Earth and Environmental Sciences

Shooting in low light condition needs prolonged exposure time to take a high quality image by increasing the amount of incident light into a camera sensor. However, long exposure images contain motion blur by object movement. Since near infrared (NIR) images are more robust to low light condition without color information than visible (VIS) images, multi-sensor fusion of VIS and NIR images provides a viable solution to low light imaging for high quality photographs. In this paper, we propose motion deblurring of long exposure images in low light condition by selective fusion of VIS and NIR images. To extract features at different depths, we combine attention with edge feature distillation in the fusion network. We obtain foreground and background features from VIS and NIR images using a segmentation mask. In the training phase, we use the segmentation mask in the loss function to preserve foreground objects while removing motion blur and shadows. For training, we generate a dataset of VIS and NIR images in low light condition using JAI AD-130 GE camera. We manually generate the segmentation mask using Photoshop function and add motion blur into the foreground. Experimental results show that the proposed fusion network successfully removes motion blur and shadows while reproducing realistic color in the fusion results as well as outperforms competing methods on several information-based metrics. The proposed method achieves the highest mutual information (MI) and feature mutual information (FMI), indicating its superior ability to preserve complementary information from the source images while maintaining important structural features. The code is available at https://github.com/Xidian-Media-Lab/motion-deblur .

The rapid advancement of audio generation technologies has enabled the creation of highly realistic synthetic speech, underscoring the need for robust audio spoof detection techniques. However, most existing countermeasures are evaluated on monolingual datasets with utterances from native speakers, leaving their performance on code-switched multilingual speech underexplored. To address this gap, we introduce the multilingual spoofed speech (MSS) dataset, featuring bona fide and spoofed utterances in Urdu, Hindi, and English that capture the natural conversational and code-switching patterns of South Asian speakers. A fused self-supervised learning (SSL) embedding-based audio anti-spoofing framework using an integrated spectro-temporal graph attention network (AASIST) is proposed that combines representations extracted from waveform-to-vector 2.0 and universal speech representation with speaker-aware training frontends. Extensive experiments are conducted on the proposed MSS dataset and the benchmark automatic speaker verification spoofing and countermeasures (ASVspoof) 2019 logical access (LA), ASVspoof-2021 LA, and ASVspoof-2021 deepfake (DF) datasets. Cross-lingual and cross-corpora evaluations demonstrate that the proposed anti-spoofing framework achieves excellent generalization, attaining equal error rates of 0.38%, 0.99%, and 2.46% on ASVspoof-2019 LA, ASVspoof-2021 LA, and ASVspoof-2021 DF, respectively. Moreover, multilingual training on MSS significantly improves the robustness of the implemented artificial intelligence (AI) model, particularly in handling code-switched and linguistically diverse audio. Overall, the results highlight both the engineering contribution of the MSS dataset and the AI contribution of the proposed multi-SSL AASIST framework, together advancing the application of AI for cross-lingual generalizability and the reliable detection of spoofed speech across monolingual and code-switched scenarios.

Aggregate gradation is a key indicator influencing the mechanical performance of concrete structures. To enhance the efficiency of concrete aggregate gradation detection, this study proposes an artificial intelligence-based automated coarse aggregate gradation detection method based on machine vision, incorporating attention mechanisms and feature-path fusion to improve traditional segmentation networks. Additionally, it introduces the Guldinus theorem for the rapid computation of particle morphological characteristics. The proposed method consists of three main components: image pixel correction, a vision-based coarse aggregate instance segmentation neural network, and a gradation curve calculation approach derived from the Guldinus theorem. Through field verification and comparison with traditional sieving methods, the proposed automated approach achieved an average absolute deviation of less than 1% and a root-mean-square error of less than 2%, significantly outperforming the engineering tolerance limit of ±5%, while enabling batch-level detection within seconds. The improved mask region-based convolutional neural network, which integrates the shifted-window transformer and path aggregation feature pyramid network, demonstrates superior segmentation accuracy compared with the conventional mask region-based convolutional neural network, particularly for small particle sizes. Furthermore, the Guldinus-theorem-based gradation estimation method enables accurate and reliable computation of aggregate gradation. The findings of this study provide a robust and efficient artificial intelligence application framework for automated coarse aggregate gradation analysis in engineering practice.

