Amazon Deforestation Halved Under Lula as Brazil Votes

Brazilian Amazon deforestation has more than halved since Luiz Inácio Lula da Silva took office, falling from 11,594 square kilometres in the year to July 2022 to 5,731 square kilometres in the year to July 2025 on INPE's PRODES record. Brazil votes on 4 October.

Carbon Brief published an analysis on 28 September 2026 that sets those observed figures against a counterfactual: what would have happened had Brazil's forest code, the law requiring landowners to keep and restore forest on their property, gone unenforced. The gap between the two comes to at least 20,000 square kilometres of forest still standing, an area close to the size of Wales or the US state of New Jersey.

What the numbers are and are not

The observed half of that comparison is official. PRODES is INPE's annual satellite mapping, measured over a year running from August to July. The figure for August 2025 to July 2026 has not been released yet; the preliminary count from DETER, INPE's near-real-time alert system, stands at 2,874 square kilometres. DETER is designed for enforcement rather than accounting and consistently reports less than PRODES eventually finds, so the final number will be higher. Even so, Brazil's Climate Observatory expects 2026 to come in below any year since the record began in 1988.

The counterfactual half is modelling, and it should be read as such. It comes from a 2023 study in Global Change Biology by researchers at the University of Oxford, the International Institute for Applied Systems Analysis and INPE, which used the GLOBIOM-Brazil model to compare forest code implementation against a baseline in which the code is not enforced. Carbon Brief uses that baseline as a stand-in for a Jair Bolsonaro election win in 2022. Nobody observed the alternative Brazil; a model produced it.

Dr Aline Soterroni of the University of Oxford, who led the 2023 work, told Carbon Brief that the reversal under Lula "shows how quickly deforestation can respond to political will", while stressing that her no-enforcement scenario is "relatively conservative", holding deforestation high rather than allowing it to climb. Claudio Angelo, international policy coordinator at Brazil's Climate Observatory, argued the opposite way, saying that had Bolsonaro won in 2022 there is reason to believe it would have produced "much higher rates than during his first term". Both readings point the same direction, which is the useful part: 20,000 square kilometres is a floor, not a central estimate.

The protections that are already gone

Presidential policy is not the only variable. The Amazon soy moratorium, the voluntary agreement under which traders refused soya beans grown on Amazon land cleared after 2008, effectively ended this year when major grain traders withdrew. Research in Nature Food credits it with preventing around 18,000 square kilometres of deforestation between 2005 and 2016, and a 2026 study in Science estimated its collapse could add 14,000 square kilometres by 2036. That is a private-sector commitment unwinding regardless of who wins in October, and it is the same pattern Parallax Nexus reported on 27 September, when a Harvard database of Russell 3000 climate targets found that about 42 per cent of revisions lowered ambition against 15 per cent that raised it.

Nor does the presidency settle the law. Right-leaning parties, including the Liberal Party linked to the Bolsonaro family, hold half the seats across Brazil's two congressional chambers, and in 2025 congress overrode Lula's veto of parts of an environmental licensing bill its critics call the devastation bill. Voters choose governors and members of congress on 4 October as well as a president, and a runoff follows on 25 October if nobody wins a majority.

What is genuinely unknown is how durable the enforcement gains are. Deforestation did not fall the moment Lula was elected; Soterroni describes restoring the law as a "gradual and imperfect process", and the drop took three years to reach half. A reversal would not be instant either, which cuts against the tidiest version of this analysis. The figure that will test it is the PRODES number for August 2025 to July 2026, and it lands after the votes are counted.