The growing risk of climate tipping events underscores the urgent need for reliable early warning signals (EWS). Here we synthesize recent advances in EWS research for climate systems, systematically categorizing available methods into three groups, namely statistical indicators, nonlinear dynamical diagnostics, and data-driven approaches, and evaluating their theoretical applicability to practical limitations. In conclusion, statistical indicators offer theoretical clarity but are sensitive to non-stationary noise, nonlinear dynamical methods provide enhanced robustness in high-dimensional settings, and network- and deep learning-based approaches show the potential for detecting diverse tipping types, particularly when integrated with physical constraints and multi-modal observations. Given these trade-offs, we advocate for three practical strategies to enhance EWS utility, including adopting ensemble frameworks that combine multiple indicators to reduce false alarms, advancing cascade EWS to capture cross-system tipping interactions, and shifting from qualitative trend warnings toward quantitative, threshold-defined predictions with quantifiable uncertainties. We also discuss supporting techniques for these strategies, including adaptive noise filtering, dimensionality reduction, satellite data integration, and physics-informed machine learning. This review provides a structured reference for method selection across climate subsystems and data regimes, and underscores that operational EWS will require sustained integration of nonlinear dynamics, Earth observation, and data science.
Firn refreezing over the Greenland Ice Sheet (GrIS) buffers surface mass loss by retaining meltwater. However, the spatial heterogeneity of long-term firn refreezing variability and its relationship with extreme melt events remain largely unclear. In this study, the Community Firn Model (CFM) is used to investigate refreezing processes at 13 firn sites across the GrIS, forced by automatic weather station (AWS) observations (2009‒2024) and the regional climate model MAR (1940‒2024). Taking the southwestern percolation-zone site KAN_U as an example, multiple parameterizations are evaluated by comparing simulated firn density and temperature with in-situ observations, and the Brils2022 (BRI) scheme is selected for subsequent simulations. The results show that the mean annual refreezing at KAN_U during 2009‒2024 is 207 mm w.e. (45.3% refreezing ratio on average). The CFM-BRI simulations show 21%‒24% lower mean refreezing ratios relative to MAR. Long-term simulations for 1940‒2024 show relatively stable refreezing capacity at northern sites, but a 30% decline in refreezing ratio at the southwestern percolation zone (KAN_U) since the 1990s. Extreme melt events have become more frequent at KAN_U (five events during 2010‒2024), extending firn recovery from one cold season to 2‒5 years and increasing runoff, which peaked at 1160 mm w.e. in 2012. The frequency of extreme melt events is negatively correlated with the refreezing rates (r = −0.72, p < 0.001), indicating a delayed refreezing response to extreme melting. These findings improve our understanding of the long-term variability of refreezing processes in the GrIS firn zone and its response to increasing extreme melt events.
Vegetation resistance plays a vital role in sustaining ecosystem stability, but it is increasingly threatened by the rising frequency of compound drought and heat events (CDHEs). Despite the growing prevalence of these stressors, how vegetation resistance is changing and what mechanisms underlie these changes remain unclear, limiting our capacity to project future carbon sinks. This study examines the responses of leaf area index (LAI) and solar-induced chlorophyll fluorescence (SIF) to CDHEs. Using a geographic detector (GD) model, we explore underlying factors governing the spatial distribution of vegetation resistance changes globally during 2001–2020. The results show that CDHEs occur every 2.92 years on average and approximately 80% of vegetation response times (RTs) fall within a 2–4 mon window. Northern ecosystems generally exhibit longer RTs than southern ones. The spatial patterns of vegetation resistance, assessed through both LAI and SIF, are broadly consistent. However, SIF-based assessments indicate low resistance across a larger number of regions. Over 2001–2020, 52.1% of areas show a decrease in LAI-based resistance, and 47.1% show a similar trend for SIF-based resistance. Some regions, such as central North America and southern South America demonstrate a notable decline in resistance, whereas western Australia and central China show an increase in resistance. GD analysis identifies key climate factors, such as vapor pressure deficit, temperature, and surface net solar radiation, as the primary drivers of vegetation resistance changes in most hotspot regions. These factors interact through nonlinear effects, further influencing vegetation resistance. Overall, this study offers a scientific foundation for improving ecosystem conservation and enhancing regional ecological resistance.
