Xuefeng Zhang

Dr. Xuefeng Zhang


Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
30 May 2026 in npj Climate and Atmospheric Science
Verified
Journal Article
Open access
PDF file
YNYuan Niu
Yuan Niu
QHQiusheng Huang
Qiusheng Huang
XZXiaohui Zhong
Xiaohui Zhong
AGAnboyu Guo
Anboyu Guo
LCLei Chen
Lei Chen
HLHao Li
Hao Li
Accurate and physically realistic ocean forecasting at high spatial and temporal resolution is essential for resolving fine-scale dynamical features, with profound implications for hazard prediction, maritime navigation, and sustainable ocean management. Conventional numerical models can produce sub-daily, eddy-resolving forecasts but require substantial computational resources and often struggle to maintain predictive skill at such fine scales. Deep learning offers a promising alternative with significantly higher computational efficiency. However, most existing models operate at daily resolution, averaging out sub-daily variability and thus struggling to learn the complex and diverse diurnal cycle characteristics across different variables and depth layers. This limits their ability to capture rapidly evolving ocean processes, and they remain heavily dependent on atmospheric forcing from numerical weather prediction (NWP) models. Here, we introduce FuXi-Ocean, a global deep learning-based ocean forecasting model that generates 6-hourly predictions at 1/12 ∘ mesoscale-eddy-resolving resolution, extending to 1500 m depth. Unlike previous approaches, FuXi-Ocean operates independently of NWP forecasts, using near-surface atmospheric variables only for initialization, thereby establishing a fully independent, coupled ocean-atmosphere prediction framework. The model adaptively integrates forecasts across multiple temporal windows, capturing variable-specific evolution rates in ocean dynamics, while effectively addressing the technical challenge of cumulative error amplification in autoregressive forecasting. Through its innovative architectural design, FuXi-Ocean maintains stable predictive performance despite requiring four times more autoregressive steps than daily-resolution models. Based on our evaluation for 2022, FuXi-Ocean demonstrates improved performance over the latest deep learning-based GOFS (WenHai) in 10-day forecasts of temperature, salinity, zonal and meridional currents, sea surface temperature, and sea level anomaly. By advancing global ocean forecasts from daily to sub-daily resolution through independent deep learning-based ocean-atmospheric coupling, FuXi-Ocean represents a significant step forward in high-resolution ocean prediction.

31 March 2026 in Oceans
Verified
Journal Article
Open access
PDF file
SCShao Caixia
Shao Caixia
ZDZhang Dianjun
Zhang Dianjun
XZXuefeng Zhang
Xuefeng Zhang
Ocean fronts play a vital role in modulating climate variability, driving material transport, and maintaining the stability of marine ecosystems. Therefore, accurate identification of ocean fronts is of great significance for marine environmental monitoring and resource management. This study focuses on the Northwestern Pacific region and conducts a systematic comparison between two representative deep learning models—U-Net and Mask R-CNN—for automated ocean front detection. The objective is to evaluate the adaptability and strengths of different network architectures in handling multi-scale features, complex background conditions, and boundary delineation, thereby providing a theoretical basis for model selection and application-specific deployment. Experimental results show that U-Net achieves superior spatial consistency in large-scale frontal segmentation, with an IoU of 0.81 and a Dice coefficient of 0.76, while maintaining relatively high computational efficiency. In contrast, Mask R-CNN demonstrates stronger boundary modeling capabilities in detecting small-scale fronts and handling heterogeneous backgrounds, achieving an IoU of 0.78 and a Dice score of 0.73, though at the cost of increased computational demand. Overall, U-Net is more suitable for broad-scale automatic detection of ocean fronts, whereas Mask R-CNN exhibits greater potential in complex scene recognition. Integrating the structural advantages of both models holds promise for further enhancing the stability and accuracy of frontal detection, thereby offering robust technical support for ocean remote sensing analysis and environmental forecasting.

6 March 2026 in Remote Sensing
Verified
Journal Article
Open access
PDF file
XHXu Han
Xu Han
DWDaoming Wei
Daoming Wei
XZXuefeng Zhang
Xuefeng Zhang
JZJie Zhang
Jie Zhang
JSJiren Sun
Jiren Sun
ZDZhang Dianjun
Zhang Dianjun
Accurate three-dimensional (3D) salinity fields are crucial for diagnosing freshwater transport and upper-to-intermediate ocean dynamics, yet subsurface salinity observations remain uneven in space and time, especially away from primary observing corridors. Gridded fields derived from sparse measurements can also suffer from regional biases and over-smoothing, which may blur mesoscale signals in energetic regimes. This study presents an empirical orthogonal function-guided U-shaped Vision Transformer (EOF-UViT) to reconstruct daily 3D salinity in the Northwest Pacific (0–50°N, 100–150°E) from multi-source surface remote-sensing factors. Compared with the U-Net baseline and the Modular Ocean Data Assimilation System (MODAS) reconstruction, EOF-UViT produces more realistic horizontal structures and improved vertical consistency, with the largest gains in dynamically active regions. Comparison with collocated in situ profiles further supports the reconstruction skill (RMSE = 0.094 psu; R2 = 0.904), with estimates clustering more tightly around the 1:1 line than MODAS. Overall, EOF-UViT provides an efficient, observation-driven route to spatially coherent 3D salinity fields, supporting applications such as model initialization, assimilation background fields, and basin-scale salinity variability analysis.

Verified
Journal Article
Open access
PDF file
QYQi Yang
Qi Yang
ZDZhang Dianjun
Zhang Dianjun
JWJun Wang
Jun Wang
RXRuixue Xia
Ruixue Xia
XZXuefeng Zhang
Xuefeng Zhang
Accurate three-dimensional temperature and salinity initial fields are essential for regional short-term ocean forecasting, but subsurface constraints remain limited in the Western Indian Ocean because in situ profiles are sparse and satellite observations mainly describe the sea surface. This study evaluated a method to improve short-term forecasts by assimilating pretrained language model (PLM)-reconstructed thermohaline fields into the Finite Volume Community Ocean Model with a three-dimensional variational data assimilation scheme (FVCOM–3DVAR). The method was applied to the Western Indian Ocean, covering 15° S–10° N, 33° E–60° E. A non-assimilation control experiment (Control_run), Modular Ocean Data Assimilation System assimilation (MODAS_ass), and PLM reconstructed-field assimilation (PLM_ass) were conducted to evaluate the forecast performance with 5-day rolling forecasts. World Ocean Database (WOD) profiles were used for profile-based validation, and Copernicus Marine Environment Monitoring Service (CMEMS) gridded fields were used for spatially continuous reference evaluation in February, May, August, and November 2021. The results demonstrated that PLM_ass produced lower temperature and salinity root-mean-square errors (RMSEs) than Control_run and MODAS_ass at most depths and lead times. Relative to Control_run, PLM_ass reduced mean temperature RMSE by 28.0% at 100–200 m and mean salinity RMSE by 23.5% at 0–100 m; compared with MODAS_ass, the reductions were approximately 16.6% and 14.2%, respectively. Spatial diagnostics showed variable-, depth-, and region-dependent impacts with the most robust improvement in thermocline temperature forecasts. This study provides a feasible pathway for incorporating AI-reconstructed subsurface information into regional ocean forecasting systems under sparse observation conditions.