Near-term projection of Amazon rainfall dominated by phase transition of the Interdecadal Pacific Oscillation

The Amazon basin, though occupying less than 0.05% of global land area, contains ~40% of global tropical rainforests and contributes ~15% of freshwater to the oceans, making it critical to the Earth’s climate system. Its climate exhibits strong seasonality, with a wet season (December–April) that supports the regional hydrological cycle, ecosystems, and livelihoods. In the past decade, the Amazon experienced rapid rainfall decreases and widespread, severe droughts (notably in 2005, 2010, 2015) that triggered fires, forest degradation, biomass declines, biodiversity loss, and socioeconomic impacts. On interannual timescales, sea surface temperature (SST) variability, including ENSO and North Tropical Atlantic warming, reduces Amazon rainfall by weakening the Walker circulation and shifting the Atlantic ITCZ northward. On decadal scales, internal variability such as the Interdecadal Pacific Oscillation (IPO) and the Atlantic Multidecadal Oscillation (AMO) can modulate Amazon rainfall; a positive IPO weakens the Walker circulation and favors El Niño, associated with prolonged Amazon drying, with a strong historical correlation (r = −0.70) between IPO and Amazon rainfall (1950–2019). However, the mechanisms behind the recent prolonged reduction remain debated, particularly the relative roles of external forcing versus internal variability. Near-term projections are further complicated by internal variability potentially overwhelming anthropogenic signals. Leveraging single-model large ensembles can disentangle internal variability from external forcing to improve near-term projections. This study uses a 100-member CESM2 large ensemble to quantify the contributions of internal variability—especially the IPO—to the post-2010 Amazon drought and assess its constraint on near-term projections before 2040.

Prior studies link Amazon rainfall variability to Pacific and Atlantic SST anomalies: ENSO events reduce Amazon rainfall via a weakened Walker circulation and suppressed convection over tropical South America, while North Tropical Atlantic warming shifts the ITCZ northward, decreasing rainfall. Decadal-to-multidecadal modes such as the IPO and AMO also modulate regional hydroclimate: a positive IPO phase is associated with weakened Walker circulation, more frequent/intense El Niño events, and Amazon drying; a positive AMO can shift the Atlantic ITCZ northward and reduce Amazon rainfall. Despite these insights, there remains disagreement over the drivers of the prolonged post-2010 drought and the relative contributions of internal variability vs external forcing. Climate models generally project a drier Amazon under warming with more frequent droughts, driven by long-term SST changes, yet internal variability introduces substantial spread in near-term projections. Previous assessments often relied on multi-model, single-member projections, limiting the separation of internal variability from forced responses. Advances in large ensemble simulations enable clearer attribution of internal variability and its influence on near-term projections.

Data: Monthly gridded rainfall from CRU v4.06 (1950–2019; 0.5°) and SST from HadISST v1.1 (1950–2019; 1°). Model: CESM2 Large Ensemble (CESM2-LENS) with 100 members, driven by historical forcings to 2014 and SSP3-7.0 thereafter (2015–2100). A 2000-year CESM2 pre-industrial control run is used to assess intrinsic variability relationships. A 9-year running mean is applied to extract decadal signals. Separation of signals: The ensemble mean across 100 members represents externally forced response; deviations of individual members from the ensemble mean represent internal variability. Indices: The Amazon rainfall index is the 9-year running mean of DJFMA rainfall anomalies averaged over 12°S–2°N, 75°W–48°W. The IPO is defined via the Tripole Index: DJFMA SST anomaly in 10°S–10°N, 170°E–90°W minus the average of anomalies in 25°N–45°N, 140°E–145°W and 50°S–15°S, 150°E–160°W; observed IPO uses detrended SST, while model IPO uses internal SST. The AMO is DJFMA SST anomaly over 0°–65°N, 80°W–0°, with global mean removed. Attribution for 2010–2019: Members were analyzed for DJFMA rainfall trends. Two sub-ensembles (dry10 and wet10) were built from the 10 driest and 10 wettest members to highlight internal variability. To quantify IPO’s contribution, for each member the simulated IPO-related Amazon rainfall change was removed, then replaced with the observed IPO-related change using linear regression of Amazon rainfall onto IPO over 2010–2019. The adjusted rainfall trends reflect external forcing plus the observed IPO phase transition; IPO-contributed trends are isolated by difference. Near-term projection (2020–2039): Trends in Amazon rainfall were examined across members, and sub-ensembles with driest and wettest trends were identified. Regression maps of projected grid-point rainfall and SST trends against the projected Amazon rainfall trend were computed to diagnose associated SST patterns. To quantify IPO’s contribution to projection uncertainty, the IPO-induced component of each member’s rainfall trend (regression of rainfall onto IPO times the member’s IPO trend) was removed, and the spread reduction was assessed (range, inter-member SD, 5th–95th percentile spread). Scenario experiments were created by imposing a common IPO phase transition of ±1 °C over 2020–2039, adding the IPO-induced rainfall impact (from each member’s regression) to the IPO-removed trends to assess changes in ensemble mean and probabilities of drying.