Concurrent compound heatwaves (CCHWs) pose an escalating threat to global socioeconomic stability in a warming climate, yet current dynamical models remain inadequate for seasonal-scale forecasting. This predictive shortfall hampers effective early warning efforts. Here, we addressed this deficit by developing two machine learning (ML) models capable of forecasting summer CCHWs one season in advance. The ML models deliver a substantial leap in performance, near doubling the predictive skill of NCEP CFSv2 dynamical forecasts by raising the correlation score from 0.46 (and 0.35 for the conventional linear method) to 0.80. In addition to the well-documented Arctic influence, our ML-based attribution identifies sea surface temperature anomalies in the Tropical Atlantic and Southern Indian Ocean as critical precursors. Together, these three predictors dominate the forecast signal and collectively account for 76% of the interannual variance in summer CCHWs observed between 2012 and 2021. The integrated three precursors maintain a statistically significant correlation with summer CCHWs up to nine months in advance. Physically, the combined effects of these three factors are associated with a wavenumber-6 circumglobal quasi-stationary wave pattern, potentially emerging from interconnected mechanisms that involve weakened meridional temperature gradients and enhanced Rossby wave flux activity. These results establish a valuable pathway for advancing early warning systems, enhancing regional resilience to compound extremes in a changing climate.
Upper-air wind speed (UWS) is a critical indicator of atmospheric circulation and regional climate change, yet its long-term behavior has not been sufficiently examined using multiple global reanalysis products. In this study, using radiosonde observations, we evaluate the performance of five state-of-the-art reanalyses—ERA5, MERRA-2, JRA-3Q, CRA-40, and NCEP-2—in representing the UWS climatology, trends, and variability over China. All five reanalyses accurately reproduce the spatial and seasonal patterns of the UWS climatology, trends, and the first empirical orthogonal function modes (EOF1), with spatial correlation coefficients range from 0.676 to 0.895 for UWS trends and from 0.925 to 0.968 for EOF1 spatial patterns. Cross-data comparisons reveal that ERA5 performs best in capturing the trends and first modes during winter and spring, JRA-3Q excels in summer, and MERRA-2 leads in autumn. For vertical profiles, ERA5 consistently outperforms the other datasets in representing the UWS trends across all seasons, with seasonal correlation coefficients ranging from 0.659 to 0.844. The skill of these reanalyses in reproducing the UWS trends is closely linked to their ability to represent prevailing seasonal wind regimes: in winter, when westerly jet streams dominate, the reanalyses most accurately capture the zonal wind trends; in summer, when meridional flows govern, they perform best in capturing the meridional wind trends. Consequently, UWS trends are better represented in winter and summer than in spring and autumn. Despite their overall high skills in capturing nationwide spatial patterns, all reanalyses systematically underestimate the magnitudes of UWS and its trends at regional scales. This study not only supports informed data selection for future UWS-related research but also establishes a solid scientific basis for further investigations of UWS dynamics over China and offers valuable insights for improving numerical models and future reanalysis products.