Infrared thermography provides a non-contact solution for gearbox condition monitoring in complex industrial environments. However, reliable fault diagnosis from infrared images remains challenging under few-shot conditions. In this paper, a novel Multi-Scale SwinResNet-enhanced prototype network with CA-Mahalanobis distance metric (MSwinR-CMPNet) is proposed for infrared image-based gearbox fault diagnosis under small samples and multiple different working conditions. First, a Multi-Scale Squeeze and Excitation block (MSSE) is developed to realize multi-scale feature aggregation and channel-adaptive enhancement of shallow layers. Second, the residual convolution and Swin Transformer ensemble block (SwinResNet) is presented to extract local features and model global representations with long-range dependencies to improve the model's feature extraction ability. Third, the covariance-aware Mahalanobis (CA-Mahalanobis) distance metric is involved to replace the traditional Euclidean distance to enhance the class discrimination and confidence reliability of the prototype networks. Experimental results show that the proposed MSwinR-CMPNet model outperforms other benchmarks and achieves an infrared image-based gearbox fault diagnosis accuracy of up to 98.3%. The proposed MSwinR-CMPNet model can provide accurate, stable, and reliable fault identification for infrared-based gearbox monitoring in data-scarce and noise-corrupted industrial scenarios.

Amid deepening global urbanization, rural multifunctional transformation in metropolitan regions has become a key pathway toward urban–rural integration. However, existing research has yet to conceptualize this transformation as a dynamic process, limiting understanding of its evolutionary trajectories and driving mechanisms. Using 147 townships in Beijing as a case study, this study develops a six-dimensional evaluation framework encompassing agricultural production, economic development, social security, cultural recreation, ecological conservation, and technological innovation. Spatial analysis, correlation assessment, and K-means clustering are combined to examine functional evolution, interactions, and typological transitions between 2010 and 2020. The findings indicate that: (1) agricultural production declined while non-agricultural functions strengthened, producing a composite spatial pattern of distance-decay gradients, peripheral enhancement, resource-based peaks, and node-based clustering; (2) correlation analysis revealed strong synergies among agricultural, economic, and social security functions, positive associations with technological innovation, and trade-offs with ecological conservation; (3) six functional types were identified, with transitions following multiple pathways of functional substitution, strategic embedding, resource valorization, and institutional lock-in; and (4) a four-dimensional driving framework—natural endowment, urbanization dynamics, evolving urban–rural demand, and policy–institutional regulation—clarifies how interacting mechanisms generate differentiated functional gradients and typological transitions. This study advances the understanding of how rural functions evolve and reorganize under metropolitan urbanization, providing new insights into the dynamic and multi-scalar transformation of rural areas.

Urban reconstruction in conflict-affected regions is commonly understood through narratives of state failure, humanitarian substitution, or post-war planning. Less attention has been paid to how urban governance is exercised through infrastructures that remain neither fully completed nor entirely absent. Drawing on qualitative urban research conducted between 2022 and 2024, this study examines how diaspora-funded housing and collective facilities shape everyday urban governance in South Lebanon under conditions of partial and selective state planning. Based on urban ethnography, 17 semi-structured interviews, and a structured architectural survey across the city of Tyre and the towns of Jouwaya and Abbassieh, the analysis shows that diaspora-funded infrastructures do not simply compensate for gaps in public provision. Under conditions of fragmented responsibility, their material and operational incompletion can sustain a persistent mode of governance characterised by deferred functionality, moralised access, and dispersed accountability, through which authority, obligation, and responsibility are spatially organised. By conceptualising incompletion as an emergent governing relation rather than a transitional failure, the study contributes to debates on urban governance, infrastructure, and planning in post-conflict and diasporic contexts.

