AbstractThe multiannual (1993–2020) variability of sea level in the Baltic Sea is reconstructed by applying a Kalman filter approach. This technique learns how to generate data sets with the same statistics as the training data set, which in the studied case was taken from the CMEMS Baltic MFC operational model. It is demonstrated that using tide gauge data and statistical characteristics of the Baltic Sea from the model enables the generation of a high-resolution reconstruction of the sea surface height. Results obtained in this study demonstrated that the reconstruction method offers comprehensive high-resolution estimates (space and time) of sea level variability in the Baltic Sea based on tide gauge observations with high temporal resolution (e.g. hourly). The approach represents a valuable extension to the existing observing capabilities from altimetry, which do not capture sub-daily variations of sea level (e.g. storm surges). At the same time, the method consumes only a small fraction of the computational resources required by an assimilative model with comparable temporal/spatial resolution.