Increases in the temperature seasonal cycle indicate long-term drying trends in Amazonia

The Amazon rainforest is a major global carbon pool and has historically acted as a carbon sink, but warming and drying associated with climate change and deforestation threaten a transition toward a carbon source and potential forest dieback. Recent observations show longer dry seasons, increased frequency of dry days and hot extremes, and reduced resilience of rainforest ecosystems. Given sparse direct measurements of evaporation and soil moisture across Amazonia, the study hypothesizes that the amplitude of the near-surface temperature seasonal cycle (difference between the warmest and coolest monthly means each year) can serve as an indicator of surface moisture availability. Reduced evaporative cooling during longer and more intense dry seasons should increase the temperature seasonal cycle amplitude and correspond to declines in evaporative fraction (EF), a dimensionless measure of the fraction of surface net radiation used for latent heat flux. The purpose is to quantify the relationship between temperature seasonal cycle amplitude and EF using reanalyses and CMIP6 Earth System Models (ESMs), and to use this relationship with observational temperature records to infer historical and projected Amazon drying.

Prior work suggested potential Amazon dieback under climate change, with debate over model dependence. CMIP6 models with dynamic vegetation show localized abrupt dieback driven by warming and drying under elevated CO2. Observations document increasing dry-season length in southern Amazonia, more dry days in September–November, longer fire seasons, and more hot extremes; trends are attributed to greenhouse gases and deforestation. Tropical tree growth is tightly linked to dry-season rainfall, and Amazon resilience has reportedly declined since the early 2000s. CMIP5 models showed increasing temperature variability associated with decreases in evaporative fraction, and other studies linked temperature variability to EF. A strong negative correlation was found in CMIP5 between changes in hot extremes relative to mean warming and EF over Amazonia, but CMIP5 exhibited large biases in temperature variability over Amazon. CMIP6 improves biases in seasonal temperature amplitude and evapotranspiration representation over Amazonia, motivating reassessment of EF–temperature relationships and their application to observations.

Data sources include ERA5 reanalysis (primary observationally constrained dataset) and additional reanalyses (NCEP-DOE R2, MERRA-2, JRA-55) for comparison; observational near-surface temperatures from HadCRUT5 for historical reconstruction; and CMIP6 ESM outputs. CMIP6 experiments span 1900–2099, combining historical (1900–2014) with SSP5-8.5; SSP2-4.5 and idealized 1% CO2 per year runs are also analyzed for robustness. All datasets are interpolated to a 1°×1° grid. The temperature seasonal cycle amplitude is defined as the within-year difference between maximum and minimum monthly mean near-surface temperatures. Evaporative fraction is EF = LE/(LE+H), where LE is latent heat flux and H is sensible heat flux. Annual anomalies for temperature seasonal cycle amplitude and EF are computed relative to 1979. For reanalyses and CMIP6, regional analyses use the IPCC AR6 subregions over Amazonia (NWS, NSA, NES, SAM). Linear regressions are fitted between annual EF anomalies and temperature seasonal cycle amplitude anomalies to quantify relationships. For CMIP6, historical and future data are concatenated to expand anomaly ranges, and regressions are computed across all years and models (one model per modeling center). Spatial maps of correlation are derived at grid-point level. To reconstruct EF from temperature observations, regional regressions from CMIP6 are applied to temperature seasonal cycle amplitude anomalies from HadCRUT5 and from CMIP6 ensemble mean. Time series are smoothed with a 10-year running mean. To reduce scenario dependence, reconstructed EF and CMIP6 EF are plotted against global mean warming relative to 1850–1900, using nearest-neighbor interpolation onto a global-warming axis from −1 to +4 K in 0.05 K steps; no extrapolation is performed and ensemble statistics are shown only when at least 10 models provide values.