Within the methodology framework of the Intergovernmental Panel on Climate Change on national emission inventories, process-oriented modelling is referred to as Tier 3 approach. This framework covers the climatically/environmentally important nitrogen (N) gases, including ammonia (NH3), nitrogen oxides (NOx) and nitrous oxide (N2O), from managed soils. Dynamical inventories of these gases with fine resolutions are urgently needed to well elucidate the responses of coupled air pollution and climate change to human activities, e.g., heavy use of N fertilizers. Here we report a case study dynamically and synchronically inventorying the national direct emissions of NH3, NOx and N2O from croplands of China mainland, using the CNMM-DNDC which is a high-resolution, process-oriented hydro-biogeochemical model, to meet Tier 3 needs. The model performed robustly in validation against observations at 17 field sites across diverse climatic zones, showing normalized root mean square errors of 46%, 23% and 31% and Nash-Sutcliffe Index values of 0.73, 0.97 and 0.95 for the three gases, respectively. Summing up the 3-hour and 30-second simulations, the national annual direct emissions due to application of synthetic nitrogen fertilizers in 2015 were estimated at 4.66 ± 1.49 (NH3), 0.152 ± 0.041 (NOx) and 0.188 ± 0.047 (N2O) in Tg N (1 Tg = 1012 g). Logically, the simulated NH3 emissions hot-spotted in the North China Plain and peaked in May and October while NOx and N2O emissions in subtropical provinces such as Hunan and Hubei in June and October. The background emissions from croplands were equivalent to about 4%, 19% and 25% of the national direct NH3, NOx and N2O emissions, respectively. The model simulations also resulted in aggregated annual direct emission factors at logical levels for the national widely applied synthetic N fertilizers in 2015, which were on average 15.67% for NH3, 0.56% for NOx and 0.84% for N2O. This study implicates that the applied model not only acts as a reliable and robust Tier 3 approach to generate dynamic national direct emission inventories of the three N gases with fine resolutions, but also possesses the profoundly distinguished capacity in predicting direct emission factors of the gaseous species for complex/extensive conditions where field observation is impracticable.
Glacier surface albedo regulates shortwave radiation absorption and mass balance, yet its long-term and spatial variability remains poorly quantified for maritime glaciers in the southeastern Tibetan Plateau (SETP). Using MODIS Terra and Aqua products, we reconstructed daily albedo for 2003–2024 and examined annual mean, melt-season mean, and melt-season minimum albedo in relation to topography, debris cover, snowfall, and mass balance. All metrics declined significantly (p < 0.05), with Theil–Sen slopes of −0.0015, −0.0016, and −0.0026 per year, respectively. Declines occurred in all six subregions but varied in magnitude and significance, possibly reflecting regional gradients in monsoon moisture transport and snowfall replenishment. Monthly declines were strongest in August and remained significant in November and December (p < 0.05). Melt-season duration increased by approximately 6 d per decade (p < 0.05), mainly through delayed termination, extending glacier darkening into the early-winter snow-recovery period. Albedo increased with elevation by 0.106 km (p < 0.01), and annual mean albedo was higher in accumulation than ablation zones (p < 0.01). Debris cover was negatively associated with all metrics (p < 0.01). Despite greater potential solar radiation, south-facing slopes had higher albedo than north-facing slopes; their greater precipitation and snowfall suggest that fresh-snow replenishment may partly offset radiative forcing. Melt-season albedo was generally positively but weakly associated with annual mass balance, with strong subregional variation. Overall, SETP albedo variability reflects surface darkening and monsoonal snowfall replenishment, highlighting the need to consider both ablation and accumulation when assessing glacier responses to warming.
Winter mean and extreme temperature indices obscure the cumulative sub-zero temperature deficit relevant to freeze–thaw processes. However, systematic assessment of the variability and drivers of these processes remains lacking. In this study, using station observations, reanalysis fields and CMIP6 multi-model simulations, we assess the spatiotemporal variability, atmospheric drivers and future trajectory of the winter accumulated negative temperature (WANT), defined as the cumulative daily mean temperature below 0 °C, over Northeast China during 1960–2020. The WANT increases significantly at a rate of 32.86 °C d per decade (p < 0.01), with the most pronounced increases at low elevations. The winter mean temperature increases at 0.38 °C per decade (p < 0.01), whereas the winter mean sub-zero daily temperature increases at 0.33 °C per decade (p < 0.01), indicating that winter mean warming may lead to overestimation of the decline in below-freezing temperature conditions. The long-term variation of the WANT shows no systematic northward increase, whereas the interannual anomaly patterns during the cold and warm periods differ in both magnitude and latitudinal organisation. Beyond sea ice alone, cold and warm WANT periods are individually associated with distinct combinations of autumn snow cover, sea ice and sea surface temperature anomalies that jointly precondition contrasting winter waveguide structures and downstream wave energy convergence over Northeast China. The CMIP6 ensemble-mean trends reproduce the observed variations in the WANT, with less than 5% error. Using SSP5-8.5, the WANT is projected to increase by approximately 600 °C d by 2100, exceeding the full range of historical interannual variability, whereas SSP1-2.6 indicates a change limited to approximately 110 °C d. These results underscore the importance of incorporating WANT into climate assessments for cold regions and adaptation strategies for freeze–thaw sensitive ecosystems and infrastructure.
