors offers a novel technical solution for acquiring high-resolution, human-scale humid-heat data. However, large-scale monitoring remains challenging due to cost constraints. To address this critical gap, this study proposes a novel multimodal prediction framework that combines a Long Short-Term Memory network (LSTM), a Residual Neural Network (ResNet), and a FusionSelf-Attention module (FSA). The framework aims at low-cost, city-scale prediction of human-scale humid-heat parameters, including air temperature (T), relative humidity (RH), black globe temperature (BGT), and wet bulb globe temperature (WBGT). It integrates remote sensing, street-view imagery, and meteorological data, using high-resolution street-level humid-heat data collected from bicycle-mounted mobile sensing for model training. A key innovation lies in its Fusion-Self-Attention (FSA) module, which uniquely optimizes cross-modal integration through attention mechanisms, enabling more effective interaction and fusion of multi-source data. The results demonstrate that the LSTM-ResNet-FSA framework achieves high predictive accuracy (R-2: 0.70-0.79) and enables city-scale humid-heat mapping. The ablation tests reveal: (1) Remote sensing + meteorological data alone yield R-2 > 0.6; (2) Street-view data boosts performance by Delta R-2 approximate to 0.10. Furthermore, the study demonstrates that the FSA fusion module consistently and significantly outperforms both SENet-based and MLP-based fusion approaches in all metric evaluations, confirming its superior performance in cross-modal feature integration.

With rising needs for high-resolution population data in urban development, the spatial distribution of occupational groups has emerged as a pivotal element for industrial layout optimization and resource allocation. While traditional census and remote sensing data struggle to capture the fine-scale heterogeneity of specific occupational dynamics, this study proposes a spatial simulation method for grid-level occupational population modeling by integrating Tencent user big data with multi-source geographic information and employing machine learning techniques. The results demonstrate that using only highly accessible data combined with the XGBoost model achieves excellent performance in simulating occupational population distributions, with an average coefficient of determination (R2) of 0.80 across all analyzed industries. Furthermore, by employing SHAP for interpretability, we quantitatively revealed the influence intensity and complex nonlinear effects of diverse geographical features on these distributions. The findings not only deepen the understanding of how occupational agglomeration shapes urban functional dynamics but also offer actionable scientific evidence to support data-driven urban planning strategies.

Nighttime leisure activities play a crucial role in enhancing the quality of life for urban residents. However, existing studies have rarely differentiated the diverse spatial contexts in which nighttime leisure-related activities occur, limiting understanding of their spatiotemporal patterns and underlying mechanisms. This study aims to fill this gap by integrating mobile phone signaling data with areas of interest (AOI) datasets to systematically analyze the spatial and temporal distribution and determinants of nighttime leisure activities among residents in Guangzhou, China. We categorize leisure areas into four types to capture the diversity of nighttime leisure activities. Employing an optimal-parameter-based geographical detector model, we investigate the independent and interactive effects of various factors influencing nighttime leisure activities. Our findings reveal that among the four types of leisure areas, commercial service areas are the most popular venues for nighttime activity. Furthermore, high accessibility and well-developed surrounding facilities are identified as crucial factors facilitating nighttime leisure engagement. This study enriches the theoretical framework of urban leisure behavior by highlighting the spatiotemporal complexities and the distinctive driving factors of nighttime leisure. These findings advance the theoretical understanding of nighttime leisure dynamics and provide scientific guidance for urban nightlife management.

Cases in which residential-oriented new towns evolve into single-function bedroom communities (BCs) characterized by strong commuting dependence are widespread in regions experiencing rapid urbanization worldwide. While existing literature has extensively examined residents' commuting behaviors in BCs and the resulting social inequalities, their spatial patterns and internal heterogeneity remain under-explored. Addressing this gap, this study develops a comprehensive framework for identifying and analyzing BCs in Guangzhou, integrating a social sensing perspective with urban vitality theory. From the viewpoint that human activity reflects urban functions, we identify typical BCs leveraging grid-level daily commuting characteristics extracted from mobile phone data. This identification is further validated through large language model-assisted web retrieval, demonstrating high spatial consistency. Building on the identified BCs, our analysis reveals following findings. At the macro scale, the spatial pattern of BCs is shaped by Guangzhou's polycentric structure, forming two distinct concentric belts. Meanwhile, axial aggregation is observed along the metro corridors connecting urban centers, particularly toward their terminal sections. At the micro scale, BCs exhibit significant heterogeneity in urban vitality, categorized into three types: mature-development, transportation-dependent, and spatial-isolated. These results deepen our understanding of the spatial organization and evolutionary logic of BCs, offering empirical insights for targeted urban renewal and functional optimization.

