El Niño and tropical Atlantic warming can provide early- warning signals for extreme heat in the Amazon, with some precursors detectable up to seven months ahead, according to a new study.

Researchers from the ARC Centre of Excellence for 21st Century Weather (led by scientists at UNSW Sydney, with collaborators from the Australian National University, University of Melbourne, and Brazil’s National Institute of Space Research/INPE) published the findings in the journal Earth’s Future.

They analyzed monthly climate data from 1950- 2023 across six climatic sub- regions of the Amazon Basin, examining links between large- scale climate modes and temperature and rainfall extremes (assessed over three- month periods).

Key findings

• Hot extremes in the Amazon are more likely after unusually warm conditions linked to El Niño (positive ENSO phase in the tropical Pacific) and warming in parts of the tropical Atlantic. These relationships are stronger and more widespread for heat than for cold extremes or rainfall extremes.
• The clearest long- range signal is for March- May: extreme heat across much of the Amazon is associated with positive climate- mode anomalies observed as much as seven months earlier.
• Seasonal variations:
  • September- November: Hot extremes primarily linked to earlier tropical North Atlantic warming.
  • December- February: Extreme heat more often tied to recent tropical South Atlantic warming (across much of the basin), with longer- lead ENSO signals in southern and western areas (e.g., Planalto Amazon and Andes- Amazon foothills).

Compound hot- dry events

The team also assessed simultaneous extreme heat and rainfall deficits (compound hot- dry events), which raise wildfire risk, stress ecosystems, threaten health, food and water security, and can reduce the Amazon’s carbon uptake.

Using statistical (copula) methods focused on extremes plus machine learning to test predictability from lagged climate indices (ENSO, tropical North and South Atlantic SSTs, and North Atlantic Oscillation):

• Useful early- warning potential exists in some seasons and regions, especially northern and lower central Amazon.
• Predictability is dominated by ENSO, peaking in December- February (notably western Guiana highlands) and March- May (lower central Amazon). Skill is weakest in June- August and parts of the western Amazon.
• Reliable skill typically requires at least one climate index at short (e.g., 1- month) lead; the multi- month heat precursor signal does not translate into reliable seven- month forecasts of compound events.

Rainfall extremes show weaker, less widespread dependence on these large-scale ocean patterns than temperature does, partly because rainfall is more strongly influenced by regional weather, land- atmosphere feedback, and local conditions.

Context and caveats

El Niño alters tropical circulation in ways that can suppress Amazon rainfall and increase surface heating; tropical Atlantic warming can reinforce dryness and heat by shifting the Intertropical Convergence Zone. These mechanisms are well- known for seasonal means, but the study highlights their imprint on extremes, including surprisingly long lead times for heat in some seasons.

The results indicate potential predictability and identify promising regions and seasons for monitoring Pacific and Atlantic conditions. They do not constitute an operational forecast system.

Translating the findings into practical early- warning tools will require further work with dynamical climate models and additional predictors (e.g., soil moisture, vegetation state).

Lead author Dr Sanaa Hobeichi and co-author Associate Professor Andrea Taschetto (UNSW) noted that machine learning helped isolate the most informative signals while also showing where ocean indices alone are insufficient.

This research is relevant amid rising frequency and intensity of Amazon hot- dry extremes, which are expected to worsen with continued global warming and possible amplification by deforestation.

Comparison of Hobeichi et al. (2026) with related Amazon climate studies

Hobeichi et al. (2026, Earth’s Future) focuses on lagged statistical dependence (via copulas, emphasizing tail and extreme dependence) between major climate modes (primarily ENSO/Niño3.4, TNA, TSA, and NAO) and Amazon temperature and precipitation extremes, plus machine- learning assessment (Random Forest) of the predictability of compound hot-dry events from those lagged indices.

It divides the basin into six climatic sub-regions, uses 1950- 2023 observational data, and highlights a particularly long- lead (up to 7-month) precursor signal for March- May heat extremes, with more limited and shorter- lead skill for compounds (strongest in northern and lower- central regions in DJF/MAM; weakest in JJA).

This sits within a large body of work on Amazon hydroclimate teleconnections, extremes, and predictability. Below is a structured comparison across key dimensions.

1. Established teleconnections (ENSO and tropical Atlantic)

Classic and recent studies establish the physical basis that Hobeichi et al. build upon:

• ENSO: El Niño weakens the Walker circulation, promotes subsidence over northern/central Amazonia, suppresses convection/rainfall, and elevates temperatures. Major droughts (e.g., 1926, 1983, 1998, 2015- 16) are strongly linked to El Niño. La Niña tends to produce the opposite. Northern Amazonia is especially sensitive.
• Tropical Atlantic: Warm TNA (often lagged after El Niño via atmospheric bridge mechanisms) shifts the Intertropical Convergence Zone northward, delaying or weakening the wet- season onset (especially March- May in northeast Brazil and southern Amazonia) and reinforcing dryness and heat. The 2005 drought is a classic TNA- dominated example. TSA anomalies and the Atlantic Meridional Mode (AMM) also play roles. Southern Amazonia is often more Atlantic-influenced.
• Combined Pacific- Atlantic effects frequently amplify extremes; interactions with the Indian Ocean Dipole (IOD) appear in some multi-year drought analyses.

