About Roof Age
Roof Age is a Nearmap product available for properties in the US. It predicts the roof installation date and therefore age of both residential and commercial roofs. The prediction uses Nearmap imagery, Gen 6 AI data and additional third-party data including climate data, building permits and assessor year-built data to arrive at an accurate roof age estimation.
The Roof Age Gen2 product uses multiple bespoke AI models to determine roof age by detecting a new roof installation in imagery or using images of a roof to estimate its age, as well as utilizing the additional information provided by building permits and assessor year-built data. This ensures comprehensive roof age estimates for every parcel.
Roof Age predictions are aligned with property ownership by using parcel boundaries. This enables roof age predictions on only the portion of the roof within a particular parcel rather than the entire building, for example, in the case of row homes.
To deliver the unprecedented scale of property intelligence offered by Roof Age, we analyzed 2.8 billion roof images spanning 2.5 million square miles of aerial imagery and covering a total of 151 million parcels nationwide. This massive dataset is backed by a rich temporal history, with over half of all parcels featuring 10 or more historical captures, allowing for a deep understanding of roof evolution over time.
Since Roof Age’s dataset is precomputed at a national scale, it enables sub–two-second lookups, empowering insurers to quickly evaluate roof-related risks and make faster, more informed underwriting decisions.
Roof Age calculation
The method of calculation of Roof Age is dependent on the data available for a given location.
- When multiple images are available, a specialized deep learning model detects whether a new roof installation has occurred between captures. If a new roof is detected, the roof installation date is calculated as the mid-point between the capture dates.
- In cases where a new roof installation is not detected in the imagery, or only one image is available, a separate machine learning model uses the available roof imagery, along with other third-party data, to predict the age of the roof.
- In cases where no imagery is available, roof age estimates can still be provided by analyzing other data sources such as building permits or assessor year-built data where available.
In addition to providing the roof installation date and roof age, the Roof Age product also provides an Evidence Type and a Trust Score with every prediction. These measures help assess the reliability of the predicted roof age and offer additional context for every prediction.
- Evidence Type indicates what data and method was used to calculate the roof age.
- Trust Score quantifies the reliability of each prediction. It reflects both the strength of the evidence available and the model’s confidence in its prediction.
These combined measures provide transparency around the roof age prediction and allow insurance carriers and other users to evaluate prediction quality and make informed decisions such as the likelihood of a roof replacement being required. For more information about Evidence Types, see this article.
Accessing Roof Age
Currently, Roof Age is available in Betterview with additional access options now being introduced. You can access Roof Age for a property in the following ways:
- In Betterview via the Construction tab - you can view imagery and compare roof age between two different survey dates.
- Via the Roof Age API - If you're a Nearmap customer, you can use this API to retrieve roof age data that includes roof instances, age and type. You can augment this with information about roof condition, materials, objects and so on, using Roof Age API in conjunction with the AI Feature API.
- Via the Property Insights and Property Now APIs - If you're a Betterview customer, you can use these APIs to get roof age data.
Using this information, insurance carriers can assess the cost of replacing the roof at the end of its life when issuing a new policy.
Frequently Asked Questions
Each roof which is detected on a parcel will have a separate roof age prediction, except for roofs less than 30sqm or less than 3m wide which are removed from the dataset.
Supporting evidence such as assessor year-built and permit data are applied to all roofs on the parcel.
When a roof installation is not detected in the imagery, Roof Age Gen2 model will predict the age of the roof using available images. This will typically have a lower associated Trust Score than if the roof installation is visible within imagery. This is because predicting age from imagery is inherently more difficult than comparing two images for change.
However, the minimum age of the roof can be calculated by looking at the age of the minimum capture date for that roof. Due to the high performance of the model which detects roof changes from imagery, this minimum age can be used with a high degree of confidence.
For example, a roof may have a predicted installation date of Feb 12th, 2012 (which corresponds to 13.5 years old as of Oct 2025) with Evidence Type 5 and a Trust Score of 58. However, the earliest capture date for that roof is Aug 16th, 2014. This means there is a high likelihood that the roof is at least 11 years old as of Oct 2025.
This approach may be useful in scenarios where it is important to have a high confidence, but less specific, roof age prediction.
For the training and evaluation of the Roof Age product, Nearmap has gathered an extensive set of human labelled pairs of images which identifies new roof installations. A representative set of these roofs is used as the testing set for Roof Age evaluation. This set is used to determine how often Roof Age Gen2 correctly identifies new roof installations (Evidence Type 7 and 8).
For the cases where roof age is extrapolated back from the roof imagery (Evidence Types 3, 4, 5, and 6), it is challenging to obtain a ground truth dataset of roof ages. Therefore, we simulated these cases by using the ground truth from imagery and then removing the appropriate imagery from the model to assess if the model can still correctly identify the age of the roof.
For the purposes of evaluation, the main metric that was focused on was the percentage of predictions within 2 years of actual. When a roof installation is correctly detected in imagery, Roof Age Gen2 is within 2 years of actual 96% of the time.
Finding information about Roof Age