Urban Currents

Predicting barge tow size on inland waterways using vessel trajectory-derived features: proof of concept

This proof-of-concept study predicts the number of barges in a tow configuration by analyzing 39 trajectory features derived from Automatic Identification System (AIS) transponders on towing vessels. Using high-resolution satellite imagery along the Lower Mississippi River to label 26 observed tow instances, the researchers trained six regression models, finding that Poisson and support vector regressors achieved the lowest mean absolute errors of 1.920 and 1.922 barges, respectively. The most predictive features across the models were course entropy and the interaction between trip duration and speed variability.

Why it matters — It demonstrates that barge fleet sizes can be estimated in near real-time from the movement patterns of the towing vessel alone, offering a potential way to bypass the lack of tracking transponders on individual cargo barges for lock and berth scheduling.

Caveat: The predictive models were trained and validated on an extremely small sample size of 26 total observations along a single river corridor.

Transportation Research Interdisciplinary Perspectives · doi

Disruptions as catalysts for change: exploring multi-year temporal dynamics of mobility styles

This study analyzed longitudinal Point of Interest data from 15,720 panelists in the United States between February 2020 and May 2022 to track long-term behavioral adaptations to the COVID-19 pandemic. Using spectral clustering on mobility indicators, the researchers identified six distinct mobility styles, finding that high at-home dwell times persisted into mid-2022 even as high-activity travel styles rebounded past pre-pandemic levels. The analysis also revealed that higher-income individuals were more likely to transition to lower-mobility styles during peak restriction periods.

Why it matters — It provides a structured framework to reduce complex, multi-year activity-travel patterns into six interpretable profiles, demonstrating that major disruptions cause uneven, long-term structural changes in mobility habits across different income groups rather than just temporary shifts.

Caveat: The findings rely on mobile device Point of Interest data as a proxy for complete individual activity-travel behavior, which may not capture all trips or non-POI-based movements.

Transportation · pdf · doi

Perception of tertiary students on the use and importance of campus green spaces at the Kwame Nkrumah University of Science and Technology

A survey of 304 students at the Kwame Nkrumah University of Science and Technology (KNUST) analyzed how campus green spaces are utilized. The study found that students primarily use these areas for academic and social purposes, such as group discussions, meeting friends, and reading, while physical activities are rare, with 82.25% and 77.70% of respondents never using the spaces for cycling or jogging, respectively.

Why it matters — It provides empirical evidence of how African university students interact with campus nature, demonstrating that these spaces function primarily as informal academic and social hubs rather than recreational sports facilities.

Caveat: The findings are based on a self-reported survey at a single university campus, which may not represent student behaviors and green space designs at other institutions.

Discover Cities · pdf · doi

The invisible hand of God: faith-based organizations and the governance of food poverty in Rome

This study investigates the dominant role of faith-based organizations in Rome's food aid system using a qualitative research design that combines documentary analysis, non-participant observation, and semi-structured interviews. It traces how welfare retrenchment and weak public steering allowed these religious groups to become the primary providers of food assistance, establishing a fragmented, charity-based model with low policy integration.

Why it matters — It demonstrates how historically accumulated resources and organizational capacity can lock in non-state, faith-based actors as the default welfare providers, showing that even when policy discourse shifts toward a rights-based approach, actual redistribution of institutional authority and resources remains highly resistant to change.

Caveat: The findings are based on a qualitative case study of a single city with a unique historical and religious context, which may limit generalizability to cities with different welfare traditions or religious landscapes.

Cities · pdf · doi

Post-Flood disruptions to transportation access among people experiencing homelessness

This study analyzes post-disaster transportation access for people experiencing homelessness (PEH) following the 2023 Vermont floods using mixed-methods survey data collected both in-person and online. It compares the short- and long-term mobility disruptions, travel modes, and recovery assistance experiences of individuals who were unhoused prior to the flood against those displaced into homelessness by the disaster.

Why it matters — It provides rare empirical evidence showing that transportation access for unhoused populations remains severely constrained long after physical infrastructure recovers, exposing a mismatch between formal disaster recovery systems and the actual mobility needs of housing-insecure individuals.

