2026-06-22

Dynamic multi-agent deep reinforcement learning-based pricing and incentivization approach in multimodal transportation networks

The study develops a multi-agent deep reinforcement learning framework that models interactions between a public transit authority and a shared mobility service provider. The authority agent dynamically distributes spatio-temporal public transport incentives to target equity, emissions, and efficiency, while the provider agent dynamically adjusts fares to maximize revenue. Numerical simulations of a three-hour morning peak period demonstrate that this coordinated approach reduces commuter costs by approximately 20%, cuts emissions by about 10%, and nearly doubles public transit profits.

Why it matters — It demonstrates that conflicting public and private transportation goals can be balanced dynamically, showing that targeted public transit incentives can simultaneously improve social equity and environmental outcomes without destroying private operator profitability.

Caveat: The framework's performance and 20% cost reduction benefits are demonstrated only within a simulated three-hour morning peak period rather than a real-world deployment.

Transportation Research Part C Emerging Technologies

Identifying predictive thresholds and marginal benefit boundaries of urban heat environments via explainable AI

This study developed a random forest model combined with explainable AI techniques to analyze the nonlinear relationships between urban spatial features and summer land surface temperatures in Shijiazhuang, China. Using multi-source data including remote sensing imagery, street-view-derived sky view factor (SVF), traffic flow, and points of interest, the model identified key predictive thresholds such as a building density limit of 0.270 and a traffic flow threshold of 2,800 vehicles per hour. The analysis also revealed that SVF is the most temporally stable predictor, while vegetation index and building morphology show complementary diurnal patterns.

Why it matters — It establishes a quantitative framework that translates complex machine learning outputs into specific, actionable planning thresholds for urban heat mitigation, moving beyond simple linear correlations to identify precise marginal benefit boundaries.

Caveat: The empirical thresholds and predictive relationships are derived from a single high-density city in northern China, which may limit their direct applicability to cities with different climates or urban morphologies.

Computational Urban Science · pdf · doi

Dynamic micro-simulation of domestic electricity consumption: a case of Beijing

The researchers developed SelfSim-Energy, an agent-based domestic electricity consumption model integrated into an urban microsimulation platform, and simulated Beijing's urban evolution from 2021 to 2030. The simulation projects that Beijing's total domestic electricity consumption will rise from 29.1 to 32.3 billion kWh, driven by population growth and a projected decline in homeownership from 82.1% to 56.8%, which subsequently reduces the adoption of energy-efficient lighting from 78.2% to 69.1%.

Why it matters — It connects household energy demand directly to broader urban dynamics like land use, population shifts, and housing tenure, demonstrating that declining homeownership rates can actively suppress the adoption of energy-saving technologies.

Caveat: The findings and behavioral relationships are simulated specifically for the institutional and demographic context of Beijing, which may not generalize to other urban governance systems.

Computational Urban Science · pdf · doi

A scoping review of residential balcony gardens as overlooked contributors to small urban green spaces

A scoping review of 160 studies synthesizes current knowledge on the contributions of resident-led balcony gardens to urban environments. The analysis reveals that these micro-spaces improve physical and mental health, foster social interaction, mitigate heat, enhance air quality, and support urban biodiversity.

Why it matters — It establishes balcony gardens as a distinct category of decentralized, resident-led green infrastructure, shifting the focus from large public parks to the unmapped ecological and social contributions of private micro-spaces in dense cities.

Caveat: The review notes that the actual city-wide contribution and spatial distribution of balcony gardens remain unmeasured because existing literature rarely treats them as formal urban green spaces.

Discover Cities · pdf · doi

A systematic review of geospatial artificial intelligence and machine learning in urban analytics

This study presents a systematic review and bibliometric analysis of 100 journal articles retrieved from Web of Science and Scopus to map the methodologies, data types, and applications of geospatial artificial intelligence (GeoAI) in urban analytics. The analysis, supported by the VOSviewer tool, reveals that despite the diversification of urban data, traditional machine learning and deep learning algorithms still dominate the field.

