Restoring degraded land is one of the world’s great challenges.

We understand the problem: The planet loses 18 soccer fields of primary rainforest every minute — ecosystems critical for providing water, clean air, jobs and other benefits. We have the commitment to change: Governments have pledged to restore 1 billion hectares of land by 2030. What’s lacking is the finance.

Funding for nature needs to quadruple to $269 billion annually by 2030 to meet the world’s restoration commitments. At least $8 billion is needed in Africa alone for techniques like reforestation, agroforestry and mangrove restoration. There’s simply no way restoration projects — no matter how beneficial — will attract finance at the speed and scale necessary without significant public and private investment.

Fortunately, a new model is emerging — one that harnesses the power of artificial intelligence (AI) to pinpoint exactly where trees are regrowing and direct finance to the restoration projects most likely to deliver the most benefits. 

Satellite imagery of mangroves regrowing in Ghana
In Ghana, Hen Mpoano is building community nurseries to restore mangroves (right). With DINOv3, we can see the coastal ecosystem come back to life (left). Credit: Sena Affadu & Justine Spore/WRI

Why Doesn’t Restoration Attract Enough Finance?

To understand how AI could unlock restoration finance, it’s necessary to understand why investing in nature is so difficult.

Natural ecosystems are worth an estimated $125 trillion every year to the global economy. Healthy landscapes support industries like farming, forestry and tourism, which employ 1.2 billion people. Each $1 invested in restoring degraded landscapes can bring $7-30 in economic returns.

But while the investment opportunity is clear, the timing is challenging. Restoration delivers benefits that take place over time — clean water, healthy soils, higher incomes — but markets prioritize short-term returns. Communities and landowners therefore prioritize investments that turn a quick profit or abandon restoration projects before they reach maturity.  

Even if the market began to value long-term benefits, systems that track the results of restoration (known as “monitoring, reporting and verification,” or “MRV”) are not scalable or cost-effective. Without precise data, it’s impossible to know which restoration projects work and provide the best value. For instance, in Africa’s semi-arid regions, fewer than half of planted trees survive, but the best projects have survival rates above 80%. This uncertainty undermines trust in the sector and limits the finance available.

Finally, the people best qualified to restore land are the least likely to qualify for finance. Research shows that local organizations and land stewards like farmers — who manage more than 50% of the world’s land — are 6-20 times more effective at restoring land than larger organizations. Yet financiers favor well-established groups with international visibility and a track record with major funders.

Carbon and biodiversity markets could help, but certification schemes are often too complex, requiring thousands of farmers to enter long-term agreements and go through costly field verification processes. These options can work in countries with large tracts of land owned by a single person or company, but struggle in places like Rwanda and Kenya, where smallholder farmers dominate the landscape.

A New Approach for Restoration Finance, Powered by AI

Partnerships like TerraFund have built the infrastructure to find, screen and fund high-quality, locally led restoration finance projects. For example, in collaboration with One Tree Planted and Realize Impact and with anchor funding from the Bezos Earth Fund, WRI has directed $33 million in philanthropic funding and connected grants, debt and equity investments to nearly 200 restoration projects across Africa. We’ve started this work in Brazil and India, too.

But philanthropy alone cannot finance local organizations at a larger scale. We need to leverage philanthropic capital to unlock private and public finance for local groups — and to do that, we need to verify their results to build trust in the market.

This could finally be possible thanks to an open-source, AI model called DINOv3, developed by Meta with WRI’s assistance. We are adapting this model to count individual trees from space using high-resolution satellite imagery. This is a major advancement over current methods of restoration monitoring, which require on-the-ground reporting to estimate the number of trees growing in a project site.

DINOv3 counted of 7,064 agroforestry trees planted in January, 2022 in Ashanti, Ghana by Eco-Care Ghana. The blue boundary is the planting site, and each yellow dot represents one tree. Credit: WRI

What is the DINOv3 model?

Developed by Meta AI with support from Land & Carbon Lab, a geospatial data research initiative convened by WRI and the Bezos Earth Fund, DINOv3 is an advanced AI computer vision model. DINOv3 scans massive, unlabeled datasets of images and teaches itself to identify distinct objects. 

