Improving the prediction skill of El Niño-Southern Oscillation (ENSO) is of critical importance for society. Over the past half-century, significant improvements have been made in ENSO prediction. Recent studies have shown that deep learning (DL) models can substantially improve the prediction skill of ENSO compared to individual dynamical models. However, effectively integrating the strengths of both DL and dynamical models to further improve ENSO prediction skill remains a critical topic for in-depth investigations. Here, we show that these DL forecasts, including those using the Convolutional Neural Networks and 3D-Geoformer, offer comparable ENSO forecast skill to dynamical forecasts that are based on the dynamic-model mean. More importantly, we introduce a combined dynamical-DL forecast, an approach that integrates DL forecasts with dynamical model forecasts. Two distinct combined dynamical-DL strategies are proposed, both of which significantly outperform individual DL or dynamical forecasts. Our findings suggest the skill of ENSO prediction can be further improved for a range of lead times, with potentially far-reaching implications for climate forecasting.

El Niño-Southern Oscillation (ENSO) asymmetry in predictability on springtime initial condition remains unclear. From the perspective of the spring predictability barrier (SPB), this paper investigates the ENSO asymmetry in SPB and explores the potential factors that may lead to this asymmetry. Both the observation and 29 Coupled Model Intercomparison Project Phase 6 (CMIP6) models show that the spring sea surface temperature (SST) persistence is significantly higher in El Niño years than that in La Niña years, and the SPB intensity is stronger in La Niña years than that in El Niño years. Through the recharge oscillator model, observation and CMIP6 models, we demonstrate that the nonlinear wind stress response to SST anomalies in spring is the main cause of the asymmetric SPB intensity. By the mixed-layer heat budget of the tropical Pacific in the spring, we further identify that a stronger response of zonal wind stress in El Niño events can cause a stronger zonal advection feedback, which finally leads to a weaker SPB and enhances the predictability of El Niño. In contrast, the cooling SST in the spring only leads to weak easterly anomalies, the zonal advection feedback is relatively weaker, thus SPB is stronger and the predictability of La Niña is lower. From the perspective of SPB, we suggest that El Niño is more predictable than La Niña.

Shadow and layover are geometric distortion phenomenons in side-view imaging synthetic aperture radar (SAR) systems, especially in mountainous areas and densely populated urban areas. The shadow can block the target of the observation area, making it impossible to obtain the scattering characteristics of the target. The layover causes phase distortion and alters target characteristics. Shadow and layover severely hinder the interpretation of SAR images. To confront the above problems, a multi-angle fusion algorithm based on unsupervised progressive segmentation network is proposed. Firstly, inspired by mega-constellations of low earth orbit, a spaceborne SAR collaborative observation model is proposed to generate multi-angle images of fluctuant terrain. Secondly, according to the difference of echos in the shadow and layover regions, an unsupervised progressive segmentation network is designed to sequentially segment the shadow and layover regions. Finally, to improve the contrast and brightness of the fused SAR image, a single-scale weighted fusion algorithm is designed. Experiments were conducted using the simulated multi-angle SAR images. Compared with single-angle images, the accuracy of target detection and figure-of-merit of the fused SAR image are significantly higher than those of other methods.

Medical image analysis plays a crucial role in linking perceptual mechanisms with clinical diagnosis, yet conventional deep learning models often rely on statistical correlations rather than modeling the underlying generative structure, leading to limited robustness in small-sample and cross-domain scenarios. To address this issue, we propose a hierarchical feature consistency framework named “MedCSS” that integrates causal self-supervised learning. Built upon a 3D ResNet backbone, the method aligns intermediate and high-level features through distributional consistency while introducing a coding rate–based causal regularization to suppress non-causal redundancy. Experiments on the MedMNIST3D benchmark demonstrate enhanced feature stability, boundary sensitivity, and generalization across diverse medical structures. Visualization analyses further reveal improved morphological coherence and causal interpretability. This study highlights the potential of causal self-supervision for structurally robust and semantically consistent representation learning in three-dimensional medical imaging.

Time series forecasting (TSF) plays a crucial role in many applications. Transformer-based methods are one of the mainstream techniques for TSF. Existing methods treat all token dependencies equally. However, we find that the effectiveness of token dependencies varies across different forecasting scenarios, and existing methods ignore these differences, which affects their performance. This raises two issues: (1) What are effective token dependencies? (2) How can we learn effective dependencies? From a logical perspective, we align Transformer-based TSF methods with the logical framework and define effective token dependencies as those that ensure the tokens as atomic formulas (Issue 1). We then align the learning process of Transformer methods with the process of obtaining atomic formulas in logic, which inspires us to design a method for learning these effective dependencies (Issue 2). Specifically, we propose Attention Logic Regularization (Attn-L-Reg), a plug-and-play method that guides the model to use fewer but more effective dependencies by making the attention map sparse, thereby ensuring the tokens as atomic formulas and improving prediction performance. Extensive experiments and theoretical analysis confirm the effectiveness of Attn-L-Reg.