Overview
The Nearmap Damage Classifications AI provides AI-derived damage classifications for Building Lifecycle features via the AI Feature API. Based on Gen 6 AI and a specialized model built to specifically detect the severity of damage, these damage classifications enable remote and efficient triaging of properties impacted by a catastrophic event, as well as informing effective response planning. Additionally, carriers can compare the condition of a building across different survey dates to better evaluate its insurable value.
We recommend that you cross-check the results from this with MapBrowser imagery or imagery from another source. For more information, see Statistical Performance.
Version history
Version | Last Updated | Description of Changes |
|---|
B.2 | 5 November, 2025 | Addressed floating point precision issues. Training data augmented with float32 precision variants to ensure robust predictions across deployment environments. |
B.1 | 26 June, 2025 | Resolved underprediction for structures impacted by fallen trees through a targeted relabeling followed by model retraining. Following this, Class 3 performance improved significantly. |
B.0 (rc2) | 14 May, 2025 | Production release (B.0rc2 promoted to B.0).
Implemented SMOTE class balancing improved performance across all damage classes. |
B.0 (rc1) | 14 May, 2025 | Expanded from 100k to ~194k labels. Trained on ~194k buildings across 74 post cat events. Refined the 5 class damage scale to align with the
FEMA framework. |
B.0 (rc0) | 21 August, 2024 | Initial Gen 6 Nearmap AI damage model. LightGBM model with 18 bin histograms across 10 damage related classes. Added 3 new classes:
Building Under Construction (80), Exposed Roof Deck (203), Exposed Underlayment (283). Trained on the same 100k labeled buildings used in
the prior model version. |
Requirements
To access the damage detection data in Damage Classifications AI, you must have access to the following:
- AI Feature API
- Damage Classifications product
- An AI pack that provides access to the Building Lifecycle layers (ID 91987430-6739-
5e16-b92f-b830dd7d52a6)
Damage classifications are only produced for Building Lifecycle features.
Real-time damage classifications are not available on bulk requests. If the real-time damage supplementary data source is selected as a part of a bulk request, the system will return an error.
Availability
Damage Classifications AI is available on survey resources with the following Gen 6 AI system versions or newer versions:
gen6-glowing_lantern-1.0gen6-glowing_grove-1.0
Before you start
It is recommended that you review this section before you start using Damage Classifications AI.
AI Packs
Post-cat surveys are special surveys run in the context of a catastrophe. Survey resources belonging to post-cat surveys cannot be accessed using packs that are marked as non-post-cat and vice versa. In the context of damage classifications:
packs=damage is a pack that includes the Building Lifecycle class and can be used to access postcat survey resources with damagepacks=damage_non_postcat is a pack that includes the Building Lifecycle class and can be used to access non-post-cat survey resources with damage
Damage Classifications Data
The following Nearmap AI data is available in the Damage Classifications AI product:
Damage classification is based on evaluating the primary structure of the given parcel. Three tiers of classification are available:
- 5-tier (or
raw), based on the FEMA 5-tier classification system composed of Undamaged,
Affected, Minor, Major and Destroyed
- 3-tier, composed of
UndamagedOrAffected, Minor and MajorOrDestroyed
- 2-tier, composed of
UndamagedOrAffectedOrMinor and MajorOrDestroyed
The extent of damage classified using a 5-tier classification scale and their descriptions are based on the Federal Emergency Management Agency's (FEMA) damage assessment guidelines.
Triage Class | NM Score | Category | Description |
|---|
No Damage | 0 | No Damage | The building does not appear to be impacted and is livable. |
Assessment Needed | 1 | Affected | - Minimal missing tiles or shingles - May be minimal structural damage to garage/carport, deck |
Assessment Needed | 2 | Minor | - No structural damage - Some missing roof covering material (tile, shingle) - Some minor damage to associated roof objects including chimney, air conditioner, solar panel, solar hot water, skylight. |
Assessment Needed/Unlivable | 3 | Major | - Building has structural damage or loss, however overall structure still standing and can be used for shelter - Parts of interior may be visible |
Unlivable | 4 | Destroyed | - Structure can no longer be used for shelter - Majority of structure compromised (that is, walls, roof); interior exposed - Building frame bent, twisted, or broken - Only foundation remains |
As a default, Nearmap displays 5-tier classification. However, you can use the API if you choose to view a 3-tier or 2-tier classification.
Data source
In order for damage classifications to be produced and attached, damage must be included as a supplementary data source using the include query parameter. You can do this by either manually specifying damage as a parameter to include or specifying all as a parameter to include, which will then include all data sources available to the user.
Examples
include=damageinclude=damage,roofSpotlightIndex
Additional data sources such as damage and Roof Spotlight Index (RSI) are only available on gen6 system versions. The systemVersionPrefix or systemVersion query parameters must be set.
Learn how to access Damage Classifications AI.
Damage model
By default, the latest version of the damage model will be used. To specify a particular
version of the damage model, use the damageModelVersion query parameter.
Ratios
The ratios parameter provides a measure that indicates how much of the overall footprint meets at least 50% confidence level for a given detection. Each footprint has a list of ratios which indicate the ratio per detection class.
This is calculated by dividing the area where confidence exceeds 50% by the total footprint area and then rounding the result to three decimal places. For example, if a 100 sqm footprint contains 50 sqm with confidence above 50%, the ratio would be 0.5.
Examples
Browse code examples for date-based and resource-based access.
Date-based request
This example is a date-based request for an address. For details about the query parameters and the response, see Example Date-based Request and Response.
GET https://api.nearmap.com/ai/features/v4/features.json?packs=damage&include=damage&systemVersionPrefix=gen6-&since=2025-01-12&until=2025-01-13&streetAddress=15102 ALBRIGHT ST&city=PACIFICPALISADES&state=CA&zip=90272&country=US&parcelMode=true&rapid=true&excludeTilesWithOcclusion=true
Survey resource-based request
This request is based on a known surveyResourceId , so the API will not need parameters to find a survey resource. For details about the query parameters and the response, see Example Survey Resource-based Request and Response.
Two variations are possible when using an AOI geometry. Either as the polygon query parameter as a sequence of coordinates:
GET https://api.nearmap.com/ai/features/v4/surveyresources/09aee77b-3f6e-5077-b9ec-036da962642e/features.json?include=damage&classes=91987430-6739-5e16-b92f-b830dd7d52a6&systemVersionPrefix=gen6&polygon=-118.523608,34.048762,-118.523579,34.048634,-118.524001,34.048569,-118.524029,34.048697,-118.523608,34.048762
Or in the body as GeoJSON under the property aoi:
POST https://api.nearmap.com/ai/features/v4/surveyresources/09aee77b-3f6e-5077-b9ec-036da962642e/features.json?include=damage&classes=91987430-6739-5e16-b92f-b830dd7d52a6&systemVersionPrefix=gen6
Data model report
Attached below is the Damage Classifications AI data model report. Click the link to download the report.