Radar echo extrapolation is critical in short-term weather forecasting. To enhance the accuracy and adaptability of radar echo prediction, this paper proposes a novel method that integrates a linear variable convolution (LVC) module and a cross-attention (CA) mechanism into the spatiotemporal long short-term memory (ST-LSTM) framework, named LVC-LSTM. The LVC module enables dynamic adjustment of the convolutional sampling shape, allowing the network to capture the irregular and evolving structures of radar echoes more accurately, thereby offering improved flexibility and representation capability compared with traditional and deformable convolutions. The CA mechanism introduces a pyramid-based CA structure that incorporates cross-scale embeddings and hierarchical attention blocks, effectively capturing multi-scale features and long-and short-range spatial dependencies inherent in radar echo dynamics. Experimental results using a real-world radar echo dataset demonstrated that the LVC-LSTM model outperformed comparative models across multiple evaluation metrics, including the critical success index , heidke skill score, root mean square error, mean absolute error, and structural similarity index, indicating its strong potential for operational radar echo extrapolation.

As atmospheric environmental prediction continues to improve, interpretable validation of pollution mechanisms and feedback processes has become a main challenge in atmospheric chemistry. Yet mechanism validation based on complex numerical models still relies heavily on expert knowledge: mechanistic hypotheses must be operationalized into executable experiments, and model outputs must be organized into traceable evidence. We present TianJi-Environ, an auditable AI Scientist for atmospheric-chemistry mechanism validation. TianJi-Environ establishes the first WRF-Chem-based multi-agent framework that autonomously drives complex atmospheric-chemistry simulations, converting mechanistic hypotheses into executable configurations, testing experiments, and evidence criteria. Using ozone response and particulate-matter feedback as two representative examples, we demonstrate TianJi-Environ's capability for mechanism validation. In a summertime ozone case over the North China Plain, the system detects directionally consistent aerosol-radiation-interaction signals in shortwave radiation and boundary-layer height, but judges the evidence for ozone response to NOx control to be incomplete. In a wintertime PM2.5 case over the Guanzhong Basin, it localizes the unsupported link to insufficient propagation from black-carbon perturbation to particulate response and missing diagnostics of vertical absorptive heating. These results show that TianJi-Environ makes expert-driven mechanism validation explicit, structured, and auditable, offering a reproducible paradigm for multi-agent systems coupled with complex atmospheric-chemistry models.

The Seychelles Dome (SD) is a prominent thermocline dome in the southwest tropical Indian Ocean, exerting significant impacts on regional marine ecosystems and climate. Accurate simulation of the SD in climate models is therefore crucial. This study evaluates the reproducibility of key SD characteristics in models participating in the Coupled Model Intercomparison Project phase 6. Results show that the multimodel ensemble mean can effectively capture the unique semiannual oscillation of SD heat content and the close interannual relationship with El Niño–Southern Oscillation (ENSO). However, substantial discrepancies exist among individual models in both the seasonal cycle and interannual variations. Notably, models that struggle to simulate the semiannual oscillation have apparent biases in the location of the intertropical convergence zone over the tropical Indian Ocean. These biases lead to inaccurate surface wind stress over the SD, affecting Ekman pumping strength and thus the semiannual oscillation in heat content. Furthermore, models that poorly simulate the interannual ENSO–SD relationship demonstrate biases in ENSO-driven atmospheric teleconnections to the tropical Indian Ocean. Both the intertropical convergence zone (ITCZ) and atmospheric teleconnection biases are closely linked to mean-state sea surface temperatures (SSTs) in the eastern tropical Pacific. While increasing the SSTs improves the simulation of the ENSO–SD relationship, it deteriorates the simulation of the SD’s seasonal cycle. Nevertheless, Coupled Model Intercomparison Project phase 6 (CMIP6) models show an apparent improvement in simulating the SD compared to their precursors. This provides confidence in using them to project SD changes under global warming.

