Urban Currents

RTSKG: Building a Rail Transit Station Knowledge Graph Dataset

The researchers constructed the Rail Transit Station Knowledge Graph (RTSKG), a dataset that models spatial and semantic interactions between rail stations, road segments, and points of interest using a unified schema. The dataset's utility was evaluated through two downstream tasks: station-area store recommendation and knowledge-enhanced ridership prediction.

Why it matters — It provides a structured, linked-data framework that integrates heterogeneous urban entities, allowing researchers to model complex spatial and semantic relationships around transit hubs rather than treating stations in isolation.

arXiv · pdf

A new framework for vehicle trajectory prediction based on physics-informed deep learning models

The Mixed-Granularity Physics-Informed Deep Learning (MPIDL) framework predicts multi-step longitudinal vehicle trajectories by coupling a microscopic LSTM module with a macroscopic ANN module. The microscopic module predicts vehicle-level accelerations from historical speed, headway, and relative-speed sequences, while the macroscopic module estimates joint velocity-density fields regularized by the Aw-Rascle-Zhang traffic-flow model. Evaluated on the NGSIM and ZTD datasets, the framework achieves lower Average Displacement Error and Final Displacement Error than standard physics-based and deep learning baselines.

Why it matters — It bridges the gap between micro-level vehicle dynamics and macro-level traffic flow theory, ensuring that individual trajectory predictions remain physically consistent with overall traffic density and velocity fields without sacrificing real-time inference speed.

Caveat: Although the framework is trained jointly using both micro and macro modules, the macroscopic constraints are discarded during inference, meaning real-time physical consistency relies on the patterns learned during training rather than active runtime enforcement.

Transportation Research Part C Emerging Technologies · doi

A distributed data governance architecture for smart cities based on data unified registration, dynamic catalog, and local LLM data coordination

This paper presents a distributed data governance architecture for smart cities that combines a unified registration standard, a probe-driven dynamic catalog, and local Large Language Models (LLMs) for semantic coordination. The system was deployed in Suining City, China, connecting 151 systems across 50 departments. In a load test of 83,909 simulated requests, the platform achieved an average response time of 730 ms, a maximum latency of 1,443 ms, and an error rate of 0.12%.

Why it matters — It demonstrates a viable architecture for cross-departmental data sharing that respects ownership boundaries by restricting LLMs to semantic interpretation while using deterministic policy engines and human review to enforce security and compliance.

Urban Informatics · pdf · doi

Smart technologies in enhancing flood resilience in coastal African cities

This study evaluates flood risk and resilience strategies across four coastal African cities—Accra, Cape Town, Mombasa, and Tunis—by analyzing peer-reviewed articles, policy reports, and technical documents published between 2000 and 2024. The comparative thematic analysis reveals distinct local approaches: Accra focuses on community-based early warning and GIS; Cape Town utilizes real-time monitoring and predictive modeling; Mombasa pilots low-cost IoT sensors; and Tunis integrates smart tools into urban planning.

Why it matters — It documents how smart city technologies are being adapted to different institutional and resource constraints in African coastal urban centers, identifying shared systemic barriers such as funding deficits, data gaps, and fragmented governance.

Caveat: The findings are based on a descriptive literature review and qualitative case studies rather than primary empirical measurements of technology performance or flood mitigation outcomes.

Discover Cities · doi

Motor insurers: gatekeepers of future mobility?

This study investigates the role of motor insurers in shaping transport governance and future mobility innovations through 40 semi-structured interviews with insurers and stakeholders in the United Kingdom. It maps how these actors view emerging connected, automated, shared, and electric mobility trends, identifying three key domains of influence: legal obligations, safety monitoring via data practices, and financial risk management.

Why it matters — It demonstrates that insurers act as active gatekeepers whose varying appetite for risk and product innovation directly accelerates or hinders the adoption of new transport technologies.

Caveat: The findings are based on qualitative interview data from a single national context, which may reflect the specific regulatory and market conditions of the United Kingdom.

Transportation Research Interdisciplinary Perspectives · doi

Battery chemistry, climate exposure, and V2G economics in electric bus fleets: A two-stage degradation-aware dispatch framework

The study presents a two-stage dispatch optimization framework that integrates battery degradation models calibrated from accelerated aging tests of Lithium Iron Phosphate (LFP) and Nickel Manganese Cobalt (NMC) cells. Tested under representative transit conditions and weather profiles, the model co-optimizes charging and vehicle-to-grid (V2G) participation while accounting for battery wear. The results demonstrate that LFP cells sustain 2.3 to 2.4 times the throughput-to-end-of-life of NMC cells, and the degradation-aware dispatch strategy reduces annual fleet operating costs by 12.6%.