Why it matters — It synthesizes the structural limitations of current GeoAI research, specifically documenting how data dependency, a lack of model interpretability, and the omission of socioeconomic factors restrict the global applicability of these methods, particularly in developing regions.

Caveat: The findings are based on a relatively small sample of 100 selected journal articles, which may not capture the entire breadth of GeoAI literature.

Discover Cities · pdf · doi

Walking through foodscapes: Analyzing the culinary pathways of Italian migrant women in Chile

This study analyzes the everyday food practices of three Italian migrant women in the Valparaso-Via del Mar metropolitan area in Chile. Using a walking-based qualitative methodology, the research maps how participants navigate urban food spaces like markets, cafs, and restaurants, documenting how these sites connect to cultural identity, gender roles, and emotional wellbeing.

Why it matters — It demonstrates how mobile, sensory-focused qualitative methods can capture the emotional and spatial dimensions of migration that standard interviews miss, while establishing a reusable research protocol for studying urban belonging.

Caveat: The findings are based on a very small sample of three highly specific participants in a single Chilean urban area.

City Culture and Society · doi

Assessment of individualised dynamic environmental exposures within The World Avatar

The researchers developed a digital twin framework within The World Avatar project that uses a dynamic knowledge graph to calculate individual environmental exposures. The system deploys a computational agent to integrate spatial and temporal data, accounting for historical changes like greenspace development and real-time factors like food retail opening hours. The scalability of the approach was demonstrated using smartphone GPS trajectories tracking movements across both Singapore and the United Kingdom.

Why it matters — It shifts exposure modeling from static, single-variable spatial associations to dynamic, multi-variable pathways that track individuals through time and space. This allows researchers to account for both long-term urban structural changes and immediate, hour-by-hour service accessibility in health studies.

Computers Environment and Urban Systems · doi

Operationalising the urban self-driving vehicles readiness index from a policymaker perspective: the Hungarian case

The study developed an Urban Self-Driving Vehicle (SDV) Readiness Index and applied it to 50 Hungarian cities with populations over 20,000 and active public transport systems. Based on a 2025 survey of municipal policymakers, the index measures readiness across four dimensions: current mobility baselines, anticipated SDV challenges, intervention timelines, and implementation barriers. The assessed cities scored between 18% and 76% on the index, with an average score of 45% and larger cities demonstrating higher readiness.

Why it matters — It shifts the assessment of autonomous vehicle readiness from the national level to the municipal level, providing local policymakers with a standardized diagnostic benchmark to identify specific planning gaps and prioritize local infrastructure interventions.

Caveat: The index relies entirely on the subjective perceptions and expectations of municipal policymakers rather than objective physical or digital infrastructure measurements.

Case Studies on Transport Policy · doi

Machine learning for urban wind simulation: A comprehensive review

This review synthesizes recent developments in data-driven, physics-informed, and hybrid machine learning methods used as surrogate models for urban wind simulation. It categorizes the literature along the standard simulation pipeline, evaluating how these models handle modeling assumptions, validation strategies, and generalization behavior compared to traditional computational fluid dynamics.

Why it matters — It establishes a structured framework for understanding how machine learning can bypass the high computational costs and mesh-quality dependencies of conventional fluid dynamics, identifying critical gaps in robustness, benchmarking, and uncertainty quantification that must be resolved before these fast surrogates can be reliably deployed in iterative urban design.

Sustainable Cities and Society · doi

Evaluating vehicle-to-grid as a participation-dependent system flexibility resource: evidence from charging profiles and policy scenarios

This study evaluates the system-level potential of vehicle-to-grid (V2G) technology in South Korea by analyzing hourly electric vehicle charging sales data from 2021 to 2023. It models seasonal charging profiles and percentile-based peaks to project demand scenarios for 2030 and 2050, applying a five-dimensional participation framework that accounts for user enrollment, plug-in availability, technical readiness, energy constraints, and operational feasibility.

Why it matters — It demonstrates that V2G viability is constrained by human and operational participation rates rather than battery capacity, showing that even modest behavioral participation can unlock gigawatt-scale grid flexibility while low participation leaves the technology systemically irrelevant.