The China‒Nepal Himalayas (CNH) are threatened by glacial lake outburst floods (GLOFs), which endanger trade and local communities. Currently, glacial lake risk assessments in the CNH lack a reliable, quantitative, and cross-regional comparative approach. To fill this gap, this study presents a detailed and refined GLOF risk assessment using a data-driven machine learning framework. We first mapped lake boundaries in 1992, 2000, 2009, and 2022 using Landsat imagery and analyzed their distribution and temporal changes. We then identified potentially dangerous glacial lakes (PDGLs) among lakes that contact glaciers and are at least 0.1 km2 in size. A machine learning model trained on High Mountain Asia estimated the likelihood that PDGLs would trigger GLOFs using key factors. We also evaluated impacts on exposed elements through stochastic inundation modeling. The final risk level was determined by combining hazard assessment with downstream impact analysis. Results show that 2377 glacial lakes were identified, mostly at elevations between 4100 m and 5900 m, with a 27% increase in area from 1992 to 2022. The trained model accurately identified all historical GLOFs in the CNH, confirming its high reliability. Downstream exposed elements, including buildings, bridges, roads, and hydropower facilities, remain at risk. Of the 76 lakes identified as potentially hazardous, four are categorized as having a very high risk and 14 as having a high risk. This study provides data to help stakeholders and policymakers pinpoint high-risk lakes and develop effective mitigation strategies.
Interdecadal variability in China’s summer rainfall is commonly attributed to ocean forcing, but the role of upstream land heating across the Tibetan, Iranian, and Mongolian Plateaus remains unclear. Here we refer to these regions collectively as the Asian plateaus and show that sensible heating anomalies over the Asian plateaus form a coordinated upstream land-forcing system on interdecadal timescales. During 1979–2018, summer sensible heat weakened over the Tibetan Plateau but strengthened over the Iranian and Mongolian Plateaus, forming a tripolar mode that shifted in the late 1990s and was tightly linked to a rainfall tripole over China. This thermal contrast was associated with reduced lower-tropospheric thickness over the Tibetan Plateau, enhanced thickness over the Iranian and Mongolian Plateaus, a deep anticyclonic anomaly over Mongolia–Northeast Asia, upper-level jet and Rossby wave source adjustments, and downstream moisture divergence and convergence anomalies over China. A structural equation model further showed that plateau sensible heating exerted a total effect (0.52) on summer rainfall variability that was slightly larger than that (0.42) of the major sea surface temperature modes considered here. These results identify surface heating over the Asian plateaus as a major and previously underappreciated driver of China’s interdecadal hydroclimate, advancing interpretation beyond an ocean-centered framework and highlighting the need to consider land–atmosphere and ocean–atmosphere forcing together for understanding and predicting long-term rainfall variability over China.