Grasslands are important terrestrial carbon sinks and play a critical role in the global carbon cycle and climate change mitigation. China has implemented a range of grassland restoration and management practices in recent decades. However, their relative importance, nonlinear relationships with net ecosystem productivity (NEP), and interactions with environmental factors remain insufficiently understood at the national scale. This study integrates long-term remote sensing, environmental, and grassland management practice datasets and applies the XGBoost-SHAP modeling to investigate the spatiotemporal patterns and drivers of grassland NEP across China. The results indicated that NEP increased across 73.44% of China’s grasslands, with the most pronounced increases occurring in the northeastern grasslands and the eastern Qinghai-Tibet Plateau. Grassland management variables accounted for part of the variation in NEP captured by the model, with grazing intensity (GI) identified as the most important management predictor. GI showed a significant nonlinear association with grassland NEP. Across different ecological regions, GI interacted most strongly with elevation, temperature, vapor pressure deficit, and slope, highlighting marked spatial heterogeneity in the effects of grassland management. These findings underscore the critical role of grassland management in regulating grassland NEP and provide a scientific basis for optimizing region-specific management strategies and safeguarding ecological security.

Soil p CO 2 is an important driver of carbonate dissolution and links terrestrial biological processes with groundwater dissolved inorganic carbon (DIC). However, how land use alters soil p CO 2 variability and its coupling with groundwater DIC remains poorly understood, particularly in karst critical zones. Here, we combined two years of high-frequency monitoring with Random Forest (RF), Convergent Cross Mapping (CCM), and structural equation modeling (SEM) to investigate soil p CO 2 dynamics and soil CO 2 to DIC systems, with higher concentrations in shrubland and grassland than in cropland and bare soil. Environmental controls also varied among systems, which indicated that bare soil was primarily associated with moisture variation, whereas vegetated systems showed more complex temperature and moisture responses that were consistent with an additional contribution from vegetation-associated biological CO 2 production. Reconstructed spring water DIC also differed among systems, with lower concentrations in bare soil and higher values in the vegetated systems. The p CO 2 -DIC association was weakest in bare soil, where DIC variation was more closely related to hydrological pulses and soil moisture constraints, but became stronger in cropland and grassland, suggesting closer coupling between soil CO 2 supply and carbonate weathering related DIC generation. Shrubland showed high soil p CO 2 and a distinct DIC response associated with temperature and hydrochemical conditions. Future climate scenario simulations further indicated that soil p CO 2 responses to warming may also differ among land-use systems and may become nonlinear under stronger warming. Overall, our findings indicate differences related to land use in soil p CO 2 dynamics and reconstructed DIC responses in groundwater within the monitored karst simulation platform, highlighting the sensitivity of coupled water and carbon processes in karst critical zones to surface conditions and climate forcing.

Rainfall erosivity is a key driver of soil erosion and hydrologic processes, yet its estimation remains limited by sparse gauge observations and uncertainties in satellite precipitation products. This study presents a machine learning–based framework to improve global rainfall erosivity estimates derived from the Integrated Multi-satellitE Retrievals for GPM (IMERG) precipitation data. Rather than directly estimating erosivity from satellite precipitation, we focus on modeling and correcting the errors between IMERG-derived and gauge-based erosivity using a suite of hydroclimatic and land surface predictors. Global evaluation using 5935 gauge stations shows that erosivity derived from IMERG V06 and V07 systematically underestimates ground-based values by 56% and 62%, respectively, although both products capture spatial variability with moderate-to-high correlation ( R = 0.70–0.73). The proposed machine learning correction substantially reduces bias and error magnitude, yielding near-unbiased estimates (relative bias <1%) and improved agreement with observations ( R = 0.93–0.94). Feature attribution analysis indicates that errors are primarily associated with atmospheric moisture, energy fluxes, and precipitation extremes, highlighting the physical controls on satellite retrieval limitations. We further show that differences in residual structure between IMERG versions influence model learnability, and that spatial scale plays a critical role in shaping erosivity estimates, with grid-based products smoothing extreme rainfall signals. The resulting open-source framework enables reproducible, point-scale estimation of rainfall erosivity from globally available data and provides an accessible workflow for applying machine-learning-based error correction to satellite-derived erosivity estimates. This open and reproducible implementation supports broader applications in soil erosion modeling, land management, and hydrologic assessments, particularly in regions with limited ground observations.