Caveat: The study relies on self-reported survey data from a single state-level disaster event, which may limit generalizability to urban areas with different baseline transit systems and homelessness services.

Transportation Research Interdisciplinary Perspectives · doi

Mapping traffic noise annoyance: visualising environmental stress on children’s walk-quality using a psychoacoustic isochrone approach

This study introduces a spatial analysis method that models children's exposure to traffic noise on school commutes by integrating psychoacoustic annoyance as a perceptual metric into pedestrian isochrone networks. The approach was tested across catchments for 18 primary schools in metropolitan Melbourne, Australia, calculating a spatial 'Amenity Cost' to map walk-quality from a receiver-centric perspective.

Why it matters — It shifts the evaluation of pedestrian infrastructure from objective noise emission levels to human psychological perception, allowing urban planners to visually identify quiet, child-friendly routes and test the impact of noise-mitigation interventions.

Caveat: The study serves as a proof-of-concept and does not validate the modeled psychoacoustic annoyance against actual behavioral or psychological responses from the children in the study areas.

Transportation Research Interdisciplinary Perspectives · doi

Multilevel structural equation modelling of walkability in a motorcycle-dominated city: A case study of Da Nang, Vietnam

The study developed and validated a Multilevel Structural Equation Modelling framework that integrates objective field audits with 720 subjective pedestrian surveys across 30 locations in Da Nang, Vietnam. At the individual level, accessibility had the strongest influence on perceived environmental quality, followed by comfort, safety, and pleasurability, explaining 51.7% of the variance. At the location level, objective environmental quality accounted for 71.1% of the variance in perceived quality, while spatial analysis revealed that tourism and recreation zones are prioritized for pedestrian infrastructure over essential service areas.

Why it matters — It provides an empirical test of the Walking Needs Hierarchy tailored to a tropical, motorcycle-dominated Southeast Asian city, demonstrating that comfort plays a much larger role in pedestrian perception than traditional Western, car-centric models assume.

Caveat: The findings are based on a single-city study of Da Nang, which may limit direct generalizability to cities with different climate profiles or lower levels of motorcycle dominance.

Journal of Urban Mobility · doi

Unpacking integration: policy instruments in planning for land-use and transport integration

This study analyzes how policy instruments are used to implement integrated land-use and transport planning across the planning process in four European cities: Groningen, Amsterdam, Helsinki, and Turku. Using a qualitative approach that combines policy document analysis and semi-structured interviews with local and regional practitioners, the research categorizes instruments into procedural and substantive types using the NATO framework. The findings show that these cities primarily rely on information, personnel, and owned property to drive integration, but sector-specific policy packages frequently cause fragmentation during detailed planning phases.

Why it matters — It demonstrates that strategic visions for integrated planning often fail to translate into concrete projects because of a lack of inter-sectoral resource coordination and a failure to account for project-management contexts, pointing to a specific need for stronger procedural instruments that foster multisectoral collaboration.

Caveat: The findings are based on qualitative analysis and interviews within a specific sample of four Northern European cities, which may limit direct applicability to planning systems with different institutional or regulatory frameworks.

Journal of Urban Mobility · doi

Food deserts in context: how site-specific factors reshape food deserts discussion

This review synthesizes global literature on food deserts, tracking their conceptual evolution from simple distance-based metrics to multidimensional frameworks. It documents how modern analyses incorporate affordability, mobility, and local food environments, while evaluating interventions ranging from physical supermarket investments to digital food platforms.

Why it matters — It establishes that the effectiveness of food security interventions depends entirely on site-specific conditions like urban density and social capital, demonstrating that standardized, one-size-fits-all policy approaches are systematically insufficient.

npj Urban Sustainability · pdf · doi

Linking Driver Socioeconomic Profiles to Speeding and Hard-Braking Behaviors: Insights from High-Resolution Vehicle Trajectory Data and Departure-Based Locations

This study analyzed high-resolution vehicle trajectory data to correlate speeding and hard-braking behaviors with the socioeconomic profiles of drivers' departure locations. Using multiscale geographically weighted regression, the analysis mapped driving events against demographic variables such as income, education, age, race, and disability status, finding that both behaviors positively correlate with drivers aged 20–59 and Black or African American populations, while negatively correlating with disability rates and older drivers.

