The data behindevery score

Every number in a Refugia report can be traced to a named source. Where an official agency or a research institute publishes a global dataset, Refugia reads that dataset directly at your coordinates. It does not estimate from a summary of a region, and it does not use a figure it cannot cite.

This note lists the sources by subject: what each one measures, how fine its grid is and what it is used for. The live list, with each source’s weight and licence, is on the methodology page; how the sources are blended is in How a report is made.

Elevation comes from the European Space Agency’s Copernicus GLO-30, a global radar survey at 30 metres. Prominence is computed from it: a point that stands well above everything within fifty kilometres scores differently in a flood or a crust-shift scenario from one at the bottom of a basin, even at the same altitude.

WorldCover tells the report what surrounds the pin, which feeds criteria from food production to wildfire and heating with wood. SoilGrids is a machine-learning map of soil built from about 240,000 soil profiles; it decides whether the ground can grow food.

GloFAS is the European Commission’s global river flood model. In the United States, the FEMA flood map is also read, and because it is official, recent and drawn for the country, it carries more weight in the blend. Storm surge on the coast is estimated from elevation and distance to the sea. Depth to groundwater has no global open map at useful resolution, so a research agent looks it up from regional surveys and cites them.

Climate figures are ten years of observed weather from the ERA5 reanalysis of the European Centre for Medium-Range Weather Forecasts, not a single year and not a projection. NASA’s POWER service supplies the solar and wind averages that decide whether a place can make its own energy off the grid.

Distance to a hospital or a grocery is measured from the point along real map data, not from the centre of the town. Distance to large cities works the other way in the grid-down layer: closeness that is convenient today becomes exposure when supply chains stop.

Landslide and wildfire indices are calculated by Refugia from the models above, and are labelled as derived so their lower authority is visible. Where a country publishes its own hazard maps, they are the next sources to be added.

The nuclear exchange scenario combines these distances with the prevailing wind from ERA5, to estimate both blast exposure and where fallout would drift.

Nuclear plants come from Wikidata rather than the 2021 power plant database, because Wikidata records when a plant closed. Closed plants still appear on the map, in grey, but no longer count in a report’s scores.

Each site gets a danger class from its industry: refining, basic chemicals, fertiliser and pesticide plants, smelters and hazardous waste are high; other chemicals, plastics, pulp, cement, mining and fossil power are medium. Its weight then grows with the size of its toxic releases. Greenhouse gases and nutrients are left out, and bulk combustion pollutants count at 5%.

A report adds up the sites within 10 km for everyday pollution, and the high and medium sites within 30 km for the chemical hazards of a breakdown, when plants are left unattended. Nearer sites count more, and a site upwind on the prevailing wind counts up to twice as much as one downwind.

For places on a watchlist, Refugia checks four public feeds every ten minutes and alerts you to events around them:

• USGS real-time earthquakes.
• GDACS, the Global Disaster Alert and Coordination System of the UN and the European Commission, for floods, cyclones and volcanoes.
• NASA EONET for wildfires, storms and other natural events seen from orbit.
• NOAA weather alerts in the United States.

Every source’s licence is stored with it, and the methodology page shows it. Most are public domain or Creative Commons Attribution. Two, ERA5 climate and terrain prominence, are currently read through the free tier of the Open-Meteo service, which does not allow commercial use, and are marked so until they are moved to a commercial feed.

Open data has limits, and a report is clearer for stating them:

• A grid cell is an average. A 9 km climate cell cannot see a frost hollow at the bottom of one valley.
• Data has an age. Population is from 2020 and land cover from 2021; the report shows each source's year.
• Some things no map holds, such as property prices, crime or how reliable the local grid is. Research agents look these up and cite their sources, at the lowest weight in the blend.
• A score compares places. It is not a prediction and not professional advice.