New feature: Understand tree distribution in your city with Restor’s tree crown count

Having a tree crown count helps ecologists and forest managers gather data, understand the ecological processes, and monitor tree populations, health, growth rates, and biodiversity in a certain location. As a result, they can make informed decisions about conservation, reforestation, resource management, and understanding the overall health and dynamics of forest ecosystems. Previously, however, this process was conducted through labor-intensive ground-level field surveys. In line with Restor’s commitment to embedding science and transparency in the restoration movement, Restor introduces the Tree Crown Delineation (TCD) feature that uses drone photos and an AI model to count tree crowns.

What exactly is Tree Crown Delineation (TCD)

TCD means identifying and outlining the outer edges of a tree's crown, which includes all the branches and leaves extending from its trunk. This process helps to map the exact location and size of each tree, providing detailed information about the structure and health of forests and tree-covered areas. The dataset that backs this capability is one of the largest and most diverse.

TCD and tree count are similar yet different. While tree counting simply involves counting the number of trees in an area, TCD goes a step further by identifying and outlining the edges of each tree's crown (the branches and leaves). Consequently, in addition to the number of trees, we can get their precise location, size, and overall structure.

Technology behind the TCD

To create the TCD AI model, we use advanced AI to find and outline the tree cover in a photo. An extensive dataset is used to build the model:

Diverse imagery: Models trained on over 20,000 hectares of photos.
High Resolution: Images resampled to 10 cm per pixel.
Global Coverage: The dataset spans most terrestrial biomes.
Labelled Trees: Over 280,000 hand-labelled trees.

You can find more technical information about our models and dataset by reading our preprint paper, and you can even dig into the code itself. If you’re interested in contributing to this project, we also have an active Citizen Science campaign on the Zooniverse platform, where we’re asking volunteers to tag trees to help improve the dataset.

How to request a detailed report?

Log in to your Restor account and navigate to one of your sites (where you have drone imagery). Look for the "Tree Crown Delineation" button (shown in the image below). Next, click on it to initiate the process and upload drone images of your site. These images are essential for analysing each tree's crown individually, providing valuable insights into forest health and tree distribution.

Follow the instructions to upload your orthomosaic drone data in the correct format. Once uploaded, our team will generate a report within three weeks, sent directly to your email address for review. Currently, this service is available upon request, and we will do our best to meet as many requests as possible.

How can knowing the tree crown count help Restor users?

Transparency. The TCD represents a powerful tool that allows field practitioners to evaluate various indicators of restoration progress and insights into forest health. This capability enables transparent monitoring where Restor users can observe changes and enhancements over time, crucial for verifying commitments made by stakeholders. It helps ensure that restoration efforts are effectively managed, stay on target, and align with their intended objectives.

TCD vs LIDAR: An accurate and cost-effective alternative

Case in point: The Grün Stadt Zürich (Green City Zürich) using tree cover count as a re-greening target The city of Zürich is known as one of the most liveable cities in the world. The city planners have laid out an ambitious plan to improve “green cover” in the city, which includes an increase of canopy area coverage from 17% (measured in 2018) to 25%. The strategy includes modifications to building planning to allow for more tree coverage, as well as better conservation strategies in the already-wooded areas, like the Zürichberg. To assess canopy coverage, Zürich currently uses Airborne LIDAR scanning (ALS) to measure vegetation height and coverage. This is an effective and standard method, but ALS is expensive to collect and, in other countries, is not always available. In benchmark testing, we found our models applied to Swisstopo aerial imaging are around 90% accurate when identifying tree cover compared to the LIDAR data. Our models could therefore be used as an accurate and cost-effective alternative to LIDAR scanning for coverage measurements.

The current Tree Crown Delineation models work best for detecting individual trees in areas where the canopy is open and for estimating the overall percentage of tree cover in any environment. However, they are not yet reliable for counting trees in areas where the forest is dense. Restor plans to improve the model with expected future updates. Questions? Please write to us at community@restor.eco