How DINOv3 is helping World Resources Institute restore forests and farms globally

For more than a decade, World Resources Institute (WRI) has worked to protect and restore forests, farmland, and other ecosystems by partnering with local businesses, government agencies, and other nonprofits.

Core to that work is transparency. Without highly accurate, locally relevant, and low-cost data, it’s impossible to know if the billions the world will invest in environmental preservation lead to tangible impact on the ground. Since 2014, WRI has operated the world’s premier forest monitoring platform, Global Forest Watch, which tracks forest and land use change using publicly available satellite data. Their latest analysis was sobering—it showed that tropical primary forests disappeared at a rate of 18 soccer fields per minute in 2024. Building on their long record of collaboration, Meta and WRI researchers applied Meta’s DINOv2 foundation model to map the height of the world’s tree canopy in unprecedented detail in order to improve the accuracy of future datasets.

The next frontier is applying that work to monitoring local forest restoration and agroforestry projects. WRI created TerraFund in collaboration with One Tree Planted and Realize Impact to finance dozens of small-scale projects in Africa that regrow forests and restore farms—the project launched in 2022 with seed funding from the Bezos Earth Fund and support from Meta.

Existing datasets could identify large trees disappearing, but monitoring newly growing trees is much harder because of the time it takes them to grow tall enough to be seen from space. The fund’s partners needed scalable, low-cost solutions that could leverage technology to identify which projects were the most effective across thousands of project sites in 27 African countries—meaning they needed to be able to count and monitor individual saplings as they grew.

In collaboration with Meta, John Brandt—data science lead for restoration at WRI and Land & Carbon Lab, a geospatial data research initiative convened by WRI and the Bezos Earth Fund—used DINOv3 to develop an algorithm to accurately count individual trees from drone and satellite imagery. Brandt used high-resolution satellite imagery and trained the model by looking at the geospatial boundaries of TerraFund projects. The initial results are striking. Researchers can see a growing sapling as early as 8 months after it’s planted and continue to monitor it as it grows. By comparing the self-reported data from local projects with the results of the algorithm, WRI can verify which projects are meeting their targets and recommend them for more financing.

Working with such a universal open source model, pre-trained on a large amount of satellite data, enables WRI to easily adapt the model for their own needs, which Brandt says aligns perfectly with WRI’s core values of partnership and collective impact. Beyond tracking tree growth, WRI’s Land & Carbon Lab will use the model to test scalable, low-cost monitoring and land-use planning techniques for all land cover types and ecosystems. It has already accelerated their ability to deploy models across diverse Earth observation tasks, driving efficiency and lowering costs.

The emergence of faster, more precise monitoring is reshaping how restoration progress is tracked—unlocking new potential for scale and investment. The Bezos Earth Fund’s $1 billion commitment to land restoration has helped catalyze this shift, backing innovations that are driving greater accessibility across the sector.

“Open source tools like DINOv3 fuel transparency and accountability in restoration,” says Emily Averna of the Bezos Earth Fund. “We’re excited to contribute to innovative tools that help bring clarity and speed to restoration efforts around the world.”

“Since DINOv3 is publicly available, a local NGO in rural Kenya can now generate high-res forest recovery maps with a laptop and just $10 in cloud credits,” Averna adds. “For philanthropies, it transforms restoration from a leap of faith into something we can see, measure, and trust to deliver the carbon and nature gains the planet urgently needs. It’s a strong example of what’s possible when philanthropy, the private sector, and NGOs join forces to drive measurable impact for people and the planet."

By linking $61 million worth of investments across African countries, Brazil, and India with these techniques, WRI is signaling that even marginal accuracy improvements can increase the volume of these transactions. This will help finance to reach smaller organizations.

“We are really excited about the ability to have a model robust enough to underpin larger financial transactions in the climate market,” says Brandt.