Intensification of heatwaves in China in recent decades: Roles of climate modes

The study investigates how strongly large-scale climate modes modulate Chinese heatwaves under a warming climate. While China has undergone pronounced warming, urbanization, and more frequent/hotter summers, the relative contributions of internal climate variability (ENSO, AMO, IOD) versus anthropogenic warming to abrupt intensification of heatwaves remain unclear. Prior work suggested a country-mean abrupt increase around 1996–1997 but lacked a regional assessment and mechanistic attribution. The authors aim to (1) assess regional changes and abrupt shifts in heatwave intensity across seven homogeneous regions in China, (2) quantify nonstationary impacts of ENSO, AMO, and IOD using a Bayesian dynamic linear model, and (3) elucidate physical mechanisms using reanalysis and AGCM experiments, with implications for decadal prediction.

Existing literature documents growing heat extremes globally and in China under anthropogenic warming, with impacts on health, agriculture, and water. Climate modes (ENSO, AMO, IOD) modulate heatwaves and hydroclimate across regions worldwide and in parts of China. A recent study reported an abrupt rise in China’s heatwave magnitude around 1996–1997, attributing it to increased global temperatures, but lacked a systematic regional breakdown and quantification of each climate mode’s role and mechanisms. The dynamics linking ENSO/AMO to East Asian climate via Gill-type responses and Rossby waves are known, but their decadal modulation and concurrent phase-transition effects on Chinese heatwaves needed assessment.

Data and indices: The study uses daily maximum and minimum temperatures from 718 CMA stations (1961–2017). Climate mode indices (JJA means with climatology removed) include Niño3.4 (ENSO), AMO, and DMI (IOD). The IPO index is used to represent ENSO decadal variability; IPO and Niño3.4 are highly correlated. ERA5 and NCEP/NCAR reanalyses provide additional fields (air temperature, SLP, 500 hPa height, winds, vertical velocity, clouds, heat fluxes, radiation). Heatwave metric (EHF): Heatwave intensity is quantified by the Excess Heat Factor (EHF), which combines (1) EHI_sig = 3-day mean of daily mean temperature minus the 95th percentile daily mean (computed with a moving 15-day window), and (2) EHI_acci = 3-day mean minus prior 30-day mean. EHF = EHI_sig × max(1, EHI_acci), incorporating both day/night conditions and antecedent heat. Regionalization and change-point detection: Fuzzy C-Means clustering with validity indices (FPI, PI, XB) determines seven homogeneous regions (three East, two Middle, two West). Regional mean EHF (JJA) time series are tested for abrupt changes using Pettitt and moving t-tests (10-year window), identifying region-specific change years spanning 1993–2000. Bayesian Dynamic Linear Model (DLM): After detrending heatwave and climate indices (AMO trend estimated from 1920–2017 to capture its multidecadal cycle), a Bayesian DLM with time-varying coefficients models EHF as a function of ENSO, AMO, and IOD. To mitigate ENSO–IOD dependence (r=0.42), a partial DMI is computed by removing ENSO’s effect; a partial DLM also models residual EHF after removing ENSO and AMO as a function of partial DMI. The DLM captures nonstationary, time-varying impacts and is compared to conventional linear regression. Reanalysis diagnostics: ERA5 composites (post- vs pre-abrupt) of SAT, SLP, 500 hPa height, clouds, radiative and turbulent fluxes, vertical motion, and temperature advection diagnose mechanisms of surface warming and atmospheric circulation changes. AGCM experiments: ECHAM4.6 is forced with idealized SST anomalies representing IPO, AMO (interdecadal; 40-year sinusoidal modulation of observed regression patterns), and ENSO (interannual; idealized 2-year sinusoid). Experiments: CTRL (climatological SST), Exp-IPO, Exp-AMO, Exp-(IPO+AMO), and Exp-(IPO+AMO+ENSO). Each sensitivity run spans 42 years (first 2 years spin-up discarded); three-member ensembles for sensitivity runs. Impacts are quantified by differences from CTRL and by comparing last 20 years minus first 20 years to emulate pre/post-abrupt periods. Linear superposition versus combined forcing assesses nonlinearity.

