The sea level anomaly (SLA) is an important measure for describing the state and changes in the oceans. However, predicting SLA for the entire Adriatic basin is difficult due to spatially sparse measurements. These are limited to tide gauge stations and satellite altimetry, which cover a very small part of the basin. Methods such as HIDRA3 allow for accurate predictions at the locations of tide gauge stations. For a dense SLA field of the entire Adriatic Sea, however, we need a model that reconstructs the missing values from spatially sparse predictions. In this work, we address this problem with a new two-stage method. We map the spatially sparse sea surface height (SSH) predictions and the predictions of atmospheric variables into a latent representation of the SLA state. This representation is then reconstructed by a variational autoencoder decoder into an SLA field for the entire Adriatic Sea. A comparison of the experimental results with the results of existing numerical models shows the potential for further research of such an approach for SLA prediction.
|