Why it matters — It demonstrates how high-resolution telematics data can be linked to spatial demographics at departure points, offering a way to identify risky driving patterns across different socioeconomic groups without relying solely on lagging crash databases or self-reported surveys.

Caveat: The study relies on the socioeconomic characteristics of a vehicle's departure location as a proxy for the individual driver's actual demographic profile.

Journal of Transportation Engineering Part A Systems · doi

Joint Optimization of CAV Trajectories and Signal Phase Control in Mixed Traffic at Signalized Intersections

The study presents a joint optimization model that combines a multiagent deep Q-network (MADQN) for controlling connected and automated vehicle (CAV) trajectories with an adaptive signal control strategy based on queue pressure. Tested via simulation on a real-world multidirectional intersection, the model integrates real-time vehicle states and signal phases to optimize a global reward of average delay and energy consumption.

Why it matters — It demonstrates that intersection delays can be reduced even at low CAV penetration rates, with compounding benefits to both delay and energy efficiency as the proportion of automated vehicles increases.

Caveat: The performance of the optimization model is demonstrated solely through simulation experiments rather than a physical deployment.

Journal of Transportation Engineering Part A Systems · doi

The modern face of the state in postwar urban reconfiguration and post-earthquake planning: the case of the Sakarya Government House, Türkiye

This study traces the life cycle of the Sakarya Government House in Turkey, a prominent example of postwar International Style architecture, across three historical phases: its initial construction as a symbol of state modernization, its structural reinforcement following the 1967 Akyazı earthquake, and its eventual demolition in 2005 after the 1999 Marmara earthquake. The analysis details how central-local planning conflicts, neoliberal zoning expansions, and post-disaster urban policies shaped the building's physical trajectory and ultimate removal.

Why it matters — It demonstrates how state-sponsored modernist architecture in provincial Anatolia served as a physical instrument of national political alignment, and how subsequent seismic disasters were leveraged by neoliberal urban policies to justify the demolition and fragmentation of public heritage.

Caveat: The study is a historical qualitative case study focused on a single civic building, which may limit its direct applicability to non-state or residential architectural histories.

Planning Perspectives · doi

Georeferencing Non-Gazetteered Place Names using Biological Specimen Records

This study develops and compares deterministic, probabilistic, and Large Language Model (LLM) methods to georeference historical, vernacular, or colloquial place names absent from modern gazetteers. Using digitized specimen records from the Allan Herbarium in New Zealand, the approach extracts and inverts spatial relation terms from repeated occurrences of these names to constrain their locations. On a pseudo-non-gazetteer benchmark, the probabilistic model achieved the highest accuracy with a median error of 1.43 km, outperforming the LLM's median error of 1.80 km.

Why it matters — It establishes a method to recover lost historical and informal geography by leveraging the spatial context embedded in natural history collections, demonstrating that traditional probabilistic modeling still outperforms LLMs when high spatial precision is required.

Caveat: The performance of the georeferencing methods was evaluated on a constructed pseudo-benchmark rather than directly on completely unmapped historical place names.

arXiv · pdf

Walkable to Whom? Capturing Subjective Variability in Walkability Perception Using Multimodal Deep Learning

The study introduces a dataset of 29,870 walkability ratings from 1,196 respondents paired with sidewalk-view imagery across urban, suburban, and regional Australia, and develops a user-conditioned multimodal deep learning framework that fuses visual features with individual rater attributes. The user-conditioned model improved rank agreement with observed ratings by 65% over an image-only baseline, achieving a quadratic weighted kappa of 0.47 compared to 0.29. Additionally, a viewpoint comparison revealed that sidewalk-view images receive significantly higher walkability ratings than matched, vehicle-mounted street-view images.

Why it matters — It demonstrates that walkability is not perceived uniformly and provides a method to model subjective, user-specific environmental assessments rather than relying on aggregated, observer-independent scores. It also establishes that the choice between sidewalk-level and street-level imagery systematically biases walkability survey results.

Caveat: The predictive performance of the user-conditioned model, while a substantial improvement over the baseline, remains moderate with a quadratic weighted kappa of 0.47.

arXiv · pdf

Evidence on Slowing Progress in Longevity -- Is it Misleading?

