The impacts of spring frost on agricultural production and plant ecology have been frequently reported; however, understanding of the spatial and temporal distribution patterns and changes in trends of spring frost events on large scale remains limited. In this work, we explore the spatiotemporal variations of spring frost events in North America and Eurasia from 1948 to 2016. The frequency of spring frost has increased in North America, while it has decreased in Eurasia. The percentage of regions with increased spring frost is increasing in North America, whereas it is decreasing in Eurasia. It was also found that the frequency of cropland experiencing spring frost is rising in North America, whereas it is falling in Eurasia. The changes in the temporal and spatial distributions of spring frost events raise the question whether they are triggering changes in agriculture management and potential gains and losses as the world becomes warmer, especially in regions where the risk of spring frosts is increasing.

Accurate and reliable subseasonal precipitation forecasts are critical for disaster prevention and mitigation, particularly in densely populated regions like East Asia. However, substantial gaps remain between the reliability and accuracy of dynamical model forecasts and societal demands. This study proposes a machine learning-based adaptive bias correction (ABC) method to postprocess forecasts from the Climate Forecast System version 2 (CFS) and the European Centre for Medium-Range Weather Forecasts Integrated Forecasting System (EC). Results indicate that ABC effectively reduces systematic errors in 3–6 weeks lead forecasts. The method improved the precipitation prediction skills (based on uncentered anomaly correlation) of the CFS model over East Asia during 2018–2021 by 70

The Arctic Oscillation (AO) and the Interhemispheric Oscillation (IHO) are two important atmospheric variabilities that exert strong impacts individually on climate variations in the northern hemisphere (NH). Using the ERA5 reanalysis, we have investigated the combined influences of IHO and AO on NH surface air temperature anomalies (SATAs). Our results demonstrate that there indeed occur significantly strong interannual changes in wintertime circulations due to the combined influences. When IHO and AO indices are same signed, two major geopotential height anomaly (GHA) centers appear respectively over the north Atlantic and north Pacific besides of a large GHA center over Arctic, exhibiting “meridional-type teleconnection” patterns there. A “shoe-sole-like” GHA pattern appears in stratosphere where the planetary wave energy propagates downward to the bottom of troposphere. A SATA pattern with warmer (colder) Barents-northwest Eurasia against colder (warmer) Greenland emerge in years when both IHO and AO indices are positive (negative). However, when IHO and AO indices are oppositely signed, different anomalous circulation patterns emerge, exhibiting a “round-cornered triangle” GHA center over Arctic and three GHA centers in mid-to-high- latitudes. There appear more “zonal-type teleconnection” patterns in NH, including three probable “trans-Arctic teleconnections”. A SATA pattern characterized by “warmer (colder) Arctic against Colder (warmer) Eurasia” occurs in years when IHO index is positive (negative) and AO's negative (positive). All the results suggest that combined influences of IHO and AO may facilitate occurrences of extremely warm/cold events in wide regions of the NH during boreal winter.

Accurate solar energy resource assessment is essential for supporting sustainable development and reducing carbon emissions in China. China’s latest-generation geostationary meteorological satellite, Fengyun-4A (FY4A), provides an opportunity to obtain high-resolution and high-accuracy global horizontal irradiance (GHI) distribution maps over China. To fully explore the potential of FY4A, this study proposes a GHI estimation method for China based on hourly ground-based GHI data and FY4A satellite data, using a new interpretable deep learning model (RadNet). The hourly GHI estimation results are compared with other statistical estimation methods and GHI databases, demonstrating that the RadNet model exhibits significantly lower error and bias. Moreover, the RadNet model exhibits good spatial and temporal generalization capability. In terms of time, the RadNet model, trained only on GHI data from 2022, demonstrates high accuracy in 2021 and 2023. In terms of space, 65% of the test stations achieve high accuracy, with RMSE < 96 W/m2. The annual average GHI estimation results reveal that the RadNet model outperforms the reanalysis dataset and the remote sensing product in both resolution and accuracy. The RadNet model can provide a GHI resource distribution map of China with a 4 km × 4 km horizontal resolution.

