Xiaofeng Li

Sudden tropical cyclone (TC) track turns remain a major source of short range forecast error over the western North Pacific because small environmental differences can shift both the turning time and the turning location, rapidly amplifying cross track displacement. Here we examine whether these cases are better understood as a branch‐sensitive predictability grouping rather than as generic high‐error samples. To examine this grouping, we use a short range TC track forecasting framework that combines a mixture of experts design, used to represent multiple plausible track branches, with a steering‐sensitive cross track loss term that emphasizes directional errors during turning conditions. The model uses TC history, TC centered ERA5 environmental fields, and infrared imagery, and is trained for 6–24 hr displacement forecasting using 2000–2020 data and evaluated on independent cases from 2021 to 2024. For the full sample, forecast skill remains comparable to the baseline models, whereas the largest gains occur in sudden turn cases. In these cases, the improvement is more evident in turn synchronized diagnostics than in endpoint error alone, indicating that the gain comes mainly from better representation of positional mismatch. Counterfactual diagnostics further show that the improvement is more strongly associated with pressure level environmental fields than with single level surface fields. These results support the interpretation that sudden TC turns constitute a steering sensitive, branch‐dependent forecast grouping whose short range predictability is more strongly associated with mid to upper level circulation reorganization than with slowly varying lower‐boundary conditions.

Antarctic mass loss has been a major contributor to global sea-level rise for most of the last few decades, mainly driven by West Antarctica ¹ . During 2021–2023, however, a sharp increase in surface mass balance over Queen Mary Land and Wilkes Land in East Antarctica offset West Antarctic loss and slowed the rate of total ice mass loss ^2,3 . Although this slowdown is consistent with the expected long-term precipitation response to global warming through poleward-shifted storm tracks and Antarctic moistening ⁴ , our results point to a different mechanism. Here we show that the recent ice mass gain was linked to a recurrent atmospheric teleconnection driven by sea surface temperature anomalies in the tropical warm pool, which experienced unusually persistent warming from 2021 to 2023 relative to the previous two decades. On the basis of observations and model experiments, we find that tropical warm pool warming excites a poleward-propagating Rossby-wave train that induces a high-pressure anomaly over East Antarctica, enhancing Queen Mary Land and Wilkes Land precipitation and driving the observed mass gain, with moisture primarily sourced from the mid-latitude Indian Ocean. Similar multiyear warming in the tropical warm pool recurs about once per decade in observations and historical simulations, and its influence on precipitation is distinct from the effects of global warming. Therefore, the recent Antarctic Ice Sheet mass gain is probably temporary and does not yet reflect a sustained, global-warming-driven moistening of Antarctica.

Rapid and reliable coastal inundation mapping from Synthetic Aperture Radar (SAR) imagery is vital for disaster response. Although recent deep learning advances have enhanced this task's accuracy and automation, two major challenges remain: (a) geographic and imaging variations across flood events limit the generalizability of supervised models trained on static data sets, and (b) constructing large‐scale annotated flood data sets remains costly and time‐consuming. These challenges are especially pronounced in coastal regions, where precise segmentation is critical but conventional models often struggle to adapt to unseen scenarios, leading to degraded performance. To address this, we propose Coastal Inundation Mapping via Test‐Time Domain Adaptation (CIM‐TTDA) for SAR‐based coastal flood mapping. Instead of requiring new labels, CIM‐TTDA adapts a pretrained water segmentation model to each unseen event using only unlabeled post‐event SAR imagery. The core of CIM‐TTDA is a self‐supervised learning strategy tailored for coastal inundation mapping. Built upon pseudo‐label learning and exponential moving average updates, the strategy ensures stable adaptation across diverse flood events while maintaining computational efficiency. We evaluate CIM‐TTDA on three cyclone‐induced flood events (eight distinct scenes) and conduct five comprehensive experiments to assess its effectiveness, robustness, and efficiency. Results show that CIM‐TTDA improves pixel‐level mean Intersection over Union (IoU) by approximately 0.15 over conventional supervised models across various convolutional‐ and Transformer‐based architectures. With the EfficientNet‐B0 architecture as the pretrained model, CIM‐TTDA achieves a mean IoU of 0.803, outperforming supervised baselines and state‐of‐the‐art test‐time domain adaptation methods. Overall, CIM‐TTDA offers a promising solution for accurate and stable coastal inundation mapping in unseen real‐world scenarios.

Internal solitary waves (ISWs) are powerful underwater disturbances that can vertically displace water by more than 200 meters, driving ocean mixing and affecting offshore engineering. Yet, their global distributions and vertical imprints remain poorly characterized. Traditional satellite sensors are constrained by cloud cover or limited sampling. The Surface Water and Ocean Topography (SWOT) mission provides near-continuous, cloud-independent ocean observations at unprecedented resolution. Here, we develop an artificial intelligence framework to automatically detect ISWs worldwide. Using 29 SWOT orbital cycles (July 2023 to March 2025), this study reveals a global map of ISW distribution and amplitude variations, identifying energetic hot spots and distinct seasonal and tidal variations. These findings provide insights into ISW dynamics and offer a foundation for assessing their role in global energy redistribution and ocean-climate interactions.

