Sea surface temperature (SST) plays an important role in affecting global climate, weather disasters, and marine resources. Full SST data that covers large areas and spans long periods is essential for various scientific research. Nowadays, meteorological satellites (e.g., the Himawari 8) have been able to provide large-scale, high-resolution continuous observations, but have always been interfered by cloud activities. While a lot of efforts have been made for the SST analysis, limitations associated with existing tools have not been resolved. Thus, based on interdisciplinary knowledge, we propose a physically-informed machine learning approach to elegantly reconstruct daily SSTs under both cloud and cloud-free areas. To capture the advection and diffusion processes, a TS-RBFNN (i.e., Temporal-Spatial Radial Basis Function Neural Network) is developed for SST reconstruction with applications in the Northwestern Pacific Ocean (NPO) and Taiwan’s adjacent waters (TAW). Overall, compared to the conventional DINEOF (i.e., Data Interpolation Empirical Orthogonal Function), the TS-RBFNN would better perform SST reconstruction with significant improvement up to 60%.

sea surface temperature,satellite observation,physical processes,machine learning