Methodology
Every hazard on a Dwellyng report is read at the address point from its own gridded dataset, then converted to a 1–10 score against fixed, nationally consistent breakpoints — so a 7 in Florida means the same thing as a 7 in Oregon. This page documents where each number comes from and what it does not cover.
How the 1–10 scores are assigned
Each hazard layer has its own breakpoints, fixed nationally rather than derived from the neighbourhood around the address. A score is therefore comparable between properties and between states: it says how this location ranks against the whole country for that hazard, not how it ranks against its own street.
Scores roll up into categories by their worst member, so a single severe hazard is never averaged away by mild ones sharing its category. The overall rating is anchored on the hazards that cause direct structural damage — flood, wind, wildfire, earthquake, tornado. Comfort and energy-demand metrics contribute to their own categories but can never drive the overall rating on their own, because a house that is merely expensive to cool is not a house at risk.
On the report, categories are ordered by severity at that specific address, and those scoring Low or Very Low are collapsed into a drawer at the bottom. Nothing is removed — every hazard remains on the page, and unscored hazards stay visible rather than being filed under "lower-rated", because a gap in the data is not the same as a low risk.
The underlying data and scoring are shared verbatim with scores.degreeday.org, and enforced by tests rather than convention, so the two sites cannot silently disagree about the same address.
Site-context flags
Three checks shown beside the address are memberships, not scores: a property is either inside them or it is not, so they appear as labels rather than as a 1–10 value.
- Wildland–Urban Interface — read from a 100 m WUI mask. It marks where development meets or intermingles with wildland vegetation, which raises a structure's exposure beyond what the wildfire score alone conveys.
- Leveed area — from the USACE National Levee Database, named where the database names it. This one changes how to read the rest of the report: the flood scores here are undefended and take no credit for the levee. Levees can also be overtopped or fail, so residual risk remains either way.
- Hurricane zone — within roughly 100 km (62 mi) of a historical tropical-cyclone track of Category 1 or stronger, from IBTrACS. It marks exposure to the storm type; the wind speed itself is scored separately under Extreme Wind.
Each is shown only where it applies, and none has a "no" state. The absence of a flag is not a clean bill of health — it means the check did not fire, which can also happen when a dataset cannot be read.
Methodology by hazard
Every hazard scored on a report, grouped as the report groups them. Each entry describes the dataset and the modelling behind that layer.
Flooding
Water reaching the property from rivers, the coast, or heavy rain.
River & Coastal Flood Flood depth in a 1-in-100-year flood
Riverine flood depths are based on the JRC global river flood model, which combines ERA5-forced GloFAS hydrology with the two-dimensional LISFLOOD-FP hydraulic model, run on the MERIT-Hydro digital elevation model, to estimate inundation depth for the 100-year return period at approximately 90 m resolution. Coastal flooding is screened using a connectivity-based inundation model driven by the 100-year COAST-RP extreme water level, which combines ERA5-based extratropical storm surge, synthetic tropical-cyclone surge, and tides over the approximately 30 m DeltaDTM coastal terrain model. Mean dynamic topography is used to account for differences between the vertical reference frame of the coastal extreme water levels and the coastal terrain elevations, so that water levels and land elevations can be compared consistently. Areas are identified as potentially inundated where low-lying terrain is hydraulically connected to the ocean and the adjusted extreme water level exceeds the local ground elevation. The coastal screening includes a friction-based attenuation component to account for the reduction of inland water levels with distance and land-surface resistance, so the method is not a simple bathtub fill. This approach represents a static coastal inundation screen rather than a full dynamic hydrodynamic simulation. Wave setup, estimated as a fraction of the significant wave height, is added on wave-exposed open coasts identified by a fetch-based exposure classification and omitted in sheltered bays, estuaries, and fjords. Wave runup, overtopping, erosion, and local drainage interactions are not included. Both flood layers represent undefended hazard conditions; levees, seawalls, surge barriers, and other local flood-protection infrastructure are not explicitly modeled.
For a buyer: A depth above roughly 0.3 m (1 ft) at the building footprint usually means structural flood damage, not just a wet yard.