Understanding the behavior of the Indian Ocean Walker circulation (IWC) is crucial for interpreting past climate variations; however, its mechanisms during typical interglacial periods remain insufficiently understood. This study examines IWC changes during the mid-Holocene and Last Interglacial (LIG) periods compared to the preindustrial period using simulations from 16 models within the Coupled Model Intercomparison Project Phase 6 framework. The results reveal a weakened annual-mean IWC with pronounced seasonal asymmetries. These IWC variations are largely independent of the Pacific Walker circulation and instead arise from Gill-type adjustments to orbitally driven shifts in African summer monsoonal precipitation. During austral summer, reduced insolation suppresses southern African monsoon precipitation, generating eastward-propagating Kelvin waves that strengthen the IWC. Conversely, enhanced boreal-summer insolation intensifies northern African precipitation, producing a mirror-like response that substantially weakens the IWC and dominates the annual-mean signal. This summer weakening is additionally fueled by concurrently reduced zonal sea surface temperature (SST) gradients over the tropical Indian Ocean, representing a secondary role of ocean-atmosphere coupling. Additionally, during boreal spring transition, when monsoonal activity is minimal, the IWC undergoes moderate attenuation, and this change is driven by suppressed convection over the Maritime Continent and Southeast Asia, resulting from orbitally induced insolation decline and associated tropical cooling. This study highlights the dominant influence of African monsoonal systems in shaping the IWC during typical interglacial periods rather than regional oceanic conditions, offering a plausible explanation for geological evidence of wetter conditions in eastern Africa and reduced zonal SST gradients in the tropical Indian Ocean.

Global wave forecasts play a crucial role in international shipping, global trade, and various economic and social activities. However, current numerical wave models, such as WAVEWATCH III (WW3), still exhibit substantial forecast errors, particularly in wave-active regions. Therefore, we propose an Intelligent Error Correction Method (IECM) that integrates multiple U-Nets with a boundary fusion algorithm to correct global Significant Wave Height (SWH) forecasts from WW3. This approach effectively considers different wave characteristics in different ocean regions and alleviates performance degradation at forecast boundaries. Following error correction, the mean Root Mean Squared Errors (RMSEs) of WW3 global SWH forecasts at lead times of 24, 48, and 72 h are reduced from 0.44 m, 0.46 m, and 0.49 m to 0.22 m, 0.25 m, and 0.31 m, corresponding to percentage reductions of 50

Abstract Dynamical Earth System Models are extremely expensive, requiring ∼106 core‐hours for century‐scale simulations. This limits climate projections to only a few Shared Socioeconomic Pathways (SSPs) and leaves numerous policy trajectories unexplored. Here, we develop an external forcing boundary‐constrained generative emulator trained on CMIP6 model outputs, enabling rapid projection of temperature, precipitation and sea‐surface height under arbitrary CO2 forcing pathways. The model learns the continuous manifold of future climate states by conditioning on the physical boundary defined by two extreme scenarios (SSP1‐1.9 and SSP5‐8.5). It can reconstruct climate responses under pathways held out during training (e.g., SSP2‐4.5). We further apply it to four newly constructed CO2 pathways (higher‐emission, lower‐emission, overshoot, and step‐change) to demonstrate its generalization capability across distinct forcings. This allows us to capture regional shifts in extreme events and threshold timings. The emulator substantially reduces computational demand and provides a high‐throughput, user‐driven platform for evaluating flexible mitigation strategies.

A comprehensive understanding of Earth is essential to address climate change, but the fragmentation and explosive growth of data make it impossible for scientific discovery to keep pace with planetary change. We address this challenge by introducing EarthLink, the first AI “copilot” for Earth scientists that can automate the entire research process, enabling systematic, large-scale exploration across over 5 petabytes of cross-disciplinary data and more than 500 analytical tools. Evaluations based on over 900 expert scores demonstrate that EarthLink achieves performance comparable to junior scientists across core research tasks, including bias diagnosis and future climate projection. Crucially, while existing systems in other fields are often limited to textual reasoning or require human validation, we present the first demonstration of an AI autonomously formulating and verifying a novel physical mechanism. When tasked with the open-ended challenge of identifying Atlantic Niño precursors, EarthLink autonomously uncovered previously unrecognized drivers and formulated a physically interpretable mechanism. These results establish EarthLink as a robust AI-driven engine capable of generating original scientific insights previously thought exclusive to human cognition, paving the way for a fundamental shift in the pace and scale of Earth science research.

Predicting tropical cyclone (TC) precipitation is an important step in disaster prevention and mitigation. However, in the probability prediction of TC precipitation, traditional deep learning models are highly sensitive to initial conditions and can only provide deterministic forecasts, making it difficult to quantify uncertainty. In this study, we develop an AI-driven deep learning model based on diffusion models, incorporating historical data to reduce sensitivity to initial conditions and enhance precipitation distribution accuracy. Compared with traditional deep learning methods, this model outperforms other models in terms of the SSIM and PSNR for deterministic prediction of TC precipitation in 0–12 h. For probabilistic prediction, this model also achieves lower CRPS and Brier scores. Therefore, diffusion-based deep learning models not only show broad application prospects in TC-precipitation forecasting but also hold promise for providing probabilistic prediction methods for various disasters, enabling the widespread adoption of probabilistic forecasting across different prediction domains.