Against the background of intensifying global climate change, extreme precipitation events have become increasingly frequent. Improving the accuracy of short-term precipitation nowcasting is therefore essential for disaster prevention and mitigation. Traditional numerical weather prediction (NWP) approaches are constrained by computational latency and errors arising from physical parameterizations, making it difficult to satisfy real-time forecasting requirements at high spatiotemporal resolution. Using the SEVIR dataset, this study conducts a systematic comparison of two Transformer-based deep learning models-Earthformer and LLMDiff-for short-term extreme precipitation nowcasting. Model performance is evaluated using the Critical Success Index (CSI), Probability of Detection (POD), and Success Ratio (SUCR). Results indicate that, for 0-30 min lead times, Earthformer more efficiently captures both local and long-range spatiotemporal dependencies via its Cuboid Attention mechanism and shows a slight advantage for low-intensity precipitation. As the lead time extends to 60 min, LLMDiff demonstrates stronger longer-horizon skill due to its diffusion-based probabilistic modeling and a frozen large language model (LLM) module, which enhance the representation of uncertainty and longer-term evolution of precipitation systems. However, LLMDiff tends to produce a higher false-alarm rate. Overall, Earthformer is better suited for rapid early warning of light precipitation, whereas LLMDiff is more appropriate for high-accuracy nowcasting of heavy precipitation, offering useful insights for intelligent forecasting of extreme weather.

Visibility is a critical meteorological factor for ensuring the safety of maritime and bridge transportation, and accurate identification of low-visibility levels is essential for early warning and operational scheduling. Traditional methods such as Random Forest often exhibit insufficient feature-modeling capability when dealing with high-dimensional, multi-source remote sensing data. Meanwhile, satellite observations used for visibility recognition are characterized by strong inter-channel correlations, complex nonlinear interactions, significant observational noise and outliers, and the scarcity of low-visibility samples that are easily confused with low clouds and haze. As a result, existing general deep learning methods (e.g., the Saint model) may still exhibit unstable attention weights and limited generalization under complex meteorological conditions. To address these limitations, this study constructs a visibility classification task for the Jiaxing–Shaoxing Cross-Sea Bridge region in China based on multi-channel visible and infrared spectral observations from the Fengyun-4A (FY-4A) and Fengyun-4B (FY-4B) satellites. We propose a visibility classification method using the LF-Transformer for the Jiaxing–Shaoxing Cross-Sea Bridge region in China, and systematically compare it with the Random Forest and Saint models. Experimental results show that the Precision of the LF-Transformer increases significantly from 0.47 (Random Forest) to 0.59, achieving a 13% improvement and demonstrating stronger discriminative ability and stability under complex meteorological conditions. Furthermore, a combination input of FY4A+FY4B outperform the single FY4A, with a 25.5% increased Macro F1-score. With an additional ensemble strategy, the LF-Transformer further improves its precision on the FY4A+FY4B fused dataset to 0.61, a 3% compared to the original LF-Transformer, indicating enhanced prediction stability. Overall, the proposed method substantially strengthens visibility classification performance and highlights the strong application potential of the LF-Transformer in remote-sensing-based meteorological tasks, particularly for low-visibility monitoring, early warning, and transportation safety assurance.

Since the initiation of the subseasonal-to-seasonal prediction project by the World Meteorological Organization,the accuracy of model forecasts has improved notably.However,substantial discrepancies have been observed among forecast results produced by different ensemble members when applied to South China.To enhance the accuracy of sub-seasonal forecasts in this region,it is essential to develop new methods that can effectively leverage multiple predictive models.This study introduces a weighted ensemble forecasting method based on online learning to improve forecast accuracy.We utilized ensemble forecasts from three models:the Integrated Forecasting System model from the European Centre for Medium-Range Weather Forecasts,the Climate Forecast System Version 2 model from the National Centers for Environmental Prediction,and the Beijing Climate Center-Climate Prediction System version 3 model from the China Meteorological Administration.The ensemble weights are trained using an online learning approach.The results indicate that the forecasts obtained through online learning outperform those of the original dynamical models.Compared to the simple ensemble results of the three models,the weighted ensemble model showed a stronger capability to capture temperature and precipitation patterns in South China.Therefore,this method has the potential to improve the accuracy of sub-seasonal forecasts in this region.

This study analyzes the record-breaking persistent heatwave-drought compound event that occurred in Jiangxi Province, China, from July to September 2022. The results indicate that this event was the most severe in the past 44 years, with heatwave days reaching 43 and cumulative precipitation in most areas being more than 60% below the average for the same period. The analysis suggests that the abnormal eastward extension and strengthening of the South Asian High (SAH), together with the abnormal westward extension and strengthening of the Western Pacific Subtropical High (WPSH), persistently and jointly dominated the Jiangxi region. This led to prevailing subsidence and warming throughout the atmospheric column, along with a divergent water vapor flux pattern, which suppressed the formation of precipitation. Further investigation reveals that this event is closely related to abnormally warmer sea surface temperature (SST) in the southwestern Pacific (with a correlation coefficient of 0.62). This may have triggered the Pacific-Japan (PJ) teleconnection wave train, thereby modulating the anomalies of SAH and WPSH, ultimately leading to the occurrence of this extreme compound event.

