Urban Currents

From announcement to operation: The changing impact of light rail transit on housing prices across implementation phases

This study analyzed the impact of Montreal's Réseau express métropolitain (REM) light rail project on housing prices across four distinct phases: announcement, construction, testing, and initial operation. Using a difference-in-differences approach on panel data of 14,711 residential property transactions from 2013 to 2023, the researchers measured price changes within 800 meters of both operating and under-construction branches. The initial announcement premium persisted through construction, but testing and early operations offset these gains near the active branch, likely due to noise and vibration, while properties near unopened branches maintained their elevated values.

Why it matters — It demonstrates that transit-induced property value appreciation is not a simple upward trajectory, but a dynamic process where early speculative gains can be erased by the negative externalities of actual operations. This temporal variation suggests that land value capture policies must be carefully timed rather than assuming permanent post-construction price premiums.

Caveat: The findings are based on a single transit project in Montreal, which may limit generalizability to cities with different housing market dynamics or transit technologies.

Transportation Research Part A Policy and Practice · pdf · doi

Zero-shot trajectory generation with self-supervised diffusion model

The researchers developed a zero-shot trajectory generation method based on a self-supervised diffusion model that uses only aggregated traffic flow statistics and public road network topology. The model treats surveillance traffic flow distributions as conditioning information to guide its denoising process through an adaptive optimization framework, bypassing the need for complete single-vehicle trajectory annotations. The method was validated using real-world traffic datasets from the Dadukou District in Chongqing, producing synthetic trajectories that closely match the similarity and distribution of real trajectories.

Why it matters — It establishes a capability to generate realistic, high-quality vehicle trajectories in areas where individual tracking data is unavailable or restricted, relying instead on easily accessible aggregate flow statistics and public maps.

Caveat: The model's performance and applicability have only been validated using traffic datasets from a single district in Chongqing.

Computational Urban Science · pdf · doi

Optimization design of age-friendly homestay spaces integrating Space syntax and SegNet

This study develops an automated spatial optimization framework for age-friendly homestay interiors by combining a lightweight SegNet model with space syntax and Visibility Graph Analysis. The SegNet model detects architectural floor plan elements with a mean Intersection over Union of 0.89 and precision and recall exceeding 0.90. A hybrid optimization algorithm combining Non-dominated Sorting Genetic Algorithm II and Simulated Annealing then generates layouts that optimize proxy indicators for accessibility, fall-risk exposure, visual recognizability, and wayfinding, achieving a hypervolume indicator of 0.72.

Why it matters — It replaces subjective, experience-based layout practices with a quantifiable and automated design workflow, allowing designers to systematically balance caregiving safety, cognitive clarity, and physical accessibility in short-term elderly accommodations.

Caveat: The optimization relies entirely on spatial proxy indicators rather than direct measurements of actual elderly user experiences or long-term behavioral data within the spaces.

Computational Urban Science · pdf · doi

<scp>NOCTURNAL INFORMALITY</scp> : Rethinking the Temporal Politics of Urban Informality

This conceptual essay introduces the framework of 'nocturnal informality' to analyze how urban informality operates across temporal rhythms rather than just spatial dimensions. It uses the lived experiences of street traders in Harare, Zimbabwe, to illustrate how the urban poor navigate changing patterns of surveillance, visibility, and state regulation as day turns to night.

Why it matters — It shifts the theoretical focus of informal urban studies from purely spatial negotiations to temporal ones, demonstrating that the night is a politically charged space-time where state authority and informal survival strategies are actively renegotiated.

Caveat: The paper is a conceptual essay based on qualitative observations of a single city, rather than an empirical study with quantitative datasets.

International Journal of Urban and Regional Research · doi

Geospatial Modelling of Stunting and Underweight Among Children Under Five Years: Insights from Sub-Saharan Africa

This study analyzed subnational spatial variation in stunting and underweight among 198,443 children under five across 24 Sub-Saharan African countries using Demographic and Health Survey data from 2015 to 2023. Using generalized additive models with spatial smoothing, the researchers mapped localized hotspots and found that stunting prevalence ranged from 16% to 55% and underweight from 5% to 29%. Children in the highest cumulative disadvantage quartile faced significantly higher odds of growth failure, with structural and socioeconomic risk factors statistically associated with 45% to 89% of the malnutrition burden.