Caveat: The quantitative projections and behavioral profiles are based on charging data specific to South Korea, which may limit direct applicability to countries with different urban densities, driving habits, or grid structures.

Sustainable Cities and Society · doi

Can green cycling paths reduce commuter exposure to ultrafine particles? A repeated measures study in Copenhagen, Denmark

This study measured commuter exposure to ultrafine particles (UFP) by conducting 40 cycling trips along a fixed 8.3 km route in Copenhagen over six weeks in early 2024. Using a DiSCmini and GPS watch, researchers tracked particle number concentration (PNC) across different street types during rush and non-rush hours. The analysis shows that cycling along green paths reduced PNC by 32% to 50% compared to moderate- and high-traffic segments, with average green path concentrations at 5,849 pt/cm3 compared to over 11,000 pt/cm3 on high-traffic streets.

Why it matters — It quantifies the direct exposure mitigation benefit of off-road green cycling infrastructure, proving that spatial separation from motorized traffic significantly lowers a cyclist's inhaled pollutant load even within a dense urban network.

Caveat: The findings are based on a single 8.3 km route in Copenhagen, which may not represent exposure dynamics in cities with different traffic densities, meteorological conditions, or green infrastructure designs.

Sustainable Cities and Society · doi

The impact of purchase subsidy termination on electric vehicle adoption: a synthetic control approach

Using the Synthetic Control Method on monthly registration data from January 2018 to March 2025, this study analyzed the causal impact of Germany's sudden, court-mandated termination of battery electric vehicle (BEV) subsidies in December 2023. The abrupt policy end resulted in an estimated 45% decline in the German BEV market, translating to approximately 140,000 forgone registrations in 2024.

Why it matters — It provides the first empirical, real-world measurement of how an unexpected and immediate withdrawal of financial incentives affects electric vehicle adoption, demonstrating that market stability and decarbonization trajectories are highly vulnerable to sudden policy shifts.

Transportation Research Part A Policy and Practice · doi

An elephant in the bus: Violence against women

Using survey data from Cairo, Egypt, this study analyzes how gender-based violence on public transport affects women's mobility and well-being. The research documents how safety concerns act as the primary barrier to transit use, forcing women to endure unsafe conditions due to a lack of affordable alternatives and resulting in psychological trauma, financial stress, and social exclusion.

Why it matters — It demonstrates that transit-related violence is not just a safety issue but a structural driver of social and economic exclusion that restricts women's participation in daily activities and reinforces systemic inequality.

Caveat: The study relies on self-reported survey data from a single city, which may reflect specific local cultural and legal contexts regarding harassment.

Transport Policy · doi

A data-driven end-to-end framework for traffic estimation with cross-network generalization

The GTE-GC framework estimates network-wide traffic speeds using only network topology and land-use characteristics, bypassing the need for historical traffic observations. The model combines a spatial feed-forward neural network, temporal self-attention, and graph contrastive learning to perform both transductive estimation on partially observed networks and inductive estimation on entirely unseen networks. Performance was validated through experiments across ten European cities, demonstrating superior generalization in inductive scenarios when paired with appropriate source domains.

Why it matters — It provides a scalable, data-driven alternative to traditional, computationally heavy simulation models, allowing planners to estimate traffic conditions in entirely unobserved or newly planned urban areas using only static spatial and land-use data.

Caveat: The model's inductive performance on unseen networks depends heavily on the selection of appropriate source domains for training.

Transportation Research Part C Emerging Technologies · doi

Navigating complexity in offshore emergency response: lessons from the OMSAR 2024 exercise using effective management theory

This study evaluated the OMSAR 2024 maritime disaster simulation along a 175-kilometer offshore transport corridor in Malaysia, which involved a simulated vessel collision and mass rescue. Researchers analyzed 152 real-time observer logs from 37 experts using the Five-Element Model of Effective Emergency Management to assess the execution of the national Arahan NADMA No. 1 directive. The analysis identified 51 consolidated operational issues, revealing that 25% of bottlenecks occurred in resources and logistics and 22% in command structures, including a 40-minute command vacuum during fleet handovers.

