The Longmenshan fault zone in southwest China has produced two large earthquakes within the last few years: the Wenchuan earthquake with magnitude scale (Ms) 8.0 on May 12, 2008, and the Lushan earthquake with Ms 7.0 on April 20, 2013. In this paper, we investigate the Coulomb stress changes, the data from the strainmeters and pendulum tiltmeter which detected stress changes before the Wenchuan and Lushan earthquakes, and the anomalies in the outgoing longwave radiation (OLR) data observed by National Oceanic and Atmospheric Administration Satellites. We do this using our proposed new change detection method, tested through examination of ten strong earthquakes in China, which can effectively reveal changes in OLR data before and after earthquakes. Our investigation suggests that the positive coseismic Coulomb stresses of Wenchuan may have increased the probability of the earthquake occurring. The pre-earthquake synchronous increments in stresses detected by strainmeters and the pendulum tiltmeter in several seconds can be viewed as slow earthquakes and thus may have triggered the Wenchuan and Lushan earthquakes. Our study indicates that there are larger OLR anomalies in fault zones than nonfault zones and those larger anomalies were generated before the Wenchuan than Lushan earthquakes. There were also larger increments in vertical pendulum tiltmeter and four-component borehole strains before the Wenchuan earthquake than before the Lushan earthquake. Our results suggest that larger stress changes may trigger larger earthquakes and cause larger anomalies in OLR data.

This paper discusses the sedimentary environment and source supply in the sedimentary area and their coupling relationship through a detailed description of the deposits from the Ziliujing Formation to Shaximiao Formation of the Lower–Middle Jurassic on the Tieshan section in the Dazhou City, northeastern Sichuan Province, through sedimentary characterization, determination of the paleocurrent direction, analysis of heavy minerals, and detrital zircon U–Pb dating. The results show that the Zhenzhuchong Member is sufficiently supplied with detrital sediments and is dominantly composed of fluvial-delta deposits. The Dongyuemiao Member—the first member of the Lianggaoshan Formation—is dominated by lacustrine deposits, with the detrital supply increased initially in the early Lianggaoshan. The second member of the Lianggaoshan Formation suggests a significant increase in detrital supply, with shrunken lake basin and changed paleocurrent direction. The Shaximiao Formation reveals a complete disappearance of the lake basin in the northern Sichuan Basin. The comprehensive analysis on source supply indicates that the change in source property is apparently coupled with the change in the sedimentary environment, both controlled by orogenesis around the basin. It is inferred from the zircon age distribution that the changes in the sedimentary environment and source supply during the late Ziliujing period and the middle and late Lianggaoshan period resulted from the uplifting of the Micangshan Mountain, Dabashan Mountain, and Qinling Mountain, respectively.

Brain–computer interfaces, where motor imagery electroencephalography (EEG) signals are transformed into control commands, offer a promising solution for enhancing the standard of living for disabled individuals. However, the performance of EEG classification has been limited in most studies due to a lack of attention to the complementary information inherent at different temporal scales. Additionally, significant inter-subject variability in sensitivity to biological motion poses another critical challenge in achieving accurate EEG classification in a subject-dependent manner. To address these challenges, we propose a novel machine learning framework combining multi-scale feature fusion, which captures global and local spatial information from different-sized EEG segmentations, and adaptive Lasso-based feature selection, a mechanism for adaptively retaining informative subject-dependent features and discarding irrelevant ones. Experimental results on multiple public benchmark datasets revealed substantial improvements in EEG classification, achieving rates of 81.36%, 75.90%, and 68.30% for the BCIC-IV-2a, SMR-BCI, and OpenBMI datasets, respectively. These results not only surpassed existing methodologies but also underscored the effectiveness of our approach in overcoming specific challenges in EEG classification. Ablation studies further confirmed the efficacy of both the multi-scale feature analysis and adaptive selection mechanisms. This framework marks a significant advancement in the decoding of motor imagery EEG signals, positioning it for practical applications in real-world BCIs.

The decoding of motor imagery (MI) in electroencephalogram (EEG) relies on the in-depth exploration of its spatiotemporal features. However, existing methods have struggled to capture the spatiotemporal dynamics of brain networks, lacking mechanisms for noise-robust extraction of key spatial features from multiple perspectives and failing to adequately model spatiotemporal interaction relationships. To address these challenges, this study proposes MBDSTGAT, a multibranch dynamic spatiotemporal graph attention network (GAT) with topology perturbation. The proposed network integrates three core components: a multimode topology perturbation (MTP) mechanism, a multibranch dynamic strong-connection attention (MBDSCA) fusion structure, and a segmented spatiotemporal graph (SSTG) modeling strategy. The MTP mechanism enhances model robustness through three perturbation modes, namely edge, edge weight, and hybrid, which simulate the variability of brain connectivity. The MBDSCA module employs multiple branches with distinct parameter matrices to extract multiperspective spatial features, where each branch uses an annealing-based Top-K neighbor selection strategy embedded in the message-passing process of the GAT to dynamically emphasize salient and strongly connected neural interactions. Finally, the SSTG strategy integrates spatial features and temporal segments into a unified graph, in which a spatiotemporal GAT models cross-scale dependencies to refine discriminative representations. In large-scale cross-session and cross-subject experiments, the proposed network achieved average accuracies of 97.76% and 82.34%, 86.57% and 67.50%, and 98.75% and 87.55% on the High Gamma dataset (HGD), BCIC-IV-2a, and self-collected VR-MI datasets, respectively, outperforming existing methods and reaching state-of-the-art levels. In conclusion, MBDSTGAT effectively models the spatiotemporal dynamics of EEG signals and demonstrates superior MI decoding performance. The code and self-collected dataset for this study are openly available at https://github.com/shanmaojimu/MBDSTGAT