We analyze 3163 climate-related YouTube videos posted between 2006 and 2021. More sophisticated videos—those with higher production quality, richer descriptions, longer duration, and specialized terminology—draw climate-related comments that lean toward explicit position-taking and away from affective response, and they tend to receive more views. Content sophistication thus shapes the character of climate discourse, while the overall amount of climate discussion remains about the same.

Science-Policy Interfaces (SPIs) are increasingly recognized as mechanisms for linking knowledge generation with biodiversity decision-making. However, little is known about how institutionalized SPIs operate at subnational levels, particularly in megadiverse countries of the Global South, a region characterized by political institutional weaknesses and significant socio-environmental challenges. Mexico offers an illustrative case for examining these dynamics through its State Biodiversity Strategies Initiative (SBS-Initiative), one of the most extensive subnational biodiversity planning experiences worldwide promoted by the Convention on Biological Diversity. This study analyzes the institutional design of the SPIs established through the SBS-Initiative, the barriers and enablers shaping their development, and their contributions to the credibility, relevance, and legitimacy of this planning process. Using the CRELE framework as an analytical lens, we conducted a qualitative analysis of 34 institutional documents and 31 semi-structured interviews with national and subnational government officials and researchers involved in these interfaces. Three main findings stand out: the initiative combines two complementary SPI configurations, a predominantly linear model for biodiversity diagnostics and a co-productive model for strategies and action plans; barriers and enablers are context- and phase-dependent; and credibility, relevance, and legitimacy emerge dynamically from the interaction among institutional arrangements, diverse stakeholders, and multiple governance conditions. We conclude that SPIs do not depend on choosing between linear or co-production models, but on developing adaptive and hybrid institutional arrangements capable of responding to dynamic and diverse subnational contexts. These findings offer practical insights for strengthening biodiversity planning and other sustainability governance processes.

Severe convection produces localized hazards that often require warnings before radar echoes fully reveal storm development. Convective initiation and the maintenance of intense convection remain challenging for radar-only nowcasting because pre-convective signals may be absent from recent radar observations, and strong echoes often decay rapidly in forecasts. Here we present FuXi-Nowcast, an environment-conditioned deep learning system that combines high-resolution observations with three-dimensional atmospheric forecasts to predict composite reflectivity, precipitation, wind gusts, and surface variables up to 12 h ahead. In April–July 2024 evaluations over East China, FuXi-Nowcast outperforms operational numerical weather prediction, persistence, and extrapolation baselines for reflectivity and precipitation. Case studies, diagnostics, and ablation experiments suggest that atmospheric moisture information and explicit preservation of strong convective signals contribute to forecasts of convective initiation and maintenance. These results show that environmental conditioning can mitigate important failure modes of radar-only nowcasting for high-impact convective weather.