CRA-40 and CMA-RA V1.5 are two sets of global atmospheric reanalysis data released successively by the China Meteorological Administration.Previous studies have indicated that CRA-40 is inadequate in effectively repre-senting one of the most important phenomena in the tropical stratosphere—the quasi-biennial oscillation(QBO).This study evaluates the accuracy of CMA-RA V1.5 reanalysis data in depicting the QBO and compares its repre-sentation with that provided by ERA5.The results demonstrate that CMA-RA V1.5 offers substantially improved representation of key aspects of the QBO,including its index,period,and amplitude,when compared to CRA-40.Moreover,CMA-RA V1.5 exhibits a skill comparable to ERA5 in characterizing the QBO.The advanced data as-similation technology employed may be a crucial factor contributing to this enhanced capability of CMA-RA V1.5.Based on these findings,it is advisable to use CMA-RA V1.5 when employing Chinese atmospheric reanalysis data for stratospheric research.

The Northwest Pacific subtropical high (NWPSH) significantly affects East Asian weather and climate, rendering the prediction of its intensity and location critically important. This study aims to evaluate the performance of the Convolutional Long and Short-Term Memory (ConvLSTM) model for predicting the summertime 500 hPa geopotential height and NWPSH intensity and area at a lead time of three months, and to compare it with the dynamical models of the Nanjing University of Information Science and Technology Climate Forecast System (NUIST-CFS1.0) and the Canadian Seasonal to Interannual Prediction System Version 2 (CanSIPSv2). The mean latitude-weighted RMSE (RMSEw), anomaly correlation coefficient (ACC), and NWPSH indices are used as evaluation metrics. For both summer mean and monthly prediction, the ConvLSTM model outperforms the two dynamical models in terms of RMSEw and ACC for the 500 hPa geopotential height over the western Pacific region. The correlation coefficients between the NWPSH intensity index predicted by the ConvLSTM model and the observations are higher than those obtained from the two dynamical models. Regarding the NWPSH area index, the ConvLSTM model shows more stable performance. Particularly in August, the improvement of the ConvLSTM model compared to the two dynamical models is more significant, indicating the robust capability in capturing late-summer circulation patterns. Therefore, the ConvLSTM model demonstrates significant potential for summer NWPSH prediction, offering a new perspective and approach for climate prediction in this region.

Air quality forecasting has attracted increasing attention as global air pollution worsens. Spatiotemporal graph neural networks have become a leading paradigm, thanks to their ability to capture complex spatial and temporal dynamics in Air Quality Index (AQI) data. However, existing methods remain limited by weak modeling of long-range temporal dependencies and insufficient integration of meteorological factors. Building on a publicly available nationwide air quality dataset spanning eight years, we propose MADGCN, a Meteorology-Aware Decoupled Spatio-Temporal Convolutional Network that jointly addresses long-horizon temporal modeling and meteorological context fusion. MADGCN includes a dynamic causality discovery module grounded in Granger causality, which captures time-varying causal relationships between meteorological conditions and AQI dynamics. The inferred causal structures further guide a causal graph convolution module and a PatchMixer module, enabling effective spatial interaction modeling and multiscale temporal dependency learning. Extensive experiments against 16 strong baselines show that MADGCN achieves competitive performance for long-horizon air pollution forecasting and generalizes well under high-pollution regimes.

Abstract Accurate prediction of gusts is critical for risk mitigation, but the turbulent processes that generate gusts are unresolved in conventional numerical weather prediction models. Here we introduce a deep‐learning‐based wind gust estimation scheme (DL‐GUST) that uses archived diagnostics from the Weather Research and Forecasting (WRF) model within a long short‐term memory (LSTM) network as an offline, one‐way gust‐estimation framework. Evaluated over the northwestern inland region of China, DL‐GUST achieves a root‐mean‐square error (RMSE) of 1.98 m s−1 and a Pearson correlation coefficient (PCC) of 0.93 on an independent temporally separated test set. Relative to empirical, statistical, and physics‐based baseline methods, RMSE is reduced by 45%–60%, and by 50%–65% for the strongest gust events from the four test months. Additional validation over the European Alps and the southeastern coastal region of China shows useful but non‐uniform transfer skill across different surface environments and weather conditions.