This study evaluates claims of slowing human longevity progress by applying a Bayesian change-point framework to annual life expectancy gains across nearly two centuries of data from record-holding countries. The analysis reveals that while the variance of annual gains has shifted through distinct historical regimes, the expected mean annual gain has remained unchanged, with the most regular progress occurring in recent decades.

Why it matters — It demonstrates that the recent flattening of cumulative life expectancy records is a statistical artifact of a low-variance regime rather than a decline in the underlying pace of improvement. This challenges the prevailing assumption that a biological ceiling to human lifespan is approaching, reframing the best-practice frontier as an empirical benchmark of historical conditions rather than an absolute limit.

arXiv · pdf

Vehicle routing problem using deep reinforcement learning - A case study about truck planning in the industry

This study applies deep reinforcement learning (DRL) to optimize vehicle routing across three distinct external truck network logistics use cases. The DRL agent's optimized routes achieved a total cost reduction of over 10% compared to baseline routing results.

Why it matters — It demonstrates that deep reinforcement learning can successfully navigate the complex constraints, information opacity, and human behavioral uncertainties of real-world industrial logistics, outperforming traditional baseline routing methods.

arXiv · pdf

HeatCast: A Benchmark for Neighborhood-Scale LST Forecasting across 124 U.S. Cities

This study introduces HeatCast, a Landsat-based benchmark dataset for monthly Land Surface Temperature (LST) forecasting at 30 m resolution across 124 U.S. cities from 2013 through June 2025. The dataset includes monthly tiles containing LST, elevation, surface reflectance, spectral indices, broadband albedo, and Local Climate Zone labels. Evaluating two deep learning models on next-month forecasting, the Earthformer model achieved a 7.74 K RMSE, outperforming a CNN+LSTM model which scored 10.42 K.

Why it matters — It establishes the first shared, high-resolution benchmark for neighborhood-scale urban heat forecasting, demonstrating that predicting LST from non-temperature environmental channels yields higher accuracy than relying on historical temperature data alone.

arXiv · pdf

Are Visual Place Recognition Models Recognizing Places or Conditions? Distractor-Augmented Evaluation and Condition Suppression

The study evaluates eleven Visual Place Recognition (VPR) methods across six datasets using a new metric, Distractor-Augmented Recall (DAR), which measures how often models mistakenly match images based on shared environmental conditions rather than actual geographic locations. To address this vulnerability, the researchers applied two condition suppression techniques, Iterative Nullspace Projection (INLP) and Concept Erasure (LEACE), to remove weather, seasonal, and illumination features from the image descriptors.

Why it matters — It establishes that standard VPR evaluation metrics fail to account for condition-based distractors common in crowdsourced map databases, and demonstrates that explicitly suppressing environmental condition data improves retrieval robustness without degrading standard performance.

arXiv · pdf

Data collection from highways: a geometric, class-agnostic approach to embedded vehicle counting

This study develops and tests a class-agnostic, purely geometric vehicle counting pipeline designed to run on low-power Single Board Computers like the Raspberry Pi. The system uses background subtraction to detect moving blobs and applies two counting rules across a virtual line: a constant average speed rule with an analytical accuracy of about 86%, and a self-calibrating rule that reconstructs lane geometry to count vehicles and estimate their speed. Tested across four videos and a field deployment, the self-calibrating method achieved 83.3% to 100% accuracy, outperforming a blob-tracking baseline (91% versus 37.5%) under the same hardware constraints.

Why it matters — It demonstrates that reliable traffic monitoring and speed estimation can be achieved on cheap, edge-compute hardware without the need for pre-trained deep learning models, annotated datasets, or high power budgets. This provides a viable deployment path for tracking rare vehicle classes, protecting privacy, and bootstrapping training data in resource-constrained settings.

Caveat: The system's performance is highly sensitive to specific geometric thresholds, showing a collapse in accuracy below certain resolution and frame rate limits where vehicles alias past the detection line.

arXiv · pdf

EpiFlow: A framework for improving the utility of wastewater signals for disease forecasting

The researchers developed EpiFlow, a framework that processes wastewater viral loads, analyzes their causal relationship with disease burden using entropy and causality tests, and integrates them into a time-varying forecasting model. The framework was evaluated by forecasting COVID-19 hospital admissions across Virginia and its health regions, simulating the effects of reporting delays and varying disease prevalence.