Why it matters — It demonstrates that national averages obscure critical localized hotspots of child malnutrition, providing high-resolution spatial maps and population-attributable risk estimates that allow policymakers to target resources to specific subnational areas of cumulative disadvantage.

Caveat: The estimated reductions in malnutrition burden are theoretical, model-based associations under specific assumptions rather than direct evidence of preventable outcomes.

Applied Spatial Analysis and Policy · pdf · doi

Exploring the role of place visitation big data on small area health measure estimation

This study integrated smartphone-derived place visitation data across 120 categories from SafeGraph Patterns with 12 demographic and social determinants of health variables from the 2019 American Community Survey to estimate 22 CDC health measures at the census tract level. Using hierarchical regression analysis, the model evaluated how routine activity patterns affect health predictions in both urban and rural tracts classified by USDA codes.

Why it matters — The research demonstrates that dynamic mobility data significantly improves small-area health estimations over static demographic indicators alone, yielding a 7.5% average increase in R² that rises to 12.5% in rural areas. It establishes that specific visitation behaviors, particularly visits to drinking establishments, serve as powerful predictors for localized health outcomes like binge drinking and depression.

Caveat: The study relies on smartphone-derived location data, which may introduce demographic biases in who is tracked, and uses static 2019 survey data as a baseline.

Computational Urban Science · pdf · doi

Between city and countryside: The (un)sustainable development of peri-urban areas (PUA) of large cities in Poland – The case of Warsaw

This study analyzes spatial development patterns in Warsaw's peri-urban fringe and neighboring rural municipalities using spatial data from 2014 to 2024. It maps the most intensive zones of residential expansion along administrative borders, establishing a typology of development forms based on density, continuity, diversity, and multifunctionality. The analysis reveals a highly fragmented, functionally monotonous urban fabric that lacks public infrastructure and relies heavily on private car transit.

Why it matters — It provides empirical evidence of how administrative boundaries fail to contain urban sprawl, demonstrating that transitional fringe zones require formal recognition as distinct planning entities to prevent unsustainable, car-dependent development.

Caveat: The empirical analysis is limited to the metropolitan area of a single Polish city, Warsaw.

Cities · doi

We are not all the same: Preference-based market segmentation among public transport users in Soweto, South Africa

Using revealed and stated preference surveys from public transport users in Soweto, Johannesburg, this study analyzed travel choices across six modes, including minibus taxis, bus rapid transit, and rail. The researchers estimated multinomial, nested, and latent class logit models to identify preference heterogeneity among commuters. The latent class model successfully segmented the population into two distinct groups: a 'transfer sensitive' group averse to transfers and waiting, and a 'walk sensitive' group that tolerates transfers but strongly dislikes walking.

Why it matters — It provides empirical evidence of distinct commuter segments within a major Sub-Saharan African township, moving beyond simple income-based segmentation. This allows transit planners to design targeted interventions, such as minimizing transfers or reducing stop distances, to match the specific tolerance thresholds of different local user groups.

Caveat: The study relies on hypothetical stated preference scenarios alongside revealed choices, which may not perfectly capture actual commuting decisions.

Research in Transportation Economics · pdf · doi

Exploring joint travel-charging adaptation responses of battery electric vehicle users under varying weather conditions

This study analyzes how battery electric vehicle owners simultaneously adapt their travel and charging behaviors during adverse weather using a two-stage analytical framework. Latent class analysis and an integrated choice model were applied to 1,506 stated adaptation survey responses from private electric vehicle owners in Shanghai, China, evaluating reactions to temperature, wind, and rainfall. The model identified four distinct behavioral patterns: status quo maintenance, cautious adjustment, adaptive rescheduling, and protective withdrawal.

Why it matters — It establishes a direct behavioral link between weather conditions and combined mobility-charging decisions, demonstrating that rainfall and wind trigger avoidance behaviors while temperature fluctuations dictate whether drivers adjust or maintain their routines. This provides infrastructure operators with empirical evidence to predict grid load shifts and manage charging network resilience during extreme weather events.

Caveat: The findings are based on a scenario-based stated adaptation experiment rather than observed real-world GPS and charging logs, which may introduce hypothetical bias.