Long-term effects of massive building material use in China, which experienced intense urbanization in the past two decades, remain insufficiently explored. Here, to fill these gaps, we developed a high-resolution time-series database of building material stocks from 2000 to 2019 and found that China held 15

The urgent task of mitigating global warming requires efforts to reduce carbon emissions. The key is to incorporate relevant measures into urban planning strategies. Building form not only affects the daily activities of inhabitants but also significantly influences carbon emission patterns within surrounding areas. Consequently, it is crucial to understand how building form impacts carbon emission patterns. The nonlinear threshold effects of building form on carbon emissions have not been extensively studied. This study aims to address this void by examining the nonlinear connection between carbon emissions and building form and identifying specific threshold effects by using random forest and partial dependence plots. We conducted a comparative analysis of various machine learning models, including gradient boosting decision tree, extreme gradient boosting, and random forest. The random forest demonstrated the best fit. Further analysis indicated that urban building form has a substantial impact on urban carbon emissions. Notably, the floor area ratio was the most critical factor, accounting for 12.93 % of the relative importance in influencing carbon emissions. This was followed by the building congestion degree (12.24 %), the high building density (11.64 %), and the sky view factor (10.65 %). Collectively, these top four indicators, all related to building form, underscore their significant role in determining urban carbon emissions. In addition, nonlinear threshold effects were observed between the building form indicators and carbon emissions. These effects manifested as distinct patterns, such as platform, V-shaped, and N-shaped relationships, characterized by alterations in influencing trends, frequency, and distribution of thresholds. Among these relationships, platform and V-shaped types were observed with greater frequency, whereas N-shaped relationships, which are more complex, were encountered less frequently. Our findings provide insights for urban policymakers to develop targeted strategies to mitigate carbon emissions by optimizing building form in urban contexts.

Accessibility of healthcare services is a paramount determinant of elderly health outcomes. However, existing research often neglects the effects of resource-sharing pressures among different demographic groups for healthcare resources when measuring the accessibility for elderly population. To bridge this gap, this study developed a Supply Allocation Model (SAM) that considers the interactions between the elderly and non-elderly populations, as well as the distribution of healthcare resources. The model was subsequently integrated with the G2SFCA method, utilizing demographic, mobile phone, and point of interest (POI) datasets to assess the spatial accessibility of healthcare services for the elderly population in Guangzhou, China. The model’s accuracy and reliability were tested through calibration and validation processes, utilizing real-world healthcare treatment datasets. The effectiveness of the SAM was measured through the computation of the healthcare accessibility index and the Gini coefficient, utilizing both the SAM-G2SFCA and G2SFCA models. The results show that the SAM achieves the highest prediction accuracy at a 15-minute threshold. As the time threshold decreases, the role of supply factors in predicting the proportion of healthcare utilization strengthens. Moreover, the SAM-G2SFCA leads to a reduction in the equity of accessibility across all time thresholds when compared to the G2SFCA method, particularly in peripheral urban areas where elderly populations face greater resource-sharing pressures and healthcare accessibility is often overestimated. These findings provide valuable insights for policy formulation and theoretical advancement, informing the design of more equitable and efficient healthcare resource allocation strategies.

Poverty is a pervasive global issue that adversely affects human well-being. Traditional socioeconomic censuses are time-consuming and resource-intensive, suffering from temporal delays, while reliance on nighttime light data with low spatial resolution is insufficient for fine-scale identification of impoverished regions. Furthermore, the spatial heterogeneity of nighttime light in different urban functional zones has been overlooked. To address these shortcomings, we proposed a novel approach by integrating high-resolution SDGSAT-1 nighttime light data (10 m) with urban functional zoning data using a spatial overlay tool. A random forest model was then applied to predict county-level poverty identification in Guangdong, China. For comparative validation, traditional NPPVIIRS nighttime light data (500 m) were also incorporated. This method effectively explored the nonlinear relationship between nighttime light, urban functional zones, and the multidimensional poverty index (MPI, serving as the dependent variable). Our experiments demonstrate that the integration of urban functional zoning with nighttime light moderately improves the accuracy of poverty estimates. Among the models tested, the one considering functional zoning-based indicators of "number of light pixels" and "sum of pixel light values" increased the correlation coefficient by 0.0158 compared to the model without considering these indicators. Additionally, comparative analysis revealed that high-resolution data from SDGSAT-1 exhibited a better fit with the MPI when integrated with functional zoning-based indicators. Specifically, the correlation coefficient of this combination was 0.0086 higher than that of traditional NPP-VIIRS data. This highlights that SDGSAT-1 can delineate the boundaries between dark and light regions more precisely, leading to a more accurate reflection of regional poverty levels. Our findings facilitate fine-scale poverty estimation across large regions. This approach can inform policy design, such as dynamic optimization of resource allocation based on poverty estimates, thus enabling timely and accurate poverty alleviation efforts.

