2026-08-25

7 arXiv categories· 96 journals· 2212 candidates — 7 worth your time· 93 without an open abstract

Headline

Automated georeferencing of historical maps, using spatially explicit AI

Today, together

tag shift
no tag ran above its 30-day average
shares references
no items share references today counted over the 7 papers published above
on the same shoulders
4 of the day's papers cite Accessibility evaluation of land-use and transport strategies: review and research directions (Karst Geurs, Bert van Wee 2003); 4 cite Travel and the Built Environment (Reid Ewing, Robert Cervero 2010) counted over the 95 of this day's 100 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
Austrian Research Institute for Artificial Intelligence; CRC for Spatial information; Fachhochschule Salzburg; and 5 more none of them on a paper we published in the 74 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.

datapublished: International Journal of Geographical Information Systems

Yanan Wu, Yalin Yang, Chengbin Deng, et al.
International Journal of Geographical Information Systems

University of Central Arkansas

Geographic Information Systems Studies · Historical Geography and Cartography · Data Visualization and Analytics · convolutional neural network · Chicago · University of Central Arkansas

‘Give me a point and I will place the map’: automated georeferencing historical maps using spatially explicit AI

This study develops a georeferencing framework that combines vision-language models, convolutional neural networks, and graph-based reasoning to align historical maps using spatial semantics and geometry. Tested on Sanborn fire insurance maps of Chicago, the method successfully georeferenced 84.3% of the map sheets, maintaining an 80% success rate with fewer than three ground control points (GCPs) and a 74.3% success rate under uneven GCP distributions.

Why it matters — The framework shifts automated georeferencing from a rigid geometric fitting task to a spatially contextual inference problem. This allows researchers to accurately digitize historical map collections even when reference points are highly sparse or poorly distributed, a common bottleneck in historical GIS workflows.

Caveat: The evaluation and proof of concept were restricted to Sanborn maps from a single city, Chicago.

International Journal of Geographical Information Systems · doi

published: Annals of the American Association of Geographers

Zhiyong Zhou, Song Gao, Qianheng Zhang, et al.
Annals of the American Association of Geographers · was arXiv cs.AI

Zhejiang University

Human Mobility and Location-Based Analysis · Urban Transport and Accessibility · Transportation and Mobility Innovations · United States · Zhejiang University

Quantifying geographic domain shift to decouple the geospatial transferability of human mobility flow generation models

This study evaluated four human mobility generation models across 2,265 US counties using census-tract commuting data, introducing mutual information and spatial shift metrics to quantify geographic domain shift via linear mixed-effects regression. The analysis reveals significant spatial heterogeneity and directional asymmetry in cross-regional model performance, proving that domain shift components independently drive model transferability.

Why it matters — Researchers can now isolate geographic structural differences from algorithm performance when evaluating GeoAI models, preventing misleading transferability assessments across disparate regions.

Caveat: The evaluation relies exclusively on census-tract-level US commuting data, which may not capture finer-grained urban mobility patterns or non-work travel dynamics.

Annals of the American Association of Geographers

published: Transportation

Laurent Cazor, Lawrence Christopher Duncan, Corin Edward Staves, et al.
Transportation

Danish Academy of Technical Sciences

Urban Transport and Accessibility · Transportation Planning and Optimization · Economic and Environmental Valuation · Danish Academy of Technical Sciences · Technical University of Denmark

Formulation of energy exertion in bicycle networks and estimation of its impact on route choices

A physics-based energy model was developed to calculate expected energy expenditure in Joules and energy intensity in Joules per metre for individual bicycle network links, accounting for slope, cruising speeds, and intersection stopping probabilities. The model was calibrated using a large-scale GPS dataset from the Greater Copenhagen Area and integrated into Multinomial Logit, Latent Class, and Mixed Logit route choice models. The analysis revealed that cyclists have a statistically significant dispreference for high-energy routes, with a distinct subgroup willing to take major detours to avoid high-intensity links.

Why it matters — It replaces proxy measures of cycling effort with a physically grounded, individual-specific metric of energy expenditure. This allows planners to evaluate how the combined effects of topography and intersection stop-and-go dynamics directly influence cyclist route choices.

Caveat: The empirical calibration and route choice modeling are demonstrated using data from a single region, the Greater Copenhagen Area, which may limit immediate generalizability to cities with vastly different cycling cultures or topographies.

