The Amazon rainforest, despite its small land area (less than 0.05% of the global land area), plays a crucial role in the Earth's climate system, hosting ~40% of the global tropical rainforests and contributing ~15% of the freshwater input to the world's oceans. Its climate exhibits robust seasonality, with a rainy season (austral summer to autumn) vital for the hydrological cycle, agriculture, and biodiversity. Extreme rainfall changes in the Amazon basin threaten ecosystem stability, biodiversity, and human livelihoods. The past decade witnessed a rapid decrease in Amazon rainfall, marked by prolonged droughts in 2005, 2010, and 2015, resulting in devastating bushfires, forest degradation, and socioeconomic losses. Interannual variations in sea surface temperature (SST), linked to El Niño-Southern Oscillation (ENSO) and north tropical Atlantic warming, have been implicated in these decreases. However, decadal-to-multidecadal internal variability, potentially involving the Interdecadal Pacific Oscillation (IPO) and the Atlantic Multidecadal Oscillation (AMO), also influences Amazon rainfall. The relative contributions of external forcing and internal variability to the prolonged drought of the past decade remain unclear. Climate models project a drier Amazon with increased drought frequency in a warmer climate, but the influence of internal variability on near-term projections is uncertain. Improving projections requires understanding and predicting internal decadal variability. Previous studies often relied on multi-model single-member projections, hindering the separation of internal variability from external forcing. This study utilizes a 100-member ensemble simulation from the CESM2-LENS to investigate the relative roles of internal variability and external forcing in the recent Amazon drought and to constrain near-term rainfall projections before 2040.

Several studies have explored the causes of decreased rainfall in the Amazon, focusing on interannual variability linked to ENSO and the North Tropical Atlantic. During El Niño events, anomalous SST warming weakens the Walker circulation, suppressing the ascending branch over South America and reducing rainfall. North tropical Atlantic warming displaces the Intertropical Convergence Zone (ITCZ) northward, further decreasing Amazon rainfall. Decadal-to-multidecadal variability has also been implicated, with the positive phase of the IPO potentially weakening the Walker circulation and increasing El Niño frequency, leading to prolonged Amazonian drying. A strong correlation between IPO and Amazon rainfall (r = −0.70, p < 0.01) over 1950–2019 suggests significant IPO modulation of decadal rainfall variability. The AMO also plays a role, with its positive phase shifting the Atlantic ITCZ northward and reducing Amazon rainfall. However, the mechanism for the prolonged rainfall reduction in the past decade and the relative contributions of external forcing and internal variability remain debated. Previous studies often used multi-model, single-member projections, making it difficult to separate the roles of internal variability and external forcing. This study employs a large ensemble approach to address these limitations.

The study uses data from the Climatic Research Unit (CRU) v4.06 for rainfall (1950–2019) and the Hadley Centre Sea Ice and Sea Surface Temperature dataset (HadISST) v1.1 for SST (1950–2019). The core analysis relies on 100-member simulations from the Community Earth System Model Large Ensemble Project (CESM2-LENS). These simulations are driven by the same external forcing (historical and future greenhouse gas forcing under SSP3-7.0) but differ in initial oceanic and atmospheric states. The ensemble mean represents the external forcing signal, while inter-member differences represent internal variability. The study focuses on Amazon wet-season (DJFMA) rainfall changes during 2010–2019 and 2020–2039. The IPO is defined using a tripole index based on SST anomalies in the central-eastern equatorial Pacific, northwestern Pacific, and southwestern Pacific. The AMO is defined using SST anomalies in the North Atlantic. To quantify the IPO's contribution to Amazon rainfall change, the authors adjust the IPO phase in each member to match the observed trend during 2010–2019. They remove the simulated IPO-related rainfall change and add the observed IPO-related change, effectively replacing the model's random IPO evolution with the observed one. Linear regression is used to link Amazon rainfall variation to the IPO. Similarly, to assess the IPO's influence on near-term rainfall projections (2020–2039), they remove the IPO's influence from the projected rainfall trends using linear regression, and then analyze the impact of imposing a simulated IPO phase transition (+1°C and -1°C) on the adjusted rainfall trends.

