Partial-convolution-implemented generative adversarial network for global oceanic data assimilation
The oceanic data assimilation (DA) system, essential for understanding and forecasting global climate variabilities, blends prior information from numerical models with observed data to create the best possible estimates and their uncertainties of oceanic conditions. We developed DeepDA, a global oceanic DA system using deep learning, by integrating a partial convolutional neural network and a generative adversarial network (GAN). The partial convolution serves as an observation operator, mapping irregular observational data onto gridded fields, while the GAN incorporates observational information from previous time frames. Our observing system simulation experiments revealed that DeepDA significantly reduces the analysis error in three-dimensional temperature measurements, outperforming both background and observed values. DeepDA's global temperature reanalysis from 1981-2020 accurately reconstructs observed global climatological fields, seasonal cycles, major oceanic temperature variabilities, and the global warming trend. Developed solely with a long-term control simulation, DeepDA lowers the technical hurdles in creating global ocean reanalysis datasets using multiple numerical models' physical constraints, thereby would diminish systematic uncertainties in estimating global oceanic states over decades with these models.