How Hobeichi et al. differ and advance: Most prior work uses linear correlations, composites, regressions, or wavelet methods focused on seasonal means or specific historical events. Hobeichi et al. explicitly target extremes via tail dependence (copulas), quantify lag dependence systematically (up to 7 months), and test predictive skill for compounds with ML. They confirm the expected physical mechanisms while showing temperature extremes have stronger and more widespread lagged links than precipitation extremes (which are noisier due to local processes).

2. Focus on extremes and compound hot-dry events

• Multiple studies documents increasing frequency, intensity, duration, and spatial extent of Amazon heatwaves, droughts, and compounds (especially since ~2000 or 2010). Extreme heat almost always co- occurs with dryness in the most severe cases. The 2023- 2024 period stands out as record-breaking in compound dry-hot conditions, low soil moisture, and high vapor-pressure deficit, affecting large fractions of the basin.
• Heatwaves are amplified under dry conditions via land- atmosphere feedback (reduced evaporative cooling, increased sensible heat). Synoptic patterns (e.g., enhanced South Atlantic subtropical high, reduced moisture influx) contribute.
• Projections (CMIP5 and CMIP6) robustly show rising compound hot- dry risk under warming, with Amazonia as a hotspot; deforestation can further intensify them via feedback.

Hobeichi contribution: Moves from documentation and attribution of past events and future projections toward statistical early- warning potential using remote climate modes alone. It shows compounds are harder to predict than heat alone, and skill is highly region- and season- dependent, consistent with known spatial heterogeneity (northern vs. southern Amazonia).

3. Predictability and early- warning efforts

• Some prior work has proposed Atlantic SST- based early- warning indicators for central Amazon droughts, with claimed lead times of many months (up to ~18 months in network- based approaches focused on specific ocean regions).
• Seasonal forecasting systems and empirical methods already use ENSO and Atlantic indices operationally, but skill for extremes (especially compounds) is limited and variable.
• Multi- mode analyses (ENSO, IOD and TNA) highlight coupled tropical interactions for multi- year droughts and note forecasting potential by monitoring oceanic indices.

Hobeichi contribution: Provides a more rigorous, basin-wide, lag- explicit quantification using modern tools (copulas for dependence and Random Forest for predictability).

It tempers expectations: the impressive 7- month heat signal does not equate to reliable long- lead compound forecasts; useful skill usually requires shorter leads (often concurrent or 1-month) and is strongest in specific region- season windows. NAO contributes little.

4. Methodological and regional novelty

• Sub- regional analysis is common (northern vs. southern, or specific sub-basins), but Hobeichi’s k-means clustering into six climatologically coherent regions is systematic.
• Lead author Hobeichi has prior expertise in ML applications to drought metrics, extremes predictability, and climate- mode explanations of precipitation variability (mostly Australian- focused previously), so the paper extends that toolkit to the Amazon.
• Related recent work explores two- way interactions (e.g., Amazon drought itself amplifying subsequent TNA warming via land- atmosphere- ocean feedback) and emerging “hyper tropical” conditions under continued warming.

Summary of positioning

Hobeichi et al. (2026) is evolutionary rather than revolutionary: it confirms well- established ENSO and tropical Atlantic teleconnections while advancing the quantitative characterization of extreme (especially heat) dependence at long lags and the practical limits of using those modes for compound- event early warning.

Strengths relative to much of the literature include the explicit extremes focus (copulas), ML predictability testing, long consistent observational record, and clear regional and seasonal differentiation. Limitations shared with many observational studies include reliance on large- scale indices alone (local land- surface and atmospheric processes matter, especially for rainfall and western Amazonia) and the challenge of translating statistical relationships into operational systems.

It complements event- based analyses (e.g., 2023- 24 compounds), process and mechanistic studies, and climate- projection work by highlighting where monitoring Pacific and Atlantic conditions offers the most actionable lead time for heat risk, and where additional predictors will be essential for compounds.

Climate-Mode Precursors and the Predictability of Amazon Hot and Dry Extremes

Authors and affiliations

Lead and corresponding author: Sanaa Hobeichi (ARC Centre of Excellence for the Weather of the 21st Century; Climate Change Research Centre & UNSW AI Institute, University of New South Wales).
Co-authors include Lincoln Muniz Alves (INPE, Brazil), Andréa S. Taschetto (UNSW/ARC Centre), Wil Laura (ANU), Nayan Talmale (University of Melbourne), and Lisa V. Alexander (UNSW). First published online 27 August 2026.