Low-Lying Coastal Land Ground height above the local high-tide line
The Coastal Exposure Index identifies low-lying coastal land that falls below selected extreme high-water levels and is hydraulically connected to the ocean. It is intended as a screen for chronic tidal exposure and sea-level-rise susceptibility, not as a storm-surge, wave, or event-based flood simulation. Water levels are referenced to the mean higher-high-water tidal datum, using NOAA VDatum in the United States and FES2022 tides elsewhere. Outside the United States, mean dynamic topography is used to align the modeled tidal water levels with the land-elevation reference frame before comparing water levels with terrain. Elevation data come from lidar-based DEMs in the United States and the approximately 30 m global DeltaDTM coastal terrain model outside the United States. A connectivity check removes low-lying terrain that is isolated from the coast, helping distinguish potentially exposed coastal areas from inland depressions or disconnected basins. The index does not model groundwater emergence, drainage interactions, land subsidence, shoreline change, or flood defenses.
For a buyer: Negative or near-zero values indicate land already at or below high tide — expect nuisance flooding and rising insurance costs.
FEMA Flood Zone Official FEMA designation
FEMA flood zones are based on FEMA’s National Flood Hazard Layer (NFHL), the official regulatory dataset used for U.S. floodplain management and flood insurance. The card reports the mapped flood designation at the selected location, such as Zone AE, including Special Flood Hazard Areas associated with the 1% annual-chance flood and areas of 0.2% annual-chance flood hazard. Coverage is limited to the United States and depends on digitized NFHL availability. FEMA maps change over time, and some areas may be unmapped, incomplete, or updated after the data snapshot used here.
For a buyer: A lender will require flood insurance in any zone beginning with A or V. Zone X is not a guarantee of safety: roughly a quarter of NFIP claims come from outside the high-risk zones.
Wildfire
Chance of a wildfire burning at this location.
Wildfire Chance of burning in any given year
This model produces global annual burn-probability estimates at 10 arc-second (~300 m) resolution from a tree-based gradient-boosting regressor trained on observed burned area worldwide (2003–2023, filtered to wildfire). Predictors span fire weather (Canadian FWI System indicators), climate, fuels and land cover (LULC, aboveground biomass), terrain (elevation, slope, TPI), and human/ignition factors (population, roads, lightning); relationships are learned globally and past fire location is excluded as a predictor. Over the United States the delivered layer substitutes USFS FSim, and a post-processing step then “oozes” wildland burn probability into adjacent wildland–urban interface pixels to represent structure-fire exposure.
For a buyer: Insurers increasingly price — or decline — on this. Worth pairing with a defensible-space and roof-material check.
Severe Weather
Wind, tornado, hail and lightning exposure.
Extreme Wind Peak gust with a 1-in-100-year strength
Extreme wind hazard is represented by terrain-aware 3-second peak-gust return levels at 30 arc-second (~1 km) resolution. The global wind model combines separate tropical-cyclone and non-tropical-cyclone wind estimates, then applies terrain and surface-roughness adjustments to better represent local exposure.
The non-tropical-cyclone component is derived from ERA5 pressure-level winds, with tropical-cyclone days removed using historical track data. Wind profiles are vertically interpolated to represent near-surface conditions, annual maxima are extracted, and extreme-value statistics are used to estimate return levels. These estimates are refined to 30 arc-second (~1 km) resolution using elevation and surface-roughness exposure adjustments.
The tropical-cyclone component is based on CHAZ, the Columbia HAZard model, a statistical–dynamical tropical-cyclone model that generates large synthetic tropical-cyclone catalogs from large-scale environmental conditions. CHAZ represents tropical-cyclone activity through separate genesis, track, and intensity components, allowing coastal wind return levels to be estimated from many physically plausible storms rather than the limited observed record alone. In this workflow, CHAZ tropical-cyclone wind return levels are converted from 1-minute sustained winds to 3-second gusts using the WMO offshore conversion factor and then combined with the non-tropical-cyclone wind branch after surface-roughness and terrain-speedup adjustments.
For a buyer: Above about 110 mph, roof tie-downs, garage doors and window protection dominate whether a house survives intact.
