Coverage

Places and hazards on the map

The map covers 53 places in 2 regions. It has 16 hazards and 22 data sources. The system builds this page from the same registry that the data pipelines read. A place with no data therefore appears in the list, with its status.

The page uses the project configuration files. The database supplies two more facts: whether a place is on the map, and whether its grid exists. To rebuild it, run python scripts/generate_world_layers.py.

Regions

A region is the largest unit on the map. Each region has its own hazards, languages and boundary. India and the Middle East have different hazards. A dust storm in the Gulf and a monsoon flood in India need different data and different methods.

Model or index

Of the 16 hazards, 2 use machine learning: flood and extreme heat. The others use a calculated index, a rule, long-term climate data, or an official warning that we republish without change. The hazard with the best score is landslide susceptibility, at 0.85 in Himachal and 0.81 in Uttarakhand. It uses a calculated index, not machine learning. It ranks where, not when.

Places and their status

A place moves through three stages. First we declare it. Then we collect its data. Then it appears on the map. The list below shows all three stages. A place that we declared but did not collect data for is a gap in the coverage. A built grid means the 1 km map cells exist. Without them, the map cannot place an event at that location.

Zones

We group places into zones that have similar hazard conditions. A model trained in one zone can fail in another. We measured this. The urban-heat model scores 0.40 inside one city, but its score is negative when we apply it to a different city. A zone needs at least two places before we can test a model across places.

Data sources

The map uses 22 sources in 17 families. A family is a group of sources that share one supplier. Sources in the same family count as one source. Ten news sites that copy the same report therefore count once, not ten times. Open-Meteo is an example. It supplies air quality, dust, wet-bulb heat and wadi flash flood data. If all four report a hazard at one place, that is one source, not four.

We set the reliability weight for each source by hand. We have not yet calculated these weights from recorded results. See Hazards and Docs.