Transportation Research Part A Policy and Practice · doi

Influence of the COVID-19 pandemic on strategic sustainable mobility goals, discourses, plans and strategies in Norwegian urban regions

This study analyzed how the COVID-19 pandemic affected long-term sustainable land-use and transport planning in two Norwegian urban regions. The researchers conducted 21 in-depth interviews with leading planning professionals across two waves in 2021 and 2024, and reviewed planning documents spanning the pre-pandemic, mid-pandemic, and post-pandemic periods. The analysis reveals that despite temporary shifts in travel behavior, pre-existing sustainable mobility goals and strategies remained unchanged due to strong political anchoring and institutionalized agreements.

Why it matters — It demonstrates that highly institutionalized urban planning frameworks can withstand major external shocks without abandoning sustainability targets, while also identifying that these strategies remain vulnerable to rising public transit costs and threats to anti-sprawl policies.

Caveat: The findings are based on a small sample of professional interviews and document reviews restricted to two urban regions in a single country with highly specific governance structures.

Transportation Research Interdisciplinary Perspectives · doi

Space debris and outer space sustainability: legal challenges

This paper evaluates the legal and policy frameworks governing space sustainability against the physical threat of cascading space debris, utilizing data from the European Space Agency Annual Space Environment Report 2025. It analyzes how current voluntary guidelines and fault-based liability standards fail to address the Kessler Syndrome, and contrasts the market-driven regulatory approach of the proposed EU Space Act with the coordination-focused model of the United States.

Why it matters — It demonstrates that the rapid deployment of commercial mega-constellations and active debris removal technologies has created a critical governance gap, proving that Cold War-era treaty regimes are structurally inadequate for managing a congested, finite orbital environment.

Caveat: The study synthesizes existing legal discourse and reports rather than introducing novel quantitative models or new frameworks for regulatory reform.

Transportation Research Interdisciplinary Perspectives · pdf · doi

How e-commerce influences last-mile practices that foster informality

This study uses an inductive qualitative approach and an instrumental case study design to analyze how e-commerce outsourcing drives informality in last-mile logistics. The research is based on semi-structured interviews with five stakeholder groups: delivery drivers, e-commerce firms, public sector officials, logistics and road safety experts, and community representatives.

Why it matters — It demonstrates how the outsourcing of last-mile deliveries systematically transfers operational risks and costs from private e-commerce companies to informal workers and the public, establishing that current regulatory gaps institutionalize these negative externalities.

Caveat: The findings are based on qualitative case study interviews, which may limit their direct generalizability to regions with different regulatory frameworks or labor market structures.

Transportation Research Interdisciplinary Perspectives · pdf · doi

Real-Time Climate Risk Assessment for Supply Chain Resilience: A Data-Driven Nowcasting Framework for Colombian Agriculture

This paper develops a conceptual early warning system architecture that integrates short-term climate nowcasting with supply chain risk modeling for Colombian agriculture. Using historical meteorological observations, official agricultural time series, and reanalysis products, the framework translates precipitation nowcasts into threshold-based risk indicators for logistics, inventory, and sourcing without relying on satellite imagery or computer vision.

Why it matters — It establishes a method to convert immediate, localized weather forecasts directly into actionable supply chain decisions, allowing agricultural logistics operators to anticipate disruptions from extreme weather rather than reacting after they occur.

Caveat: The framework's feasibility was demonstrated using synthetic and historical data experiments in a controlled computational environment rather than a real-world deployment.

arXiv · pdf

AirFlow: Context Preserving and Multi-Rate State Modeling for Air Quality Forecasting

The AirFlow forecasting framework models air quality using a dual-stream state space architecture that processes station-level multivariate observations without relying on graph propagation or predefined signal decomposition. It introduces a statistic-guided normalization routing mechanism based on 24-hour autocorrelation and distribution drift, alongside a hierarchical dual-stream state model with gated bidirectional cross-attention. Tested on real-world multi-city datasets, the model outperformed state-of-the-art baselines in 34 out of 36 metric comparisons, reducing root mean square error by up to 11.11% while requiring only 0.0483M parameters and 0.0215G FLOPs.

Why it matters — It demonstrates that air quality forecasting can be significantly improved by tailoring normalization and temporal modeling to the distinct periodicities and drift rates of individual pollutants, rather than forcing all channels through a single shared backbone. This achieves superior accuracy with a highly compact parameter footprint suitable for resource-constrained urban monitoring systems.

arXiv · pdf

Full-Feature versus Limited-Input Machine Learning for Residential Energy Estimation: A Comparative Analysis of RECS and ResStock Under Realistic Input Constraints

This study evaluated five machine learning models (CatBoost, XGBoost, LightGBM, Random Forest, and Neural Networks) to estimate residential energy use under constrained data conditions, using the RECS and ResStock datasets. While full-feature CatBoost models achieved R2 scores of 0.90 on ResStock and 0.73 on RECS, restricting inputs to ten easily accessible variables (such as weather and basic administrative records) dropped performance to R2 scores of 0.62 and 0.61 respectively. However, restricting the low-input model to a homogeneous subset of single-family detached homes in Climate Zone 6A built between 2000 and 2010 restored predictive accuracy to R2 = 0.85.