Transportation · pdf · doi

published: Computational Urban Science

Marjorie Mattes, Dorian Arifi, Bernd Resch
Computational Urban Science

University of Salzburg

Human Mobility and Location-Based Analysis · Urban Transport and Accessibility · Data-Driven Disease Surveillance · text analysis · spatial analysis · social media data

Understanding pre- and post-COVID urban neighborhood dynamics: a study analyzing georeferenced tweets for New York City

This study analyzed geotagged Twitter data from New York City between 2018 and 2022 to track urban behavioral shifts across five domains: Transportation, Retail Activity, Cultural/Social Activity, Healthcare, and Work/Remote Work. The researchers developed a classification workflow combining Wikipedia-derived keyword filtering with BERTopic modeling to categorize posts and measure spatial and temporal deviations from long-term baselines. The analysis revealed distinct patterns of disruption and partial recovery, particularly showing sharp declines and intensifications in transportation and cultural activities.

Why it matters — It demonstrates how geotagged social media can serve as a bottom-up sensor to capture how citizens perceive and experience urban disruptions, moving beyond traditional top-down mobility traces that only record physical movement.

Caveat: The method proved less effective for domains with sparse data, such as Healthcare and Work/Remote Work, which produced fragmented patterns that limited clear interpretation.

Computational Urban Science · doi

datapreprint

Maciej Adamiak, Sascha Fendrich, Julian Psotta, et al.
arXiv cs.CV

Automated Road and Building Extraction · Traffic Prediction and Management Techniques · Traffic and Road Safety · openstreetmap · University of Łódź · GeoInformation (United Kingdom)

Vehicle speed dataset for the major European road network derived from Sentinel-2 imagery, 2022-2026

This dataset provides individual vehicle speed observations across the European E-road network for the years 2022 to 2026, extracted from Copernicus Sentinel-2 Level-2A optical satellite imagery. By measuring the spatial displacement of moving vehicles across the blue, green, and red spectral bands caused by short sensor acquisition delays, the processing pipeline reconstructs per-vehicle trajectories and calculates ground speeds. The resulting dataset is formatted as GeoPackage files containing trajectory geometries, headings, estimated speeds, timestamps, and quality indicators.

Why it matters — It establishes a continental-scale empirical speed dataset derived entirely from satellite observations, offering a standardized way to analyze traffic patterns, model transport, and calibrate road network attributes across national borders without relying on ground-based sensors or proprietary GPS probe data.

Caveat: The dataset is restricted to the major European E-road network and its accuracy is bound to the specific acquisition times and cloud-free passes of the Sentinel-2 satellites.

arXiv · pdf

preprint

Kishu Gupta, Deepika Saxena, Ashutosh Kumar Singh, et al.
arXiv cs.CR

graph neural network

Neighbor-embedded Graph Neural Network-based Crowd Delivery Traffic Management in Smart City

The researchers developed the Neighbor-Embedded Graph Neural Network-based Crowd Delivery Traffic Management (NeCDM) model, which combines a traffic congestion prediction unit with a vehicle selection unit to optimize crowd delivery requests. Tested against smart city parameters, the model reduces L1 loss by up to 4.03%, L2 loss by up to 16.66%, and computation time by 7.64% compared to baseline methods.

Why it matters — It shifts the focus of smart transportation management from merely predicting congestion to actively mitigating it by intelligently routing and selecting crowd-sourced delivery vehicles.

Caveat: The performance improvements are evaluated purely on computational loss metrics and processing time rather than real-world traffic flow or physical deployment outcomes.

arXiv · pdf · doi

preprint

Valeria Legaria-Santiago, Amadeo Arguelles, Magdalena Saldana-Perez, et al.
arXiv cs.LG

machine learning · random forest · gradient boosting · London

Leveraging Remote Traffic Data for Local Air Pollutant Estimation: A Scenario-Based Machine Learning Study Across London Monitoring Sites

This study evaluated four tree-based machine learning models (Random Forest, Extra Trees, LightGBM, and XGBoost) across six predictor scenarios to estimate NO2, PM10, PM2.5, and O3 concentrations across multiple monitoring sites in London. Using remote traffic data, meteorological variables, and temporal data alongside measurements from up to four neighboring stations, the models improved NO2 estimation accuracy, reducing the RMSE from a range of 9.73–11.66 μg/m³ without traffic data to 8.72–11.52 μg/m³ when traffic data was included.

Why it matters — It demonstrates that remotely acquired traffic data can substitute for physical monitoring infrastructure in traffic-dominated urban environments, as SHAP analysis reveals its predictive contribution to NO2 estimation is comparable to actual pollutant measurements from neighboring stations.

Caveat: The performance improvements and variable importance metrics were evaluated specifically within London's monitoring network, which may limit generalizability to cities with different traffic patterns or sparser sensor distributions.

arXiv · pdf

Still cited

Waldo Tobler (1970), A Computer Movie Simulating Urban Growth in the Detroit Region 1 of today's items cite it · 30 of 3240 in the archive stand on it

Also published today

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