2026-09-15

7 arXiv categories· 96 journals· 2482 candidates — 7 worth your time· 105 without an open abstract

Headline

Pedestrian volume estimation from open GIS data, using gradient boosting models

Today, together

tag shift
no tag ran above its 30-day average
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) counted over the 104 of this day's 112 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
San Diego State University; Texas Health and Science University none of them on a paper we published in the 95 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.

"Estimating Pedestrian Volumes from GIS-Derived Built-Environment Features" developed a machine learning pipeline to predict peak pedestrian volumes at urban intersections in Portland, Oregon, using open GIS data. "Global maps of travel time to emergency and tertiary hospitals" mapped global drive-time access to over 217,000 hospitals using OpenStreetMap routing and population data. "Linking COPD Prevalence with Income Distribution" developed a geographically weighted penalized compositional regression model to analyze income distribution data as compositional covariates while detecting regional clusters with shared regression effects.

codedatapreprint

Bahareh Golchin, Banafsheh Rekabdar, Sirisha Kothuri, et al.
arXiv cs.LG

machine learning · gradient boosting · traffic sensor data

Estimating Pedestrian Volumes from GIS-Derived Built-Environment Features: A Machine Learning Framework

A machine learning pipeline was developed to predict 2-hour PM peak pedestrian volumes at 101 urban intersections in Portland, Oregon, using open GIS data on built-environment, land-use, and street-network features. The best-performing model, a histogram-based gradient boosting model with Poisson loss and L1 Lasso feature selection, reduced cross-validated RMSE by 12% (from 89.8 to 78.7) and holdout RMSE by 19% (from 108.0 to 87.9) compared to a traditional Negative Binomial baseline.

Why it matters — It demonstrates that combining open GIS data with gradient boosting models can significantly improve the accuracy of pedestrian volume estimation across an urban network, reducing the reliance on expensive manual counts for safety planning.

Caveat: The predictive model was trained and validated on a relatively small sample of 101 intersections within a single city.

arXiv · pdf

published: Annals of the American Association of Geographers

Jingwen Deng, Shujie Ma, Sergio J. Rey, et al.
Annals of the American Association of Geographers · was arXiv

Texas Health and Science University

Spatial and Panel Data Analysis · Geochemistry and Geologic Mapping · Data-Driven Disease Surveillance · United States · Texas Health and Science University · San Diego State University

Linking COPD Prevalence with Income Distribution: A Spatial Heterogeneous Compositional Regression via Geographically Weighted Penalized Approach

The researchers developed a geographically weighted penalized compositional regression model that handles income distribution data as compositional covariates while detecting both contiguous and noncontiguous regional clusters with shared regression effects. They applied this framework to analyze the relationship between household income distribution brackets and chronic obstructive pulmonary disease (COPD) prevalence across the United States, utilizing a pairwise fusion penalty and the minimax concave penalty to capture abrupt spatial variations.

Why it matters — The model overcomes the limitations of traditional spatial regressions that assume smooth, continuous geographic transitions, allowing researchers to identify shared socioeconomic-health relationships across geographically isolated regions that share similar structural characteristics.

Annals of the American Association of Geographers · doi

codepreprint

Johan Emilsson
arXiv physics.soc-ph

Global maps of travel time to emergency and tertiary hospitals

The study mapped global drive-time access to 217,841 hospitals using OpenStreetMap routing and GHS-POP 2025 population data, categorizing facilities into nested capability tiers. It found that while 96.2% of the global population lives within a one-hour drive of any hospital, this share drops to 78.5% for hospitals with documented emergency departments, and to 40.3% for tertiary care facilities. Approximately 1.45 billion people live within an hour of a basic hospital but lack access to an emergency hospital, concentrated heavily in South Asia, East Asia, and Nigeria.

Why it matters — It demonstrates that treating all healthcare facilities as equal severely overestimates actual access to life-saving, time-critical emergency and specialized care, providing a more realistic baseline for global health planning.