The extensive expansion of impervious surfaces encroaches on green spaces and causes frequent urban waterlogging disasters. Previous studies have focused mainly on the influence of green space landscape pattern on waterlogging, with less attention given to green space morphological spatial pattern (MSPA). MSPA can be used to differentiate various types of land use morphologies from a microscopic perspective and reveal visualized spatial characteristics. Therefore, this study selected Shenzhen, a city with serious waterlogging problems, as the study area. The anthropogenic/natural environments and green space morphological spatial pattern were considered. Pearson correlation analysis and random forest regression were combined to investigate the influence of these drivers on the density of waterlogging hotspots and quantify the degree of importance for each driver. The results were supplemented with explanations using SHapley Additive exPlanations and Partial Dependence Plots. Pearson correlation analysis revealed that green space morphological spatial pattern, the proportion of green spaces, and the proportion of impervious surfaces were the dominant drivers. Additionally, the random forest regression showed that incorporating green space morphological spatial pattern and average tree height as potential drivers could strengthen the model’s goodness-of-fit. While the proportion of impervious surfaces, the proportion of green spaces, and population density were important drivers, the green space morphological spatial pattern, specifically the “loop”, “edge”, and “core”, was even more crucial and had an optimal design range. Therefore, green space morphological spatial pattern should be emphasized during the planning of “sponge cities” to maximize the ability of green spaces to mitigate waterlogging. In summary, our findings are expected to provide feasible suggestions for waterlogging control and green space planning.

The rise of dockless bike-sharing systems has led to increased interest in using bike-sharing data for sustainable transportation and travel behavior research. However, these studies have rarely focused on the individual daily mobility patterns, hindering their alignment with the increasingly refined needs of active transportation planning. To bridge this gap, this paper presents a two-layer framework, integrating improved flow clustering methods and multiple rule-based decision trees, to mine individual cyclists' daily home-work commuting patterns from dockless bike-sharing trip data with user IDs. The effectiveness and applicability of the framework is demonstrated by over 200 million bike-sharing trip records in Shenzhen. Based on the mining results, we obtain two categories of bike-sharing commuters (74.38 % of Only-biking commuters and 25.62 % of Biking-with-transit commuters) and some interesting findings about their daily commuting patterns. For instance, lots of bike-sharing commuters live near urban villages and old communities with lower costs of living, especially in the central city. Only-biking commuters have a higher proportion of overtime than Biking-with-transit commuters, and the Longhua Industrial Park, a manufacturing-oriented area, has the longest average working hours (over 10 h per day). Moreover, massive users utilize bike-sharing for commuting to work more frequently than for returning home, which is intricately related to the over-demand for bikes around workplaces during commuting peak. In sum, this framework offers a cost-effective way to understand the nuanced non-motorized mobility patterns and low- carbon trip chains of residents. It also offers novel insights for improving the operations of bike-sharing services and planning of active transportation modes.

Global forest cover has been shrinking at an accelerating rate over the past decade due to deforestation and forest degradation. Connecting fragmented forest patches can effectively promote ecosystem health and sustainability. However, previous studies have rarely conducted forest network analysis at the national scale. Therefore, we aim to provide a comprehensive solution for the establishment of large-scale forest networks. We focused on China and combined morphological spatial pattern analysis with connectivity indicators for recognizing forest ecological sources at different distance thresholds. Moreover, the linkage mapper was employed for determining practicable ecological corridors. We found 734 ecological sources and 1717 practicable corridors within the national forests at a distance threshold of 3000 m. At an increase threshold of 5000 m, the number of ecological sources reached 934, with 2176 practicable ecological corridors. Notably, smaller ecological sources dominated the country at both distance thresholds, but more small ecological sources acted as "stepping stones" when the distance threshold was 5000 m. The forest patches in Northeast China and the Eastern Himalayas had high centrality values because they are crucial for maintaining connections between ecological sources. Our findings underscore the importance of connecting dispersed and fragmented forest patches at large scales to promote ecosystem health and sustainability. These results not only contribute to the understanding of forest networks but also offer practical guidance for national-scale forest conservation endeavors. The construction of such forest networks could be a pivotal strategy for conserving biodiversity and ensuring the long-term well-being of ecosystems.

