AtlasCast
The storm,
address by address.
Ten storm types. Ten days out. Four times a day.
I · The idea
Which addresses could a storm reach next?
AtlasCast, part of Atlas United, is a forecast model suite that works address by address. It looks days ahead, and it names the places a storm could touch, not just the region it is heading for.
II · Ten storm types
Ten storms. One forecast.
Tornado
Rotating storms
Hail
Ice stones from the sky
Wind
Damaging gusts
Flash flood
Water that rises fast
Heavy rain
Intense downpours
Winter
Snow and blizzards
Ice
Freezing rain
Coastal surge
Sea water inland
Hurricane
Tropical cyclones
Wildfire
Fire danger
III · Ten days out
Six windows, 24 hours to 10 days.
The forecast covers six windows, from the next 24 hours out to ten days. Counties also carry an A to F grade as a quick map label; the address list is the product.
IV · Address by address
A list, not a map shading.
Inside each forecast footprint, AtlasCast lists the individual addresses: where they are, which storm type could reach them, and in which window. You can read it on the live map, or take it as data.
V · Begin
See it live,
or take the data.
Open the live map to watch the forecast. When you want the addresses themselves, get a free API key for ML Storm Forecast by Address. Five cents per address, no contracts, pay as you go.
Open the live map Get the dataFurther detail · AtlasCast, part of Atlas United
Storm forecasts, by address.
Atlas United risk models for every U.S. address: address-level storm forecasts for the next 24 hours to 10 days, plus every reported storm near that address back to 2003. Pay per address, no contract. AtlasCast is the forecast model suite inside Atlas United: a separate machine-learning model for each of 10 perils, producing address-level forecasts 24 hours to 10 days ahead. Built so decision-makers see the threat with time to act.
10 perils
·
address-level
·
24 hours to 10 days
·
refreshed 4×/day
Further detail
10
Perils, each with its own model
10
days
Longest forecast window
4
×
Full re-forecasts per day
Further detail · In the API
Five products, one endpoint.
Forecast products first, then the record of what already happened. Every product has a free preview that shows counts and the exact price.
ML Storm Forecast by Address
Probabilistic address-level forecasts for 10 perils from per-peril machine-learning models, 24 hours to 10 days ahead, refreshed 4 times a day. Each address record leads with the structure replacement value and estimated people at that address, plus a forecast time and a risk label. Ten perils: TORNADO, HAIL, WIND, FLASH_FLOOD, HEAVY_RAIN, WINTER, ICE, COASTAL_SURGE, HURRICANE, WILDFIRE.
You get: structure_value (replacement cost, not market value), people, full_address, city, state, zip, county, lat, lon, storm_type, forecast_time, risk_label, storm_hits.
Price: $0.05 per address. Free preview first.
Forecasts describe what could happen, not what will happen. Not a prediction of damage to any structure. The free preview shows counts, per-peril totals and the exact cost; the county grade is a teaser label, not a property score.
Model-Based Weather by Address
Point weather for any U.S. coordinate: current conditions, active severe weather alerts and a short-term rain outlook. The lean response is free with no account. Add details=true for the full pull of over 170 variables, including hail size, rotation, lightning, reflectivity and cloud measures.
You get: location, valid_time, alerts, temperature_f, dewpoint_f, humidity_pct, wind_speed_mph, wind_gust_mph, wind_direction_deg, precip_in, precip_type, pressure_mb, visibility_mi. With details=true, the full variable set.
Price: Lean response free. $0.0002 per call with details=true. Batch: $0.0002 per location, up to 500 per call.
Conditions, not a hazard or damage assessment. Forecast snapshots carry fewer fields than current conditions.
Model-Based Hourly Forecast by Address
Forecast hours from 0 to 48 ahead for any U.S. coordinate. The response snaps to the nearest available hour and reports forecast_hour_used. With details=true it returns about 40 core fields plus tornado_rotation_score.
