2026-10-18

7 arXiv categories· 96 journals· 2812 candidates — 2 worth your time· 51 without an open abstract

Headline

Crowdsourced disaster-photo geolocalization via Sinkhorn re-ranking and vision-language models

Today, together

tag shift
no tag ran above its 30-day average
on the same shoulders
no foundation work was cited by four of the day's papers counted over the 47 of this day's 53 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
no institution here was new to the last 128 days none of them on a paper we published in the 128 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.

codedatapreprint

Hao Li, Wenping Yin, Fabian Deuser, et al.
arXiv cs.AI

Geographic Information Systems Studies · Disaster Management and Resilience · Remote Sensing and Land Use · street view imagery · satellite imagery · openstreetmap

A Decade of PhotoMappers: A Longitudinal Study of Crowdsourced Disaster Photos Geolocalization with Vision-Language Models and Geospatial Reasoning

This study analyzes a ten-year dataset from the PhotoMappers crowdsourced disaster-mapping initiative and tests GeoAI models on cross-view image geolocalization and geographic reasoning. Fusing volunteered geographic information, street-view imagery, and remote sensing imagery via Sinkhorn re-ranking achieved a top-1 retrieval recall rate of 40.90%, with yearly performance fluctuating between 25.00% and 65.07%. Evaluation across 30 geographic indicators showed that the Qwen3.6-27B vision-language model achieved 67% agreement with human geographic reasoning, outperforming InternVL3-8B at 51%, though both models struggled to interpret fine-grained, localized place cues.

Why it matters — Automated GeoAI models can scale up the retrieval of candidate locations for disaster imagery, but human interpretation remains essential for verifying ambiguous, context-dependent local features.

arXiv · pdf · code

datapreprint

Davide Torre, Miguel Artigues, Benjamin Faucher, et al.
arXiv physics.soc-ph

agent-based model · census data

Synthetic populations for epidemic, demographic and mobility modeling

The authors developed a harmonized dataset and pipeline generating synthetic populations across Spain, France, and Italy at the census-tract level. The pipeline integrates census, commuting flow, and economic activity data to reproduce age-sex distributions, household structures, population density, and municipal workplace patterns.

Why it matters — Researchers can now run standardized, cross-border individual-based simulations using identical data structures for three major European nations, bypassing the need to reconcile disparate country-specific census formats.

arXiv · pdf

Also published today

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