Low-visibility events, particularly dense fog, pose significant risks to navigation and transportation safety in coastal estuarine regions, making accurate and timely early warning systems essential. This study develops a lightweight low-visibility warning model based on the light gradient boosting machine (LightGBM) algorithm, using hourly meteorological observation data from the Qiantang River Estuary region forthe period 2021-2024.The forecast lead time was set to 3 h, and the model's performance was evaluated for predicting both low-visibility events (visibility <= 2,000 m) and fog events (visibility <= 500 m), with interpretability analysis conducted using Shapley additive explanations (SHAP). The results show that: (i) Validation against actual low-visibility events confirms that the model provided effective warnings across the study area, achieving an average accuracy of 98.6% for low-visibility events. (ii) The original LightGBM model requires parameter optimization to handle imbalanced classification, particularly for rare fog events. By adjusting class weights, the false-negative rate for dense fog was effectively reduced, improving recall from 44% to 70%. (iii) Global SHAP analysis revealed that relative humidity is the meteorological factor contributing most to dense fog warnings. Sample characteristics such as low wind speed, high humidity, and a small air-ground temperature difference consistently contribute to the model's prediction of dense fog events.

With the emergence of the 'Arctic amplification' phenomenon in recent years, vegetation in the circum-Arctic region has exhibited significant greening trends, while summer Arctic cyclones have also shown a notable increase in both frequency and intensity. To explore the potential relationship between Arctic vegetation changes-particularly in the northern margin of the Eurasian continent (NME)-and the intensification of Arctic cyclone activity under the backdrop of accelerated Arctic warming, this study investigates the impact of vegetation greening on the intensity of summer Arctic cyclones in this region. Using GIMSS 3G+ NDVI data from 1982 to 2022, combined with the WRF numerical weather model, the analysis reveals that, compared to the 1980s, tundra regions such as the central and eastern parts of the NME experienced the most pronounced increase in vegetation, corresponding to significant warming in these areas. As temperatures rose markedly, the north-south land-sea temperature gradient in the NME intensified, enhancing atmospheric baroclinicity in coastal regions. Consequently, cyclone intensity increased accordingly. The rise in the Green Vegetation Fraction (GVF) led to an increase in the Leaf Area Index (LAI) while reducing surface albedo, allowing the surface to absorb more shortwave solar radiation. This process plays a critical role in driving the rise in near-surface temperatures and the development of cyclones.

In recent decades, there has been a notable increase in the frequency and severity of drought events in Southwest China (SWC), which have significantly impacted agriculture and the social economy. Using sea surface temperature (SST) data from the Hadley Centre, daily meteorological drought composite index grid data and ERA5 reanalysis data, we investigate the characteristics of autumn drought in SWC and discuss its possible causes using empirical orthogonal function (EOF) analysis. The results indicate a distinct 'Northeast-Southwest' dipole pattern of autumn drought over SWC in the past decade, which we define as the 'Chuan-Yu'-type drought. A British-Okhotsk Corridor (BOC) pattern Rossby wave train, presenting over Eurasia, is identified as the key factor influencing the 'Chuan-Yu'-type drought in SWC during the autumn. The BOC-pattern Rossby wave train not only obstructs water vapour transport channels in SWC but also induces anomalous descending motions in the lower to middle troposphere over the region, leading to high temperatures and insufficient precipitation. Further investigation reveals that North Atlantic dipole SST anomalies trigger the BOC-pattern Rossby wave train in autumn. The results of the linear baroclinic model sensitivity simulations support the above conclusion. These findings will make valuable contributions to autumn drought prevention and mitigation in SWC.

The cloud system characteristics within satellite cloud imagery play a crucial role in the meteorological operational analysis of cold fronts, and integrating satellite cloud imagery into automated frontal identification schemes can provide valuable insights for accurately determining the position and morphology of cold fronts. This study introduces Cloud-DETR, a deep learning identification method that uses the DETR model with satellite cloud imagery, to identify cold fronts from extensive datasets. In the Cloud-DETR method, preprocessed satellite cloud imagery is used to generate training images, which are then put into the DETR model for cold front identification, achieving excellent results. The alignment between the Cloud-DETR cold fronts and weather systems during continuous periods and extreme weather events is assessed. The Cloud-DETR method exhibits high accuracy in both the position and morphology of cold fronts, ensuring stable identification performance. The high matching rate between the Cloud-DETR cold fronts and the manually identified ones in the test set, image dataset and labels from 2017 is verified. This indicates that the Cloud-DETR method can provide an accurate cold fronts dataset. The cold fronts dataset from 2005 to 2023 was obtained using the Cloud-DETR method. It was found that over the past 18 years, the frequency of cold fronts displays distinct seasonal patterns, with the highest occurrences observed during winter, particularly along the mid-latitude storm tracks extending from the east coast of East Asia to the Northwest Pacific. The methodology and findings presented in this study could help advance further research on the characteristics of cold front cloud systems based on long-term datasets.