Analysis of the CESM2-LENS simulations revealed that internal variability, rather than external forcing, is the dominant driver of the recent Amazon drought. Observed rainfall decreased by -0.80 mm day⁻¹ decade⁻¹, while externally forced change was weak (-0.098 mm day⁻¹ decade⁻¹), explaining only ~12% of the observed drought. Internal variability, as reflected in the inter-member spread, showed a wide range of rainfall changes (-0.97 to +0.63 mm day⁻¹ decade⁻¹). Sub-ensembles of the 10 driest and 10 wettest members highlighted the significant impact of internal variability. The IPO was identified as the primary driver of internal variability influencing Amazon rainfall. The dry sub-ensemble exhibited a positive IPO-like SST pattern in the Pacific, strongly correlated (r = 0.72, p < 0.01) with the observed SST change. A significant negative correlation (r = -0.56, p < 0.01) existed between IPO changes and Amazon rainfall across all ensemble members. Adjusting the IPO phase in the simulations to match the observed phase transition revealed that the IPO contributes to ~45% (~40–49%) of the observed drought, much greater than the contribution from external forcing. Regarding near-term projections (2020–2039), the ensemble mean showed a slight negative trend, but large uncertainty remained due to internal variability. The difference in rainfall trends between the driest and wettest sub-ensembles was significant, emphasizing the role of internal variability. The spatial pattern of SST trends in the dry and wet sub-ensembles resembled a positive IPO-like pattern, highlighting the IPO's influence as a major uncertainty source in the near-term projections. Removing the IPO's influence from the projected rainfall trends reduced the uncertainty by ~38%, narrowing the range from -0.73 to +0.31 mm day⁻¹ decade⁻¹ to -0.42 to +0.23 mm day⁻¹ decade⁻¹. Simulations with imposed IPO phase transitions (positive or negative) showed that a positive IPO transition significantly increased the probability of a decreasing Amazon rainfall trend.

This study demonstrates the critical role of the IPO in driving the recent Amazon drought and influencing near-term rainfall projections. The finding that internal variability, specifically the IPO phase transition, explains a much larger portion of the observed drought than external forcing highlights the importance of considering internal climate variability in regional climate projections. The significant reduction in uncertainty achieved by removing the IPO's influence emphasizes the potential for improved projections by incorporating IPO forecasts. The study successfully utilizes a large ensemble approach to disentangle the contributions of external forcing and internal variability, offering a more nuanced understanding of the Amazon's hydrological system. While the study focuses on a single climate model, the robustness of the findings suggests their broader applicability. Future research could benefit from expanding the analysis to include multi-model ensembles to assess the generalizability of these findings and reduce uncertainty associated with a single-model approach.

This research demonstrates that the Interdecadal Pacific Oscillation (IPO) phase transition is the dominant driver of the recent Amazonian drought and significantly impacts near-term rainfall projections. The IPO explains a substantially larger proportion of the observed rainfall decrease than external forcing. Removing the IPO's effect from the projections significantly reduces the uncertainty range. Predicting the IPO's near-term phase transition is crucial for improving Amazon rainfall projections. Future work should include multi-model ensemble analyses to strengthen the findings' robustness and reduce reliance on single-model results.

The study's reliance on a single climate model (CESM2-LENS) is a limitation, potentially limiting the generalizability of the findings. While the large ensemble approach mitigates some uncertainty, multi-model studies are needed to confirm the results and assess model-specific biases. The focus on the IPO as the primary driver of internal variability does not preclude other contributing factors, which could be explored in future studies. The study's focus on a specific emission scenario (SSP3-7.0) could also affect the results, and investigations under other scenarios would provide a more comprehensive analysis.