Aug 2026· Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2· pp. 11576-11587· 0 citations· 19 references
TL;DR
Jingwei is a 4D spatiotemporal graph transfer learning model that reconstructs ocean oxygen levels globally over the past six decades on a 1◦ x 1◦ grid, covering depths of 0-5500 meters, and foresee Jingwei revolutionizing observation-based ocean modeling and deepening the authors' understanding of the breathless ocean.
Abstract
Climate change and anthropogenic activities have exacerbated hypoxic conditions in the global ocean, posing a serious threat to marine ecosystems. Quantifying changes in dissolved oxygen levels is crucial for understanding the impact of these factors on marine life and earth sustainability. However, dissolved oxygen observations are severely sparse, limiting comprehensive analysis. Here we present Jingwei, a 4D spatiotemporal graph transfer learning model that reconstructs ocean oxygen levels globally over the past six decades on a 1◦ x 1◦ grid, covering depths of 0-5500 meters. Jingwei simultaneously leverages intra-profile knowledge transfer using a pattern bank from simulations and captures inter-profile correlations through zoning-varying message passing among observations. It significantly reduces reconstruction error by 27.26% compared to CMIP6, demonstrates high consistency with cruise surveys, and provides satisfactory visual quality. Furthermore, Jingwei produces interpretable results, effectively identifying vertical profile patterns and distinguishing fine-grained spatial distributions. Jingwei provides cartography and quantitative analysis of oxygen minimum zones (OMZ) evolution since 1960. We foresee Jingwei revolutionizing observation-based ocean modeling and deepening our understanding of the breathless ocean. Alongside, we have released an open-source online platform (https://jingwei.acemap.info/ https://jingwei.acemap.info/), providing data visualization, resource sharing and ongoing updates.
The first oceanic 4D sparse observation reconstruction dataset, named OceanVerse, is presented, providing a novel large-scale dataset that meets the MNAR (Missing Not at Random) condition, supporting more effective model comparison, generalization evaluation and potential advancement of scientific reconstruction architectures.
Bin Lu, Jingjing Shen, Ze Zhao et al.· Proceedings of the 32nd ACM...· 0 citations
Accurate ocean temperature forecasting underpins fishery management, maritime safety, and climate early warning, yet its central difficulty lies in pronounced spatiotemporal heterogeneity that challenges uniform modeling: the same periodic forcing can elicit region-dependent responses across latitudes, and the same spatial proximity can reflect different physical relationships across oceanic regimes. Spectral analysis of multidecadal daily observations from three Chinese sea areas reveals that the dominant periodic structure varies by three orders of magnitude across latitudes, accompanied by a sevenfold contraction in seasonal amplitude. Motivated by these observations, we propose OceanMind, a framework that equips each stage of representation learning with explicit regime awareness. Along the temporal dimension, the model adapts its candidate frequency set to the spectral structure of the target domain and learns station-specific amplitude and phase to produce spatially differentiated periodic encodings. Along the spatial dimension, it learns a node-to-region projection and performs differentiated graph convolution in the compressed region space, while a complementary topology-free routing pathway models teleconnection-inspired remote associations beyond local graph topology. On three long-term daily sea-area benchmarks spanning 7.5°N to 41°N, OceanMind reduces mean MSE by 8%–14% over the strongest baseline across all forecast horizons while requiring one to several orders of magnitude less training time and memory than its closest competitors; an additional Himawari-8 clear-sky satellite benchmark further supports its cross-product robustness. The model-learned regional projection patterns and period importance weights are consistent with oceanographic priors, suggesting that heterogeneity-aware design improves accuracy while providing post-hoc interpretability. The source code is publicly available at https://github.com/OvOYu/OceanMind
Shiyu Yang, Hengyu Guo, C. Fan et al.· IEEE Transactions on Geoscie...· 0 citations
Ocean dissolved oxygen (DO) is a key indicator for assessing the health of marine ecosystems. However, constrained by sparse observational data and insufficient spatiotemporal resolution, traditional reconstruction methods struggle to accurately characterize its nonlinear spatial distribution. This paper proposes a conditional denoising diffusion probabilistic model (CD-DDPM) for ocean dissolved oxygen super-resolution reconstruction. Taking the South China Sea as the study area, we perform a reconstruction task from low-resolution (one-degree by one-degree) to high-resolution (quarter-degree by quarter-degree) using the Copernicus Marine Environment Monitoring Service (CMEMS) global reanalysis product. The model employs U-Net as the denoising network, integrating temperature and salinity (as auxiliary fields) with dissolved oxygen to form a multi-channel input. Additionally, a loss function is introduced to exclude interference from land pixels, and an exponential moving average strategy is adopted to stabilize the training process. Experimental results on the test set show that CD-DDPM achieves a Pearson correlation coefficient of 0.9532, significantly outperforming SRCNN. From the perspectives of PSNR, RMSE, and SSIM, our method also exhibits superior performance compared with traditional optimal interpolation and baseline deep learning models. Ablation experiments further demonstrate that removing both temperature and salinity auxiliary fields increases root mean square error by 98 percent and decreases peak signal-to-noise ratio by 6.72 dB, confirming the irreplaceable constraint effect of multivariate physical information on reconstruction accuracy. This study is the first to apply diffusion models to ocean dissolved oxygen super-resolution reconstruction, providing a new technical pathway for generating high-resolution dissolved oxygen data under sparse observation conditions.