Why it matters — It provides transit operators with a quantified, climate-sensitive comparison of battery chemistries and a scheduling method that prevents V2G revenue from being erased by premature battery wear.

Transportation Research Interdisciplinary Perspectives · doi

Impact of green logistics on export performance: an empirical study in Vietnam on the comprehensive and progressive agreement for trans-Pacific partnership nations

Using an augmented gravity model and panel data spanning 2010 to 2018 and 2022, this study analyzed how green logistics performance affects Vietnam's exports to other CPTPP nations. Feasible Generalized Least Squares estimation revealed that while Vietnam's domestic green logistics performance is positively associated with its export values, the green logistics performance of importing partner countries is negatively associated with Vietnam's export performance.

Why it matters — The study demonstrates that green logistics operates as a double-edged sword in international trade, acting as a domestic driver of export competitiveness while simultaneously functioning as a non-tariff barrier when enforced by importing nations against a developing economy.

Caveat: The analysis relies on macro-level panel data and is restricted to Vietnam's trade relationships within the CPTPP framework, which may not generalize to other developing nations or bilateral trade agreements.

Transportation Research Interdisciplinary Perspectives · doi

Untold stories. On women, gender and architecture in Denmark

This study reconstructs the historical contributions of women to Danish architecture, centering on the 1936 telegram sent to architect Ragna Grubb. It uses archival fragments and marginal anecdotes to recover the overlooked professional networks, collaborations, and spatial practices of female architects in Denmark during the mid-twentieth century.

Why it matters — It recovers a documented history of female architectural practice and solidarity in Denmark, challenging the male-dominated canon of architectural history by demonstrating that women were active, networked participants rather than isolated exceptions.

Caveat: The analysis relies on highly specific, qualitative archival fragments and anecdotes, which may not represent the broader systemic conditions of the profession at the time.

Planning Perspectives · doi

Folk Engineering: Planning Southern Regionalism

This historical analysis traces the rise of the Institute for Research in the Social Sciences (IRSS) at the University of North Carolina at Chapel Hill during the 1920s and 1930s. It details how the institute developed 'folk engineering' as a regional planning methodology to study and manage the social and economic transition of the American South from an agrarian society to an industrial one.

Why it matters — The study uncovers the intellectual origins of Southern regional planning, demonstrating how early academic social science was explicitly designed as an interventionist tool to shape regional development and modernization.

Caveat: The provided abstract is truncated and focuses on a historical case study of a single academic institution, which limits the generalizability of the planning practices described.

Planning Perspectives · doi

Behavioural mechanisms of driver compliance under lane-level variable speed limits: field evidence and predictive modeling

This study analyzes driver compliance with lane-level variable speed limits using real-world vehicle trajectory data. It introduces a predictive framework combining Recursive Feature Elimination with a Light Gradient Boosting Machine (LGBM) model, which achieved 93.5% accuracy in predicting compliance. The analysis reveals that compliance is primarily driven by lane speed limits and vehicle type, with lower limits yielding significantly lower compliance rates, especially in the left lane when both lanes share the same limit.

Why it matters — It demonstrates that driver compliance is not uniform across lanes or vehicle types, providing empirical evidence that can be used to design more realistic, static lane-level speed control strategies that account for anticipatory deceleration and lane-specific behaviors.

Transportmetrica A Transport Science · doi

Rethinking urban design for health: a review of built environment and physical activity correlates in high-density Asian cities

This scoping review synthesizes quantitative evidence from 34 studies to map the relationships between the neighborhood built environment and physical activity in high-density Asian cities. Using the '5Ds' framework, the analysis evaluates how density, diversity, design, destination accessibility, and distance to transit correlate with active behavior, finding that design and destination accessibility are the most consistent predictors of leisure-time physical activity.

Why it matters — It demonstrates that standard urban design frameworks behave differently in hyper-dense Asian contexts, where density and land-use diversity exhibit complex, non-linear threshold effects rather than the straightforward positive correlations typically observed in Western cities.

Caveat: The findings are based on a scoping review of existing literature, which relies on the varying methodologies and potential biases of the 34 included studies.

Cities & Health · doi

Satellite Censorship and the Geographies of (Un)Seeing: Crisis, Power, and Withheld Imagery

This paper analyzes the governance and politics of satellite visibility, tracing how orbital imagery is restricted through tasking priorities, resolution limits, and temporal release controls. Using comparative case studies of Gaza, Ukraine, and the Red Sea, it demonstrates how states and corporate actors maintain near-real-time observation while journalists, researchers, and communities receive degraded or delayed data.

Why it matters — It conceptualizes satellite imagery not as a neutral data source but as a governed infrastructure, revealing how systematic asymmetries in access undermine the reliability of geospatial science, humanitarian verification, and public accountability during crises.

Caveat: The study relies on qualitative comparative case studies rather than a quantitative audit of image availability or delay metrics.