Why it matters — It demonstrates that incorporating wastewater data improves hospital admission forecast accuracy and increases forecast coverage by 20 percentage points, proving that wastewater signals remain reliable predictors even during low-prevalence periods or when data reporting is delayed.

Caveat: The framework's performance was evaluated using a single state's health regions and specifically for COVID-19, meaning its utility for other pathogens or different regional governance structures remains to be tested.

arXiv · pdf

Towards Sustainable Artificial Intelligence: A Comprehensive Review and Comparative Analysis of Deep Learning Models' Carbon Footprint

This paper reviews Green AI optimization techniques and carbon measurement tools, and conducts an empirical evaluation of six deep learning models performing a multi-label classification task on a CPU-based setup. The experiment quantifies carbon emissions across different lifecycle stages, finding that the training phase is the primary source of emissions and that increased model complexity does not yield proportional gains in accuracy.

Why it matters — It demonstrates that model selection can be optimized for environmental impact without necessarily sacrificing predictive performance, providing a basis for integrating carbon footprint metrics directly into AI system design.

Caveat: The empirical evaluation is limited to a CPU-based experimental setup for a single multi-label classification task, which may not represent the emission dynamics of GPU- or TPU-accelerated environments.

arXiv · pdf · doi

Educational Short Videos: Bibliometric Trends, Thematic Structure, and Operationalisation

This study mapped the research landscape of educational short videos by analyzing 2,169 records from Web of Science and Scopus up to June 2026. Using bibliometric analysis, non-negative matrix factorization topic modeling, and structured content analysis, it identified 16 primary topics, with 'Skill Development in Educational Contexts' forming the structural core. The analysis also revealed that while knowledge, achievement, engagement, and motivation are the most common outcome domains, only 511 records specified a representative video duration, showing no consistent numerical threshold for what constitutes 'short'.

Why it matters — It provides a unified working definition of educational short videos based on their multimedia design and learning purpose rather than arbitrary duration limits. This helps researchers and educators synthesize evidence across highly fragmented disciplines, platforms, and methodologies that previously used the same label for vastly different resources.

Caveat: The study is limited to literature indexed in Web of Science and Scopus, which may omit relevant research published in other databases or non-academic platforms.

arXiv · pdf

Integrating spectral and morphological plant features with decision-tree models for early-season cotton biomass and nitrogen status estimation from multi-year UAV data

The study developed and evaluated machine learning models to estimate cotton dry biomass weight, plant nitrogen uptake, and plant nitrogen concentration using multispectral and morphological data from unmanned aerial vehicles. Drawing on a three-year field study in the Texas Coastal Plains, the researchers compared simple regression, multiple linear regression, random forest regression, and extreme gradient boosting models. The random forest and extreme gradient boosting models achieved the highest validation accuracies, yielding R2 values up to 0.88 for biomass and 0.86 for nitrogen concentration.

Why it matters — It establishes a reliable method for non-destructive, early-season monitoring of crop nitrogen status, allowing farmers to make precise decisions on fertilizer timing and application rates before flowering.

Caveat: The critical nitrogen dilution curve and resulting nutrition index were calibrated specifically for high-yielding, medium-to-tall cotton varieties in the Texas Coastal Plains, which may limit direct application to other cotton varieties or growing regions.

arXiv · pdf

LMM Modality Transfer: A Pre-requisite for Autonomous GIS Agents

This study introduces a modality transfer task to evaluate how well Large Multimodal Models (LMMs) translate spatial information between visual and textual formats. The experiment requires one OpenAI LMM instance to describe a grid of colored squares in text, and a second instance to reconstruct the original visual grid based solely on that description.

Why it matters — It establishes a fundamental benchmark for autonomous GIS agents, demonstrating that current state-of-the-art LMMs fail to reliably preserve simple spatial layouts when moving between text and image modalities, revealing a critical bottleneck in multi-modal alignment.

Caveat: The evaluation is limited to synthetic grids of colored squares rather than real-world geographic maps or complex spatial datasets.

arXiv · pdf · doi