The Amazon's Dieback Threshold Is Closer Than Climate Models Predicted

Luciana Gatti flies into the Amazon at dawn, when the air is still and the forest breathes out. From the cargo plane's open bay, she lowers a five-metre tube that drinks the sky — sampling carbon dioxide and methane every few seconds as the aircraft spirals down through 4,500 metres of atmosphere. She has been doing this since 2010, flying the same grid of transects over Amazônia Legal, Brazil's vast northern basin. For years, the numbers told the story everyone expected: the forest inhaled more carbon than it exhaled. It was a lung, a sink, a planetary thermostat.

Then, in 2019, the numbers flipped. The southeastern Amazon — a region larger than France — was emitting more carbon than it absorbed. Gatti, an atmospheric chemist at Brazil's National Institute for Space Research, ran the analysis again. The result held. By 2023, it had spread. The central Amazon, too, was now a net source of carbon. And in March 2026, Gatti's team published findings that shocked even the pessimists: across 38% of the Amazon basin, photosynthesis no longer compensates for respiration, decomposition, and fire. The forest is becoming a contributor to the problem it once buffered against.

The thing is, this was not supposed to happen yet. Climate models — the integrated assessment models used by the Intergovernmental Panel on Climate Change — have long predicted that the Amazon could cross a tipping point, beyond which it transitions irreversibly from tropical rainforest to savanna. But those models placed the threshold somewhere between 2060 and 2100, contingent on global temperatures rising 2.5 to 3 degrees Celsius above pre-industrial levels. Gatti's atmospheric data, corroborated by satellite measurements of tree mortality and soil moisture, suggest the process is already underway — at 1.2 degrees of warming.

What the Satellites Revealed

The Amazon dieback hypothesis has haunted climate science since the early 2000s, when Carlos Nobre and Thomas Lovejoy first articulated it in formal terms. Their argument was straightforward: deforestation reduces moisture recycling. The forest generates half its own rainfall through evapotranspiration — trees pull water from the soil and release it into the air, where it condenses and falls again. Cut down enough trees, and the rain stops. Without rain, the remaining forest dries out, becomes vulnerable to fire, and dies back. The system crosses a threshold and does not return.

For two decades, the debate centred on where that threshold sat: 20% deforestation? 25%? The Amazon has lost approximately 17% of its original forest cover since 1970, most of it in Brazil's south and east. Nobre and Lovejoy warned in 2018 that 20 to 25% deforestation, combined with global warming of 2 degrees, would trigger self-reinforcing collapse. Policymakers cited those numbers as a red line. Brazil's Forest Code, passed in 2012, was designed in part to keep deforestation below the 20% threshold.

But Gatti's data suggest the threshold is not a single number tied to deforestation alone. It is a moving target, influenced by three reinforcing pressures: direct deforestation, regional climate change, and global temperature rise. Her team analysed carbon flux data from 590 flight transects between 2010 and 2025, cross-referenced with NASA's GRACE satellite measurements of groundwater depletion, and the European Space Agency's Sentinel-2 imagery of canopy health. The southeastern Amazon, they found, has experienced a 30% reduction in wet-season rainfall since 2015. Dry-season length has increased by three weeks. Soil moisture in the top two metres — the layer accessible to most tree roots — has declined by 18%.

The Mechanism No One Saw Coming

Here is what makes this finding so unsettling: the forest is not dying primarily from chainsaws. It is dying from heat stress and drought — even in areas that remain nominally intact. Paulo Brando, a tropical ecologist at the University of California, Irvine, has spent 15 years running controlled burn experiments in the Amazon. In a 50-hectare plot near Santarém, his team lights low-intensity fires every three years — mimicking the kind of surface fires that now occur naturally in drought years. The trees survive the first burn. Some survive the second. By the third, mortality jumps to 60%.

Brando's work, published in Nature Climate Change in January 2026, shows that repeated low-intensity fire — the kind that would have been harmless in a wetter climate — fundamentally alters forest structure. Canopy cover drops. Light penetrates to the forest floor. Humidity falls. The microclimate shifts, and suddenly the forest is flammable in a way it never was before. "We are not talking about catastrophic fires," Brando says. "We are talking about fires you can step over. But they come back every few years, and the forest cannot recover between events."

