Amazon Deforestation Is Falling. The Forest Is Still Burning.

Brazil's National Institute for Space Research (INPE) reports that Amazon deforestation is continuing to fall, and is on pace for a possible record low.

The numbers are real and they come from satellites, not from a ministry's press office. They deserve to be stated plainly before the caveats arrive.

The numbers

Annual official figure (PRODES). In the deforestation year running August 2024 to July 2025, clearing in the Brazilian Amazon totalled 5,796 km² — down 11.08% on the previous year, and the lowest level in eleven years. It is the third consecutive annual decline, and roughly half the 2022 figure.

Near-real-time alerts (DETER). Between 1 August 2025 and 31 January 2026, DETER detected 1,325 km² of clearing, down from 2,050 km² in the same window a year earlier — the lowest for that interval since 2014.

The Cerrado too. The savanna bordering the Amazon, and increasingly the more active agricultural frontier, fell to 1,905 km² from 2,025 km². The Cerrado gets a fraction of the Amazon's coverage and is one of the most biodiverse savannas on Earth, as well as the water source for Brazil's major river basins.

Environment Minister Marina Silva has said Brazil could record the lowest Amazon deforestation rate since record-keeping began in 1988 if current efforts hold, attributing the decline to enforcement and to municipal cooperation — most of the 81 highest-deforestation municipalities have now joined a federal compact against illegal clearing, funded partly through the Amazon Fund.

PRODES and DETER are not the same number

Two systems, two jobs, and conflating them is the most common error in coverage of this subject.

PRODES is the official annual survey. Higher-resolution imagery, produced once a year for the August–July cycle. It is the number of record.

DETER is a near-real-time alert system. Its purpose is to point enforcement at active clearing this week, not to produce a precise total. It is faster and coarser.

DETER is a reliable indicator of short-term trend. It is not a substitute for PRODES, and a DETER figure and a PRODES figure for overlapping periods will not match. If you see two different Amazon numbers, this is usually why — and neither source is being dishonest.

Why monthly numbers lie

Neither satellite system sees through cloud.

Brazil's rainy season runs roughly November to April. Monthly figures in that window are volatile — not because clearing stops, but because the satellites are looking at cloud tops. Most large-scale clearing happens in the dry months anyway.

This is why the deforestation year ends on 31 July, and why year-over-year comparisons across a full cycle are the only ones that mean much. A cheerful December number is close to meaningless, in either direction.

The trailing-12-month DETER figure illustrates how live this data is: reported at 3,770 km² in February 2026 against 4,245 a year prior, and down to 3,182 km² by June 2026 — the lowest for any 12-month period in DETER's record back to July 2014. Both figures are correct for their dates. The number moves.

The thing the headline doesn't say

Here is the distinction that matters most, and it is a definitional one:

INPE tracks deforestation separately from forest loss due to fire.

PRODES measures clearing — the deliberate removal of forest cover, the chainsaw-and-bulldozer conversion of forest to something else. It does not measure a forest that burned.

So a year can post a falling deforestation rate and a rising area of burned forest, and both statements are true and neither is spin. Reporting through 2025–26 has repeatedly found exactly that shape: clearing down, fire up.

A burned forest is not a cleared forest in the statistics. It is, however, a burned forest.

This matters more each year, because Amazon fire is increasingly drought-driven rather than chainsaw-driven. As Mongabay put it, the fate of the forest is increasingly shaped by forces beyond the reach of chainsaws alone. Enforcement can stop a rancher. It cannot stop a drought.

Deforestation policy is succeeding at the thing it targets. The threat is partly moving to something that policy doesn't target.

What we are not telling you

One widely-syndicated account of this data included an independent monitoring figure roughly an order of magnitude below every other dataset for a comparable period. We could not reconcile it, and we are not going to reprint a number we cannot make sense of just because a good outlet printed it. If it was a transcription error, repeating it would launder it into fact.

The datasets we do cite — PRODES, DETER — are INPE's own, and their dashboard is public and linked below. You can check us.

Why this is on AEDO1

Conservation coverage has a habit of treating good news as either a victory lap or a trap. This is neither.

Three consecutive years of decline, a halving since 2022, an eleven-year low, and a credible shot at an all-time low: that is what a policy working looks like, and pretending otherwise would be its own kind of dishonesty.

And the forest is still burning, and the fires are increasingly not about enforcement at all.

Both. At the same time. That's the story.