It’s a “foundation model,” designed to be broadly reusable across countless visual tasks. Unlike many AI models designed for a specific purpose — such as satellite monitoring, medical imaging or manufacturing — DINOv3 requires only minimal training data to be adapted for each new task, like counting trees shown in a satellite image. This dramatically reduces the need for human data labelers and makes it faster to align the model with data collected in the field. Without the need to build a new model for every use case, DINOv3 represents a major leap forward for applying AI to Earth observation.

DINOv3 models are already outperforming specialized models for geospatial monitoring without the need to adapt to each specific satellite, aerial or drone dataset analyzed. Earlier versions like DINOv2, which WRI and Meta used to map tree canopy height in unprecedented detail, required large amounts of labeled satellite imagery to produce accurate results. With DINOv3, we can build, test and improve models more quickly, delivering consistent results across different image sources — whether from a satellite, a researcher’s drone or a farmer’s smartphone.

For example, we can compare the number of trees the AI model counts within a satellite image of a restoration project site to the number a field crew counted manually. If the numbers don’t align, researchers can adjust the model  and then apply it to thousands of other sites without needing to collect more field data.

Calibrated with field-collected data from TerraFund projects, this AI algorithm allows us to remotely measure the growth of individual trees as soon as 8 months after planting. Preliminary results are 80% as accurate as traditional forestry methods that measure trees in the field, at 3% of the cost.

Through this low-cost, scalable verification, we can bring new transparency to the restoration sector. We can learn which projects lead to the highest survival rates for planted trees, target projects within the most important landscapes for revitalizing biodiversity and water quality, and track the impact of each project for decades at a very low cost.

Importantly, we can also use AI to create a new model for financing restoration. If we can count each planted sapling and track its growth over time, we can develop financial structures that reward local organizations for growing the right trees in the right places. We can then re-sell the impact of those verified trees to buyers that depend on healthy land — everyone from global agricultural companies to local utilities.

This approach could significantly scale up restoration finance by bringing together funding from philanthropies, governments and the private sector.

How Would New Restoration Finance Models Work?

This new restoration model requires three key actors to work together:

  • Local organizations and enterprises who lead high-quality restoration projects. These are community groups like Rwanda’s Forest of Hope Association, which works with farmers to regrow a nearby forest, or businesses like Fanteakwa Cooperative Cocoa Farmers and Marketing Union, which helps Ghanaian farmers revitalize their underproductive fields by incorporating native shade trees.
  • An investor or grantor willing to cover upfront costs and take on early risk. This role can be filled by return-seeking investors like banks, development finance institutions or funds; by grantors like philanthropies or governments; or streams of finance that blend both. Realize Impact, for example, is a non-profit investor that facilitates loans and grants for local enterprises, with the explicit goal of minimizing transaction costs.
  • An outcome buyer who may not be able to pay upfront, but can bear the entire cost of restoration over time by paying for surviving trees for every year they can be verified. These could include global commodity suppliers like a cocoa company, an international grocer that sources fruits, local water or energy companies, or any business with a direct economic incentive to invest in healthy landscapes.

Dinov2 v Dinov3 models.

The DINOv3 model could transform Earth observation. In some regions of the world, it estimates tree canopy 50-80% better than the model’s previous version. Credit: WRI

Three financing structures could leverage AI data and bring these three actors together to finance verified trees:

1) Financing Productive, Tree-Filled Farms

The transition from conventional farming to agroforestry, where trees are integrated alongside food crops, isn’t bankable for most farmers given high upfront costs and slow returns. But growing sustainability regulations and consumer pressure are driving demand for creative market solutions.

Imagine a group of companies that source cocoa from Ghana. To demonstrate progress toward their sustainability commitments, they want to grow trees on cocoa farms in their supply chain, but they don’t have the upfront capital to pay for the trees. The companies could ask a grantor to pay a local organization to help farmers plant native trees on each farm. Every year that independent AI data can verify that those trees survive, cocoa companies pay the grantor. After a few years, the grantor is fully reimbursed — and can even revolve that capital to new projects, beginning the cycle again.

