Forest fires have a significant impact on terrestrial ecosystems, leading to harm to biodiversity and environment. To mitigate the ecological damage caused by forest fires, it was necessary to develop prediction models of fire risk. In this study, by evolving the optimal architecture and parameters using the particle swarm optimization (PSO) algorithm, a convolutional neural network (CNN) deep learning model was proposed to predict forest fire risk on a national scale. Utilizing fire data and fire risk factors from 2001 to 2020 in China, the PSO-CNN-based deep learning model (PSO-CNN) was utilized and tested. Compared to logistic regression, random forest, support vector machine, k-nearest neighbors, and CNN models, the PSO-CNN model exhibited superior performance with an accuracy of 82.2% and an AUC value of 0.92. These results clearly highlighted the effectiveness of the PSO-CNN model in enhancing the accuracy of forest fire prediction. Furthermore, the forest fire risk prediction level estimated by the proposed model on a national scale for the entire country was mostly consistent with actual fire data distribution, indicating its potential to be used as an important direction for deep learning in forest fire prediction research.

From 10 July to 12 July 2021, a long-lived (~66 h) southwest vortex (SWV), moved from Southwest China to Northeast China and caused a series of heavy rainfall events. This SWV case was rarely seen, as its lifespan was much longer than the SWVs’ mean lifespan, and the vast majority of SWVs showed a quasi-stationary behavior. It was found that the SWV formed and sustained in favorable background environments, which were characterized by a strong upper-level divergence (related to the South Asia High), a notable middle-tropospheric warm advection (related to a shortwave trough), and a vigorous low-level jet. The SWV showed remarkable interactions with a middle-tropospheric mesoscale vortex. The strong southwesterly wind in the eastern section of a deep shortwave trough east of the Tibetan Plateau acted as the steering flow for the northeastward movement of both vortices. Vorticity budget showed that the convergence-related vertical stretching dominated the SWV’s formation and development; the convection-related upward transport of cyclonic vorticity was the most favorable factor for the SWV’s sustainment, whereas, during the decaying stage, the SWV dissipated mainly due to the tilting effects and the net export transport of cyclonic vorticity. Backward trajectory analyses showed that most of the air particles that formed the SWV (at its formation time) were sourced from the lower troposphere. These air particles mainly ascended and experienced a rapid increase in cyclonic vorticity during the SWV’s formation stage. The topography of the Yunnan–Guizhou Plateau was crucial for the SWV’s formation, as around a half of the air particles (that formed the SWV) came from this region. Most of these air particles enhanced in their cyclonic vorticity and convergence when they descended along the topography of the plateau.

Sichuan is a high-incidence area of thunderstorm activity in China. Based on the data of the total lightning location system from 2018 to 2022, the total lightning, cloud-to-ground (CG) lightning, and intracloud (IC) lightning activity regularity for the Sichuan province (SC) and its three climate subregions: Sichuan Basin (SB), Panxi district (PD), and West Sichuan Plateau (WSP) are analyzed, and the influences of different climate and topography conditions on lightning activities are also discussed. The results show that (1) for the whole province, the annual mean value of total lightning is about 850 thousand. The SB has the most lightning occurrences, and the WSP has the largest IC and +CG proportion. The southwest of PD, the north-center of PD, and the southeast of SB are the three high-value centers of lightning density. (2) For SB, the better thermodynamic and moisture conditions account for the most lightning occurrences. For PD, the lightning distribution is attributed to the joint effect of specific meteorological conditions and mountainous topography. For WSP, the convections are weak and shallow, which lead to high IC and +CG proportion. (3) The IC lightning mainly occurs below 12 km, and the IC height is much lower on WSP. The spatial and seasonal variation of IC height corresponds well to the cloud base height (CBH) and cloud top height (CTH). (4) The seasonal lightning frequency distribution in the three regions is similar, but the diurnal variation is quite different. The lightning activity mainly occurs between 1400 and 2200 LT on WSP but shows obvious nocturnal in SB. (5) Most CG intensity concentrates in the range below 50 kA, and IC concentrates in the range below 75 kA.

Radar echo extrapolation is a key method for weather nowcasting. Deep learning has become a hotspot of radar echo extrapolation, but it suffers from problems such as prediction fuzziness and low prediction metrics, mainly due to two reasons: firstly, existing deep learning methods still have insufficient capabilities in learning spatial–temporal features and focus on critical regions. Second, atmospheric motion is chaotic and uncertain, and only the confidence interval of the future atmospheric state can be predicted. Existing pixel‐level loss functions such as MSE loss require the predicted values to be completely consistent with the gold standard values numerically, resulting in the model being unable to converge to the optimal solution. To address these issues, we propose the Spatial–Temporal‐Attention Dual‐path Extrapolation Network (STADEN) model and a novel multithreshold loss function (MTL). STADEN introduced a dual‐branch network based on the inception and metaformer to enhance the ability to learn spatial–temporal information. MTL allows predicted values to fall within a ground‐truth interval to achieve the minimum loss value, enhancing model robustness. In the 2‐h extrapolation experiment on the Sichuan dataset, STADEN achieved critical success index (CSI) metrics of 0.488, 0.388, and 0.211 at thresholds of 15, 25, and 35, respectively, outperforming all comparative methods. MTL were applied to all comparative methods directly, and improve CSI metrics by 0.078, 0.058, and 0.024. Robustness experiments on the HKO‐7 dataset further demonstrate that both STADEN and MTL achieve optimal performance. These results highlight the potential of STADEN and MTL for practical applications in radar echo prediction.