ABSTRACT Drought is a complicated and recurrent natural hazard that creates substantial challenges to sustainable water management and climate adaptation. To address these challenges, Multi‐Model Ensembles (MMEs) of GCM simulations are extensively used for assessing future drought conditions. However, to attain correct and precise characterization of drought there is a need to enhance the development of MMEs to reduce uncertainties and augment spatial consistency. Since the geospatial performance of individual GCMs varies considerably across different regions, incorporating these spatial characteristics into the ensemble formation process is essential for developing an efficient and regionally robust ensemble framework. The study proposes a new framework of drought assessment under GCMs simulations based MMEs to enhance the reliability of future drought projections called—Precipitation‐Concentration Kriging Ensemble Drought Assessment Framework (PCK‐EDAF). The construction of PCK‐EDAF comprises two sequential phases. The first phase, termed Geostatistical Homogeneity Analysis and Ensemble Weight Optimization (GHA‐EWO), derives optimal weights based on the spatial and temporal consistency of GCM outputs with observed precipitation data. The second phase, called the Kriging‐Weighted Ensemble Standardized Drought Index (KWESDI), applies these optimized weights to future ensemble simulations and standardizes the resulting drought estimates. Application of the PCK‐EDAF is based on the 103 grid points across Pakistan using simulations from 22 GCMs under multiple Shared Socioeconomic Pathways (SSP1–2.6, SSP2–4.5, and SSP5–8.5). Under PCK‐EDAF, the performance of the GHA‐EWO is evaluated as compared to conventional Equal Weighted Ensemble (EWE) and Mutual Information based ensemble in terms of Root Mean Square Error, Mean Average Error and correlation measures. Findings indicate that the GHA‐EWO is always the most effective, with lower values of RMSE and MAE and higher correlation coefficients, which prove that the model is more accurate and more consistent with the observed precipitation patterns. To determine the trend in drought under KWESDI we applied Mann‐Kendall test and the slope estimator of Sen. Results related to trend analysis show a significant increasing tendency in drought, particularly under SSP1–2.6 and SSP5–8.5 scenarios at longer time scales. SSP2–4.5 shows a weaker drought signal with statistically insignificant trends due to moderate emissions forcing. These results also suggest an amplification of drought and wetness cycles, which indicate increased climate variability in the future. Overall, the findings reveal that integrating kriging‐derived spatial weights within PCK‐EDAF significantly enhances ensemble reliability. Moreover, the developed PCK‐EDAF framework is modular and generalizable, allowing its application to other regions and datasets. This integration further enables a more realistic and spatially consistent assessment of future drought conditions under changing climate scenarios.

Abstract The contrast between SSTs in the convective and non-convective regimes of the tropics is closely coupled to the tropical atmospheric temperature structure. Prevailing theory for the tropical atmosphere suggests that this warm-cold contrast is the aspect of the tropical SST pattern most related to tropics-wide top-of-atmosphere fluxes, providing a potential mechanism for a tropical SST ‘pattern effect’ (Stevens et al. 2016). However the response of the tropical cold-warm contrast to an increase in atmospheric CO 2 is unknown. Here we quantify it for the first time. We find that in models and observation-based products over the historical period, and in the simulated response to a CO 2 increase, the warm-cold contrast increases. This increased contrast in response to abrupt quadrupling of atmospheric CO 2 is linked to a wind-induced pattern of latent heat flux. This organized change in wind speed represents an enhancement of the mean pattern in wind speed (surface winds are low in the convective regime which is characterized by low-level convergence). The amplified patterns in wind and SST are accompanied by amplified patterns in cloud radiative effect and precipitation which may suggest a large-scale convective aggregation in the tropics.