- Observed Amazon wet-season (DJFMA) rainfall decreased significantly during 2010–2019 by −0.80 mm day−1 decade−1 (areal mean), indicating widespread drying over much of the basin. - Externally forced ensemble-mean trend is weak: −0.098 mm day−1 decade−1, explaining ~12% of the observed drought; internal variability across members spans −0.97 to +0.63 mm day−1 decade−1, encompassing observations. - Sub-ensembles: dry10 yields −0.69 mm day−1 decade−1 (comparable to observed), wet10 yields +0.51 mm day−1 decade−1, underscoring internal variability’s dominance. - SST patterns: Observed and dry10 members show a positive IPO–like Pacific pattern; dry10–wet10 SST trend difference strongly correlates with observations (pattern r = 0.72; p < 0.01). Across members, IPO–Amazon rainfall changes correlate negatively (r = −0.56; p < 0.01); AMO shows no significant correlation for this period. - IPO attribution (2010–2019): After adjusting members to the observed IPO phase transition, the ensemble-mean Amazon rainfall reduction is −0.45 (−0.50 to −0.40) mm day−1 decade−1, ~56% (50–62%) of observed. The IPO-induced reduction alone is −0.36 (−0.39 to −0.32) mm day−1 decade−1, ~45% (40–49%) of observed—far exceeding the external forcing contribution (~12%). - Near-term projections (2020–2039, SSP3-7.0): Ensemble mean trend is −0.13 mm day−1 decade−1, but dry10 is −0.47 and wet10 is +0.16 mm day−1 decade−1; the dry–wet difference is significant and larger than the forced trend. Associated SST trend differences resemble a positive IPO pattern. - IPO as leading uncertainty source: Removing IPO-induced components narrows the trend range by ~38% (from −0.73 to +0.31 to −0.42 to +0.23 mm day−1 decade−1), reduces inter-member SD by 31% (0.182 to 0.126), and shrinks the 5th–95th percentile spread by 22% (0.58 to 0.45 mm day−1 decade−1). - Imposed IPO phase transitions (2020–2039): A +1 °C negative-to-positive IPO transition shifts the ensemble-mean trend to −0.35 mm day−1 decade−1 and raises the probability of a decreasing trend from 80% to 94%. A −1 °C positive-to-negative transition yields +0.12 mm day−1 decade−1 and lowers the probability of a decreasing trend to 27%.

Findings demonstrate that internal decadal variability, specifically the IPO’s phase transition, is the primary driver of the recent Amazon decadal drought and a dominant source of uncertainty in near-term rainfall projections. External forcing alone cannot account for the observed magnitude of drying since 2010. The IPO’s positive phase weakens the Walker circulation, favoring conditions that suppress Amazon convection and rainfall, consistent with the observed and modeled SST patterns and the strong statistical relationships. Accounting for the IPO markedly improves attribution of recent drying and narrows near-term projection uncertainty. Predicting the IPO’s phase transition therefore has outsized value for near-term hydroclimate risk assessment in Amazonia, informing ecosystem management and socioeconomic planning. However, robust projections further require broader assessment across models to gauge structural uncertainties beyond internal variability captured by a single model ensemble.

The study shows that a negative-to-positive IPO transition explains approximately 45% (40–49%) of the observed post-2010 Amazon decadal drought, far exceeding the ~12% contribution from external forcing. The IPO substantially modulates near-term Amazon rainfall projections, and removing its influence reduces projection uncertainty by ~38%, with additional reductions in inter-member spread metrics. Anticipating IPO phase transitions can significantly improve near-term projections and risk assessments for Amazon hydroclimate. Future work should extend constraints to multi-model large ensembles, refine mechanisms linking IPO to Amazon convection and moisture transport, and improve decadal prediction capabilities for the IPO to enhance actionable near-term climate information.

- Single-model dependence: Results rely on CESM2-LENS; structural model uncertainties are not assessed. Authors note the need for multi-model large ensemble assessments to reduce single-model limitations. - Linear regression framework: IPO influence is estimated via linear relationships; potential nonlinearities or state dependencies may not be fully captured. - Assumed IPO predictability: Scenario adjustments presuppose accurately predicting IPO phase transitions; real-world predictability may be limited. - Scenario specificity: Near-term projections are evaluated under SSP3-7.0; different forcing pathways may alter forced trends and interactions with internal variability. - Regional focus and index definitions: Results depend on chosen Amazon and IPO index domains and the 9-year smoothing, which may influence quantitative estimates.