Tornado Chance of a strong tornado within 25 miles per year
Tornado hazard is represented by a U.S. climatology of significant tornado occurrence based on NOAA Storm Prediction Center records. The metric is the mean number of days per decade with an EF2 or stronger tornado reported within 25 miles (40 km) of the location over 1986–2015. The 25-mile search radius accounts for spatial uncertainty in tornado tracks and provides a smoother local climatology, while the EF2+ threshold focuses on events most likely to cause substantial damage. This layer represents historical tornado occurrence, not a forecast or event-based tornado simulation. Coverage is limited to the United States.
For a buyer: The 25-mile radius reflects how narrow tornado tracks are — it measures the neighbourhood’s climate, not the odds for this specific roof.
Derecho (Straight-Line Wind) Derecho events per year
Derecho hazard is represented by a U.S. climatology of long-lived, widespread straight-line windstorms produced by organized thunderstorm complexes. The layer is based on a 4 km observational climatology for 2004–2021 that uses machine-learning bow-echo detection and storm tracking to identify derecho events. Event wind-damage footprints are summed, annualized, and smoothed to produce a derecho-frequency surface expressed as events per year. This layer captures organized convective wind hazard that is not represented in the Extreme Wind layer. Coverage is limited to the United States east of the Rocky Mountains.
For a buyer: The classic cause of "the whole street lost its trees and power for a week".
Large Hail Chance of 2-inch-plus hail per year
Large-hail hazard is represented by a global climatology of very large hail occurrence, defined as hailstones at least 5 cm (2 in) in diameter. The layer is based on the AR-CHaMo statistical hail model applied to ERA5 reanalysis for 1950–2023. AR-CHaMo estimates hail occurrence as the product of thunderstorm probability and the conditional probability of very large hail given a storm, using convective-environment predictors such as instability, vertical wind shear, and moisture. The model is calibrated against hail reports from Europe, the United States, and Australia. Because ERA5 does not explicitly simulate hailstones, occurrence is inferred from atmospheric environments favorable for very large hail rather than observed or modeled hail impacts at a specific site.
For a buyer: Hail is the single largest driver of roof-replacement claims in the central US, and a common reason for high deductibles.
Lightning Hours of thunder per year
Lightning activity is represented by annual thunder hours, defined as hours with at least two lightning strokes detected within 15 km. The metric is derived from a satellite-calibrated global lightning dataset beginning in 2013 and aggregated to annual totals on an approximately 5 km grid. Thunder hours provide a proxy for thunderstorm activity and electrical-storm exposure, but they do not directly measure storm severity, hail, tornadoes, damaging winds, or rainfall intensity. Detection efficiency and reporting quality can vary by region, so the layer should be interpreted as a broad climatological screen rather than a site-specific lightning risk estimate.
For a buyer: Mostly a prompt for surge protection on well pumps, HVAC and home electronics.
Heat
Extreme summer temperature and humidity.
Extreme Heat Hottest day in a typical decade
Extreme heat is represented by the annual maximum dry-bulb air temperature at a specified annual-chance level. Quality-controlled NOAA GHCN station records are fit with a Generalized Extreme Value (GEV) distribution in a nonstationary Bayesian framework, allowing distribution parameters to vary with global mean temperature. The resulting station-level estimates are regionalized to a 30 arc-second (~1 km) global grid using machine-learning methods trained on high-resolution climatological predictors derived from reanalysis, terrain-aware downscaling, and global land-surface climatologies, together with terrain and climate-zone variables. This layer represents extreme outdoor air-temperature hazard, not indoor heat exposure or heat stress, which also depend on humidity, radiation, wind, building characteristics, acclimatization, and exposure duration.
For a buyer: Drives cooling-system sizing and, in older housing stock, whether the property is habitable during a heat wave.
