31 Geospatial Papers, Three Working Demonstrations
Most of the June 2026 digest is academic and off-target for disaster GIS. A handful map directly onto Red Cross work — damage assessment, situational awareness, volunteer deployment, vulnerability mapping. Below, the three most operationally relevant techniques run live in your browser, framed for American Red Cross use.
The demos use illustrative synthetic data so the method — not any one dataset — is what you see working.
Source: Milan Janosov, Substack digest, received 30 June 2026. · Three demos below correspond to papers #6, #8, and #31.
Relevant to Red Cross
Five papers earn their place in disaster operations. Three of them are demonstrated live further down.
Composite2Change
Landsat time-series change detection on Google Earth Engine that detects, dates, and measures disturbance and recovery, per pixel per year.
Global Street Networks
OSM-derived network models + node-level betweenness for 10,351 urban areas worldwide. Downloadable now.
Urban Vulnerability Framework
A multi-indicator geospatial framework ranking vulnerability across 1,000 cities. (Topic is air pollution; the method transfers.)
LandSegmenter
A land-use / land-cover foundation model that transfers zero-shot with weak labels across modalities and class taxonomies.
Wildfires in 2025
Record-low burned area but deadlier WUI fires; Canada's third boreal extreme; LA January fires ~US$140B, 31 dead, ~12,000 homes.
Demo · Betweenness Centrality for Deployment Routing
Every city in the dataset ships with node-level betweenness centrality — a measure of how many shortest paths run through each intersection. High-betweenness nodes and edges are the choke points: lose a bridge or arterial and movement across the city collapses. This demo computes real betweenness (Brandes' algorithm) on a synthetic riverside city.
Two bridges cross the river. Betweenness concentrates on them and on the arterials feeding them — exactly the segments a deployment plan must keep open or route around. The real dataset gives you this for any of 10,351 urban areas, no synthesis required.
Demo · Before / After Damage Assessment
C2C builds gap-free annual composites from the Landsat archive, then detects, dates, and measures disturbance per pixel. The operational read for the Red Cross: drag the divider across a pre- and post-event scene, toggle the detected-change overlay, and read magnitude — the same signal that drives a damage-assessment triage map.
Red outlines mark cells C2C would flag as disturbed; brighter red = larger magnitude. In production this runs on real Landsat composites across an entire impact zone the morning after landfall.
Demo · Weighted Community Vulnerability Index
The paper's method — combine several indicators, weight them, rank places — is the engine behind any community-vulnerability map. Here it runs on five Red-Cross-style indicators across 30 synthetic communities. Move the weights; the choropleth and the priority ranking recompute live.
Equal weights give a baseline composite; emphasize No vehicle and the ranking shifts toward access-isolated communities — the ones that need pre-positioned transport. Swap in real ACS + RC geography and this is a deployable pre-positioning tool. Indicators here are illustrative.
Worth Tracking, Not Yet Operational
Geospatial foundation models
Built to adapt across sensors and spectral bands. The architecture pattern for your own AI-GIS pipelines — one model, many inputs.
LLM-mined heat perception
An LLM extracts heat discomfort from social posts (94.5% true-positive on extreme cases). The technique — LLM + social media for live situational awareness — transfers straight to extreme-heat response.
UK national climate-damage framework
Disaggregated damage/risk to 406 local areas, channel-specific, with catastrophic and "missing" risks. A model for Community Adaptation–style analyses.
Methodological
AADT on local roads (routing/logistics), multi-year droughts (hazard intelligence), and spatial interpretability (sanity-checking your own spatial models and dashboards).
Full 31-Paper Digest
The remaining ~18 papers — crop yield, Mexico City businesses, plastic-litter SAR, Himalaya carbon, water allocation, green-space equity — are interesting but off-target for disaster operations. The complete annotated digest, with every summary, source, and access status, is in the PDF.