The company doesn’t have to take on the risk of providing up-front finance, and the donor gets its money back. Crucially, the payments for tree survival make agroforestry more financially viable for farmers by helping them in the years before their new agroforestry systems are more profitable than conventional treeless farms.

2) Financing Trees for Climate & Nature

Degraded ecosystems can hurt a company’s bottom line. For example, if deforestation destroys habitat for iconic wildlife, a high-end tour operator has nothing to show its visitors. It’s simple: Fewer trees, fewer animals, few visitors, lower profits.

But ecological restoration projects often struggle to attract finance because they have no way to value the services they produce. For example, the Rwanda Forestry Authority and Wildlife Conservation Society identified 10 priority sites that could be restored to vibrant habitat for chimpanzees and sequester significant amounts of carbon, but couldn’t mobilize the required finance.

New models could help. For example, buyers like the tour operator and government agencies could partner with researchers to build ‘proxy’ indicators that associate trees with biodiversity, carbon, water or other outcomes — and pay a premium for each tree that remains standing. A multilateral development bank could issue a bond on the financial market, which would be purchased by investors at a near-market interest rate. The investors take on a risk because a portion of the bond’s proceeds will be channeled to community organizations, who will plant trees in priority areas to enhance chimpanzee habitat.

When the community organizations complete the work and the AI dataset verifies that an agreed number of new trees are growing in priority areas, the wildlife tour operator agrees to pay the bank for those trees at an agreed price. The bank then uses the capital to reimburse investors — and can issue another bond to pay for more restoration.

3) Financing Restorative Agricultural Value Chains

Restorative agricultural value chains offer huge investment potential — Kenya alone presents a $1.5 billion investment opportunity — but local enterprises face low market prices due to weak demand and supply chain gaps. Available finance often carries high interest rates and short timelines that don’t match businesses’ needs. As a result, many fail, and harvests go to waste.

Let’s take the example of a Kenyan company that sells mango seedlings to smallholder farmers, helps them plant in locations with sufficient water and the right soil, collects fruit from farmers at a guaranteed price, and then turns it into shelf-stable juice for the local market. The company needs a new processing line to boost its export capacity to a European buyer with whom they have an established relationship. But an impact investor will only issue a loan at 15% interest per year, and the company can only take on an 8% rate.

The investor and the buyer could sign an agreement with the company: The investor will issue the market-rate loan at 15% interest, and the buyer will agree to pay a premium price for the mango pulp, worth the equivalent of 7% interest, if the company hits an annual target on the number of verified trees grown. The company can use that premium payment to bring the effective interest rate of the loan down to 8%, aligned perfectly with their needs. With more of this type of creative funding, businesses can demonstrate the ability to pay back debt, making them attractive to commercial investors.

Money Can Grow on Trees

We know that counting trees is an incomplete stand-in for the vast amount of ecosystem services and economic benefits that holistic land restoration projects provide. But it’s what we can measure at scale, at low cost, and with high accuracy thanks to new AI models. Let’s start there.

DINOv3’s ability to count trees at unprecedented accuracy, like at this Eco-Care Ghana site, can help unlock finance for restoration. Credit: WRI

We need philanthropies to finance local organizations and test this approach, as well as private and public investors who can create a market for verifiable solutions. We need policymakers to understand how to equitably share the benefits of verified trees, once sold. We need data ethicists and local communities to develop guardrails for protecting sensitive data.

We also need data scientists around the world to build on this model and verify the more important outcomes associated with the humble tree. For example, mapping the size and distribution of newly planted trees can tell us how much carbon they store, or the impact they’re making on soil health or water security. This could radically simplify the data required to measure the important work communities are doing to repair the world’s ecosystems — and then pay them for it.

Restoration goals remain unfulfilled. The climate crisis continues. Now is the time to explore how AI can reset the conversation and bring prosperity to people, nature and climate.