Abstract Accurate high-resolution estimation of fine particulate matter (PM2.5) remains challenging because of sparse monitoring networks and missing satellite observations. We developed a multistage deep learning framework to generate daily PM2.5 concentrations at 100 m resolution across the contiguous United States (CONUS) from 2000 to 2024. The framework first reconstructed missing satellite aerosol optical depth (AOD) using a U-Net-based encoder–decoder informed by reanalysis data, refined temporal dependencies using a bidirectional long short-term memory network, and downscaled reconstructed AOD to 100 m using terrain information. Daily PM2.5 was subsequently estimated using a multistream deep learning architecture integrating reconstructed AOD, meteorological, spatiotemporal, and geospatial predictors. Evaluation using strict site-level data partitioning yielded strong predictive performance (R2 = 0.82, RMSE = 2.85 μg/m3, MAE = 1.84 μg/m3), with high spatial (R2 = 0.94) and temporal (R2 = 0.78) performance. The framework generated spatially continuous daily PM2.5 surfaces across more than 766 million 100 m grid cells while capturing broad spatial gradients and fine-scale variability. These long-term, high-resolution estimates provide an exposure surface suitable for epidemiological, environmental justice, and air-pollution assessment applications.

Large-scale hydrological models like CTRIP and MGB are essential for simulating river dynamics and supporting large-scale climate studies. Their accuracy can be significantly improved through satellite data assimilation. This study leverages the stand-alone value of 20 years of ESA Climate Change Initiative (CCI) high-resolution discharge and water surface elevation (WSE) products (2000–2020) for improving large-scale hydrological simulations through data assimilation. To evaluate the added-value of these produtcs across contrasting modelling and hydrological contexts, we assimilate altimetry-derived discharge, multispectral-imagery-derived discharge, and WSE anomalies into two existing ensemble Kalman Filter frameworks: HyDAS in CTRIP, a global-scale physically based and uncalibrated river-routing model, and HYFAA in MGB, a calibrated semi-distributed regional hydrological model. The experiments are conducted over the Niger and Congo basins, which differ in hydrological variability, river-network structure, wetland influence, and CCI product availability. Across the experiments, discharge assimilation generally outperformed WSE anomaly assimilation because discharge is directly represented in the routing models, providing a more direct and physically consistent correction, while WSE requires consistency between observed and simulated rating curves to be converted into effective discharge corrections. In the Niger basin, where the seasonal signal and station coverage better constrain the main river dynamics, assimilating altimetry-derived discharge led to the strongest improvement in MGB, increasing the median Nash-Sutcliffe Efficiency (NSE) to 0.83 and the correlation coefficient to 0.94. WSE anomaly assimilation was beneficial in specific cases, particularly when the baseline simulation was poor and when observed and simulated rating curves were well aligned. Temporal data density in discharge assimilation emerged as a key driver of performance gains. Assimilating high-frequency discharge data from multispectral imagery significantly reduced bias, from 1.2 to near 1 in MGB, and from 2.3 to 1.78 in CTRIP (median values), supporting hydrological assessments related to long-term variability. Furthermore, the higher temporal resolution allowed for better capture of flow variability, with Kling-Gupta Efficiency γ approaching 1.0 in MGB, which is relevant for both seasonal climate studies and short-term predictions, such as extreme hydrological events. The comparison with the Congo basin highlights the limits of transferability across hydrological contexts and emphasizes the trade-offs between temporal resolution, spatial sampling, and product quality. Improvements within the Congo basin were more modest and more product-dependent because the available CCI stations provide a weaker spatial constraint on a basin where discharge integrates contributions from large tributaries, and because of lower discharge product quality at some stations. Overall, the results demonstrate that ESA CCI WSE and discharge products can improve large-scale hydrological simulations, but the magnitude and reliability of the improvements are not uniform and depend on the interaction between product type and quality, temporal and spatial sampling, model configuration, and basin-specific hydrological processes. Future work includes merging altimetry and multispectral discharge data, improving discharge retrieval algorithms using SWOT data, and refining data assimilation techniques to support climate studies and river system modeling in complex, climate-impacted basins.

