Previously undetected disturbances are the dominant driver of carbon loss in protected Amazonian forests

Liang Wan , Philippe Ciais, Aurélien de Truchis, Ewan Sean, Fabian Jörg Fischer, Ricardo Dalagnol, Clément Bourgoin, Yidi Xu, Nan Meng, Xiaojun Li, Arthur Fendrich, Clement Albergel, Youngryel Ryu, Stephen A. Sitch, Thais Michele Rosan, Jean-Pierre Wigneron, Fabien H. Wagner, Shidong Liu, Lei Zhu, Agnès Pellissier-Tanon, Ibrahim Fayad, Gabriel Belouze, Clement J. F. Delcourt, Sassan Saatchi, Luiz Eduardo Oliveira e Cruz de Aragão, Jean P. Ometto

Amazonian forests face increasing pressures from human-induced deforestation and degradation, compounded by natural disturbances such as droughts and windthrow events. In response, forest protection policies strengthened and formalised large protected areas and Indigenous lands, now covering 43% of the region. However, clearings and degradation in these protected areas are complex and hard to detect with current medium-resolution satellite observations. Here we mapped canopy height (10 m) and aboveground biomass carbon (30 m) changes from 2019 to 2022 by combining ~727,000 spaceborne LiDAR measurements with Sentinel-1/2 imagery using a deep learning framework. We estimate a net aboveground carbon loss of −257 ± 47 Tg C yr−1 across the entire Amazon forest, with changes in carbon density aligning with field inventory data. Gross losses from agricultural expansion, logging, fire, road construction, and drought (−437 ± 42 Tg C yr−1) exceed gross gains of 180 ± 20 Tg C yr−1, with secondary-forest regrowth contributing 38 ± 7 Tg C yr−1. Approximately 28% of net carbon loss occurred within protected areas, including 6% in Indigenous lands. We found that 80% of total carbon losses were associated with small disturbance patches and low height losses, which remain largely undetected by Landsat-based disturbance products yet supported by independent airborne LiDAR and PlanetScope imagery. As a result, carbon losses exceeded Landsat-based estimates by 84% in non-protected areas and 166% in protected areas, and were 36–55% higher than passive-microwave-based estimates3,4 and consistently greater than those from dynamic vegetation models. These findings highlight the critical role of high-resolution monitoring in uncovering hidden carbon losses and underscore the need for improved conservation strategies, even within formally protected Amazonian forests.