Why it matters — It quantifies the exact performance penalty of omitting detailed physical and behavioral data from energy models, demonstrating that algorithmic complexity cannot overcome missing inputs unless the target housing stock is highly homogeneous.

Caveat: The findings rely on one empirical survey dataset and one synthetic simulation dataset, meaning the observed model performance is bound to the specific generation methods and biases of RECS and ResStock.

arXiv · pdf

GeoAI-based post-segmentation quality validation of building footprints via spatial feature engineering

The researchers developed a GeoAI quality control framework to detect and remove erroneous building footprints generated by deep learning models (U-Net and SAM-LoRA) from UAV imagery. Using 24 geometric, spatial-contextual, spectral, and texture predictors, they trained machine learning classifiers on three sites in Bangladesh and tested them on an independent site. A Decision Tree classifier achieved 95.31% accuracy and a 91.06% F1-score in identifying boundary deformations, ultimately reducing the proportion of erroneous footprints in the database from 27.32% to 4.62%.

Why it matters — It provides an automated, highly transferable post-segmentation validation method that purifies vector databases without manual inspection, bridging the gap between raw deep-learning outputs and production-ready GIS databases.

Caveat: The framework was developed and validated using data from only five UAV survey sites within a single country, which may not capture the full diversity of global building typologies and sensor types.

arXiv · pdf

Bayesian Node Edge Modeling of Road Crashes in Central Bogotá

The study fitted a Bayesian negative binomial node-edge model to 8,169 road segments and 8,398 intersections across six central districts of Bogotá to analyze crash counts. It evaluated separate predictors for road hierarchy, pavement, speed, signalization, intersection configuration, and land-use treatment, finding that intersections with at least four incident segments and higher maximum incident speeds had higher expected crash counts.

Why it matters — It demonstrates that treating intersections and segments as distinct network elements provides a more interpretable baseline for urban crash analysis than conventional models, which often obscure their differing exposure and connectivity patterns.

Caveat: The segment-level component of the model exhibited weak raw-scale predictive performance and numerical uncertainty for some pavement categories.

arXiv · pdf

WhichTok? Comparing Three TikTok Data Acquisition Tools

This study systematically compares three TikTok data collection tools—the official TikTok Research API, Pyktok, and Apify—across five endpoints: User, Hashtag, Keyword, Comment, and Related Video. The evaluation reveals substantial discrepancies in retrieved content for hashtag and keyword searches, driven by the tools' reliance on either back-end API calls (Research API) or front-end web scraping (Apify and Pyktok). Only the user endpoint produced consistent and comprehensive results across all three methods.

Why it matters — It demonstrates that different data acquisition tools introduce systematic biases regarding the time periods and popularity levels of retrieved content, challenging the assumption that researchers can obtain randomized samples from the platform.

Caveat: The study evaluates the tools based on their performance at a specific point in time, which may change as TikTok updates its API and scraping protections.

arXiv · pdf

Mitigating Bus Bunching with Reinforcement Learning Enhanced by Semantic Stop Embedding

The study introduces a deep Q-learning controller for event-driven bus holding that incorporates offline, LLM-generated semantic stop embeddings representing physical attributes, surrounding activities, and historical operations. Tested on stochastic simulations calibrated with real-world data from two bus routes, the semantic controller reduced headway variability by 32.0%, bus bunching events by 69.2%, and passenger waiting times by 24.0% compared to a calibrated Daganzo baseline.

Why it matters — It demonstrates that embedding semantic context about transit stops, rather than relying solely on instantaneous operational data or route-specific IDs, improves control trade-offs and accelerates policy adaptation when transferring reinforcement learning models to new routes.

