2026-09-10

7 arXiv categories· 96 journals· 2600 candidates — 4 worth your time· 235 without an open abstract

Headline

Freeway emission reductions, geospatial foundation models, and LLM next-location prediction accuracy

Today, together

tag shift
no tag ran above its 30-day average
on the same shoulders
4 of the day's papers cite Travel demand and the 3Ds: Density, diversity, and design (Robert Cervero, Kara M. Kockelman 1997) counted over the 109 of this day's 239 named papers whose reference lists we have — standing on the same older work is not the same as being about the same thing
first appearance
Department of Mathematical Sciences none of them on a paper we published in the 90 days before this issue

Two counts, not one counted twice: every row here but “on the same shoulders” counts the papers published above, and that one counts every paper this day named — those and the ones under Also published today, whose abstracts we could not read.

"Emission reduction potential of freeway stop-and-go wave smoothing" establishes a counterfactual wave-smoothing benchmark using a quadratic program to reconstruct smooth vehicle trajectories. "Assessing extreme flood impacts on urban rail transit: A passenger-oriented, resilience-informed framework" develops a passenger-oriented framework to classify journey impacts into six categories during transit disruptions. "Who You Are Adds Nothing Detectable to Where You Go Next: Sociodemographic Conditioning in LLM Next-Location Prediction" evaluates the predictive value of sociodemographic attributes in LLM-based next-location prediction using a closed-set benchmark of 5,000 Shenzhen residents.

published: Transportation Research Part D Transport and Environment

Junyi Ji, Derek Gloudemans, Gergely Zachár, et al.
Transportation Research Part D Transport and Environment · was arXiv

Vanderbilt University

Traffic control and management · Vehicle emissions and performance · Transportation Planning and Optimization · Vanderbilt University · Integrated Software (United States) · Rutgers, The State University of New Jersey

Emission reduction potential of freeway stop-and-go wave smoothing

The study establishes a counterfactual wave-smoothing benchmark using a quadratic program with fixed boundary conditions and a maximum-gap constraint of 0.1 miles to reconstruct smooth vehicle trajectories. Using the MOVES model on nine weeks of weekday peak-period empirical data from the I-24 MOTION testbed, the researchers estimated average CO2 reductions of 9.80% to 14.08% across lanes for passenger cars, alongside reductions of up to 25.70% in CO, 29.76% in HC, and 36.03% in NOx.

Why it matters — It quantifies the realistic, large-scale emission reduction potential of smoothing freeway traffic waves under practical gap constraints, revealing that targeting trucks yields roughly double the emission benefits of targeting passenger cars.

Caveat: The emission reductions are estimated using a counterfactual simulation model rather than observed from physical, real-time vehicle control deployments.

Transportation Research Part D Transport and Environment · doi

preprint

Wei Bi
arXiv physics.soc-ph

Infrastructure Resilience and Vulnerability Analysis · Flood Risk Assessment and Management · Evacuation and Crowd Dynamics · London

Assessing extreme flood impacts on urban rail transit: A passenger-oriented, resilience-informed framework

This study develops a passenger-oriented framework to classify journey impacts into six categories, accounting for rerouting, alternative station use, and delay thresholds during disruptions. The framework was tested using hourly dynamic simulations of 15 London rail transit lines under 30-year, 100-year, and 1,000-year flood scenarios, revealing that unsatisfied travel demand is primarily caused by completely unavailable routes rather than unacceptable delays.

Why it matters — It shifts flood impact assessments from simple delay-based metrics to a model that captures completely blocked journeys and passenger rerouting behavior, showing that rerouting is the dominant factor in travel disruption during severe events.

Caveat: The framework's behavioral assumptions and simulation parameters were demonstrated using a single transit system in London.

arXiv · pdf

preprint

Xin Wang, Paraic Carroll, Kerry Nice, et al.
arXiv cs.CY

Human Mobility and Location-Based Analysis · Transportation and Mobility Innovations · Urban Transport and Accessibility · Shenzhen

Who You Are Adds Nothing Detectable to Where You Go Next: Sociodemographic Conditioning in LLM Next-Location Prediction

The study evaluates the predictive value of sociodemographic attributes in LLM-based next-location prediction using a closed-set benchmark of 5,000 Shenzhen residents. Across three LLMs and a supervised reranker, adding age, gender, occupation, and income to prompts changed top-1 accuracy by only -0.8 to +0.5 percentage points, showing no detectable predictive gain. Conversely, mobility trajectories alone could predict income with an AUC of 0.708, while candidate destination sampling methods shifted top-1 accuracy by up to 22.3 percentage points.

Why it matters — It demonstrates that adding demographic profiles to LLM mobility prompts does not improve next-location prediction accuracy, challenging a common practice in agent-based travel simulations. It also reveals that reported LLM performance in mobility tasks is highly sensitive to how candidate destinations are sampled rather than the model's actual predictive capability.

Caveat: The findings are based on a closed-set benchmark where models rank 100 candidate destinations, which may not fully capture open-ended or generative next-location prediction scenarios.

arXiv · pdf

preprint

Ayush Khot, Wen Zhou, Shaowen Wang
arXiv cs.LG

Land Use and Ecosystem Services · Impact of Light on Environment and Health · Remote-Sensing Image Classification · satellite imagery

Applying foundation model embeddings towards urban livability evaluation

This study evaluates how physical features encoded within geospatial foundation models—specifically AlphaEarth, AnySat, and TerraMind—can predict urban livability indicators. The researchers developed a systematic framework to identify the most predictive geospatial features from satellite imagery and analyze how different spatial data types influence these predictions while accounting for complex spatial dependencies.

Why it matters — It establishes a principled methodology to extract actionable livability metrics in data-scarce regions, providing a way to substitute high-resolution satellite imagery and foundation model embeddings for missing ground-level socioeconomic observations.

Caveat: The abstract does not specify the exact livability indicators used, the geographic location of the evaluation, or the baseline performance metrics of the models.

arXiv · pdf · doi

Also published today

These appeared today in journals we track. Their abstracts are not openly available, so we cannot summarise them.