On the potential of mapping sea level anomalies from Copernicus Marine Service with Random Forest Regression

The sea level observations from satellite altimetry are characterised by a sparse spatial and temporal coverage. For this reason, along-track data are routinely interpolated into daily grids provided by the Copernicus Marine Service. These are strongly smoothed in time and space and are generated using an optimal interpolation routine requiring several pre-processing steps and covariance characterisation.

In this study, we assess the potential of Random Forest Regression to estimate daily sea level anomalies. One-year-long records of along-track sea level are used to build a training dataset whose predictors are the neighbouring observations. The validation is based on the comparison against daily averages from tide gauges. The generated dataset is on average 10% more correlated to the tide gauge records than the commonly used product from Copernicus. As an example, four time series estimated from satellite altimetry from this study (ML, blue) and CMEMS (orange) at the closest point to four tide gauges (green) are shown in the attached figure. Also shown as text is the Root Mean Square Error (RMSE) of the altimetry dataset considering the tide gauges as ground-truth. Moreover, improvements in the temporal characterisation of the sea level variability will be shown by means of a coherence analysis to be spread over all subannual periods. While the current Copernicus daily sea level anomalies are more optimised for the detection of spatial mesoscales, we show how the methodology of this study can improve the characterisation of sea level variability, particularly in the coastal zone.

Our study fits into the use of Copernicus Marine Service data in the context of pan-European coastal zone monitoring, since this innovative machine-learning based technique is validated along the coast of the North Sea. A publication of this study is in advanced state of review in Ocean Dynamics, a pre-print of the first draft is freely available from https://doi.org/10.48550/arXiv.2207.11962.