Increased snow accumulation on the Antarctic Ice Sheet mitigated global sea level rise by ∼ 11 mm during 1901–2000 according to ice core reconstructions. However, in the most recent 40 years of more intense observation and warming, the trend in the Antarctic-wide accumulation rate has been negligible. We attribute these trends by evaluating Earth system model experiments in comparison with dynamically consistent reconstructions of surface climate. Single-forcing experiments reveal that rising concentrations of greenhouse gases (GHGs) have been the underlying driver of increased accumulation, yet acting alone would have caused twice the observed accumulation-related sea level mitigation during 1901–2000. Aerosol-driven cooling partially compensates this overprediction, but the reconstructions provide evidence that poorly modeled processes can explain observation-model trend discrepancies. In particular, these data support a hypothesis that high-latitude winds have been working together with ice-shelf meltwater fluxes to dampen Southern Ocean surface warming and suppress the GHG-driven accumulation increase since the initiation of West Antarctic ice shelf thinning in the mid-20th Century. The wind pattern associated with strengthening of the Southern Hemisphere westerlies and deepening of the Amundsen Sea Low distributes accumulation unevenly across the continent in an orographic pattern that is consistent across models and the reconstructions. In reconstructions, these same wind and accumulation patterns are associated with muted surface warming across the eastern Pacific and Southern Ocean, a pattern not captured in climate projections including the all-forcings large ensemble studied here. However, the westerly wind history constrained by paleoclimate data assimilation largely reconciles differences between the model's ensemble-mean response and the observed world for both Antarctic-wide accumulation and large-scale warming patterns. Although the large ensemble simulates similar wind histories to the real one – driven by internal variability and anthropogenic forcing – its corresponding responses in SSTs and Antarctic-wide snow accumulation are decoupled from the wind. We discuss how this significant observation-model discrepancy, which has implications for projecting regional climate change, likely arises from omitted meltwater forcing and/or resolution limitations. As a component of the sea level budget and a gauge of the magnitude and spatial pattern of climate change, Antarctic snow accumulation is a critical target for models to replicate.

Outputs from complex Earth system models (ESMs) participating in the Coupled Model Intercomparison Project (CMIP) are a major source of information for policy-relevant assessments of climate change. The way we view and interpret the CMIP archive shapes our understanding of the real world, yet many assessments present analysis choices only implicitly. We describe a fully Bayesian approach and software package for analyzing CMIP data and present a series of interpretive models with gradually increasing complexity to illustrate the methodology. We also show how to update CMIP-derived posteriors using additional evidence, including observations, emergent constraints, and information about processes in the Earth system models themselves. Using these methods, we show that even apparently strong relationships between observable processes and equilibrium climate sensitivity (ECS) in ESMs do not necessarily tightly constrain ECS. We also show that estimates of β, the terrestrial carbon dioxide fertilization effect, are revised downward when considering the individual processes included in ESMs. Our results illustrate how these and other estimates may be updated as new information arrives.

Study Region The Yangtze River is a large regulated river system with pronounced variations in hydrology, sediment transport, and river–lake interactions. This study focuses on the mainstem from Yibin to Shanghai, particularly reaches influenced by the Three Gorges Dam (TGD), Dongting Lake, and Poyang Lake. Study Focus An integrated satellite-driven framework was developed to reconstruct turbidity dynamics during 2016–2024. Landsat-8/9 imagery was combined with ensemble machine learning, causal inference, and GeoDetector analysis to characterize turbidity variability and identify environmental controls. The Voting Regressor achieved the best performance (R² = 0.934, RMSE = 5.221 NTU, MAE = 3.874 NTU, MAPE = 18.618%). New Hydrological Insights for the Region Basin-wide turbidity declined by 32.2% during 2016–2024, reaching a minimum in 2022 followed by a slight rebound in 2023–2024. Turbidity was generally higher during the low-water period than during the high-water period. Spatially, turbidity generally increased downstream but exhibited pronounced heterogeneity. A marked reduction occurred downstream of the TGD, whereas turbidity increased farther downstream in reaches influenced by Dongting Lake. Near the Yangtze–Poyang Lake confluence, turbidity showed a positive correlation with runoff (r = 0.54), contrasting with upstream reaches. Hydrological factors exerted the strongest influence, while meteorological and land-use factors mainly contributed through their interactions with hydrological conditions. These findings underscore the importance of considering reservoir regulation, river–lake interactions, and watershed environmental factors in understanding turbidity dynamics in large regulated rivers.