In February 2022, a persistent low-temperature rain and snow event (LRSE) occurred in the central Pan-Pearl River Delta (CPPRD) region of southern China, causing severe damage and economic losses. During the LRSE, both the temperature and precipitation fields exhibited quasi-biweekly oscillation (QBWO) signals over the CPPRD region. Circulation analysis revealed that the eastward propagation of Rossby waves at mid–high latitudes enhanced the Baikal blocking high and the Mongolian high, facilitating the continuous southward migration of cold air. The strengthening India–Burma trough (i.e., the southern branch trough) brought abundant warm and humid airflow, converging with cold air from the north in the CPPRD region. Moreover, deep convective activity originating in the northern Indian Ocean became exceptionally active, propagating to southern China and providing dynamic lifting conditions for precipitation in the study region. The combined effects of tropical and extratropical weather systems resulted in the LRSE occurrence. Partial lateral forcing (PLF) experiments were performed to quantify the contributions of the QBWO signals from different boundaries of the region. The extratropical QBWO signal from the northern boundary led to a temperature decrease of 1.61°C, with 77.83

Global warming has accelerated Arctic warming and vegetation expansion. Concurrently, summer Arctic cyclones pose increasing threats to shipping and ecosystems. Here, we demonstrate that enhanced vegetation growth (“Arctic greening”) since 2000 has significantly contributed to the intensification of summer Arctic cyclones. Using satellite-derived vegetation indices and atmospheric reanalysis data (1982–2022), we show that both cyclone intensity and Arctic vegetation greenness have exhibited significant positive trends post-2000. This greening has reduced land surface albedo by approximately 5

Seven extreme marine heatwave (MHW) events that occurred in the central–eastern tropical Pacific over the past four decades are divided into high-(MHW#1 and #2), moderate-(MHW#3–5), and low-predictive (MHW#6 and #7) categories based on the accuracy of the 30–60d forecast by the Nanjing University of Information Science and Technology Climate Forecast System (NUIST CFS1.1). By focusing on high- and low-predictive MHWs, we found that metrics indicative of strong and severe warming (S > 2 and S > 3, where S is MHW severity index) pose greater challenges for accurate forecasting, with the biggest disparity observed for S > 2. All events are intertwined with the El Niño–Southern Oscillation (ENSO), yet a robust ENSO forecast does not guarantee a good MHW forecast. Heat budget analysis within the surface mixed layer during the rapid warming periods revealed that the moderate and severe warming in MHW#1, #2, #6 are primarily caused by heat convergence due to advection (Adv), whereas MHW#7 is mainly driven by air–sea heat flux into the sea surface (Q). The NUIST CFS1.1 model better captures Adv than Q. High-predictive events exhibit a greater contribution from Adv, especially the zonal component associated with the zonal gradient of sea surface temperature anomalies, which may explain their higher sub-seasonal forecast skills.

Extended-range forecast has long maintained a difficult point for the seamless forecast system due to the lack of predictability, with intraseasonal oscillation (ISO), an important signal in many high-impact weather events, being an important source of that. To improve the accuracy of ISO extended-range forecast and make up the gaps in previous researches in this regard, a data-driven model ISOX is proposed for the intraseasonal components of atmospheric fields. Compared with the subseasonal forecast results from climate forecast system (CFS), and the climatological forecast, ISOX achieves higher accuracy for lead times longer than 13 days, with few spatial or temporal weak points. It also performed better in predicting the positive 2 m temperature ISO and lower tropospheric conditions in a heatwave event, surpassing CFS for lead times longer than 13 days. Finally, through gradient evaluation, the model is proved to be able to study the ISO signal movements of atmospheric systems. Thus, the success of this model may shed light on improving extended-range forecast skills and assist the timely detection and prevention of possible meteorological disasters.

