90% of Amazon Deforestation Happened Outside Indigenous Lands. That Number Is Not a Model.

On 28 October 2025, ahead of the COP30 climate talks in Belém, the Environmental Defense Fund published an analysis titled The Importance of Protected Areas in Reducing Jurisdictional Deforestation in the Legal Amazon.

Its headline projection: without Indigenous lands and other protected areas, deforestation across the Brazilian Amazon would be 35% higher and carbon emissions 45% higher.

Before we go further, a distinction that most coverage of this study collapsed — and that changes how much weight each number can carry.

One of these numbers is measured. One is modelled.

The observation. Between 1985 and 2020, 90% of Amazon deforestation occurred outside Indigenous lands. Inside them, just 1.2% of native vegetation was lost over those 35 years.

That is not a projection. It is a count of what happened, from satellite records, over three and a half decades. It requires no assumptions about land prices or counterfactual worlds. Indigenous territories are, empirically, the best-preserved land in the Amazon.

The projection. The 35% and 45% figures are model outputs. They describe a world that does not exist — the Amazon as it would have been without these designations. To produce them, researchers used cattle-ranching and agricultural land prices as a proxy for the economic incentive to clear, cross-checked against geographic and environmental factors and current deforestation trends.

That is a reasonable method. It is still a model, and a model is an argument with numbers in it. Change the assumed land prices and the counterfactual moves.

Both are worth knowing. They are not the same kind of fact, and an article that presents them in the same sentence with the same confidence is doing its reader a disservice.

We are also going to say plainly: EDF is an advocacy organisation, and this report was released to influence COP30 delegates. That does not make it wrong — advocacy groups fund good research, and the observed figures are corroborated elsewhere. It does mean the modelled figures deserve the scepticism you would apply to any number produced by someone who wants a specific outcome. The lead researcher, environmental economist Breno Pietracci, is explicit that the model captures avoided deforestation both inside boundaries and beyond them through spatial interactions — which is a real effect and also the kind of assumption that moves a result.

What the study proposes

63.4 million hectares of Brazilian Amazon forest currently sit undesignated — neither Indigenous land nor protected area nor private title. Almost two-thirds of it is in Amazonas state.

Undesignated public forest is the most vulnerable category there is. It is what land grabbing takes.

Designating that 63.4 Mha would, the model projects:

• cut Amazon-wide deforestation by up to 20% — 2.5 million hectares — by 2030
• cut deforestation-linked carbon emissions by up to 26% — 1.2 GtCO₂e, roughly Japan's annual emissions — by 2030

And the study estimates that existing Indigenous lands and protected areas will prevent 4.3 million hectares of deforestation between 2022 and 2030, avoiding 2.1 GtCO₂e — more than Russia's annual emissions, around 5.6% of the world's.

The state-by-state numbers are where it gets specific

Projected increase in deforestation without Indigenous lands or other protected areas:

The spread is the interesting part. Roraima at 71% and Maranhão at 11% are not telling the same story.

A high number does not mean a state is doing well — it means protection is doing a lot of work there, because the pressure is high and the protected share is large. Maranhão's 11% partly reflects that much of its forest is already gone. You cannot protect what has been cleared. A low counterfactual can be a symptom rather than a success.

The carbon ledger

Between 2001 and 2021, Amazon territories managed by Indigenous Peoples with recognised land rights:

• released ~120 million tonnes of CO₂ per year
• removed ~460 million tonnes per year
• net: 340 million tonnes removed annually — about the UK's annual fossil fuel emissions

These territories are not carbon-neutral. They are a net sink, at national scale, every year, and they achieve it as a by-product of people living on their own land.

Recognition is the variable, not presence

The evidence points at legal recognition specifically, not merely Indigenous occupancy.

A 2021 UN report found deforestation rates up to 50% lower in Indigenous territories in Latin America. Research in the Brazilian Amazon found deforestation drops significantly once territories are legally recognised, with communities holding collective property rights seeing roughly a 66% decrease in border deforestation.

The mechanism is not mysterious. A recognised boundary is a boundary a court will enforce and a land grabber's lawyer has to argue with. An unrecognised one is a line on somebody's map.

That is what makes the 63.4 Mha of undesignated forest the live question. It is not a proposal to change how anyone uses the land. It is a proposal to write down who it belongs to.

The pressure that recognition doesn't stop

In 2025, Indigenous and community lands in Brazil lost 676,000 hectares of forest — and 80% of that loss was fire, not clearing.

This is the same crack that runs through the deforestation numbers generally: as we covered in Amazon Deforestation Is Falling, clearing is down sharply while fire is not.

A legal boundary stops a bulldozer. It does not stop a drought. Land tenure is demonstrably one of the most effective forest-protection instruments available — and it is not sufficient on its own, and the 2025 fire figure is the evidence.

Why this is on AEDO1

The Legal Amazon's nine states hold about 60% of the rainforest and 28 million people, 36% of whom live in poverty. Any account of Amazon protection that treats it as an empty wilderness problem is describing somewhere else.

The strongest sentence in this material is not the 35% projection that made the headlines. It is the measured one: over 35 years, 1.2%. That number was not produced by a model, a campaign, or a summit. It was produced by people living where they have always lived, and it is the highest-performing conservation result in the basin.