- ERA5 (1979–2020) shows decreasing trends in EF (drying) accompanied by increasing temperature seasonal cycle amplitude in three of four Amazon subregions (NSA, NES, SAM); NWS shows little trend, likely due to coastal influence. The observed temperature seasonal cycle amplitude increased by about 0.4 °C over the last three decades. - Interannual anomalies in ERA5 reveal significant negative correlations between EF anomalies and temperature seasonal cycle amplitude anomalies: NSA r = −0.61 (slope implies ~0.02 EF decrease per +1 °C in amplitude); NWS and NES exhibit weaker correlations; SAM r negative as well. Across alternative reanalyses, NSA correlations are generally r < −0.45 (except JRA-55), with slopes ranging ~0.01 to 0.07 EF decrease per +1 °C. - CMIP6 models (historical + SSP5-8.5, 1900–2099) show strong negative EF–amplitude correlations in all regions, notably NSA r = −0.75. The CMIP6 NSA slope is approximately double the ERA5-derived slope, yet within the range across reanalyses. At grid level, CMIP6 correlations are spatially homogeneous and predominantly negative over the basin, while ERA5 shows more heterogeneity with weaker or positive correlations near coasts. - Model robustness: 23 of 25 CMIP6 models have correlations ≤ −0.5 in NSA (17 of 25 ≤ −0.7); 22 of 25 models project drying over NSA. The two models with weak correlations are those projecting greatest wetting, consistent with evaporation not being soil-moisture limited. - Reconstruction: Using regressions and temperature seasonal cycle amplitude, reconstructed EF from HadCRUT5 aligns well with CMIP6 ensemble EF historically, indicating a continuous EF decline since 1900. Under future warming, CMIP6 indicates continued drying; in NSA, reconstructed EF suggests an EF decrease of ~0.05 at 3 °C global warming. Across regions, EF may decrease by as much as ~5% at 2 °C global warming. - Overall, an increase of 1 °C in annual temperature seasonal cycle amplitude is associated with an EF reduction up to ~0.04 (range ~0.01–0.07 across reanalyses), linking amplified seasonal temperature contrast to reduced evaporative cooling and drying.

The study demonstrates that increased amplitude of the near-surface temperature seasonal cycle serves as a robust indicator of declining evaporative fraction and surface moisture availability in Amazonia. This addresses the challenge of sparse direct hydrological observations by leveraging widely available temperature data to infer drying trends. Consistent negative EF–temperature amplitude correlations across reanalyses and CMIP6 models, including in idealized and alternative scenario runs, indicate that longer and more intense dry seasons reduce evaporative cooling, thereby amplifying seasonal temperature contrasts. Agreement between reconstructed EF from observed temperatures and CMIP6 EF supports the method’s validity and indicates persistent drying since 1900. Model projections show continued amplification of the temperature seasonal cycle and further EF declines with global warming, highlighting elevated risk of Amazon drying and potential forest dieback. The spatial patterns, regional differences (weaker relationships near coasts and mountainous regions), and strong ensemble agreement underscore the reliability of the emergent relationship for large-scale assessments.

The paper establishes a strong, quantitative link between the amplitude of the temperature seasonal cycle and evaporative fraction over Amazonia. Observations and CMIP6 models jointly indicate that recent increases in seasonal temperature amplitude correspond to declining EF, evidencing ongoing drying. The approach enables reconstruction of historical EF using temperature records, revealing continuous drying since 1900, and suggests drying will intensify with future warming, with EF reductions on the order of a few percent per degree of global warming. This emergent relationship offers a practical diagnostic for monitoring and projecting Amazon moisture availability in the absence of dense flux networks.

- Observational sparsity: Weather and flux stations are sparse across Amazonia, particularly in the west, limiting reanalysis accuracy and contributing to heterogeneous correlations in ERA5. - Coastal and orographic influences: Near-coastal and Andean-adjacent regions are less soil-moisture limited, weakening or reversing EF–temperature amplitude correlations. - Model structural dependencies: Multiple CMIP6 models share components, raising the possibility of systematic biases. The study mitigates this by using one model per center, but residual shared biases may remain. - Scenario and sensitivity dependence: Time-based projections depend on emissions scenarios and differing model climate sensitivities, necessitating plotting against global warming levels to reduce dependence. - Regression linearity and generality: Relationships are derived from linear regressions at regional scales; applicability may vary at finer scales or under conditions of strong wetting where evaporation is not moisture-limited.