You get: forecast_hour_used, forecast_note, valid_time, alerts, temperature_f, dewpoint_f, humidity_pct, wind_speed_mph, wind_gust_mph, precip_in, visibility_mi, pressure_mb, tornado_rotation_score.
Price: $0.0002 per call.
Forecast hours carry fewer fields than current conditions. Forecasts describe what could happen, not what will happen.
Storm History by Address: Reported and Radar-Detected Events Since 2003
One property, all 11 reported-event categories (hail, wind, flood, lightning, tornado, hurricane, winter storm, extreme heat and cold, drought, wildfire, coastal) from 2003 to today, as one citable report with a reference ID. Optional date_of_loss=YYYY-MM-DD answers whether an event was reported or radar-detected near the address on that date, and lists events before and after it. A summary cut, hazard_score, is $0.50 per property.
You get: One line per event: date, type, magnitude with units where measured, distance in miles and county. Significant thresholds: hail 0.75 in, wind 50 mph, winter storm 6 in.
Price: $0.05 per event line, $10 minimum per report. The free preview shows the exact line count and price.
A record of reported and radar-detected events, not a damage assessment. It does not confirm a claim or a loss.
Past Storm Leads by Address: Reported and Radar-Detected, Last 365 Days
Storms that already happened, by address. Address-level records of events reported or radar-detected at or near each address over the last 365 days, filterable by county, peril and measured severity.
You get: full_address, city, state, zip, county, lat, lon, storm_type, event_time, severity, severity_unit, distance_mi, storm_hits.
Price: $0.05 per address. The free preview shows counts, severity ranges, addresses with repeat storms and the exact cost.
A record of reported and radar-detected events, not a damage assessment, inspection or repair estimate. Severity describes the storm, not damage to a structure.
Further detail · What it’s used for
Lead time is the product.
AtlasCast turns forecasts into an address-level warning the teams responsible for people, property and infrastructure can act on, days before impact.
Emergency management
Pre-position crews and shelters, flag schools, hospitals and other facilities by exact address, and buy lead time before impact.
Insurance & reinsurance
See which addresses on the book face incoming severe weather, warn policyholders to mitigate, and pre-stage catastrophe response, reserves and claims.
Property managers & restoration
Know which properties could be reached next and stage crews and materials ahead of the storm.
Solar & energy
Pre-stage crews ahead of wind and ice and prepare solar assets for hail.
Data and AI-agent builders
One endpoint, pay per address, with x402 and an MCP server for agents.
Further detail · Ten perils
Ten perils, every season.
Each address is forecast for the perils that could reach it, from spring storms to winter ice.
Tornado
Rotating storms
Hail
Stone size
Wind
Damaging gusts
Flash flood
Water that rises fast
Heavy rain
Intense downpours
Winter
Snow and blizzards
Ice
Freezing rain
Wildfire
Fire weather
Hurricane
Tropical storms
Coastal surge
Storm surge
Further detail · How it works
Models in. A per-address forecast out.
AtlasCast uses a separate machine-learning model for each of 10 perils, producing a calibrated probability and severity estimate that drive each address’s risk label. Models are trained and validated chronologically on held-out reported events: they are scored only on periods they never saw in training. Inputs are restricted to information available at forecast issue time, so no future information leaks into a forecast. Forecasts refresh 4 times a day, are timestamped, versioned and reproducible, and every paid response carries a reference_id receipt. Forecasts describe what could happen, not what will happen.
Further detail · Why it’s different
Not a radar app. A forecast, per address.
01
Per address
A forecast for the individual address, with the structure replacement value and estimated people attached. Not a county shading, not a ZIP, not a radar tile.
02
Forward-looking
What could happen 24 hours to 10 days out, not a static grade of what a property is.
03
Lead time
The threat is flagged ahead of time, so crews, carriers and agencies can act before it lands.
04
Replacement cost, not market value
The value on each address is what it would cost to rebuild the structure.
Further detail · Get started
See the storm before it arrives.
Launch the live national map, or get an API key for your addresses. A forecast shows what could happen, not what will happen.
Private briefing