Caveat: The classification of tertiary care facilities relies on proxies like helipads or university names in OpenStreetMap, which may not capture all high-level medical centers uniformly across different regions.

arXiv · pdf · code

preprint

Bibandhan Poudyal, Mariana Macedo, Ronaldo Menezes, et al.
arXiv physics.soc-ph

network analysis

The hierarchical organization of urban social overlap: How shared social spaces diverge from physical mobility

The study constructs co-presence overlap networks across 13 Brazilian cities to analyze urban social organization using two new metrics: colocation, which measures recurrent co-presence at specific sites, and co-connectedness, which measures how locations share co-present pairs. The analysis reveals that social centers differ in both quantity and location from mobility centers, and that co-connectedness forms a non-random hierarchical structure where locations of similar activity levels preferentially share recurrent pairs.

Why it matters — It demonstrates that physical mobility patterns do not dictate social overlap, showing that cities with nearly identical mobility structures can possess vastly different spatial organizations of recurrent social interaction.

Caveat: The findings are based on urban areas within a single national context, which may limit generalizability to countries with different urban forms or social dynamics.

arXiv · pdf

preprint

Miao Zeng, Roberto Murcio, Camilo Vargas-Ruiz, et al.
arXiv physics.soc-ph

street network data

Unveiling healthcare-access inequality in Ghana using a multiscale approach

This study maps healthcare access across Ghana by combining street-level accessibility with road network hierarchy using an extended percolation divergence tree framework. It reveals that approximately 25% of the national population resides more than 5 kilometers from the closest healthcare facility. The analysis identifies distinct spatial patterns of deprivation, showing that poor access manifests as isolated fringe pockets in well-connected monocentric regions, but spans entire nested subsystems in poorly connected polycentric regions.

Why it matters — It shifts the analysis of healthcare inequality from simple proximity mapping to a multiscale connectivity approach, demonstrating how regional road network structures dictate whether healthcare isolation is a localized issue or a systemic regional barrier.

arXiv · pdf

preprint

Shangyu Lou, Ziqi Cui
arXiv cs.SI

New York City

Enhancing Human Mobility Prediction with Spatially Aware LLM-based Multi-Agent Systems

The researchers developed a multi-agent LLM framework that decomposes next-POI prediction into pattern extraction, spatial reasoning, and decision synthesis. Tested on the NYC benchmark dataset using two LLM backbones, the framework incorporates geographic distance, road network distance, and neighborhood affiliation to constrain predictions. The system achieved up to a 493% improvement in Hit@1 and a 37% relative improvement in Hit@5 compared to baseline methods.

Why it matters — It demonstrates that LLMs can overcome their inherent spatial reasoning limitations when structured as multi-agent systems that explicitly process real-world spatial constraints alongside behavioral history.

Caveat: The framework's performance was evaluated using only a single benchmark dataset from New York City.

arXiv · pdf · doi

preprint

Ting-Hsuan Chang, Minghao Qiu, Yaguang Wei, et al.
arXiv stat.AP

weather and climate data · United States

Evaluating differential air quality impacts of prescribed fire and wildfire in the United States

Using fire records from California, Florida, and Georgia spanning 2006 to 2020, this study applied a causal forest model to compare the air quality impacts of prescribed burns and wildfires. The analysis revealed that prescribed fires generate over six times less population-level smoke exposure per square kilometer burned than wildfires. Additionally, the smoke exposure from prescribed burns was found to be lower in areas characterized by higher relative humidity and precipitation.

Why it matters — It provides the first large-scale, multi-state quantitative evidence proving that preventive prescribed burns significantly reduce population-level smoke exposure compared to uncontrolled wildfires, offering specific environmental conditions under which these burns minimize public health risks.

Caveat: The study relies on data from three specific states, which may not fully represent the fire dynamics and meteorological conditions of other regions in the United States.

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.