Artificial intelligence (AI) is rapidly transforming Earth science, offering unprecedented capabilities to tackle the most pressing challenges in the field. This work explores significant advances and emerging challenges across the AI for atmosphere-ocean sciences, while outlining critical ways forward. We review deep-learning methods and their application in weather and climate forecasting, which outperforms dynamical models in accuracy and computational efficiency. The role of AI in detecting complex phenomena, enhancing data assimilation and reconstruction, bias correction and downscaling coarse model outputs is also examined. However, the 'black-box' nature of complex AI models necessitates a focus on explainable AI to build trust and extract mechanistic insight. The most promising path forward is identified as the development of hybrid physics-AI modeling, which integrates the data-driven power of AI with the foundational constraints of physical laws to ensure generalizability and causal consistency. A new framework for AI-based model intercomparison is essential for rigorous benchmark performance. Finally, we contextualize these technical developments by discussing the usefulness and applicability of AI to society, including the improvement of multi-hazard early-warning systems and green energy production. We conclude by envisioning the future of AI agents for Earth science-autonomous, goal-oriented systems capable of designing and running experiments, generating and testing hypotheses, and learning dynamics from multisource data. This synthesis underscores that AI is not merely a tool, but a paradigm shift, which will significantly improve how we understand and adapt to a changing climate.

Despite the rapid warming in other oceans under increased anthropogenic radiative forcing, the eastern tropical Pacific has experienced a robust cooling trend over the recent decades since early 1980’s, which has exerted significant and prolonged impacts on global climate, partly mitigating the global warming and even inducing an intermittent hiatus trend in early 2000’s. This counterintuitive multidecadal cooling in the eastern tropical Pacific has perplexed the climate community regarding its underlying causes. Many previous studies have proposed different potential explanations and/or hypotheses, including the influence of internal climate variability (e.g., the negative phase of interdecadal Pacific Oscillation), the inter-basin influence from the strong warming in the tropical Indian Ocean and Atlantic Ocean, the influence of aerosol forcing change, and the impact of Atlantic Multidecadal Variation (AMV), and so on. In this study, we identify a distinct and previously underappreciated driver. By conducting a series of sensitivity coupled model experiments based on CESM2, we demonstrate that an intensified annual cycle of sea surface temperature in the tropical Atlantic can also drive a cooling in the eastern tropical Pacific over the past decades. The intensified annual cycle significantly alters the distribution of seasonal precipitation, leading to a net reduction of annual mean precipitation in the equatorial Atlantic. This dry state is accompanied by lower-troposphere divergence, inducing surface easterly anomalies along the equatorial Pacific and ultimately leading to a strong Pacific cooling. Our findings unveil a previously overlooked role of the tropical Atlantic annual cycle change in regulating tropical Pacific climate and shaping large-scale climate patterns.

Accurate tropical cyclone track forecasting is of crucial importance for effective disaster risk management in coastal regions. These storms are recognized as one of nature's most hazardous weather phenomena. However, traditional typhoon track prediction methods face certain limitations in handling complex meteorological data and multi-dimensional features. To improve prediction accuracy, this paper proposes an improved Wide & Deep framework. This framework utilizes a Stacked Two-layer Long Short-Term Memory (S2-LSTM) encoder in the Wide branch to model the sequential features of typhoon tracks, and incorporates a Multi-ConvGRU in the Deep branch to handle three-dimensional meteorological data. Following a dynamic/static feature decomposition, the S2-LSTM encoder processes a subset of dynamic features, effectively integrating typhoon position, wind speed, and rate-of-change information from multiple time steps to better capture the temporal evolution patterns of typhoon movement. Furthermore, a Coordinate Attention (CA) module is introduced, which captures positional dependencies along the vertical and horizontal dimensions within the feature maps through a factorized average pooling operation. This enhances the model's ability in identifying key areas within atmospheric fields, thus ensuring improved spatial precision in typhoon track forecasting. The model's efficacy was assessed with the 2020-2024 typhoon track dataset published by the China Meteorological Administration (CMA) and the ERA5 reanalysis product maintained by the European Centre for Medium-Range Weather Forecasts (ECMWF). The findings indicate how the two introduced enhancements to the model lead to higher precision for deep learning-based forecasts to a certain extent. (c) 2026 The Authors. Publishing services by Elsevier B.V. on behalf of KeAi Communications Co. Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