The feedback loop works like this: warming temperatures increase evapotranspiration demand — the atmosphere can hold more water vapour, so it pulls harder on the trees. If rainfall does not increase proportionally, trees experience water stress. Stressed trees grow more slowly, produce less leaf litter, and become more susceptible to pests and disease. Some die. Dead trees release carbon. Carbon increases warming. Warming increases evapotranspiration demand. The loop tightens.

And then there is fire. Between 2015 and 2025, the Amazon experienced five severe drought years — 2015, 2016, 2019, 2023, and 2024. Each drought year corresponded with record fire activity. In 2024 alone, 180,000 square kilometres burned, an area larger than Cambodia. Most of those fires were human-ignited — set by ranchers clearing land or by smallholders burning crop residue — but they spread farther and burned hotter than they would have in a wetter climate. The fires are not the cause of dieback. They are the symptom. The forest has become kindling.

The Scientific Debate: How Close Is the Edge?

Not everyone agrees that the Amazon is on the brink. The debate within the climate science community is sharp, technical, and consequential. On one side are researchers like Gatti, Brando, and Nobre, who argue that observational data — atmospheric carbon measurements, satellite imagery, field studies — show the system is already destabilising. On the other side are Earth system modellers who argue that the data are noisy, that regional variability is high, and that a few bad years do not constitute a trend.

Pierre Friedlingstein, a climate modeller at the University of Exeter and lead author of the Global Carbon Budget, points out that the Amazon has experienced droughts before — in 2005, 2010, and 2015 — and recovered each time. "We need to be very careful about extrapolating from short-term variability," he says. "The 2023 and 2024 droughts were driven in part by El Niño, which is a natural mode of climate variability. When we move into a La Niña phase, rainfall will return, and the forest may begin absorbing carbon again."

The thing is, La Niña did return in late 2025 and early 2026. Rainfall increased by 12% across the central Amazon between December 2025 and March 2026. And yet Gatti's latest measurements, taken in February 2026, still show net carbon emissions. The forest is not bouncing back the way the models say it should.

Part of the disagreement stems from what scientists are measuring. Climate models simulate the Amazon as a single entity — a biome with average properties. But the Amazon is not uniform. The western Amazon, fed by rivers flowing from the Andes, remains relatively wet and stable. The southeastern Amazon, where deforestation is concentrated and rainfall has declined most sharply, is collapsing. Gatti's approach — measuring atmospheric carbon at high spatial resolution — captures this heterogeneity in a way that global models do not.

There is also a deeper methodological divide. Models project future states based on equations that describe how the world works. Observations describe how the world is working. When the two diverge, it usually means the equations are missing something. In this case, the models appear to underestimate the speed and severity of drought-fire feedbacks, and they do not fully capture the effect of repeated sublethal stress on tree mortality.

What This Means for the Carbon Budget

The stakes extend far beyond the Amazon itself. The forest currently stores between 150 and 200 gigatonnes of carbon in its trees, soil, and leaf litter. For context, humanity emits about 40 gigatonnes of CO₂ per year. If even half the Amazon's stored carbon were released over several decades — through fire, decomposition, and tree mortality — it would overwhelm any plausible emissions reduction scenario. The Paris Agreement's 1.5-degree target assumes that tropical forests will continue acting as carbon sinks, absorbing roughly 30% of annual anthropogenic emissions. If the Amazon flips, the global carbon budget collapses.

Climate scientists describe this as a "tipping element" — a component of the Earth system that can undergo a rapid, non-linear shift in state when pushed beyond a threshold. Other tipping elements include the West Antarctic Ice Sheet, the Atlantic Meridional Overturning Circulation (the current system that includes the Gulf Stream), and Arctic permafrost. The Amazon is different in one critical respect: its collapse would accelerate global warming, which in turn would push other tipping elements closer to their own thresholds. It is a cascade.

