Reconstruction of Subsurface Temperature Anomaly in the West Pacific Ocean Based on Spatial Clustering Methods

Reconstruction of ocean subsurface temperature anomalies (STA) is of great significance for the study of complex and variable ocean phenomena. Nowadays, STA reconstruction using deep learning is an important method, but the current reconstruction method still has the problem of incomplete feature extraction for multi-scale ocean dynamical processes. To address this problem, this study proposes a Transformer Network based on K-means clustering, referred to here as KMT, to reconstruct the STA in the Western Pacific using Argo measurements and multi-source satellite remote sensing data, and compares it with the Transformer Network inversion. The results show that the reconstruction accuracy of KMT is significantly better than that of Transformer Network, especially in the upper 200 m ocean where the temperature change situation is complicated.The average RMSE of the STA reconstruction accuracy of Transformer Network is 0.91 °C, and the average MAE is 0.73 °C , while the average RMSE of the STA reconstruction accuracy of KMT reaches 0.74 °C , and the average MAE reaches 0.57 °C. This study provides a more efficient and accurate STA reconstruction method for the West Pacific region, which can help to better understand and predict complex dynamical processes at different depths in the ocean.