Development of continuous dynamic vertical reference for maritime and offshore engineering by applying geodetic and machine learning strategies

This study develops real-time continuous dynamic vertical reference for quantifying hydrodynamic processes with respect to high-resolution and accurate marine geoid model. In particular, this can now be realised through dynamic topography (DT), which is defined as the instantaneous sea surface height (SSH) deviation from the marine geoid (DT=SSH–geoid) and represents one of the most useful parameters of marine dynamics.

In regions of good quality and dense coverage of gravity data a 5 cm accurate marine geoid modelling is achievable. Due to the underlying accurate geoid model the DT values can now be estimated, eg. from a suitable hydrodynamic model, with the dm level accuracy. This corresponds to the most strict requirement for vertical accuracy at cargo handling in ports, dredging, maritime engineering, hydrography and under-keel clearance (UKC) management. This DT accuracy range also creates pre-conditions for identifying realistic sea level variations and circulation patterns of oceanic currents, seamlessly from the coastline toward the offshore over large basins.

Due to strict safety regulations various maritime and offshore applications require short term realistic sea level forecasts for hours to days in advance. Such near-real time DT estimations create a breakthrough opportunity for advancing from the “static” marine geoid referred vertical datum to the development of a new type vertical datum – a dynamic (both spatially and temporally) vertical reference frame. This continuous (and liquid!) DT field represents (either retrospectively or in the forecasting mode) the realistic sea level in absolute sense. Once DT is solved with sufficient accuracy then this dynamic DT field serves then as a reference (hence the name!) for developing further data products. These continuous DT field estimates are used for computing its spatio-temporal derivatives (eg. horizontal gradient), that might reveal ocean circulation patterns.

The Baltic Sea countries have been fortunate to have access to a wide range of marine data products, including that of a high-resolution marine geoid and hydrodynamic models that allow further capabilities to be explored in terms of sea level accuracy and validation. Accordingly, this study proposes a geodetic methodology that synergizes different sea level data sources by utilization of the marine geoid. The methodology applied utilized mathematical, statistical and machine learning strategies to obtain a spatio-temporally continuous dynamic topography of the Baltic Sea level. Sea level forecasting using DT is examined using machine learning approaches such as Convolution Neural Network. By using deep learning methods a DT modelling accuracy of within 10 cm has been achieved, which appears to better than the traditional data assimilation based forecasting.

Accomplishing this creates a marine dynamic vertical reference frame, which allows novel opportunities for marine digital twins, navigation, oceanographic processes and marine forecasting abilities.