Arctic cyclone activity is an important component of the local climate, and the frequent occurrence of extreme summer storms has raised widespread scientific interest. In this paper, we investigated the distinctive structural characteristics of intense summer Arctic cyclones by utilizing ERA-Interim reanalysis data and employing a deep learning algorithm for cyclone detection. We found that the northern edge of Eurasia (i.e., the marginal ice zone (MIZ)) and the Alpha Ridge of Arctic Ocean (AR, i.e. central Arctic) are the two most active regions for intense Arctic cyclone activities in summer (from June to September). However, the surface conditions and coupling frequency between surface cyclone and tropopause polar vortices (TPVs) are distinct over these two regions. By further analysis of 100 intense cyclone activities in these two areas, respectively, we found that cyclones in MIZ are often smaller in size but higher in intensity at their maximum intensity, and their life cycles are generally shorter. MIZ cyclones are typically accompanied by a large Eady growth rate and frontal structure in the lower troposphere and their intensification primarily attributed to the thermal-baroclinic process. In contrast, cyclones in AR are more frequently associated with higher potential vorticity (PV) values and pronounced PV downward intrusion from the stratosphere, as well as notable "upper warm-lower cold" structures. The downward intrusion of TPVs and stratosphere vortices contribute to a decrease in the upper and column air mass deficit, leading to the intensification of surface Arctic cyclones in these regions. In this study, we researched intense summer storms in the Arctic. We found that there are two main areas where these storms occur: the marginal ice zone (MIZ) near Eurasia and the Alpha Ridge (AR) in the central Arctic. However, storms in these two areas have different characteristics. In the MIZ, the storms are smaller but stronger, and they do not last as long. They are mainly driven by instability in the lower troposphere. On the other hand, the storms in AR respond more to the downward intrusion of potential vorticity from the stratosphere. These storms have a unique structure where the upper air is warmer than their surroundings, and the lower air is colder than their surroundings, especially in AR. This structure makes them more intense and longer-lasting. Exploring these differences helps us understand how Arctic storms work, and how they might be affected by climate change. Both the marginal ice zone (MIZ) and Alpha Ridge exhibit active summer Arctic cyclone activities, especially for intense storms In the MIZ, baroclinic instability plays a more prominent role in the intensification and maintenance of cyclones Cyclones in Alpha Ridge are more commonly accompanied by potential vorticity downward intrusion, and "upper warm-lower cold" structures

In recent decades, the atmospheric moisture capacity has increased globally in concert with global warming, with a particularly notable warming trend in Arctic regions. However, due to limited observational data, the variation and causes of polar precipitation, especially large-scale precipitation events associated with Arctic cyclones, remain unclear. In this paper, GPM satellite data are compared with ERA5 reanalysis data to explore the characteristics of summer precipitation at the northern margin of the Eurasian region (NMER) and the influence of cyclone activity on precipitation. It is revealed that high precipitation values in the Arctic region, as indicated by the GPM and ERA5 data, are mainly concentrated at the NMER. However, the GPM data show an overall larger precipitation amount, while the station observations more closely agree with the ERA5 precipitation changes at the NMER. The cyclone identification results indicate that summer cyclones at the NMER are mainly distributed in the Barents, Kara and Laptev Seas, and the precipitation contribution rate of ERA5-derived cyclones is 37.35%, which is significantly higher than that of GPM-derived cyclones (29.47%). Furthermore, high cyclone activity results in more intense precipitation, with the top 5% of the strongest cyclones contributing 60% (GPM) and 40% (ERA5) to the total cyclonic precipitation.

Few-shot segmentation was proposed to obtain segmentation results for a image with an unseen class by referring to a few labeled samples. However, due to the limited number of samples, many few-shot segmentation models suffer from poor generalization. Prototypical network-based few-shot segmentation still has issues with spatial inconsistency and prototype bias. Since the target class has different appearance in each image, some specific features in the prototypes generated from the support image and its mask do not accurately reflect the generalized features of the target class. To address the support prototype consistency issue, we put forward two modules: Data Augmentation Self-knowledge Distillation (DASKD) and Prototype-wise Regularization (PWR). The DASKD module focuses on enhancing spatial consistency by using data augmentation and self-knowledge distillation. Self-knowledge distillation helps the model acquire generalized features of the target class and learn hidden knowledge from the support images. The PWR module focuses on obtaining a more representative support prototype by conducting prototype-level loss to obtain support prototypes closer to the category center. Broad evaluation experiments on PASCAL-5i and COCO-20i demonstrate that our model outperforms the prior works on few-shot segmentation. Our approach surpasses the state of the art by 7.5% in PASCAL-5i and 4.2% in COCO-20i.