This dataset provides absolute dynamic topography (DT) of the Baltic Sea, obtained through a synergistic integration of multiple data sources: hydrodynamic model (HDM), tide gauge records, and satellite altimetry. The backbone of the dataset is the Nemo-Nordic model because of its high temporal and spatial resolution. To improve the accuracy of the sea level derived from the HDM, a correction was applied using a deep neural network. The neural network learned to predict the "modeling errors" in both time and space dimensions and identified relationships between causal (spatiotemporal) input variables and these errors across different locations. As demonstrated in Jahanmard et al. (2023), the neural network successfully reduced the modeling errors and limited them to a high-frequency band. This approach significantly improved the RMSE of the corrected HDM from 7.6 cm to 3.5 cm when compared to tide gauge records and from 6.5 cm to 4.1 cm when compared to satellite altimetry data (used as an external validation source). Furthermore, by employing geodetic and oceanographic approaches for DT determination, we accurately referenced the corrected HDM to the Baltic Sea Chart Datum (BSCD2000) with a reference bias of 18.1 ± 2.9 cm. As a result, the zero reference surface of the corrected HDM aligns with BSCD2000, which is a common geoid-based chart datum for Baltic countries. The corrected DT is in European Vertical Reference System (EVRS2000) with the origin zero level of Normaal Amsterdams Peil (NAP) and reference epoch of 2000.0.

For more details about absolute dynamic topography, sea level correction, and the neural network, please refer to the following references:

Jahanmard, V., Delpeche-Ellmann, N. and Ellmann, A., 2022. Towards realistic dynamic topography from coast to offshore by incorporating hydrodynamic and geoid models. Ocean Modelling. https://doi.org/10.1016/j.ocemod.2022.102124

Jahanmard, V., Hordoir, R., Delpeche-Ellmann, N. and Ellmann, A., 2023. Quantification of Hydrodynamic Model Sea Level Bias Utilizing Deep Learning and Synergistic Integration of Data Sources. Ocean Modelling. https://doi.org/10.1016/j.ocemod.2023.102286