Precipitation nowcasting refers to the high‐resolution forecasting of rainfall and hydrometeors within 0–6 hr according to the official definition of the World Meteorological Organization, which has relied on numerical models for decades. Recently, artificial intelligence (AI) has shown promise in addressing precipitation nowcasting. However, three key scientific issues have not been studied: (a) the researcher does not know what semantic knowledge the AI model has learned in the high‐dimensional space that influences precipitation forecasts. (b) If the learned semantic knowledge can be quantified, which aspects of precipitation does each factor control? And (c) can the semantic knowledge contribute to improving the accuracy of precipitation nowcasting? Hence, this study proposes a human supervised spatial‐temporal disentanglement model (STNet) that perturbs high‐dimensional vectors in the latent space and enforces orthogonality constraints to disentangle the spatial‐temporal semantic knowledge learned by the model—referred to as disentangled latent factors (a set of orthogonal high‐dimensional vectors). Qualitative and quantitative experiments on the specialized nowcasting dataset (SEVIR) reveal that the disentangled latent factors primarily fall into two categories: temporal factors, which predominantly control the evolving precipitation pattern (dynamics), and spatial factors, which mainly regulate precipitation intensity (static). Specifically, the temporal factor constraint provides more significant improvements in precipitation area forecasting. The spatial factor constraint, on the other hand, focuses on optimizing precipitation intensity.

Coastal tropical cyclones are high-impact events whose nearshore wind structures are difficult to observe with conventional observing systems alone. Available SAR wind products are commonly distributed as scene-level products rather than organized cyclone-focused event records. This fragmentation limits reproducibility, makes quality screening repetitive, and obscures the relationship among storm identity, scene geometry, coastal context, and derived wind fields. We present TC-SARCoast, a global event-based Sentinel-1 Level-1 Interferometric Wide Swath (IW) SAR dataset for coastal tropical cyclones from 2015 to 2026. The current release contains 277 tropical-cyclone (TC) events, 411 event-track directories, and 3128 SAR scenes. The dataset is organized into four traceable layers: L0 scene provenance and source linkage, L1 SAR–IBTrACS matchup records, L2 scene-level statistics and hierarchical scene-usability classes, and L3 value-added wind-product files. We develop an event-track-scene data model that preserves provenance, storm context, coastal information, scene usability, and product diagnostics in one hierarchy. Dataset assessment includes completeness and consistency assessment, spatiotemporal coverage analysis, scene-quality statistics, external validation against SFMR, NDBC, and SMAP, and wind-regime-dependent comparison with ECMWF fields. Validation against SFMR aircraft observations demonstrates that the released approximately 1 km L3 wind product captures tropical-cyclone wind fields with high matchup consistency, achieving an RMSE of 3.97 m s−1, bias of -0.36 m s−1, and correlation of 0.82 for the quality-screened matchup sample. TC-SAR-Coast advances SAR tropical- cyclone data use from isolated scene inspection to reproducible event-based analysis by integrating source provenance, storm context, coastal exposure information, quality diagnostics, validation records, and value-added wind fields in a single open archive.

ENSO teleconnections are a key source of Antarctic sea‐ice predictability, particularly in the Antarctic Dipole (ADP) region, but whether this predictability remains stable under the recent transition toward Central Pacific (CP) El Niño is unclear. Here, using a Markov model, we reveal an 83% decrease in sea ice concentration (SIC) predictability, measured by the anomaly correlation coefficient (ACC), during austral winter and spring since 2002. We show that this decline is linked to a weakened ENSO–Antarctic teleconnection, in which the CP El Niño‐triggered Rossby wave train shifts northward away from the Antarctic sea‐ice zone and toward the open ocean. This northward shift reduces the wave train's influence on Antarctic sea ice, dampens sea‐ice persistence, lowers its signal‐to‐noise ratio, and consequently limits predictability. Our results suggest that ENSO–Antarctic teleconnections may provide a useful benchmark for evaluating and improving Antarctic sea‐ice prediction models.

Tropical instability waves (TIWs) generate mesoscale sea surface temperature (SST) fluctuations in the eastern equatorial Pacific and influence the evolution of the El Niño‐Southern Oscillation (ENSO). Yet satellite‐based prediction of TIW‐related SST anomalies remains largely SST‐centric and makes limited use of sea surface salinity (SSS) observations. Here we show that satellite SSS provides a practically useful constraint on TIW‐related SST forecasts. We develop a dual‐branch deep learning framework trained on Copernicus Marine Service SST and SSS products, in which a ConvLSTM‐UNet3D trend branch represents slowly varying evolution and a frame‐difference branch highlights transient TIW disturbances. A Mean Deviation Loss strengthens learning in regions of strong anomalies, including the TIW triangular zone and the cold tongue. SST is the sole prognostic output, while SSS is prescribed at each rollout step under three protocols with distinct operational meanings: a deployable, leakage‐free SST‐recursive forecast using training‐derived monthly SSS climatology; a diagnostic upper bound using observed SSS at the corresponding forecast dates; and a constant‐field Zero‐map stress test. Compared with the SST‐only configuration, incorporating time‐varying SSS reduces the one‐step 5‐day SST RMSE from 0.30 to 0.27°C, while the deployable climatological‐SSS setting attains 0.28°C, within 0.01°C of the observed‐SSS upper bound at the single‐step horizon and closely tracking it under multi‐step rollouts. Cross‐spectral and time‐lagged regression analyses further reveal that coherent leading SSSA modes can lead SSTA by 5–12 days on TIW timescales, indicating that seasonal‐background and event‐scale SSS carry physically meaningful, complementary predictive information for TIW‐related SST evolution.