Climate change poses substantial risks to ski tourism globally, making it increasingly important to identify where ski resorts can remain suitable. Across China, however, suitability has not been comprehensively quantified at the national scale. Here, we constructed a comprehensive indicator framework encompassing natural and socioeconomic dimensions and integrated machine learning with multi-model climate projections to assess ski resort suitability across China under current and future warming scenarios. Currently, the total suitable area (Suitability Index ≥ 0.3) is approximately 1.98 × 106 km2, concentrated in Northeast and North China, with winter temperature and population identified as the dominant contributing factors. Under SSP126 and SSP245, the suitable area increases slightly in 2021‒2050, then decreases by 9.36% and 9.24%, respectively, in 2051‒2070. Under SSP585, it remains relatively stable in 2021‒2050 but decreases sharply by 19.90% in 2051‒2070, with Southern China exhibiting the largest decline (55.71%), followed by the Northeast (21.23%), North (18.68%), and Northwest (10.43%). The suitability centroid shifts northwestward from North China toward higher elevations and latitudes. Adaptation requires region-specific strategies. Northeast and Northwest should prioritize expanding the scale and capacity of high-grade ski resorts and improving their service quality, while North China requires relocation to higher elevations, advanced snowmaking technology effective above −2 °C, and optimization of resource utilization. Southern regions should transition to indoor ski tourism supplemented by outdoor activities. This framework offers a transferable approach for assessing the impacts of warming on climate-sensitive industries.
Crop migration is widely regarded as a critical adaptation strategy to mitigate climate-induced production losses; however, its quantitative efficacy and inherent limitations in offsetting systemic production declines remain poorly quantified. To address this gap, we developed an integrated climate change impact assessment framework—MaxEnt-SPAM-EPIC—that couples spatially explicit crop distribution modeling (MaxEnt-SPAM) with biophysically grounded growth simulation (EPIC) as parallel, interoperable submodules for jointly quantifying future crop migration and yield. Applied to the North China Plain (NCP), a globally notably grain-producing region, the framework is rigorously validated against historical (1995–2014) winter wheat distribution and yield data before projecting dynamics under SSP126, SSP245, and SSP585 across near-, mid-, and long-term periods. Results reveal a robust northeastward shift in the spatial centroid of winter wheat cultivation (4.57–6.66 km), driven by internal redistribution of planting intensity: high- and extremely high-density zones expand by 1.89% on average, while medium- and low-density zones contract, yielding a marginal net increase in cultivated area (1.61%–2.03%). Critically, despite this migration, total winter wheat production declines by 3.60%–8.01% relative to baseline, with losses escalating under higher-emission scenarios. Yield-related losses account for 73.59%–80.43% of production variance and drive absolute declines of 5.53%–10.23%; in contrast, migration contributes only 1.95%–2.14% to promote production, offsetting just 17.86%–27.65% of climate-induced production losses. The interaction between climate change and migration explains <2% of variance, confirming its negligible role. These findings demonstrate that crop migration provides only a partial, non-sustaining buffer: it cannot reverse the systemic downward trend in NCP winter wheat output. Consequently, migration alone is insufficient as a long-term adaptation strategy. This study provides key insight into the practical effectiveness of crop migration as a mitigation strategy, guiding policymakers in developing comprehensive region-specific adaptation measures to ensure future food security.
To bridge the gap between large-scale climate projections and local adaptation, it is critical to quantify how future extreme sea levels (ESLs) are disproportionately amplified across mainland deltas and offshore islands. Utilizing a high-resolution hydrodynamic framework driven by over 1300 synthetic tropical cyclones, we project 100-year ESLs and coastal vulnerability for the northern South China Sea (NSCS) under scenarios SSP2-4.5 and SSP5-8.5. Our findings reveal that the amplification of ESLs is strongly heterogeneous across three sub-regions—East, Central, and West—rather than uniformly across the basin. Compared to SSP2-4.5, SSP5-8.5 drives the greatest elevation of 100-year ESLs over shallow, semi-enclosed shelves and mainland deltas, with the western and eastern sub-regions experiencing average amplifications of 35% and 30%, respectively. In contrast, deep-water offshore islands only increase 18%. However, hazard frequencies intensify drastically across all regions: the 100-year ESLs under SSP2-4.5 shortens to a 40-year return period across large mainland coastal sectors under SSP5-8.5, and plummets to a 20-year event for low-lying islands. Vulnerability also escalates unevenly, with high-risk zones expanding most in the West, where high-risk zones nearly double in spatial extent (from 3.4% to 6.7%). Meanwhile, low-lying islands exhibit disproportionately high vulnerability despite lower absolute ESLs. These findings underscore the urgent need for targeted coastal adaptation strategies that address the distinct vulnerabilities of mainland deltas and offshore islands.