The Intergovernmental Panel on Climate Change's Sixth Assessment Report, published in 2021, assessed Amazon dieback as "low confidence" for warming below 2 degrees, and "medium confidence" for warming above 3 degrees. That assessment was based on models that, as Gatti's data now suggest, underestimated the risk. In internal discussions ahead of the IPCC's next report, due in 2028, authors are debating whether to revise the likelihood upward. Several scientists involved in the process, speaking on condition of anonymity, said the latest observational evidence has shifted the conversation.

Policy Implications: Can Dieback Be Reversed?

If the Amazon is approaching a tipping point, the urgent question becomes: can it be stopped? The honest answer is that no one knows. Tipping points, by definition, involve non-linear dynamics — small changes in forcing can produce large changes in outcome. That cuts both ways. It means degradation can accelerate suddenly. But it also means that aggressive intervention might stabilise the system if applied early enough.

Brazil's government, under President Lula da Silva, has renewed enforcement against illegal deforestation. Between 2023 and 2025, deforestation rates in the Brazilian Amazon fell by 34%, the sharpest two-year decline since monitoring began. But Gatti's findings suggest that stopping deforestation is no longer sufficient. The forest is degrading in areas that remain legally intact, driven by climate change that originates outside Brazil's borders. "We can protect every tree," Nobre says, "and the forest will still die if the planet warms another degree."

Some scientists are exploring more radical interventions. Restoration ecologist Robin Chazdon, at the University of the Sunshine Coast in Australia, has proposed large-scale reforestation along the Amazon's degraded southern and eastern edges — not to replace lost forest, but to restore moisture recycling and buffer the core. Models suggest that reforesting 10% of the basin's cleared area could increase regional rainfall by 3 to 5%, potentially enough to stabilise the central Amazon. The cost would be in the tens of billions of dollars. No international fund currently exists at that scale.

Other researchers advocate for "adaptive management" — accepting that parts of the Amazon will transition to savanna and focusing resources on defending the wetter, more resilient western Amazon. This is a form of triage, and it is deeply controversial. Who decides which forests to defend and which to abandon? Indigenous communities in the southeastern Amazon, whose territories are drying out, have not been consulted.

What We Still Don't Know

The most important uncertainty is speed. Gatti's data show that parts of the Amazon are already emitting carbon. But emission is not the same as collapse. The forest could stabilise at a new, degraded equilibrium — fewer trees, less carbon storage, reduced biodiversity, but still recognisably a forest. Or it could cross into a self-reinforcing die-off, where each dead tree makes the next one more likely. The models do not agree on which scenario is more probable, because the models were not designed to capture this level of granularity.

The second uncertainty is regional variability. The Amazon is not a single system; it is a patchwork of microclimates, soil types, and species assemblages. Some areas may prove more resilient than expected. Others may collapse faster. Predicting which is which requires much finer-scale monitoring than currently exists. Gatti's flights cover the basin at a resolution of approximately 200 kilometres. That is groundbreaking for atmospheric science, but it is still too coarse to identify which specific forests are most at risk.

The third uncertainty is whether dieback, once triggered, is reversible. Some tipping points are one-way doors. The Sahara Desert was once green; it transitioned to desert roughly 5,000 years ago due to shifts in Earth's orbit, and it has not returned. Other tipping points are reversible if forcing is reduced quickly enough. The Amazon may fall into either category, or neither. The science simply does not know yet.

Gatti continues to fly. Every month, the plane spirals down through the Amazon sky, and the tube drinks the air, and the numbers accumulate. She does not talk about tipping points in apocalyptic terms. She talks about them as thresholds that we are approaching and might still avoid — but only if the world acts as though the data are real. "The forest is telling us something," she says. "The question is whether we will listen before it is too late."

The Amazon has survived ice ages, megadroughts, and the arrival of humans. It has been resilient for 55 million years. Now, in the span of a single human lifetime, it is being asked to survive a rate of change for which evolution has no answer. Whether it does may determine not just the fate of the forest, but the trajectory of the climate system itself. And the window to find out is closing faster than the models predicted.