- Abrupt intensification: Robust abrupt increases in JJA EHF occurred across China during 1993–2000. Northern and western regions (East I, Middle I, West I, West II) show the strongest increases; East II also shows significant changes, while East III and Middle II are less consistent across tests. - Magnitude changes: Post-abrupt mean EHF is 4.8× (East I), 6.7× (Middle I), 4.8× (West I), and 6.1× (West II) relative to pre-abrupt; EHF variability (STD) also increases markedly, indicating more intense individual events. - Trend removal: After removing linear trends, abrupt EHF increases remain evident in northern and western regions, implying anthropogenic warming alone cannot explain the abrupt intensification. - DLM performance: Bayesian DLM reproduces observed interannual to decadal EHF variability and abrupt increases with high correlations (r ≈ 0.95–0.98 between observations and DLM fits across regions). Relative to conventional linear regression, DLM improves hindcast skill by >10% and reduces mean square error by >25%. - Climate modes’ contributions: During the post-abrupt period, combined ENSO+AMO+IOD explain 62.35% (East I), 70.01% (Middle I), 66.20% (West I), and 63.64% (West II) of observed EHF intensification (STD_modes/STD_EHF). Within the DLM simulations, these modes account for 85.56%, 89.33%, 86.72%, and 88.11% (STD_modes/STD_DLM) respectively. ENSO provides the largest contribution in East I, Middle I, and West I; in West II, ENSO, IOD, and AMO have comparable impacts. - Nonstationary coefficients: Despite similar amplitudes of ENSO/AMO/IOD indices pre/post, their DLM coefficients intensify after the mid-1990s, making the same-strength mode anomalies produce larger EHF responses. ENSO coefficients become strongly negative post-abrupt, indicating La Niña (negative IPO) enhances heatwaves. AMO shows time-varying impacts (negative before ~2008, positive thereafter), consistent with changing SSTA patterns. IOD impacts strengthen post-abrupt; in West II, IOD coefficient changes sign around ~2007, with uncertain cause. - Spatial composites: La Niña and negative IPO phases, and positive AMO/IOD phases, are associated with enhanced EHF over China, strongest in the north. - Mechanisms: Post-abrupt period features enhanced Eurasian high (higher SLP and 500 hPa height), leading to subsidence, reduced cloud cover, increased downward shortwave radiation, reduced latent heat loss, increased sensible heat flux, and regional horizontal advection, collectively warming surface air and intensifying heatwaves. - AGCM support and nonlinearity: Exp-IPO enhances EHF in central, northeastern, and western China; Exp-AMO weakly enhances southern China. Combined Exp-(IPO+AMO) yields strong EHF increases over northern China, exceeding linear superposition, indicating nonlinear interactions. Exp-(IPO+AMO+ENSO) shows same-amplitude La Niña events produce stronger heatwaves post-abrupt in East I and West II, evidencing interdecadal modulation of ENSO’s impact. - Synthesis: The concurrent mid-1990s transition to negative IPO and positive AMO phases intensified Eurasian high-pressure and amplified atmospheric internal variability and climate modes’ impacts, explaining a major portion of the abrupt EHF increase, atop ongoing anthropogenic warming.

The study demonstrates that climate modes, particularly ENSO and its decadal variability (IPO), substantially modulate Chinese heatwaves in a nonstationary manner. The abrupt mid-1990s intensification in northern and western China is linked to strengthened atmospheric circulation anomalies (Eurasian high) that enhance surface warming via subsidence, radiative, and turbulent flux changes. The Bayesian DLM reveals intensified sensitivity of EHF to climate modes post-1990s rather than stronger mode amplitudes per se. AGCM experiments confirm that the concurrent decadal phase transitions of IPO (to negative) and AMO (to positive) can both intensify heatwaves directly and amplify ENSO’s interannual impacts, with nonlinear interactions being important. These findings address the research question by quantifying the extent to which climate modes, against a warming background, regulate regional heatwave intensity and by elucidating the dynamical mechanisms, emphasizing the need to account for decadal variability in prediction and risk management.

The paper provides a comprehensive regional assessment of China’s heatwave intensification, identifying an abrupt increase during 1993–2000 that persists after removing long-term warming. It quantifies that combined ENSO, AMO, and IOD impacts explain a majority of the observed intensification in northern and western regions and shows that post-1990s the climate system became more sensitive to these modes. Mechanistically, concurrent IPO and AMO phase transitions intensified the Eurasian high and warmed surface air, thereby amplifying heatwaves and ENSO’s effects; nonlinear interactions matter. These insights can improve decadal prediction and near-term projection of heatwaves using models that incorporate nonstationary climate mode impacts (e.g., Bayesian DLM). Future research should investigate sources and dynamics of atmospheric internal variability, regional nonlinearity, potential changes in climate mode characteristics under continued greenhouse warming, and enhance observations (e.g., over the Tibetan Plateau) to reduce uncertainties.

- Spatial data limitations: Sparse meteorological stations over the Tibetan Plateau (West II) may affect robustness; NCEP/NCAR reanalysis underestimates EHF changes outside northeast China compared to ERA5. - Model assumptions: The DLM assumes predictor independence; removing ENSO’s effect from DMI may overestimate ENSO and underestimate IOD impacts, though tested via partial DLM. Time-varying coefficients capture nonstationarity but can still be influenced by unmodeled processes. - Attribution uncertainties: The sign change of IOD coefficients in West II after ~2007 remains unexplained and may reflect internal variability or data sparsity. - AGCM idealizations and errors: Forcing uses idealized sinusoidal SSTAs (40-year IPO/AMO; 2-year ENSO), which simplify real-world variability and diversity; model structural errors persist. Conclusions rely on consistency across idealized experiments and observations rather than perfect realism.