Caveat: While warm-start fine-tuning accelerated early learning during cross-route transfers, cold-start training still yielded the best final performance, indicating limits to immediate zero-shot policy generalization.

arXiv · pdf

Deep Learning based Detection of Fishing Vessels and Fishing Monitoring using Nightlight Images

A dual-branch YOLO11 deep learning model was developed to detect small-scale fishing vessels by fusing 10-meter panchromatic and 40-meter RGB nighttime light imagery from the SDGSAT-1 satellite. Applied to the western coast of India across a 2022-2023 dataset, the model detected 31,525 vessel instances, achieving a precision of 0.99 and an mAP@50 of 0.96. Cross-matching with Automatic Identification System (AIS) data revealed that 77.3% (24,379) of the detected vessels operated without active AIS transmissions.

Why it matters — It establishes a highly accurate method for identifying 'dark vessels' that evade traditional tracking systems, revealing that more than three-quarters of the fishing fleet along India's western coast operates without active AIS monitoring.

arXiv · pdf

Unsupervised Detection of Groundwater Storage Anomalies in Ghana Using GRACE Satellite Data

This study analyzed groundwater storage anomalies in Ghana from 2004 to 2024 using GRACE satellite data, Z-score standardization, and an ensemble-based Isolation Forest machine learning model. The framework identified 12 anomalous months, consisting of 5 deficit and 7 surplus events, revealing a persistent groundwater deficit from 2004 to 2009 and a shift to positive anomalies after 2018. Spatial patterns showed more frequent deficits in northern Ghana and stronger surplus events in the south.

Why it matters — It establishes a practical, unsupervised monitoring framework that can detect subtle groundwater deviations in data-scarce regions without relying on extensive in-situ observations or conventional rigid statistical thresholds.

Caveat: The study relies entirely on satellite-derived GRACE data and unsupervised modeling due to a lack of long-term, in-situ groundwater measurements to validate the detected anomalies on the ground.

arXiv · pdf

Joint return levels of maximum temperature and minimum relative humidity by combining copulas with an extreme value framework for bimodal data

This study models the joint return periods of extreme heat and low humidity in Brasília, Brazil, using a copula-based approach combined with a novel extreme value framework for bimodal data. The researchers evaluated multiple copula families, including rotated versions, to capture the asymmetric dependence structure and the multimodal joint density of maximum air temperature and minimum relative humidity.

Why it matters — The model successfully reproduces multimodal patterns that suggest the presence of multiple distinct climate regimes, providing a more accurate statistical tool to quantify the frequency of concurrent hot and dry events that trigger wildfires.

Caveat: The empirical application and validation of this bimodal extreme value framework are limited to a single city.

arXiv · pdf

From Rating Factors to Crash Mechanisms: A Multiscale Causal DAG Framework Linking Motor Insurance and Road Safety

The study introduces a multiscale causal directed acyclic graph (DAG) framework that bridges the gap between annual motor insurance claims and short-term crash-generating processes. The framework integrates a crash-occurrence graph built from 72 study-edge records, an observation layer mapping rating variables to latent behaviors, and a downstream claim-administration process. It was tested using diagnostic examples from the French freMTPL2freq portfolio and a Spanish age-mediation dataset to demonstrate how aggregate insurance metrics fail to isolate specific crash mechanisms.

Why it matters — It establishes the mathematical limits of using annual insurance rating factors to infer actual road safety behaviors, proving that stronger causal claims about crash mechanisms require trip-level data and linked crash-claim records rather than aggregated policy-year statistics.

Caveat: The framework's empirical demonstrations are limited to aggregate, coarse-resolution insurance portfolios from France and Spain, which lack the high-resolution trip telemetry needed to resolve specific physical crash mechanisms.

arXiv · pdf

GeoRoute: Geometry-Aware Hybrid Inference for Traffic Future-Frame Prediction

The researchers developed GeoRoute, a training-free inference framework that improves long-horizon traffic video prediction by stabilizing static structures. For front-camera footage, it uses a multi-frame depth-layered renderer to project static geometry from historical frames onto the generated future frames, while a frozen vision-language model routes heterogeneous camera views to specialized motion predictors. The system was validated on the AI City Challenge Track 5 benchmark, achieving competitive performance without requiring any retraining or fine-tuning of the underlying video diffusion models.

Why it matters — It provides a way to eliminate temporal ghosting, geometry drift, and structural instability in video diffusion models at inference time, bypassing the need for computationally expensive model retraining.

Caveat: The framework's performance and validation are demonstrated primarily on a single competitive benchmark dataset.

arXiv · pdf