Why it matters — It exposes how friction between distinct corporate and state maritime protocols can cause critical command gaps during deep-water transit crises, demonstrating that tactical seamanship alone cannot compensate for systemic coordination failures.

Caveat: The findings are based on a single simulated exercise within a specific national regulatory framework, which may limit direct generalizability to other international maritime jurisdictions.

Transportation Research Interdisciplinary Perspectives · doi

Exposed yet unmapped: evidence of differential flood exposure in deprived urban areas using citizen science

This study models flood exposure across six Sub-Saharan African cities by combining a lightweight, low-cost flood model, global remote sensing datasets, and citizen science data. The analysis reveals that deprived urban areas experience up to 200% higher flood exposure than other areas, with shallow floods under 10 cm causing over 50% of the documented property damage, disease outbreaks, and infrastructure failures. It also identifies systematic spatial biases in global built-up surface datasets, which overestimate flood exposure in peri-urban zones while omitting dense, deprived urban neighborhoods entirely.

Why it matters — It demonstrates that conventional global remote sensing datasets systematically misrepresent flood risk in informal settlements, proving that local citizen science is necessary to correct these mapping omissions and prevent the misallocation of resilience resources.

Caveat: The study relies on a lightweight, low-cost flood model and citizen-reported data, which may introduce different uncertainties compared to high-resolution physical hydraulic modeling.

npj Urban Sustainability · pdf · doi

One country, multiple portraits: representativeness in GPS-based mobility data is source-specific and spatially dependent

The study quantifies coverage bias in mobile phone location data across 2,478 municipalities in Mexico by comparing population estimates from Facebook and the multi-app aggregator Veraset against the 2020 Mexican Population Census. The analysis reveals that Facebook's data has higher and more evenly distributed coverage, whereas Veraset's data concentrates users in larger, wealthier, and more digitally connected municipalities. Explainable machine learning models show that digital access and material resources drive bias in the multi-app data, while demographic and population structures drive bias in the Facebook data.

Why it matters — It demonstrates that demographic bias in GPS-based mobility data is not uniform but highly source-specific and spatially clustered. This provides a framework for researchers and planners to statistically adjust and correct for representation bias when using mobile data in highly unequal, data-scarce settings.

Caveat: The findings are based on two specific commercial data providers in a single country, meaning the exact bias profiles and drivers may differ for other platforms or national contexts.

arXiv · pdf

Simulation-Free Estimation of Traffic Flows from Sparse Count Data

The researchers developed a simulation-free method to estimate time-varying traffic flows by partitioning a study area into regions, generating feasible region-to-region routes, and solving a weighted least-squares optimization problem. The model uses a weighted contribution matrix to align route allocations with sensor coverage, deriving edge-level trajectories from sparse, aggregated vehicle counts. The approach was validated on the Brussels road network using both real and synthetic traffic data, successfully reproducing daily traffic profiles.

Why it matters — It provides a computationally efficient alternative to traditional traffic simulation models, allowing planners to estimate detailed edge-level traffic flows from sparse sensor data at a fraction of the usual processing cost.

Caveat: The method's performance was evaluated using data from a single city network, and its accuracy depends on the spatial distribution and coverage of the existing physical sensors.

arXiv · pdf

Transfer learning-based method for automated ewaste recycling in smart cities

The study developed an automated e-waste classification model by fine-tuning the output layers of the pre-trained AlexNet architecture using transfer learning. The model was trained and evaluated on a small dataset consisting of 12 smartphone classes across 6 different brands, utilizing data augmentation to prevent overfitting. The optimized configuration, using a Stochastic Gradient Descent with Momentum optimizer and a learning rate of 3e-4, achieved a classification accuracy of approximately 98%.

Why it matters — It demonstrates that high-accuracy smartphone classification for circular economy recycling can be achieved with limited training data, reducing the need for massive, custom-labeled e-waste datasets.

Caveat: The model's high accuracy was validated only on a small dataset of 12 smartphone classes, which may not represent the diverse and degraded state of real-world e-waste streams.

arXiv · pdf · doi