Annals of the American Association of Geographers · doi

Surviving the Buffalo Blizzard: Online Interactions, Mutual Assistance, and the Power of Weak Ties

This study analyzed online-offline mutual aid interactions during the December 2022 Buffalo blizzard by manually collecting Facebook conversations, classifying mutual-aid messages with machine learning, and mapping user interactions with social network analysis. The analysis categorized the digital exchanges into practical, informational, and emotional support, mapping how these connections formed networks of weak ties among disconnected residents.

Why it matters — It demonstrates that virtual spaces function as digital urban commons, proving that emergent social capital among strangers can organize and deliver life-saving physical resources when formal emergency services are completely halted.

Caveat: The findings are based on a single extreme weather event in one city and rely on data from a single social media platform.

Annals of the American Association of Geographers · doi

Shaping the notion of #wellbeing in the therapy culture context: an analysis through Instagram narratives

The study analyzed 9,844 Instagram posts tagged with #wellbeing using hierarchical clustering, non-parametric statistical tests, and computer vision object detection models. It mapped the textual and visual themes of these posts, measured user engagement across different categories, and evaluated gender representation in the images.

Why it matters — It demonstrates that online wellbeing discourse is heavily feminized and dominated by mental, psychological, and spiritual narratives rather than physical fitness or nutrition. This provides empirical evidence of how social media platforms shift the public understanding of wellness toward therapeutic and self-discovery frameworks.

Caveat: The findings are based on a relatively small sample of under 10,000 posts from a single social media platform, which may not represent broader cultural shifts outside of Instagram's specific user demographics.

arXiv · pdf · doi

Towards an approach to multivariate outlier detection for District Heating System data

The study tested five outlier detection methods—Z-score, Mahalanobis distances, Principal Component Analysis (PCA), Isolation Forest, and Hotelling's T-squared test—on transmitted heat energy and ambient temperature data from a single district heating substation. Domain experts evaluated the results, leading to the development of an ensemble model that identifies anomalies based on the consensus of PCA, Isolation Forest, and the Hotelling method.

Why it matters — It establishes a validated, domain-specific anomaly detection approach that filters out irrelevant zero-energy data points, enabling operators to pinpoint irregular plant behaviors and target gas consumption and carbon emission reductions.

Caveat: The proposed ensemble method was developed and tested using data from only a single selected substation.

arXiv · pdf · doi

Mediatised Participation: Citizen Journalism and the Decline in User-Generated Content in Online News Media

This paper presents a critical review of audience participation in news media by conducting a systematic literature review of studies published over the last two decades. It synthesizes findings on citizen journalism and user-generated content, documenting a general decline in audience interest in creating news content, persistent professional resistance from journalists, and corporate strategies focused primarily on monetizing participation.

Why it matters — It reframes the once-celebrated concept of citizen journalism as a failed innovation, demonstrating that active audience participation has been constrained and transformed into a highly controlled, mediatised phenomenon rather than a democratic equalizer.

Caveat: The findings are based on a literature review of existing academic studies rather than new empirical data on current newsroom practices.

arXiv · pdf · doi

Policy Convergence and Divergence Across National and Within Regional AI Strategies: A Policy Design Element Analysis

This study coded and analyzed 74 national and 3 regional AI strategies from a global scan of 205 states using a latent-inductive approach. It evaluated horizontal and vertical policy convergence across three design elements: goals, approaches, and principles. The analysis revealed strong horizontal convergence around economic competitiveness, research support, and ethical AI use, but persistent divergence regarding human rights, participatory governance, and human-centric principles.

Why it matters — It provides the first systematic empirical evidence of how national AI policies align with or diverge from regional frameworks, mapping out which policy elements are becoming global norms and which remain highly localized.

Caveat: The analysis is based on a qualitative coding of policy documents, which reflects stated strategic intentions rather than actual regulatory enforcement or implementation outcomes.

arXiv · pdf

A Systematic Sample Size Analysis of ML-Based Path Loss Prediction for LPWAN

This study evaluates machine learning models for LoRa path loss prediction using real-world measurements from an urban deployment, testing how training set size affects accuracy. A Random Forest model using LiDAR-derived terrain features and a k-Nearest Neighbors model using coordinate data were compared against empirical and specialized LPWAN baselines. At maximum training size, both machine learning models achieved root-mean-square errors below 6.5 dB, outperforming the best baseline model which scored 9.7 dB.

Why it matters — It demonstrates that machine learning models can significantly improve path loss prediction accuracy for smart city network planning compared to traditional empirical models, though their performance depends heavily on whether the model must generalize to unseen gateway locations.