The narrative of glaciers as a water resource has led to the perception that glaciers are critical to basin runoff, especially in arid regions. Substantial discrepancies exist among quantifying the contribution of glacier meltwater to basin runoff for the same basin. To address this knowledge gap, multiple precipitation datasets were evaluated to identify the most suitable one, based on which a threshold-constrained framework was established to evaluate glacier meltwater contributions across basins on the northern slope of the Qilian Mountains (QM). The stability and reliability of the framework were verified through sensitivity analyses. The results demonstrate that the contribution of glaciers to annual runoff did not exceed 2.6%, 1.4%, 10.3%, 22.9% and 17.9% in the upper reaches of the Shiyanghe (URSYH), Heihe (URHH), Beidahe (URBDH), Shulehe (URSLH) and Danghe (URDH) River basins, respectively. However, previous studies have estimated their contributions to be 4.1%–6.1%, 2.7%–3.5%, 12.1%–23.1%, 23.3%–51.2% and 46.8%–47.8%, respectively, all higher than the upper threshold derived in this study, suggesting that the contribution of glacier meltwater to runoff at the mountain outlets on the northern slope of the QM may have been overestimated. It is observed that annual precipitation in glacierised areas of the URSYH, URHH, URBDH, URSLH and URDH River basins was remarkably higher than that in non-glacierised areas. However, since glacierised areas occupy only a small fraction of each basin (0.32%–4.37%), the contribution of precipitation in glacierised areas to total basin precipitation is correspondingly limited (0.34%–5.27%). This study provides important guidance for water resource management in the Hexi Corridor and offers a reference framework for glacier meltwater contribution evaluations in other glacierised regions worldwide.
Quantifying changes in the vegetation carbon sinks under drought stress is crucial for advancing carbon neutrality goals under climate change. However, knowledge of these changes during the period from drought onset to subsequent sustained carbon loss remains limited. In this study, a comprehensive drought index was constructed for the global scale over the period 2001–2024, and the spatiotemporal patterns of drought events were quantified by applying run theory to identify continuous drought episodes. On this basis, drought-induced carbon losses and their variation in vegetation were calculated, and the probability of vegetation carbon loss under different drought scenarios was further quantified. The results indicated a general growth trend in global drought in 2001–2024, with pronounced increases observed in regions such as South America, Eurasian mid- to high-latitude regions and central Africa. Notably, drought duration decreased from arid to humid climate zones. The spatiotemporal heterogeneity of drought leads to regional differences in global vegetation carbon sink losses. A decline in global carbon sink capacity was observed across 72.90% of the study areas, with the Eurasian mid- to high-latitude regions experiencing the most severe drought impact on net ecosystem exchange, reaching a maximum loss of 2.01 gC/(m2·d). However, drought was not universally harmful; in coastal areas of eastern North America, vegetation carbon sink capacity even strengthened. Vegetation type and climate play important roles in determining the differential responses of vegetation to drought. For example, deciduous broadleaf forests in humid regions tend to lose more carbon than other vegetation types, reaching a mean loss of 1.04 gC/(m2·d). Vegetation carbon loss probability increases with increasing drought grade; under extreme drought, more than one-fifth of the dry sub-humid area faces a loss probability exceeding 0.6.