This study examines the combined effects of sudden stratospheric warming (SSW) events and El Ni & ntilde;o-Southern Oscillation (ENSO) on late winter (January-March) climate in China. It is found that SSWs significantly modulate El Ni & ntilde;o's impact over China. During El Ni & ntilde;o winters with SSWs, southern China tends to experience notable warming; conversely, without SSWs, the surface air temperatures there are generally colder. However, SSWs do not effectively alter La Ni & ntilde;a's impact; southern China remains anomalously colder during La Ni & ntilde;a winters even when SSWs occur. The net warming effect of SSWs over much of China during La Ni & ntilde;a, defined as the difference between scenarios with and without SSWs, is found considerably weaker than during El Ni & ntilde;o. These discrepancies arise from the contrasting tropospheric pathway of ENSO teleconnection over the East Asia-Pacific region. Specifically, El Ni & ntilde;o influences the occurrence of SSWs primarily by enhancing the tropospheric planetary wave 1. This enhancement corresponds to an emergence of a dipole pattern of geopotential height anomalies, characterized by an anomalous low over northeastern Eurasia and an anomalous high over China. In contrast, La Ni & ntilde;a contributes to SSW occurrences via an increase in the tropospheric wave 2, which corresponds to only an anomalous low over northeastern Eurasia. The consistently significant high over China both before and after SSWs during El Ni & ntilde;o, but much weakened or even absent during La Ni & ntilde;a, is mainly responsible for the stronger net warming in late winter over China in El Ni & ntilde;o years.

ABSTRACT The rapid and unprecedented warming of the Indian Ocean intensifies climate risks across socio‐economically vulnerable regions, highlighting the need of accurate prediction over the coming decade. This study first assesses the decadal predictability of the Indian Ocean Dipole (IOD) using the Decadal Climate Prediction Project (DCPP) models of Coupled Model Intercomparison Project Phase 6 (CMIP6), then develops a deep learning model based on a bidirectional gated recurrent unit (BiGRU) to enhance its predictive skill. The BiGRU model is designed to process multisource sequential data and is trained on the IOD index derived from DCPP simulations. Results show the MME achieves moderate skill on detrended IOD with an anomaly correlation coefficient (ACC) and mean squared skill score (MSSS) of 0.51 and 0.20 during 1963–2020, respectively. The BiGRU model improved the skill with an anomaly correlation coefficient (ACC) of 0.88 and a mean squared skill score (MSSS) of 0.72 during the testing period of 2009–2020, and accurately captured extreme IOD events. Based on the enhanced IOD index, we further used it to improve the prediction skill of the IOD‐related Australian rainfall, achieving ACC and MSSS values of 0.90 and 0.70 during 2009–2020, compared to 0.29 and −1.62 from the MME, underscoring the reliability of the BiGRU model for both IOD and related precipitation forecasts.

A significant wetting trend has been observed in the Yangtze–Huai River Basin (YHRB) during winter, suggesting an increasing risk of seasonal-scale extremely pluvial winters under global warming. Unlike previous studies focusing on daily precipitation extremes, this study investigates future changes in the total precipitation amount, frequency, and intensity of extremely pluvial winters in the YHRB under the SSP2-4.5 and SSP5-8.5 scenarios, based on simulations from 29 CMIP6 models. The total precipitation, frequency, and intensity of extremely pluvial winters are projected to increase substantially under both scenarios, with larger increases under SSP5-8.5 and for more severe events defined by the 95th percentile threshold. Under SSP2-4.5, for events defined by the 90th percentile threshold, the total precipitation may increase by about 50%–160%, the frequency by 20%–200%, and the intensity by 15%–35% relative to the historical period. The intensification of extremely pluvial winters is primarily associated with enhanced net moisture transport into the YHRB, driven by circulation changes and increased humidity gradients between Southwest China and the Indochina Peninsula. Projection uncertainty in the intensity of extremely pluvial winters mainly reflects the inter-model spread associated with differences in simulated circulation responses, moisture transport processes, and vertical motion changes. In contrast, the uncertainty in frequency projections remains considerable. These findings highlight an increasing risk of severe pluvial winters in the YHRB under continued warming and emphasize the importance of better constraining associated dynamic and thermodynamic processes in future climate projections.

Typhoon disaster research is hampered by data fragmented across incompatible spatial scales and temporal resolutions, hindering the development of data-driven risk modeling. This article presents a track-centric, multi-source dataset of western North Pacific typhoons spanning 2000–2024, comprising 756 tracks and 47,903 track points, organized around IBTrACS trajectories. For each track point, a storm-centered quasi-Lagrangian reference frame integrates ocean variables (Copernicus), socioeconomic dynamics (nighttime lights, population, and GDP-LitPop exposure), and hazard-loss metrics (ERA5 wind/rainfall, EM-DAT) into annual NetCDF products, simultaneously capturing storm evolution and local impact. Internal quality checks confirm numerical reproducibility and physical consistency, while external validation against GDIS event locations and IMERG satellite rainfall further corroborates spatial registration. The dataset supports typhoon risk modeling and impact attribution, as well as disaster chain analysis and spatiotemporal deep learning.