Caveat: The models show limited spatial generalizability, with the coordinate-only k-NN model degrading substantially and the Random Forest model showing highly placement-dependent errors when tested on unseen gateways.

arXiv · pdf

Technology interactions reshape the economics of China's coal power decarbonization

An interaction-aware optimization framework was developed to evaluate energy conservation, biomass co-firing, and carbon capture across 1,885 coal-fired power plants in China. The model accounts for plant-level heterogeneity alongside shared biomass and carbon dioxide storage resources to map marginal abatement cost curves. The analysis reveals that 1.2 Gt CO2 per year can be mitigated at negative marginal cost, while achieving full carbon neutrality requires a marginal abatement cost of US$56 per tonne of CO2.

Why it matters — It demonstrates that evaluating decarbonization technologies in isolation miscalculates both mitigation costs and emission reduction potentials, providing a coordinated framework to optimize retrofit investments and infrastructure planning.

arXiv · pdf

XGBoost "is all you need": the case of forecasting transmitted heat energy in District Heating Systems

This study compares the performance of XGBoost and Long Short-Term Memory (LSTM) networks for forecasting transmitted heat energy using a real-world District Heating Systems dataset. The evaluation demonstrates that XGBoost consistently outperforms the deep learning model, particularly because LSTM produces larger errors during intervals with limited data availability.

Why it matters — It demonstrates that conventional machine learning can outperform complex deep learning models for specific utility forecasting tasks, offering a way to reduce both computational costs and the carbon footprint of data analysis in energy systems.

Caveat: The comparison is based on a single, unspecified real-world District Heating Systems dataset, which may limit how well the performance hierarchy generalizes to other infrastructure networks.

arXiv · pdf · doi

Unlocking CCUS-Ready Coal Power Investments: A Spatial Real Options Approach

The study introduces a Spatial Real Options framework that combines national-scale siting screening with site-level valuation to assess the viability of Carbon Capture, Utilization, and Storage (CCUS)-ready coal plants. It evaluates a national inventory of 194,027 technically feasible sites across China, incorporating temporal uncertainty, province-specific conditions, CO2 transport distances, and policy instruments. The analysis reveals that the investment window for conventional coal closes by the early-to-mid 2040s, and that operational incentives like generation-hour compensation are far more effective than upfront capital subsidies.

Why it matters — It demonstrates that the economic viability of CCUS-ready coal investments is highly sensitive to spatial factors like transport distance and storage type, shifting the policy focus from subsidizing capital costs to addressing ongoing operational costs.

Caveat: The framework's findings are highly dependent on the specific regulatory and geological conditions of China, which may limit direct applicability to other national energy markets.

arXiv · pdf

MARS: A framework for modelling register-based social networks

The Multiplex Affiliation-based Random Spatially-embedded (MARS) graph framework replicates the generative processes of register-based social networks. The framework's statistical properties were derived and tested using a simplified model calibrated against the population-scale register-based social network of the Netherlands, analyzing how spatial tie strength influences network closure.

Why it matters — It provides a theoretical framework capable of replicating and analyzing government-curated administrative microdata networks, demonstrating that increased spatial freedom correlates with a decline in social cohesion.

Caveat: The empirical validation is demonstrated using a simplified model implementation rather than the full MARS framework.

arXiv · pdf

A Comparative Evaluation of Deep Learning Object Detection Models on a Real-World Multi-Plant Dataset from Africa

The study evaluated six object detection models—YOLOv5, YOLOv8, YOLO11, YOLO26, Faster R-CNN, and RT-DETR—using the AgriAISeg dataset, which contains 3,382 manually collected images of sesame, cabbage, and tomato crops from Nigerian farms. The transformer-based RT-DETR model achieved the highest performance with a precision of 0.768 and an mAP@0.5:0.95 of 0.624, while Faster R-CNN performed poorly with an mAP@0.5 of 0.466.

Why it matters — It establishes performance benchmarks for modern computer vision models on real-world, uncontrolled African agricultural data, proving that one-stage and transformer-based detectors are more resilient to complex field conditions like occlusion and changing illumination than older two-stage detectors.

Caveat: The evaluation is limited to three specific crop types (sesame, cabbage, and tomato) within a single national context.

arXiv · pdf

Technology, education and critical media literacy: potential, challenges, and opportunities

This study analyzes the integration of technology in media education by surveying 141 university students in Communication and Education programs and conducting semi-structured, in-depth interviews with academic experts. The research evaluates how emerging digital challenges like deepfakes, disinformation, and incidental information exposure are addressed in higher education curricula.

Why it matters — It documents a critical gap where technological tools are integrated into classrooms only superficially, identifying specific deficiencies in teacher training that prevent students from developing the critical media literacy needed to navigate modern misinformation.

Caveat: The empirical findings are limited to the perspectives of a specific cohort of communication and education university students and selected academic experts, which may not represent broader educational contexts.

arXiv · pdf · doi