Regime shifts in shallow lakes can lead to great changes in ecosystem structures and functions, making ecosystem management more complicated. Lake Yilong, located in Yunnan Province, is one of the most eutrophic lakes in China. Although there is a high possibility that this lake has undergone regime shift one or more times, the presence of regime shifts and their drivers remain unknown. Here, we employed the sequential t-test analysis of regime shifts to detect the regime shifts based on the long-term (1989–2018) dataset of the lake. We further determined their potential drivers, and explored the nutrient thresholds of regime shifts and hysteresis. The results showed that during the testing period, three regime shifts occurred in 1996 (restorative type), 2009 (catastrophic type) and 2014 (restorative type). The potential key drivers for the first two regime shifts (1996 and 2009) were both related to aquaculture. The abolition of cage fish culture may have led to the restorative regime shift in 1996, and the stocking of crabs and excessive premature releasing of fry possibly caused the catastrophic regime shift in 2009. However, the third regime shift, which occurred in 2014, was possibly related to the drought and succedent hydration. These results indicate that adjustments of aquaculture strategy and hydrological conditions are critical for the lake ecosystem’s recovery. Moreover, the total phosphorus thresholds were identified to be lower than 0.046 mg/L (restorative type) and higher than 0.105 mg/L (catastrophic type), respectively. In addition, an obvious hysteresis was observed after 2014, suggesting that nutrient reduction is important for this lake’s management in the future.

Landslide disasters are complex spatiotemporal phenomena. Existing deep learning (DL) models for remote sensing (RS) image analysis primarily exploit shallow visual features, inadequately incorporating critical geological, geographical, and environmental knowledge. This limitation impairs detection accuracy and generalization, especially in complex terrains and diverse vegetation conditions. We propose Knowledge Graph Embedding in Swin Feature Pyramid Networks (KGE–SwinFpn), a novel RS landslide segmentation framework that integrates explicit domain knowledge with deep features. First, a comprehensive landslide knowledge graph is constructed, organizing multi-source factors (e.g., lithology, topography, hydrology, rainfall, land cover, etc.) into entities and relations that characterize controlling, inducing, and indicative patterns. A dedicated KGE Block learns embeddings for these entities and discretized factor levels from the landslide knowledge graph, enabling their fusion with multi-scale RS features in SwinFpn. This approach preserves the efficiency of automatic feature learning while embedding prior knowledge guidance, enhancing data–knowledge–model coupling. Experiments demonstrate significant outperformance over classic segmentation networks: on the Yuan-yang dataset, KGE–SwinFpn achieved 96.85% pixel accuracy (PA), 88.46% mean pixel accuracy (MPA), and 82.01% mean intersection over union (MIoU); on the Bijie dataset, it attained 96.28% PA, 90.72% MPA, and 84.47% MIoU. Ablation studies confirm the complementary roles of different knowledge features and the KGE Block’s contribution to robustness in complex terrains. Notably, the KGE Block is architecture-agnostic, suggesting broad applicability for knowledge-guided RS landslide detection and promising enhanced technical support for disaster monitoring and risk assessment.

In recent years, deep neural network-based knowledge graph embedding (KGE) has become prevalent for link prediction. However, existing methods for KGE adopt single embedding arrangements, which limits the fine-grained interaction between entities and relations. Furthermore, these models inadequately incorporate global structural information, and thus lack the ability to capture long-range dependencies of entity-relation embedded elements. To address the above problems, we propose a novel Randomized and Global-aware Knowledge graph Embedding model, named RGaKE, which integrates randomized embedding strategy with global-aware module. Specifically, a randomized perturbation embedding strategy is first proposed to enhance the semantic expression diversity of entity-relation embedding through randomized arrangement. Furthermore, a global awareness module is introduced to strengthen the ability to capture global contextual information through spatial attention and channel recalibration. Finally, the multi-scale feature attention fusion module dynamically aggregates interaction features across different receptive fields and captures complex relational patterns and contextual semantics. The experimental results on four representative knowledge graph benchmark datasets, i.e. FB15K-237, WN18RR, YAGO-10, and KINSHIP demonstrate that the proposed RGaKE exhibits significant advantages in link prediction task and superior generalization ability.