Qian-Hao Li, Jianfeng Chen, Shenyao Wu et al.· International Conference on...· 0 citations
Sea surface temperature (SST), as a key variable in the ocean-climate system, plays a crucial role in global climate evolution, the occurrence of extreme weather events, and changes in marine ecosystems. Accurate SST prediction is of great significance for improving medium- and long-term climate forecasting capabilities and supporting effective marine environmental management. In recent years, deep learning methods have been widely applied to SST prediction tasks and have achieved promising results. However, existing approaches often suffer from inadequate modeling of temporal periodic structures and multilevel spatial features, making it difficult to effectively capture the multiscale dynamic variations of SST under complex spatiotemporal contexts. To address this issue, this article proposes a multiscale periodic spatiotemporal graph convolutional network (MPSGCN) to model the multiscale periodic spatiotemporal dependencies of SST. The proposed model explicitly captures periodic patterns in SST sequences through a periodic modeling mechanism, integrates multiscale adaptive graph convolution to dynamically learn regional dependencies at varying spatial scales, and employs multilayer spatial convolutions to mitigate the over-smoothing issue introduced by spectral-based graph convolution. Extensive experiments conducted on three representative marine regions with distinct climatic characteristics demonstrate that the proposed model significantly outperforms existing state-of-the-art methods across multiple evaluation metrics, validating the effectiveness and robustness of MPSGCN in modeling the complex spatiotemporal evolution of SST. Code available at https://github.com/xiaoling-0621/MPSGCN
Ling Xiao, Peihao Yang, Lang He et al.· IEEE Transactions on Geoscie...· 0 citations
Marine biogeochemical forecasting is increasingly important for managing marine ecosystems and the carbon cycle, yet global, seasonal forecast products lag far behind physical oceanography, held back by the complexity of the processes involved and by data scarcity. We introduce BG4Sea, which to our knowledge is the first global, data-driven system to produce multivariate seasonal forecasts of the marine biogeochemical state. BG4Sea is a modular architecture with a column autoencoder that compresses the vertical column into a low-dimensional latent space, a latent forecaster propagates this representation forward in time, a surface-forcing conditioner that injects physical boundary information via Feature-wise Linear Modulation (FiLM), and a horizontal-coupling module that incorporates neighboring-column context through cross-attention. The model is trained and evaluated on the global ocean reanalysis BIORYS4 (NEMO/PISCES), and produces six-month forecasts at 1/4 degree, monthly resolution for dissolved chemistry, biology, and carbon-pool variables, outperforming persistence and climatology across most variables and lead times. We position BG4Sea as an interpretable baseline for future, more expressive approaches, and discuss predictability attribution to each component, alongside the model's structural limitations.
Gabriela Martinez Balbontin, A. Charantonis, Dominique Béréziat et al.· 0 citations
Soil temperature is a key component of the climate system, influencing surface–atmosphere heat exchange, ecosystem processes and agricultural microclimates. Its temporal evolution is governed by complex interactions among environmental factors, surface disturbances and vertical heat conduction, making accurate prediction challenging. To address these challenges, this study introduces a Physics‐Inspired Time‐Gradient Adaptive Graph Convolutional Network (PTAGCN) coupled with a Physics‐Inspired attention mechanism. The framework captures local temporal dynamics within soil layers while modelling thermally coupled interactions across layers and enforces gradient and scale consistency to ensure physically plausible predictions. Experiments conducted across distinct climate regions in China—Fengyun Village, Chongqing and Naiman, Inner Mongolia—demonstrate the effectiveness of the proposed approach, achieving an RMSE of 1.5633°C, an MAE of 1.2670°C and a Nash–Sutcliffe efficiency (NSE) of 0.9759 in forecasting 20 cm soil temperature. These results indicate that the model outperforms conventional data‐driven approaches while maintaining strong physics‐inspired interpretability. The framework provides a robust and generalisable tool for climate system modelling, regional environmental monitoring and data‐driven agricultural management.
Yuqiang Yang, Kun Song, Huanzhi Luo· International Journal of Cli...· 0 citations