Climate adaptation increasingly requires knowledge systems that are place-based, practice-ready and legitimate to the communities most exposed to risk. This review synthesises how Traditional Knowledge (TK), Indigenous Knowledge (IK) and Local Knowledge (LK) have been studied and mobilised in climate-change research from 2000 to 2025. Using a curated SCI/SSCI corpus of 1693 articles, we combine evidence mapping, keyword analysis, LDA topic modelling and harmonised country coding. Publication volume expanded sharply after 2015, indicating a shift from peripheral recognition of TK/IK/LK toward a mainstream adaptation evidence field. Topic modelling identifies five thematic clusters: biodiversity/forest and plant resource management, food-health-governance nexus and resilient systems, climate perceptions, weather variability and livelihood impacts, social-ecological management, coastal/marine systems and environmental change, and agricultural adaptation strategies, vulnerability and resilience. The evidence base is geographically concentrated, led by Canada (167), the United States (133), India (93), Australia (58) and China (47), but country profiles differ substantially across TK, IK and LK. Methods are dominated by qualitative, participatory and governance-oriented designs, while systematic reviews, modelling and scenario approaches are growing. Urban-related studies remain limited (169 articles; 9.98% of the corpus), with the highest share in LK research (12.4%). Overall, the field has shifted from observing climate impacts to governing and implementing adaptation. We argue that TK/IK/LK should be treated not as generic local information, but as plural evidence and adaptive capacity requiring clear classification, co-production, rights-aware safeguards, and evaluation metrics that can support climate-resilient development without undermining knowledge sovereignty.
The year of 2025 was the warmest year in China since 1951, with the summer witnessing widespread and intense heatwaves. It also marked the first quasi-operational deployment of the Beijing Climate Centre's (BCC) event attribution prototype. To document the results and identify pathways for future improvement, this study reviews the rapid attribution analyses conducted for the extreme high-temperature events of June, July, and August 2025. Observations reveal that multiple regions experienced record-breaking temperatures in June and July, with monthly mean anomalies exceeding 3 to 4 standard deviations above the 1961–1990 baseline. Attribution analyses consistently demonstrate a substantial anthropogenic influence on the occurrence probability of these events. Human activities increased the likelihood of record-breaking heat, with risk ratios exceeding a factor of 10 in most affected regions and reaching 64.5 (90% CI: 46.1–97.9) in Northwest China in June—corresponding to a probability shift from 0.09% under natural conditions to 5.62% under anthropogenic influence. While natural variability modulated the regional expression of heat extremes, the dominant role of human-induced climate change is clear. Future projections under the SSP2-4.5 medium emissions scenario indicate that such extreme summer heat will become substantially more frequent. By mid-century many regions will experience such heat every 1–2 years, compared with present-day return periods of approximately 10–20 years, effectively transitioning towards commonplace conditions. This consolidated assessment underscores the urgent imprint of climate change on contemporary extreme heat and the critical need for enhanced adaptation and mitigation strategies.
Western Arctic sea ice has experienced a prolonged melt season since the 1980s, but the oceanic processes that sustain delayed freeze-up remain unclear. In particular, the role of subsurface heat stored in the Near-Surface Temperature Maximum (NSTM) layer has not been well quantified on multi-year timescales. This study analyzes melt season duration in the Canada Basin over 1990–2023, focusing on NSTM evolution and quantifying the impact of NSTM heat release on freeze-up timing over 2005–2023. Results show that the melt season duration in the Canada Basin has lengthened by 15 d per decade from 1990 to 2023 (p < 0.01), mainly attributed to a delayed freeze-up. In the marginal ice zone, freeze-up is closely linked to the heat release from the NSTM layer. In certain years prior to 2013, subsurface heat was largely retained in a long-lasting NSTM. After 2013, the duration of the NSTM has shortened remarkably. The excessive subsurface heat was released into the mixed layer in autumn, leading to surface warming and delayed freezing of sea ice. The shortened NSTM retention time after 2013 was linked to three factors: reduced summer net heat flux (∼13%), deepened mixed layer (∼7.3 m), and weakened halocline stratification (∼41%). This NSTM heat-release compensation mechanism explains the continuous winter ice loss over the past decade despite a slowdown in summer, providing valuable theoretical support for improving sea ice forecast models.