Abstract The sixth assessment report (AR6) of the Intergovernmental Panel on Climate Change (IPCC) presented regional summaries of the scientific knowledge of changing weather extremes and the human influence upon this. The hexagon figures, found in the Summary for Policymakers (SPM), were a key part of this presentation. They provided clear visual overviews of this understanding on three hazard types (hot extremes, heavy precipitation and agricultural and ecological drought) for policymakers, scientists and communicators alike. We have shown that since 2021, when AR6 was published, there has been a rapid advancement in knowledge of changing extremes on the timescale of individual years, with a greater than doubling in the number of extreme event and trend attribution studies coupled with advances in methodologies, rising levels of anthropogenic forcing, and a growing sample of manifested extreme events. The coming seventh assessment report (AR7) is an opportunity to update and iterate these figures, expand to other hazard types and, we argue, to facilitate more regular updates in line with this rapidly advancing field of knowledge. To this end, we have updated the AR6 hexagons in several ways: we have added evidence levels to the figures to provide additional information without sacrificing visual clarity; we have built on the AR6 regional synthesis process with a step-by-step procedure and accompanying expert guidance and discussion; we have applied this to six illustrative regional case studies for heavy precipitation; we have updated the evidence tables and figures directly for hot extremes and heavy precipitation.

Accurate quantitative precipitation estimation (QPE) is critical for responding to severe weather events such as heavy rainfall. Deep learning (DL) methods, which can establish nonlinear mappings between radar observations and rain rate (R), have been widely applied to reduce QPE errors. This study evaluates point-based and spatial DL approaches for radar QPE across precipitation intensity ranges over Hainan Island using dual-polarization radar and rain gauge data. Four DL-based QPE models, including R-DNNNet, R-IncNet, R-Res-IncNet, and R-DenseNet, are trained using point-based or spatial grid-based datasets, and evaluated across retrospectively defined subsets based on the observed rain-gauge rain rate: low-intensity (R < 10 mm h−1), moderate-intensity (10 ≤ R ≤ 20 mm h−1) and high-intensity precipitation (R > 20 mm h−1). Results show that DL models generally outperform traditional empirical relationships. For overall precipitation, the point-based three-parameter (ZH-ZDR-KDP) R-DNNNet achieves the best performance (CC = 0.94, RMSE = 5.88 mm h−1), improving CC and RMSE by 4% and 23%, respectively, over the best traditional empirical method. R-DNNNet also performs best for low-intensity and moderate-intensity precipitation. For high-intensity precipitation, the spatial convolutional neural network (CNN) model R-Res-IncNet showed the best overall performance, with maximum improvements of 8%, 14%, and 19% in CC, RMSE, and MAE, respectively. R-Res-IncNet achieved an RMSE of 12.45 mm h−1 compared with 12.81 mm h−1 for R-DNNNet, and the paired-bootstrap 95% confidence interval for the model-to-model RMSE difference was 0.018–1.006 mm h−1, indicating a modest but statistically supported improvement. This result suggests that retaining neighborhood radar information may be beneficial under high-intensity conditions. These intensity-specific rankings are based on retrospective evaluation within subsets defined by observed rain-gauge rain rate and characterize conditional model performance across precipitation intensity ranges. Overall, the results show that the relative performance of the evaluated point-based and spatial-grid QPE configurations varies with rainfall intensity, while the three-variable polarimetric input provides consistent benefits across the evaluated datasets. In summary, the findings highlight the precipitation-intensity-dependent performance characteristics of point-based and spatial DL models and suggest their potential for improving radar QPE performance over Hainan Island, particularly for severe convective rainfall.