The AI Feature API allows you to assess the impact of hurricanes via the Hurricane Vulnerability score. The sections below explain how this score is calculated and how it should be interpreted.
Hurricane Vulnerability score
The Hurricane Vulnerability Score is a 1-5 score that differentiates between buildings based on their relative susceptibility to sustaining wind damage if exposed to hurricane-force winds. A score of 1 represents the most vulnerable to hurricane damage, and a score of 5 represents the most resilient. The score enables insurance companies to:
- Accurately assess property-level vulnerability to hurricane wind damage
- Differentiate risk between properties in similar geographic areas
- Make informed underwriting and pricing decisions
- Proactively engage with policy holders for risk mitigation
The Hurricane Vulnerability Score is calculated in real time using a machine learning model trained on actual hurricane damage outcomes. For this reason, it is omitted from the API response by default, and comes with a latency cost when requested on the API.
Quick info
* Data is available in Australia and New Zealand where it assesses vulnerability to tropical cyclones (the regional term for hurricanes). However, the model was trained on US hurricane data.
Understanding the Hurricane Vulnerability Score
The Hurricane Vulnerability score is calculated using a combination of the model input features and the surrounding environment. Details of the calculation of this score are given below.
Model input features
The Hurricane Vulnerability Score is calculated using a machine learning model trained on actual hurricane damage outcomes. The following features are used as inputs to the model.
Roof Characteristics
- Roof Material (
shinglePresent): Presence of asphalt shingle roofing - Roof Shape (
flatPresent, hipPresent): Whether the roof is flat or hip-shaped - Roof Condition Indicators:
roofStainingRatio: Ratio of roof area affected by stainingmissingRoofTileOrShingleRatio: Ratio of roof area with missing tiles or shingleswornShinglesRatio: Ratio of roof area with worn shinglesroofPatchingRatio: Ratio of roof area with visible patchingroofRustingRatio: Ratio of roof area with visible rusting
Surrounding environment
treeOverhangRatio: Ratio of roof area covered by tree overhangdefensibleSpaceWeightedCoverage: Weighted coverage metric from Defensible Space zones (0-5 ft, 5-30 ft, 30-100 ft)
All ratio features range from 0 to 1, where 0 indicates the condition is not present and 1 indicates it covers the entire roof area. Binary presence features are either 0 (not present) or 1 (present). In the response, the modelInputFeatures object provides transparency into the specific factors that influenced the vulnerability score for each building.
Hurricane Vulnerability vs Hurricane Hazard
It is important to distinguish between vulnerability and hazard when assessing hurricane risk.
Hurricane Vulnerability
What this score measures:
- Property-specific susceptibility to wind damage
- Based on building characteristics and surroundings
- Independent of location
- Two identical buildings in Louisiana and California can have the same vulnerability
Hurricane Hazard
Not included in this score:
- Probability and intensity of hurricanes in a given location
- Based on geographic factors and historical hurricane activity
- Varies significantly by region
- The building in Louisiana faces higher hazard than the one in California
Overall Hurricane Risk = Hazard × Vulnerability
The Hurricane Vulnerability score provides the vulnerability component.
Geographic Applicability
The Hurricane Vulnerability Score is theoretically applicable to any building exposed to risk of wind damage. The model was trained on data from Lake Charles, Louisiana (Hurricane Laura, 2020) and validated on 141,889 buildings across 7 U.S. hurricanes.
Validated Events
- Hurricane Beryl (2024, Category 1)
- Hurricane Dorian (2019, Category 2)
- Hurricane Henri (2021, Tropical Storm)
- Hurricane Ian (2022, Category 4)
- Hurricane Ida (2021, Category 4)
- Hurricane Idalia (2023, Category 3)
- Hurricane Laura (2020, Category 4)
If roof types in a geography are substantially different from those the model was trained on (e.g., different roof material patterns), reduced performance may occur. Validation on tropical cyclones from geographies outside the United States is forthcoming pending sufficient data availability.
Peril Applicability
The Hurricane Vulnerability Score correlates specifically with damage caused by hurricane-force winds. It does not apply to:
- Flood-induced damage
- Storm surge damage
- Rain-related water intrusion
The score is not expected or designed to perform well on flood damage or flood-related insurance claims.
Parcel Mode behavior
For information about the AI Feature API’s behavior when parcelMode=true is included in the request, see the Parcel Mode documentation.
Model performance and regulatory approvals
The Hurricane Vulnerability Score and regulatory approvals can be found within the Perils Hurricane Release Report.