Heat + Humidity Stress Peak wet-bulb temperature
Human heat stress is represented by the annual maximum wet-bulb temperature at the 10% annual-chance level. Wet-bulb temperature combines air temperature and humidity into a single thermodynamic measure related to the body’s ability to cool through evaporation. It is computed from quality-controlled station air-temperature and humidity records. Extremes are fit with a Generalized Extreme Value (GEV) distribution in a non-stationary Bayesian framework, allowing distribution parameters to vary with global mean temperature, and are then regionalized to a 30 arc-second (~1 km) global grid using the same machine-learning framework applied to extreme heat and cold. This layer represents large-scale outdoor heat-humidity hazard; it does not account for direct solar radiation, wind speed, clothing, activity level, indoor conditions, acclimatization, or exposure duration.
For a buyer: High wet-bulb values make air conditioning a safety system rather than a comfort one, and raise the stakes on power outages.
Cold
Extreme winter temperature.
Extreme Cold Coldest night in a typical decade
Cold stress is represented by the annual minimum dry-bulb air temperature at the 10% annual-chance level, with lower temperatures indicating greater hazard. Quality-controlled station minimum-temperature records are fit with a Generalized Extreme Value (GEV) distribution in a nonstationary Bayesian framework, allowing distribution parameters to vary with global mean temperature. The resulting station-level estimates are regionalized to a 30 arc-second (~1 km) global grid using the same machine-learning framework applied to the heat metrics, with gridded predictors representing local climate, elevation, terrain, geographic location, and climate-zone context. This layer represents outdoor cold-temperature hazard; actual impacts depend on asset design, exposure duration, wind, snow/ice conditions, building characteristics, and operational tolerances.
For a buyer: Sets the bar for pipe-freeze protection, insulation, and back-up heat.
Geophysical
Ground movement: earthquakes, landslides, sinking land.
Earthquake Peak ground shaking (1-in-475-year)
Seismic hazard is represented by peak ground acceleration (PGA) with a 10% probability of exceedance in 50 years, equivalent to an approximately 475-year return period or about a 0.2% annual chance. Values represent firm-rock conditions and are provided on an approximately 1 km global raster. The layer combines the best available regional probabilistic seismic-hazard models, including USGS, ESHM20, GEM, and other regional sources, with GSHAP used as background coverage where higher-resolution regional models are unavailable. Where models overlap, the maximum PGA value is used. The layer does not include local soil amplification, liquefaction, earthquake-triggered landslides, tsunami hazard, or other secondary effects. Boundaries or discontinuities may appear where independently developed regional models meet.
For a buyer: Above about 0.2 g, foundation bolting and cripple-wall bracing are the retrofits that matter most for older houses.
Landslide Susceptibility index
Landslide hazard is represented by a relative susceptibility index that reflects terrain conditions and rainfall-triggering potential. Outside the United States, the layer uses a global rainfall-driven landslide model that combines ERA5 daily rainfall with five susceptibility classes derived from factors such as slope, lithology, vegetation, and soil moisture to estimate landslide-initiation potential at approximately 90 m resolution. Within the United States, the layer uses USGS slope–relief susceptibility models trained on more than 600,000 documented landslides and 10 m elevation data. The outputs represent relative susceptibility and screening-level landslide potential, not event-specific probability, runout extent, depth, velocity, or site-scale slope stability.
For a buyer: Elevated values on a sloped lot are a reason to ask for a geotechnical report before closing.
Sinking Land Rate the ground is sinking
Land subsidence represents the estimated near-present-day rate of vertical ground-surface lowering. The layer is derived from a global deep-learning model trained on more than 46,000 quality-controlled observation points and 23 predictors related to groundwater abstraction and recharge, climate, geology, soil conditions, and topography. Groundwater pumping is an important driver in many high-subsidence regions, but local rates can also reflect sediment compaction, natural consolidation, hydrogeology, and land-use history. Estimates are provided at 30 arc-second (~1 km) resolution and should be interpreted as a broad screening layer, since subsidence can vary sharply over short distances due to local pumping, aquifer properties, construction history, and other factors not fully captured in global datasets. The source dataset does not provide estimates for urban areas, so subsidence may be unavailable or underrepresented in some locations where exposure is otherwise important.
For a buyer: Subsidence quietly worsens flood risk over a mortgage term and is a known cause of foundation cracking.
Drought
Long, severe dry spells affecting the water supply.