Abstract and Key findings

The study characterizes lagged dependence between large- scale climate modes and Amazon temperature and precipitation extremes, then evaluates whether monthly precursor signals enable early warning of compound hot- dry events.

• Methods: Copula- based analysis of tail dependence (focusing on extremes rather than average relationships) between climate indices and local anomalies at multiple lags. Machine- learning models (Random Forest classification) then test predictability of binary compound hot- dry months from lagged indices.
• Climate drivers examined: Niño3.4 (ENSO), Tropical North Atlantic (TNA), Tropical South Atlantic (TSA), and North Atlantic Oscillation (NAO).
• Data: CRU TS v4.08 monthly temperature and precipitation (1950- 2023) over the Amazon, divided into six climatic sub- regions via k -means clustering on seasonal precipitation and temperature climatologies. Indices from NOAA sources.

Main results:

• Hot extremes are more likely following positive ENSO phases and tropical Atlantic warming; these relationships are stronger and more widespread than for cold extremes.
• Strongest and most persistent signal: March- May (MAM) temperatures. Hot extremes are more likely across much of the Amazon after positive climate -mode anomalies at lags of up to 7 months → substantial long- lead early- warning potential for heat.
• September- November: Hot extremes mainly associated with prior TNA warming.
• December- February: Short- lag TSA warming across the basin; longer ENSO lags in the Planalto Amazon and Andes and Amazon foothills.
• Precipitation extremes show limited tail dependence with the climate indices. NAO generally shows weak dependence with individual hot or dry extremes.
• Compound hot- dry predictability: Dominated by ENSO. Highest skill in DJF (especially western Guiana highlands and northern regions) and MAM (lower central Amazon). Lowest skill in June-August. Highest success rates in specific region- season pairs (e.g., ~81% in one central region for MAM, ~79% in a northern region for DJF). Reliable predictions typically require at least one climate index at ~1-month lead.

Plain- language summary (from the paper)

Unusually high or low phases of major climate modes (especially ENSO and tropical Atlantic variability) months in advance are linked more strongly to extreme heat than to rainfall deficits. The clearest long- lead heat signal is in March- May (up to 7 months). Forecast skill for simultaneous heat and dryness peaks in DJF and MAM and is generally strongest when ENSO information is included, but useful forecasts usually need relatively short lead times.

Context and aims

Hot-dry compound events threaten Amazon ecosystems, carbon uptake, fire risk, health, and food and water security. They have become more frequent and intense and are projected to worsen with continued warming (and possible amplification by deforestation). While mean- state teleconnections of ENSO and Atlantic SSTs with Amazon climate are well known, the paper focuses specifically on extremes, lagged relationships, and quantitative early- warning potential using modern statistical and ML tools.

The paper emphasizes that the 7- month signal is a long- range precursor for heat extremes, not evidence of reliable multi- month forecasts of compound events. Further work with dynamical models and additional local predictors (soil moisture, vegetation, etc.) would be needed for operational systems.

Journal information: Earth’s Future (2026)

First published: 27 August 2026

DOI: 10.1029/2026ef008496

Provided: ARC Centre of Excellence for 21st Century Weather

Authors: Sanaa Hobeichi, Lincoln Muniz Alves, Andréa S. Taschetto, Wil Laura, Nayan Talmale, Lisa V. Alexander

Abstract

We characterize the dependence between lagged large-scale climate modes and Amazon climate extremes and assess whether monthly precursor signals enable early warning of compound hot-dry events. A copula-based framework analyses tail dependence between major climate drivers and precipitation and temperature anomalies at multiple lags, identifying where and how far in advance extreme phases of modes of variability likely precede local extremes. Machine learning models then evaluate the predictability of compound hot-dry events from lagged climate indices, determining which modes and lead times yield skillful forecasts. Copula modeling shows hot extremes are more likely following positive phases of El Niño-Southern Oscillation (ENSO) and tropical Atlantic variability, with relationships more prevalent than for cold extremes. The most persistent signal occurs for March–May temperatures, when hot extremes are more likely across the Amazon following positive climate-mode anomalies at lags up to 7 months, indicating substantial long-lead early-warning potential. In September–November, hot extremes are associated with prior tropical North Atlantic warming. During December–February, they occur at short lags with tropical South Atlantic warming across the basin, and at longer ENSO lags in the Planalto Amazon and Andes–Amazon foothills. In contrast, precipitation shows limited tail-dependence with climate indices, and North Atlantic variability exhibits weak dependence with individual hot and dry extremes. Predictability of compound hot-dry events is dominated by ENSO and peaks in December–February, especially in the western Guiana highlands, and in March–May in the lower central Amazon, with lowest skill in June–August. Reliable predictions typically require at least one climate index at 1-month lead.