Existing studies underscore the significance of traditional public transportation systems (e.g., buses and metros) in shaping residents' daily mobility patterns and social interactions. However, limited research has examined the spatiotemporal interaction patterns among users of dockless bike-sharing (DBS), an emerging low-carbon mode of public transportation. To address this gap, we investigate the daily spatiotemporal interaction patterns and disparities among bike-sharing users across different socioeconomic statuses (SES) in Shenzhen, China. Leveraging massive DBS trip datasets with user IDs, we delineate the individual daily activity spaces of frequent users across different SES groups. Taking into account spatiotemporal proximity and distance decay effects, we compute multiple activity-space-based differentiation indices, followed by comprehensive analyses. The results reveal that middle-SES users experience the greatest diversity in interactions with users from different socioeconomic backgrounds during daily cycling activities, followed by high-SES users, while low-SES users exhibit the least interaction. Furthermore, the activity-space-based differentiation for each SES group follows a consistent periodic temporal pattern, characterized by two alternating peaks and troughs throughout the day. Peaks emerge in midday and late-night hours, whereas troughs align with the morning and evening commuting periods. The degree of activity-space-based differentiation is strongly associated with proximity to central city and land-use functions. These findings provide valuable implications for promoting social integration and promoting equity in non-motorized and sustainable transportation services.

Urban cycling is a key component of sustainable transportation, and its development depends on creating cyclist-friendly environments. However, traditional data collection, including surveys and street view imagery (SVI) annotations, is constrained by limited scale and high cost, creating a bottleneck for systematic, city-level analysis. The development of Multimodal Large Language Models (MLLMs) offers new possibilities, but recent studies often focus on numerical scores while neglecting the explanatory textual rationales produced. This study first validates the MLLM's performance against a human-rated dataset, then develops a Cascaded Perception Parsing Framework (CPPF) to explain its decision-making behavior. The framework features a two-stage process: a secondary Large Language Model (LLM) first parses the MLLM's textual rationales into structured data, which then informs an interpretable XGBoost proxy model for analysis based on eXplainable Artificial Intelligence (XAI). First, the MLLM shows high agreement with human ratings for Traffic Safety and Public Security but struggles with the more subjective Scenic Beauty. Second, the proxy model analysis attributes this discrepancy to the MLLM's simplistic aesthetic logic: a mechanistic preference for orderliness that ignores complex human intuition. This research provides critical empirical evidence on the capability boundaries and cognitive limitations of AI in urban perception, while also offering a novel methodology for developing more explainable AI-assisted tools.

The choice of transportation mode by residents significantly affects road traffic carbon emissions. Recent studies have explored the carbon reduction effects of green travel behaviors, such as bicycle-sharing and metro travel. However, there remains a gap in research estimating the carbon reduction potential associated with the transition from motorized transport to low-carbon alternatives. In this study, we propose a carbon emission reduction scenario simulation framework based on vehicle trajectory big data. This framework is designed to evaluate the impact of shifts in residential travel modes on carbon emissions at a fine spatial and temporal scale. Our analyses indicate that only 7.2 % of car trips are suitable for a shift to active transportation options, while over 64 % of trips qualify as multimodal, particularly involving e-bikes in combination with metro, which can result in annual carbon reductions of up to 3,138 tons. This highlights the importance of multimodal transport in reducing transportation-related carbon emissions. Regarding the spatial pattern, peripheral areas present substantial carbon reduction potential, accounting for nearly 50 % of the total. Moreover, significant carbon reduction potential exists in road sections connecting central and peripheral areas. In terms of timing, we observe two peaks in emission reductions on weekdays, occurring between 7-9 AM and 4-6 PM, with an additional peak on weekends around 9 PM. Ultimately, our research highlights that multimodal transportation, especially the combination of walking or conventional cycling with metro, may offer greater carbon reduction efficiency than relying solely on active transportation options. The findings of this study can significantly inform urban transportation policy-making and guide residents toward sustainable travel choices.