Drought Composite drought score
The composite drought score is a simple, globally comparable screening metric designed to identify basins where severe, multi-year drought could result in low remaining precipitation availability. ERA5 precipitation is aggregated over major river basins, and rolling 48-month basin-total precipitation accumulations are analyzed with extreme-value statistics to estimate the precipitation deficit associated with a 25-year drought. This deficit is subtracted from normal 48-month precipitation to estimate the remaining precipitation available to the basin under severe drought conditions. The resulting remaining-precipitation values are then ranked across basins worldwide, with higher drought scores assigned to basins with lower remaining precipitation. This structure prevents very wet regions from being classified as high drought risk solely because they can experience large absolute precipitation deficits.
The score is intentionally simplified and emphasizes absolute precipitation scarcity under severe drought conditions rather than departure from local norms. It does not represent full water availability, which depends on runoff generation, groundwater, snowpack, reservoirs, interbasin transfers, irrigation, water rights, infrastructure, demand, and water-management operations. These factors are not observed or modeled consistently at global scale, and available global water-stress datasets can depend heavily on uncertain assumptions about withdrawals, storage, routing, and management. Water availability is highly system-specific, so this layer should be interpreted as a first-pass drought-scarcity screen rather than a basin water-balance or water-security model.
For a buyer: A basin-scale screen. It does not know about this property’s well, reservoir, or water rights — check those separately.
Energy Demand
How much heating and cooling this location needs.
Cooling Demand Cooling degree days per year
Cooling demand is represented by annual Cooling Degree Days (CDD), the cumulative amount by which daily mean temperature exceeds the 18.3 °C (65 °F) balance point. CDD is computed from a 30 arc-second (~1 km) daily temperature climatology derived from reanalysis using terrain-aware statistical downscaling for 2004–2023. The metric approximates temperature-driven cooling energy demand, but does not account for humidity, solar gains, wind, building characteristics, occupancy, equipment loads, or cooling-system efficiency.
For a buyer: Compare against the seller’s actual utility bills; this measures the climate, not the building’s efficiency.
Heating Demand Heating degree days per year
Heating demand is represented by annual Heating Degree Days (HDD), the cumulative amount by which daily mean temperature falls below the 18.3 °C (65 °F) balance point. HDD is computed from a 30 arc-second (~1 km) daily temperature climatology derived from reanalysis using terrain-aware statistical downscaling for 2004–2023. The metric approximates temperature-driven heating energy demand, but does not account for wind, solar gains, building characteristics, occupancy, internal heat gains, heating-system efficiency, fuel type, or behavioral responses.
For a buyer: Compare against the seller’s actual utility bills; this measures the climate, not the building’s efficiency.
Property outlines and imagery
The lot line drawn on the aerial comes from a national parcel snapshot held by this site, with live county and state assessor services as a fallback where the snapshot cannot answer. Parcel data is a record of the legal lot, not a survey: boundaries carry the source assessor's accuracy, and a geocoded point can land on a driveway, a lot centroid, or the wrong side of a building. Owner names and mailing addresses are never shown.
Aerial imagery is Esri World Imagery. Street-level photography is Google Street View, and the report opens on it instead of the aerial when tree canopy covers the roof — measured from the aerial itself, so a house hidden under a mature canopy still gets a usable photo.
Data, time periods and uncertainty
These layers describe current average hazard conditions from the best recent data available. The exact period differs by hazard, because each relies on different sources, observation windows and methods — some use recent climate records, some use current observations, and some estimate rare events from many years of data.
Using recent data does not assume a static climate. It reflects that recent observations remain the strongest starting point for describing hazard conditions today. Climate change affects how often hazards occur, how severe they are and where — but not uniformly across hazards, places or periods, and that remains an active area of research.
Many values are modelled estimates rather than direct measurements, describing how often a rare event is expected or a long-term average. They are uncertain because observations are limited, extreme events are rare, models make assumptions, and national datasets cannot capture every local detail. That uncertainty grows the further ahead you look, so these scores should not be read as predictions of distant future conditions. Long-term decisions deserve analysis specific to the hazard and to the decision being made.