By the fastCRW team · Catalogue figures checked 2026-11-05 against GET /v1/data · Verify independently before relying on any figure.
A hazard agent needs three answers before it can say anything useful: what has just happened near a place, whether a storm is forming, and how exposed the area is to wildfire. Pergamon exposes official feeds for each. usgs covers earthquakes, nhc covers the National Hurricane Center, wildfirerisk covers USFS wildfire risk, and usastructures covers FEMA's USA Structures building footprints. One API key reaches all four.
Earthquakes near a point
usgs/query filters the USGS catalogue by time, magnitude and radius. The call below asks for events of magnitude 2.5 or more within 300 km of San Francisco since 2026-10-01.
curl "https://api.fastcrw.com/v1/data/usgs/query?starttime=2026-10-01&minmagnitude=2.5&latitude=37.77&longitude=-122.42&maxradiuskm=300&limit=3" \
-H "Authorization: Bearer $CRW_API_KEY"
Real output, three events from the response's features array:
mag 2.79, 8 km E of Lower Lake, CA
mag 2.51, 8 km E of Lower Lake, CA
mag 2.71, 7 km NW of The Geysers, CA
Each feature is GeoJSON, and properties holds mag, place and time. time is milliseconds since the Unix epoch, so convert it before you show it to a person. Without a point, the same call returns events worldwide: at check time a magnitude 5.1 event in Tanzania led the list.
Each feature also has a root-level id, the USGS event ID, such as nc75461677. Store it. An agent that polls the feed every few minutes can use it to skip events it has already handled, and to notice when USGS revises an event it has already seen.
Storms: the National Hurricane Center
nhc/active-storms lists the storms that are active now. nhc/forecast-cone returns the forecast cone, and returnGeometry=false drops the shape when you only need the metadata.
curl "https://api.fastcrw.com/v1/data/nhc/active-storms" \
-H "Authorization: Bearer $CRW_API_KEY"
On 2026-11-05 both calls returned empty results, and the storm list was {"activeStorms": []}. An empty answer is the normal result when no storm is active. It is not an error, so give your agent an explicit branch for it. The nhc provider has 32 capabilities in total, including forecast-track, wind-probabilities-34kt and past-track, and you call them the same way.
Wildfire burn probability at a point
wildfirerisk/fsim-burn-probability takes a geometry option. It must be JSON that includes a spatialReference. The example below uses WGS 84 (wkid 4326) and was run live with the Python SDK (crw 0.41.0).
# pip install crw
from crw import CrwClient
client = CrwClient() # reads CRW_API_KEY
fire = client.get_data("wildfirerisk", "fsim-burn-probability", {
"geometry": '{"x": -120.5, "y": 38.9, "spatialReference": {"wkid": 4326}}',
})
print(fire["data"]["value"])
Real output: 5000.
The spatialReference is not optional. The same point sent as {"x":-120.5,"y":38.9} is read as Web Mercator metres and returns "value": "NoData". When you get NoData, check the spatial reference first, and do not read it as a sign that the area is safe.
This post does not state a unit for the value. Show the raw number, and point readers to the dataset documentation, which Pergamon returns through the dataset-documentation capability. The wildfirerisk provider has 90 capabilities, and its terms are set per dataset, not once for the whole provider. Read the licence field of the dataset you use.
Structures near a point
usastructures/footprints-near-point returns building footprints within a distance of a point. distance is in metres. The call below looks within 100 m of a point in San Francisco and returned 7 features.
curl "https://api.fastcrw.com/v1/data/usastructures/footprints-near-point?longitude=-122.42&latitude=37.77&distance=100" \
-H "Authorization: Bearer $CRW_API_KEY"
Each feature is one building footprint, a building outline with its own OBJECTID. Its properties carry fields such as OCC_CLS, PRIM_OCC, PROP_CITY and HEIGHT. The provider also has count and bounding-box capabilities, such as count-in-bbox, which answer "how many" without downloading every footprint. Check their options with find before you build on them.
Footprint fields come from the source layer as published. Do not treat an occupancy code as a verified fact about a particular building. The layer's licence statement, returned in provenance, reads: "This work is licensed under a Creative Commons by Attribution (CC BY 4.0) license. See Credits and Map for Attribution." Show that attribution wherever you show the footprints.
What each feed does and does not answer
| Capability | Answers | Does not answer |
|---|---|---|
usgs/query | Which earthquakes were recorded near a point, with magnitude, place and time | Whether a building was damaged, or whether another event is coming |
nhc/active-storms | Which storms are active now | What a storm will do at a specific address |
wildfirerisk/fsim-burn-probability | The raw burn-probability value at a point in the USFS layer | The unit of that value, or whether an area must be evacuated |
usastructures/footprints-near-point | Which building footprints lie within a distance of a point | Replacement value, insurance value or verified occupancy |
Put the right column into the agent's instructions. A model that reads only the left column tends to answer more than the data supports.
A risk-at-this-address sketch
Pergamon returns the raw values. It does not return a risk score, and your agent should not invent one. A workable flow:
- Turn the address into a latitude and longitude with your own geocoder.
- Call
usgs/queryfor recent events within a radius you choose. - Call
nhc/active-stormsand branch on an empty list. - Call
wildfirerisk/fsim-burn-probabilitywith the point and itsspatialReference. - Call
usastructures/footprints-near-pointand count the buildings found. - Write the answer from those values only, with the
attribution_textof each source next to the figure it supports.
One run of this flow makes four server calls (the storm check, the earthquake query, the burn probability and the structures query), so it costs four credits. Add the calls you make for each extra address, and cache the results for the places you check often.
Limits
- These feeds are inputs to a decision, not a rating of a property or a region.
- Earthquake records can change after they are first published, because magnitudes and locations get revised. Query again before a decision that depends on them.
- Storm feeds are empty outside active storms. An empty result does not mean the risk is zero.
- This is not emergency advice. For warnings that affect life or property, use the official alert channels of the agency responsible.
Sources
- USGS copyrights and credits
- NWS disclaimer, which covers the NHC data
- Pergamon data sources reference
Related: weather.gov API in Python · Can you use public API data commercially? · Public data API for AI agents · Pergamon catalogue
